article stringlengths 1.98k 169k | summary stringlengths 1.01k 4.15k | section_headings listlengths 2 38 | keywords listlengths 0 12 | year stringclasses 11 values | title stringlengths 30 189 |
|---|---|---|---|---|---|
The neuron specific RNA-binding proteins NOVA1 and NOVA2 are highly homologous alternative splicing regulators . NOVA proteins regulate at least 700 alternative splicing events in vivo , yet relatively little is known about the biologic consequences of NOVA action and in particular about functional differences between NOVA1 and NOVA2 . Transcriptome-wide searches for isoform-specific functions , using NOVA1 and NOVA2 specific HITS-CLIP and RNA-seq data from mouse cortex lacking either NOVA isoform , reveals that NOVA2 uniquely regulates alternative splicing events of a series of axon guidance related genes during cortical development . Corresponding axonal pathfinding defects were specific to NOVA2 deficiency: Nova2-/- but not Nova1-/- mice had agenesis of the corpus callosum , and axonal outgrowth defects specific to ventral motoneuron axons and efferent innervation of the cochlea . Thus we have discovered that NOVA2 uniquely regulates alternative splicing of a coordinate set of transcripts encoding key components in cortical , brainstem and spinal axon guidance/outgrowth pathways during neural differentiation , with severe functional consequences in vivo .
During central nervous system ( CNS ) development , a neuron extends its axon through a complex yet precise path to reach its final destination by sensing extracellular signals called guidance cues . These cues are sensed by the growth cone , a motile structure at the extending axon edge , and they control growth cone motility through directed cytoskeletal remodeling . Netrins , slits , semaphorins , and ephrins are the major classic guidance cues and elicit attractive or repulsive responses in growth cones via specific receptors ( Brose et al . , 1999; Cheng et al . , 1995; Drescher et al . , 1995; Fan and Raper , 1995; Kapfhammer and Raper , 1987; Kennedy et al . , 1994; Kidd et al . , 1999; Serafini et al . , 1994 ) . An important aspect of axon guidance is the spatial and temporal control of response to the guidance cues . For example , the spinal cord commissural axon reaching the midline senses netrin-1 , secreted from the floorplate as a chemoattractive cue; however , once it has crossed the floorplate , this cue becomes repulsive ( Kennedy et al . , 1994; Kidd et al . , 1998; Tessier-Lavigne et al . , 1988; Zou et al . , 2000 ) . Furthermore , the spatiotemporally restricted expression of Robo3 alternative splicing isoforms in spinal cord commissure axons are essential for the switching of the growth cone response to the axon guidance cues ( Chen et al . , 2008 ) , indicating that spatiotemporally regulated protein isoform expression and diversity is crucial to establish proper neuronal networks . Alternative splicing and alternative polyadenylation can produce multiple messenger RNAs ( mRNAs ) possessing distinct coding and regulatory sequences from a single gene . The regulated processes that generate such mRNA diversity are orchestrated by RNA-binding proteins ( RBPs ) . In the nervous system , alternative splicing has many important roles , including controlling the spatial and temporal expression of protein isoforms that are necessary for neurodevelopment and the modification of synaptic plasticity ( Li et al . , 2007; Licatalosi et al . , 2008; Ule and Darnell , 2006 ) . Significantly , human genetic studies have indicated that RNA misregulation resulting from defects in RBP expression and function are linked to numerous diseases , including Fragile X syndrome , spinal muscular atrophy , spinocerebellar ataxias , motoneuron disease and others ( Cooper et al . , 2009; Darnell , 2010; Lukong et al . , 2008 ) . NOVA1 and NOVA2 , RBPs initially identified as targets in autoimmune motor neuron disease ( Buckanovich et al . , 1993; Darnell and Posner , 2003 ) , are RNA-binding splicing regulators specifically expressed in neurons in the CNS ( Buckanovich et al . , 1996; Yang et al . , 1998 ) . NOVA1 and NOVA2 harbor three KH-type RNA binding domains , and in vitro RNA selection ( Buckanovich and Darnell , 1997; Jensen et al . , 2000b; Yang et al . , 1998 ) and X-ray crystallography ( Lewis et al . , 2000; Teplova et al . , 2011 ) demonstrate that NOVA1 and NOVA2 bind directly to RNA sequences harboring YCAY motifs . The development of RNA:RBP crosslinking and immunoprecipitation ( CLIP ) ( Ule et al . , 2003; 2005a ) followed by high-throughput sequencing ( HITS-CLIP ) ( Licatalosi et al . , 2008 ) methods has enabled identification of genome-wide RBP:RNA interaction maps in vivo . The bioinformatic integration of HITS-CLIP binding data and functional outcome revealed by exon junction microarray and RNA-seq data sets led to the conclusion that NOVA regulates alternative splicing following the rule that the position of NOVA binding to pre-mRNA predicts its action to enhance or inhibit alternate exon inclusion ( Licatalosi et al . , 2008; Ule et al . , 2005b; 2006; Wu et al . , 2013 , Zhang et al . , 2010 ) . These analyses revealed ~700 NOVA1/NOVA2 target alternate exons and allowed to predict the biological process regulated by NOVA1/NOVA2 . However , these studies have found no biochemical actions that were unique to NOVA1 versus NOVA2 paralogues , reflecting their nearly identical KH-type RNA binding domains . In the present work , we find that NOVA2 uniquely regulates a series of alternative splicing events of axon guidance related genes , including deleted in colorectal carcinoma ( Dcc ) , Roundabout , Axon Guidance Receptor , Homolog 2 ( Robo2 ) , Slit homolog 2 ( Slit2 ) , and EPH Receptor A5 ( Epha5 ) . These findings derived from combining transcriptome-wide NOVA1 and NOVA2 specific HITS-CLIP with RNA-seq analysis in the cortex of mice lacking either Nova1 or Nova2 . Nova2 deficiency results in some common phenotypes , such as severe growth retardation , as previously reported in Nova1-/- mice ( Jensen et al . , 2000a ) , as well as unique phenotypes , including agenesis of the corpus callosum ( ACC ) in Nova2-/- mice . Unexpectedly , given that motoneurons express both NOVA1 and NOVA2 , a subset of motoneuron axons directed to the ventral diaphragm show outgrowth defects in Nova2-/- but not Nova1-/- mice . In addition , the efferent innervation to the cochlea which is derived during development from a ventral subpopulation of facial motoneurons , is normal in Nova1-/- mice , reduced in Nova2-/- mice , and completely stalled in the absence of both Nova1 and Nova2 , suggesting that alternative splicing regulator NOVA2 plays critical roles in axon pathfinding context in mammals and that NOVA1 may have a cooperative role in this process . Based on these observations from genome-wide and histological analyses of Nova2-/- mice , we conclude that NOVA2-mediated RNA regulation is essential for CNS development by regulating neural networks wiring .
We utilized targeted mutagenesis to disrupt Nova2 genetic function in mice . A targeting cassette was designed to replace the first exon of Nova2 gene ( Figure 1A ) . The genotype of the resulting progenies was confirmed both by Southern blot ( Figure 1B ) and PCR ( Figure 1C ) . The progeny displayed the expected Mendelian ratio of mice for the heterozygous and homozygous Nova2 mutation . To confirm that homozygous Nova2 mutant mice did not express NOVA2 protein , Western blot analysis was carried out on protein extracts from P0 cortex ( Figure 1D ) of wild-type , Nova2-/- , and Nova1-/- mice , using anti-NOVA1 , anti-NOVA2 , and antisera from a patient with paraneoplastic opsoclonus-myoclonus ataxia ( POMA ) , which recognizes all NOVA protein species ( Yang et al . , 1998 ) . Expression of NOVA2 protein isoforms was absent in Nova2-/-mice ( Figure 1D ) . 10 . 7554/eLife . 14371 . 003Figure 1 . Generation of Nova2 null mice and characterization of SuperNOVA2 . ( Ai ) The wild-type Nova2 locus illustrated contains the first exon ( green box , with initiator ATG indicated ) . ( Aii ) A targeting construct was generated harboring a genomic fragment ( left: 2 . 2 kb ) flanking the initiator methionine , an IRES-Cre FRT-NEO-FRT ( FNF ) insertion , and an intronic genomic fragment flanking the first coding exon ( right: 6 kb ) . ( Aiii ) The Nova2 null locus following FLP-mediated excision of FNF cassette . Restriction enzyme sites were indicated for BamHI ( B ) , HindIII ( H ) , SacI ( S ) , SmaI ( Sm ) and XbaI ( X ) . The probes position used for Southern blot was indicated in red . ( B ) Genotypic analysis of Nova2 null mice . Southern blot analysis was performed on tail DNA digested with BamHI , using the probe described in ( A ) . ( C ) Genotyping PCR analysis of Nova2 null mice . ( D ) Western blot analysis of NOVA1 and NOVA2 proteins . Extracts of mouse cortex ( 10 μg/lane ) were made from age-matched P0 wild-type , Nova2-/- , and Nova1-/- mice , loaded on SDS-PAGE gels , and blotted with anti-pan NOVA ( POMA antisera ) , anti-NOVA1 specific , anti-NOVA2 specific , anti-PTBP2 , and anti-GAPDH antibodies . Quantification and comparison of NOVA1 and NOVA2 proteins expression amounts in the cortex of wild-type , Nova2-/- , and Nova1-/- mice . Data are presented as mean ± SD . *p<0 . 05 , **p<0 . 01 ( n = 3 , Tukey’s multiple comparison test ) . ( E ) Characterization of superNOVA2 . Left Diagram showing a putative superNOVA2 initiator methionine ( green ) positioned 192 nt upstream of the known Nova2 initiator methionine . Right panel showing the NOVA proteins mobility on electrophoresis . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 00310 . 7554/eLife . 14371 . 004Figure 1—figure supplement 1 . Growth retardation of Nova2-/- mice . ( A ) Growth curves of Nova2-/- , Nova2+/- , and Nova2+/+ littermates . Weights of Nova2-/- ( blue ) , Nova2+/- ( green ) , and Nova2+/+ ( red ) mice were studied from birth to age 14 days ( n > = 7 ) . Data are presented as mean ± SD . ( B ) Survival curves of Nova2-/- ( blue ) , Nova2+/- ( green ) , and Nova2+/+ ( red ) littermates . Nova2-/- mice died on average between P14-P18 , whereas Nova2+/- showed no difference in survival from Nova2+/+ mice . Data are presented as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 00410 . 7554/eLife . 14371 . 005Figure 1—figure supplement 2 . Anti-NOVA2 and anti-NOVA1 antibodies specificity . Coronal brain sections of E16 . 5 wild-type , Nova1-/- , and Nova2-/- , stained for NOVA1 ( green ) , NOVA2 ( red ) , and DAPI ( blue ) . Scale bar; 500 μm . ( B ) Specificity of antibodies for immunoprecipitation . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 005 Interestingly , more than one protein isoform was absent in protein extracts from Nova2 null mice: a single 50–52 kDa NOVA2 protein species corresponding to the predicted size of the previously described Nova2 ORF , and a series of previously described ( Ule et al . , 2005a; Eom et al . , 2013 ) but uncharacterized protein isoforms of ~70 kDa recognized by anti-NOVA2 antibody and POMA antisera , which we named SuperNOVA2 . We found three putative initiator ATG sequences upstream of the known Nova2 start codon on the 5’ UTR of the transcript . The mobility of SuperNOVA2 on electrophoresis was comparable with the protein product expressed from SuperNOVA2 plasmid containing full-length sequence of the 5’ UTR ( Figure 1E ) , confirming that SuperNOVA2 translation started from an upstream codon in the Nova2 5’ UTR . Using anti-NOVA1 or SuperNOVA2/NOVA2 specific antibodies , we found that NOVA1 protein levels were significantly increased , by ~40% in Nova2-/- mice cortex ( Figure 1D ) but NOVA1 anatomic expression pattern was similar to wild-type mice sections ( data not shown ) . Thus NOVA1 has partially overlapping distributions with SuperNOVA2/NOVA2 ( hereinafter we defined SuperNOVA2/NOVA2 as NOVA2 ) , is upregulated by unknown means in the absence of Nova2 , and may have partially redundant functions with NOVA2 in the cortex . Nova2 null animals were born indistinguishable from littermates but failed to thrive , demonstrating progressive motor dysfunction and overt motor weakness , and they died an average of 14–18 days after birth ( Figure 1—figure supplement 1 ) . There was no apparent phenotype in Nova2 heterozygotes , although we previously found that when assessed by electroencephalography , 6 month old Nova2+/- ( and Nova1+/- ) mice had frequent synchronous cortical interictal discharges consistent with epilepsy ( Eom et al . , 2013 ) . In conclusion , we demonstrate that the Nova2 gene encodes for multiple protein isoforms , including an identified ~70 kDa SuperNOVA2 isoform , and that Nova2 is necessary for post-natal survival . The NOVA1/2 splicing-regulatory network and its target alternative splicing events have been previously defined by utilizing NOVA2 and pan-NOVA HITS-CLIP data , splicing sensitive microarray data , NOVA1/2-binding motifs , and integrative modeling ( Licatalosi et al . , 2008; Ule et al . , 2005b; 2006; Zhang et al . , 2010 ) , yet these data sets were integrated data sets from multiple developmental stages , CNS regions , and Nova1/Nova2 genotypes . Relatively little is known about the biologic consequences of NOVA1 and NOVA2 action and about the functional differences between NOVA1 and NOVA2 in a specified CNS region and at a particular developmental stage . To identify NOVA1 and NOVA2 direct RNA targets in E18 . 5 cortex , a transcriptome-wide library of NOVA1-RNA and NOVA2-RNA interactions were generated by HITS-CLIP from E18 . 5 wild-type mice cortex and sequenced . NOVA1 and NOVA2 CLIP reads were filtered and aligned to the mouse genome ( mm9 ) . A total of 2 , 318 , 553 unique NOVA2 CLIP tags were obtained from three biological replicates ( 483 , 446 , 836 , 678 , and 998 , 429 unique tags , respectively ) , with a total of 139 , 007 clusters from at least two or three biologic replicates ( defined as biological complexity of two ( BC2 ) ) . A total of 910 , 342 unique NOVA1 CLIP tags were obtained from three biological replicates ( 319 , 904 , 292 , 709 , and 297 , 729 unique tags , respectively ) , with a total of 54 , 546 clusters ( BC2 ) . To identify sequence features most commonly associated with either NOVA1-RNA or NOVA2-RNA interactions , 4 nucleotide ( nt ) long motifs were counted in BC2 NOVA1 and NOVA2 clusters and in a set of shuffled control sequences as a control . Each of the top five ranking tetramers in NOVA1 and NOVA2 clusters were the same ( UCAU , CAUC , CAUU , UCAC , and UUCA ) ( Figure 2A ) and coincided with the previously identified YCAY NOVA binding motif . Enrichment of the YCAY motif around NOVA1 and NOVA2 clusters was comparable between NOVA1 and NOVA2 and greatly enriched within NOVA1 and NOVA2 CLIP clusters ( Figure 2B ) , suggesting that NOVA1 and NOVA2 recognized the same YCAY RNA sequence . NOVA1 and NOVA2 CLIP tag positions on known NOVA regulated alternatively spliced exons ( Zhang et al . , 2010 ) were comparable ( Figure 2C and D ) . For example , alternative splicing of Agrn minigene containing exon 31–34 were regulated by NOVA1 and NOVA2 in the same splicing direction , in this case mediating the inclusion of the exon ( Figure 2—figure supplement 1 ) , indicating that NOVA1 and NOVA2 have similar RNA interaction profiles on NOVA1/2 target alternative splicing sites and are capable of regulating the same alternative splicing events . 10 . 7554/eLife . 14371 . 006Figure 2 . NOVA2 and NOVA1 HITS-CLIP . ( A ) Top 5 tetramers present in BC2 NOVA2 and NOVA1 sequences compared with the shuffled control sequences . The sequences corresponding to YCAY motif were high-lighted with magenta . ( B ) Enrichment of YCAY motif near NOVA2 ( blue ) and NOVA1 ( red ) clusters . ( C ) Distribution of BC2 NOVA2 ( blue ) and NOVA1 ( red ) CLIP tags ( Y-axis: CLIP tags density in each 25 nt window ) near known NOVA1/2 targeted alternative exons silencing ( top panel ) and splicing ( bottom panel ) events . ( D ) Examples of NOVA2 and NOVA1 CLIP cluster location near known NOVA1/2-target alternative splicing exon ( top panel; silencing event , Neo1 exon 25–27 , bottom panel; splicing event , Epha5 exon 6–8 ) . ( E ) Summary of robust NOVA2 ( blue ) and NOVA1 ( red ) CLIP BC2 clusters consisted of a minimum of 10 tags . ( F ) Genomic distribution of BC2 NOVA2 ( left ) and NOVA1 ( right ) clusters consisted of a minimum of 10 tags . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 00610 . 7554/eLife . 14371 . 007Figure 2—figure supplement 1 . Alternative splicing regulation of Agrn minigene by NOVA proteins . Alternative splicing impact on Agrn minigene was examined in HEK293T cells with or without NOVA proteins conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 00710 . 7554/eLife . 14371 . 008Figure 2—figure supplement 2 . Genomic distribution of BC2 NOVA2 specific ( left ) , NOVA1 specific ( middle ) , and NOVA2/NOVA1 common clusters consisted of a minimum of 10 tags . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 00810 . 7554/eLife . 14371 . 009Figure 2—figure supplement 3 . NOVA1 and NOVA2 expression in the mouse embryonic brain . ( A ) Immunohistochemistry of NOVA1 ( b , f , j , n: green ) and NOVA2 ( c , g , k , o: red ) in E18 . 5 coronal sections ( a–d ) , sagittal sections ( e–h ) , neocortex ( i–p ) . Scale bars; 500 μm ( a–h ) , 50 μm ( i–l ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 009 Twenty-six percent or eighteen percent , respectively , of the NOVA2 or NOVA1 BC2 clusters consisted of a minimum of 10 tags ( defined as peak height ( PH ) ; NOVA2; 36 , 591 of 139 , 007 , NOVA1; 9988 of 54 , 546 ) . 80 . 1% of NOVA1 clusters were shared with NOVA2 cluster positions , while 22 . 1% of NOVA2 clusters overlapped with NOVA1 ( Figure 2E ) . Interestingly , the genomic distribution of NOVA2 clusters and NOVA1 showed different distributions within their target transcripts ( Figure 2F and Figure 2—figure supplement 2 ) . 77 . 7% of NOVA2 clusters were located in introns and 13 . 9% in exons ( 5’ UTR , CDS , and 3’ UTR ) compared to 45 . 4% of NOVA1 clusters that were in introns and 45 . 1% in exons ( Figure 2F ) . 84 . 9% of NOVA2 specific BC2 clusters ( PH>=10 ) were enriched in introns , while 74 . 1% of NOVA1 specific BC2 clusters ( PH>=10 ) were in exons ( Figure 2—figure supplement 2 ) , indicating that NOVA2 binds preferentially to introns than exons , suggesting that NOVA2 may play a greater nuclear role than NOVA1 , and demonstrating that RNA-interaction profiles on a genome-wide scale are different between NOVA homologues . The distribution of NOVA2 throughout the brain mirrored previous immunohistochemical and in situ hybridization data ( Yang et al . , 1998 ) showed that NOVA2 was expressed at high levels in cortex and hippocampus , and at lower levels in midbrain and spinal cord , where NOVA1 was expressed at high levels in a generally reciprocal fashion , with low levels in the cortex and relatively high levels in the midbrain and spinal cord ( Figure 2—figure supplement 3 ) . The NOVA2 expressed cell in the cortical plate of neocortex was ubiquitously distributed at comparable expression level , yet NOVA1 was expressed in the specified cell types . Taken together , the HITS-CLIP and immunohistochemical data suggest that NOVA1 and NOVA2 perform unique biological functions in different brain areas and cell types , and that in those few cortical neurons expressing NOVA1 , NOVA2 might be expected to have some redundant activity , while the reciprocal may not so often be the case . To identify NOVA1 or NOVA2 target transcripts whose abundance or alternative splicing was changed in either Nova1-/- or Nova2-/- cortex , paired-end 125 nucleotide RNA-seq libraries were prepared from littermate E18 . 5 wild-type , Nova1-/- , and Nova2-/- mice cortex . A total of 225 , 054 , 059 and 213 , 596 , 515 unique reads were obtained from three biological replicates of wild-type and Nova1-/- mice , respectively , and a total of 190 , 282 , 241 and 180 , 200 , 564 unique reads were obtained from three biological replicates of wild-type and Nova2-/- mice , respectively . These experiments showed that the abundance of 60 transcripts were significantly changed in Nova2-/- mice compared to only 2 transcripts which showed steady-state changes in Nova1-/- mice compared to wild-type littermates ( FDR<0 . 05 and |log2FC|>=1 ) ( Figure 3A ) , consistent with lower levels of NOVA1 expression in cortex . More changes were seen in analyzing splicing-dependent changes , and this pattern of differential NOVA2 and NOVA1 effects was also seen . We integrated NOVA1 and NOVA2 specific HITS-CLIP data , RNA-seq data , and NOVA binding YCAY motifs to define 1991 NOVA2 and 58 NOVA1-mediated alternative splicing events in the mouse cortex ( Figure 3B and Figure 3—figure supplement 1 ) . Interestingly , there was no correlation between delta I ( ΔI ) of wild-type versus Nova1-/- and wild-type versus Nova2-/- ( R = 0 . 09 ) ( Figure 3C ) . Taken together , this data shows that Nova1 and Nova2 have different RNA regulatory networks in developing cortical neurons; while the quantitative differences likely relate to the greater and more unique cellular expression of NOVA2 relative to NOVA1 in cortex , they also support the suggestion that the two NOVA homologues may regulate unique biological processes . 10 . 7554/eLife . 14371 . 010Figure 3 . RNA-seq analysis in either Nova2-/- or Nova1-/- versus littermate wild-type control . ( A ) RNA abundance changes in Nova2-/- and Nova1-/- mice cortex and comparison of its correlation in Nova2-/- with in Nova1-/- mice . X-axis and Y-axis indicated log2 ( Nova2-KO/wild-type fold change ( FC ) ) and log2 ( Nova1-KO/wild-type FC ) , respectively . Transcripts significantly changed in Nova2-/- and Nova1-/- mice cortex were shown in blue and red , respectively ( FDR<0 . 05 and log2|FC|>=1 ) . ( B ) Summary of NOVA2- and NOVA1-dependent alternative splicing changes that showed |ΔI| >= 0 . 1 ( FDR < 0 . 1 ) ( Ule et al . , 2005b ) and that contained YCAY motif ( s ) within BC2 NOVA1 or NOVA2 HITS-CLIP clusters on alternative splicing exons and/or its upstream/downstream introns . 1991 NOVA2-target alternative splicing events on 540 genes ( known [Zhang et al . , 2010]; 780 events , novel; 1211 events ) , 58 NOVA1-target events on 20 genes ( known; 0 events , novel 58 events ) . ( C ) Correlation of Nova2- and Nova1-deficient impact on alternative splicing events . X-axis and Y-axis indicated ΔI of wild-type ( WT ) vs Nova2-/- and vs Nova1-/- , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 01010 . 7554/eLife . 14371 . 011Figure 3—source data 1 . Gene Ontology ( GO ) terms associated with NOVA2-target alternative splicing exons . Summary of GO analysis of NOVA2-regulated genes performed using DAVID Bioinformatics Resources ( https://david . ncifcrf . gov ) ( FDR<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 01110 . 7554/eLife . 14371 . 012Figure 3—source data 2 . The transcriptome abundance of the neuronal differentiation markers in the E18 . 5 cortex of Nova1-/- and Nova2-/- mice . log2FC: log2 ( Nova1 or Nova2 mutants/wild-type ) fold change . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 01210 . 7554/eLife . 14371 . 013Figure 3—figure supplement 1 . Summary of the number of alternative splicing events detected by RNA-seq analysis with/without NOVA2 or NOVA1 CLIP clusters and/or YCAY clusters . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 01310 . 7554/eLife . 14371 . 014Figure 3—figure supplement 2 . KEGG pathways over-represented among NOVA2-target genes . Axon guidance pathway ( FDR<0 . 05 ) . Red stars indicating NOVA2-target alternative splicing genes identified by combining RNA-seq with NOVA2 HITS-CLIP analysis . Figures obtained from GO analysis using DAVID Bioinformatics Resources ( https://david . ncifcrf . gov ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 01410 . 7554/eLife . 14371 . 015Figure 3—figure supplement 3 . KEGG pathways over-represented among NOVA2-target genes . Adherens junction pathway ( FDR<0 . 05 ) . Red stars indicating NOVA2-target alternative splicing genes identified by combining RNA-seq with NOVA2 HITS-CLIP analysis . Figures obtained from GO analysis using DAVID Bioinformatics Resources ( https://david . ncifcrf . gov ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 01510 . 7554/eLife . 14371 . 016Figure 3—figure supplement 4 . RNA-seq analysis in the midbrain and hindbrain of Nova1-/- versus littermate wild-type control . ( A ) Summary of NOVA2-dependent ( cortex ) and NOVA1-dependent ( midbrain and hindbrain ) alternative splicing changes that showed |ΔI|>=0 . 1 ( FDR<0 . 1 ) . ( B ) Correlation of Nova2-deficient ( cortex ) and Nova1-deficient ( midbrain and hindbrain ) impact on alternative splicing events . X-axis and Y-axis indicated ΔI of wild-type vs Nova2-/- ( cortex ) and vs Nova1-/- ( midbrain and hindbrain ) , respectively . ( C ) Examples of alternative splicing changes of the NOVA2-regulated axon guidance genes in the cortex of Nova2-/- and Nova1-/- and in the midbrain and hindbrain of Nova1-/- mice . Low expression genes in midbrain and hindbrain samples were filtered out . Y-axis indicated ΔI . Blue; Nova2-/- ( cortex ) , red; Nova1-/- ( cortex ) , orange; Nova1-/- ( midbrain and hindbrain ) . *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 016 We explored the biologic pathways regulated by NOVA1 and NOVA2 with gene ontology ( GO ) analysis on the transcripts harboring NOVA-regulated alternative exons . Among the KEGG ( Kyoto Encyclopedia of Genes and Genomes ) pathway terms associated with NOVA2-regulated genes ( FDR<0 . 05 ) were axon guidance signaling ( FDR=0 . 0039 ) and adherens junctions signaling ( FDR = 0 . 017 ) ( Figure 3—figure supplement 2 and 3 ) . In the GO terms associated with NOVA2-regulated genes ( FDR<0 . 05 ) , axonal projection related terms ( cell morphogenesis , neuron projection morphogenesis , axonogenesis , cell morphogenesis involved in neuron differentiation , and cell projection morphogenesis ) were also enriched ( Figure 3—source data 1 ) . NOVA2-dependent alternative splicing events of axon guidance related genes were validated by semi-quantitative RT-PCR with RNA prepared from E18 . 5 wild-type Nova2-/- , and Nova1-/- cortex ( Figure 4 and Figure 4—figure supplement 1 ) . The alternative splicing events ( Dcc exon 17 , Slit2 exon 28b , Robo2 exons 6b and 21 , Epha5 exon 7 , Arhgef12 exon 4 , Ppp3cb exon 10b , Neo1 exon 26 , and Rock1 exon 27b ) were significantly changed in Nova2-/- but not in Nova1-/- mice cortex , coinciding with RNA-seq results . No association was detected in the NOVA1-regulated alternative exons GO or KEGG pathways terms , likely due to the smaller number of NOVA1-regulated transcripts in cortex . 10 . 7554/eLife . 14371 . 017Figure 4 . NOVA2 unique alternative splicing events of axon guidance related genes in E18 . 5 mice cortex . Left Diagrams showing RefSeq annotation genes , changed alternative splicing events , RNA-seq results of wild-type ( grey ) and Nova2-/- ( blue ) , NOVA2 CLIP clusters ( light blue ) , RNA-seq results of wild-type ( grey ) and Nova1-/- ( red ) , and NOVA1 CLIP clusters ( pink ) . Right panels and graphs showing RT-PCR results and quantification data in E18 . 5 wild-type , Nova2-/- , and Nova1-/- mice cortex , respectively . *p<0 . 05 , **p<0 . 01 ( n = 3 , Tukey’s multiple comparison test ) . Data are presented as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 01710 . 7554/eLife . 14371 . 018Figure 4—figure supplement 1 . NOVA2 unique alternative splicing events of axon guidance related genes in E18 . 5 mice cortex . Left Diagrams showing RefSeq annotation genes , changed alternative splicing events , RNA-seq results of wild-type ( grey ) and Nova2-/- ( blue ) , NOVA2 CLIP clusters ( light blue ) , RNA-seq results of wild-type ( grey ) and Nova1-/- ( red ) , and NOVA1 CLIP clusters ( pink ) . Right panels and graphs showing RT-PCR results and quantification data of Ahrgef12 , Ppp3cb , Neo1 , and Rock1 in E18 . 5 wild-type , Nova2-/- , and Nova1-/- mice cortex , respectively . **p<0 . 01 ( n=3 , Tukey’s multiple comparison test ) . Data are presented as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 018 RNA-seq analysis in the E18 . 5 midbrain and hindbrain of wild-type and Nova1-/- mice , where NOVA1 is more abundantly expressed , identified that 119 of 1991 NOVA2-dependent alternative splicing events were changed in Nova1-/- mice ( |ΔI|>=0 . 1 , FDR<0 . 1 ) ( Figure 3—figure supplement 4A ) . There was no correlation between ΔI of wild-type versus Nova1-/- ( mid- and hindbrain ) and wild-type versus Nova2-/- ( cortex ) ( R=0 . 20 ) ( Figure 3—figure supplement 4B ) as well as identified between ΔI of wild-type versus Nova1-/- ( cortex ) and wild-type versus Nova2-/- ( cortex ) . In the midbrain and hindbrain of Nova1-/- mice , only one alternative splicing event ( Robo2 exon 6b ) was significantly changed in the NOVA2-regulated genes list of KEGG axon guidance pathway but it was smaller change when compared with ΔI of wild-type versus Nova2-/- ( cortex ) ( Figure 3—figure supplement 4C ) . Taken together , these data suggest that NOVA2 regulates a distinct set of transcripts related , in particular , to axon guidance , in the E18 . 5 mouse cortex . To pursue the potential connection between NOVA2 and axonal guidance , we examined NOVA2 expression during cortical mouse brain development . We found that NOVA2 expression level detected by immunofluorescence was high in the cortical plate ( CP ) and subplate ( SP ) that consists of post-mitotic neurons at E18 . 5 ( Figure 2—figure supplement 3 ) . Similarly a high-expression of NOVA2 in CP and SP was also observed at E12 . 5 , E14 . 5 , and E16 . 5 ( Figure 5A ) , indicating that cortical NOVA2 expression level was progressively up-regulated during neural differentiation from neural progenitor cells ( NPC ) . 10 . 7554/eLife . 14371 . 019Figure 5 . NOVA2 switches developmentally regulated exons usage of axon guidance related genes . ( A ) NOVA2 expression in developing neocortex ( left panels ) . High NOVA2 expression regions were high-lighted with red in middle panels . MZ: marginal zone , CP: cortical plate , SP: subplate , IZ: intermediate zone , SVZ: subventricular zone , VZ: ventricular zone . ( B ) Analysis of NOVA2-regulated exons in Dcc , Robo2 , Epha5 , Slit2 , and Neo1 in E12 . 5 wild-type , E18 . 5 wild-type , and E18 . 5 Nova2-/- cortex ( left panels ) . Quantification data of RT-PCR products were shown in right graphs . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 ( n = 3 , Tukey’s multiple comparison test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 019 To assess whether NOVA2 regulated alternative splicing is also regulated in a developmentally regulated manner , we compared expression of alternatively spliced NOVA2 target exons in E12 . 5 wild-type , E18 . 5 wild-type , and E18 . 5 Nova2-/- cortex ( Figure 5B ) . This revealed that NOVA2-target splicing events were developmentally regulated between E12 . 5 and E18 . 5 , such that splicing in E18 . 5 Nova2-/- cortex reverted to patterns seen earlier in E12 . 5 wild-type cortex . Based on these observations , we conclude that NOVA2 switches the alternative splicing patterns of a series of axon guidance genes during mouse cortical development . The observations that NOVA2 regulates alternative splicing events in transcripts encoding axon guidance signaling factors led us to test whether axon guidance itself is affected in Nova2 null mice . When coronal and horizontal sections of E18 . 5 mice brains were double-immunostained for L1 , an axonal marker , and TAG-1 , a corticofugal axon marker , we discovered that Nova2-/- mice , but not wild-type littermates , had agenesis of the corpus callosum ( ACC ) ( 14/14 mice ) ( Figure 6 and Figure 6—figure supplement 1A ) . ACC was in all Nova2-/- mice serial sections examined ( Figure 6—figure supplement 1B ) , was seen along with Probst bundles ( abnormal collections of cells characteristically seen in patients with ACC; arrows in Figure 6A–o and Figure 6—figure supplement 1A–l ) , and was also clearly evident in sections stained with the pioneer axon marker Neuropilin-1 ( Figure 6—figure supplement 1C ) . Interestingly , Nova1-/- mice showed normal CC formation ( Figure 6—figure supplement 1D ) . The anterior commissure axons were observed in both Nova2-/- and Nova1-/- mice ( Figure 6—figure supplement 2 ) . 10 . 7554/eLife . 14371 . 020Figure 6 . Agenesis of corpus callosum in Nova2-/- mice . ( A ) Immunohistochemistry of L1 ( green; a , e , i , m ) , TAG1 ( red; b , f , j , n ) proteins , DAPI ( blue; c , g , k , o ) and merged views ( d , h , l , p ) on coronal sections in E18 . 5 wild-type ( a–d , i–l ) and Nova2-/- ( e-h , m–p ) littermates . ( i–p ) Higher magnified view of anterior commissure region of a–h . Arrows indicated Probst bundles . Scale bars; 500 μm . ( B ) Commissure axons pathfinding defect in Nova2-/- mice . Coronal sections of the anterior ( a , b ) and posterior ( c , d ) telencephalon in P0 wild-type ( a , c ) and Nova2-/- ( b , d ) mice showing anterogradely labeled fibers after DiI crystal placements in the cingulate cortex . Arrows indicated the cortico-thalamic axon terminal into dorsal thalamus . Asterisks: DiI placed positions . Scale bars; 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 02010 . 7554/eLife . 14371 . 021Figure 6—figure supplement 1 . Loss of corpus callosum in Nova2-/- but not Nova1-/- mice . ( A ) Immunohistochemistry of L1 ( a , d , g , j ) , TAG1 ( b , e , h , k ) proteins , and DAPI ( c , f , I , l ) on horizontal sections in E18 . 5 wild-type ( a–c , g–i ) and Nova2-/- ( d–f , j–l ) littermates . ( g–l ) Higher magnified view of anterior commissure region of ( a-–f ) . Arrows indicated Probst bundles . Scale bars; 500 μm . ( B ) Serial section images of L1 immunostaining from anterior brain position to posterior . Wild-type images shows on left panels and Nova2-/- images on right panels . Scale bar; 500 μm . ( C ) Coronal brain sections of E18 . 5 wild-type ( a–d ) and Nova2-/- ( e–h ) littermates , stained for L1 ( green; aand e ) , Neuropilin-1 ( a pioneer axon marker; red; b and f ) , DAPI ( blue;cand g ) , and merged views ( d and h ) . Scale bar; 500 um . ( D ) Normal formation of corpus callosum in Nova1-/- mice . Coronal brain sections of E18 . 5 wild-type and Nova1-/- littermates , stained for L1 ( upper panels ) , NOVA1 ( middle panels ) , and DAPI ( lower panels ) . Scale bar; 500 um . ( E ) Commissure axons pathfinding defect in Nova2-/- mice . Coronal sections in wild-type ( a ) and Nova2-/- ( b ) mice showing anterogradely labeled fibers after DiI crystal placements in lateral neocortex . Asterisks: DiI placed positions . Scale bars; 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 02110 . 7554/eLife . 14371 . 022Figure 6—figure supplement 2 . Normal formation of anterior commissure axons in Nova1-/- and Nova2-/- mice . Coronal brain sections of E18 . 5 wild-type , Nova1-/- and Nova2-/- , stained for DAPI . Scale bar; 500 um . Red arrows: anterior commissure axons . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 02210 . 7554/eLife . 14371 . 023Figure 6—figure supplement 3 . Alternative splicing changes of the genes associated with mouse ACC phenotypes and human ACC syndromes . Left Diagrams showing RefSeq annotation genes , changed alternative splicing events , RNA-seq results of wild-type ( grey ) and Nova2-/- ( blue ) , NOVA2 CLIP clusters ( light blue ) , RNA-seq results of wild-type ( grey ) and Nova1-/- ( red ) , and NOVA1 CLIP clusters ( pink ) . Right panels and graphs showing RT-PCR results and quantification data in E18 . 5 wild-type , Nova2-/- , and Nova1-/- mice cortex , respectively . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 ( n = 3 , Tukey’s multiple comparison test ) . Data are presented as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 023 To independently corroborate these observations , corpus callosal commissural axons were visualized with DiI anterograde tracer placed into the cingulate cortex . No CC commissural axons were detected that crossed the midline in Nova2-/- brain , however , were clearly detected in anterior and posterior sections of wild-type brain at P0 ( Figure 6B ) . Corpus callosal commissural axons from the lateral neocortex also failed to cross the midline in Nova2-/- mice ( Figure 6—figure supplement 1E ) . Interestingly , corticothalamic axons of both wild-type and Nova2-/- mice terminated in a similar thalamic region at E18 . 5 ( arrows in Figure 6B ) , indicating that only specific aspects of axon guidance are disrupted in Nova2 null mice . These observations suggest that Nova2 but not Nova1 regulate a series of alternative splicing events that are necessary for forming the CC , and demonstrate that NOVA2 and NOVA1 have different functions in CC axon guidance pathways . To understand the axon guidance phenotype in more detail , we compared the pathway of L1 positive axons traversing the upper cortical layer in the cortex of Nova2-/- and wild-type mice at E16 . 5 and E18 . 5 ( E16 . 5; Figure 7 and E18 . 5; Figure 6A ) . In wild-type mice , L1 positive axons passing between the subplate and subventricular zone were detected as one wide bundle , whereas in Nova2-/- mice L1 positive axon routes were separated into two bundles ( Figure 7A ) . The L1 positive axons detected in the deeper cortical layer in Nova2-/- appeared normal , in that they passed through similar cortical layers as did axons in wild-type mice . In contrast , separated axons passing through the upper L1 positive axonal pathway , which were Netrin-G1 ( NTNG1 ) positive axons , were only seen in Nova2-/- mice and were passed along a subplate path enriched in NURR1 ( a subplate marker ) , as revealed by L1/NTNG1 or L1/NURR1 double-immunostaining ( Figure 7B and Figure 7C ) . These results indicate that only a portion of the L1 positive axonal path , which is NTNG1 positive , is specifically affected by the absence of Nova2 . We conclude that NOVA2 plays a role as a modifier for a unique set of neuron-specific axons in the developing cortex . 10 . 7554/eLife . 14371 . 024Figure 7 . Abnormal thalamo-cortical path in the cortex of Nova2-/- mice . ( A ) Immunohistochemistry of L1 ( c , d ) and NOVA2 ( e , f ) on coronal sections in wild-type ( a , c , e , g ) and Nova2-/- ( b , d , f , h ) at E16 . 5 . ( a , b ) Merged coronal section views of L1 ( green ) , NOVA2 ( red ) , and DAPI ( blue ) . ( c-h ) Higher magnified view of neocortex of ( Aa ) and ( Ab ) . Scale bars; 200 μm ( a , b ) , 50 μm ( c–h ) . ( B ) Immunohistochemistry of L1 ( c , d ) and NTNG1 ( e , f ) on coronal sections in wild-type ( a , c , e , g ) and Nova2-/- ( b , d , f , h ) at E16 . 5 . ( a , b ) Merged coronal section views of L1 ( green ) , NTNG1 ( red ) , and DAPI ( blue ) . ( c-h ) Higher magnified view of neocortex of ( Ba ) and ( Bb ) . Scale bars; 500 μm ( a , b ) , 50 μm ( c–h ) . ( C ) Immunohistochemistry of L1 ( c , d ) and NURR1 ( e , f ) in the cortex of wild-type ( a , c , e , g ) and Nova2-/- ( b , d , f , h ) at E16 . 5 . ( a , b ) Merged views of L1 ( green ) , NURR1 ( red ) , and DAPI ( blue ) . Scale bar; 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 02410 . 7554/eLife . 14371 . 025Figure 8 . Innervation defect in ventral diaphragm of Nova2-/- mice . Whole mount staining images of axons ( green ) and muscle ( white in A , blue in B ) in diaphragm of E18 . 5 ( A ) , E14 . 5 and E16 . 5 ( B ) wild-type and Nova1/2 mutants . No differences were observed between left and right diaphragm in each genotypes . Scale: 2 mm . D: Dorsal , L: left , R: right , V: ventral . ( C ) Quantification of innervation percentage of muscle in each quadrant of diaphragm that is covered by the phrenic nerve in E18 . 5 wild-type and Nova1/2 mutants . Using the phrenic nerve as reference , measurements of muscle length were taken from the insertion point of the phrenic nerve to the tip of the nerve ( X ) , and from the tip of the nerve to the end of hemidiaphragm ( Y ) , for the ventral and dorsal quadrants of left and right hemidiaphragms . The values indicate the ration of X/ ( X+Y ) , expressed as percentage ± standard deviation . **p<0 . 01 by t-test . Data are presented as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 025 We previously reported that NOVA1/NOVA2 regulate the alternative splicing of the agrin Z exons ( Ruggiu et al . , 2009 ) . Mice that lack both Nova family members fail to cluster AChR at neuromuscular junctions , including those of the diaphragm . Given the role for NOVA2 in axon guidance in the developing cortex ( Figure 6 , Figure 6—figure supplement 1 , and Figure 7 ) , we revisited the role of NOVA1 and NOVA2 on motoneuron axon guidance . The axons of peripheral motor neurons innervating the diaphragm muscle were visualized by immunostaining of neurofilaments at E18 . 5 ( Figure 8A ) and at E14 . 5 and E16 . 5 ( Figure 8B ) and the innervation percentage of muscle in each quadrant of diaphragm at E18 . 5 was quantified ( Figure 8C ) . Although the motor innervation patterns and percentages of the dorsal diaphragm were normal in Nova2-/- and Nova1/Nova2-double knockout ( dKO ) , those of the ventral diaphragm were severely abnormal with approximately 60–85% of the muscle uninnervated in either Nova2-/- and Nova1/Nova2 dKO mice ( p<0 . 001 ) . Similar abnormalities were observed in Nova2-/- and Nova1/Nova2 dKO mice at E14 . 5 and E16 . 5 ( Figure 8B ) . These innervation defects were specific to Nova2-deficient mice , as defects in ventral diaphragmatic innervation were never observed in wild-type or Nova1-/- mice . We cannot rule out a role of NOVA1 in the axon guidance of a subset of motoneurons given the slight decrease in ventral innervation in the Nova1/2 double-KO compared to the Nova2-/- mice . Nevertheless , taken together , these data demonstrate that in addition to what was observed in CC axons , Nova2 is also essential for phrenic nerve innervation of the ventral but not dorsal diaphragm . The inner ear is innervated by two types of fibers , afferent innervation consisting of both vestibular and cochlear ( spiral ganglion neurons; Mao et al . , 2014 ) and from efferent fibers derived from rhombomere 4 of the hindbrain ( Simmons et al . , 2011 ) . Efferents are guided by and extend along afferents to reach the sensory epithelia of the ear arriving at approximately the same time ( Fritzsch et al . , 1998; Ma et al . , 2000 ) . Efferents are a unique population of ventral brainstem neurons derived from facial motoneurons that project to the inner ear in contrast to facial muscle fibers or glands ( Karis et al . 2001 ) . Some transcription factors have been hypothesized to be relevant for guiding inner ear efferents ( Duncan and Fritzsch , 2013 ) , however , essentially nothing is known about how these neurons diverge from facial motoneurons to selectively reach the ear to innervate hair cells and afferent processes ( Simmons et al . , 2011 ) . Dye tracing analysis comparing with control mice at E14 . 5 ( Figure 9A , A’ ) showed that Nova2 ( Figure 9C , C’ ) but not Nova1 ( Figure 9B , B’ ) was essential for proper progression of efferent growth along afferents . Moreover , although efferent neurons expressed both NOVA1 ( Figure 9—figure supplement 1I–J ) and NOVA2 ( Figure 9—figure supplement 1F–H ) , Nova1 by itself was not capable of maintaining the function . In the absence of both Nova1/Nova2 , the efferents stalled when they reached vestibular ganglion neurons ( Figure 9D , D’ ) while afferent fibers that also expressed both NOVA1 and NOVA2 ( Figure 9—figure supplement 1A–E , K ) reached the cochlea normally . Our data suggested that Nova2 played a crucial role for efferent guidance relative to Nova1 , but both cooperate for normal efferent fiber extension . Interestingly , vestibular efferent innervation which segregated from cochlear efferents at E14 . 5 ( Bruce et al . , 1997 ) was unaffected by the absence of Nova1 or Nova2 alone ( see green efferents reaching the posterior canal crista ( PC ) in Figure 9B , C ) but was also completely stalled in Nova1/Nova2 double knockouts ( green label , Figure 9D ) . Postnatal Nova2-/- mice vestibular innervation was also normal ( not shown ) . 10 . 7554/eLife . 14371 . 026Figure 9 . NOVA2 expression is necessary for efferent innervation , targeting to the cochlea during embryonic development . ( A–D’ ) Olivocochlear and vestibular efferents were labeled with lipophilic dye application to the crossing bundle in rhombomere 4 ( green ) and afferent ( red ) by dye application to the cochlear nuclei . This view shows the left ear viewed from medial . Anterior ( a ) is to the left and dorsal ( d ) is up . ( A , A' ) wild-type mice spiral ganglion neurons ( SGN ) afferents reach the developing organ of Corti at E14 . 5 in the base and middle turn . Vestibular ganglia ( VGN ) project to the posterior canal crista ( PC ) and other vestibular sensory epithelia . ( B , B’ ) Dye labeling of efferent fibers to the ear is comparable to wild-type in Nova1-/-mice and ( C , C' ) shows reduced fiber growth in Nova2-/- mice ( D , D' ) . There is virtually no efferent fiber growth to the ear in Nova1-/- , Nova2-/- double knockout mice . Note that in D' the afferent signal was reduced to reveal how far efferents are of the target ( E–L ) . Efferent innervation ( green ) is shown in cochlea cryosections of Nova;YFPJ E18 . 5 mice . Nuclei are stained with DAPI ( blue ) and actin with Alexa-labeled Phalloidin ( magenta ) . Apical ( E–H ) and basal ( I–L ) turns are shown . ( E , I ) wild-type ( F , J ) and Nova1-/- mice show equivalent axonal innervation , ( G , K ) Nova2-/- mice have reduced innervation , ( K ) especially in the base . ( H , L ) Nova1-/- , Nova2-/- double knockout show no efferent innervation of the cochlea . Arrowhead shows approximate location of OHC rows ( white ) or IHC ( orange ) . Some auto-fluorescence is visible in L . Inset shows nuclear stained ( DAPI ) . Scale bars: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 02610 . 7554/eLife . 14371 . 027Figure 9—figure supplement 1 . NOVA1 and NOVA2 expression in spiral ganglion ( SG ) and in superior olive neurons but not in hair cells . ( A ) Sagittal sections of Rosa26 ( R26R ) ; Nova2-/- mice at six month of age . β-gal staining shows that the expression of Nova2 was switched on during development in the spiral ganglion ( SG ) were the soma of the afferent innervating neurons are located ( B ) The enlarged organ of Corti area ( dashed line in A ) shows no expression of Nova2 in the cochlea , including IHC ( red arrowhead ) and OHC ( green arrowhead ) . ( C ) Control Rosa26 ( R26R ) ; Nova2+/+ mouse shows no staining . ( D ) Immunostaining of P20 SG cells with an antiserum that recognizes NOVA1 and NOVA2 ( E ) showing the expression of NOVA1 in Nova2-/- mouse and demonstrating that NOVA1 and NOVA2 are expressed in SG . ( F ) In addition there is label in the soma of the medial olivococlear ( MOC ) neurons projecting efferent axons to the OHC which are located in the Medial superior olive ( MSO ) and lateral olivocochlear ( LOC ) neurons projecting axons the afferents below the IHC whose somas are located in the lateral superior olive ( LSO ) of the brainstem . ( G ) Enlarged area showing stained neuronal soma in MSO and LSO . ( G–I ) Immunostaining of a comparable section at P14 ( H ) wild-type section stained with C-16 a NOVA2 specific antibody . ( I–J ) Nova2-/- section stained simultaneously with C-16 antibody ( showing unspecific background ) and ( I ) with an antiserum that recognizes NOVA1 and NOVA2 showing the expression of NOVA1 in the superior olive . ( K1–K3 ) Total RNA was extracted from micro dissected organ of Corti ( dashed line rectangle ) , analysed by qPCR and normalized by actin . The plots show Nova1 and Nova2 gene expression relative to the expression of Nova1 in control P7 mouse at ( K1 ) P7 and ( K2 ) P14 . ( K3 ) the expression of the hair cell marker Myosin 7a ( Myo7a ) is shown at the indicated time points as a control . The reduced expression at P18 reflects problems in the dissection due to increased calcification . There is no apparent problem at P14 , suggesting that the reduction in NOVA1 and NOVA2 expression at P14 is a biological phenomenon that probably reflects pruning of innervation . V: motor trigeminal nucleus , VII: Facial Nucleus . Scale bars; 200 μm ( A ) , 50 μm ( B ) 10 μm ( H–J ) . *p<0 . 05 , **p<0 . 01 ( n = 3 , t test ) . Data are presented as mean ± SE . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 027 The cochlea has a tonotopic map where the sensory hair cells at each position along the organ are most sensitive to a particular frequency , forming a continuous gradient from high frequency in the base to low frequency in the apex ( Howard et al . , 1988; Hudspeth , 1989; Liberman , 1982 ) . Dye tracing analysis showed that the loss of efferent innervation was frequency specific; it was almost completely lost in the base of the Nova2-/- cochlea and only partially lost towards the apex ( compare Figure 9A’ and C’ ) . A more detailed analysis was possible at E18 . 5 ( Figure 9E–L ) , when the medial olivocochlear ( MOC ) efferent axons expand through the tunnel of Corti , prior to their reaching the outer hair cells ( OHC ) at P2 and establishing functional synapses at P7 ( Simmons et al . , 2011 ) . These axons stained strongly with the Thy-1 , YFPJ line that also has high expression in motor neurons , demonstrating the motor origin of auditory efferents . Thy-1 positive axons reached the inner hair cells ( IHC ) at E18 . 5 in control ( Figure 9E , I ) and Nova1-/- ( Figure 9F , J ) mice but this innervation was reduced in the Nova2-/- ( Figure 9G , K ) mice , and this was especially evident in the base ( Figure 9K ) . In the absence of both Nova1/Nova2 there were no labeled axons reaching the cochlea ( Figure 9H , L ) , showing that efferents remain stalled . Imaging of efferent innervation at the onset of hearing ( P14 ) showed a clear reduction in axons reaching IHC and specially OHC in the base when comparing Nova2-/- mice ( Figure 10C , D , G , H ) to controls ( Figure 10A , B , E , F ) . At this age we also found a reduction in efferent synapse formation as demonstrated by reduced labeling with the pre-synaptic marker synaptophysin ( blue label at the base of IHC and OHC ) . Whole mount apical cochlea ( Segment 2 , see map in Figure 10O ) at P10 showed a reduction in axons reaching IHC and an almost complete absence of axons reaching OHC ( Figure 10K , L ) compared to controls ( Figure 10I , J ) . 10 . 7554/eLife . 14371 . 028Figure 10 . Efferents innervation defect and hearing impairment in postnatal Nova2-/- mice . ( A–H ) Cochlea cryosections immunostaining of in Nova2-/-;YFPJand control Nova2+/+;YFPJ P14 mice . The images show reduced efferent axons in green ( YFP , see arrows ) , afferent neurofilaments ( NF200 ) in red and phalloidin ( Phall ) labeling of actin-rich hair bundles in magenta . The presynaptic marker synaptophysin ( syn ) in blue , shows a reduction of functional efferent innervation . ( A–F ) control mouse , ( C–H ) Nova2-/-;YFPJ mouse . ( A–D ) Apical or ( E-H ) Basal organ of Corti . ( I–L ) Cochlea apical turn segment 2 ( see below ) whole mount preparations immunostaining of ( K , L ) Nova2-/-;YFPJ and ( I , J ) Nova2+/+;YFPJ P10 mice . The images show reduced efferent axons in green ( YFP ) , NF200 in red and Phall in blue . ( A–L ) Arrowheads show approximate location of OHC rows ( white ) or IHC rows ( orange ) . ( O ) Representation of the dissection map of the cochlea indicating segments 1–5 and the approximate localization along the length of the cochlea . ( M ) The thresholds of the auditory brainstem response ( ABR ) to pure-tone stimuli ranging from 10 to 80 dB SPL are increased significantly in Nova2-/- mice ( n = 4–5 for each frequency ) compared to wild-type mice ( n = 5–6 ) at P21-P22 . ( N ) The distortion-product otoacoustic emissions ( DPOAEs ) at 2f1-f2 measured at 55 , 65 and 75 dB SPL show significant differences between wild-type mice ( n = 4 ) and Nova2-/- animals ( n = 4 ) at P21-P22 . The noise floor ( NF ) that was measured simultaneously is also shown for both groups . The approximate acoustic representation of segments 2–4 of the dissected cochlea are shown over each plot and segment 2 is overshadowed in gray . Scale bars: 50 μm . *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 02810 . 7554/eLife . 14371 . 029Figure 10—figure supplement 1 . Reduced efferent innervation and increased afferent innervation to the apex of the Nova2-/- mice cochlea . ( A–C ) Quantitation of efferent neuronal axons by colorimetric detection of the enzyme acetyl cholinesterase ( AChE ) in the cochlear apex of ( B , B’ ) Nova2-/- and control ( A , A’ ) wild-type mice . Representative images are shown . Green arrows indicate the 3 rows of OHC and red Arrow the row of IHC . These preparations were done in non-decalcified cochlea and only the apex could be preserved after dissection . ( B ) Plot showing quantification of the colorimetric signal . Heterozygote Nova2+/- mice were included in the plot to show that defect is dependent on the dose of NOVA2 expressed ( n = 4 ) . P15-P25 mice were used **p<0 . 01 . ( D , E ) Cochlea whole mount and ( D’ , E’ ) Cryosections immunostaining in wild-type and Nova2-/- P20 mice . The images show apparent increased afferent innervation staining ( NF200 , red ) . Hair cells are labeled with a MYOVI antibody ( green ) and hair bundles with Phall . ( magenta ) . ( F–I ) Afferent innervation to the IHC was quantified by counting individual ribbon synapses stained with CTBP2 antibody at P18 . ( F–H ) wild-type ( G–I ) Nova2-/- , ( F , G ) segment 2 , apex and ( H , I ) segment 4 , apex ( see Figure 10-O ) . ( F’–I’ ) Respective amplified afferent IHC synapses . ( J ) A total of 30 wild-type and 44 Nova2-/- cells from 2 mice of each genotype were quantified . Each dot in the plot represents 2–5 averaged cells . The figure shows the approximate location of cells from segment 1–4 along the% length of the cochlea from apex to base on the lower x axes and the approximate frequency representation in the upper x axes . The number of dots ( synapses ) per cell was quantified by counterstaining with MYOVI which allows to outline the IHC perimeter . **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 029 These defects were also evident in apical cholinergic innervation ( Dalian et al . , 2001; Maison et al . , 2003 ) after the onset of hearing ( P15-P25; Figure 10—figure supplement 1A–C ) . At these stages , we were able to demonstrate that the loss of axonal innervation was functionally significant , as assessed by auditory brainstem responses ( ABRs ) and distortion products otoacustic emissions ( DPOAEs ) recordings ( Figure 10M–N ) In addition , an apparent increase in afferent innervation was found in Nova2-/- mice by neurofilament immunodetection ( Figure 10A–L and Figure 10—figure supplement 1D–E ) . The difference was significant in the apex as determined by quantitation of IHC ribbon synapses ( Meyer et al . , 2009 ) ( Figure 10—figure supplement 1F–J ) . Taken together , these data demonstrate a role for NOVA in cochlear innervation , efferent pathfinding , and normal hearing function in vivo .
In the present work , we provide evidence for a novel and unique role for the tissue specific splicing regulator NOVA2 as an axon pathfinding modifier in cortical CC axons , motoneuron , and auditory efferents . In prior studies , integration of transcriptome-wide HITS-CLIP analysis , exon junction microarray data sets , and bioinformatics led to the identification of a large set of NOVA1/2 RNA functional interaction sites and major biological pathways which NOVA1/2 orchestrates in vivo ( Licatalosi et al . , 2008; Ule et al . , 2005b; Ule and Darnell , 2006; Zhang et al . , 2010 ) . Given identical binding characteristics in shared transcripts ( e . g . Agrin , others in Figure 2 ) , we infer the following: 1 ) specificity is determined primarily by biochemical interactions: kD of NOVA KH domains binding to accessible YCAY motifs; this might include differential levels of NOVA1/NOVA2 expression within a cell; 2 ) differential binding is a function of differences in cell expression and/or differences in cell biology ( e . g . subcellular distribution of NOVA1 and NOVA2 ) . Evidence for differences in cell expression is clear in the cortex . Differences in actions within cells such as motoneurons that express both NOVA1 and NOVA2 isoforms could arise from different levels of NOVA1 and NOVA2 within the neurons , or due to differential localization of NOVA1 and NOVA2 within the motoneuron , as suggested by the differences in CLIP binding ( intron versus 3’ UTR ) of the two proteins ( Figure 2F ) . The development of single cell-type CLIP would be able to address this issue . Prior genomic and biochemical studies have found no actions that were unique to either NOVA1 or NOVA2 paralogues . The present study identifies NOVA2 as playing a unique role in axonal pathfinding in embryonic development by combining histological analysis with genome wide NOVA1 and NOVA2 specific HITS-CLIP and RNA-seq analysis in Nova1-/- and Nova2-/- mice . These data build upon previously identified roles for NOVA1 and NOVA2 in early development as modifiers of neuronal migration in late-generated cortical neurons and Purkinje neurons through control of splicing of Dab1 exon 7b/7c ( Yano et al . , 2010 ) , for Nova1/Nova2 as essential factors for AChR clustering at the NMJ through the regulation of Agrin Z exons ( Ruggiu et al . , 2009 ) , and for Nova2 in the induction of long term potentiation of slow inhibitory , but not excitatory , post synaptic potential currents after coincidence detection in CA1 hippocampal neurons ( Huang et al . , 2005 ) . The current study provides the first evidence that the formation of corpus callosum in brain , proper innervation of ventral diaphragm , and efferent innervation to the cochlea at developing embryonic stage all require specific actions mediated by NOVA2 ( Figure 11 ) . 10 . 7554/eLife . 14371 . 030Figure 11 . Model summarizing a unique role for NOVA2 in the consequence to neuronal axon guidance and outgrowth . The results of HITS-CLIP and RNA-seq analysis combined with histological analysis in Nova2-/- mice suggest a model that NOVA2 starts to act on an axon guidance regulatory cascade during neural differentiation in the neurons located on cortical plate . The breakdown of NOVA2-mediated RNA regulation results in the collapse of the normal axon guidance/outgrowth properties in either/both the neuron extending axon or/and the neuron expressing axon guidance cue . DOI: http://dx . doi . org/10 . 7554/eLife . 14371 . 030 Immunohistochemical and DiI tracer analysis showed that Nova2 deficiency results in severe ACC ( Figure 6 , and Figure 6—figure supplement 1 ) . There are over 50 mouse genes known to be required for the formation of the corpus callosum ( reviewed in Paul et al . , 2007; Edwards et al . , 2014 ) . NOVA2 HITS-CLIP and RNA-seq analysis revealed that 11 NOVA2 target alternative splicing events on 9 genes ( Dcc , Robo2 , Epha5 , Slit2 , Ank2 , Dclk1 , Enah , Cask , Ptprs ) intersected with the genes associated with mouse ACC phenotypes and human ACC syndromes ( Figure 6—figure supplement 3 ) , suggesting that the ACC phenotype in Nova2-/- mice may be caused by splicing dysregulation of multiple coordinately NOVA2-regulated transcripts , consistent with previous observations that NOVA proteins regulate subsets of transcripts that mediate coordinate biologic functions ( Ule et al . , 2005b ) . Given that Netrin-1 null mice and Dcc null mice shows similar ACC phenotype to Nova2 null mice ( Fazeli et al . , 1997; Serafini et al . , 1996 ) , we tried to rescue the ACC phenotype with the DCC-long isoform , that is decreased in the cortex of Nova2-/- mice , by in utero electroporation ( data not shown ) . The expression of DCC-long ( long isoform of exon 17 ) isoform in cingulate cortex at E14 . 5 was not sufficient to form the commissure corpus callosum , indicating that the DCC-long isoform was not sufficient to rescue the ACC phenotype , and suggesting the possibility that the ACC phenotype in Nova2-/- mice either does not involve the DCC-long isoform or is caused by the dysregulation of multiple transcriptomes . We found that in the cortex , NOVA2 regulates alternative splicing events of two netrin receptors: Dcc exon 17 and Neo1 exon 27 in developmentally regulated manner ( Figure 4 , Figure 4—figure supplement 1 , and Figure 5 ) . In the spinal cord , Dcc exon 17 is regulated by NOVA1/2 and required for the spinal commissural neuron development ( see accompanying paper by Leggere et al . ) . Recent structure analysis of netrin-1/NEO1 and netrin-1/DCC-short ( short isoform of exon 17 ) revealed that netrin-1/NEO1 form a 2:2 heterotetrameric netrin-1/NEO1 complex and netrin-1/DCC-short have a continuous netrin-1/DCC assembly ( Xu et al . , 2014 ) . Netrin-1 recognizes the fourth and fifth fibronectin type III domains ( termed FN4 and FN5 , respectively ) of both NEO1 and DCC . Interestingly , the NOVA2-regulated Dcc exon 17 is located in the linker region between the FN4 and FN5 domains , and this splicing switch from DCC-short to DCC-long isoform is enough to support the architectural switch of netrin-1/DCC complex from the continuous assembly to the 2:2 assembly ( Xu et al . , 2014 ) . This suggests that NOVA2-dependent splicing regulation results in architecturally distinct netrin-1/DCC complexes that elicit distinct downstream signaling . Changes in NOVA2-dependent regulation of Dcc exon 17 splicing during development ( Figure 5 ) further indicate that NOVA2-regulated splicing controls the ability of cortical and spinal neurons to utilize distinct netrin-1/DCC assemblies to respond to a variety of cellular demands . NEO1 , another member of the Dcc family , shares ~50% amino acid identity with DCC ( Vielmetter et al . , 1994 ) and has four known alternative splicing sites . The NEO1 extracellular region is a receptor for repulsive guidance molecule ( RGM ) family molecules in addition to netrins . Upon ligand binding to the extracellular domain of NEO1 and DCC , their intracellular domains can activate multiple downstream signal transduction pathways , that modulate chemotropic axon guidance ( Bradford et al . , 2009 ) . The inclusion of Neo1 exon 27 inserts a 53 amino acid sequence into the intracellular region . Inclusion is increased in E18 . 5 Nova2-/- cortex , suggesting that NOVA2-dependent alternative splicing of Neo1 exon 27 regulates downstream signal transduction . It is well-established that DCC receptors mediate chemo-attraction upon netrin-1 binding , while the UNC5 receptor alone or heterodimerized with DCC elicits chemo-repulsive activity in neuronal tissues ( Hong et al . , 1999 ) . Interestingly , UNC5B interacts with NEO1 as a co-receptor for RGMa and associates with leukemia-associated guanine nucleotide exchange factor ( LARG , also termed as ARHGEF12 ) to mediate the signal transduction leading to growth cone collapse ( Hata et al . , 2009 ) . Although the abundance of Arhgef12 RNA in cortex was comparable between E18 . 5 wild-type and Nova2-/- mice ( log2FC = 0 . 10 , FDR = 0 . 78 ) , Arhgef12 exon 4 inclusion was significantly increased in Nova2-/- mice ( Figure 4—figure supplement 1 ) , suggesting the possibility that NOVA2 modulates RGM signal transduction mediated by NEO1 , UNC5b , and ARHGEF12 through coordinate regulation of alternative splicing of those RNAs . Interestingly , exon 7–8 deletion of Srrm4/nSR100 gene , an alternative splicing regulator , mice shows partial ACC phenotype and neurite outgrowth defect in diaphragm ( Quesnel-Vallieres et al . , 2015 ) in addition to hearing defects ( see below ) , indicating that alternative splicing regulation mediated by RBPs ( e . g . NOVA2 , SRRM4/nSR100 ) is required for the formation of corpus callosum and neurite outgrowth into diaphragm . Our analyses were focused on axon guidance related genes but do not rule out the possibilities that axonal pathfinding defect phenotypes in Nova2-/- were caused by deficits of multiple biological processes . Several relevant processes were enriched in GO and KEGG pathway analysis , including protein modification , phosphorylation , cellular protein metabolic process , synaptic transmission , chromatin modification , adherens junction , and cell morphogenesis involved in neuron differentiation . Our previous work has showed that the proliferation of cortical neural progenitor cells was not disrupted in Nova2-/- ( Yano et al . , 2010 ) . RNA-seq data showed that the transcriptome abundance of the neuronal differentiation markers were comparative between wild-type and Nova2-/- mice ( Figure 3—source data 2 ) , indicating that the differentiation from neuronal progenitor to neuron is comparable between wild-type and Nova2-/- mice . Despite the important role of alternative splicing in the formation of the cochlear tonotopic gradient ( Navaratnam et al . , 1997; Rosenblatt et al . , 1997; Miranda-Rottmann et al . , 2010 ) and in hair cell development and function ( Kollmar et al . , 1997; Liu et al . , 2007; Webb et al . , 2011 ) to date only two alternative splicing factors have been implicated in hearing: Srrm4/nSR100 ( Nakano et al . , 2012 ) and Sfswap ( Moayedi et al . , 2014 ) . Both are expressed in hair cells necessary for the development of the sensory epithelium , in contrast with Nova1 and Nova2 , which are the first neuron-specific splicing factors described to regulate cochlear innervation . Interestingly auditory efferents are a ventral population of rhombomere 4 facial brachial motoneurons that segregate during development ( Karis et al . , 2001 ) and while facial innervation is normal , the ventral auditory efferents are selectively affected in the Nova2-/- mice ( Figure 9A–D and data not shown ) . This specificity is similar to the observation that dorsal motoneuron innervation of the diaphragm is normal while ventral innervation ( Figure 8 ) is affected in the Nova2-/- mouse , suggesting a more general ventral specialization of axonal guidance regulation by NOVA2 . These observations are reminiscent of those seen with defects in GATA-3 innervation of facial motor neurons ( Karis et al . , 2001 ) and dorsal/ventral pathfinding of limb motoneurons regulated by EphA4 ( Helmbacher et al . , 2000 ) , suggesting the possibility of an integrated role for NOVA2 in dorsal/ventral axonal pathfinding . To assess OHC function we measured DPOAEs ( Lonsbury-Martin and Martin 1990 ) in Nova2-/- mice . There was a significant reduction in DPOAEs in the apex of Nova2-/- mice cochlea ( Up to 10 dB SPL , Figure 10N ) where a significant reduction in cholinergic innervation was also found ( Figure 10—figure supplement 1A–C ) . Nevertheless the reduction in efferent innervation is comparatively more severe in the base of the cochlea ( Figure 9 and 10A–H ) suggesting that OHCs sensitive to low frequency sounds are more dependent on Nova2 controlled efferent innervation . Direct testing of efferent function by electrical activation of these nerves has been shown to reduce the baseline DPOAE intensity ( Liberman et al . , 1996; Vetter et al . , 1999 ) . Preliminary testing in Nova2-/- mice showed that this inhibitory function was not affected ( Stéphane S Maison , personal communication ) , suggesting that the residual efferent innervation was able to sustain the function of further reducing the DPOAEs . Only an indirect mouse model of efferent innervation defect has been available to date: a mutation in the α9α10 hair cell nicotinic receptor ( Taranda et al . , 2009 ) in which the DPOAEs threshold levels are reduced by approximately 7–14 dB SPL ( a reduction in 10 dB SPL is perceived approximately as half the intensity ) . In addition , the reduction in DPOAEs found in the Nova2-/- mice could explain the ABR threshold elevation at the same frequency range ( Figure 10M ) ( Sun and Kim 1999; Taranda et al . , 2009 ) . The opposite possibility , that the defect evidenced by ABRs is responsible for the reduction in DPOAEs is unlikely since it has been reported that even with 80% loss of IHC ( with intact OHC ) the DPOAEs are only reduced by 10–20 dB SPL ( Trautwein et al . , 1996 ) . It has also been recently shown that afferent innervation to the OHC has no effect on DPOAEs ( Froud et al . , 2015 ) . Despite the reduction in ABRs which measure afferent pathway function , we found an increase in afferent innervation ( Figure 10A–L ) quantified by synaptic ribbon count in Nova2-/- mice ( Figure 10—figure supplement 1E–J ) although further study will be necessary to determine if this synapses are functional . Increase in afferent innervation could be a compensatory mechanism to decreased lateral olivocochlear ( LOC ) efferents , which form synapses on IHC afferents , or a direct effect of the loss of Nova2 expression in afferent neurons through an ephrin-dependent pathway ( Defourny et al . , 2013 ) . An interesting possibility is that the transient synapses formed between MOC efferents and IHC during development ( Simmons et al . , 2011 ) have an important role in IHC development that is impaired in Nova2-/- mice . Taken together , our observations suggest that protein/RNA diversity provided by NOVA2-mediated RNA regulation is required for proper axon pathfinding and formation of complex synapses/neural networks , particularly in dorsal/ventral choices , and that alternative splicing switches mediated by NOVA2 may regulate key developmental steps in mammalian biology and pathogenesis of neurological diseases .
Two fragments from the Nova2 genomic locus of 2 . 2 and 6 kb , flanking a 1 . 5 kb DNA fragment harboring the exon containing the initiating methionine , were cloned into a targeting vector . An IRES-Cre-FRT-Neo-FRT targeting cassette was inserted into the Nova2 locus upstream of the initiator ATG in the first known coding exon of the Nova2 gene . Linearized targeting vector plasmid was electroporated into ES , and G418-resistant clones ( 150 ug/mL G418 for 24 hr ) harboring homologous recombinants were screened by Southern blot with a 5’ Nova2 genomic probe and injected into mouse blastocysts to produce germline transmitted chimeras . The neomycin cassette was excised by breeding germline-transformed mice with transgenic animals expressing Flp recombinase under the control of the CMV promoter . The original Nova2+/- ( and Nova1+/- ) mice obtained in a B57B/6 background were backcrossed for 10 generations to CD-1 and FVB strains and to the thy1-YFP line J ( YFPJ ) strain ( Jackson labs Stock 003709 ) which has high expression of YFP in motor neurons ( Feng et al . , 2000 ) or crossed to the R26R strain ( Jackson lab stock 003309 ) for Nova2 tissue-wide expression analysis . Most experiments were done in CD-1 background , but mice in the inbred FVB background showed no phenotypic differences . FVB mice were used for the hearing tests reported here because they present normal hearing thresholds ( Zheng et al . , 1999 ) and less inter individual differences . All procedures in mice were performed in compliance with protocols approved by the Institutional Animal Care and Use Committee ( IACUC ) of the Rockefeller University or the Comité de déontologie de l’expérimentation sur les animaux ( CDEA ) of the Univeristy of Montreal . Primary antibodies used for immunohistochemistory and western blot were as follows; goat anti-NOVA2 ( C-16 ) ( sc-10546 , Santa Cruz ) , rabbit anti-NOVA1 [EPR13847] ( ab183024 , abcam ) , human anti-pan NOVA ( anti-Nova paraneoplastic human serum ) , rabbit anti-PTBP2 ( Polydorides et al . , 2000 ) , rat anti-L1 ( MAB5272 , Millipore ) , goat anti-TAG1/Contactin-2 ( AF4439 , R&D systems ) , goat anti-NetrinG1a ( AF1166 , R&D systems ) , rabbit anti-NURR1 ( M-196 ) ( sc-5568 , Santa Cruz ) , rabbit anti-neurofilament ( AB1981 , Chemicon ) , mouse anti-NF200 ( N5389 , SIGMA ) , mouse anti-ctbp2 ( 612044 , BD ) , rabbit anti-myoVI ( sc-50461 , Santa Cruz ) , and goat anti-neuropilin-1 ( AF566 , R&D systems ) . Anti-NOVA1 and anit-NOVA2 antibody specificity for immunohistochemistory and immunoprecipitation was confirmed ( Figure 1—figure supplement 2 ) . E16 . 5 , E18 . 5 mice brains were fixed with 4% praraformaldehyde ( PFA ) /PBS at 4 degrees overnight , and sequentially replaced to 15% sucrose/PBS and 30% sucrose/PBS at 4 degrees for cryo-protection , then embedded in OCT compound . Frozen brains were sliced into 80 μm thick sections on a cryostat ( CM3050S , LEICA ) . Brain sections and whole embryos fixed with 4% PFA/PBS were subjected to floating or whole-mount immunohistochemistry . These samples were washed three times with PBS at room temperature ( RT ) , incubated with 0 . 2% Triton X-100/PBS at R . T . , blocked with 1 . 5% normal donkey serum ( NDS ) /PBS at RT , and then incubated overnight at 4 degrees with primary antibodies or Alexa-dye conjugated phalloidin in 1 . 5% NDS/PBS followed by incubation with Alexa 488 , 555 , or 647 conjugated donkey secondary antibodies ( 1:1000 ) in 1 . 5% NDS/PBS . Images of immunostained specimens were collected by BZ-X700 ( KEYENCE ) , fluoscence microscopes Axioplan 2 ( Zeiss ) , or an inverted TCS SP8 laser scanning confocal microscope ( LEICA ) at The Rockefeller University Bio-Imaging Resource Center . Embryos were obtained by C-section of time-mated dams and fixed by overnight immersion in 4%PFA . Postnatal mice were intracardialy perfused with 4% PFA , cochleae immediately removed and perfused through the round window with a 1 ml syringe . After post-fixation 2:30 hr at 4°C cochleae were washed in PBS , directly dehydrated in 30% sucrose at 4°C or previously decalcified for 24–48 hr in 120 mM EDTA in PBS . For beta-galactosidase expression analysis the decalcified cochleae were perfused through the round window with M-1 embedding matrix ( Thermo Scientific ) and frozen in a block of the same matrix , 40 µm cryosections were incubated in staining solution ( 4 mM potassium ferricyanide , 4 mM potassium ferrocyanide , 1 mg/ml x-gal ) dissolved in wash buffer ( 2 mM magnesium chloride , 0 , 01% sodium deoxycholate , 0 , 1% NP40 and 0 . 2 mM buffer phosphate pH 8 . 0 ) overnight at 37°C , washed in wash buffer 3 hr at 4°C , post-fixed again in 4% PFA and washed in PBS before mounting . For AChE staining the non-decalcified cochleae were dissected by carefully chipping the bone away and incubated in a solution containing 0 . 5 mg/ml acetylcholine iodide as explained elsewhere ( Willott , 2001 ) . For whole mount immunostainings the decalcified cochleae were dissected cutting into half turns ( See Figure 10O ) , permeabilized and blocked in 0 . 1% TX100 , 5% donkey serum in PBS . Primary and secondary antibodies were incubated sequentially 24 hr at 4°C . Washes were done in 0 . 05% TX100 . For cryosections the decalcified cochleae perfused through the round window with M-1 embedding matrix ( Thermo Scientific ) or the fixed embryos were cut into 14–20 µm sections , permeabilized and blocked in 0 . 1% Triton-X100 , 5% donkey serum in PBS . Primary antibodies were incubated 24 hr at 4°C and secondary antibodies for 2 hr at room temperature . Washes were done in PBS . Alexa labeled Phalloidin ( 1:100 ) was added to the secondary antibody mix . Images were collected using a Zeiss LSM 510 confocal microscope at The Rockefeller University Bio-Imaging Resource Center and a Zeiss LSM 700 confocal microscope at the IRIC core of the University of Montreal . For DiI trace analysis , DiIC18 ( 3 ) crystal ( D3911 , molecular probes ) were placed in cingulate cortex or lateral neocortex of P0 brain . After 8–21 days at 37 degrees in 4% PFA/PBS to allow dye diffusion , samples were cryo-protected with 30% sucrose/PBS at 4 degrees , embedded in OCT compound , and cut into 100 μm sections on cryostat . Counterstaining was with 1 μg/mL DAPI ( 4’ , 6-Diamidino-2-phenylindole dihydrochloride ) . The images of immunostained specimens were collected by a fluoscence microscope Axioplan 2 ( Zeiss ) . Inner ear afferents and efferents were labeled with lipophylic ( NeuroVue ) dyes ( Molecular Targeting Technologies; MTTI ) . These dyes were placed into specific areas for selective labeling of nerve fibers ( Fritzsch et al . , 2005; Duncan et al . , 2015 ) Dyes were placed into the cerebellum and the alar plate of rhombomere 2 to label the afferent fibers to the ear . Dye was also placed into the basal plate of rhombomere 4 for efferent nerve labeling . The preparations were then incubated in 4% PFA at 36 degrees C for three days . After dye diffusion the ear was dissected out and placed on a glass slide with glycerol and a cover slip on top to be imaged with a Leica SP5 confocal microscope . Western blotting and RT-PCR analyses were performed as described elsewhere ( Yano et al . , 2010 ) . For quantification , three biological replicates were used and quantified from obtained images by using ImageJ free software ( http://rsb . info . nih . gov/ij/index . html ) . ABRs were recorded from deeply anesthetized mice . The positive needle electrode was inserted subdermally at the vertex , the negative electrode was placed beneath the pinna of the left ear , and the ground electrode was located on the hind leg . ABRs were evoked by tone bursts ranging from 4–64 kHz and were produced by a closed-field electrostatic speaker connected to a driver ( EC‑1 and ED‑1; Tucker-Davis Technologies ) . The 5 ms tone bursts were presented 33 . 3 times per second; their 0 . 5 ms onsets and offsets were tapered with a squared cosine function . The speaker’s audio output was transmitted into the ear through a custom acoustic assembly . Sound-pressure levels were measured with a calibrated microphone and preamplifier connected to a conditioning amplifier ( 4939‑A‑011 and 2690‑A‑0S1 , Brüel and Kjær ) . The response was amplified x10 , 000 and bandpass filtered at 0 . 3‑3 kHz ( P55 , Natus Neurology Inc . ) . The amplified response was then digitally sampled at 10 μs intervals with a data acquisition device ( PCIe-6353 , National Instruments ) controlled by custom software ( LabVIEW 2010 , National Instruments ) . The responses to 1000 bursts were averaged at each intensity level to determine the threshold , defined as the lowest level at which a response peak was distinctly and reproducibly present . For each frequency , sound-pressure level was decreased from 80 dB SPL in 5 dB steps until threshold was reached . DPOAEs were then elicited with an acoustic coupler that allowed for recording ear-canal sound pressure levels via a probe tube concentrically situated within the common sound-delivery tube . The acoustic coupler consisted of two electrostatic speakers ( EC‑1 , Tucker-Davis Technologies ) to generate primary tones and a miniature microphone ( EK‑23103 , Knowles ) to measure ear-canal sound pressure ( coupler from Mike Ravicz , EPL , Eye&Ear inf , Boston ) . The speakers and the microphone were both calibrated using the calibrated microphone described above . The 2f1 − f2 distortion product was measured with f2 = 6–54 kHz , f2/f1 = 1 . 2 , and L1 = L2 = 55‑75 dB SPL . The acoustic signal was amplified by a preamplifier ( ER‑10B+ , Etymotic Research ) and the sound pressure measured in the ear canal was digitally sampled at 10 μs intervals with the data-acquisition system described above . Each frequency pair was presented for 1 s . After fast Fourier transforms had been computed and averaged over ten consecutive traces , the amplitudes of the 2f1 − f2 distortion product and the surrounding noise floor ( +/- 100 Hz of 2f1 − f2 ) for each frequency pair were determined , a procedure requiring 17 s of data acquisition and processing time . NOVA1 and NOVA2 CLIP was performed on E18 . 5 wild-type cortex using three biological replicates , as described elsewhere ( Licatalosi et al . , 2008 ) . High-throughput sequencing was performed at the Rockefeller University Genome Resource Center . Sequence tags were aligned to the mouse genome ( mm9 ) by novoalign . Unique tags were collected by eliminating PCR duplicates . E18 . 5 mouse cortex RNA from wild-type and Nova2-/- littermate and E18 . 5 mouse cortex and midbrain/hindbrain RNA from wild-type and Nova1-/- littermate ( three biological replicates for each genotype ) was prepared using Trizol ( ambion ) and ribosomal RNA was removed from 1 μg RNA using Ribo-Zero rRNA removal Kit ( epicentre ) . Standard RNA-seq libraries were prepared using TruSeq RNA Sample Preparation Kit v2 ( illumina ) following manufacturer’s instructions . High-throughput sequencing was performed on Hi-seq 2500 ( illumina ) to obtain 125 nucleotide paired-end reads at New York Genome Center . Reads were aligned to the mouse genome ( mm10 ) using OLego ( Wu et al . , 2013 ) . The downstream splicing analysis was performed as described previously ( Yan et al . , 2015 ) and then liftovered to mm9 . Differential expression analysis was performed with the R package edgeR ( Robinson et al . , 2009 ) available on Bioconductor 3 . 0 ( Gentleman et al . , 2004 ) . RNA-seq and HITS-CLIP data have been deposited in GEO under accession number GSE69711 ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE69711 ) . | The first step of producing a protein involves the DNA of a gene being copied to form a molecule of RNA . This RNA molecule can often be processed to create several different “messenger” RNAs ( mRNAs ) , each of which are used to produce a different protein by a process known as alternative splicing . A class of proteins that bind to RNA molecules controls alternative splicing . These “splicing factors” ensure that the right protein variant is produced at the right time and in the right place to carry out the appropriate activity . Many genes that play important roles in the nervous system have been reported to undergo alternative splicing to generate different protein variants . However , it is unclear whether alternative splicing is important for controlling how the nervous system develops , during which time the neurons connect to the cells that they will communicate with . Forming these connections involves part of the neuron , called the axon , growing along a precise path through the nervous system to reach its destination . Two RNA-binding proteins called NOVA1 and NOVA2 are produced exclusively in the central nervous system , where they regulate a number of actions including alternative splicing . So far , no differences in the roles of NOVA1 and NOVA2 have been identified , and relatively little is known about their actions in the brain . Saito et al . have addressed these missing puzzle pieces by combining RNA analysis methods with an analysis of the structure of the nervous system of mice that lack either NOVA1 or NOVA2 . This approach identified where NOVA1 and NOVA2 bind on mRNAs , and showed that the mRNAs are processed in different ways in the developing mouse brain depending on which form of the NOVA protein is bound to it . Further analysis of the data revealed that NOVA2 , and not NOVA1 , regulates splicing in a series of RNA molecules that help to guide axons to the correct locations in the developing mouse brain . A related study by Leggere et al . also reported on the role that NOVA proteins play in the alternative splicing of one of these genes , called Dcc . Saito et al . also found defects in the nervous systems of the mice that lacked NOVA2 that only occurred in these mice and resulted from certain axons being unable to follow the correct path to their target cells . These led to major defects , such as agenesis of the corpus callosum ( a complete lack of connection between the right and left sides of the brain ) . Further defects affected how specific subsets of motor neurons connect to muscles and how cochlear neurons in the brainstem connect to the inner ear . The next steps are to explore how the processing of RNA molecules by NOVA2 causes these defects , and to assess whether these actions relate to developmental brain disorders in humans . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | NOVA2-mediated RNA regulation is required for axonal pathfinding during development |
Numerous genetic variants associated with MEF2C are linked to autism , intellectual disability ( ID ) and schizophrenia ( SCZ ) – a heterogeneous collection of neurodevelopmental disorders with unclear pathophysiology . MEF2C is highly expressed in developing cortical excitatory neurons , but its role in their development remains unclear . We show here that conditional embryonic deletion of Mef2c in cortical and hippocampal excitatory neurons ( Emx1-lineage ) produces a dramatic reduction in cortical network activity in vivo , due in part to a dramatic increase in inhibitory and a decrease in excitatory synaptic transmission . In addition , we find that MEF2C regulates E/I synapse density predominantly as a cell-autonomous , transcriptional repressor . Analysis of differential gene expression in Mef2c mutant cortex identified a significant overlap with numerous synapse- and autism-linked genes , and the Mef2c mutant mice displayed numerous behaviors reminiscent of autism , ID and SCZ , suggesting that perturbing MEF2C function in neocortex can produce autistic- and ID-like behaviors in mice .
An imbalance of excitatory and inhibitory synaptic transmission in the brain is an emerging theory of the pathophysiology of multiple neurodevelopmental and neuropsychiatric disorders ( Garber , 2007; Zoghbi , 2003 ) , including autism and SCZ . However , the genes and molecules that regulate the number of excitatory and inhibitory synapses formed and maintained on neurons remain poorly understood . The MEF2 transcription factor genes are expressed in both excitatory and inhibitory neurons throughout development and adulthood in overlapping , but unique , expression patterns ( Lyons et al . , 2012; Shalizi and Bonni , 2005; McKinsey et al . , 2002 ) , and they have been shown to regulate excitatory synapse density on multiple neuron types ( Flavell et al . , 2006; Li et al . , 2008; Barbosa et al . , 2008; Pulipparacharuvil et al . , 2008 ) . For example , MEF2A and MEF2D can regulate activity-dependent elimination of glutamatergic synapses on both hippocampal pyramidal neurons and medium spiny neurons of the striatum in a cell-autonomous manner ( Flavell et al . , 2006; Pulipparacharuvil et al . , 2008 ) . Expression of a constitutively-active form of MEF2C ( MEF2C-VP16 ) promotes excitatory synapse elimination in hippocampal pyramidal neurons in a complex process that requires the RNA-binding protein , Fragile X mental retardation protein ( FMRP ) ( Flavell et al . , 2006; Pfeiffer et al . , 2010; Tsai et al . , 2012; Wilkerson et al . , 2014 ) . Brain-wide deletion of Mef2c was reported to cause an increase in dendritic spine density on dentate granule neurons of the hippocampal dentate gyrus ( Barbosa et al . , 2008 ) , whereas another group reported that Mef2c deletion in embryonic neural stem cells ( nestin-Cre ) , caused deficits in cortical neuron migration and excitatory synaptic transmission in a subset of animals ( Li et al . , 2008 ) . Recent genetic studies have linked human MEF2C to a syndromic form of intellectual disability with autistic features , and single-nucleotide polymorphisms ( SNPs ) near MEF2C produce significant risk for SCZ ( Paciorkowski et al . , 2013; Mikhail et al . , 2011; Novara et al . , 2010; Le Meur et al . , 2010; Cardoso et al . , 2009; Engels et al . , 2009 ) , which highlight the importance of this gene for normal brain development and function . However , the functional role ( s ) of MEF2C in early neuronal development , particularly in the neocortex , remains unclear . In the central nervous system , MEF2C is highly expressed very early in brain development ( ~E11 . 5 ) , and its expression is enriched in differentiated forebrain neurons within the neocortex and dentate gyrus ( Lyons et al . , 1995; Leifer et al . , 1993 , 1997 ) . Here , we sought to evaluate the role of MEF2C in differentiated cortical excitatory neurons , and to determine whether loss of MEF2C function in these neuronal populations might produce behavioral and synaptic phenotypes with potential relevance to its associated neurodevelopmental disorders .
Mef2c mRNA is enriched in the developing cortical plate , mature cortex and dentate gyrus ( Leifer et al . , 1993 ) . Immunostaining of mature brain slices with MEF2C-specific monoclonal antibodies revealed that >99% of the MEF2C-positive cells co-localized with the neuronal marker , NeuN ( Figure 1A ) , indicating that MEF2C expression in the cortex is primarily restricted to neurons . To generate conditional gene disruption of Mef2c selectively in differentiated forebrain excitatory neurons , we bred homozygous floxed Mef2c mutant mice ( Lin et al . , 1997 ) with mice heterozygous for Cre recombinase inserted into the endogenous Emx1 gene ( Iwasato et al . , 2008 ) , which expresses Cre in ~90% of differentiated neocortical and hippocampal excitatory neurons and in some forebrain glia starting as early as embryonic day 11 . 5 ( E11 . 5 ) . The Mef2c cKO ( Mef2cfl/fl;Emx1IRES-Cre/+ ) mice show selective and dramatic reduction of MEF2C protein levels throughout the cortex and hippocampus , but no reductions were observed in the striatum or thalamus ( Figure 1B ) – Emx1-negative regions that express low levels of MEF2C . 10 . 7554/eLife . 20059 . 003Figure 1 . Generation of Mef2c cKO mice . ( A ) MEF2C protein ( green ) is enriched in NeuN-positive cortical neurons ( red ) . ( B ) Western blot of MEF2C in various brain regions . ( C ) Somatosensory cortical thickness was slightly reduced in Mef2c cKO brains ( ~10% ) compared to control littermates . Thickness was averaged over 4 slices/brain from 5 brains per genotype . Data are represented as mean ± SEM . Statistical significance was determined by unpaired t-test . *p<0 . 05 , ns=not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 20059 . 00310 . 7554/eLife . 20059 . 004Figure 1—figure supplement 1 . Neuronal characterization of Mef2c cKO mice . ( A ) There was no difference in body weight between Mef2c cKO mice and control littermates during behavioral testing ( 12 weeks ) . ( B ) Basal MEF2 transcriptional activity is reduced by ~40% in Mef2c cKO neuronal cultures , and neuronal depolarization with KCl ( 60 mM ) promotes MEF2 activity in both control and Mef2c cKO cultures . Cortical cultures from control and Mef2c cKO animals were transfected with MEF2-response element ( MRE ) -Luciferase at DIV5 , and activity was monitored at DIV7 . ( C ) Nissl staining of adult control and Mef2c cKO brains show no gross morphological changes in the brain . ( D ) NeuN staining of adult control and Mef2c cKO brains reveal normal neuronal migration and layering . Data are represented as mean ± SEM . Statistical significance was determined by unpaired t-test . *p<0 . 05 , **p<0 . 005 , ns=not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 20059 . 004 The Mef2c cKO offspring were viable and healthy , and their body weights , growth trajectories and Mendelian frequency appear indistinguishable from their Cre-negative littermates ( Figure 1—figure supplement 1A and data not shown ) . Using a 3XMRE ( MEF2 response element ) -luciferase reporter in cultured cortical neurons from control or Mef2c cKO offspring , we detected a ~35% decrease in basal MEF2 activity , but no deficit in depolarization-induced , MEF2-dependent transcriptional activity ( Figure 1—figure supplement 1B ) , suggesting that MEF2A and MEF2D are sufficient to mediate normal levels of depolarization-dependent MEF2 activity . In young adult Mef2c cKO mice , we observed normal gross brain morphology and cortical layer organization ( Figure 1—figure supplement 1C , D ) , but Mef2c cKO mice did exhibit a slight decrease ( ~10% ) in neocortical thickness compared to controls ( Figure 1C ) . MEF2 transcriptional activity promotes excitatory synapse and dendritic spine elimination in the hippocampus ( Flavell et al . , 2006; Pfeiffer et al . , 2010; Tsai et al . , 2012; Zang et al . , 2013 ) . Therefore , we sought to test whether loss of Mef2c in the cortex alters cortical synaptic transmission in vivo . To achieve this , we measured cortical UP states , which are spontaneous , synchronous oscillations of neocortical networks that are driven by recurrent excitatory and inhibitory synaptic circuitry ( Hays et al . , 2011; Gibson et al . , 2008 ) , to assess overall synaptic function and excitability of the neocortical circuit within the somatosensory cortex ( SSC ) of the Mef2c cKO mice . Surprisingly , we observed large reductions ( ~90% ) in the frequency of spontaneous UP states in acute slices from the SSC of Mef2c cKO mice ( Figure 2A ) . In addition , the UP states in the Mef2c cKO mice were shorter in duration ( ~50% ) and smaller in amplitude ( ~50% ) ( Figure 2A ) . To further explore this decrease in neocortical circuit activity , we performed patch-clamp recordings of layer 2/3 pyramidal neurons from the SSC acute slices . In the Mef2c cKO slices , we detected small decreases in both the frequency and amplitude of miniature excitatory postsynaptic currents ( mEPSCs ) ( Figure 2B ) , although the decrease in frequency did not quite reach statistical significance ( p=0 . 07 ) . We also observed large increases in both the frequency and amplitude of miniature inhibitory postsynaptic currents ( mIPSCs ) ( Figure 2C ) . Together , these findings indicate that the embryonic loss of MEF2C in cortical excitatory neurons results in a small decrease in glutamatergic synaptic transmission and a large increase in inhibitory synaptic transmission – the combination of which likely contributes to the dramatic reduction in cortical network activity as detected by spontaneous UP states . 10 . 7554/eLife . 20059 . 005Figure 2 . Increased cortical inhibition in Mef2c cKO mice . ( A ) UP states in 3-week old Mef2c cKO mice . Mef2c cKO mice have fewer spontaneous UP states than control mice . Additionally , the duration and amplitude of each spontaneous UP state was significantly reduced in the Mef2c cKO mice . Representative recordings from control and Mef2c cKO organotypic slices . Scale bar = 50 µV/1 s . ( B ) Mef2c cKO mice have reduced mEPSC frequency and amplitude in cortical layer 2/3 pyramidal neurons from 3-week old mice . Scale bar = 200 ms/10 pA . ( C ) Mef2c cKO mice have increased mIPSC frequency and amplitude in cortical layer 2/3 pyramidal neurons from 3-week old mice . Scale bar = 200 ms/10 pA . Data are represented as mean ± SEM . Statistical significance was determined by unpaired t-test using GraphPad Prism . *p<0 . 05 , **p<0 . 01 , ****p<0 . 0001 . Numbers of slices/neurons ( n ) are reported in each bar for respective experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 20059 . 005 Since changes in UP states or the frequency of mEPSCs or mIPSCs can result from an alteration in synapse number , we next analyzed the density of excitatory and inhibitory synapses and overall dendritic complexity . While the Mef2c cKO cortical pyramidal neurons have a normal dendritic complexity ( Figure 3—figure supplement 1A , B ) , we detected a ~2-fold increase in dendritic GABAergic synapse density on Mef2c cKO cortical pyramidal neurons ( Figure 3A ) as determined by quantifying co-localization of pre- and postsynaptic markers ( GAD65 and GABRG2 , respectively ) . We also detected a significant reduction in dendritic spine density on Mef2c cKO neurons ( Figure 3B and Figure 3—figure supplement 1E ) , suggesting that MEF2C regulates , directly or indirectly , the densities of both excitatory and inhibitory synapses . The direction of these changes in E/I synapse density argues against the likelihood that the increase in inhibitory synapse density is a compensatory reaction to the reduction in excitatory synapse density , and vice versa , as these changes work in the same direction to reduce overall network synaptic activity . 10 . 7554/eLife . 20059 . 006Figure 3 . MEF2C functions as a transcriptional repressor to regulate synapse development in postsynaptic cortical pyramidal neurons . ( A ) Representative image of a GFP expressing mouse cortical neuron immunostained with antibodies against GAD65 ( pre-synaptic ) and GABRG2 ( post-synaptic ) . Quantification of inhibitory synapse density ( see Materials and methods ) on Mef2c cKO neurons showed an increase compared to wildtype control neurons . ( B ) Representative image of spine density across a dendritic stretch . Quantification of spine density on Mef2c cKO neurons showed a reduction compared to wildtype control neurons . ( C ) Quantified GABAergic synapse density onto Mef2cfl/flcortical pyramidal neurons transfected at DIV4 with either Cre-GFP ( Cre ) or an enzyme-dead mutant of Cre-GFP ( ΔCre ) . ( D ) Quantified spine density onto Mef2cfl/fl cortical pyramidal neurons transfected at DIV4 with either Cre-GFP ( Cre ) or an enzyme-dead mutant of Cre-GFP ( ΔCre ) . ( E ) Quantified GABAergic synapse density onto Mef2cfl/flGAD65 positive interneurons transfected at DIV4 with either Cre-GFP ( Cre ) or an enzyme-dead mutant of Cre-GFP ( ΔCre ) . ( F ) Quantified spine density onto WT cortical pyramidal neurons transfected at DIV4 with either an empty vector , a constitutive transcriptional promoting form of MEF2C ( MEF2-VP16 ) , or a constitutive transcription repressor form of MEF2C ( MEF2-EN ) . ( G ) Quantified spine density onto wildtype or Mef2c cKO neurons transfected with either an empty vector or MEF2-EN . ( H ) Quantified GABAergic synapse density onto WT cortical pyramidal neurons transfected at DIV4 with an empty vector , a constitutive transcriptional promoting form of MEF2C ( MEF2-VP16 ) , or a constitutive transcription repressor form of MEF2C ( MEF2-EN ) . ( I ) Quantified GABAergic synapse density onto wildtype or Mef2c cKO neurons transfected with an empty vector or MEF2-EN . Data are represented as mean ± SEM . Number ( n ) of neurons ( A , C , E , H , I ) or of dendritic stretches ( B , D , F , G ) are reported in each bar . Statistical significance was determined by unpaired t-test ( A–E ) , One-way ANOVA ( F , H ) or Two-way ANOVA ( G , I ) using GraphPad Prism . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , ****p<0 . 0001 . Also see Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 20059 . 00610 . 7554/eLife . 20059 . 007Figure 3—figure supplement 1 . Structural synaptic changes in Mef2c cKO neurons . ( A ) Representative images of GFP-transfected primary cortical neurons at DIV18 . Sholl analysis of pyramidal neurons at DIV18 reveals no significant changes in dendritic complexity in Mef2c cKO neurons in vitro . n=57 neurons for control and n=48 neurons for Mef2c cKO . ( B ) Representative traces of in vivo golgi-stained cortical layer 2/3 pyramidal neurons from somatosensory cortex in adult mice . Sholl analysis of pyramidal neurons reveals no significant changes in dendritic complexity by genotype . n=9 neurons from 3 control animals and n=8 neurons from 3 Mef2c cKO animals . ( C−D ) Immunocytochemical analysis of inhibitory GAD65-presynaptic ( B ) and GABARγ2-postsynaptic ( C ) puncta in cultured cortical neurons at DIV18 . Cortical neurons from Mef2c cKO mice show no change in inhibitory presynaptic puncta ( GAD65 positive ) or postsynaptic puncta ( GABARγ2 positive ) compared to controls as measured by co-localization of GAD65 ( presynaptic ) and GFP ( neuron mask ) . ( E ) Mef2c cKO neurons have fewer dendritic spines than control neurons . Primary cortical neurons were grown to DIV18 , and spines were visualized using myristoylated-GFP . Reduced spine density was observed in both secondary and tertiary dendrites , resulting in an overall reduction in spine density in all dendrites ( Figure 3B ) . Numbers of dendritic stretches ( n ) are reported in each bar from at least 30 control and 22 Mef2c cKO neurons . ( F ) MEF2C-Engrailed and MEF2C-VP16 regulate MEF2 transcriptional activity in cultured cortical neurons . ( left ) Representative images of 3XMRE-mCherry expression in cortical neurons transfected with either empty vector , MEF2C-EN or MEF2C-VP16 . ( right ) Quantification of average 3XMRE-mCherry intensity normalized to GFP intensity . Data are represented as mean ± SEM . Statistical significance was determined by repeated measures ANOVA ( A , B ) , unpaired t-test ( C–E ) , or one-way ANOVA followed by Kruskal-Wallis post-hoc test ( F ) . **p<0 . 01 , ****p<0 . 0001 , ns=not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 20059 . 007 To determine if the MEF2C-dependent synaptic changes are cell autonomous or represent a consequence of indirect network changes , we cultured Mef2cfl/fl cortical neurons and generated a sparse population of individual Mef2c cKO neurons ( transfection efficiency <1% ) by transient transfection of Cre-recombinase or an enzyme-dead mutant form of Cre ( △Cre ) . In this way , >99% of the presynaptic inputs onto the transfected neurons are wild-type , so noted changes in the synapse density onto Cre-expressing neurons are likely postsynaptic cell-dependent functions for MEF2C . Using this approach , we observe that Cre-expressing cortical pyramidal neurons displayed a significant increase in GABAergic synapse density ( Figure 3C ) and a significant decrease in dendritic spine density ( Figure 3D ) , similar to the direction and magnitude of E/I synapse changes observed in the global Mef2c cKO neurons ( Figure 3A , B ) . Interestingly , we did not observe any significant changes in GABAergic synapse density onto GAD65-positive Mef2cfl/fl interneurons transfected with Cre ( Figure 3E ) , suggesting the increase in inhibitory synapses is specific to excitatory cortical pyramidal neurons . Overexpression of a transcription-promoting form of MEF2C ( MEF2C-VP16 ) reduces structural and functional glutamatergic synapse density in hippocampal pyramidal neurons ( Flavell et al . , 2006; Pfeiffer et al . , 2010; Tsai et al . , 2012; Zang et al . , 2013 ) , similar to the effect of Mef2c loss-of-function mutation in cortical neurons . As such , we considered the possibility that MEF2C-VP16 and MEF2C loss-of-function could both produce excitatory synapse reductions if endogenous MEF2C is functioning as a transcriptional repressor on key target genes that regulate synapse density . Consistent with this basic idea , we found that overexpression of MEF2C-VP16 , a constitutively-active form of MEF2C ( Figure 3—figure supplement 1F ) , in wild-type cortical pyramidal neurons significantly reduced dendritic spine density ( Figure 3F ) , whereas a dominant-repressor form of MEF2C ( MEF2C-EN ) ( Figure 3—figure supplement 1F ) failed to alter dendritic spine density ( Figure 3F , G ) . However , cell-autonomous expression of MEF2C-EN in Mef2c cKO neurons rescued the decrease in dendritic spine density ( Figure 3G ) , strongly suggesting that endogenous MEF2C functions as a transcriptional repressor to regulate dendritic spine density . In wild-type cortical pyramidal neurons , MEF2C-VP16 also significantly increased GABAergic synapse density ( Figure 3H ) , whereas MEF2C-EN had no effect on GABAergic synapse density ( Figure 3H , I ) . Similar to the dendritic spine density findings , cell-autonomous expression of MEF2C-EN rescued the increase in GABAergic synapse density observed in the Mef2c cKO cortical pyramidal neurons ( Figure 3I ) . These results suggest that endogenous MEF2C functions predominantly as a transcriptional repressor to inhibit target genes that promote excitatory synapse elimination and inhibitory synapse formation and/or stability . Since MEF2C is a nuclear transcription factor , we sought to identify differential gene expression that results from early embryonic deletion of Mef2c . To this end , we performed deep sequencing of polyA-enriched mRNAs ( RNA-Seq ) isolated from the somatosensory cortex of Mef2c cKO or control littermates and identified differentially expressed genes ( DEGs ) ( Figure 4A; Supplementary file 1 ) . Using a stringent cut-off ( |log2FC| > 0 . 3 , FDR < 0 . 05 ) , we detected 1076 DEGs in Mef2c cKO cortex compared with controls , including 598 genes with a decreased expression level and 478 genes with an increased expression level . Comparing the Mef2c cKO DEGs to genes expressed in specific neuronal populations ( Cahoy et al . , 2008 ) , the DEGs were highly enriched for neuron-expressed genes ( Figure 4—figure supplement 1A , B ) , consistent with the observation that Mef2c is not expressed at appreciable levels in non-neuronal cells ( Figure 1A ) . Notably , the expression levels of Mef2a and Mef2d , which are close family members , were not significantly altered in the Mef2c cKO mice ( Figure 4C ) . 10 . 7554/eLife . 20059 . 008Figure 4 . Characterization of Mef2c cKO RNA-Seq differentially expressed genes . ( A ) Heatmap showing the disorder-related genes differentially expressed in Mef2c cKO ( KO ) compared with wild-type ( WT ) . In red , genes with higher expression; in blue , genes with lower expression . ( B ) Overlap between Mef2c cKO DEGs and gene sets of interest . Marked , the overlap p-values . Number of genes for each gene sets are indicated . ( C ) Relative expression of selective down-regulated ASD-associated genes from Mef2c cKO DEGs compared to controls . Both RNA-Seq and qPCR ( P21 and Adult ) show similar expression changes for most genes . ( D ) Relative expression of selective up-regulated Mef2c cKO DEGs compared to controls . Both RNA-Seq and qPCR ( P21 and Adult ) show similar expression changes . Data are represented as mean ± SEM . See Materials and methods for statistical analysis . n=3 animals/genotype for RNA-Seq; n=6 animals/genotype for adult qPCR; n=6 control and n=4 Mef2c cKO for P21 qPCR . Also see Table 1 , Supplementary file 1 and Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 20059 . 00810 . 7554/eLife . 20059 . 009Figure 4—figure supplement 1 . Differential gene expression in Mef2c cKO cortical tissue . ( A ) Mef2c cKO DEGs were significantly enriched for neuron-specific genes . ( B ) Mef2c cKO DEGs were significantly enriched for S1-Pyramidal and CA1-Pyramidal neurons . Gene ontology enrichment for Mef2c cKO DEGs . In red , the up-regulated genes; in blue , the down-regulated genes . Circle size is correlated with the adjusted p-value . Gene ontology categories are alphabetically listed on the y-axis . Differentially expressed genes showed enrichment for categories involved in neuronal development and synaptic transmission . DOI: http://dx . doi . org/10 . 7554/eLife . 20059 . 009 Gene ontology enrichment analysis revealed multiple distinct categories for up-regulated and down-regulated genes ( Figure 4—figure supplement 1C; Table 1 ) . For instance , genes with an increased expression level in the young adult Mef2c cKO cortex were significantly enriched for cellular processes such as neuron differentiation and development , suggesting that MEF2C might function as a repressor for some of these DEGs . Analysis of down-regulated genes in Mef2c cKO mice revealed a significant enrichment for cellular processes including synaptic transmission and ion transport ( p<0 . 01 , BH correction; Figure 4—figure supplement 1C ) , suggesting an important role for MEF2C in regulating synapse function and neuronal excitability . 10 . 7554/eLife . 20059 . 010Table 1 . Gene-ontology of Mef2c cKO DEGs . DOI: http://dx . doi . org/10 . 7554/eLife . 20059 . 010CategoryTermCountBenjaminilogUPneuron projection development222 . 49E-0546 . 03800653UPneuron development247 . 33E-0541 . 34896025UPaxonogenesis189 . 25E-0540 . 33858267UPneuron projection morphogenesis180 . 0001438 . 53871964UPcell projection organization240 . 00017537 . 56961951UPcell morphogenesis involved in neuron differentiation180 . 00018137 . 42321425UPcell projection morphogenesis180 . 0005532 . 59637311UPneuron differentiation260 . 00056932 . 44887734UPcell morphogenesis involved in differentiation180 . 00082230 . 85128182UPcell part morphogenesis180 . 00082230 . 85128182UPcell morphogenesis200 . 00807524220 . 92844454UPcellular component morphogenesis200 . 0353826414 . 51209766DOWNpotassium ion transport287 . 00E-09−8 . 15E+01DOWNmetal ion transport445 . 62E-10−92 . 50263684DOWNcation transport451 . 63E-087 . 79E+01DOWNion transport501 . 25E-06−59 . 03089987DOWNsynaptic transmission213 . 13E-05−45 . 04455662DOWNtransmission of nerve impulse237 . 95E-05–40 . 99632871DOWNcell-cell signaling250 . 00035−34 . 55931956 To further characterize the Mef2c cKO DEGs , we compared our gene list with the recently updated risk genes from the Simons Foundation Autism Research Initiative ( SFARI database , 667 genes ) ( Basu et al . , 2009 ) , mRNAs associated with FMRP ( Darnell et al . , 2011 ) , ID-associated genes from multiple sources ( Inlow and Restifo , 2004; Lubs et al . , 2012; Ropers , 2008; van Bokhoven , 2011 ) , and synaptic-associated genes ( Synaptome DB ) ( Pirooznia et al . , 2012 ) . In the Mef2c cKO DEGs , we identified a significant overrepresentation of ASD-risk genes ( p=0 . 0007 , hypergeometric test , perm=0 . 001 ) and synapse-linked genes ( p=0 . 0004 , hypergeometric test , perm=0 . 001 ) , ( Figure 4B ) , including the autism-linked genes Ntng1 , Nlgn1 , Nrxn1 , Nrxn3 , Pcdh19 , Shank2 , Shank3 , Pten and Htr1b . We also detected a significant enrichment for FMRP-associated RNAs ( p=3x10−07 , hypergeometric test , perm=0 . 001 ) , which is interesting since FMRP is required for MEF2C-VP16-induced excitatory synapse elimination ( Pfeiffer et al . , 2010 ) . Using quantitative PCR ( qPCR ) , we validated dysregulation of several Mef2c cKO DEGs in both postnatal day 21 ( P21 ) and adult SSC tissue ( Figure 4C , D ) . Interestingly , there were several dysregulated genes that are reported to regulate inhibitory GABAergic transmission ( Gabra5 and Nos1 ) , and we observed a significant increase in Pcdh10 mRNA , a factor we previously implicated in MEF2/FMRP-dependent glutamatergic synapse elimination ( Tsai et al . , 2012 ) . Overall , our RNAseq data analysis suggests that MEF2C , either directly or indirectly , influences a large , complex gene expression program that influences neuronal and synapse development , and numerous syndromic and idiopathic autism-linked genes . Since we observed significant dysregulation of many ASD-related genes in Mef2c cKO mice , and since MEF2C is linked to human neurodevelopment disorders with autistic features and cognitive deficits , we examined whether loss of MEF2C function in forebrain excitatory neurons might produce autism- and ID-like behavioral phenotypes . In humans , impairments in communication and social interactions are common symptom domains of autism and SCZ . Mef2c cKO mice displayed dramatic abnormalities in a putative form of oral social communication in mice – ultrasonic vocalizations ( USVs ) produced by a young adult male mouse when placed in the presence of a female in estrous or upon young pup separation from its mother ( Figure 5 ) ( Ey et al . , 2013; Hanson and Hurley , 2012 ) . In the presence of a sexually-receptive female , Mef2c cKO males generated far fewer USV calls ( ~70% reduction , Figure 5B ) , and showed a significant increase in the latency to the first call ( ~3-fold , Figure 5—figure supplement 1C ) . Wild-type littermate mice produced a range of distinct simple and complex stereotyped USV call subtypes ( Figure 5—figure supplement 1A ) ( Ey et al . , 2013 ) . In contrast , Mef2c cKO mice produced a ~5-fold increase in unstructured USVs , and a corresponding decrease in complex , but not simple , USVs ( Figure 5C and Figure 5—figure supplement 1B ) . While many basic USV parameters were unchanged by genotype ( Figure 5—figure supplement 1D–F ) , we detected significant reductions in the maximum USV frequency and the mean frequency at the end of calls ( Figure 5—figure supplement 1G–H ) . Similar to the adults , Mef2c cKO mice at postnatal days 4–10 produced significantly fewer USVs upon separation from the mother ( distress calls ) ( Figure 5D ) , but at this age , the USV structures , subtypes and basic call parameters were indistinguishable from WT littermates ( Figure 5—figure supplement 1I–J ) . Together these data indicate that Mef2c cKO mice produce significantly fewer USVs in a species-specific form of putative oral communication . 10 . 7554/eLife . 20059 . 011Figure 5 . Social behavior abnormalities in Mef2c cKO mice . ( A ) Representative spectrograms of ultrasonic vocalizations ( USVs ) recorded from adult male mice in the presence of an estrous female mouse . ( B ) Adult Mef2c cKO male mice emit fewer USVs to an estrous female than control littermates . ( C ) Adult Mef2c cKO male mice show different call types than control littermates . Mef2c cKO mice have more unstructured USVs ( % ) and fewer complex USVs than control mice . Representative images of call types and further breakdown of USV sub-type are presented in Figure 5—figure supplement 1A , B . ( D ) Juvenile Mef2c cKO mice ( pups ) emit fewer USVs during maternal separation than control littermates . USVs were recorded on postnatal days ( P ) 4 , 6 , and 10 . ( E ) Mef2c cKO mice show reduced preference for interacting with a novel social target . ( F ) Mef2c cKO mice show normal olfactory response to novel social scent . ( G ) Mef2c cKO mice fail to build structured nest when utilizing a nest score system ( Deacon , 2006 ) . ( H ) Mef2c cKO mice induce control littermates to withdrawal in the tube test for social dominance in >90% of trials . ( I ) Mef2c cKO mice show reduced preference for a natural reward , sucrose . Both genotypes showed aversion to the bitter solution , 0 . 04% quinine . Data are represented as mean ± SEM . Statistical significance was determined by unpaired t-test ( B , G , H–I ) or 2-way ANOVA with Sidak’s post-hoc comparison ( C–F ) . #p<0 . 1 , *p<0 . 05 , **p<0 . 005 , ***p<0 . 0005 , ns=not significant . Numbers of animals ( n ) are reported in each bar for respective experiment . Also see Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 20059 . 01110 . 7554/eLife . 20059 . 012Figure 5—figure supplement 1 . Characterization of Mef2c cKO mouse USVs . ( A ) Representative images of the different classes of call types , modified from previous studies ( Shalizi and Bonni , 2005 ) . Example of different USV types: Short/Simple ( flat and upward ) , Complex ( 1-Frequency jump and complex ) , and Unstructured ( mixed and unstructured ) , respectively . ( B ) Breakdown of the call usage frequency of each sub-type of USVs emitted by adult Mef2c cKO male mice to an estrous female mouse . ( C ) Adult Mef2c cKO male mice take longer to emit the first USV to an estrous female . ( D ) No significant difference in the average duration of each adult USV was observed between control and Mef2c cKO mice . ( E ) No change in USV amplitude was recorded at either the start or end of each call . ( F ) The average maximum amplitude of USVs was not different between control and Mef2c cKO mice . ( G ) Mean frequency of adult USVs at the start and end of each call . While both genotypes show the same average start frequency for each USV , control mouse USVs ended at a higher frequency than the call started while Mef2c cKO mice show reduced frequency at the end of the call . ( H ) The average maximum frequency for USVs was reduced in Mef2c cKO mice compared to control mice . ( I ) No significant differences in the duration of USVs from pups during maternal separation were recorded . ( J ) No difference in the call types ( simple , complex , unstructured ) was observed between control and Mef2c cKO juvenile mice . ( K ) Mef2c cKO mice show reduced time interacting with a social target in a 2-choice interaction assay . Time spent interacting with a social target ( novel mouse ) and novel object ( black paper binder ) are reported . Solid bars represent empty holding cages ( targets absent ) . Striped bars represent targets present ( mouse and object ) . Data are represented as mean ± SEM . Statistical significance was determined by unpaired t-test ( B–D , F , H ) or 2-way ANOVA ( E , G , J–K ) . #p<0 . 10 , *p<0 . 05 , **p<0 . 005 , ****p<0 . 0001 . ns = not significant . Numbers of animals ( n ) are reported in each bar for adult and juvenile mouse USV and social interaction experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 20059 . 012 In a social interaction test , the wild-type littermates spent significantly more time interacting with an unfamiliar mouse than an empty chamber ( Figure 5E ) . The Mef2c cKO mice also spend more time interacting with the social animal vs . the empty chamber , but the Mef2c cKO mice spent significantly less time interacting with the social animal than the control mice ( Figure 5E ) . In another cohort of mice , we observed a significant reduction in social interaction even in the presence of a competing novel inanimate object ( Figure 5—figure supplement 1K ) . The reduction in social interaction did not appear to be due to deficits in olfactory recognition of social animals or basic novelty detection , since Mef2c cKO mice showed a strong preference to interact with a social-related smell from an unfamiliar mouse ( Figure 5F ) . In addition to social interaction deficits , Mef2c cKO mice showed significant reductions in another social-related behavior , nest building ( Figure 5G ) ( Etherton et al . , 2009; Kwon et al . , 2006; Deacon , 2006 ) , and they displayed abnormal social behavior in the tube test for social dominance ( Figure 5H ) . Deficits in brain reward function have been proposed to contribute to some autistic behaviors , including social interaction ( Insel , 2003; Dichter et al . , 2012 ) , and a lack of motivation is a common negative symptom of SCZ . Interestingly , Mef2c cKO mice show significant reductions in a hedonic-related behavior in the sucrose preference test ( Figure 5I ) , an assay that measures an animal’s preference for a sweet solution vs . water . The reduction in sucrose preference is not likely due to basic gustatory deficits since Mef2c cKO mice showed normal avoidance of a bitter-tasting solution ( quinine ) ( Figure 5I , right ) . Taken together , Mef2c cKO mice demonstrate multiple abnormalities in mouse social behaviors and a strong deficit in an appetitive reward-related behavior . Autism is characterized by restricted or repetitive patterns of behavior , interests or activities ( American Psychiatric Association , 2013 ) . Interestingly , Mef2c cKO mice spent a significantly greater fraction of time in a repetitive jumping behavior ( ~3-fold increase , Figure 6A ) , which was visually observed in both novel and home cage settings . In addition , we detected a significant increase in repetitive fine motor movements ( Figure 6B ) , which is often interpreted as a motor stereotypy behavior ( Avale et al . , 2004 ) . In contrast , no differences by genotype were observed in time spent self-grooming or digging ( Figure 6—figure supplement 1A , B ) . Mef2c cKO displayed normal motor coordination as measured in the accelerating rotarod test ( Figure 6C ) , but significant motor hyperactivity was detected in a novel environment ( Figure 6D ) . 10 . 7554/eLife . 20059 . 013Figure 6 . Repetitive behaviors and hyperactivity in Mef2c cKO mice . ( A ) Mef2c cKO mice spend more time jumping than control animals in an operant chamber over a 1-hr interval . ( B ) Mef2c cKO mice have more fine motor movements in an operant chamber , reflective of stereotypic activity . ( C ) Latency to fall off an accelerating rotarod is not different in the Mef2c cKO mice . ( D ) Mef2c cKO mice are hyperactive compared to control littermates . Activity was monitored for 1 hr , and data is plotted by beam breaks/5 min ( left ) and cumulative beam breaks ( right ) . Data are represented as mean ± SEM . Statistical significance was determined by unpaired t-test ( A–D ) or 2-way ANOVA ( D ) . **p<0 . 005 , ***p<0 . 0005 , ns=not significant . Numbers of animals ( n ) are reported in each bar for respective experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 20059 . 01310 . 7554/eLife . 20059 . 014Figure 6—figure supplement 1 . Mef2c cKO mice do not exhibit repetitive grooming or digging . ( A ) No difference in time spent grooming is observed in Mef2c cKO mice compared to controls . ( B ) No difference in the number of digging bouts was observed between control and Mef2c cKO mice . Data are represented as mean ± SEM . Statistical significance was determined by unpaired t-test . ns=not significant . Numbers of animals ( n ) are reported in each bar for respective experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 20059 . 014 Intellectual disability is also a common associated symptom of autism ( American Psychiatric Association , 2013 ) , and cognitive deficits comprise one of the three major symptom domains in SCZ . Mef2c cKO and control littermates showed similar startle responses to a broad range of foot shock intensities and acoustic white-noise volumes ( Figure 7A , B ) , suggesting that nociception and auditory sensory sensitivity are not significantly altered in the mutant mice . However , the Mef2c cKO mice showed profound deficits in threat-related learning and memory in the classic fear-conditioning assay ( Figure 7C–E ) . Unlike the WT controls , the Mef2c cKO mice failed to develop robust freezing behaviors in the 1 min . periods following each tone-shock pairing ( Figure 7C ) . Twenty-four hours later , Mef2c cKO mice showed significantly reduced freezing behaviors when re-exposed to the shock-paired context ( Figure 7D ) or after presentation of the tone cue ( conditioned stimulus ) in an altered context ( Figure 7E ) . Together these findings suggest that embryonic loss of MEF2C in excitatory forebrain neurons causes significant deficits in fear learning and memory , multiple social behaviors , socially-motivated ultrasonic vocalizations , and reward-related behaviors . Mef2c cKO mice also show significant increases in repetitive motor behaviors and overall hyperactivity – all symptom domains with potential relevance to human neurodevelopmental disorders such as autism , ID and SCZ . 10 . 7554/eLife . 20059 . 015Figure 7 . Cognitive deficits in Mef2c cKO mice . ( A–B ) Both control and Mef2c cKO mice showed similar force plate response to various intensities of shock ( A ) or acoustic startle ( B ) . Grey bars highlight the intensities used in fear conditioning ( FC ) . ( C ) During training for fear conditioning , Mef2c cKO mice fail to increase freezing during the 1-minute intervals after each tone/shock pairing . ( D ) Fear Conditioning . Mef2c cKO mice show deficits in contextual memory . ( E ) Fear Conditioning . In a novel context , Mef2c cKO mice show deficits in cue-dependent memory . Data are represented as mean ± SEM . Statistical significance was determined by 2-way ANOVA ( A–C ) or unpaired t-test ( D , E ) . *p<0 . 05 , **p<0 . 005 , ***p<0 . 0005 , ns=not significant . Numbers of animals ( n ) are reported in each bar for respective experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 20059 . 015
We conditionally deleted Mef2c in Emx1-lineage populations , including cortical excitatory neurons , during early embryogenesis , and observed a large reduction in cortical network activity ( UP states ) , a small decrease in layer 2/3 pyramidal neuron excitatory synaptic transmission and a large increase in inhibitory synaptic transmission in primary sensory cortex . Consistent with these ex vivo acute slice recordings , cultured Mef2c cKO cortical neurons displayed a significant decrease in dendritic spine density and an increase in GABAergic synapse density . Our functional rescue findings suggest that MEF2C in cortical pyramidal neurons regulates E/I synapse densities in early development by acting in the postsynaptic neuron as a cell-autonomous , transcriptional repressor on key target genes . Loss of MEF2C alters the gene expression , directly or indirectly , of numerous autism- and synapse-linked genes and RNAs known to associate with FMRP , the leading genetic cause of ID and autism . These findings position MEF2C as a critical transcriptional regulator functioning at the nexus of numerous synapse- and neurodevelopmental disorder-linked genes . Finally , the Mef2c cKO mice display multiple behavioral phenotypes that are reminiscent of core autism symptoms in humans , including abnormal social behaviors , reduced USVs in multiple ages and contexts , and an increased frequency of some repetitive motor behaviors . We also observed learning and memory deficits and motor hyperactivity , which are common associated symptoms of ASDs and observed in patients with MEF2C mutations . Taken together , our findings suggest that the loss of MEF2C’s repressor function in cortical excitatory neurons produces an E/I synapse imbalance and that this synaptic phenotype might contribute to the numerous , neurodevelopmental disorder-related behavioral phenotypes observed in the Mef2c cKO mice . Previous studies have demonstrated an important role for MEF2A and MEF2D in the process of activity-dependent excitatory synapse elimination in hippocampal neurons ( Flavell et al . , 2006; Pfeiffer et al . , 2010; Tsai et al . , 2012 ) . However , similar to a previous report ( Li et al . , 2008 ) , we found that loss of MEF2C produced a decrease in cortical excitatory synaptic transmission , suggesting that MEF2C is a positive regulator of excitatory synapses in cortical neurons . However , overexpression of MEF2C-VP16 , a strong , constitutive transcriptional activator form of MEF2C , in wild-type cortical neurons also decreased dendritic spine density and increased inhibitory synapse density ( Figure 3F , H ) – essentially phenocopying the Mef2c cKO neurons . In contrast , overexpression of MEF2C-EN , a dominant repressor form of MEF2C , in WT neurons had no effect on dendritic spines or GABAergic synapse densities; whereas , expression of MEF2C-EN in Mef2c cKO cortical pyramidal neurons normalized excitatory and inhibitory synapse densities to WT levels ( Figure 3G , I ) . Together , these findings suggest that endogenous MEF2C functions as a cell-autonomous , transcriptional repressor at one or more key target genes involved in excitatory synapse elimination and GABAergic synapse formation/stability . These observations are consistent with previous reports demonstrating that MEF2A can also function as a transcriptional repressor to promote cerebellar excitatory synapse maturation ( Shalizi et al . , 2006 ) , highlighting a critical role for diverse mechanisms of transcriptional regulation by MEF2 family members to regulate synapse development in the brain . It’s interesting to note that membrane depolarization of Mef2c cKO cortical neurons stimulated MEF2 reporter gene activity levels that are comparable to responses in WT littermates , suggesting that the activity-dependent induction of MEF2-sensitive target genes can be fully compensated for by endogenous MEF2A and MEF2D . In the future , it will be important to study the overlapping and potentially distinct functions of the various MEF2 family members , and their activity-dependent regulation , in cortical neuron function and development . Loss of MEF2C in forebrain excitatory neurons produces an increase in structural and functional excitatory synapses formed onto hippocampal dentate granule neurons ( DG ) ( Barbosa et al . , 2008; Adachi et al . , 2016 ) , suggesting that MEF2C might have cell-type specific functions and/or that the increase in DG excitatory synaptic transmission is an indirect , homeostatic effect of decreased cortical stimulation of DG neurons . In the future , cell autonomous manipulations of the DG neurons will be important to resolve this question . Also , postnatal Mef2c deletion in forebrain excitatory neurons did not produce social or repetitive behavioral phenotypes , despite the increase in DG dendritic spine density ( Adachi et al . , 2016 ) , suggesting a dissociation of hippocampal DG spine density and postnatal MEF2C deletion from several ASD-related behaviors . As such , the role ( s ) for MEF2C in embryonic and/or early postnatal cortical development might be more critical for producing the behavioral phenotypes observed in our Mef2c cKO mice . Finally , while we observe significant synaptic and gene expression differences in the SSC of the Mef2c cKO mice , Emx1-Cre expression is also detected in other brain regions and cell types , including the olfactory bulb and some glia , and as such , it is not possible to attribute the behavioral phenotypes observed in the Mef2c cKO mice to the synaptic changes in the SSC , and future studies that further dissect the Emx1-lineage populations will be necessary to link specific Mef2c cKO behavioral phenotypes with a specific brain region ( s ) . Imbalances in excitatory and inhibitory synaptic transmission are proposed to underlie many neuropsychiatric disorders , including ASDs ( Rubenstein , 2010; Cellot and Cherubini , 2014 ) and SCZ ( Coyle et al . , 2016 ) . Genetic analyses of patients affected by these disorders revealed mutations in many synapse-related genes ( Garber , 2007; McCarthy et al . , 2014 ) . In mice , increased excitatory synaptic function has been reported in several mouse ASD models , including mutant mice lacking Fmr1 , Pten , and Tsc1/2 genes ( Gibson et al . , 2008; Williams et al . , 2015; Bateup et al . , 2013 ) . In contrast , only a few prior studies have examined the role of altered inhibitory synapse function in ASD-related behaviors . For example , mice containing a human disease mutation in the Nlgn3 gene ( Nlgn3 R451C ) displayed an increase in inhibitory synaptic transmission and several autism-associated behaviors ( Tabuchi et al . , 2007 ) , suggesting that altered E/I balance in either direction can produce behavioral phenotypes with potential relevance to neurodevelopmental disorders . Little is currently known about the mechanisms that control GABAergic synapse density ( Petrini and Barberis , 2014 ) . Interestingly , GABAergic synapse formation during development precedes glutamatergic synapse formation ( Ben-Ari et al . , 2007 ) , and it can strongly affect the subsequent development of glutamatergic synapses and neuronal morphology ( Wang and Kriegstein , 2008 ) . Our finding here show that MEF2C in postsynaptic cortical neurons function to cell-autonomously modulate the density of GABAergic synapses formed on the dendrites of cortical pyramidal neurons . In contrast , loss of MEF2C on cortical interneurons did not alter GABAergic synapse density ( Figure 3E ) , suggesting a specific role in cortical pyramidal neurons . Future studies will focus on exploring the precise regulation , molecular mechanisms and developmental period ( s ) when MEF2C regulates inhibitory and/or excitatory synapse density . Comparison of differentially expressed genes ( DEGs ) in the SSC of Mef2c cKO mice by RNA-Seq revealed ~1000 significantly altered genes . Nearly half of the dysregulated genes in the Mef2c cKO cortex showed an increased expression level compared to controls , consistent with the idea that MEF2C can function as a transcriptional repressor , directly or indirectly , on a subset of these DEGs . Among all the dysregulated genes , we observed significant overlap of Mef2c cKO DEGs and autism-linked genes ( 78 ) , synaptome genes ( 191 ) , and FMRP-target RNAs ( 110 ) , many of which are associated with multiple disease and neuronal function groups ( Figures 4B and Figure 4—figure supplement 1C ) , including Nlgn1 ( ASD , Synaptome ) , Nrxn1/3 ( ASD , Synaptome , FMRP-bound RNAs ) , Grm4 ( ASD , FMRP-bound RNAs ) , and Shank2/3 ( ASD , Synaptome , FMRP-bound RNAs ) . Additionally , we found reduced Cdkl5 expression in the MEF2C cKO DEGs , consistent with the reduced expression of CDKL5 mRNA reported in human patients with MEF2C mutations ( Zweier et al . , 2010 ) . Recent human genetic studies have revealed that deletion of , or non-synonymous mutations in , the MEF2C gene is associated with a severe neurodevelopmental disorder with features of autism and ID ( Paciorkowski et al . , 2013; Mikhail et al . , 2011; Novara et al . , 2010; Le Meur et al . , 2010; Cardoso et al . , 2009; Engels et al . , 2009 ) . These MEF2C haploinsufficient patients present with a number of symptoms , including motor abnormalities ( e . g . dyskinesias , stereotypies and hyperactivity ) , impairments in reciprocity , severe deficits in verbal communication , and severe intellectual disability ( Paciorkowski et al . , 2013 ) . In humans , MEF2C haploinsufficiency appears to be sufficient , at least in reported individuals , to produce this complex and severe neurodevelopmental disorder . Generally consistent with previous reports ( Li et al . , 2008; Barbosa et al . , 2008 ) , our preliminary studies indicate that loss of one gene copy of Mef2c ( Mef2cfl/+;Emx1Cre/+ ) in the Emx1-cell lineage produces mice with behaviors indistinguishable from their Cre-negative WT controls ( A . J . H . and C . W . C . , unpublished observations ) . It is possible that in humans there are other factors that influence disease penetrance and severity , including unique or sensitized functions for MEF2C through human evolution , human-specific genetic modifiers and/or environmental influences that increase symptom penetrance or additional cell populations where reduction in MEF2C is required . Nevertheless , our findings here indicate that MEF2C plays an essential role in early cortical synaptic development , and that reduction in MEF2C function in forebrain excitatory neurons can produce behaviors potentially relevant to multiple intellectual and developmental disorders . Schizophrenia is a debilitating mental illness with neurodevelopmental origins that affects nearly 1% of the world’s population , and there is significant overlap in risk genes for ASDs and SCZ . In contrast to ASDs , human postmortem brain analysis of SCZ brains revealed a thinning of the cortex and a decrease in dendritic spine density , observations supporting the leading hypothesis that hypofunction of excitatory synaptic transmission underlies the pathophysiology of SCZ ( Coyle et al . , 2016 ) . Recently , 108 genomic loci were identified by SNP meta-analysis as conferring significant risk for SCZ , and MEF2C was identified as a candidate risk gene ( Schizophrenia Working Group of the Psychiatric Genomics Consortium , 2014 ) . In the Mef2c cKO mice , we observed a thinning of the cortex ( Figure 1C ) , a decrease in dendritic spine density of Mef2c cKO cortical neurons ( Figure 3B ) , and behavioral phenotypes that are reminiscent of cognitive and negative symptoms of SCZ ( e . g . learning and memory deficits , lack of pleasure and motivation , reduced sociability and poverty of speech ) . While the potential relevance of Mef2c cKO phenotypes to the pathophysiology and symptoms of ASDs , ID and/or SCZ is not yet clear , our findings reveal an essential role for MEF2C in cortical neuron development and typical animal behaviors . In summary , we show here that Mef2c is required for proper synapse development on excitatory forebrain neurons , and its embryonic loss in these populations produces mice with behavior phenotypes reminiscent of multiple neurodevelopmental disorders , including ASDs and ID . The behavior changes are associated with a reduction in cortical network activity and alterations in E/I synapse densities and function . We also show that MEF2C likely regulates E/I synapse density by functioning as a cell-autonomous , transcriptional repressor , and that Mef2c loss-of-function in Emx1-lineage populations produces , directly or indirectly , a dramatic dysregulation of hundreds of neuronal genes , positioning it at the nexus of numerous critical neurodevelopment genes reported to influence neuronal and synaptic development and risk for neurodevelopmental disorders .
Mice ( mus musculus ) were group housed ( 2–5 mice/cage; unless specified ) with same-sex littermates on a 12 hr light-dark cycle with access to food and water ad libitum . MEF2Cfl/fl mice were previously described ( Arnold et al . , 2007 ) , as were Emx1Cre/+ knock-in mice ( Iwasato et al . , 2008 ) . Test mice were bred and maintained on a mixed SVeV-129/C57BL/6J background . Experimental mice ( Mef2cfl/fl; Emx1Cre/+ ) were compared to Cre-negative littermates ( Mef2cfl/fl ) . Experimenters were blinded to the mouse genotype during data acquisition and analysis . Every experiment was independently replicated at least twice , and total numbers of animals/cells are reported in the respective figures . All procedures were conducted in accordance with the Institutional Animal Care and Use Committee ( IACUC ) and National Institute of Health guidelines . All experiments were independently replicated at least twice ( typically 3–4 times ) . The numbers of animals/neurons/dendritic stretches are reported in each figure , and these numbers were estimated based on previous reports . Outliers were determined using GraphPad’s Outlier calculator and excluded from data analysis . Mice were terminally anesthetized with sodium pentobarbital ( Sigma ) and perfused transcardially with PBS followed by 4% ( w/v ) paraformaldehyde ( PFA ) . Brains were post-fixed overnight at 4°C in 4% PFA then cryoprotected in 30% sucrose . Brains were coronally sectioned at 30 µm using a sliding microtome and stored in PBS with 0 . 02% sodium azide . Sections were blocked in 3% albumin from bovine serum ( BSA ) and 3% normal donkey serum in PBS with TritonX-100 ( 0 . 3% ) and Tween20 ( 0 . 2% ) for 2 hr at room temperature ( RT ) . Sections were immunostained overnight at 4°C anti-NeuN ( A60 , 1:200; Millipore; RRID:AB_177621 ) or anti-Mef2c ( 1:250; Abcam ab197070; RRID:AB_2629454 ) antibody followed by AlexaFluor-555 or AlexaFluor-488 conjugated secondary antibodies and dehydration . Cover slips were mounted using DPX mountant ( Sigma ) . For Nissl staining , slices were mounted on Superfrost Plus microscope slides ( Fisher ) , dried , and dehydrated . Slices were stained for 6 min in 37°C cresyl violet ( 0 . 1%; Fisher ) , differentiated for 2 min , dehydrated ( 100% ethanol ) , and cleared ( xylenes ) . Cover slips were mounted using DPX mountant ( Sigma ) . Cortical thickness was assessed using ImageJ software ( NIH ) . Adult ( 8–12 week old ) male mice were euthanized by CO2 asphyxiation followed by decapitation . Tissue from different brain regions was rapidly dissected and frozen on dry ice . Tissues were sonicated on ice in a SDS lysis buffer: 1% ( w/v ) SDS , 300 mM sucrose , 10 mM NaF ( Sigma ) , 50 mM HEPES ( Sigma ) , and 1X Complete Protease Inhibitor cocktail ( Roche ) . Samples were boiled for 10 min , then centrifuges at 16 , 000 x g for 10 min . Total protein concentration was determined by the DC protein assay kit ( BioRad ) , and 20 µg of total protein was resolved using 10% SDS-PAGE . Proteins were transferred to Immobilon-FL PVDF ( Millipore ) , blocked in Odyssey blocking buffer ( Li-Cor ) for 2 hr , and incubated overnight with either anti-Mef2c ( AbCam ab197070; 1:2500; RRID:AB_2629454 ) or anti-Neuronal class III β-tubulin ( Tuj1 , Covance; 1:10 , 000; RRID:AB_2313773 ) antibodies . Blots were developed with enhanced chemiluminiscence western blotting detection reagent ( Amersham ECL-Prime; GE Healthcare ) or Odyssey CLx Western blot system ( LiCor Biosciences ) . Primary cortical neurons were cultured from individual P0 mouse pups as previously described ( Beaudoin et al . , 2012 ) , with modifications . Briefly , the neocortex was isolated from P0 individual pups ( after removing hippocampus and midbrain ) , dissociated with 0 . 25% trypsin for 20 min , and plated on PDL ( Sigma ) - and laminin ( Invitrogen ) -coated 12 mm glass coverslips ( Bellco ) in a 24-well plate at 150 , 000 cells/well in Neurobasal media ( NB ) ( Invitrogen ) supplemented with 10% fetal bovine serum ( FBS ) ( Invitrogen ) , 1% penicillin/streptomycin ( P/S ) ( Sigma ) , and 1% L-glutamine ( Q ) ( Sigma ) and incubated at 37°C/5% CO2 in a humidified incubator . Three to four hours after plating , the media was changed to NB supplemented with 2% B27 ( Sigma ) , 1% P/S , and 1% Q . Every 3–4 days , half of the media was removed and replaced with fresh NB+B27+P/S+Q media . Dissociated cortical neurons were transfected at 5 days in culture ( DIV ) using calcium phosphate as previously described ( Flavell et al . , 2006 ) . For synaptic staining experiments , neurons were transfected with myristolated-GFP and incubated till DIV 17–18 . Dissociated cortical neurons were transfected at 5 DIV using calcium phosphate as previously described ( Flavell et al . , 2006 ) . Cultures were stimulated for 5 hr using 60 mM KCl 42–48 hrs post-transfection and harvested for dual luciferase activity ( Promega ) . For the luciferase assay , MEF2-response element ( MRE ) -firefly luciferase activity was divided by TK-renilla luciferase activity to control for transfection efficiency in each independent well . To test effects of MEF2C-EN and MEF2C-VP16 on MEF2-dependent transcription , P0 cortical cultures were co-transfected using calcium phosphate method at DIV4 with a myristolated GFP , 3XMRE-mCherry , and either MEF2C-EN , MEF2C-VP16 , or an empty vector control . After 72 hr , cultures were fixed with 4% PFA/Sucrose and imaged for mCherry expression . To quantify normalized MEF2 activity in each transfected neuron , 3XMRE-mCherry fluorescence intensity was divided by GFP signal intensity . Image acquisition and quantification were performed in a blinded manner . Sixteen-bit images of neurons were acquired on a Leica SP8 confocal microscope using a 63x objective . Within each experiment , images were acquired with identical settings for laser power , detector gain , and amplifier offset . Images were acquired as a z-stack ( 5–20 optical sections and 0 . 5 µm step size ) . Maximum intensity projections were created from each stack . For GAD65/GABAAR γ2 experiments , synapse density was quantified as the overlap of GFP , α-GAD65 ( Millipore MAB351 , 1:1000; RRID:AB_11214081 ) and α-GABAAR γ2 ( Millipore AB5559 , 1:100; RRID:AB_177523 ) staining using a custom ImageJ macro ( see Source code 1 , 2 and 3 ) . For each neuron , the threshold for GFP , GAD65 and GABARγ2 was determined from the sum of the average pixel intensity and standard deviation for each independent neuron . This thresholding method was then consistently applied across all images within the experiment . A binary mask including all pixels above the threshold was created for all channels for each image and the 'Analyze particles' function was used to determine regions of triple co-localization at least one pixel in size . To calculate synapse density , this number was divided by the area of the neuron as measured using the GFP mask minus the cell body ( for dendritic synapse density ) or using the GFP mask of only the cell body ( for soma synapse density ) . Approximately 5–20 images from 2–3 separate coverslips were acquired and analyzed for each condition within an experiment for a total of at least two experiments . Synapse density values within each experiment were normalized to account for the variation in antibody staining and neuronal density from experiment to experiment . Within an experiment , the average synapse density value was obtained for the control and for experimental conditions . The normalized value of each experiment is the average experimental value divided by the average control value . For experiments involving quantification of inhibitory synapse density onto interneurons specifically , the observation of GAD65 expression in the cell body was used to identify and distinguish these interneurons from other excitatory cell types . Statistical significance was determined by t-test by using GraphPad Prism ( RRID:SCR_002798 ) . Error bars denote standard error . Spine density was quantified as the number of spines divided by the length of a 15–20 µm stretch of dendrite . Spines from at least one secondary and one tertiary dendrite per image were manually counted using the 'Cell Counter' function in ImageJ software ( NIH; RRID:SCR_003070 ) . Statistical significance was determined by unpaired t-test by using GraphPad Prism ( RRID:SCR_002798 ) . Error bars denote standard error . Images were acquired in a similar manner as described previously , except images were taken using a 20x objective . Maximum Intensity projections were generated from each stack and morphology was assessed using the 'Concentric Circles' plugin for ImageJ ( NIH ) . The parameters for concentric circles plugin were set to generate 11 concentric circles at a line width of 1 . 0 . X and Y values were set at the center of the soma of the transfected neuron and the inner radius and outer radii were calculated to produce a distance of 10 µm between circles . Dendritic morphology was then determined by manually counting the number of dendrite intersections per circle . The number of dendrite intersections per circle for each neuron within either the control group or experimental group were averaged together to generate an average number of intersections per radius for either control or experimental condition . Statistical difference was determined by using a two-way ANOVA with repeated measures in GraphPad Prism ( RRID:SCR_002798 ) . Error bars denote standard error . Acute neocortical slices of somatosensory , or 'barrel' cortex , were prepared from male or female MEF2Cfl/fl or MEF2Cfl/fl; CreEmx1 littermates from age P20-25 ( 3 week ) and bred on a mixed SVeV-129/C57BL/6J background . Mice were anesthetized with an I . P . injection of Ketamine ( 125 mg/kg ) /Xylazine ( 25 mg/kg ) and the brain removed . Coronal slices , 250–300 μm thick , were prepared in partially frozen dissection buffer consisting of ( in mM ) : 110 choline chloride , 2 . 5 KCl , 1 . 25 Na2H2PO4 , 25 NaHCO3 , 25 D-glucose , 3 . 1 Na pyruvate , 11 . 6 Na ascorbate , 1 kynurenate , 7 MgCl2 , and 0 . 5 CaCl2 , aerated with 95% O2 and 5% CO2 prior to and during the slicing procedure . Slices for some experiments were prepared in 4°C dissection buffer consisting of ( in mM ) : 75 sucrose , 87 NaCl , 3 KCl , 1 . 25 NaH2PO4 , 7 MgSO4 , 26 NaHCO3 , 20 dextrose , and 0 . 5 CaCl2 , aerated with 95% O2 and 5% CO2 . All solutions were pH 7 . 4 . Genotypic differences using these different dissection solutions were the same so the results were pooled . For experiments in animals aged ≥P21 , the mice were transcardially perfused with dissection buffer containing 1 mM kynurenic acid . Slices were then transferred to a 300 mOsM artificial cerebrospinal fluid ( ACSF ) solution containing in mM: 125 NaCl , 2 . 5 KCl , 1 . 25 Na2H2PO4 , 25 NaHCO3 , 10 D-glucose , 1 kynurenic acid , 2 MgCl2 , and 2 CaCl2 , to recover at 35°C for 25 min , and then transferred to room temperature ( ~21°C ) for 30 min prior to recording . Whole-cell recordings were performed in layer 2/3 neurons ( resting Vm < −50mV , input resistance > 80 MΩ ) centered above a barrel hollow , and cells were targeted with IR-DIC optics in an Olympus FV300 microscope . Recordings were performed at room temperature . Data were collected with a 10 kHz sampling rate and a 3 KHz Bessel filter . All data were analyzed using unpaired t-tests or 2-way ANOVA , as indicated , using GraphPad Prism ( RRID:SCR_002798 ) . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , ****p<0 . 0001 . Tissue from the somatosensory cortex of adult male mice were rapidly dissected and frozen at −80°C . Samples were thawed in TRIzol ( Invitrogen ) , homogenized , and processed according to manufacturer’s protocol . Total RNA was reverse-transcribed using Superscript III ( Invitrogen ) with random hexamers . Total RNA was isolated from somatosensory cortex as described above . Sequencing was performed by the Harvard Biopolymer Facility using PolyA mRNA isolation , directional RNA-seq library preparation , and the Illummina HiSeq2500 sequencer . Reads were aligned to mm9 using TopHat ( Trapnell et al . , 2009 ) ( RRID:SCR_013035 ) and Bowtie ( Langmead et al . , 2009 ) ( RRID:SCR_005476 ) . Gene counts were calculated by HTSeq package ( Anders et al . , 2015 ) ( RRID:SCR_005514 ) using the relative UCSC mm9 gtf file . Counts were normalized by RPKM ( Mortazavi et al . , 2008 ) . We applied a treatment specific RPKM filtering considering genes with RPKM values more than 0 . 5 either in treatments or control . DESeq ( Anders and Huber , 2010 ) ( RRID:SCR_000154 ) was used to detect the differentially expressed genes ( |logFC| > 0 . 3 , FDR < 0 . 05 ) . The RNA-Seq data discussed in this publication have been deposited in NCBI’s Gene Expression Omnibus ( RRID:SCR_005012 ) and are accessible through GEO Series accession number GSE87202 ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE87202 ) . We assumed that the samples were normally distributed . P-values for overlaps were calculated with hypergeometric test using a custom made R script . We retained an independent background for population size ( Allen brain expressed genes ) . P-values were subsequently adjusted for multiple comparisons using Benjamini-Hochberg FDR procedure . Two-way permutation test of 1000 was adapted to validate the overlaps . First we randomize the external gene sets ( e . g ASD , ID , SynDB , FMRP ) randomly selecting same number of genes from independent brain expressed genes list ( MGI and/or Allen brain database ) and subsequently calculating the overlap p-values . The second approach randomized the internal gene sets ( e . g . DEGs gene set ) randomly selecting same number of genes from RNA-seq expressed genes and subsequently calculating the overlap p-values . Moreover we adapted a permutation test to evaluate the detected DEG , randomizing 1000 times the RNA-seq data and recalculating the DEG . Analysis for RNA-seq were performed using custom made R scripts implementing functions and adapting statistical designs comprised in the libraries used . For all behavior tests , test mice were acclimated in a holding room outside the test room for 1 hr prior to testing . Behaviors in Mef2c cKO mice were compared to Cre-negative littermates tested on the same day . All behavioral tests were conducted using young adult male mice ( 8–12 weeks ) , except juvenile communication ( USV ) recordings . All behavior tests were conducted during the light-phase . All data are presented as mean ± SEM . All comparisons were between littermates using appropriate two-sided statistical tests ( specified in figure legends ) . We assumed that the samples were normally distributed . Outliers were determined using GraphPad’s outlier calculator and excluded from analysis . P-values were calculated with unpaired t-test ( two-tailed ) or two-way ANOVA followed with Sidak’s multiple comparisons post-hoc test using GraphPad Prism ( RRID:SCR_002798 ) , with specific test described in figure legends . All data are presented as mean ± SEM . All comparisons were between littermates using appropriate two-sided statistical tests ( specified in Figure legends ) . We assumed that the samples were normally distributed . Outliers were determined using GraphPad’s outlier calculator and excluded from analysis . P-values were calculated with unpaired t-test ( two-tailed ) or two-way ANOVA followed with Sidak’s multiple comparisons post-hoc test using GraphPad Prism , with specific test described in Figure legends . | Abnormal development of the brain contributes to intellectual disability , as well as to a number of psychiatric disorders , including schizophrenia and autism . As the brain develops , neurons establish connections with one another called synapses , which are either excitatory or inhibitory . At excitatory synapses , an electrical signal in the first cell increases the likelihood that the second cell will also produce an electrical signal . At inhibitory synapses , electrical activity in the first cell reduces the chances of the second cell producing an electrical signal . An imbalance between excitatory and inhibitory activity is one of the factors thought to give rise to neurodevelopmental disorders . Many individuals with schizophrenia , autism or intellectual disability possess mutations in , or near , a gene called MEF2C . This gene , which is active in both excitatory and inhibitory neurons , encodes a protein that regulates the activity of many other genes during brain development . Harrington , Raissi et al . therefore hypothesized that alterations in MEF2C might predispose individuals to neurodevelopmental disorders by disrupting the balance of excitatory and inhibitory synapses in the developing brain . To test this idea , Harrington , Raissi et al . generated mice that lack the Mef2c gene in a large proportion of their neurons throughout development . As predicted , the animals showed an imbalance of excitatory and inhibitory synapses in the brain’s outer layer , the cortex . They also displayed changes in behavior like those seen in autism . These included a loss of interest in social interaction and a reduction in vocalizations , suggesting impaired communication . Other behavioral changes included hyperactivity , repetitive movements and severe learning impairments: all features commonly observed in human neurodevelopmental disorders . The next challenge is to understand when , where and how MEF2C acts in the cortex to shape the balance of excitatory and inhibitory connections . Once this is known , further studies can test whether disrupting these processes leads directly to behaviors characteristic of autism , schizophrenia and intellectual disability . This may lead to the development of new drugs that can reverse or improve the symptoms of these conditions in affected individuals . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | MEF2C regulates cortical inhibitory and excitatory synapses and behaviors relevant to neurodevelopmental disorders |
Bre1 , a conserved E3 ubiquitin ligase in Saccharomyces cerevisiae , together with its interacting partner Lge1 , are responsible for histone H2B monoubiquitination , which regulates transcription , DNA replication , and DNA damage response and repair , ensuring the structural integrity of the genome . Deletion of BRE1 or LGE1 also results in whole chromosome instability . We discovered a novel role for Bre1 , Lge1 and H2Bub1 in chromosome segregation and sister chromatid cohesion . Bre1’s function in G1 and S phases contributes to cohesion establishment , but it is not required for cohesion maintenance in G2 phase . Bre1 is dispensable for the loading of cohesin complex to chromatin in G1 , but regulates the localization of replication factor Mcm10 and cohesion establishment factors Ctf4 , Ctf18 and Eco1 to early replication origins in G1 and S phases , and promotes cohesin subunit Smc3 acetylation for cohesion stabilization . H2Bub1 epigenetically marks the origins , potentially signaling the coupling of DNA replication and cohesion establishment .
Bre1 is a conserved E3 ubiquitin ligase containing a C3HC4 zinc-finger RING domain at its C-terminus , which forms a complex with Lge1 and associates with the E2 ubiquitin-conjugating enzyme Rad6 to mediate histone H2B monoubiquitination ( H2Bub1 ) on lysine 123 ( H2BK123 ) in Saccharomyces cerevisiae ( Hwang et al . , 2003; Robzyk et al . , 2000; Wood et al . , 2003 ) . H2Bub1 is one of the histone posttranslational modifications that has been implicated in diverse cellular functions , including: transcription regulation ( Fleming et al . , 2008; Minsky et al . , 2008; Pavri et al . , 2006; Sansó et al . , 2012 ) that is mediated through cycles of ubiquitination and deubiquitination ( Henry et al . , 2003; Osley , 2006 ) and by cross-talk effects on histone H3 methylation on residues K4 and K79 ( Briggs et al . , 2002; Dover et al . , 2002; Nakanishi et al . , 2009; Ng et al . , 2002; Sun and Allis , 2002 ) ; DNA replication progression ( Trujillo and Osley , 2012 ) ; modulation of nucleosome dynamics ( Chandrasekharan et al . , 2009; Fierz et al . , 2011 ) ; DNA double-strand breaks ( DSBs ) repair ( Chernikova et al . , 2010; Moyal et al . , 2011; Nakamura et al . , 2011; Northam and Trujillo , 2016 ) ; DSB in meiosis ( Yamashita et al . , 2004 ) ; maintenance of functional , transcriptionally active centromeric chromatin in fission yeast ( Sadeghi et al . , 2014 ) ; methylation of kinetochore protein Dam1 ( Latham et al . , 2011 ) ; apoptosis ( Walter et al . , 2010 ) ; and cell size control ( Hwang et al . , 2003; Jorgensen et al . , 2002 ) . The human homologs of yeast Bre1 , the RING-finger proteins Rnf20 and Rnf40 , form a heterodimer complex and are also required for H2Bub1 on lysine 120 ( H2BK120 ) ( Zhu et al . , 2005 ) . RNF20 and RNF40 , which are implicated as tumor suppressor genes , are mutated or misregulated in various types of cancers ( Johnsen , 2012 ) . H2Bub1 is also downregulated during tumor progression ( Thompson et al . , 2013 ) , suggesting a role for H2Bub1 in tumor suppression . In budding yeast , bre1Δ and lge1Δ mutants have been identified in multiple genome-wide screens as exhibiting structural and numerical chromosomal instability ( CIN ) phenotypes ( Yuen et al . , 2007 ) . The structural CIN phenotype involving gross chromosomal rearrangements ( GCR ) observed in bre1Δ and lge1Δ can be explained by the known functions of H2Bub1 in DNA damage response and repair , but the underlying cause of numerical CIN phenotypes involving whole chromosome gains or losses in bre1Δ and lge1Δ is currently not clear , though Bre1’s function in replication origins has been implicated in minichromosome maintenance ( Rizzardi et al . , 2012 ) . Accurate chromosome segregation requires the coordination of many cell-cycle-regulated processes , including sister chromatid cohesion , spindle assembly checkpoint , kinetochore function and centrosome function ( Yuen , 2010 ) . RNF20 was one of the five human homologs of yeast CIN genes that are somatically mutated in colorectal cancers ( Barber et al . , 2008 ) . The other four genes regulate sister chromatid cohesion , affecting cohesin subunits SMC1–SMC1L1 , SMC3–CSPG6 , SCC3–STAG3 and cohesin-loading complex subunit SCC2–NIPBL , implying that cohesion gene mutations are enriched in colorectal cancers ( Barber et al . , 2008 ) . Whether RNF20 also functions in sister chromatid cohesion is unknown . Cohesion between the replicated sister chromatids is established from S phase until the onset of mitotic anaphase , which ensures that an identical set of genetic information is inherited by both daughter cells . Sister chromatid cohesion is mediated by a conserved multi-subunit ring-shaped protein complex called cohesin , which consists of four subunits: the coiled-coil proteins Smc1 and Smc3 are linked by the globular SMC hinge domains at one end , at the other end , the ATPase head domains bind to Scc1–Mcd1–Rad21–Klesin together with Scc3 ( Haering et al . , 2002 , 2004; Michaelis et al . , 1997; Tóth et al . , 1999 ) . Cohesin is proposed to hold DNA topologically ( Haering et al . , 2008 ) . The cohesin complex is loaded onto chromosomes in late G1 by the cohesin-loading complex Scc2–Scc4 ( Ciosk et al . , 2000 ) through opening of the SMC hinge region ( Gruber et al . , 2006; Nasmyth , 2011 ) . In budding yeast , cohesin preferentially accumulates between convergently transcribed genes and at centromeres ( Lengronne et al . , 2004; Tanaka et al . , 1999 ) . Establishment of sister chromatid cohesion during S phase requires an essential acetyltransferase , Eco1/Ctf7 , which acetylates the cohesin subunit Smc3 at K112 and K113 ( Rolef Ben-Shahar et al . , 2008; Skibbens et al . , 1999; Tanaka et al . , 2000; Tóth et al . , 1999; Unal et al . , 2008 ) to inhibit cohesin’s interaction with the Wpl1–Pds5 complex , which destabilizes the cohesin on chromatin ( Rolef Ben-Shahar et al . , 2008; Kueng et al . , 2006; Rowland et al . , 2009; Sutani et al . , 2009; Terret et al . , 2009 ) . In addition , two non-essential cohesion establishment pathways , including Ctf4 and Ctf18 , contribute to cohesion establishment ( Hanna et al . , 2001; Mayer et al . , 2001 ) . Cohesion can no longer be established once replication is complete ( Uhlmann and Nasmyth , 1998 ) , except during DSBs in G2 , when cohesin is recruited to DSBs for Eco1-dependent cohesion establishment and efficient break repair by homologous recombination ( HR ) ( Ogiwara et al . , 2007; Ström et al . , 2004 , 2007; Unal et al . , 2007 ) . The destruction of cohesion at the onset of anaphase is mediated by separase-induced proteolysis of Scc1 , thereby triggering the segregation of sister chromatids ( Nasmyth and Haering , 2009; Peters et al . , 2008 ) . Emerging evidence suggests that establishment of cohesion between sister chromatids is coupled to replication fork progression . A number of replication proteins , including the replication factor C ( RFC ) core subunit Rfc4 ( Mayer et al . , 2001 ) , the DNA sliding clamp Proliferating Cell Nuclear Antigen ( PCNA ) ( encoded by POL30 ) ( Lengronne et al . , 2006; Moldovan et al . , 2006 ) , the helicase Chl1 involved in processing Okazaki fragments ( Samora et al . , 2016; Skibbens , 2004 ) , the leading-strand DNA polymerase ε ( Edwards et al . , 2003 ) , the replication checkpoint proteins Tof1 and Csm3 ( Mayer et al . , 2004 ) , and subunits of the origin recognition complex ( ORC ) subunits Orc2 and Orc5 ( Shimada and Gasser , 2007; Suter et al . , 2004 ) play important roles in sister chromatid cohesion . In turn , cohesion establishment factors localize to replication forks , affecting fork progression and stability ( Gambus et al . , 2009; Lengronne et al . , 2006; Terret et al . , 2009 ) . Smc3 acetyltransferase Eco1 associates with the replication fork through PCNA ( Moldovan et al . , 2006; Skibbens et al . , 1999 ) . Ctf18 , a component of the replication factor C ( RFCCtf18 ) complex , can load and unload PCNA ( Bylund and Burgers , 2005; Lengronne et al . , 2006; Mayer et al . , 2001; Murakami et al . , 2010; Shiomi et al . , 2007; Terret et al . , 2009 ) and physically interacts with Eco1 ( Kenna and Skibbens , 2003 ) . The localization of Ctf18 at replication origins partially depends on Ctf4 ( Lengronne et al . , 2006 ) . Ctf4 is a component of the replisome progression complex ( RPC ) ( Gambus et al . , 2006 ) that recruits DNA polymerase α ( Polα ) /primase for lagging-strand synthesis ( Gambus et al . , 2009; Zhu et al . , 2007 ) and recruits Chl1 helicase ( Samora et al . , 2016 ) to the replisome through its physical association with the GINS ( go-ichi-ni-san ) complex . This complex is part of the Cdc45–Mcm2-7–GINS ( CMG ) helicase complex that is important for origin unwinding , establishment of the replication fork at origins and fork progression ( Gambus et al . , 2009 ) . Ctf4 in turn depends on GINS and the replication factor Mcm10 for its localization ( Perez-Arnaiz et al . , 2016; Terret et al . , 2009; Wang et al . , 2010; Zhu et al . , 2007 ) . Mcm10’s localization to origins is facilitated by the presence of inactive Mcm2-7 complex ( Douglas and Diffley , 2016; Ricke and Bielinsky , 2004; Wohlschlegel et al . , 2002 ) . Next , Mcm10 recruits Cdc45 and GINS to inactive Mcm2-7 complex , and activates the CMG replicative helicase ( Perez-Arnaiz et al . , 2016; Quan et al . , 2015; Thu and Bielinsky , 2014 ) . Thus , Mcm10 is crucial in replication initiation and elongation . Previous work has shown that H2Bub1 is not required for the association of pre-replication complex ( ORC and Mcm4 ) and Cdc45 with origins in G1 phase , but is required for Mcm4 , Cdc45 , Psf2 ( a component of GINS ) , Polα , Polε , RPA and Spt16’s chromatin association in S phase , both for replisome stability and for nucleosome assembly onto nascent DNA at active replication forks ( Trujillo and Osley , 2012 ) . Whether Bre1 and H2Bub1 could affect sister chromatid cohesion through its function in DNA replication has not been explored . Here we show that Bre1 RING-domain- and Lge1-mediated H2Bub1 is critical for accurate chromosome segregation , and specifically sister chromatid cohesion . Bre1’s role in G1 and S phase contributes to cohesion establishment , but it is dispensable for cohesin component loading . Bre1 facilitates the localization of the upstream replication factor Mcm10 and cohesion establishment factors ( Ctf18 , Ctf4 and Eco1 ) to chromatin and early replication origins in G1 and S phases . The recruitment of these factors by Bre1 not only stabilizes the replisome progression complex , advancing the replication fork in S phase ( Trujillo and Osley , 2012 ) , but also couples the establishment of sister cohesion to maintain whole-chromosome stability .
The genome-wide CIN screens in budding yeast have revealed that the E3 ubiquitin ligase BRE1 and its interacting partner LGE1 are important for maintaining chromosome stability ( Yuen et al . , 2007 ) . To assess the chromosome transmission fidelity ( CTF ) of bre1Δ and lge1Δ , we monitored artificial chromosome fragment loss rate as described previously ( Spencer et al . , 1990; Yuen et al . , 2007 ) . Haploid cells in ade2–101ochre mutation background are red in color . An artificial chromosome III fragment ( CF ) with a yeast centromere and telomeres at the ends , resembling natural chromosomes’ structure and stability , was introduced to the cells . The CF also contains a selectable marker and the SUP11 tRNA suppressor gene , which suppresses the ochre mutation ( Spencer et al . , 1990 ) . Cells containing the CF are white , whereas cells that have lost the CF are red . Thus , red sectors within a white colony indicate that the CF is lost in some mitoses during the formation of the colony . To quantify the CF loss rate per cell division , individual cells from selective medium were plated onto non-selective medium , and the percentage of colonies that were half-red or more than half-red on non-selective medium , representing cells that have lost the CF during the first cell division , was calculated ( Figure 1A ) . Consistent with prior work ( Yuen et al . , 2007 ) , deletion of BRE1 or LGE1 resulted in significant CF loss rates ( 0 . 78% and 0 . 85% , respectively ) ( Figure 1A ) , which were 5 . 7- and 6 . 3-fold higher than that in wildtype cells ( 0 . 13% ) , suggesting a role for Bre1–Lge1 in accurate chromosome segregation . To dissect whether chromosome loss arises from a cohesion defect in BRE1- or LGE1-deleted cells , we utilized a MATa haploid strain containing Lac operator tandem repeats integrated 22 kb from the centromere of chromosome III and expressing a GFP–Lac repressor fusion protein as described by Straight et al . ( 1996 ) . The separation of the two sister chromatids can be visualized by the GFP signals during G2/M phase . If sister chromatid cohesion is normal , only one GFP focus can be observed due to the tight tethering of replicated sister chromatids by cohesion . However , two GFP foci can be seen if the sister chromatids prematurely separate ( Figure 1B ) . Only 3 . 5% of WT cells with a large bud ( 60–75 min release from G1 arrest by alpha-factor [α-F] ) had two GFP signals , and fluorescence-activated cell sorter ( FACS ) analysis showed that these large budded cells have replicated DNA content ( Figure 1C ) . By contrast , bre1Δ and lge1Δ showed significant increases in the frequency of G2/M cells containing two GFP signals ( 20 . 2% and 19 . 9% , respectively ) ( Figure 1D ) . Alpha-factor ( α-F ) was added back at 60 min post G1 release to arrest cells in the next G1 phase , and mitotic anaphase and cytokinesis finished by 120–150 min after G1 release ( Figure 1C ) . Each G1 cell should contain one GFP focus if chromosomes are accurately segregated , whereas cells with missegregated chromosome III may have no GFP focus or two GFP foci ( Figure 1B ) . In WT cells , 99% of cells contained only one GFP dot , indicative of proper chromosome separation . By contrast , ~9% of bre1Δ or lge1Δ cells had two or no GFP signals , indicating a gain or loss of chromosome ( Figure 1E ) . Together , these results demonstrate that the chromosome missegregation phenotype exhibited by BRE1 or LGE1 deletion could be caused by the defect in sister chromatid cohesion , though we cannot rule out the possibility that other cellular functions of Bre1 could also contribute to whole-chromosome stability . The cohesin ring complex is loaded onto chromosomes in late G1 , whereas sister chromatid cohesion is established in S phase , and maintained through G2/M before anaphase ( Mehta et al . , 2012 ) . To determine the cell-cycle stages at which Bre1 is important for sister chromatid cohesion , we exploited the Auxin-Inducible Degron ( domain for inducing degradation ) ( AID ) system ( Nishimura et al . , 2009 ) to conditionally control the expression of Bre1 at specific cell-cycle stages and examined the cohesion phenotype in G2/M-arrested cells . We constructed an AID*−9Myc tag at the C-terminus of Bre1 at the endogenous locus and expressed F-box Transport Inhibitor Response 1 ( TIR1 ) from rice Oryza sativa ( OsTIR1 ) in a MATa haploid strain containing the GFP–LacI:LacO system for cohesion assay . Auxin binds to AID* and OsTIR1 , which interacts with the E3 ubiquitin ligase SCF ( Skp1 , Cullin and F-box ) and targets AID*−9Myc-tagged Bre1 for polyubiquitination and proteasome-mediated degradation . AID*−9Myc-tagged Bre1 and OsTIR expression did not affect growth and sister chromatid cohesion function ( Figure 2A and B ) , but affected G1-S transition cyclin gene expression mildly ( Figure 2—figure supplement 1A ) ( Zimmermann et al . , 2011 ) and hydroxyurea ( HU ) sensitivity ( Figure 2—figure supplement 1B ) . To assess the efficiency of auxin-induced Bre1-AID*−9Myc degradation , we monitored the degradation time course after treatment with 1 mM auxin in combination with G1 , S and G2/M arrest . First , Bre1-AID*−9Myc-expressing cells were synchronized in early G1 phase by adding α-F for 3 hr , and then released into media containing α-F or HU for G1 or S phase arrest , respectively , together with 1 mM auxin to induce Bre1 degradation . Alternatively , to induce Bre1 degradation in G2/M phase , G1-arrested cells were released into HU-containing medium for 2 hr , and then released into Nocodazole ( Noc ) -containing medium with the addition of 1 mM auxin . Samples were collected every 15 min for FACS analysis of DNA content ( Figure 2C ) and for western blotting analysis of Bre1 protein level ( Figure 2D ) . Specifically , over 90% of Bre1 was degraded after 45 min in G1 , 75 min in S and 60 min in G2/M phase after auxin induction ( Figure 2D ) . To examine the timing of Bre1’s function in the cohesion cycle , we degraded Bre1 at specific cell-cycle stages . Once the arrest in G1 , S or G2/M was achieved by α-F , HU or Noc , respectively , with or without auxin induction , the medium was washed and released into the next cell cycle . Samples were collected every 30 min for FACS analysis of DNA content ( Figure 2E ) and western blotting analysis of Bre1 protein level to confirm the recovery of Bre1 ( Figure 2F ) . Finally , cohesion phenotype in G2/M-arrested cells was assessed to determine whether Bre1 is functional in the G1 , S , or G2/M phase . The no degradation control ( without auxin in all stages ) resulted in only 4 . 6% of cells showing a cohesion defect ( Figure 2G ) , which was comparable to rates in WT cells ( Figure 1D , Figure 5B and C ) , suggesting that no cohesion defect was caused by the effects of cell cycle arrest by different drugs . As expected , degradation of Bre1 in all cell-cycle stages showed 19 . 2% of cells with a cohesion defect , which was similar to the proportion in bre1Δ . Degradation of Bre1 in G1 only or in G1 and S phases showed similar proportions of cohesion defects ( 20 . 7% and 17 . 1% , respectively ) , which were comparable to the proportion of cells in which Bre1 was degraded in all cell-cycle stages ( 19 . 2% ) , suggesting that Bre1’s role in G1 phase contributes to sister chromatid cohesion . To our surprise , degradation of Bre1 in S phase only or in S and G2/M phases resulted in fewer cells with cohesion defects ( 12 . 4% and 10 . 2% , respectively ) , but still significantly more defective cells compared to control . The occurrence rate of the cohesion defect caused by degradation of Bre1 in G1 and S phases ( 17 . 1% ) was significantly higher than that in cells in which Bre1 is degraded in S only ( 12 . 4% ) . It is possible that for degradation of Bre1 in S phase , the residual level of Bre1 during early S-phase time points may suffice for some function ( Figure 2F ) . However , degradation of Bre1 in G2/M phase only resulted in an occurrence rate for cohesion defects similar to that in the the negative control , implying that Bre1 is not required in G2/M phase for cohesion . Taken together , these findings demonstrated that Bre1’s role in sister chromatid cohesion is most prominent in G1 phase , but also in S phase , consistent with the timing of cohesin loading in G1 phase or cohesion establishment in S phase , but Bre1 is not required in G2/M phase for cohesion maintenance . To further delineate the role of Bre1 in cohesion , we tested its effects on cohesin loading and cohesion establishment . Cohesin associates with chromatin in late G1 , then accumulates at regions of convergent transcription ( Lengronne et al . , 2004 ) . Bre1 regulates the transcription of genes involved in G1-S transition ( Zimmermann et al . , 2011 ) , but not that of cohesin components or coehsion establishment genes . The change in the transcription pattern of the G1-S transition genes may affect the binding of cohesin on chromatin . Next , we constructed strains expressing Scc1 or Smc3 3HA-tagged and examined their chromatin enrichment in G1- , S- and G2/M-arrested cells in WT and bre1Δ by chromatin spreads . The protein levels of Scc1-3HA and Smc3-3HA in bre1Δ were comparable to those in WT cells ( Figure 3—figure supplement 1B and D ) , consistent with the mRNA level result ( Figure 3—figure supplement 1A ) . As expected and consistent with the study by Baetz et al . ( 2004 ) , there was no detectable Scc1-3HA and Smc3-3HA signal in α-F-arrested early G1 phase WT or bre1Δ cells . The cohesin components were associated with chromatin in S and G2/M-arrested cells in WT cells ( Prinz et al . , 1998 ) , which were unaffected by the deletion of BRE1 ( Figure 3—figure supplement 1C , E , F and G ) . As cohesin subunit Scc1 is transiently enriched at active early replication origins and spreads along DNA as replication fork progresses ( Tittel-Elmer et al . , 2012 ) , and as Bre1 is also present at origins of replication throughout the cell cycle ( Trujillo and Osley , 2012 ) , we tested whether Bre1 affects the enrichment of Scc1 at early origins in HU-arrested S phase cells using chromatin immunoprecipitation followed by quantitative real-time PCR ( ChIP-qPCR ) . In agreement with the chromatin spreads result , the enrichment of Scc1–3HA at early origins ( ARS305 and 306 ) was not affected by BRE1 deletion ( Figure 3—figure supplement 1H ) . Therefore , our results suggest that Bre1 , like cohesion establishment factor Eco1/Ctf7 ( Tóth et al . , 1999 ) , is dispensable for cohesin binding to chromatin . To confirm the physical interaction of Bre1 with Smc3 in a yeast two-hybrid experiment ( Newman et al . , 2000 ) , we performed co-immunoprecipitation in a strain containing Flag–Bre1 and Smc3–13Myc using anti-Myc antibody to immunoprecipitate Smc3–13Myc . We cannot , however , confirm interaction between Bre1 and Smc3 ( Figure 3—figure supplement 1I ) . Our Bre1 degron mutant experiment suggested that Bre1 is important in G1 and S phases , and may be involved in sister chromatid cohesion establishment in S phase . To elucidate whether Bre1 affects the recruitment of cohesion establishment factors Ctf4 , Ctf18 and Eco1 to chromatin , chromatin spread was analyzed at different arrested cell cycle stages , and we found that Ctf4 , Ctf18 and Eco1 associate with chromatin during all stages of the cell cycle in WT cells ( Figure 3A–C and Figure 3—figure supplement 2A–F ) . This observation is consistent with previous studies ( Hanna et al . , 2001; Tóth et al . , 1999 ) , but our chromatin spread assay did not detect the degradation of Eco1 in G2/M phase as shown in a previous study ( Lyons and Morgan , 2011 ) . Nevertheless , the association of Ctf4 and Ctf18 with chromatin in S phase was significantly reduced in bre1Δ cells , whereas this association was not affected in G1 and G2/M phases ( Figure 3A–C ) . The association of Eco1 with chromatin was reduced in bre1Δ cells in both G1 and S phases . As Ctf4 , Ctf18 and Eco1 also localize at early replication origins and at forks in HU-treated early S-phase cells ( Lengronne et al . , 2006 ) , and as Bre1 and H2Bub1 are present at origins ( Trujillo and Osley , 2012 ) , we hypothesized that Bre1 is required for the recruitment of Ctf4 , Ctf18 and Eco1 to early origins . Their occupancy at origins was measured by ChIP in HU-arrested cells . The occupancy of Ctf4 , Ctf18 and Eco1 at early origins ARS305 and ARS306 , and of Eco1 at early origin flanking regions was reduced significantly in bre1Δ compared to WT ( Figure 3D–G ) . In addition , the occupancy of Eco1 at early origin was also reduced significantly in G1 cells ( Figure 3H and Figure 3—figure supplement 2G ) . Ctf4 , Ctf18 and Eco1 were present at lower , but comparable , levels at a late origin ARS501 in WT and bre1Δ cells . The decreased occupancy of cohesion establishment factors at chromatin or early origin is not due to any change in mRNA or protein levels ( Figure 3—figure supplement 1A and Figure 3—figure supplement 2A–C ) . Our results suggest that Bre1 facilitates the localization of these cohesion establishment factors to the chromatin , and specifically to the early origins . In the absence of Bre1 , however , a proportion of cohesion factors remained on chromatin and at origins , suggesting that there are probably redundant pathways that function in their recruitment . Since Ctf18 and Eco1 are both required for the acetylation of Smc3 during S phase to generate a stably chromosome-bound cohesin pool for enduring sister chromatid cohesion ( Beckouët et al . , 2010; Terret et al . , 2009 ) , we postulated that Bre1 may also affect Smc3 acetylation in S phase . We monitored acetylated Smc3 in WT and bre1Δ cells by western blotting using anti-Smc3-Ac antibody . In the absence of Bre1 , Smc3 acetylation was significantly diminished , whereas the protein levels of Smc3-13Myc in WT and bre1Δ cells were similar ( Figure 3I ) . These results suggest that Bre1 affects the acetylation of Smc3 , but not the protein level of Smc3 or its association with chromatin . Collectively , we showed that despite the fact that Bre1 is dispensable for cohesin association to chromatin in S phase , it is important for recruiting cohesion establishment factors to the replication origins in G1 and S phases , and for Smc3 acetylation . However , if Bre1’s only function in cohesion is to facilitate Smc3 acetylation , removal of wpl1Δ , the destabilizer of chromatin-bound cohesin , may rescue bre1Δ’s cohesion defect , as for ctf18Δ ( Borges et al . , 2013 ) . Surprisingly , wpl1Δ partially rescues bre1Δ’s cohesion defect at 60–75 min after G1 release , suggesting that Bre1 could play a role in reducing cohesion turnover on chromatin , counteracting Wpl1 ( Figure 3—figure supplement 2H ) . Alternatively , it may reflect a slight delay in the cell cycle progression of wpl1Δ bre1Δ ( Figure 3—figure supplement 2I ) . However , wpl1Δ has 9 . 2% of cells with a cohesion defect at 90 min after G1 release , whereas wpl1Δ bre1Δ has 18 . 8% , similar to the proportion in bre1Δ alone . This suggests that stabilizing cohesin alone on chromatin does not fully rescue the cohesion defect in bre1Δ as it does in ctf4Δ ( Borges et al . , 2013 ) . To investigate the genetic relationship between BRE1 and the non-essential cohesion establishment factors CTF4 and CTF18 ( Hanna et al . , 2001; Mayer et al . , 2001; Xu et al . , 2007 ) , we monitored the cohesion phenotypes in single and pair-wised double mutants . The cohesion defect in bre1Δ or lge1Δ was slightly less severe than that in ctf4Δ ( 23 . 7% ) or ctf18Δ ( 28 . 7% ) ( Figure 3J ) . The cohesion phenotype difference in single mutants is consistent with the chromatin association results , suggesting that other parallel pathways independent of bre1Δ can help to recruit Ctf4 and Ctf18 . Interestingly , cohesion defects in the double mutants , ctf18Δ bre1Δ ( 23 . 5% of cells ) , ctf4Δ lge1Δ ( 19 . 9% ) and ctf4Δ bre1Δ ( 23 . 9% ) , were not significantly more frequent than those in the more severe single mutant of ctf4Δ and ctf18Δ , suggesting that BRE1 and LGE1 epistatically interact with CTF4 and CTF18 , and that BRE1 and LGE1 may function upstream of both CTF4 and CTF18 genetic pathways in cohesion , affecting their recruitment partially . We hypothesized that Bre1 may signal to replication factors upstream of Ctf4 and Ctf18 to affect their localizations . Indeed , Bre1-mediated H2Bub1 is required for the association of some replisome proteins ( Polα , Polε and RPA ) ( Sun et al . , 2016 ) and replisome progression complex ( RPC ) components ( Psf2 and Spt16 ) ( Gambus et al . , 2006 ) with early origins , and for stable replication fork progression ( Trujillo and Osley , 2012 ) . Psf2 is required for Ctf4 localization at origin of replications ( Gambus et al . , 2006 ) , and in turn , Ctf4 is required for Polα localization ( Gambus et al . , 2009; Zhu et al . , 2007 ) . Thus , we attempted to search for replication factors more upstream than Psf2 that are regulated by Bre1 . Yet , H2Bub1 is dispensable for the loading at origins of the most upstream origin recognition complex ( ORC ) component Orc2 ( Trujillo and Osley , 2012 ) . Previous studies showed that both the loading of GINS , Polα and Ctf4 onto chromatin and CMG helicase activation depends on replication initiation and elongation factor Mcm10 ( Perez-Arnaiz et al . , 2016; Quan et al . , 2015; Ricke and Bielinsky , 2004; Zhu et al . , 2007 ) . Therefore , we asked whether Bre1 affects the recruitment of Mcm10 . We verified that Bre1 , like H2Bub1 , is required for the association of Psf2 and Polα with chromatin and origins ( Trujillo and Osley , 2012 ) , and tested whether Mcm10 recruitment is also affected . Chromatin spreads showed that Psf2 , Polα and Mcm10 associate with chromatin in G1 , S and G2/M phases in WT cells ( Figure 4A–C and Figure 4—figure supplement 1D–F ) , consistent with previous reports ( Falconi et al . , 1993; Gambus et al . , 2006 ) . However , the levels of Psf2 , Polα and Mcm10 associated with chromatin in bre1Δ cells in G1 and S phases were significantly reduced , whereas those in G2/M phase was not affected . The reduced chromatin association of Psf2 , Polα and Mcm10 in bre1Δ cells was not the consequence ofreduced protein expression levels , as confirmed by western blotting analysis ( Figure 4—figure supplement 1A–C ) . ChIP-qPCR further showed the occupancy of Psf2 at early replication origins in bre1Δ cells in S phase , the occupancy of Mcm10 at early origins in G1 and S phase , and the occupancy of Mcm10 at early origin-flanking regions in S phase were all significantly reduced , but that at a late origin their occupancy levels remained low but comparable to those in WT and bre1Δ ( Figure 4D–G ) ( Sekedat et al . , 2010 ) . Interestingly , co-immunoprecipitation showed that Bre1 interacts weakly with Mcm10 ( Figure 4—figure supplement 1G ) . However , Bre1 is dispensable for Mcm10 diubiquitination ( Mcm10[Ub]2 ) ( Figure 4—figure supplement 1H ) . Collectively , we demonstrated that Bre1 plays a role in regulating the localization of an upstream replication factor , Mcm10 , which is important for CMG activation , replication initiation at origins , replication fork progression , and recruitment of replication-coupled cohesion establishment factors . E3 ubiquitin ligase Bre1 is known to function with E2 ubiquitin conjugating enzyme Rad6 to monoubiquitinate histone H2B at K123 ( H2Bub1 ) through the conserved RING domain of Bre1 ( Hwang et al . , 2003; Robzyk et al . , 2000 ) . We tested whether Rad6 plays a role in cohesion as Bre1 does , and whether Bre1 functions in cohesion through its ubiquitin ligase activity on its target H2Bub1 . We examined rad6Δ , and made use of the Bre1 RING domain truncation mutant bre1-RINGΔ , which lacks E3 activity and cannot ubiquitinate H2B as shown in a prior study ( Hwang et al . , 2003 ) ( Figure 5A ) . In addition , we constructed a H2B mutant that cannot be ubiquitinated . To do this we used K123R point mutation in one of the H2B genes , HTB1 , and deletion of another H2B gene HTB2 ( htb1K123R htb2Δ ) as described in prior work ( Robzyk et al . , 2000 ) . We then performed the cohesion assay in G2/M-arrested cells by adding nocodazole for 3 hr . The cohesion defect in rad6Δ ( 19 . 6% of cells ) , bre1-RINGΔ ( 18 . 0% ) and htb1-K123R htb2Δ ( 17 . 9% ) occured at frequencies comparable to that in bre1Δ in nocodazole ( 18 . 8% ) ( Figure 5B ) and that in bre1Δ in 60–75 min post G1 arrest and release ( 20 . 2% , Figure 1D ) . On the other hand , htb2Δ showed a less frequent but significant cohesion defect ( 10 . 0% of cells ) , and Flag-HTB1 ( 7 . 3% ) showed low levels of cohesion defect comparable to those of WT controls ( 5 . 9% ) ( Figure 5B ) . These results suggest that Rad6 and Bre1-catalyzed H2Bub1 accounts for Bre1’s function in sister chromatid cohesion . By contrast , deletion of the other known substrate of Bre1 , swd2Δ ( 5 . 6% ) ( Vitaliano-Prunier et al . , 2008 ) , showed only a WT level of cohesion defect . The cohesion defect in the bre1Δ lge1Δ double mutant in nocodazole-arrested G2/M cells ( ~17 . 4% of cells ) was similar to that in single bre1Δ or lge1Δ ( 18 . 8% or 18 . 3% of cells , respectively ) ( Figure 5B ) , consistent with Bre1 and Lge1 functioning together in a complex . To distinguish the role of Rad6 , the Bre1 RING domain and H2Bub1 in sister chromatid cohesion establishment in S phase versus maintenance in G2/M phase , we repeated the G1 arrest and release cohesion assay over a time course ( Figure 5C , as in Figure 1B and C ) , and confirmed that the WT has a low percentage of premature sister chromatid separation up to 90 min after release from G1 . On the other hand , rad6Δ , bre1Δ , bre1-RINGΔ , and htb1-K123R htb2∆ have progressively elevated frequencies of premature sister chromatid separation from 30 to 90 min after release from G1 , shortly after DNA replication begins ( Figure 5—figure supplement 1A ) , which are comparable to frequencies seen in cohesion establishment factor mutants such as ctf18∆ ( Hanna et al . , 2001 ) . Consistent with the cohesion defect shown by bre1-RINGΔ and htb1K123R htb2Δ ( Figure 5B and C ) , we found that these two mutants also reduce the association of cohesion establishment factor Ctf4 and replication factor Mcm10 at early origins in S phase ( Figure 5D–G ) . These results further suggest that the role of Bre1 in cohesion is through its catalytic RING domain and H2B monoubiquitination . As Bre1 stability has been shown to be affected by its RING domain’s catalytic activity , and by its ubiquitination level at H2BK123 ( Wozniak and Strahl , 2014 ) , we checked the endogenously tagged Bre1 protein level in WT , bre1-RINGΔ and htb1-K123R htb2Δ cells . Consistent with previous findings , Bre1 protein level was reduced in the bre1-RINGΔ mutant . Thus , the effect of a bre1-RINGΔ mutation on cohesion could be due to either the loss of Bre1’s E3 catalytic activity or reduced Bre1 level . However , Bre1 protein level was comparable in htb1-K123R htb2Δ and WT ( Figure 5—figure supplement 2A and B ) , suggesting that Bre1-mediated H2B monoubiquitination , but not Bre1 protein level , accounts for the cohesion defect in htb1-K123R htb2Δ . As Bre1 and H2Bub1 are required for the recruitment of Ctf4 and Mcm10 , and as Ctf4 interacts with Mcm10 ( Wang et al . , 2010 ) , we checked whether bre1∆ disrupts this interaction . By performing co-immunoprecipitation between Ctf4 and Mcm10 , we found that bre1∆ mutation does not affect the interaction of Ctf4 with Mcm10 ( Figure 5H ) .
The conserved E3 ubiquitin ligase Bre1 , responsible for H2B monoubiquitination , contributes to structural chromosome integrity , which is well evidenced by its characterized roles in DNA replication , transcription , DNA damage response and repair processes through modulating nucleosome dynamics and histone crosstalk signaling . However , the underlying cause of whole chromosome instability ( CIN ) in BRE1-deletion mutants is not fully understood . In this study , we have identified a novel role for Bre1 , its interacting partner Lge1and H2Bub1 , catalyzed by the RING-finger domain of Bre1 , in precise chromosome segregation and sister chromatid cohesion . Whereas Bre1 is non-essential , and so the deletion mutant is viable , our degron mutant together with assays of cohesion establishment and replication factors recruitment help to pinpoint the timing of Bre1's function in cohesion to G1 and S phases . Although Bre1 is dispensable for the loading of cohesin subunits Scc1 and Smc3 onto chromatin , it facilitates the recruitment of cohesion establishment factors Ctf4 , Ctf18 and Eco1 to chromatin and to early origins in S phase to promote Smc3 acetylation . It is known that H2Bub1 is required to regulate the occupancy of active Mcm4 , Cdc45 , Psf2 and Polα at the early origins in S phase , but this protein is not required for the localization of ORC , inactive Mcm4 and Cdc45 in G1 phase , as shown in a prior study ( Trujillo and Osley , 2012 ) . Here we identified a further upstream , essential replication initiation and elongation factor , Mcm10 , which is important for CMG ( Cdc45-Mcm2-7-GINS ) helicase assembly and activation ( Perez-Arnaiz et al . , 2016 ) and whose recruitment to chromatin and early origins is at least partially affected by non-essential Bre1 . These findings are compatible with a model in which Bre1 localizes to origins and monoubiquitinates H2B ( Trujillo and Osley , 2012 ) , which acts as an upstream epigenetic mark to signal the recruitment of both the replication factors ( Mcm10 , Psf2 , and Polα ) and the cohesion establishment factors ( Ctf4 , Ctf18 and Eco1 ) to origins ( Figure 6 ) . Mcm10 , Psf2 and Polα each interacts with Ctf4 , an RPC component that is also required for sister chromatid cohesion ( Gambus et al . , 2006 , 2009; Simon et al . , 2014; Tanaka et al . , 2009; Wang et al . , 2010; Wittmeyer and Formosa , 1997; Zhu et al . , 2007 ) . The partially reduced level of chromatin-associated replication factors ( Psf2 and Mcm10 ) in BRE1 null mutant affects the localization of Ctf4 , which in turn affects the localization of Ctf18 , and thus Eco1 , leading to reduced Smc3 acetylation and resulting in defective cohesion establishment . Surprisingly , PCNA’s association at early origins was unaffected in the absence of H2Bub1 ( Trujillo and Osley , 2012 ) . This could be because the partial reduction of Ctf18’s chromatin association is not severe enough , or because PCNA can associate with origins through a Ctf18-independent pathway . Bre1 and H2Bub1 play a role in the progression of replication forks , in which these replication and cohesion establishment proteins travel together to allow cohesion establishment on replicated sister chromatids . Our findings provide new supporting evidence for the proposed model in which the establishment of cohesion is coupled with DNA replication ( Lengronne et al . , 2006; Terret et al . , 2009 ) . Whether the replication fork could slide through the existing cohesin ring , and the exact orientation of cohesin rings , is still unclear . However , as Bre1 is non-essential and as in its absencesome level of replication factors and cohesion factors still localize to chromatin and replication origins , additional factors in independent pathways possibly also contribute to the recruitment of these factors to origins . Our data suggest that Bre1’s function in recruiting replication factors to origins is important for its function in cohesion establishment . However , H2Bub1 has also been shown to function in nucleosome reassembly and is retained in newly replicated DNA ( Trujillo and Osley , 2012 ) . Whether H2Bub1 level , which doubled in replicated DNA ( Trujillo and Osley , 2012 ) , and H3K56Ac , an epigenetic mark for newly synthesized H3 which assembles upon DNA replication ( Kaplan et al . , 2008 ) , could play a direct role in signaling successful replication to the cohesion establishment pathway is unexplored . In addition , how Bre1 is temporally and spatially regulated , for example during its recruitment to origins in G1 phase and its maintenance in sisters in S phase , needs to be addressed in the future . As does the question of whether the deubiquitinisation of H2bub1 at a later stage in the cell cycle is relevant to cohesion function . As H2Bub1 is important for DSB repair ( Moyal et al . , 2011; Nakamura et al . , 2011; Yamashita et al . , 2004 ) , it will be interesting to investigate whether it is important for DSB-induced cohesion establishment . The human homologs of yeast BRE1 , RNF20 and RNF40 are mutated and misregulated in different types of cancers . Whether defective Rnf20 or Rnf40 leads to defective cohesion and CIN in human cells , and whether this contributes to tumorigenesis initiation and progression , is worth pursing in order to reveal the genetic basis of CIN in cancers .
The gene deletion strains and strains that expressed 3HA- or 13Myc-tagged proteins were generated by the PCR-based gene deletion and modification methods as described before ( Longtine et al . , 1998 ) . The yeast strains and plasmids used in this study are listed in Supplementary files 1 and 2 , respectively . To generate htb-K123R mutant strains , a fragment consisting of Flag-HTB1-K123R with the URA3 marker was amplified from WYYp30 and integrated at the endogenous HTB1 locus of YPH1343 by homologous recombination , in addition to deleting HTB2 . Flag tagged HTB1 from WYYp19 was integrated at the endogenous HTB1 locus of YPH1343 by homologous recombination as a WT control . To make the BRE1-degron strain , plasmid WYYp74 was used to amplify the fragment AID*−9Myc ( AID*: minimum functional size region [71–114 amino acids] of full-length AID [229 amino acids] ) with KanMX6 marker at the 3’ end , which was transformed into YPH1343 , generating WYYY250 . In addition , the plasmid containing OsTIR1 ( pNHK53 ) was linearized with StuI and integrated into WYYY250 , creating WYYY326 . Yeast cells were routinely grown in YPD ( 1% yeast extract , 2% peptone , and 2% dextrose ) -rich media at 30˚C . Synthetic complete ( SC ) medium lacking a specific amino acid was used for selection . An SC with limiting adenine plate was prepared as described previously ( Hieter et al . , 1985 ) . A final concentration of 400 µg/ml of G418 ( Cat#: G4185 , Formedium , England ) antibiotic was used for the selection of gene deletions and epitope tagging with KanMX6 marker . The CTF assay was performed as described previously ( Spencer et al . , 1990; Yuen et al . , 2007 ) . Briefly , WT and gene-deletion cells containing an ade2-101 ( ochre ) mutation and a SUP11-marked chromosome III fragment ( CFIII ) were picked from plates selecting for the CFIII ( SC-URA ) , and then plated onto minimal ( SD ) medium non-selective for the CFIII ( SC with 20% limiting adenine [10 µg/ml] ) at a density of ~200 colonies per plate . The plates were incubated at 30°C for 2–3 days and then placed at 4°C to facilitate red pigment development . Cells containing the CFIII were white , whereas those that had lost the CFIII were red . Therefore , a white-and-red sectored colony was observed if the CFIII is lost in some mitoses during the formation of the colonies . Colonies that were at least half red were considered as having a chromosome loss event during the first division . The loss frequency of the CFIII was calculated as the ratio of the number of over half-red colonies to the total number of colonies . At least 2000 cells were scored in each experiment , and three independent experiments were performed . G1 arrest and release cohesion assays and chromosome segregation assays were carried out as previously reported ( Straight et al . , 1996 ) with minor modifications . Basically , early log phase cultures with optical density ( OD ) at 600 nm around 0 . 2 to 0 . 4 were collected , washed with water and arrested in the G1 phase with 5 µg/ml alpha-factor ( α-F ) for 3 hr . Cells were washed with water and released into YPD medium . α-F was added back to the culture at 60 min post G1 release to restrict cells in the next cell cycle at G1 . Samples were collected every 15 min for fluorescence-activated cell sorting ( FACS ) analysis . Cohesion assays were performed on large-budded cells at 60–75 min after release from G1 arrest , at which time the majority of cells reached G2/M phase by FACS and budding index , or at 15 min-intervals between 30–90 min after release from G1 arrest . Chromosome segregation assays were carried out on unbudded cells at 120–150 min after release from G1 arrest , at which time most cells had completed cytokinesis and had no bud , and the FACS profiles showed that the majority of cells were in G1 phase . Cells at G2/M phase and G1 phase were fixed with freshly prepared 4% paraformaldehyde at room temperature for 15 min , followed by a wash with SK buffer ( a 1% potassium acetate [Kac]−1M sorbitol solution ) and centrifugation at 2000 rpm for 2 min . Pellets were resuspended in SK buffer for cohesion assessment . The nocodazole ( Noc ) -arrested cohesion assay was performed as reported previously with slight modification ( Hanna et al . , 2001 ) . Early log phase cells were harvested and washed with water and released into YPD medium containing 15 µg/ml Noc for 3 hr . Samples were collected , fixed and resuspended in SK buffer as mentioned above for the G1 arrest and release cohesion assay . Cells were imaged on a Carl Zeiss LSM 710 NLO confocal laser scanning microscope using an EC Plan-Neofluar 40x/1 . 30 Oil Ph3 M27 oil objective and a conventional FITC excitation filter . Z-stacked images were acquired ( six z-sections were acquired at 1 µm intervals ) . In cohesion experiments , at least 100 large budded cells were scored as containing one or two GFP foci . In the chromosome segregation assays , at least 100 G1 cells were scored , and data were averaged from at least three independent experiments . The Bre1-AID*−9Myc cells were arrested in G1 with 5 µg/ml alpha factor ( α-F ) for 3 hr and washed with water before splitting into three cultures . The first and second split cultures were released into YPD containing 5 µg/ml α-F to maintain G1 phase or 0 . 2 M hydroxyurea ( HU ) to arrest cells in S phase , in the presence of 1 mM auxin for 2 hr . The third G1 culture was released into YPD consisting of 0 . 2 M HU to arrest cells in S phase for 2 hr , and subsequently released into 15 µg/ml nocodazole ( Noc ) -containing YPD to arrest cells into G2/M phase with 1 mM auxin for 2 hr . Upon auxin addition , samples were collected every 15 min for FACS analysis as described in a previous study ( Hanna et al . , 2001 ) and for western blotting analysis of Bre1 protein level using anti-Myc antibody and Pgk1 as the loading control . Bre1-AID*−9Myc protein levels were quantified by Image J software . The normalized signal density value for each sample band at an indicated time point was calculated as the ratio of the relative density of each sample lane ( after subtracting background ) over the relative density of the Pgk1 loading control for the same lane ( Miller , 2010 ) . The normalized protein amounts relative to that before auxin addition were plotted in graphs . Auxin was added to the cell cultures arrested at the indicated cell-cycle stages to induce the degradation of Bre1 as described below . ( 1 ) For the no auxin-induced degradation control , early log-phase cells containing Bre1-AID*−9Myc were arrested in G1 with 5 µg/ml alpha-factor ( α-F ) for 3 hr before releasing into YPD medium containing 5 µg/ml α-F for another 2 hr to maintain G1 arrest . Then , G1 phase cells were released into hydroxyurea ( HU ) -containing media to arrest cells in S phase for 2 hr . Finally , S phase cells were released into Noc-containing media to arrest cells in G2/M phase for 3 hr . ( 2 ) For Bre1 degradation in all the stages , cell-cycle arrest procedures were the same as described in ( 1 ) , except that cells in HU-containing medium were arrested for 4 hr instead of 2 hr due to a delayed G1-S transition and progression in S phase in the presence of auxin in G1 . Bre1 degradation at each stage was induced by the addition of 1 mM auxin to the same medium . ( 3 ) For Bre1 degradation in G1 and S , procedures were the same as in ( 2 ) except in the last step , when S phase cells were released into Noc-containing medium without auxin . ( 4 ) For Bre1 degradation in G1 phase , procedures were the same as in ( 3 ) except that G1-arrested cells were released into HU-containing medium without auxin . ( 5 ) For Bre1 degradation in S phase , procedures were the same as in ( 1 ) except that G1-arrested cells were released into HU-containing medium with auxin . ( 6 ) For Bre1 degradation in G2/M phase , procedures were the same as in ( 1 ) except that S-phase-arrested cells were released into Noc-containing medium with auxin . ( 7 ) For Bre1 degradation in S and G2/M phases , procedures were the same as in ( 1 ) except that auxin was added into both HU- and Noc-arrested cells . Samples were collected every 30 min for flow cytometry analysis of DNA content and western blotting analysis for quantification of Bre1 protein levels as described above . In all the cases , G2/M-phase cells were collected after Noc arrest for 4 hr for cohesion assessment . Yeast whole cell extracts were prepared using the trichloroacetic acid ( TCA ) precipitation method as described by the Dohlman lab ( http://www . med . unc . edu/~dohlmahg//TCA . html ) . The protein concentration was determined with the Bio-Rad DC protein assay kit . Equal amounts of protein samples were boiled in 4xSDS-PAGE sample buffer and subjected to SDS-PAGE gel , before being transferred to Immobilon PVDF membrane ( CAT#: 1620177 , Millipore , Ireland ) . The membrane was blocked in TBST ( 20 mM Tris-HCl , 125 mM NaCl , 0 . 1% Tween-20 ) containing 5% non-fat dry milk for 1 hr at room temperature before probing using antibodies against HA ( 12CA5 , 1:1000 , Cat#: 11583816001 , Roche , Germany ) , Myc ( 9E10 , 1:2000 , Cat#: 05–419 , Millipore , Billerica MA , USA ) , Rad53p ( 1:2000 , Cat#: ab104232 , Abcam , San Francisco , CA , USA ) , Pgk1 ( 1:6000 , Cat#: ab113687 , Abcam , Frederic , MD , USA ) , acetyl-Smc3 ( 1: 500 , a gift from the Dmitry Ivanov Lab ) , Flag ( M2 , 1:2000 , Cat#: F1804 , Sigma-Aldrich , St . Louis , MO , USA ) or ubiquitin ( P4G7-H11 , 1:1000 , Cat#: ab90376 , Abcam , San Francisco , CA , USA ) and incubated overnight at 4°C . The membrane was washed three times with TBST for 10 min each time , and subsequently incubated with the secondary antibody ( Goat polyclonal Secondary Antibody to Rabbit IgG H and L [HRP , 1:100000 , Cat#: ab97051 , Abcam , San Francisco , CA , USA or Goat polyclonal Secondary Antibody to Mouse IgG H and L [HRP , 1:100000 , Cat#: ab97023 , Abcam , San Francisco , CA , USA] ) for 30 min at room temperature . After washing , the blots were detected using the Amersham ECL Select western blotting detection reagent ( GE Healthcare Life Sciences , UK ) and developed using X-ray film . Early log-phase cells were arrested in G1 , S or G2/M phase with 5 µg/ml α-F , 0 . 2 M HU or 15 µg/ml Noc , respectively , for 3 hr . Chromatin spread was performed on slides as described previously ( Grubb et al . , 2015; Rockmill et al . , 2009 ) . The slides were blocked with 300 µl 1% bovineserum albumin ( BSA ) /PBS in a moist chamber for 15 min at room temperature and subsequently incubated with 100 µl of primary antibody against HA ( 1:200 , Cat#: sc-57592 , Santa Cruz Biotechnology , Dallas , TX , USA ) or Myc ( 1:200 , Cat#: sc-40 , Santa Cruz Biotechnology , Dallas , TX , USA ) in 1% BSA/PBS at room temperature for 2 hr . Samples were then washed three times with PBS , and incubated with CY3-conjugated goat anti-mouse secondary antibody ( 1:500 , Cat#: 115-166-062 , Jackson Immunoresearch Labs , West Grove , PA , USA ) in 1% BSA/PBS at room temperature for 2 hr . Finally , the slides were stained with 4’ , 6-diamidino-2-phenylindole ( DAPI ) ( 1 µg/ml ) for 5 min , washed with PBS , and mounted with mounting media . The slides were then processed for immunostaining . Indirect immunofluorescence was observed using a Carl Zeiss LSM 710 NLO confocal laser scanning microscope with a 40x/1 . 4 NA oil objective and a conventional FITC excitation filter . The percentage of chromatin masses associated with epitope-tagged protein was calculated as the ratio of the number of chromatin masses with epitope tag signals over the number of chromatin masses with DAPI signals . At least 100 chromatin masses were scored , and data were averaged from at least three independent experiments . ChIP was carried out according to the methods used in previous studies with slight modifications ( Wahba et al . , 2013; Zakari et al . , 2015 ) . In brief , 100 ml early-log-phase cells were arrested in S phase with 0 . 2 M HU for 3 hr at 30˚C ( Ricke and Bielinsky , 2004 ) . 1% formaldehyde was used for crosslinking for 20 min at room temperature . Spheroplasts were prepared as described for the chromatin spread assay . Spheroplasts were resuspended in SDS lysis buffer ( 1% SDS , 10 mM EDTA , 50 mM TRIS , pH 8 . 1 ) with proteinase K inhibitor and sonicated by an AFA focused-ultrasonicator ( Covaris ) : 10% duty cycle , 75 watts intensity of peak incident power , 200 cycles per burst , 4 min to obtain sheared DNA fragments of 250–1000 bp in length ( average 500 bp ) . Crosslinked proteins were immunoprecipitated with monoclonal anti-HA antibodies ( 12CA5 ) or anti-Myc antibodies ( 9E10 ) , as well as Mouse IgG ( Cat: 12–371 , Millipore , Temecula , CA , USA ) as a control for specificity , overnight at 4°C . The immune complexes were harvested by the addition of 50 µl of protein A dynabeads ( Cat: 10001D , Thermo Fisher Scientific Inc , Norway ) . Formaldehyde crosslinks were reversed by incubation at 65°C for 5 hr , followed by protease K treatment at 42°C for 1–2 hr , and purification of recovered DNA was achieved using the ChIP DNA Clean and Concentrator kit ( Zymo Research Corporation , Cat#: D5205 , Irvine , USA ) . The purified DNA was subjected to quantitative real-time PCR using the StepOnePlus Real-Time PCR System ( ABI ) . Primers for origins used were the same as those used previously ( Trujillo and Osley , 2012 ) . The % Input for each IP ( 2-∆Ct ) at origins was calculated , and the specific enrichment ( % Input Ab IP – % Input IgG IP , according to SABiosciences ChIP-qPCR Array data analysis method , http://www . sabiosciences . com/chippcrarray_data_analysis . php ) was further normalized to the specific enrichment at the ASI1 locus ( 40 kb from the nearest origin ) . The normalized ratios of three independent IP experiments , each with duplicates or triplicate qPCR reactions for each primer set , were averaged and plotted on each graph as the relative enrichment of proteins . Immunoprecipitation was performed as described previously with modifications ( Gerace and Moazed , 2014 ) : yeast whole-cell extracts at early log phase were prepared by bead-beating in lysis/IP buffer ( 50 mM Tris-HCl [pH 7 . 5] , 100 mM NaCl , 5 mM EDTA [pH 8 . 0] , 0 . 1% NP40 , 1 mM DTT and protease inhibitors ) from 100 mL of log-phase culture . Extracts were precipitated with anti-HA ( 12CA5 ) or anti-Myc ( 9E10 ) antibody at 4˚C overnight , followed by incubation with protein A dynabeads for 2 hr at room temperature . Dynabeads were then washed three times with lysis buffer . Associated proteins were eluted by incubating beads with SDS sample buffer for western blotting . 10 mL log-phase cells ( OD600 ∼0 . 5 ) were used for RNA isolation as described previously ( Ares , 2012 ) . A total of 1 µg RNA was used for cDNA synthesis using the ThermoScript RT-PCR System . cDNA was analyzed by RT-qPCR using primers specific for the cell cyclin genes CLN2 and CLB5 , cohesion genes SCC1 , SMC3 , CTF4 , CTF18 and ECO1 , and actin gene ACT1 ( primer sequences are shown in Supplementary file 3 ) . PCR was carried out using Applied Biosystems SYBR Green PCR Master Mix . For each gene , relative expression levels were calculated by the comparative CT method ( StepOne software v2 . 3 , from Applied Biosystems ) obtained by qPCR assays of cDNA samples . Finally , CLN2 , CLB5 , SCC1 , SMC3 , CTF4 , CTF18 and ECO1 expression levels were normalized to that of ACT1 . Data are expressed as the mean ± standard error of the mean ( SEM ) from the number of independent experiments indicated in the figure legends . Student’s t-test was used to analyze statistical significance . | Most of the DNA in a cell is stored in structures called chromosomes . During every cell cycle , each cell needs to replicate its chromosomes , hold the two chromosome copies ( also known as “sister chromatids” ) together before cell division , and distribute them equally to the two new cells . Each step must be executed accurately otherwise the new cells will have extra or missing chromosomes – a condition that is seen in many cancer cells and that can cause embryos to die . Since these processes are so essential to life , they are highly similar in a range of species , from single-celled organisms such as yeast to multicellular organisms like humans . However , it was not clear when and how sister chromatids first join together , or how this process is linked to DNA replication . The DNA in the sister chromatids is wrapped around proteins called histones to form a structure known as chromatin . An enzyme called Bre1 plays roles in gene transcription and DNA replication and repair by adding ubiquitin molecules to a histone called H2B . Now , by using genetic , molecular and cell biological approaches to study baker and brewer yeast cells , Zhang et al . show that the activity of Bre1 helps to hold sister chromatids together . Specifically , Bre1 recruits proteins to the chromatin before and during DNA replication , which help to initiate replication and to establish cohesion between the sister chromatids . The ubiquitin molecule attached to H2B by Bre1 is also essential for establishing cohesion , acting as a mark that helps to link the two processes . In the future it will be worthwhile to investigate whether genetic mutations that prevent sister chromatids adhering to each other is a major cause of the chromosome abnormalities seen in cancer cells . This knowledge may be useful for diagnosing cancers . Drugs that prevent the activity of Bre1 and other proteins involved in holding together sister chromatids could also be developed as potential cancer treatments that kill cancer cells by causing instability in their number of chromosomes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"cell",
"biology"
] | 2017 | E3 ubiquitin ligase Bre1 couples sister chromatid cohesion establishment to DNA replication in Saccharomyces cerevisiae |
The Hippo tumor suppressor pathway regulates tissue growth in Drosophila by restricting the activity of the transcriptional coactivator Yorkie ( Yki ) , which normally complexes with the TEF/TEAD family DNA-binding transcription factor Scalloped ( Sd ) to drive the expression of growth-promoting genes . Given its pivotal role as a central hub in mediating the transcriptional output of Hippo signaling , there is great interest in understanding the molecular regulation of the Sd-Yki complex . In this study , we identify Nerfin-1 as a transcriptional repressor that antagonizes the activity of the Sd-Yki complex by binding to the TEA DNA-binding domain of Sd . Consistent with its biochemical function , ectopic expression of Nerfin-1 results in tissue undergrowth in an Sd-dependent manner . Conversely , loss of Nerfin-1 enhances the ability of winner cells to eliminate loser cells in multiple scenarios of cell competition . We further show that INSM1 , the mammalian ortholog of Nerfin-1 , plays a conserved role in repressing the activity of the TEAD-YAP complex . These findings reveal a novel regulatory mode converging on the transcriptional output of the Hippo pathway that may be exploited for modulating the YAP oncoprotein in cancer and regenerative medicine .
The Hippo signaling pathway is a conserved mechanism that regulates organ size , tissue regeneration and stem cell biology in diverse animals ( Halder and Johnson , 2011; Harvey and Tapon , 2007; Pan , 2010; Zhao et al . , 2010 ) . Central to this pathway is a kinase cascade comprising tumor suppressors Hippo ( Hpo , MST1/2 in mammals ) and Warts ( Wts , LATS1/2 in mammals ) , which are activated at the cell cortex by various upstream inputs . Wts/LATS in turn phosphorylates and inactivates the oncoprotein Yorkie ( Yki , YAP/TAZ in mammals ) . As a transcriptional coactivator , Yki/YAP/TAZ does not bind DNA directly and its ability to regulate target gene expression relies on its obligatory DNA-binding partner encoded by the TEF/TEAD family transcription factor Scalloped ( Sd , TEAD1/2/3/4 in mammals ) . Consistent with the importance of the Sd/TEAD-Yki/YAP/TAZ transcription factor complex in Hippo signaling , loss of Sd completely rescues Yki-induced overgrowth , and mutations in Yki/YAP/TAZ that disrupt its physical interactions with Sd/TEAD abolishes the growth-promoting activity of Yki/YAP/TAZ . Given its critical role in dictating transcriptional output of the Hippo pathway , there is great interest in understanding the function and regulation of the Sd/TEAD-Yki/YAP/TAZ complex . Recent studies in Drosophila have led to a default repression model concerning Sd function: in the absence of Yki , Sd functions by default as a transcriptional repressor that actively represses the transcription of Hippo target genes , and Yki promotes growth by de-repressing Sd’s repressor function ( Koontz et al . , 2013 ) . This model provides a plausible explanation for the perplexing observation that while Yki is required for normal tissue growth , loss of Sd has a negligible effect in growth in most Drosophila tissues: unlike loss of Yki , which leads to repression of Hippo target genes and tissue undergrowth , loss of Sd would lead to de-repression of Hippo target genes and therefore a much weaker effect on tissue growth . Indeed , despite its negligible effect on normal tissue growth , loss of sd completely rescues the undergrowth phenotype caused by loss of yki ( Koontz et al . , 2013 ) . Further support for this model came from the identification of an Sd-binding protein called Tondu-domain-containing Growth Inhibitor ( Tgi , Vgll4 in mammals ) ( Koontz et al . , 2013 ) , which competes with Yki to bind to the C-terminal region of Sd in a mutually exclusive manner . As expected of a Sd corepressor , loss of tgi rescues the undergrowth phenotype of yki mutant cells . However , unlike the full rescue of yki mutant by loss of sd , the rescue by tgi is partial , suggesting the existence of additional co-repressor ( s ) of Sd ( Koontz et al . , 2013 ) . Identification of such corepressors should provide important insights into transcriptional control of the Hippo signaling pathway . Cell competition was first described in Drosophila ( Morata and Ripoll , 1975 ) whereby underperforming cells ( aka loser cells ) , such as those with reduced ribosomal activities ( the Minute mutations ) , are actively eliminated by cell death when juxtaposed with wildtype cells ( aka winner cells ) ( Moreno et al . , 2002 ) . It has since been extended to many additional contexts involving social interactions between cells of different fitness , such as the elimination of neoplastic tumor cells by neighboring wildtype cells , the elimination of cells lacking the Dpp receptor TKV by their wildtype neighbors , or the elimination of wildtype cells by cells with higher Myc activity ( de la Cova et al . , 2004; Moreno and Basler , 2004; Moreno et al . , 2002; Rhiner et al . , 2010; Yamamoto et al . , 2017 ) . Recent studies further suggested that cell competition is conserved in mammals and may contribute to diverse physiological processes such as embryogenesis and tumor suppression ( Gogna et al . , 2015 ) . Several lines of evidence have implicated the Hippo signaling pathway in cell competition . It was reported that cells with higher Yki , like those with higher Myc , can eliminate their wildtype neighbors ( Neto-Silva et al . , 2010; Ziosi et al . , 2010 ) . Furthermore , increased Yki activity could rescue the elimination of neoplastic tumor cells or Minute cells by their wildtype neighbors ( Chen et al . , 2012; Menéndez et al . , 2010; Tyler et al . , 2007 ) . Lastly , the TEAD transcription factors were implicated in Myc-mediated cell competition in cultured mammalian cells ( Mamada et al . , 2015 ) . A caveat of these studies is that they often involve conditions in which Yki is massively activated at supraphysiological level . Whether Yki is required for cell competition at its endogenous physiological level remains an open question . Here , we describe the identification of Nerfin-1 as a transcriptional repressor that antagonizes the Sd-Yki complex by binding to the TEA DNA-binding domain of Sd . Not only does ectopic expression of Nerfin-1 result in tissue undergrowth in an Sd-dependent manner , loss of Nerfin-1 enhances the ability of winner cells to eliminate loser cells in multiple scenarios of cell competition . We also provide evidence showing the conserved function of a mammalian ortholog of Nerfin-1 in repressing the activity of the TEAD-YAP complex .
In an effort to identify additional regulators of the Sd-Yki transcription factor complex , we searched Drosophila Interaction Database ( DroID , http://www . droidb . org ) for proteins that physically associate with Sd and/or Yki . Nerfin-1 emerged as a candidate that interacts with both Sd and Yki . Specifically , a genome-wide yeast-two-hybrid screen using Nerfin-1 as bait identified Sd as a Nerfin-1-binding protein ( Giot et al . , 2003 ) . In addition , a large-scale affinity purification screen in Drosophila S2R+ cells identified both Sd and Yki in Nerfin-1 immunoprecipitates ( Rhee et al . , 2014 ) . Nerfin-1 was initially identified as a protein that contains three zinc fingers and that is highly expressed in the nervous system ( thus the name Nervous Finger ) ( Stivers et al . , 2000 ) . Besides Nerfin-1 , the Drosophila genome also encodes a related protein called Nerfin-2 , which shares sequence similarity with Nerfin-1 . However , unlike Nerfin-1 , which is expressed in imaginal discs , the expression of Nerfin-2 is undetectable in imaginal discs ( Brown et al . , 2014 ) . We therefore focused our analysis on Nerfin-1 unless otherwise stated . To explore the relationship between Nerfin-1 , Sd and Yki , we expressed epitope-tagged constructs for the three proteins in Drosophila S2R+ cells and examined their interactions by coimmunoprecipitation ( co-IP ) assays . Consistent with the DroID data , Sd robustly interacted with Nerfin-1 or Yki in pairwise co-IP assays ( Figure 1A ) . In contrast , only weak interaction was detected between Nerfin-1 and Yki ( Figure 1A ) . We then examined how each pairwise interaction was influenced by the co-expression of the third protein . While co-expression of Yki did not affect Sd-Nerfin-1 interaction and co-expression of Nerfin-1 did not affect Sd-Yki interaction , the modest Nerfin-1-Yki interaction was greatly potentiated by co-expression of Sd ( Figure 1A–B ) . These findings suggest that Sd may simultaneously bind Nerfin-1 and Yki and thus function as an intermediary protein to bridge Nerfin-1 and Yki in a common protein complex . Consistent with this notion , the modest co-IP between Nerfin-1 and Yki was abolished by RNAi knockdown of Sd ( Figure 1C ) , demonstrating that the co-IP between Nerfin-1 and Yki constructs was mediated by endogenous Sd present in S2R+ cells . After demonstrating physical interactions between Nerfin-1 and the Sd-Yki complex , we wished to determine the functional effect of these interactions on the activity of the Sd-Yki complex . For this purpose , we assayed the transcriptional activity of the Sd-Yki complex using a luciferase reporter carrying multimerized minimal Hpo Responsive Element ( HRE ) derived from the Hippo target gene diap1 ( Wu et al . , 2008 ) . As expected , the HRE-luciferase reporter was increased 35-fold by co-expression of Sd and Yki ( Figure 1D ) . Interestingly , co-expression of Nerfin-1 repressed the Sd-Yki-stimulated HRE-luciferase reporter close to the basal level ( Figure 1D ) , suggesting that Nerfin-1 can antagonize the transcriptional activity of the Sd-Yki complex . To probe the molecular mechanisms underlying Nerfin-1’s repressive activity , we searched DroID database for other Nerfin-1-interacting proteins . Interestingly , CtBP , a component of the CtBP corepressor complex that represses gene transcription through the associated histone deacetylase ( HDACs , Rpd3 in Drosophila ) , was identified as a Nerfin-1-associated protein in a large-scale IP/MS screen ( Guruharsha et al . , 2011 ) . By co-expressing epitope-tagged CtBP or Rpd3 together with Nerfin-1 in S2R+ cells , we found that both proteins could be immunoprecipitated by Nerfin-1 ( Figure 1E–F ) , suggesting that the CtBP complex may be involved in Nerfin-1-mediated transcriptional repression . Consistent with this notion , treating S2R+ cells with the HDAC inhibitor Trichostatin A ( TSA ) significantly reversed Nerfin-1’s inhibition of Sd-Yki-stimulated HRE-luciferase reporter , despite that TSA itself had negligible effect on the basal activity of the HRE-luciferase reporter ( Figure 1D ) . Taken together , these findings support a model whereby Nerfin-1 antagonizes the transcriptional activity of the Sd-Yki complex by binding to Sd and recruiting transcriptional corepressors such as the CtBP corepressor complex . To test whether the function of Nerfin-1 is evolutionarily conserved , we examined Insulinoma-associated 1 ( INSM1 ) , a mammalian homologue of Nerfin-1 . Similar to their Drosophila counterparts , TEAD1 and YAP were immunoprecipitated by INSM1 in 293T cells ( Figure 1G ) . Further echoing our findings in Drosophila , YAP-INSM1 interaction was much weaker than TEAD1-INSM1 interaction , and co-expression of TEAD1 significantly enhanced YAP-INSM1 ( Figure 1G ) . Indeed , INSM1 potently suppressed YAP-induced transcriptional activation of luciferase reporters driven by multimerized TEAD binding sites ( 8xGTIIC-luc ) ( Dupont et al . , 2011 ) or CTGF promoter ( CTGF-luc ) ( Zhao et al . , 2008 ) ( Figure 1H–I ) . To further examine its role in transcriptional regulation , we measured the mRNA levels of a common set of YAP target genes in a lung carcinoid tumor cell line ( H727 ) with high endogenous INSM1 expression . In agreement with our luciferase reporter assay , siRNA knockdown of INSM1 resulted in the upregulation of several YAP target genes , most notably CTGF ( Figure 1J ) . These results suggest a conserved role for INSM1 in regulating the transcriptional output of Hippo signaling in mammalian cells . To further corroborate our model that Nerfin-1 antagonizes the transcriptional output of the Sd-Yki complex by binding to Sd , we assayed Nerfin-1 activity in vivo . Consistent with its inhibitory activity on the Sd-Yki transcription complex in S2R+ cells , MARCM ( mosaic analysis with a repressible marker ) clones with Nerfin-1 overexpression were much smaller than control clones in the eye imaginal discs ( Figure 2A–B ) . Importantly , this small-clone phenotype was completely rescued by removal of Sd , demonstrating that Nerfin-1’s growth inhibitory activity is dependent on Sd ( Figure 2C–D ) . Further supporting a link between Nerfin-1 and Hippo signaling , expression of Nerfin-1 by the engrailed-Gal4 driver resulted in smaller posterior compartment in the wing discs accompanied by decreased expression of the Hippo pathway reporter expanded-lacZ ( Figure 2E–F ) . That Nerfin-1 antagonizes the activity of the Sd-Yki complex was further supported by genetic interactions among these genes in vivo . Overexpression of UAS-Nerfin-1 by the GMR-Gal4 driver resulted in 100% lethality at late pupal stage ( Figure 2G–H ) . When Nerfin-1 was co-expressed with Yki by the GMR-Gal4 driver , we observed a mutual suppression: on the one hand , the lethality of GMR>Nerfin-1 flies was suppressed by Yki co-expression to 55% ( Figure 2M ) ; on the other hand , Nerfin-1 overexpression significantly suppressed the enlarged eye size resulting from Yki overexpression ( Figure 2I–J ) . A similar mutual suppression was observed in flies expressing Nerfin-1 , Sd and Yki: Nerfin-1 overexpression significantly suppressed the enlarged eye size resulting from Sd-Yki overexpression ( Figure 2K–L ) . Consistent with our MARCM analysis showing that Nerfin-1’s growth inhibitory activity is dependent on Sd , halving the dosage of endogenous sd dominantly rescued the GMR>Nerfin-1 flies from 100% pupal lethality to 47% survival to adulthood ( Figure 2M ) . These dosage-sensitive genetic interactions further support our model implicating Nerfin-1 as a transcriptional repressor that impinges on the Sd-Yki complex . The data presented so far suggest that Nerfin-1 can inhibit the activity of the Sd-Yki complex by binding to Sd . Furthermore , Nerfin-1 and Yki do not compete with each other for Sd-binding , and the three proteins can co-exist in a trimeric protein complex . The simplest model to account for these findings is that Nerfin-1 and Yki non-competitively bind to different regions of Sd , with Nerfin-1 repressing while Yki activating the transcriptional outcome of Sd . We tested this model by mapping the domains in Sd and Nerfin-1 that are required for Sd-Nerfin-1 interactions . Sd contains two previously characterized domains , an N-terminal TEA domain that is known to bind DNA and a C-terminal domain that is known to bind Yki . Unlike full-length Sd , a truncated Sd protein lacking the N-terminal TEA domain did not interact with Nerfin-1 in pairwise co-IP assays , and this mutant also lost the ability to enhance the interactions between Nerfin-1 and Yki ( Figure 3A–B ) . These results suggest that the TEA domain of Sd is required for binding to Nerfin-1 . In a previous study , X-ray structure of the TEA domain of human TEF-1 revealed a three-helix bundle structure in which helix 3 ( H3 ) recognizes DNA while a hydrophobic surface formed by helix 1 ( H1 ) and helix 2 ( H2 ) was speculated to provide a docking surface for unknown protein ( s ) ( Anbanandam et al . , 2006 ) ( Figure 3—figure supplement 1A–B ) . We tested whether this hypothetical protein docking surface is required for Nerfin-1-binding by mutating two conserved hydrophobic residues on this surface , Y108 and L130 ( Figure 3—figure supplement 1A–B ) . Similar to truncated Sd lacking the entire TEA domain , an SdY108A or SdL130G mutation also abolished Nerfin-1-Sd interaction ( Figure 3C ) . In contrast , mutation of residues in the loop linking H1 and H2 ( SdD120N/E121N ) , or residues in the DNA-contacting H3 ( SdQ153A and SdR157A ) , had no effect on Nerfin-1-Sd interactions ( Figure 3C ) . Thus , Nerfin-1 represents a strong candidate for the unknown protein that was previously speculated to bind to the hydrophobic surface on the TEA domain ( Anbanandam et al . , 2006 ) . To further corroborate the functional significance of Nerfin-1-Sd binding in inhibiting the transcriptional output of the Sd-Yki complex , we took advantage of a fusion protein between the DNA-binding domain of Gal4 and Yki ( Gal4-Yki ) ( Huang et al . , 2005 ) . Since Gal4-Yki can directly activate a UAS-luciferase reporter in an Sd-independent manner , one would expect Gal4-Yki to be immune to Nerfin1-mediated repression . Indeed , despite Nerfin-1’s potent inhibitory activity on Sd-Yki-stimulated HRE-luciferase reporter ( Figure 1D ) , Nerfin-1 was completely inactive towards the transcriptional activity of Gal4-Yki ( Figure 3D ) . In contrast , as reported before ( Huang et al . , 2005 ) , Gal4-Yki was significantly inhibited by co-expression of the upstream tumor suppressor Hpo ( Figure 3D ) . These results provide further support for our model that Nerfin-1 binding to the TEA domain of Sd is required for Nerfin-1-mediated transcriptional repression . As an additional test of the importance of Sd’s TEA domain in Nerfin-1-mediated transcriptional repression , we engineered a fusion protein containing the N-terminal TEA domain of Sd and the C-terminal transactivation domain of Yki . As expected , the SdN-YkiC fusion protein potently activated the HRE-luciferase reporter ( Figure 3M ) . Importantly , unlike the Gal4-Yki fusion protein , Nerfin-1 significantly inhibited the transcriptional activity of the SdN-YkiC fusion protein ( Figure 3M ) . To test whether human INSM1 also executes its repressor function through the TEA domain of TEAD , we engineered a fusion protein containing the TEA domain of TEAD1 and the VP160 transcription activation domain . As expected , INSM1 also repressed the transcriptional activity of the TEA-VP160 fusion protein , as measured by the TEAD-responsive 8xGTIIC-luciferase reporter ( Figure 3E ) . Taken together , these findings highlight the TEA domain of Sd/TEAD as the molecular target of Nerfin-1-mediated transcriptional repression ( Figure 3F ) . The most prominent structural feature of Nerfin-1 is the presence of three zinc fingers , referred to as ZF1 , ZF2 and ZF3 hereafter ( Figure 3—figure supplement 1C–D ) . To examine whether the zinc fingers are required for Sd-binding , we generated C-terminal truncations lacking all or some of the ZFs , as well as full-length Nerfin-1 carrying Cys-to-Ala mutations of two conserved cystine residues in each ZF ( ZF1CA , ZF2CA or ZF3CA ) ( Figure 3G and Figure 3—figure supplement 1D ) . In co-IP assays , a truncated Nerfin-1 protein lacking all ZFs ( NT ) completely abolished Nerfin-1-Sd interactions , while a truncated protein retaining ZF1 ( NT-ZF1 ) could still bind Sd , suggesting that ZF1 is required for Nerfin-1-Sd interaction ( Figure 3G–H ) . Consistent with this notion , the ZF1CA but not ZF3CA mutation abolished Nerfin-1-Sd interaction , while the ZF2CA mutation partially reduced Nerfin-1-Sd interaction ( Figure 3G–H ) . Next , we assayed the activity of the Nerfin-1 mutants in vivo . While overexpression of wildtype Nerfin-1 or ZF3CA by the GMR-Gal4 driver resulted in 100% pupal lethality , overexpression of ZF1CA by the same driver resulted in 100% flies surviving to adulthood and these flies had normal eye size ( Figure 3I–J and L ) . Overexpression of ZF2CA had an intermediate effect , with 25% flies surviving to adulthood with reduced eye size ( Figure 3K ) . A similar trend was observed in HRE-luciferase reporter assay: ZF1CA was inactive and ZF3CA was fully functional in inhibiting the transcriptional activity of the SdN-YkiC fusion protein , while ZF2CA was partially functional ( Figure 3M ) . These structure-functional analyses further support our model implicating Nerfin-1 as a Sd-binding transcriptional repressor . To further investigate the molecular mechanisms by which Nerfin-1 represses Sd-mediated transcription , we performed chromatin immunoprecipitation ( ChIP ) to examine the physical interactions between these proteins and the endogenous diap1 HRE site in S2R+ cells ( Figure 3—figure supplement 2 ) . As expected , both Sd and Nerfin-1 associated with the diap1 HRE . Interestingly , the binding of Sd to the HRE was decreased by the co-expression of Nerfin-1 but not a Nerfin-1 mutant protein defective in Sd-binding ( ZF1CA ) . Together with our data implicating CtBP/HDAC in Nerfin-1 function ( Figure 1D–F ) , these results suggest that Nerfin-1 represses Sd-dependent transcription both by compromising Sd-DNA interaction and by recruiting CtBP/HDAC corepressors . To investigate the physiological requirement of Nerfin-1 , we examined loss-of-function mutant of nerfin-1 . It was reported before that nerfin-1 is an essential gene , and homozygous mutant animals die at late embryonic stage ( Kuzin et al . , 2005 ) . We therefore examined nerfin-1 mutant clones in several developmental contexts that are known to involve Hippo signaling . Given the antagonistic relationship between Nerfin-1 and Yki in dictating the transcriptional output of Sd , we expected that removal of endogenous Nerfin-1 may result in phenotypes resembling those caused by elevated Yki activity . Unexpectedly , we failed to detect such mutant phenotypes in multiple developmental contexts . First , mosaic eyes containing mutant clones for a null allele of nerfin-1 did not show increased representation of mutant tissues ( Figure 4A–B ) . Second , nerfin-1 mutant clones in the pupal retina contained normal number of interommatidial cells , which is known to be exquisitely sensitive to Yki activity ( Figure 4—figure supplement 1A–B ) . Third , nerfin-1 mutant cells in the eye or the wing imaginal discs showed normal expression of the Hippo pathway targets Diap1 , Expanded and Merlin ( Figure 6A–B and Figure 4—figure supplement 1E–F ) . Fourth , posterior follicle cells mutant for nerfin-1 in stage seven egg chambers did not show elevated expression of Cut , a hallmark of activated Yki ( Figure 4—figure supplement 1G ) . Suspecting that loss of Nerfin-1 may be compensated for by the related Nerfin-2 protein , we generated a predicted null mutant allele of nerfin-2 by CRISPR/Cas9 ( Figure 4—figure supplement 1H ) . nerfin-2 is a non-essential gene , as nerfin-2 homozygotes were fully viable without detectable abnormalities . Like nerfin-1 mutant clones , double mutant clones lacking both nerfin-1 and nerfin-2 still contained normal number of interommatidial cells in pupal retina ( Figure 4—figure supplement 1C–D ) . Thus , at least in the developmental contexts we have examined , loss of Nerfin-1 does not lead to visible phenotypes resembling those caused by increased Yki activity . The apparent dispensability of Nerfin-1 in normal development promoted us to examine its requirement in other biological processes that involve Hippo signaling . Recent studies in both Drosophila and mammalian cells have implicated the Hippo pathway in cell competition: not only does high Yki/YAP activity lead to a winner cell fate , it also protects loser cells from being eliminated in Minute-mediated cell competition ( Neto-Silva et al . , 2010; Tyler et al . , 2007; Ziosi et al . , 2010 ) . We therefore examined whether Nerfin-1 may be required for the regulation of Yki activity in cell competition . We first used eyeless-Flp to generate adult mosaic eyes containing Rps174 heterozygous Minute cells labeled by red pigmentation and white-colored wildtype cells lacking red pigmentation . Thus , the relative representation of red-colored Minute tissue and white-colored wildtype tissue in the adult eyes provides an easy readout for cell competition between the loser ( Minute ) and the winner ( wildtype ) cells ( Figure 4E; quantified in Figure 4I ) . Interestingly , when compared to mosaic eyes containing wildtype winner cells ( Figure 4E ) , mosaic eyes in which the winner cells were mutant for nerfin-1 had significantly lower percentage of red tissue and significantly higher percentage of white tissue ( Figure 4F; quantified in Figure 4I ) , suggesting that loss-of-nerfin-1 enhanced the ability of winner cells to eliminate loser cells in Minute-mediated cell competition . For simplicity , we will use the term ‘super-winner’ to describe the ability of loss-of-nerfin-1 to further enhance the advantage of wildtype cells when juxtaposed with less fit cells , in contrast to the term ‘super-competitor’ , which describes cells with higher fitness than wildtype cells ( Moreno and Basler , 2004 ) . Interestingly , halving the dosage of endogenous yki was sufficient to reverse this ‘super-winner’ phenotype conferred by loss-of-nerfin-1 ( Figure 4G; quantified in Figure 4I ) , consistent with our model implicating Nerfin-1 as an antagonist of Yki . Besides Rps17 , loss-of-nerfin-1 conferred a similar ‘super-winner’ phenotype when mutation of another ribosomal subunit RpL14 was used to generate Minute tissues in the eyes ( Figure 4—figure supplement 2A–D ) . To ensure that the ‘super-winner’ phenotype was caused by loss of nerfin-1 , we performed a rescue experiment and found that the ‘super-winner’ phenotype was rescued by a tubulin-nerfin-1 transgene ( Figure 4H; quantified in Figure 4I ) . Next , we took advantage of a genetic setup that enabled us to mutate nerfin-1 specifically in the loser cells in the context of Minute-mediated competition ( Figure 4—figure supplement 2E ) ( Tyler et al . , 2007 ) . In contrast to loss-of-nerfin-1 in the winner cells , loss-of-nerfin-1 in the loser cells had no effect on the representation of winner vs . loser tissues in adult eyes ( Figure 4C–D ) . Taken together , these results suggest that Nerfin-1 is preferentially required in the winner cells to suppress Minute-mediated cell competition . After demonstrating the requirement of Nerfin-1 in suppressing winner cell advantages in Minute-mediated cell competition , we extended our analysis to another condition that involves the juxtaposition of cells with different fitness ( Hafezi et al . , 2012 ) . Instead of competition between Minute ( red ) and wildtype ( white ) cells , we assayed interactions between cells heterozygous for a recessive cell-lethal mutation ( red ) and wildtype ( white ) cells ( Figure 4J; quantified in Figure 4N ) . Interestingly , analogous to what we have observed in Minute-mediated cell competition , loss-of-nerfin-1 increased the representation of wildtype cells relative to cells heterozygous for the cell-lethal mutation and halving the dosage of endogenous yki similarly reversed this phenotype conferred by loss-of-nerfin-1 ( Figure 4K–L; quantified in Figure 4N ) . Furthermore , the increased representation of nerfin-1 mutant cells in this context was also rescued by a tubulin-nerfin-1 transgene ( Figure 4M; quantified in Figure 4N ) . Taken together , we conclude that Nerfin-1 suppresses winner cell advantages in Minute- and cell lethal-mediated cell competition , both in a Yki-dependent manner . Importantly , in both contexts , the overall eye size was not changed , suggesting that the observed changes in red/white ratio results from modulation of cell competition , not simply tissue overgrowth . To trace the developmental origin of the ‘super-winner’ phenotype conferred by loss-of-nerfin-1 , we used eyeless-Flp to generate 3rd instar mosaic eyes containing Rps174 heterozygous Minute cells that were positively labeled by β-galactosidase . We then followed cell death , a hallmark of loser cell fate in cell competition ( Moreno et al . , 2002 ) , by staining for active effector caspase Dcp-1 . Consistent with results from the adult eyes , compared to mosaic eyes containing wildtype winner cells , we observed a 50% decrease of Minute tissues ( β-gal-positive ) when winner cells were mutant for nerfin-1 ( Figure 5A–C ) . Concomitantly , we saw a two-fold increase of apoptotic Minute cells ( cells positive for both β-gal and Dcp-1 ) when winner cells were mutant for nerfin-1 ( Figure 5D ) . Next , we repeated the same analysis in mosaic wing discs containing Rps174/+Minute cells . As in the eye discs , we observed a similar decrease of Minute tissues in the wing discs accompanied by an increase of apoptotic Minute cells when winner cells were mutant for nerfin-1 ( Figure 5E–H ) . These data further support our model that the Nerfin-1 is required in the winner cells to suppress Minute-mediated cell competition . To further characterize the nerfin-1-dependent ‘super-winner’ phenotype in Minute-mediated cell competition , we examined the expression of the Yki target gene diap1 . In mosaic wing discs containing Rps174/+ Minute cells and wildtype cells , we observed a modest increase of Diap1 expression in the winner cells relative to the loser cells , especially when comparing winner and loser cells immediately abutting the clonal boundary ( Figure 6C ) . This observation is consistent with previous reports implicating Hippo signaling in cell competition . Interestingly , when we examined mosaic wing discs containing Rps174/+ Minute loser cells and nerfin-1 mutant winner cells , a more dramatic difference in Diap1 expression was observed across the winner/loser boundary ( Figure 6D , quantified in Figure 6E ) . These results provide further support for our model that Nerfin-1 suppresses winner cell advantages in cell competition by regulating the transcriptional output of Hippo signaling .
In this study , we present the characterization of Nerfin-1 as a corepressor that binds the TEA domain of Sd and antagonizes Sd-Yki-mediated transcriptional activation . Several lines of evidence support this conclusion . First , Nerfin-1 physically associates with Sd and Yki , and the three proteins can form a trimeric complex . Second , in the eye discs , overexpression of Nerfin-1 suppresses tissue growth in an Sd-dependent manner . Third , Nerfin-1 genetically interacts with sd and yki in a dosage-dependent manner . Interestingly , the relationship between Nerfin-1 and Sd appears to be evolutionary conserved , since the mammalian counterparts of Nerfin-1 , Sd and Yki also associate with each other and cooperatively regulate YAP target gene expression . Along this line , we note that loss of function of the C . elegans ortholog of Nerfin-1 and Sd , egl-46 and egl-44 , respectively , resulted in similar defects in the specification of FLP and HSN neurons ( Wu et al . , 2001 ) . Thus , the functional interactions between Nerfin-1 and Sd may have a deep evolutionary origin . It is interesting to compare Nerfin-1 with another Sd-binding co-repressor , Tgi . While both proteins confer transcriptional repression by binding to Sd , there are clear differences in their mode of action . While Tgi and Yki bind to the C-terminal domain of Sd in a mutually exclusive manner ( Koontz et al . , 2013 ) , Nerfin-1 binds to the TEA DNA-binding domain of Sd , apparently independent of the binding of Yki to the C-terminal domain of Sd . Thus , Sd-Tgi binding , but not Sd-Nerfin-1 binding , is modulated by the strength of Hippo signaling . We have examined whether these proteins play redundant roles in Sd-mediated repression in vivo by generating nerfin-1 tgi double mutant combination , and found that loss of tgi does not enhance the proliferation of nerfin-1 mutant tissues , or the ‘super-winner’ phenotype of nerfin-1 mutant in cell competition ( Figure 4—figure supplement 2F–H ) . These results suggest that Nerfin-1 and Tgi are likely required in different contexts to suppress Sd function , although we cannot exclude the possibility that they are co-required in yet-to-be-identified biological contexts . At present , how Tgi mediates transcriptional repression , especially the identity of the co-repressors recruited by Tgi to repress target gene transcription , remains unclear ( Koontz et al . , 2013 ) . By contrast , Nerfin-1 appears to repress target gene transcription by recruiting repressive histone modifying proteins such as CtBP-HDAC . Consistent with this notion , HDACs and CoREST , which are components of the CtBP-HDAC corepressor complex , were identified as proteins associated with the mammalian Nerfin-1 homologue INSM1 by IP/MS ( Welcker et al . , 2013 ) . Interestingly , contrary to the Sd co-repressor Nerfin-1 , the Sd co-activator Yki is known to confer transcriptional activation by recruiting the Trithorax-related ( Trr ) histone methyltransferase complex ( Oh et al . , 2014; Qing et al . , 2014 ) . Thus , the transcriptional output of Sd is dictated by the integration of positive and negative chromatin modifications at the target loci ( Figure 3F ) . Another insight from this study concerns the role of Nerfin-1 in cell competition . Although Nerfin-1 is dispensable for the growth of imaginal discs and ovarian follicle cells , we show that Nerfin-1 normally suppresses winner cell advantage in cell competition . Accordingly , loss of Nerfin-1 specifically in the winner cells confers a ‘super-winner’ phenotype , resulting in greater elimination of loser cells and increased representation of winner cells in mosaic tissues . Such ‘super-winner’ phenotype is mediated by increased Yki activity in the nerfin-1 mutant winner cells , as reflected by both increased expression of the Yki target diap1 in the winner cells and suppression of the ‘super-winner’ phenotype by halving the dosage of endogenous yki . Together with our molecular characterization of Nerfin-1 , these results suggest that Nerfin-1 normally suppresses winner cell advantage by antagonizing the Sd-Yki complex in the winner cells . Since cell competition is conserved in mammals ( Gogna et al . , 2015 ) , it will be interesting to examine whether the mammalian counterpart of Nerfin-1 ( INSM1 ) also plays a conserved role in cell competition . Our findings uncovering a Yki-dependent requirement for Nerfin-1 in cell competition have several implications . First , although a number of genetic perturbations are known to cause cell competition , to our knowledge , nerfin-1 is the first example of mutations that do not confer cell competition per se , but instead modulate the degree of cell competition conferred by others . Thus , cell competition is not simply the constitutive outcome of juxtaposition of cells of different fitness; the process itself is subjected to additional regulation . Second , although previous studies have implicated the Hippo signaling pathway in cell competition ( Neto-Silva et al . , 2010; Tyler et al . , 2007; Ziosi et al . , 2010 ) , those studies involved conditions in which Yki is massively activated at supraphysiological level . Our study therefore provides the first evidence that Yki is required for cell competition at its endogenous physiological level . Lastly , given that Nerfin-1 is dispensable for the growth of imaginal discs and ovarian follicle cells but is required for antagonizing Yki activity in cell competition , it is possible that Nerfin-1 is preferentially required in physiological contexts that involve interactions between cells with different Yki activity . Besides cell competition , cells with differential Yki activity have been documented in several examples of epithelial regeneration ( Grusche et al . , 2011; Losick et al . , 2013 ) . It will thus be interesting to examine the requirement of Nerfin-1 in these processes .
FLAG-Nerfin-1 construct was generated from cDNA clone LD18634 obtained from Drosophila Genomics Resource Center ( DGRC ) . INSM1 expressing construct was generated from ORF clone ( OHu02156 ) purchased from Genscript . Nerfin-1 zinc finger domain mutant constructs were generated by mutating following residues using site-directed mutagenesis: C252 and C255 were mutated to alanine in ZF1CA; C280 and C283 were mutated to alanine in ZF2CA; C336 and C339 were mutated to alanine in ZF3CA . Sd mutant constructs Y108F , D120N/E121N , L130G , Q153A and R157A were generated by site-directed mutagenesis . The following luciferase reporter constructs have been described previously: diap1 HRE-luciferase ( Wu et al . , 2008 ) , 8xGTIIC-luciferase ( Dupont et al . , 2011 ) , and CTGF-luciferase ( Zhao et al . , 2008 ) . Primary antibodies used in this study include the following: FLAG and HA ( Sigma ) ; Myc ( Calbiochem ) ; Diap1 ( gift from Bruce Hay ) ; Expanded and Merlin ( gift from Richard Fehon ) ; β-galactosidase and Cut ( Developmental Studies Hybridoma Bank ) . The following flies have been described previously: ykiB5 and UAS-Yki ( Huang et al . , 2005 ) ; sd47M , UAS-Sd and the diap1-lacZ reporter thj5c8 ( Wu et al . , 2008 ) ; nerfin-154 ( Kuzin et al . , 2005 ) ; Df ( 1 ) R194 w/FM7; P[Rpl36+w+] arm-lacZ FRT80B/TM3 ( Tyler et al . , 2007 ) ; Rps174; RpL141 and l ( 3 ) CL-L1 ( Bloomington Stock Center ) ; UAS-Nerfin-1 , UAS-Nerfin-1 ( ZF1CA-ZF3CA ) , tubulin-Nerfin-1 transgenes were made in this study . nerfin-2m6-8 was generated by CRISPR/Cas9-mediated mutagenesis with the facilitation of the following gRNAs selected by CRISPR Optimal Target Finder ( http://tools . flycrispr . molbio . wisc . edu/targetFinder/ ) : GGTCTCATCTTCCACGTAGA and TGACTACAATGAGTACGCCA . Mutants were identified by PCR selection and verified by Sanger sequencing . Representative genotypes used for clonal analysis in imaginal discs are as follows: GFP+ MARCM clones: GFP- mutant clones: To quantify the Minute area/disc area ratio ( Figure 5C and G ) , the pixels of the β-Gal-positive Minute area ( green ) in each disc was measured by ImageJ and divided by the pixels of the whole eye disc or wing disc . To quantify apoptotic Minute cells in eye discs ( Figure 5D ) , we first calculated the total number of Minute cells in each disc by dividing the pixels of the β-Gal-positive Minute area by the average pixel size of each cell ( 70 pixels/cell ) . The apoptotic Minute/total Minute ratio was then calculated by dividing the number of cells positive for both Dcp-1 and β-Gal by the number of total Minute cells in each disc . To quantify apoptotic Minute cells in wing discs ( Figure 5H ) , we calculated the ratio of Dcp-1-positive apoptotic Minute cells per micron of clonal boundary between the winner and loser cells , as described previously ( Li and Baker , 2007 ) . Drosophila S2R+ cells were cultured in Schneider’s medium supplemented with 10% fetal bovine serum ( FBS ) and antibiotics at 25°C . HEK293T and H727 cells were maintained at 37°C in DMEM and RPMI-1640 medium supplemented with 10% FBS and antibiotics , respectively . Plasmid Transfection were performed using Effectene reagent ( Qiagen ) . siRNA transfection was performed using Lipofectamine RNAiMax reagent . ON-TARGET plus siRNA SMARTpools were purchased from Dharmacon . Luciferase assay was performed as previously described ( Wu et al . , 2008 ) . Briefly , Drosophila S2R+ or 293T cells were seeded on a 48-well plate . After 24 hr , cells were transfected with the desired Firefly luciferase reporter constructs together with Pol III-Renilla luciferase reporter plasmid ( as the internal control ) using Effectene transfection reagent ( Promega ) . The HRE-luciferase reporter contains 24 tandem copies of the 26 bp minimal HRE from the diap1 gene ( Wu et al . , 2008 ) . Luciferase assay was performed at 24 hr post-transfection using Dual Luciferase Assay system ( Promega ) following the manufacturer’s instructions and a FLUOstar Lumiometer ( BMG Lab Technologies ) . ChIP assays were performed according to a previously described protocol ( Qing et al . , 2014 ) . Briefly , ∼5 × 106 S2R+ cells were cross-linked with 1% formaldehyde and sonicated to an average fragment size between 200 bp and 500 bp . Two micrograms of primary antibodies and 50 μl of protein G agarose were used in each assay . The immunoprecipitated DNA was quantified using quantitative real-time PCR . All values were normalized to the input . The primers for analyzing the ChIP are provided as follows: For quantitative RT-PCR , RNA was extracted using Trizol Reagent ( Invitrogen ) and purified using RNeasy Mini Kit ( Qiagen ) . For each sample , 1 ug RNA was used for reverse transcription using iScript cDNA synthesis kit ( BioRad ) . 100 ng cDNA was used for real-time PCR with SYBR master mix ( BioRad ) . Real-time PCR was performed using CFX96 real-time system ( BioRad ) . | Animals uses a range of mechanisms to stop their organs from growing once they have reached the right shape and size . One of these processes , a set of chemical messages called the Hippo pathway , controls the balance of cell death and cell division . In fruit flies , Hippo works by repressing a complex formed of two proteins , Yorkie and Scalloped , which normally switch genes on to encourage cells to grow . Yorkie is also involved in cell competition , a process in which cells in a tissue compare themselves to each other . Healthier ‘winner’ cells then kill neighboring ‘loser’ cells that are weaker or damaged . This ensures that the tissue keeps working properly . Despite Yorkie and Scalloped being key to control the growth and health of tissues , how the activity of these proteins is regulated was not well understood . To investigate , Guo et al . conducted a series experiments on fruit flies and found that a protein called Nerfin-1 can bind onto Scalloped to stop the Scalloped-Yorkie complex from switching on genes . As a result , flies with too much Nerfin-1 had stunted tissue growth . In addition , Guo et al . confirmed that the Nerfin-1 equivalent in mammals acts in the same way . Further work revealed that Nerfin-1 also plays a role in cell competition: without this protein , ‘winner’ cells became 'super winners' , eliminating even more of the loser cells . Besides regulating the size of organs , the Hippo pathway is also involved in stopping cells from dividing uncontrollably and becoming cancerous . Further research may therefore focus on Nerfin-1 and its equivalent in mammals to understand how this protein could contribute to the emergence of cancer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"developmental",
"biology"
] | 2019 | Nerfin-1 represses transcriptional output of Hippo signaling in cell competition |
Little is known about enteropathogen seroepidemiology among children in low-resource settings . We measured serological IgG responses to eight enteropathogens ( Giardia intestinalis , Cryptosporidium parvum , Entamoeba histolytica , Salmonella enterica , enterotoxigenic Escherichia coli , Vibrio cholerae , Campylobacter jejuni , norovirus ) in cohorts from Haiti , Kenya , and Tanzania . We studied antibody dynamics and force of infection across pathogens and cohorts . Enteropathogens shared common seroepidemiologic features that enabled between-pathogen comparisons of transmission . Overall , exposure was intense: for most pathogens the window of primary infection was <3 years old; for highest transmission pathogens primary infection occurred within the first year . Longitudinal profiles demonstrated significant IgG boosting and waning above seropositivity cutoffs , underscoring the value of longitudinal designs to estimate force of infection . Seroprevalence and force of infection were rank-preserving across pathogens , illustrating the measures provide similar information about transmission heterogeneity . Our findings suggest antibody response can be used to measure population-level transmission of diverse enteropathogens in serologic surveillance .
A broad set of viral , bacterial , and parasitic enteropathogens are leading causes of the global infectious disease burden , with the highest burden among young children living in lower income countries ( GBD 2016 DALYs and HALE Collaborators , 2016 ) . Infections that result in acute diarrhea and related child deaths drive disease burden estimates attributed to enteropathogens , but asymptomatic infections are extremely common and the full scope of sequelae is only partially understood ( Liu et al . , 2016; Platts-Mills et al . , 2018 ) . Much of what we know about enteropathogen transmission is based on passive clinical surveillance , which reflects a small fraction of all infections . For example , antibody-based incidence of infection to Salmonella enterica and Campylobacter jejuni were 2–6 orders of magnitude higher than case-based surveillance in European populations ( Simonsen et al . , 2008; Falkenhorst et al . , 2012; Teunis et al . , 2012; Teunis et al . , 2013 ) , and a study of Salmonella enterica serotype Typhi in Fiji found similarly high discordance between antibody-based incidence and case-based surveillance ( Watson et al . , 2017 ) . A more complete picture of enteropathogen infection in populations would help understand drivers of transmission , disease burden , naturally acquired immunoprotection , as well as to design public health prevention measures , and measure intervention effects . Stool-based , high-throughput PCR assays have helped solve the logistical difficulties of single-pathogen testing for enterics and have provided new insights into pathogen-specific infections and disease burden ( Liu et al . , 2016; Platts-Mills et al . , 2018 ) . Yet , stool is not routinely collected in population-based surveys , and infection with many globally important enteric pathogens can be sufficiently rare and relatively short-lived to require designs with almost continuous surveillance ( Platts-Mills et al . , 2018; Lin et al . , 2018 ) . At the same time , large-scale serological surveillance platforms create new opportunities for expanded enteropathogen surveillance alongside other infectious diseases ( Metcalf et al . , 2016; Arnold et al . , 2018 ) . These challenges and opportunities have generated interest in antibody-based measurement as a complement to PCR for population-based enteropathogen surveillance ( Griffin et al . , 2011; Exum et al . , 2016; Moss et al . , 2014; Arnold et al . , 2017 ) , and for endpoints in observational and randomized studies ( Crump et al . , 2007; Zambrano et al . , 2017; Chard et al . , 2018; Vargas et al . , 2017; Mosites et al . , 2018; Wade et al . , 2018; Egorov et al . , 2018 ) . After infection , many enteropathogens elicit a transiently elevated antibody response that wanes over time . In lower transmission settings where antibody responses could be monitored longitudinally after distinct infections , Salmonella enterica , Campylobacter jejuni , Cryptosporidium parvum , and Giardia intestinalis ( syn . Giardia lamblia , Giardia duodenalis ) immunoglobulin G ( IgG ) levels in blood have been shown to wane over a period of months since infection; IgM and IgA levels decline even more quickly ( Teunis et al . , 2012; Strid et al . , 2001; Priest et al . , 2001; Priest et al . , 2010; Falkenhorst et al . , 2013; Hjøllo et al . , 2018 ) . Compared with permanently immunizing infections such as measles , transient immunity adds a layer of complexity to seroepidemiologic inference and methods . To our knowledge , there has been no detailed study of enteropathogen seroepidemiology among children in low-resource settings where transmission is intense beginning early in life ( Platts-Mills et al . , 2018 ) . Such studies are needed to determine if serology is a viable approach to measure enteropathogen transmission in low-resource settings . We conducted a series of analyses in cohorts from Haiti , Tanzania , and Kenya that measured serological antibody responses to eight enteropathogens using multiplex bead assays . Our objectives were to identify common patterns in antibody dynamics shared across enteropathogens and populations , and to evaluate serological methods to compare between-pathogen heterogeneity in infection , including estimates of force of infection . Our results provide new insights into the seroepidemiology of enteropathogens among children living in low-resource settings , and contribute advances to inform the design and analysis of surveillance efforts whose goal is to quantify heterogeneity in enteropathogen transmission through antibody response .
The analysis included measurements from cohorts in Haiti , Kenya , and Tanzania . Blood specimens were tested for IgG levels to eight enteropathogens using a multiplex bead assay on the Luminex platform ( Table 1 ) . The Haitian cohort included repeated measurements among children enrolled in a study of lymphatic filariasis transmission in Leogane from 1990 to 1999 ( Lammie et al . , 1998; Hamlin et al . , 2012 ) . Leogane is a coastal agricultural community west of Port au Prince . At the time of the study its population was approximately 15 , 000 , most homes had no electricity and none had running water . In total , the Haiti study tested 771 finger prick blood specimens collected from 142 children ages birth to 11 years old , with each measurement typically separated by one year ( median measurements per child: 5; range: 2 to 9 ) . In Kenya , a 2013 prospective trial of locally-produced , in-home ceramic water filters enrolled 240 children in a serological substudy ( Morris et al . , 2018 ) . Study participants were identified through the Asembo Health and Demographic Surveillance System , which is located in a rural part of Siaya County , western Kenya along the shore of Lake Victoria . Only 29% of the population had piped drinking water ( public taps ) , water source contamination with E . coli prevailed ( 93% of samples tested ) , and the average age children began consuming water was 4 months ( Morris et al . , 2018 ) . Children aged 4 to 10 months provided dried blood spot specimens at enrollment ( February 2013 ) , and again 6 to 7 months later ( August to September , 2013; n = 205 children measured longitudinally ) . The Kenya study period encompassed seasonally heavy rains from March through June . In Tanzania , 96 independent clusters across eight trachoma-endemic villages in the Kongwa region were enrolled in a randomized trial to study the effects of annual azithromycin distribution on Chlamydia trachomatis infection ( Wilson et al . , 2019 ) . The population is very rural , and water is scarce in the region: at enrollment , 69% of participants reported their primary drinking water source , typically an unprotected spring , was >30 min’ walk one-way . From 2012 to 2015 , the Tanzania study collected dried blood spots from between 902 and 1577 children ages 1–9 years old in annual cross-sectional surveys that took place from October through December at the conclusion of the dry season before heavy seasonal rains ( total measurements: 4 , 989 ) . Although children could have been measured repeatedly over the four-year study in Tanzania , they were not tracked longitudinally . There was no evidence that the Kenya and Tanzania interventions reduced enteropathogen antibody response ( Supplementary file 1 ) , so this analysis pooled measurements from the study arms in each population . We estimated seropositivity cutoffs using three approaches: receiver operator characteristic ( ROC ) curve analyses for Giardia , Cryptosporidium , and Entamoeba histolytica including a panel of external , known positive and negative specimens , as previously reported ( Moss et al . , 2014; Morris et al . , 2018 ) ; Gaussian mixture models ( Benaglia et al . , 2009 ) fit to measurements among children ages 0–1 year old to ensure a sufficient number of unexposed children; and , presumed seronegative distributions among children who experienced large increases in antibody levels ( Table 1 ) . Classification agreement was high between the different approaches ( agreement >95% for most comparisons; Supplementary file 2 ) . Among children < 2 years old , antibody levels clearly distinguished seronegative and seropositive subpopulations , but there were not distinct seronegative and seropositive subpopulations by age 3 years for most pathogens measured in Haiti ( Figure 1 ) and Tanzania ( Figure 1—figure supplement 1 ) . By age 3 years , the majority of children were seropositive to Cryptosporidium , enterotoxigenic Escherichia coli heat labile toxin B subunit ( ETEC LT B subunit ) , and norovirus GI . 4 and GII . 4; in all cases antibody distributions were shifted above seropositivity thresholds . In contrast , there was a qualitative change in the antibody response distributions to Giardia , E . histolytica , Salmonella and Campylobacter with increasing age , shifting from a bimodal distribution of seronegative and seropositive groups among children ≤ 1 year old to a unimodal distribution by age 3 years and older ( Figure 1 , Figure 1—figure supplement 1 ) . A direct comparison of age-dependent shifts in antibody distributions to Giardia VSP-3 antigen and Chlamydia trachomatis pgp3 antigen in Tanzania illustrates stark differences in enteropathogen- generated immune responses versus pathogens like Chlamydia that elicit a response that consistently differentiates exposed and unexposed subpopulations as children age ( Figure 1—figure supplement 2 ) . Distributions of IgG levels in the younger Kenyan cohort ( ages 4–17 months ) showed distinct groups of seropositive and seronegative measurements for most antigens ( Figure 1—figure supplement 3 ) . IgG responses to ETEC LT B subunit and cholera toxin B subunit were near the maximum of the dynamic range of the assay for nearly all children measured in the three cohorts ( Figure 1 , Figure 1—figure supplement 1 , Figure 1—figure supplement 3 ) , and these IgG levels waned as children aged , presumably from acquired immunity ( Figure 1 , Figure 1—figure supplement 1 ) . We hypothesized that IgG responses to closely related antigens would co-vary but that IgG responses to unrelated antigens would be uncorrelated . Joint variation in individual-level IgG responses aligned with hypothesized relationships based on antigenic overlap and shared epitopes . Responses to Giardia VSP antigens were strongly correlated in Haiti ( Spearman rank: ρ=0 . 99 ) , Kenya ( ρ=0 . 84 ) and Tanzania ( ρ=0 . 97 ) , as would be expected for antigens with conserved conformational epitopes ( Figure 2 ) . Cryptosporidium ( Cp17 , Cp23 ) and Campylobacter ( p18 , p39 ) antigens were strongly correlated , but high within-individual variability suggests that measuring responses to multiple unique recombinant protein antigens yields more information about infection than measuring responses to one alone ( Figure 2 ) . High correlation between Salmonella LPS Groups B and D , between norovirus GI . 4 and GII . 4 , and between ETEC and V . cholerae likely reflected antibody cross-reactivity . Correlation could also result from multiple previous infections with different Salmonella serogroups or different norovirus genogroups . A comparison across all antigens revealed no other combinations with high correlation ( Supplementary file 3 ) . We excluded cholera toxin B subunit antibody responses from remaining analyses because of the difficulty of interpreting its epidemiologic measures in light of high levels of cross-reactivity with ETEC LT B subunit . Heat labile toxin-producing ETEC is very common among children in low-resource settings ( Platts-Mills et al . , 2018 ) , and there was no documented transmission of cholera in the study populations during measurement periods . For most pathogens , mean IgG levels and seroprevalence rose quickly and plateaued by ages 1 to 3 years in Haiti ( Figure 3 ) , Kenya ( Figure 3—figure supplement 1 ) , and Tanzania ( Figure 3—figure supplement 2 ) . Despite enormous individual-level variation , age-dependent mean IgG curves exhibited characteristic shapes seen across diverse pathogens , and reflected high levels of early-life exposure ( Arnold et al . , 2017 ) . In Haiti , seroprevalence ranged from 66% ( E . histolytica ) to 100% ( ETEC LT B subunit ) by age 3 years ( Figure 3B ) , and in Tanzania , the majority of 1 year olds were already seropositive for Giardia ( 77% ) and Cryptosporidium ( 85% ) ( Figure 3—figure supplement 2B ) . There was some evidence of maternally-derived IgG among children under 6 months old with a drop in mean IgG levels by age , but this pattern was only evident for norovirus GI . 4 in Haiti ( Figure 3A ) and Cryptosporidium Cp17 and Cp23 in Kenya ( Figure 3—figure supplement 1A ) . Age-dependent mean IgG responses to many pathogens declined from a young age , presumably from exposure early in life and acquired immunity . Mean IgG levels declined after age 1 year for Giardia in Haiti ( Figure 3A ) , Giardia and Campylobacter in Tanzania ( Figure 3—figure supplement 2A ) . Based on age-dependent shifts in IgG distributions ( Figure 1 ) , we hypothesized that conversion of IgG levels to seropositive and seronegative status could mask important dynamics of enteropathogen immune response above seropositivity cutoffs , particularly among ages beyond the window of primary infection . In Haiti and Kenya we examined longitudinal IgG profiles among children . In the Haitian cohort , which was followed beyond the window of primary infection , children commonly had >4 fold increases and decreases in IgG while remaining above seropositivity cutoffs—a pattern observed across pathogens but particularly clear for Cryptosporidium ( Figure 4 ) . In Kenya , 4-fold increases in IgG largely coincided with a change in status from seronegative to seropositive , presumably because increases in IgG followed primary infection in the young cohort ages 4–17 months ( Figure 4—figure supplement 1 ) . Many Kenyan children exhibited >4 fold increases and decreases in IgG response to Campylobacter p18 and p39 antigens above the seropositivity cutoff , a result of earlier primary infection and/or additional infection and boosting during the study period ( Figure 4—figure supplement 1 ) . The Kenya study monitored diarrhea symptoms in weekly visits between enrollment and follow-up . The study collected stool from children whose caregivers reported diarrhea symptoms in the past 7 days , and tested stool for Cryptosporidium and Giardia infections using an immunoassay with additional PCR testing for Cryptosporidium as previously described ( Morris et al . , 2018 ) . Among 132 children with paired serology measurements , those infected with Cryptosporidium ( n = 17 ) and Giardia ( n = 25 ) enabled us to compare stool-based measures of infection with IgG responses . Children with confirmed infections in diarrheal stools had higher IgG levels and seroprevalence at both time points compared with those who did not have confirmed infections , but many children without stool-confirmed infections seroconverted during the study . Among children without confirmed Giardia infection in diarrheal stools , seroprevalence to VSP-3 or VSP-5 antigens increased from 1% ( 95% CI: 0% , 5% ) at enrollment to 22% ( 14% , 32% ) at follow-up; among children without confirmed Cryptosporidium infection , seroprevalence to Cp17 or Cp23 antigens increased from 16% ( 10% , 24% ) at enrollment to 47% ( 37% , 57% ) at follow-up . These findings suggest that many children were not shedding genetic material at the time of diarrheal stool collection , or many infections with these two pathogens were asymptomatic . Supplementary file 4 includes additional details . The seroconversion rate , an instantaneous rate of seroconversion among those who are susceptible , is one estimate of a pathogen’s force of infection and a fundamental epidemiologic measure of transmission ( Hens et al . , 2012 ) . Serologically derived force of infection is useful for pathogens that commonly present asymptomatically , such as many enteric infections . Across diverse pathogens , steeper age-seroprevalence curves typically reflect higher transmission intensity ( Corran et al . , 2007; Pinsent et al . , 2018 ) , and age-adjusted seroprevalence equals the area under the age-seroprevalence curve ( a summary measure ) ( Arnold et al . , 2017 ) . We therefore hypothesized that seroprevalence and prospectively estimated force of infection should embed similar information about infection heterogeneity across pathogens . We also hypothesized that standard methods to estimate force of infection from age-structured seroprevalence would underestimate force of infection derived from longitudinal data because of significant antibody boosting and waning above seropositivity cutoffs . Longitudinal designs in Haiti and Kenya enabled us to use individual child antibody profiles to estimate average rates of prospective seroconversion and seroreversion during the studies . We defined incident seroconversions and seroreversions as a change in IgG across a pathogen’s seropositivity cutoff and estimated force of infection as incident changes in serostatus divided by person-time at risk . In a secondary analysis , we defined incident boosting as a ≥ 4 fold increase in IgG to a final level above a seropositivity cutoff and incident waning as ≥4 fold decrease in IgG from an initial level above a seropositivity cutoff . The secondary definition captured large changes in IgG above seropositivity cutoffs , which aligned with repeated boosting and waning observed in the Haitian cohort ( Figure 4 ) . We found a rank-preserving relationship between pathogen seroprevalence and average force of infection in Kenya and Haiti ( Figure 5 ) . Overall levels and steepness of the relationship differed between cohorts , presumably because Kenya measurements were within a window of primary infection for most children ( 4–17 months ) whereas Haiti measurements extended from birth to 11 years and captured lower incidence periods with overall higher seroprevalence as children aged . Consistent with this interpretation , when we progressively narrowed the age range of the Haitian cohort and repeated the analysis , the relationship was steeper when estimated among children ages 0–2 years and flattened as measurements among older children were added ( Figure 5—figure supplement 1 ) . Force of infection varied widely across pathogens in Kenya , ranging from 0 . 1 seroconversions per year for E . histolytica to >5 for Campylobacter ( Figure 5 ) . In Haiti , force of infection ranged from 0 . 3 E . histolytica seroconversions per year to 1 . 1 ETEC seroconversions per year ( Figure 5 ) . Force of infection estimated from 4-fold changes in IgG led to more events and slightly higher rates compared with those estimated from seroconversion alone ( Table 2 ) . For example , Cryptosporidium incident cases increased from 70 to 204 ( a 2 . 9 fold increase ) and the average rate increased from 0 . 6 ( 95% CI: 0 . 5 , 0 . 8 ) to 0 . 9 ( 95% CI: 0 . 7 , 1 . 0 ) per child-year when using a 4-fold IgG change criteria because of substantial IgG boosting and waning above the seropositivity cutoff ( Figure 4 ) . Sensitivity analyses that defined incident boosting over a range of 2-fold to 10-fold increases in IgG showed force of infection estimates were relatively stable across a wide range of definitions . In Haiti , the only pathogen for which force of infection estimated using a 4-fold increase in IgG was significantly higher than the seroconversion rate was Cryptosporidium ( Supplementary file 5 ) . We evaluated whether model-based force of infection estimates from age-structured seroprevalence could accurately recover estimates from the longitudinal analyses . We focused on the Kenya cohort since children were measured repeatedly during the ages of primary infection and because longitudinal force of infection and seroreversion rate estimates varied considerably across pathogens ( Figure 5 ) . We estimated force of infection from seroprevalence curves using methods developed for cross-sectional , ‘current status’ data , a common approach in serosurveillance of vaccine preventable diseases ( Hens et al . , 2012 ) , malaria ( Corran et al . , 2007 ) , and dengue ( Ferguson et al . , 1999; Katzelnick et al . , 2018 ) . Force of infection estimates from semiparametric spline models were similar to estimates from the longitudinal analysis for all pathogens , but had substantially wider confidence intervals owing to the loss of information from ignoring the longitudinal data structure ( Figure 6 ) . Parametric approaches including an exponential survival model ( Jewell and Laan , 1995 ) and a reversible catalytic model ( Corran et al . , 2007 ) yielded narrower confidence intervals than the semiparametric model but tended to underestimate force of infection compared with longitudinal estimates ( Figure 6 ) . Across pathogens , model-based force of infection estimates derived from seroprevalence were rank-preserving compared with nonparametric longitudinal analyses . We conducted a simulation study to investigate whether longer sampling intervals in the cohorts ( 6 months in Kenya , 12 months in Haiti ) could lead us to miss more frequent exposures and thus under-estimate force of infection . For each cohort , we created 100 imputed datasets that reconstructed a child’s daily IgG levels , assuming that each infection boosted IgG and that it would wane exponentially . The simulation drew IgG boosts from empirical distributions in each cohort and used antibody-specific decay rates . We allowed for the maximum number of intermediate exposures between measurements as long as IgG levels could wane sufficiently to follow a child’s empirical measurements , thus providing an approximate upper bound of the seroconversion rates ( force of infection ) that could plausibly be detected for each antibody . We down-sampled the daily datasets at intervals of 30 , 90 , 180 , and 360 days to reflect realistic measurement intervals and estimated seroconversion rates . We found that for most pathogens studied , higher resolution sampling would not substantially increase seroconversion rates in absolute terms . Rates estimated through simulation increased from between 0 . 1 to 0 . 9 episodes per child-year at risk if measured with a sampling interval of 30 days instead of annually ( Haiti ) or every six months ( Kenya ) . However , for pathogens with highest seroconversion rates , ETEC and Campylobacter , increases in rates estimated with 30 day sampling intervals detected a median of 4 to 8 additional seroconversions per child-year at risk compared with empirical rates ( Figure 7 ) . In relative terms , seroconversion rates in Haiti had a larger discrepancy; rates more than doubled when using a 30 day sampling interval , compared with the annual interval used in the study .
Among children in Haiti , Kenya , and Tanzania , antibody-based measures of enteropathogen infection reflected high transmission with primary exposure to most pathogens occurring by age 1–2 years . In low-resource populations , seroincidence rates and force of infection estimated beyond the age range of primary infection ideally should account for IgG boosting and waning above seropositivity cutoffs . Antibodies are a promising approach to measure population-level enteropathogen infection , and seroepidemiologic measures of heterogeneity and transmission are central considerations for their use in trials or in serologic surveillance . Our findings show that for most enteropathogens studied , the ideal window to measure heterogeneity in antibody response closes by ages 2 to 5 years in low-resource settings , and studies that plan to estimate force of infection should favor longitudinal designs with multiple measurements in this early age window .
In Haiti , the human subjects protocol was reviewed and approved by the Ethical Committee of St . Croix Hospital ( Leogane , Haiti ) and the institutional review board at the US Centers for Disease Control and Prevention ( CDC ) . After listening to an overview of the study , individuals were asked for verbal consent to participate . Verbal consent was deemed appropriate by both review boards because of low literacy rates in the study population . With each longitudinal visit , the study team re-consented participants before specimen collection . Mothers provided consent for children under 7 , and children 7 years and older provided additional verbal assent . In Kenya , the human subjects protocol was reviewed and approved by institutional review boards at the Kenya Medical Research Institute ( KEMRI ) and at the US CDC . Primary caretakers provided written informed consent for their infant child’s participation in the trial and blood specimen collection and testing ( Morris et al . , 2018 ) . The original trial was registered at clinicaltrials . org ( NCT01695304 ) . In Tanzania , the human subjects protocol was reviewed and approved by the Institute for Medical Research Ethical Review Committee in Dar es Salaam , Tanzania and the institutional review board at the US CDC . Parents of enrolled children provided consent , and children 7 years and older also provided verbal assent before specimen collection . We transformed IgG levels to the log10 scale because the distributions were highly skewed . Means of the log-transformed data represent geometric means . We summarized the distribution of log10 IgG response using kernel density smoothers . In the Tanzania and Haiti cohorts , where children were measured across a broad age range , we stratified IgG distributions by each year of age <3 years to examine age-dependent changes in the population distributions . To assess potential cross-reactivity between antigens , we estimated pairwise correlations between individual-level measurements in each cohort using a Spearman rank correlation ( Zar , 2005 ) and visualized the relationship for each pairwise combination with locally weighted regression fits ( Cleveland and Devlin , 1988 ) . We compared three approaches to estimate seropositivity cutoffs . Approach 1: External known positive and negative specimens were used to determine seropositivity cutoffs for Giardia VSP-3 and VSP-5 antigens , Cryptosporidium Cp17 and Cp23 antigens , and E . histolytica LecA antigen . Cutoffs were determined using ROC analysis as previously described ( Moss et al . , 2014; Morris et al . , 2018 ) for all antigens except for LecA , VSP-3 , and VSP-5 in Haiti; in these cases , the mean plus three standard deviations of 65 specimens from citizens of the USA with no history of foreign travel were used to estimate cutoffs ( Moss et al . , 2014 ) . Approach 2: We fit a 2-component , finite Gaussian mixture model ( Benaglia et al . , 2009 ) to the antibody distributions among children 0–1 years old , and estimated seropositivity cutoffs using the lower component’s mean plus three standard deviations . The rationale for restricting the mixture model estimation in Haiti and Tanzania to children 0–1 years old was based on initial inspection of the age-stratified IgG distributions that revealed a shift from bimodal to unimodal distributions by age 3 ( Figure 1 ) . This approach ensured that there was a sufficiently large fraction of unexposed children in the sample to more clearly estimate a distribution among seronegative children . Approach 3: In the longitudinal Haiti and Kenya cohorts we identified children < 1 year old who presumably seroconverted , defined as an increase in MFI-bg values of +two or more on the log10 scale . A sensitivity analysis showed that an increase of 2 on the log10 scale was a conservative approach to identify seroconversion for most antibodies considered in this study; an increase of between 0 . 3 to 2 . 16 MFI-bg lead to optimal agreement with ROC-based and mixture model-based classifications in Kenya , and an increase of 0 . 92 to 2 . 41 led to optimal agreement across antigens and references in Haiti ( Supplementary file 5 ) . We then used the distribution of measurements before seroconversion to define the distribution of IgG values among the presumed unexposed . We used the mean log10 MFI-bg plus three standard deviations of the presumed unexposed distribution as a seropositivity cutoff . We summarized the proportion of observations that were in agreement between the three classification approaches , and estimated Cohen’s Kappa ( Cohen , 1960 ) . Additional details and estimates of seropositivity cutoff agreement are reported in Supplementary file 2 . Mixture models failed to estimate realistic cutoff values if there was an insufficient number of unexposed children , which was the case for ETEC LT B subunit and cholera toxin B subunit in all cohorts , and for nearly all antigens in Tanzania where the study did not enroll children < 1 year old ( Table 1 ) . In analyses of seroprevalence and seroconversion , we classified measurements as seropositive using ROC-based cutoffs if available , and mixture model-based cutoffs otherwise . There were three exceptions . By age 1 year , a majority of children across the cohorts had IgG levels near the maximum of the assay’s dynamic range for ETEC LT B toxin and cholera toxin B subunit . The absence of a sufficient number of unexposed children to ETEC LT B toxin , cholera B toxin , and in some cases Campylobacter p18 or p39 led mixture models either to not converge or to estimate unrealistically high seropositivity cutoffs beyond the range of quantifiable levels . For these pathogens , we used seropositivity cutoffs estimated from presumed unexposed measurements in the longitudinal Haiti and Kenya cohorts ( approach 3 , above ) . High levels of agreement between classifications ( Supplementary file 2 ) meant results were insensitive to choice of approach in these cohorts . We classified children as seropositive to Giardia , Cryptosporidium , Campylobacter , or Salmonella if antibody levels against either of the antigens from each pathogen were above estimated seropositivity cutoffs . We estimated mean IgG levels and seroprevalence by age using semiparametric cubic splines in a generalized additive model , specifying binomial errors for seroprevalence , and random effects for children or clusters in the case of repeated observations ( Wood , 2017; Wood , 2012 ) . We also estimated the relationships by age using a stacked ensemble approach called ‘super learner’ that included a broader and more flexible library of machine learning algorithms ( Arnold et al . , 2017; van der Laan et al . , 2007; Polley et al . , 2018 ) , and found similar fits to cubic splines . We estimated approximate , simultaneous 95% confidence intervals around the curves using a parametric bootstrap from posterior estimates of the model parameter covariance matrix ( Ruppert et al . , 2003 ) . Supplementary file 6 includes additional details . In the Kenya and Haiti longitudinal cohorts , we estimated prospective seroconversion rates as a measure of force of infection by dividing the number of children who seroconverted by the person-time at risk between measurements . We defined incident seroconversions and seroreversions as a change in IgG across a pathogen’s seropositivity cutoff . Vaccine immunogenicity and pathogen challenge studies among healthy adults often use a 4-fold increase in antibody levels ( difference of +0 . 6 on the log10 scale ) as a criterion for seroconversion ( Bernstein et al . , 2015; Jin et al . , 2017; Chakraborty et al . , 2018 ) . In a secondary analysis aimed to capture significant changes above a pathogen’s seropositivity cutoff , we defined incident boosting episodes as a ≥ 4 fold increase in IgG to a final level above a seropositivity cutoff , and incident waning episodes as ≥ 4 fold decrease in IgG from an initial level above a seropositivity cutoff . In the secondary definition , individuals were considered at risk for incident boosting episode if they were seronegative , if they experienced a ≥ 4 fold increase in IgG in their first measurement period , or if they experienced a ≥ 4 fold decrease in IgG in a preceding period ( Haiti ) . To estimate person-time at risk used for rates and force of infection , we assumed incident changes were interval-censored and occurred at the midpoint between measurements . We estimated 95% confidence intervals for rates with 2 . 5 and 97 . 5 percentiles of a nonparametric bootstrap distribution ( Wasserman , 2004 ) that resampled children with replacement to account for repeated observations . In the Kenya cohort , we estimated force of infection through age-structured seroprevalence using multiple approaches . There is a long history methods development to estimate force of infection from age-dependent seroprevalence ( Hens et al . , 2012 ) , which is of particular interest to large-scale , cross-sectional surveillance platforms ( Arnold et al . , 2018 ) . Our rationale was to determine if force of infection estimates from age-structured seroprevalence were comparable to estimates from the longitudinal analysis based on incident changes in serostatus . As we show in Supplementary file 7 , the age dependent seroprevalence curve is the difference between the cumulative distribution functions of seroconversion times and seroreversion times . In a special case of no seroreversion , age-specific seroprevalence is thus the cumulative hazard function . The age-specific force of infection can then be estimated as the hazard of seroconverting at age A = a: λ ( a ) = F′ ( a ) / [1 – F ( a ) ] , where F ( a ) =P ( Y | A = a ) is the proportion of the population who are seropositive at age a and F′ ( a ) is the derivative of F ( a ) with respect to a . Key assumptions include stationarity/homogeneity ( i . e . , no intervention or cohort effects ) and that there is no seroreversion ( Hens et al . , 2012 ) . There was no evidence for large changes in transmission during the studies , even due to intervention ( Supplementary file 1 ) . We know for many enteric pathogens children in the Kenya cohort did serorevert ( e . g . , Figure 6 ) ; when assumption is violated , estimates provide a lower-bound of a pathogen’s force of infection . We considered three different estimation approaches for force of infection from age-structured seroprevalence . We conducted a simulation study to investigate whether longer sampling intervals used in the Kenya and Haiti cohorts could lead to under-estimates of seroconversion rates . We modeled a single antigen per pathogen ( e . g . , Cp17 for Cryptosporidium ) . For each cohort , we created 100 simulated datasets that imputed each child’s daily IgG levels . The simulation was designed provide an approximate upper bound on force of infection measured with serology . IgG levels were allowed to continuously boost and wane as long as each child’s IgG trajectory remained consistent with their empirical measurements . We drew IgG boosts for each antibody from the empirical distribution of >4 fold increases in each cohort . We assumed IgG levels decayed exponentially . We estimated exponential decay parameters for each antibody using a subsample of adjacent measurements separated by <1 year with the largest decline in IgG ( bottom quintile of declines ) , selected to reduce the possibility of intermediate exposures . IgG half-life estimates ranged from 51 to 169 days across antibodies in Haiti . In Kenya , too few children experienced reductions in IgG between measures for us to reliably estimate antibody decay rates , so the simulation assumed a fixed IgG decay rate across antibodies that corresponded to a half-life of 69 days , which was broadly consistent with the majority of IgG half-life estimates in Haiti and with 12 week half-life estimates for Cryptosporidium IgG levels among Canadian adults ( Priest et al . , 2001 ) . After simulating daily IgG trajectories for all children , we down-sampled the imputed data at intervals of 30 , 90 , 180 , and 360 days and estimated seroconversion rates and seroreversion per the main analysis . We found that sampling intervals of 30 days adequately summarized even the most dynamic IgG trajectories . As an internal validity check , we compared seroconversion and seroreversion rates estimated from the simulations with empirical rates at comparable sampling intervals ( 180 days in Kenya , 360 days in Haiti ) and found excellent agreement . Across the 100 simulated datasets , we quantified the potential influence of measurement frequency on force of infection estimates by estimating median rates for each pathogen and sampling interval , as well as the median difference in rates between sampling intervals . Supplementary file 8 includes the full simulation and all details . Analyses were conducted in R version 3 . 5 . 3 . Data and computational notebooks used to complete the analyses are available through GitHub ( Arnold , 2019; copy archived at https://github . com/elifesciences-publications/enterics-seroepi ) and the Open Science Framework ( osf . io/r4av7 ) . | Diarrhea , which is caused by bacteria such as Salmonella or by viruses like norovirus , is the fourth leading cause of death among children worldwide , with children in low-resource settings being at highest risk . The pathogens that cause diarrhea spread when stool from infected people comes into contact with new hosts , for example , through inadequate sanitation or by drinking contaminated water . Currently , the best way to track these infections is to collect stool samples from people and test them for the presence of the pathogens . Unfortunately , this is costly and difficult to do on a large scale outside of clinical settings , making it hard to track the spread of diarrhea-causing pathogens . The body produces antibodies – small proteins that can detect specific pathogens – in response to an infection . These antibodies help ward off future infections by the same pathogen , so if they are present in the blood , this indicates a current or previous infection . Scientists already collect blood samples to track malaria , HIV and vaccine-preventable diseases in low-resource settings . These samples could be tested more broadly to measure the levels of antibodies against diarrhea-causing pathogens . Now , Arnold et al . have used blood samples collected from children in Haiti , Kenya , and Tanzania to measure antibody responses to 8 diarrhea-causing pathogens . The results showed that many children in these settings had been infected with all 8 pathogens before age three , and that all of the pathogens shared similar age-dependent patterns of antibody response . This finding enabled Arnold et al . to combine antibody measurements with statistical models to estimate each pathogen’s force of infection , that is , the rate at which susceptible individuals in the population become infected . This is a key step for epidemiologists to understand which pathogens cause the most infections in a population . The experiments show that testing blood samples for antibodies could provide scientists with a new tool to track the transmission of diarrhea-causing pathogens in low-resource settings . This information could help public health officials design and test efforts to prevent diarrhea , for example , by improving water treatment or developing vaccines . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"epidemiology",
"and",
"global",
"health"
] | 2019 | Enteropathogen antibody dynamics and force of infection among children in low-resource settings |
Manipulation of the gut microbiota holds great promise for the treatment of diseases . However , a major challenge is the identification of therapeutically potent microbial consortia that colonize the host effectively while maximizing immunologic outcome . Here , we propose a novel workflow to select optimal immune-inducing consortia from microbiome compositicon and immune effectors measurements . Using published and newly generated microbial and regulatory T-cell ( Treg ) data from germ-free mice , we estimate the contributions of twelve Clostridia strains with known immune-modulating effect to Treg induction . Combining this with a longitudinal data-constrained ecological model , we predict the ability of every attainable and ecologically stable subconsortium in promoting Treg activation and rank them by the Treg Induction Score ( TrIS ) . Experimental validation of selected consortia indicates a strong and statistically significant correlation between predicted TrIS and measured Treg . We argue that computational indexes , such as the TrIS , are valuable tools for the systematic selection of immune-modulating bacteriotherapeutics .
The intestinal microbiota has been shown to critically influence a multitude of host physiological functions , often through modulation of the immune system ( Gevers et al . , 2014; Paun et al . , 2016 ) . Evidence includes studies in germ-free mice , which show that both pro-inflammatory T-helper 17 ( Th17 ) cells and anti-inflammatory regulatory T-cells ( Treg ) are reduced in numbers in the intestinal lamina propria relative to conventional or specific pathogen-free mice ( Atarashi et al . , 2008 ) , and that repopulation with specific bacteria can reconstitute them ( Atarashi et al . , 2015; Ivanov et al . , 2009 ) . One route to how the microbiota influences immune system activation is by the production of small molecules ( Donia and Fischbach , 2015 ) . Microbially produced short-chain fatty acids ( SCFAs ) were shown to facilitate extrathymic differentiation of immune system modulatory Treg ( Arpaia et al . , 2013; Atarashi et al . , 2013; Smith et al . , 2013 ) and are implicated in Treg-dependent anti-inflammatory properties of a mix of 17 human-derived Clostridia strains ( Atarashi et al . , 2013; Tanoue et al . , 2016 ) . Interestingly , individual Clostridia strains only have a modest effect on Treg induction – optimal induction relies on the synergistic interplay of several strains ( Atarashi et al . , 2013; Belkaid and Hand , 2014 ) . Because of these properties , there is currently great interest in the manipulation of this community for the treatment of inflammatory and allergic diseases ( Honda and Littman , 2012; Olle , 2013; Vieira et al . , 2016 ) . Efforts involving transplantation of bacteria from healthy humans to humans with C . difficile infections or with metabolic diseases have provided evidence that microbiota repopulation could be used as possible strategy for disease prevention and treatment ( van Nood et al . , 2013; Vrieze et al . , 2012 ) . For this reason , intestinal supplementation of defined compositions of gut bacteria to treat a range of diseases is currently being pursued by multiple bio-pharmaceutical companies ( Garber , 2015 ) . Determining what microbial consortia can colonize the host and stably coexist with an already resident microbial community , while inducing the desired immune response , is still a major challenge that hinders the translation of these efforts into the clinic ( Maldonado-Gómez et al . , 2016; Weil and Hohmann , 2015 ) . In previous work , some of us identified a set of 17 potent Treg-inducing Clostridia strains to select candidate subsets from ( Atarashi et al . , 2013 ) . The determination of maximally immune-phenotype inducing combinations from these 17 strains is , however , experimentally infeasible as it would require the testing of 217–1 = 131071 possible subsets in mice ( Faith et al . , 2014 ) . Therefore , a computational approach to prioritize which subsets to validate experimentally would have great utility . To our knowledge , to date there has been no model that allows for the simultaneous prediction of the dynamics of both the microbiota and the host immune response . Building on the approach to select for optimal bacterial combinations from ( Faith et al . , 2014 ) and other efforts aimed at identifying potential immune-modulating microbes ( Geva-Zatorsky et al . , 2017; Schirmer et al . , 2016 ) , we overcome this problem by proposing a mathematical modeling-based framework that , by resolving microbiome–immune system interactions with parameters constrained to experimental observations of microbiome and immune effectors , allows computational optimization of immune-stimulating bacterial combinations . To achieve this , we use a series of logically connected analyses that capture CD4+FOXP3+ Treg accumulation in the colonic lamina propria ( Omenetti and Pizarro , 2015; Round and Mazmanian , 2010 ) in response to the dynamics of human-derived Clostridia strains in germ-free mice , which had previously been shown to potently induce Treg expansion ( Atarashi et al . , 2013 ) ( Figure 1 ) . To constrain this proposed model framework to experimental data , we combine previously published ( Atarashi et al . , 2013 ) and newly generated fluorescence-activated cell scanning ( FACS ) data with simulated and newly generated time-series colonization data from germ-free mice ( Figure 2A ) . Subsequently , by applying a microbiome–Treg mathematical model to this combined data set , we infer each strain’s individual contribution to the CD4+FOXP3+ Treg pool ( Figure 2B ) and obtain an estimate of every possible consortium’s Treg induction potential . To justify the usage of predicted mono-colonization concentrations from a previously developed microbiome ecological model ( Bucci et al . , 2016 ) , we validate its ability in predicting temporal dynamics in response to different subsets of Treg-inducing strains ( Figure 3A ) and quantify the deviation of data and predictions ( Figure 3B ) . Introduction of the TrIS , which assigns a score to every predicted steady-state microbial composition , enables us to identify combinations that robustly maximize Treg induction ( Figure 4A ) . We analyze the relationship of TrIS and biomass ( Figure 4B ) as well as metabolic features of the consortia ( Figure 4C and D ) . Most importantly , we demonstrate the utility of our approach in predicting the potency of selected microbial combinations by validating the Treg induction and colonization ability of model-predicted strong , intermediate and weak Treg-inducing consortia in vivo ( Figure 4E ) . We envision that our framework , while in this study tailored to finding combinations ameliorating auto-inflammatory conditions , may also have direct relevance to other immune-system enhancing applications such as the optimal delivery of probiotic-based cancer immunotherapies ( Garrett , 2015 ) .
The goal of this study is to develop a mathematical modeling-based framework to rapidly and systematically select microbial consortia that maximize a desired immune outcome when introduced in a specific host microbial background . To achieve this , we combine ( i ) a microbiome ecological model that accurately describes the dynamics of these bacteria in the host with ( ii ) a microbiome–Treg mathematical model that characterizes the contribution of every strain to the immune phenotype of interest given corresponding microbiome colonization data ( Figure 1 ) . For ( i ) , we gathered newly produced quantitative polymerase chain reaction ( qPCR ) colonization data from gnotobiotic mice and combined it with a previously published microbiome ecological model of the dynamics of 12 Treg-inducing Clostridia strains that are part of an original consortium of 17 Treg-inducing strains discovered by some of us ( Atarashi et al . , 2013 ) ( see also Table 1 for a breakdown of strains used in each study ) . In contrast to the original study of ( Bucci et al . , 2016 ) , we included only 12 of the 13 strains used there because , based on the modeling and analysis of Bucci et al . ( 2016 ) , Strain 6 from the 13-strain set was predicted to not stably colonize in the presence of the other 12 strains . The published colonization data have been reported in our previous work ( Bucci et al . , 2016 ) and include time-series measurements of microbial abundances by qPCR under dietary perturbations . These are used to derive a predictive microbiome ecology model in gnotobiotic conditions based on an extension of the generalized Lotka–Volterra ( gLV ) equations ( Hofbauer and Sigmund , 1998 ) as introduced in Stein et al . ( 2013 ) . For the newly generated dataset , we gavaged 14 mice with one of three possible 11-strain subsets from the 12-strain subset of original 13 strains ( Figure 2 ) and used the derived stool measurements to validate the ability of our mathematical model in predicting unseen conditions . The three 11-strain subsets were chosen based on our ‘keystoneness’ definition , a measure describing the marginal predicted quantitative effect of removing each strain from the full community ( Bucci et al . , 2016 ) . Specifically , we included two 11-strain combinations each missing one of the two highest keystoneness-scoring strains ( VE202 Strain 15 and VE202 Strain 4 ) and one 11-strain combination which lacks the lowest keystoneness scoring strain ( VE202 Strain 29 ) . In analogy to Bucci et al . ( 2016 ) , each strain’s density was profiled over time by qPCR with strain-specific primers ( see Materials and methods ) . To resolve the contribution of each of these to Treg induction for point ( ii ) we coupled the colonization data from fecal content with newly collected and published FACS measurements of the CD4+FOXP3+ Treg population in the lamina propria of these mice ( Figure 2A ) . As it is crucial to capture each strain’s contribution alone and in combination with others , we also included CD4+FOXP3+ Treg measurements from our previously published mono-colonization experiments ( Atarashi et al . , 2013 ) ( Figure 2A ) . Due to the fact that the single-strain mono-colonization concentrations were not measured in these experiments ( Atarashi et al . , 2013 ) , we simulated them using the microbiome ecological model and parameters from ( Bucci et al . , 2016 ) . This choice was supported by the model's capability in predicting unseen validation data ( Figure 3A ) . Spearman’s rank-order correlation coefficient ranges from 0 . 92 to 0 . 98 ( p-value<10−16 ) between observations and predictions depending on the time point ( Figure 3B ) . We used the described microbiome colonization data and corresponding CD4+FOXP3+ Treg measurements to determine the contribution of each strain to the Treg pool . We begin by subdividing the population of CD4+ T-cells into two major subpopulations depending on their intracellular FOXP3 expression: CD4+FOXP3+ Treg and the remainder among the CD4+ T-cells , the conventional CD4+FOXP3− T-cells ( Bilate and Lafaille , 2012; Rudensky , 2011 ) . The concentration of CD4+ T-cells at time t in the colonic lamina propria , cT ( t ) , is then the combination of these two T-cell populations , cT ( t ) =cFOXP3+ ( t ) +cFOXP3− ( t ) . To include a variety of effects into our model , we assume T-cell dynamics to follow the gLV equations ( Gerber , 2014; Hofbauer and Sigmund , 1998 ) , which also account for the effect of the microbial strains in the lumen on the CD4+FOXP3− T-cell subpopulation . In addition , we use an extension of the standard gLV equations to include the impact of the Clostridia strains in terms of external perturbations ( Stein et al . , 2013 ) . The resulting microbiome–Treg mathematical model is found as , ( 1 ) dcFOXP3+ ( t ) dt=cFOXP3+ ( t ) ( αFOXP3++βFOXP3+FOXP3+cFOXP3+ ( t ) ∑k=1K+βFOXP3+FOXP3−cFOXP3− ( t ) +∑k=1Kεikcstrainik ( t ) ) where αFOXP3+ denotes the basal growth rate and βFOXP3+FOXP3+ the self-interaction term of the CD4+FOXP3+ Treg population . The interaction terms βFOXP3+FOXP3- and βFOXP3-FOXP3+ represent the effect of the CD4+FOXP3− T-cell on the CD4+FOXP3+ Treg population and of the CD4+FOXP3+ Treg on the CD4+FOXP3− T-cell population , respectively ( d’Onofrio , 2005 ) . Consequently , positive interaction parameters correspond to activation , negative ones to inhibition . Moreover , εik denotes the effect of strain ik on the CD4+FOXP3+ Treg population . For long-term observations , t→∞ , the dynamics of cFOXP3+ ( t ) are given by its ( non-trivial ) steady-state solution , which simplifies the then static microbiome–Treg mathematical model of the relative CD4+FOXP3+ proportion in steady state , rFOXP3+ , ss , to a linear regression problem , ( 2 ) rFOXP3+ , ss=α~+∑k=1Kε~ikcstrainik , ss . Here , we use that the absolute and relative abundance of the CD4+FOXP3+Treg population are coupled by cFOXP3+/− , ss=cT , ss⋅rFOXP3+/− , ss ( see Materials and methods ) . We assume that the steady-state CD4+ T-cell concentration is constant , cT , ss=const . , across all microbial compositions . We justify this because we are dealing with genetically similar mice and a set of closely related Clostridia . This assumption is however not justifiable when comparing non-colonized and colonized germ-free mice ( Faith et al . , 2014 ) . Assigning to rFOXP3+ , ss the measured FACS-derived CD4+FOXP3+ Treg proportions after 35 days and to cstrainik , ss the corresponding microbial profiles , we infer each strain's contribution to the CD4+FOXP3+ Treg pool ( Figure 2B ) by solving Equation ( 2 ) with an ℓ2-penalized least-square regression with one shrinkage parameter , which is determined in a leave-one-sample-out cross-validation ( Stein et al . , 2013 ) . The resulting normalized root-mean square deviation on left-out samples was found to be 12% . After deriving a model to predict our candidate strains dynamics in germ-free conditions and having resolved each strain’s contribution to Treg expansion , we aim to use this information to computationally select for consortia that maximize Treg induction while being ecologically robust ( Bucci et al . , 2016; Stein et al . , 2013 ) . To specify a measure of ecological robustness as well as immune induction potential for microbial consortia in germ-free mice , we define the Treg Induction Score ( TrIS ) as the average predicted regulatory T-cell activation of a given consortium of K strains straini1 , ⋯ , strainiK while ignoring contributions from the host , ( 3 ) TrIS ( {straini1 , ⋯ , strainiK} ) =1N∑n=1N∑k=1Kε~ikcstrainik , ss ( n ) . If the predicted steady state of the microbial consortium ( cstraini1 ( n ) , … , cstrainiK ( n ) ) ss that is computed from the n-th Markov Chain Monte Carlo ( MCMC ) parameter estimate ( Bucci et al . , 2016 ) is biologically meaningful , i . e . positive , and stable , then cstrainik , ss ( n ) denotes the steady-state concentration of strain ik; otherwise cstrainik , ss ( n ) is set to 0 . Hence , the value of the TrIS is indicative of the expected CD4+FOXP3+ Treg induction ( after removing the host contribution ) and it is of the same unit as the FACS measurements . We evaluated TrIS for every possible strain combination that would stably colonize the gut in germ-free background . In our computation , of the 212–1 = 4095 possible steady-state strain configurations evaluated in N = 22 , 500 MCMC parameter estimates , 84% are found to be biologically meaningful and stable . Interestingly , while the average TrIS increases with consortium size , our analysis shows that a subset size of seven already contains bacterial combinations maximizing induction ( Figure 4A ) . Furthermore , in addition to the strong correlation between TrIS and the predicted total bacterial abundance in the consortium , we observed that high-induction consortia display an especially large enrichment in the abundance of Strain 27 ( Figure 4B , Figure 4—figure supplement 1 ) . Because short-chain fatty acids have been previously associated with colonic Treg induction ( Arpaia et al . , 2013 ) and increase in density upon supplementation with these strains ( Atarashi et al . , 2013 ) , we decided to test if modeling-predicted high Treg-inducing consortia were also enriched in SCFAs . We therefore compared the top 5-inducing microbial consortia of size seven against their same-size counterpart bottom 5 ( Figure 4C ) . We predicted the SCFA concentration for each of the predicted compositions by summing the scaled metabolic outputs of each strain measured in mono-colonization experiments ( Atarashi et al . , 2013; Narushima et al . , 2014 ) and normalized by the strain’s model-predicted mono-colonization density . We performed a Welch two-samples t-test for each of the predicted SCFAs concentrations and found significant enrichment for all estimated SCFAs ( p<0 . 05 , one tailed ) in the high-TrIS consortia compared to the low ones ( Figure 4D ) . We decided to experimentally test our approach's ability to correctly predict consortium ranking with respect to Treg-induction . Due to regulatory constraints on the used probiotic strains – limiting us to a maximum of four strains at a time in follow-up experiments – we selected five 4-strain combinations with a variety of predicted immune effects . We measured CD4+FOXP3+ Treg induction for five microbial consortia of size four and the germ-free control . We chose the two highest TrIS consortia ( H1 , H2 with rank 1 and 2 , respectively ) , the lowest one ( L with rank 495 ) and two TrIS-intermediate consortia of interest ( M1 , M2 with rank 129 and 452 , respectively ) . Strains contained in each of the five consortia are detailed in Table 1 . We correlated the TrIS score with the mean observed CD4+FOXP3+ Treg percentage and found a significant Pearson correlation with coefficient of 0 . 97 and p-value<0 . 01 ( Figure 4E ) . Importantly , when using 16S rRNA sequencing to investigate the resulting colonization profiles for these combinations , we observed that the high TrIS-scoring consortia ( H1 , H2 , M1 ) all stably colonized while the two low-scoring consortia only displayed a subset of the introduced strains . This result remarkably reflects the nature of our scoring system which , in addition to immune activation potential , also incorporates colonization success ( Equation 3 ) . Because the SCFA enrichment analysis from in vivo measurements of individual strains ( Figure 4C ) predicted acetate to be significantly increased in the five high versus low Treg-inducing consortia of size seven , we decided to compare the measured acetic acid concentrations ( Figure 4—figure supplement 2A ) to predictions of acetate in the two high ( H1 and H2 ) and low ( L ) Treg-inducing 4-strain consortia ( Figure 4—figure supplement 2B ) . ANOVA and subsequent post-hoc Tukey test showed significance in the measured enrichment of H1 compared to L ( adjusted p-value<0 . 05 ) and of H1 compared to H2 ( adjusted p-value<0 . 05 ) ( Figure 4—figure supplement 2A ) . However , no statistical difference was observed between measurements in H2 and L ( adjusted p-value<0 . 05 ) . Remarkably , the model-predicted germ-free normalized acetate enrichment and the observed acetic acid concentration ( normalized by the mean in germ-free conditions ) are well correlated ( Figure 4—figure supplement 2B ) .
Manipulation of the intestinal microbiota with defined bacterial consortia for the treatment of disease is a promising route for future therapeutics ( Hansen and Sartor , 2015 ) . However , choosing bacterial combinations from the vast combinatorial space of microbes that effectively colonize a host ( possibly with a dysbiotic microbiota ) and at the same time maximize a desired host phenotype requires an enormous number of expensive and time-consuming experimental trials ( Faith et al . , 2014 ) . While data mining and statistical approaches could aid in this process , the majority of available microbiome analysis methods is still based on correlations ( Gerber , 2014; Morgan et al . , 2015 ) and inept at predicting unseen phenotypes . Identification of causal associations between microbes and host phenotype has been recently achieved by using an experimentally based microbe–phenotype triangulation ( Surana and Kasper , 2017 ) . However , this approach remains impractical when the goal is to explore and rank all possible microbiome consortium combinations with respect to a host phenotype of interest . In this study , we leveraged our previous computational methods to forecast temporal microbiome dynamics and make predictions on stability in the context of infectious and inflammatory diseases including C . difficile colonization and inflammatory bowel disease ( Bucci et al . , 2016; Buffie et al . , 2015; Stein et al . , 2013 ) . Specifically , we have presented a novel modeling-based method that combines dynamical predictions from data-driven models of the microbiome and the interactions between microbes and the immune response ( such as CD4+FOXP3+ Treg ) and thereby enables the rational design of immune-modulating bacterial consortia . Limiting our scope to long-term steady-state dynamics , we were able to derive a microbiome–Treg mathematical model which is constrained by experimental observations from multimodal data . In the presented work , the used data modalities are qPCR and 16S rRNA sequencing data to estimate microbial abundances and FACS to assess CD4+FOXP3+ Treg induction , respectively . However , the proposed framework is naturally extensible to other host data , for example data from metabolite profiling , immune readouts ( e . g . CD8 T-cell activation , Th1/Th17 cell depletion ) or host transcriptome profiling ( Atarashi et al . , 2017; Morgan et al . , 2015 ) . To evaluate candidate consortia that could be relevant to the treatment of auto-inflammatory diseases , we introduced the Treg Induction Score , TrIS , as a novel metric that accounts for both ecological stability and efficacy in immune modulation . This metric allowed us to rationally identify combinations that would stably colonize and at the same time produce substantial Treg induction in germ-free mice . Remarkably , in validation experiments of five distinct modeling-guided consortia with predicted diverse induction potential , we proved the ability of our approach to successfully select microbial combinations with a desired therapeutic activity . To our knowledge , this is to date , the first study in which observation-constrained in silico modeling of microbiome and host phenotype has successfully guided the rational design of drugs of defined bacterial consortia . Our work relies on the gLV model assumption for both microbiome and CD4+ regulatory T-cell dynamics which , despite being very versatile , has some limitations including the lack of third or higher order interactions or saturation effects ( Wangersky , 1978 ) . Moreover , with the data available to us , we needed to assume that the overall T-cell density at steady state is constant across mice and different microbiome compositions . While this assumption was reasonable for the data we analyzed ( see Results ) , the addition of future measurements on total T-cell abundance will likely improve the prediction accuracy of our model as well as provide important insight into the variability of T-cell population size in response to microbial immune induction . We performed our data and modeling analysis in germ-free mice . While previous work by some of us showed that the amount of Treg induction is independent of the germ-free background ( IQI , Balb/c or C57BL/6 ) ( Atarashi et al . , 2015 ) , it is noteworthy that our model , which was trained on IQI mice data , robustly predicted both Treg induction and acetate enrichment of C57BL/6 colonized mice as used in the validation experiments . Our predictions suggest some degree of consistency in terms of functional outcome ( e . g . enrichment in acetate ) of high-scoring consortia relative to low-scoring ones in germ-free mice . However , before translating these findings into therapeutic development , future studies will need to be performed in not germ-free settings ( e . g . SPF mice , humanized mice ) in order to account for the effects of an already-established flora . Individual-specific microbiome features may constrain the colonization potential of the selected strains due to specific ecological network effects ( Smits et al . , 2016 ) , which suggests the need for a careful characterization of the ecological interactions between the proposed probiotic product and a specific recipient community ( e . g . a single ulcerative colitis patient ( Atarashi et al . , 2013 ) ) . Treatment of auto-immune diseases overall is not necessarily achieved by the exclusive optimization of one objective function ( e . g . for ulcerative colitis , the maximization of Treg activation ) , but may need the simultaneous manipulation of a multi-process host immune spectrum , which could include the concomitant reduction of pro-inflammatory phenotypes ( Atarashi et al . , 2015 ) . Given new data availability and careful experimental design , we believe that the modeling framework proposed in this study could overcome these problems by introducing constraints into the scoring metric that account for pre-colonized mouse model features . To the limit , in the presence of a large training data set that would include sufficient information on individual patient variation , our workflow will give us the unprecedented potential of testing microbiota manipulation on a personalized level in silico . Mathematical modeling-based methods have the potential to greatly accelerate the development of treatment of human disease ( Michor and Beal , 2015 ) . In this work , we develop and demonstrate in a first-of-a-kind experimental validation the usefulness of a mathematical microbiome–immune system model . It can be considered a stepping-stone to the accelerated prototyping and rational design of microbiome therapies .
We assume the following dynamics for the CD4+FOXP3+ regulatory T-cell population:dcFOXP3+ ( t ) dt=cFOXP3+ ( t ) ( αFOXP3++βFOXP3+FOXP3+cFOXP3+ ( t ) + βFOXP3+FOXP3−cFOXP3− ( t ) +∑k=1Kεikcstrainik ( t ) ) . Here , αFOXP3+ denotes the basal growth rate and βFOXP3+FOXP3+ the self-interaction term of the CD4+FOXP3+ Treg population , while the interaction parameters βFOXP3+FOXP3- and βFOXP3-FOXP3+ characterize the effect of the CD4+FOXP3− T-cells on the CD4+FOXP3+ Treg population and of the CD4+FOXP3+ Treg on the CD4+FOXP3− T-cell population , respectively ( d’Onofrio , 2005 ) . Moreover , εik denotes the effect of strain ik on the CD4+FOXP3+ Treg population . The non-trivial steady-state solution ( i . e . , the algebraic solution of the right-hand side of Equation ( 1 ) set to 0 with cFOXP3+≠0 ) is found as , cFOXP3+ , ss=−1βFOXP3+FOXP3+ ( αFOXP3++βFOXP3+FOXP3−cFOXP3− , ss+∑k=1Kεikcstrainik , ss ) . Using cT , ss=cFOXP3+ , ss+cFOXP3− , ss , the steady-state concentrations of the CD4+FOXP3+ Treg and CD4+FOXP3− populations , cFOXP3+/− , ss , are derived from the FACS-based relative abundances rFOXP3+/− , ss by , cFOXP3+/− , ss= cT , ss⋅rFOXP3+/− , ss=cT , ss ( 1−rFOXP3−/+ , ss ) . Finally , the linear relationship between the relative abundances , rFOXP3+ , ss , and the strain densities , cstrainik , ss , is found as , rFOXP3+ , ss=1βFOXP3+FOXP3−−βFOXP3+FOXP3+ ( αFOXP3+cT , ss+βFOXP3+FOXP3−+1cT , ss∑k=1Kεikcstrainik , ss ) ≡α~+∑k=1Kε~ikcstrainik , ssassuming constant concentration of CD4+ T-cells , cT , ss=const . , across all possible microbiome compositions . The unknown parameters α~ and ε~ik are estimated in an ℓ2-penalized least-square regression ( so-called ridge regression ) with a positive shrinkage parameter λ , which is determined in a leave-one-sample-out cross-validation as λ*=2 . The colons were collected and opened longitudinally , washed with PBS to remove all luminal contents and shaken in Hanks’ balanced salt solution ( HBSS ) containing 5 mM EDTA for 20 min at 37°C . After removing epithelial cells , muscle layers and fat tissue using forceps , the lamina propria layers were cut into small pieces and incubated with RPMI1640 containing 4% fetal bovine serum , 0 . 5 mg/ml collagenase D , 0 . 5 mg/ml dispase and 40 mg/ml DNase I ( all Roche Diagnostics , Risch-Rotkreuz , Switzerland ) for 1 hr at 37°C in a shaking water bath . The digested tissues were washed with HBSS containing 5 mM EDTA , resuspended in 5 ml of 40% Percoll ( GE Healthcare , Boston , MA ) and overlaid on 2 . 5 ml of 80% Percoll in a 15 ml Falcon tube . Percoll gradient separation was performed by centrifugation at 850 g for 25 min at 25°C . The lamina propria lymphocytes were collected from the interface of the Percoll gradient and suspended in ice-cold PBS . For analysis of Treg , isolated lymphocytes were labeled with the LIVE/DEAD fixable dead cell stain kit ( Life Technologies , Carlsbad , CA ) to exclude dead cells from the analysis . Then , surface and intracellular staining of CD3 , CD4 and FOXP3 was performed using the BV605-labeled anti-CD3 ( 17A2 , Biolegend , San Diego , CA ) , BV421-labeled anti-CD4 ( RM4-5 , Biolegend ) , Alexa700-labeled anti-FOXP3 antibody ( FJK-16 s , eBioscience , San Diego , CA ) , and FOXP3 staining buffer set ( eBioscience ) . The antibody-stained cells were analyzed with LSR Fortessa and data were analyzed using FlowJo software ( Tree Star , Ashland , OR ) . Organic acid concentrations in caecal contents were determined by gas chromatography-mass spectrometry ( GC-MS ) . Caecal contents ( 10 mg ) were disrupted using 3 mm zirconia/silica beads ( BioSpec Products ) and homogenized in extraction solution containing 100 ml of internal standard ( 100 mM crotonic acid ) , 50 ml of HCl and 200 ml of ether . After vigorous shaking using a Shakemaster neo ( Bio Medical Science ) at 1500 rpm for 10 min , homogenates were centrifuged at 1000 g for 10 min and then the top ether layer was collected and transferred into new glass vials . Aliquots ( 80 ml ) of the ether extracts were mixed with 16 ml of N-tert-butyldimethylsilyl-N-methyltrifluoroacetamide ( MTBSTFA ) . The vials were sealed tightly by screwing and heated at 80°C for 20 min in a water bath , and left at room temperature for 48 hr for derivatization . The samples were then run through a 6890N Network GC System ( Agilent Technologies ) equipped with HP-5MS column ( 0 . 25 mm 330 m 30 . 25 mm ) and 5973 Network Mass Selective Detector ( Agilent Technologies , Santa Clara , CA ) . Pure helium ( 99 . 9999% ) was used as a carrier gas and delivered at a flow rate of 1 . 2 ml/min . The head pressure was set at 10 psi with split 10:1 . The inlet and transfer line temperatures were 250 µC and 260 µC , respectively . The following temperature program was used: 60 µC ( 3 min ) , 60–120°C ( 5°C/min ) , 120–300°C ( 20°C/min ) . One microliter quantity of each sample was injected with a run time of 30 min . Organic acid concentrations were quantified by comparing their peak areas with the standards . We used 22 , 500 sets of Markov Chain Monte Carlo generalized Lotka–Volterra parameter sets determined by applying the Bayesian Variable Selection algorithm within MDSINE to the data of Bucci et al . ( 2016 ) , and the first time point of measured microbial profiles for each of the 14 validation mice as initial condition , to simulate the gLV system of differential equations corresponding to each mouse microbiome ( Figure 3A ) . Prediction accuracy was evaluated by calculating the Spearman correlation coefficient between observed and predicted data ( Figure 3B ) . As the experimental data from Atarashi et al . ( 2013 ) only provided CD4+FOXP3+ T-cell levels for the mono-strain colonization experiments at 35 days after inoculation but no measurement of the long-term microbial concentrations in the gut , we used instead the corresponding estimated mono-strain colonization densities obtained from the gLV model ( section above and Figure 2A ) . In general , the long-term behavior of the gLV system is determined by its steady states which are uniquely defined by the inferred model parameters ( Stein et al . , 2013 ) . We computed the parameter median in each single model variable from the 22 , 500 MCMC parameter sets . These median parameters were then used to deduce the steady-state densities , which were together with the measured CD4+FOXP3 proportions included into the training of the microbiome–Treg model . Bacterial strains 4 , 7 , 9 , 14 , 15 , 16 , 27 , 28 , 29 were grown anaerobically in PYG broth ( Peptone , Yeast and Glucose broth from Anaerobe Systems , Cat no: AS-822 ) until they reached stationary phase ( 48 hr for strains 27 and 29 , 24 hr for the remaining strains ) . Each 200 µl-mouse dose of a 4-strain LBP contained 50 µl of 20 times concentrated stationary phase culture . Germ-free C57BL/6 mice aged 6–8 weeks were randomized and gavaged with a total dose of 5·107–2·108 bacteria in a 200 µl , and maintained under gnotobiotic conditions for four weeks . Use of a C57BL/6 background for these experiments was motivated by availability of animals at the facility where we performed the validation and justified by the fact that previous work from us has shown that Treg induction by our Clostridia strains does not differ between Balb/c , IQI , and C57BL/6 mice ( Atarashi et al . , 2015 ) . Mice were then sacrificed , colons harvested , and lamina propria leukocytes isolated and stained for CD3+CD4+FOXP3+ Treg as described above . Eight mice each were used for consortia High 1 ( H1-strains: 7 , 27 , 28 , 29 ) and High 2 ( H2-strains: 4 , 7 , 27 , 29 ) . Five mice each were used for the intermediate high ( M1-strains: 4 , 7 , 14 , 28 ) , and intermediate low consortia ( M2-strains: 9 , 16 , 27 , 29 ) . Three mice were used for Low 1 ( L1-strains: 14 , 15 , 16 , 29 ) . Colonization profiling was determined through 16S rRNA sequencing ( as above ) and verified by blasting representative sequences to a 16S VE202 fasta database . The 4-strain validation experiments were performed in the Massachusetts Host Microbiome Center under IACUC protocol 2016N000141 . | The role of the immune system is to protect the body from infection . It does this by using a powerful toolkit that isolates pathogens and removes damaged tissue . When directed against bacteria and viruses , the system helps to keep the body safe , but an imbalance of the components of the immune system can lead to inflammatory or allergic diseases . The body has built-in mechanisms to shut off the immune response . For example , regulatory T-cells are immune cells with an anti-inflammatory effect , meaning they can switch off inflammation after the immune system has cleared an invasion . If their numbers are too low , it can contribute to unwanted inflammation . In ulcerative colitis , for instance , an unbalanced immune system mistakenly attacks healthy tissue . This causes inflammation and ulcers in the intestine , resulting in bleeding and weight loss . Previous studies revealed that bacteria in the gut might help to control regulatory T-cell numbers . Certain combinations of bacteria can stimulate these regulatory T-cells and potentially dampen inflammation . Tweaking the different populations of microbes in the gut could provide a new way to treat diseases like ulcerative colitis . But , identifying the best combination to use is a major challenge . Testing them all one by one would be extremely challenging . Now , Stein , Tanoue et al . present a mathematical model designed to predict the best microbe-mix to use . The model incorporated data from regulatory T-cells and microbes gathered from mice to estimate the contribution that different strains of bacteria make to regulatory T-cell numbers . This then fed into an ecological model predicted how different combinations of bacteria would behave in mice . The model aimed to fulfill two key criteria . First , to find combinations that can form stable , long-term bacterial colonies alone and when faced with competing microbes . Second , to boost regulatory T-cell numbers , helping them to expand to correct the immune imbalance . To test the predictions , mice received combinations of bacteria suggested by the model . The model had predicted some combinations to be 'weak' at inducing regulatory T-cells , some 'intermediate' and some 'strong' . The results matched the predictions , validating the method . This new model allows the rapid design of microbes that could dampen the immune response in the gut and paves the way for new treatments to correct imbalances of the immune system . By using already available data , it should be possible to expand the model to other diseases with imbalance in the immune system . In future , similar models could also find combinations for other medical uses . For example , to optimise the delivery of cancer immunotherapy , or to modulate the immune system after an organ transplant . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"immunology",
"and",
"inflammation"
] | 2018 | Computer-guided design of optimal microbial consortia for immune system modulation |
Cytoplasmic dynein plays critical roles within the developing and mature nervous systems , including effecting nuclear migration , and retrograde transport of various cargos . Unsurprisingly , mutations in dynein are causative of various developmental neuropathies and motor neuron diseases . These ‘dyneinopathies’ define a broad spectrum of diseases with no known correlation between mutation identity and disease state . To circumvent complications associated with dynein studies in human cells , we employed budding yeast as a screening platform to characterize the motility properties of seventeen disease-correlated dynein mutants . Using this system , we determined the molecular basis for several classes of etiologically related diseases . Moreover , by engineering compensatory mutations , we alleviated the mutant phenotypes in two of these cases , one of which we confirmed with recombinant human dynein . In addition to revealing molecular insight into dynein regulation , our data provide additional evidence that the type of disease may in fact be dictated by the degree of dynein dysfunction .
Motor-mediated intracellular transport is essential for numerous critical cellular processes ( Roberts et al . , 2013; Kardon and Vale , 2009; Vallee et al . , 2012 ) . This is especially apparent in motor neurons , in which cargoes must be transported over long distances to support various neuronal functions . For instance , the soma is the primary site of metabolic function where RNAs and proteins are synthesized and degraded . Thus , to maintain neuronal health , it is critical that cargoes are transported from the soma to the axon terminus ( i . e . , along the axon ) and vice versa . For example , neurofilaments , which provide structural stability to a cell , are transported to the axon terminus by plus end-directed microtubule motors ( kinesins; Xia et al . , 2003 ) and to the soma by the minus end-directed microtubule motor , cytoplasmic dynein ( hereafter referred to as dynein ) ( He et al . , 2005; Wagner et al . , 2004; Shah et al . , 2000 ) . Dynein has also been shown to play a key role in trafficking numerous other neuronal cargoes in various model organisms ( He et al . , 2005; Wagner et al . , 2004; Shah et al . , 2000; Bowman et al . , 2000; Berg et al . , 2000; Barkus et al . , 2008; Hendricks et al . , 2010; Rosa-Ferreira and Munro , 2011; Maday et al . , 2012; Kamal et al . , 2000; Lazarov et al . , 2005; Fu and Holzbaur , 2013; Almenar-Queralt et al . , 2014; Rao et al . , 2017 ) , including autophagosomes ( Ravikumar et al . , 2005; Katsumata et al . , 2010; Cheng et al . , 2015 ) , mitochondria , and ionotropic glutamate receptors ( Horak et al . , 2014 ) . Defects or perturbations in dynein function lead to mislocalization of these cargoes , and results in an accumulation of protein aggregates at the neurite tip ( Ravikumar et al . , 2005; Levy et al . , 2006 ) . In fact , accumulation of misfolded protein aggregates in the neuronal cytoplasm is a common pathological hallmark for motor neuron disease ( He et al . , 2005; LaMonte et al . , 2002; Lin and Schlaepfer , 2006 ) . Consistent with an important role for dynein in neuronal health ( Schiavo et al . , 2013 ) , mouse models with dynein mutations exhibit severe neuropathy , and decreased rates of retrograde axonal transport , among other defects ( Hafezparast et al . , 2003; Chen et al . , 2007; Ori-McKenney et al . , 2010 ) . In addition to its key role in the retrograde trafficking of cargoes , dynein also plays a critical and conserved role during neuronal development by promoting interkinetic nuclear migration ( INM ) in neuronal progenitor cells , and in the subsequent migration of young postmitotic neurons . During the former process , which is critical for neurogenesis , nuclear envelope anchored dynein motors promote migration of the nucleus from the basal to the apical surface of the neuroepithelium where mitotic divisions occur ( Tsai et al . , 2010; Hu et al . , 2013; Del Bene et al . , 2008 ) . Thus , defects in dynein-mediated nuclear migration can lead to defects in early brain development . Given its myriad roles in neuronal processes , it is unsurprising that mutations within the catalytic component of the dynein complex ( the dynein heavy chain , or DHC , which is encoded by the DYNC1H1 gene ) are found in individuals suffering from a wide array of neurodegenerative diseases . For instance , mutations in dynein underlie many cases of malformations of cortical development ( MCD ) , spinal muscular atrophy with lower extremity dominance ( SMA-LED ) , congenital muscular dystrophy ( CMD ) , and Charcot-Marie-Tooth disease ( CMT ) ( Harms et al . , 2012; Tsurusaki et al . , 2012; Scoto et al . , 2015; Peeters et al . , 2015; Strickland et al . , 2015; Weedon et al . , 2011; Fiorillo et al . , 2014 ) . Although mutations in the dynein regulator LIS1 are causative of the MCD disease lissencephaly ( Wynshaw-Boris , 2007; Wynshaw-Boris and Gambello , 2001; Reiner et al . , 1993 ) ( characterized by a smooth brain due to reduced or absent cortical folding ) , mutations in dynein more often lead to polymicrogyria , which is characterized by excessive small folds in the cerebral cortex ( Poirier et al . , 2013; Laquerriere et al . , 2017; Willemsen et al . , 2012; Vissers et al . , 2010 ) . Although a clear link has been established between dynein dysfunction and various neurological diseases , the underlying molecular basis for disease onset or progression is unknown . An unambiguous mechanistic dissection of mutant dynein function in animal cells is complicated by the diverse cellular functions in which dynein participates ( e . g . , axonal transport , centrosome separation , spindle assembly , nuclear envelope breakdown , spindle checkpoint inactivation ) ( Rusan et al . , 2002; Gönczy et al . , 1999; Salina et al . , 2002; Howell et al . , 2001; Wojcik et al . , 2001; Chevalier-Larsen and Holzbaur , 2006 ) , and the difficulties and expense associated with generating and analyzing mutant cell lines ( e . g . , compromised viability , pleiotropism , heterozygosity ) . To overcome these issues , we have employed the versatility and power of the budding yeast Saccharomyces cerevisiae to understand how mutations found in individuals suffering from various neurological diseases lead to dynein dysfunction . In addition to their genetic amenability , low maintenance costs , and rapid generation time , the study of dynein function in budding yeast is simplified by several factors . In contrast to animal cells in which dynein performs numerous functions , the only known function for dynein in budding yeast is to position the mitotic spindle at the future site of cytokinesis ( Li et al . , 1993; Eshel et al . , 1993; Carminati and Stearns , 1997 ) , making functional studies of dynein mutants in this organism simple and unambiguous . As in higher eukaryotes , the yeast dynein complex is comprised of light ( Dyn2 ) , light-intermediate ( Dyn3 ) , intermediate ( Pac11 ) , and heavy chains ( Dyn1 ) , the latter of which is the ATPase that powers motility along microtubules ( see Figure 1A ) ( Markus and Lee , 2011a ) . Whereas in humans , the non-catalytic subunits exist in different isoforms encoded by multiple genes and tissue-specific isoforms ( Pfister et al . , 2006; Raaijmakers et al . , 2013 ) , each of the accessory chains in budding yeast is encoded by only a single gene , enabling simple genetic analysis and manipulation . Moreover , studies have revealed a high degree of structural similarity between yeast and human dynein ( Carter , 2013; Schmidt and Carter , 2016 ) , rendering structure-function studies in this organism relevant and translatable to animal cells . Compounded by the genetic amenability , ease of imaging , and the simple one-step method for isolation of recombinant , motile dynein motors ( Reck-Peterson et al . , 2006; Markus et al . , 2012; Markus and Lee , 2011b ) , budding yeast are a powerful model system for studies of dynein function . For this study , we focused on a library of 17 single point mutations in the dynein heavy chain ( Dyn1 ) , which are found in patients suffering from a broad spectrum of neurological diseases ( Figure 1A ) . These mutations were selected based on their conservation with corresponding residues in yeast dynein . Our findings reveal phenotypic signatures associated with the mutant library that range from partial to complete loss-of-function , and even gain-of-function in some respects . In some cases , the altered function was intrinsic to dynein , while in others , the effects could be attributed to alterations in the activity of the holo-dynein-dynactin complex . The rapid nature of our phenotypic analysis combined with the wealth of structural information available for dynein has enabled us to assess the likely structural basis for dysfunction in two instances . Moreover , consistent with recent findings from another group ( Hoang et al . , 2017 ) , our work reveals a correlation between the degree of dynein dysfunction and disease type . Overall , we describe a rapid and unambiguous set of tools that can be used to understand how disease-correlated mutations in dynein genes lead to onset or progression of disease .
Genomic alterations in individuals suffering from a variety of neurological diseases have been mapped to numerous unique sites throughout the dynein heavy chain ( Figure 1A ) . The tail domain ( ~1400 amino acids ) is the site of interaction for accessory chains ( light-intermediate and intermediate chains ) , as well as adaptors that link dynein to the dynactin regulatory complex and various cellular cargoes ( Urnavicius et al . , 2015 ) . The motor domain ( ~3000 amino acids ) forms the catalytic core of dynein , where ATP binding and hydrolysis is translated into movement of the linker element that powers motility along microtubules ( Carter , 2013; Schmidt et al . , 2015; Roberts et al . , 2013 ) . Although the genetic basis for dyneinopathies is known , there exists very limited data on how these mutations are causative of motor dysfunction . Indeed , the mutations map throughout the entire heavy chain , with no clear correlation between disease state ( i . e . , symptoms , severity , age of onset ) and mutation identity . To understand how disease-correlated point mutations affect dynein function , we employed a series of well-established methodologies to assess the function of 17 single point mutants in budding yeast . The first such assessment was to determine if the mutant motors were capable of correctly positioning the mitotic spindle , the only known function for dynein in budding yeast . To this end , we performed a spindle positioning assay using haploid yeast cells expressing mRuby2-Tub1 ( α-tubulin; to visualize the mitotic spindle ) and the mutant dynein from the native dynein locus . Given the dispensable nature of dynein function for yeast cell viability ( Li et al . , 1993; Eshel et al . , 1993 ) , a complete loss-of-function mutant would not be expected to compromise viability , but to simply affect spindle position . In this assay , single time-point images of mutant cells are acquired , and the position of the mitotic spindle is deemed to be either correct ( i . e . , the anaphase spindle extended through the bud neck along the mother-bud axis ) or incorrect ( see Figure 1B , left ) . This analysis revealed that eight mutants exhibited varying degrees of spindle positioning defects that significantly differed from wild-type ( Figure 1B , right ) . All of those mutants with defects were those with substitutions in the motor domain , which encompasses the six AAA ( ATPase associated with various cellular activities ) domains , the linker element that performs the powerstroke , and the microtubule-binding domain ( MTBD ) . All but one of the motor domain mutations led to significant defects . Seven out of the nine mutations linked to malformations in cortical development ( MCD ) were among those with defects in this assay , whereas none of the mutations associated with the other neurological diseases exhibited significant defects . Although most of these mutants differed to a small but significant degree from wild-type , H3639P exhibited defects as severe as loss of DYN1 ( dyn1∆ ) . Given the somewhat binary nature of the spindle positioning assay , it provides only a coarse assessment of mutant functionality . Thus , as a more sensitive readout of mutant dynein function , we imaged dynein-mediated spindle movements in yeast cells and quantitatively assessed various parameters of these movements . Given the reliance of dynein on dynactin for this activity in cells , assessment of spindle movements is in fact a read-out of dynein-dynactin activity . To ensure that spindle translocation events were a consequence of dynein-dynactin activity , we performed these assays in cells deleted for KAR9 , a genetic component of a pathway that promotes orientation of the spindle along the mother-bud axis ( Yin et al . , 2000; Hwang et al . , 2003; Liakopoulos et al . , 2003 ) ( see Figure 2—figure supplement 1 ) . Moreover , we treated these cells with hydroxyurea ( HU ) , an inhibitor of DNA synthesis that arrests yeast in a prometaphase-like state that precludes anaphase onset . This relatively simple but sensitive assay permits detection of subtle defects ( or enhancements ) in the motility parameters of dynein-dynactin ( Moore et al . , 2009 ) . In addition to obtaining velocity and displacement values , this assay also provides a readout for relative activity ( i . e . , how active dynein-dynactin is within the cell ) . Moreover , by scoring for ‘neck transit’ success frequency – events in which a dynein-dynactin-mediated spindle translocation results in successfully transiting the narrow mother-bud neck – we are also able to determine if there are potential defects in force production . Previous studies have shown that neck transits are compromised in cases where dynein’s microtubule-binding affinity is weakened ( Ecklund et al . , 2017 ) , or when the CAP-gly domain of Nip100 ( homolog of human p150 component of the dynactin complex ) is genetically ablated ( Moore et al . , 2009 ) . Subsequent to HU arrest , full Z-stacks of the mitotic spindle and astral microtubules ( via GFP-Tub1 ) were acquired every 10 s ( see Figure 2A for example ) , and the position of the spindle was subsequently tracked using a combination of manual ( e . g . , to assess neck transit success ) and automated 3-dimensional tracking ( using a custom written Matlab-based routine ) . This analysis revealed a broader more nuanced array of defects in our library of mutants ( Figure 2B–F , and Figure 2—figure supplement 2A and B ) . Strikingly , all mutants exhibited varying degrees of alterations in their motility parameters with respect to wild-type cells . At the most severe end of the spectrum , H3639P cells – which also had the most severe spindle positioning phenotype – exhibited only 10 dynein-mediated spindle displacement events from all cells observed ( compare to a mean of 244 events for all other strains; see Figure 2—figure supplement 2A and B for scatter plots illustrating density of datasets ) . Consequently , this mutant had extremely low ‘activity’ parameters ( i . e . , total spindle displacement , Figure 2E; and , number of dynein-mediated spindle movements per minute , Figure 2F ) . The mutant with the second most severe spindle positioning phenotype , R1852C , exhibited significant defects in all motility metrics , including velocity , displacement ( per event ) , neck transit success , and the two activity parameters . The relative phenotypic severity of these two mutants is consistent with findings from another group in which disease-correlated dynein mutants were assessed using recombinant human dynein complexes ( Hoang et al . , 2017 ) . This group identified the human equivalents of R1852C ( R1962C ) and H3639P ( H3822P ) as being the most severe loss-of-function mutants in their reconstituted motility assays ( see Discussion ) . Another noteworthy mutant was K1475Q , a substitution within the linker domain , the mechanical element that is responsible for the powerstroke ( Figure 2—figure supplement 3 ) . This mutation led to a reduction in spindle velocity , displacement , and also reduced the activity metrics of the motor . Given the position of this mutation , it may affect dynein activity by compromising linker remodeling during the priming or powerstroke movement of the linker . However , a recent study identified the equivalent human residue ( R1567 ) as being at least partly required for the formation of an autoinhibited conformation of human dynein called the phi-particle ( Zhang et al . , 2017 ) ( due to its resemblance to the Greek letter; Amos , 1989 ) , thus raising the possibility that yeast dynein also adopts this conformation . Thus , the altered motility of this mutant may be a consequence of altered activity regulation ( see below and Discussion ) . All three mutations within the AAA3 module of the motor domain led to varying degrees of spindle motility alterations ( D2439K , R2543K , and L2557M; Figure 2—figure supplement 3 ) . Interestingly , the most striking defect we observed with L2557M was a reduction in the neck transit success rate ( Figure 2D ) , which is the likely cause of the spindle positioning defect . This suggests that this mutation – which is buried within the large subdomain of AAA3 , and makes hydrophobic contacts with a closely apposed helix – is likely affecting force generation by the motor . All three microtubule-binding domain ( MTBD ) mutants exhibited fairly similar degrees and types of defects in effecting spindle movements ( reductions in velocity , displacement , and neck transit success ) . Structural analysis revealed that all three mutations mapped to the surface of the MTBD that makes contacts with the microtubule ( Figure 2—figure supplement 3 ) . Given all three substitutions lead to loss of a positive charge , it is likely that these mutations each lead to a reduction in the affinity of the motor for the negatively charged surface of the microtubule . The reduced activity metrics for R3152N and R3201N could thus be a reflection of a reduction in association kinetics of the mutants for the microtubule . This is supported by a study that found reduced microtubule binding for similar amino acid substitutions in a mouse dynein MTBD fragment ( Poirier et al . , 2013 ) . Although none of them led to significant spindle positioning defects , all of the tail domain mutations led to altered spindle motility parameters . Structural analysis revealed that most of these mutations ( 7 out of the 8 ) clustered to two distinct regions: ( 1 ) adjacent to , or within the N-terminal dimerization domain , or ( 2 ) at a surface that interfaces with the intermediate chain of a neighboring heavy chain in a two dynein:1 dynactin complex ( Figure 2—figure supplement 4 ) . This latter region was recently identified as being important to stabilize the binding of a second dynein complex to dynactin , and ensuring that all four heavy chains are properly aligned for efficient motility of the human dynein-dynactin complex ( Urnavicius et al . , 2018 ) . Our data suggest that the ability to recruit two dynein complexes to dynactin is conserved in yeast , and that disrupting this complex can compromise force generation ( Figure 2D ) or activity ( Figure 2E and F ) . Although the last tail domain mutation , W612C , is in close proximity to this latter region , it is sufficiently far from the contact point with the second dynein complex to suggest that it is likely affecting some other facet of dynein-dynactin function . Given the heterozygous nature of these mutations in affected patients , we wondered how the mutants would behave in the presence of a second , wild-type copy of dynein . It is currently unclear whether cells with two copies of DYN1 ( e . g . , wild-type and mutant ) assemble dynein complexes comprised of two different copies ( e . g . , wild-type/mutant heterodimers ) , or whether they are comprised of only one copy ( e . g . , wild-type homodimers , or mutant homodimers ) . This latter phenomenon could be a consequence of co-translational dimerization ( Natan et al . , 2017 ) . To distinguish between these two possibilities , we generated diploid yeast strains that contained one copy of a GFP-tagged dynein heavy chain ( DYN1-GFP ) , and a second copy that was fused to an N-terminal affinity tag and a C-terminal HALO tag ( ZZ-DYN1-HALO ) , the latter of which could be used to fluorescently label the motor ( Figure 3A ) . Lysate from these cells was subjected to affinity chromatography , and subsequent to incubation with a red fluorescent HALO ligand ( HALO-TMR ) , the bound protein was eluted and used in a single molecule imaging experiment . If heterodimers assemble within cells , we expected to observe dual-color labeled molecules ( green and red ) ; however , if only homodimers form , then we expected to observe only red molecules ( see Figure 3A ) . Although we observed a small number of motile green molecules ( 0 . 9% of the total; likely due to contaminating Dyn1-GFP molecules in the protein preparation ) , and a single dual labeled molecule ( 0 . 3% of the total; Figure 3B , arrow ) , the vast majority of motile molecules ( 98 . 8% ) were exclusively red , indicating that dynein very rarely , if ever , forms heterodimers ( Figure 3C ) . This suggests a co-translational dimerization model for dynein complex assembly , similar to what has been observed for p53 ( Nicholls et al . , 2002 ) . Next , we wished to recapitulate the heterozygous nature of some mutations in our budding yeast system . To this end , we chose three mutants – E545V , R1852C , H3639P – and determined whether they could affect dynein activity in heterozygous diploid yeast strains . We mated haploid wild-type cells with haploid mutant cells to generate heterozygous diploid cells ( e . g . , DYN1/dyn1R1852C; Figure 3D ) . In addition to comparing dynein activity in these cells to that from homozygous wild-type cells ( i . e . , DYN1/DYN1 ) , we also compared them to hemizygous cells with only one copy of DYN1 ( i . e . , DYN1/dyn1∆; Figure 3E–I , and Figure 3—figure supplement 1 ) . Although homozygous wild-type cells exhibited similar dynein-dynactin activity to the hemizygotes , the velocity was somewhat reduced in the latter , suggesting a critical concentration of dynein is required for effecting maximal spindle velocity ( Figure 3F , and Figure 3—figure supplement 1 ) . Analysis of the heterozygous mutants revealed that all three exhibited partial loss-of-function phenotypes in almost all assays with respect to the hemizygous cells , indicating that they are all indeed dominant alleles . Taken together , these data indicate that homodimers of mutant dynein are sufficient to compromise the activity of wild-type homodimers within the cell . Findings from our spindle tracking assay revealed the consequences of mutations on various parameters of dynein-dynactin-mediated spindle movements . These movements are mediated by cortically anchored dynein-dynactin complexes that are regulated at various levels by numerous effector molecules ( e . g . , Pac1 , Ndl1 , Num1 , She1; Markus et al . , 2012; Markus and Lee , 2011b; Li et al . , 2005; Lammers and Markus , 2015 ) . For instance , cortical targeting of dynein is affected by various molecules , including Pac1 , which tethers dynein to microtubule plus ends ( Lee et al . , 2003 ) , and Num1 , which anchors dynein-dynactin complexes to the cortex ( Heil-Chapdelaine et al . , 2000 ) . Thus , mutations that alter the ability of dynein to interact with or be affected by these molecules will result in alterations in dynein-dynactin-mediated spindle movements . To determine whether mutations affect dynein-intrinsic activities , it is therefore important to specifically assess dynein activity without these complicating factors . To this end , we employed a single molecule motility assay , in which the movement of individual purified dynein motors is quantitatively assessed . Performing this assay using yeast dynein has one key advantage over human dynein: unlike human dynein , which requires dynactin and one of several adaptor molecules for processive single molecule motility ( e . g . , BicD2 , Hook3; McKenney et al . , 2014; Schlager et al . , 2014 ) , yeast dynein is a processive motor without these factors ( Reck-Peterson et al . , 2006 ) . This somewhat unique property of yeast dynein thus permits an unbiased assessment of dynein-intrinsic motility . Single molecule motility analysis of each mutant revealed that thirteen of those that exhibited defects in the spindle tracking assay showed some degree of motility alteration in the in vitro assay ( Figure 4; see Figure 4—figure supplement 1 for scatter plots and some example kymographs ) . Interestingly , the precise defect in vivo was not always predictive of the alteration in dynein motility in vitro . Although the reasons for this are unclear , they are likely due in part to the differences in requirements for spindle transport versus those for unloaded single molecule motility ( i . e . , in which there is no resistance to transport ) . Additionally , small defects in single motor motility may lead to more pronounced defects in the context of a motor ensemble , as may be the case for dynein at the cell cortex ( Markus et al . , 2011 ) . Such mutants included W612C and R1852C , both of which led to a reduction in all spindle tracking metrics in cells , but had very little effect on the displacement ( run length ) of single molecules in vitro . Similarly , R2543K , which reduced spindle velocity by ~50% had only a minor effect on single molecule velocity values . In a few cases , we observed little or no difference from wild-type in single molecule motility parameters in spite of differences in the spindle tracking assay ( i . e . , E109I , K540C , E545V , I554M , D2439K , L2557M ) . In two of these cases ( E545V and L2557M ) , one of these changes was a reduction in the neck transit success rate ( see Figure 2D ) . As discussed above , this phenomenon is potentially a readout of force generation . Given the unloaded nature of single molecule motility , this assay would be unable to detect differences in force generation by dynein . Although the molecular determinants of dynein force production are not well understood , it is possible that these mutations specifically affect the ability of dynein to remain bound to microtubules under conditions of high load . Although most mutants exhibited loss-of-function phenotypes in the in vitro assay , a few mutations led to gain-of-functions . For instance , R241L , N283R , and K1475Q all led to an increase in velocity , while K1475Q also caused an increase in run length and a small but significant increase in the fraction of active motors ( see Figure 4—figure supplement 1C for example kymograph ) . We confirmed the increased run length for K1475Q was not a consequence of motor aggregates , which could presumably lead to an increase in apparent processivity ( Derr et al . , 2012 ) ( Figure 4—figure supplement 2 ) . As mentioned above , R241 and N283 are adjacent to the N-terminal dimerization domain , and K1475 is within the linker domain ( see Figure 2—figure supplements 3 and 4 ) . In the cases of R241L and N283R , the gain-of-functions observed in vitro are possibly causative of the in vivo deficiencies , indicating that a faster or more processive motor is not advantageous for the spindle positioning function ( see below and Discussion regarding K1475Q ) . This is consistent with recent studies that showed gain-of-functions in dynein ( or dynein regulators ) can lead to defects in dynein-mediated neuronal maturation ( Huynh and Vale , 2017 ) , or spindle orientation ( Zhang et al . , 2017 ) . In summary , in all cases in which motility parameters were altered in vitro as a consequence of a particular mutation , we are able to conclude that the underlying molecular defect is most likely intrinsic to dynein itself , and is likely not a consequence of an altered interaction with either accessories or regulators . Our in vivo and in vitro functional data described above revealed the particular parameters of dynein motility that were altered by the mutations , and also whether the altered motility was in fact intrinsic to dynein ( i . e . , if there was an in vitro phenotype ) , or was possibly a consequence of alterations in interactions with regulators such as dynactin . As discussed above , proper dynein function in yeast relies on the coordinated action of various molecules to localize dynein to microtubule plus ends , from where it is offloaded to its site of action: the cell cortex ( Figure 5A ) ( Markus and Lee , 2011b ) . For instance , dynein plus end localization occurs in a dynactin-independent manner ( Moore et al . , 2008 ) , but requires an interaction with Pac1 ( the LIS1 homolog ) ( Lee et al . , 2003 ) , as well as the accessory chains Dyn3 ( Markus and Lee , 2011a ) ( light-intermediate chain ) and Pac11 ( intermediate chain ) ( Lee et al . , 2005 ) . In contrast , dynactin is required for dynein to localize to cortical Num1 receptor sites ( Moore et al . , 2008 ) . Thus , quantitative assessment of dynein localization can reveal potentially altered interactions with various regulators or accessories . For this analysis , we focused on a select group of mutants: those with substitutions within the N-terminal tail domain ( which is the site for interaction with the accessory chains , dynactin , and Num1; Markus et al . , 2009 ) , those with the most severe phenotypes ( R1852C , R2543K , and H3639P ) , and two of the MTBD mutants . We acquired time-lapse images of haploid cells expressing fluorescently-labeled tubulin ( mRuby2-Tub1; α-tubulin ) and 3GFP-tagged copies of each dynein mutant . Plus end and cortical foci were identified from movies , and their targeting frequency ( i . e . , percent cells with foci ) and fluorescence intensities ( a readout of molecule number per site ) were quantified . This analysis revealed significantly altered localization frequencies or intensities for most of the mutants , including L213I , N283R , E545V , I554M , W612C , K1475Q , R1852C , R2543K , and H3639P ( Figure 5B–D ) . For instance , the K1475Q substitution ( within the linker ) led to a large increase in the frequency of observing cortical foci ( 2 . 2-fold; p=0 . 0011 ) , but a concomitant reduction in the number of molecules per cortical focus ( 38% reduction in fluorescence intensity; p=0 . 0117 ) , whereas W612C , R1852C and H3639P all exhibited a reduction in dynein levels at plus end and cortical sites ( see Discussion ) . Finally , immunoblotting revealed that protein expression differences likely do not account for the observed localization or activity phenotypes ( Figure 5—figure supplement 2 ) . The increased frequency of cortical foci for K1475Q suggests that this mutant may exhibit higher affinity for Pac1 and/or dynactin , the former of which is limiting for plus end association , and the latter of which is limiting for offloading to cortical Num1 sites ( Markus et al . , 2011 ) . Given the importance of the human equivalent of K1475 ( R1567 ) in stabilizing the autoinhibited ‘phi’ particle conformation – which exhibits lower affinity for dynactin than the open , uninhibited state ( Zhang et al . , 2017 ) – we wondered whether K1475Q interacts more readily with dynactin . To address this , we measured the relative ratio of dynein ( using Dyn1-3GFP ) to dynactin ( Jnm1-3mCherry , homolog of p50/dynamitin ) at microtubule plus ends , where dynactin recruitment is wholly dependent on dynein ( Moore et al . , 2008 ) . Since assembled dynein-dynactin complexes are readily offloaded to cortical Num1 receptor sites ( Markus and Lee , 2011b ) , we chose to perform this analysis in cells lacking Num1 ( num1∆ ) to exclude any contribution from the offloading process to the relative ratio of dynein to dynactin at plus ends ( Figure 5—figure supplement 1A ) . Interestingly , this analysis indeed revealed a significantly increased ratio of dynactin to dynein at plus ends ( Figure 5—figure supplement 1B; from a mean of 1 . 07 to 1 . 52; p<0 . 0001 ) , indicating that K1475Q interacts more readily with dynactin than wild-type dynein , and that this mutation may in fact be disrupting a potential autoinhibited state of yeast dynein . Given the wealth of structural information currently available for dynein , we sought to determine the structural basis for dysfunction in two of the mutants: R1852C and H3639P . Analysis of available dynein structures revealed that H3639 is situated within a conserved loop that connects two alpha helices , one of which is an extension of the buttress , a structural element that helps to communicate nucleotide-dependent structural rearrangements within the AAA ring to the MTBD ( Figure 6A ) ( Schmidt et al . , 2015 ) . We first asked whether gain of proline or loss of histidine is the cause for the severe loss-of-function . To this end , we substituted H3639 with either a serine ( to preserve the polar nature and hydrogen-bonding capabilities of histidine ) , valine ( a hydrophobic residue ) , or asparagine ( found in the equivalent position of human dynein-2 ) , and assessed the activity of these mutants in the spindle positioning assay . None of these substitutions led to significant spindle positioning defects , indicating that gain of proline is the reason for dynein dysfunction in H3639P ( Figure 6B ) . We next asked if proline substitutions are tolerated at other sites within this inter-helical loop . Using the spindle positioning assay , we found that proline was well tolerated at all sites within the inter-helical loop except for position 3641 ( 2 residues C-terminal to 3639; Figure 6B ) . Thus , introduction of a proline at two distinct sites within this loop are sufficient to severely compromise dynein function . We hypothesized that prolines are not tolerated in these two regions ( residues 3639 and 3641 ) of the inter-helical loop because of the structural constraints imposed by prolines due to bond angle restrictions . If this were true , then we reasoned that increasing the structural flexibility in the immediate vicinity of P3639 might rescue the proline-dependent defects . To test this , we introduced glycine substitutions at either the N-terminal residue ( F3638 ) , the C-terminal residue ( W3640 ) , or both . For reasons that are unclear , both double mutants ( i . e . , F3638G H3639P , and H3639P W3640G ) led to spindle positioning defects that were significantly more severe than H3639P and dynein knock-out cells ( dyn1∆; p<0 . 0001; Figure 6B ) . Strikingly , however , the triple mutant – in which P3639 is flanked by glycines ( i . e . , F3638G H3639P W3640G ) – exhibited spindle positioning defects that were significantly less severe than H3639P ( p<0 . 0001 ) . To confirm these findings , we assessed the activity of the triple mutant using the spindle tracking assay . Although the activity of this mutant was much less than that of wild-type dynein-expressing cells , it was significantly greater than the H3639P single mutant ( p≤0 . 0361; Figure 6C–E ) . In spite of the increase in activity , the quality of the spindle movements ( velocity and displacement per event ) were nearly identical between H3639P and the triple mutant ( Figure 6—figure supplement 1A–D ) . Interestingly , the single molecule assay revealed that the triple mutant was no better than the single mutant in any metrics ( Figure 6—figure supplement 1E–I ) . This indicates that the triple mutant does not rescue dynein motility , but some other metric of in vivo dynein ( or dynein-dynactin ) activity . We next performed live cell imaging to determine if the triple mutant alters any aspects of dynein localization . Although the flanking glycines did not rescue plus end or cortical localization , it did lead to a large increase in the fraction of cells exhibiting dynein foci at the spindle pole bodies ( SPBs; Figure 6F; p<0 . 0001 ) , suggesting that the triple mutant shifts the balance toward more semi-active – or properly folded – dynein within cells with respect to the single H3639P . The relevance of dynein localization to the SPB is unclear , but previous work from our lab demonstrate that this localization requires the MTBD ( Lammers and Markus , 2015 ) . In light of our other observations , the increased localization frequency of the triple mutant suggests that the flanking glycines rescue H3639P cellular defects by reducing proline-dependent inflexibility , which in turn may prevent global misfolding of Dyn1 , thus increasing the relative concentration of active dynein within the cell . If H3639P causes some fraction of Dyn1 to be misfolded within the cell , then we reasoned that increasing the total amount of protein in cells would lead to an increase in the number of active Dyn1 molecules , and thus an increase in the apparent degree of H3639P localization . To test this , we assessed dynein localization in the absence or presence of MG132 , a proteasome inhibitor , addition of which would lead to an increase in total protein content , including dynein . Addition of MG132 increased the frequency of wild-type cortical dynein foci by approximately 2-fold ( p=0 . 0013 ) , but had no significant effect on the frequency of either SPB or plus end targeting ( Figure 6G ) . Consistent with our hypothesis , MG132 treatment increased the frequency of plus end targeting of H3639P by 3 . 7-fold ( p<0 . 0001 ) to levels comparable to wild-type dynein; however , the frequency of cortical targeting of H3639P was unaffected by proteasome inhibition . Close inspection revealed a cysteine situated within very close proximity to R1852 ( ~3 Å; Figure 7A ) , which lies within the first AAA module ( AAA1 ) . Given the mutation results in a cysteine substitution at this site , we reasoned that an ectopic disulfide bond may be responsible for the phenotypic consequences . To determine whether this was the case , we mutated the closely apposed , highly conserved cysteine to a serine ( C1822S; Figure 7B ) , which would eliminate the potential for disulfide bond formation at this site . We then used several of our assays to quantitate the degree to which C1822S might rescue defects due to R1852C . Although C1822S alone compromised all parameters of dynein-dynactin-mediated spindle movements , this substitution led to a partial restoration of most of the parameters in the R1852C mutant ( i . e . , C1822S R1852C double mutant; Figure 7C and D , and Figure 7—figure supplement 1A–C ) . We observed a similar restoration of function in a C1822A R1852C double mutant . We ruled out the possibility that the partially hydrophobic nature of cysteine is the cause for the dysfunction in R1852C by assessing an R1852V mutant , which exhibited defects that were significantly less severe than R1852C . Moreover , combining C1822S with R1852V led to no degree of rescue with respect to the single R1852V mutant , which is in stark contrast to our observations with the double C1822S R1852C mutant ( Figure 7C and D , and Figure 7—figure supplement 1A–C ) . We also found that C1822S rescued the localization of R1852C to plus ends and the cell cortex almost to wild-type levels ( Figure 7—figure supplement 1D ) . As with the spindle tracking assay , although R1852V led to localization defects , the C1822S R1852V double mutant was almost identical to R1852V alone . Before assessing the extent of rescue with the single molecule assay , we first tested whether using a minimal dynein fragment that is sufficient for processive motility ( Reck-Peterson et al . , 2006 ) would rescue any of the motility defects we observed with the R1852C mutant ( GST-dynein331 ) . This fragment – a glutathione S-transferase ( GST ) -dimerized motor domain that exhibits motility parameters that are very similar to the full-length molecule ( Reck-Peterson et al . , 2006 ) – lacks the tail domain , and thus does not co-purify with or rely on any of the accessory chains . If protein misfolding is partly to blame for mutant dysfunction – as suggested from the single molecule ( only 23 . 2% active motors; Figure 4D ) and localization assays ( severe reduction in localization frequency; Figure 5B ) – then we reasoned that the simplicity and compact fold of GST-dynein331 might rescue some of these defects . In striking contrast to the full-length mutant , the fraction of active GST-dynein331 R1852C motors was almost identical to that of the wild-type GST-dynein331 motor ( note the same construct did not rescue the H3639P mutant; Figure 7—figure supplement 1E ) , suggesting that the holoenzyme complex is more susceptible to defects arising from this mutation than is the truncated motor domain . However , much like the full-length molecule , the velocity of the GST-dynein331 mutant was severely reduced with respect to wild-type ( Figure 7E and Figure 7—figure supplement 1G ) . Consistent with the spindle tracking data , C1822S was indeed sufficient to significantly rescue the single molecule velocity defect in the minimal dynein fragment ( Figure 7E and Figure 7—figure supplement 1G ) . Taken together , these findings indicate that an ectopic disulfide bond is the likely cause for R1852C dysfunction . Finally , we sought to determine whether the compensatory cysteine to serine substitution would also rescue motility defects in human dynein . To this end , we engineered the equivalent mutations ( C1932S , R1962C , or C1932S R1962C ) into an insect cell expression plasmid encoding an affinity-tagged human dynein complex ( ZZ-SNAPf-DYNC1H1 , IC2 , LIC2 , Tctex1 , Robl1 , LC8; a kind gift from Andrew Carter ) . Subsequent to purification of recombinant complexes , we assessed their function using a microtubule gliding assay in which free microtubules are translocated by dynein complexes non-specifically adsorbed to the glass surface . This assay permits an assessment of dynein-intrinsic motility parameters without the need for additional factors that are required for single molecule motility of human dynein ( i . e . , dynactin and a requisite adaptor [McKenney et al . , 2014; Schlager et al . , 2014] ) . Consistent with previous findings , the single R1962C mutation ( equivalent to R1852C ) was sufficient to completely disrupt microtubule gliding activity ( Hoang et al . , 2017 ) ( Figure 7F–H ) ; however , as with yeast dynein , substitution of the closely apposed cysteine to a serine ( C1932S ) was sufficient to restore some of the lost function . Although the compensatory mutation did not restore activity to wild-type levels ( note the low velocity and degree of activity with respect to wild-type in Figure 7G and H ) , the same was true for the full-length yeast dynein complex , which was only partially rescued by the mutation ( see Figure 7C and D , and Figure 7—figure supplement 1A–C ) . Although unclear , the different degrees of rescue by the compensatory mutation with yeast dynein ( observed in vivo; Figure 7C and D , and Figure 7—figure supplement 1A–D ) versus human dynein ( observed in vitro; Fig . F-H ) may be due to the presence of quality control mechanisms at play in live yeast cells which ensure that mostly properly folded motors are delivered to cortical receptor sites . No such mechanism is present to prevent adsorption of inactive recombinantly produced human dynein motors to the glass in the in vitro gliding assay . In summary , these results confirm our findings with yeast dynein , and moreover validates yeast as a powerful model system to understand the molecular basis for dynein dysfunction in patients . In an effort to identify potential correlations between degree of dynein dysfunction and disease , we summarized the findings from all of our various assessments into a heat map in which the variance ( or lack thereof ) from wild-type was assigned a color based on the statistical significance ( e . g . , green , p≥0 . 100; red , p≤0 . 005; Figure 8A ) . Unsurprisingly , on average , mutations within the highly conserved motor domain exhibited more statistically significant differences from wild-type than those in the tail domain . Moreover , this broad view of dysfunction in the mutant library revealed that the more severe loss-of-function mutants appeared to correlate with MCD , while those with lesser defects mainly correlate with motor neuron diseases ( SMA-LED , CMT or CMD ) . To more quantitatively assess our data for potential correlations between dynein dysfunction and disease type , we developed a system in which the degrees of variance from wild-type for each mutant were assigned numerical scores that we then used to tabulate a single value that represents the degree of dynein dysfunction for each mutation ( coefficient of dynein dysfunction , or CDD ) . For tabulation of the CDD value , we focused on the in vivo data only . The reasons for this were two-fold: ( 1 ) defects observed in the in vivo assays were largely reflective of the degree of dysfunction , whereas the in vitro data generally revealed the mechanisms of dysfunction ( e . g . , whether the mutation affected dynein-intrinsic or extrinsic function ) ; and , ( 2 ) all mutants with in vivo defects also exhibited in vitro defects , and so we chose to include only the in vivo data to avoid redundancies in the CDD tabulation . In addition to the spindle positioning assay , values from the following spindle dynamics metrics were used to compile the CDD: velocity , displacement per event , neck transit success rate , total spindle displacement per minute , and number of events per minute . The latter two were both included in the CDD tabulation since they revealed two different aspects of dynein activity: the extent to which the mutant dynein-dynactin complexes could move the spindle ( total displacement per minute ) , and the ability to initiate a spindle movement event ( number of events per minute ) . Since the spindle positioning assay reveals gross perturbations in dynein function , we increased the weight of this score with respect to each of the other metrics , such that the summed spindle dynamics metrics were weighed equally with that of the spindle positioning assay ( see Figure 8—figure supplement 1 for additional details on CDD tabulation ) . This analysis revealed a broad range of dynein dysfunction ranging from low ( <10 ) to high CDDs ( >30 ) for the mutant library . For comparison , the CDDs for wild-type and dyn1∆ cells were set to 0 and 100 , respectively . Thus , considering the phenotypic severity of H3639P ( CDD = 82 ) – which had a spindle positioning defect as severe as dyn1∆ , but exhibited some dynein activity in the spindle dynamics assays – we find that the CDD score accurately reflects the degree of dysfunction for each mutant . Ranking the mutants according to their CDD scores ( from low to high; Figure 8B ) revealed an apparent correlation between dynein dysfunction and disease type . Specifically , we found that above a CDD value of ~ 18 , the likelihood of the mutation correlating with MCD increased , while its correlation with SMA-LED , CMT or CMD decreased . This suggests that the two general types of diseases ( motor neuron disease , and defects in brain development ) are each caused by different degrees of dynein dysfunction , and that there exists a lower threshold of dynein activity that is required for neurological development ( see Discussion ) .
Neurodegenerative or developmental diseases that arise as a consequence of mutations within the dynein gene – or dyneinopathies – are a broad range of devastating diseases that include muscular atrophy , muscular dystrophy , and malformations in cortical development ( MCD ) . The ages of onset for these diseases range from birth to late adulthood , while the severity of the symptoms associated with them also cover a broad range ( Poirier et al . , 2013; Laquerriere et al . , 2017; Willemsen et al . , 2012; Vissers et al . , 2010 ) . Although the underlying reasons behind this symptomatic diversity are unclear , our findings suggest that the degree of dysfunction is at least a potential genetic determinant of the type of disease . Specifically , motor neuron diseases appear to be more susceptible to even small degrees of dysfunction ( CDD from 5 to 18 ) , while MCD tends to correlate with larger degrees of dynein dysfunction ( CDD ≥ 19 ) . The reasons for this are unclear , but we hypothesize that the differences are due to the somewhat distinct types of dynein-mediated transport required to maintain motor neuron health , versus that required to effect nuclear or neuronal migration in the developing neocortex . Motor neurons possess extremely long axons ( ≤1 m ) , and thus require a high degree of processive transport for the myriad vesicular cargoes that are moved from the soma to the axon terminal and back . This transport takes place not only during development and in growing children , but also in fully matured adults . Thus , it stands to reason that even subtle loss-of-functions in dynein-mediated transport can , over time , compromise motor neuron health . This is apparent from the wide range in ages-of-onset ( from birth to adulthood ) , and in the range of severity for dynein-based motor neuron diseases ( from weakness in the lower limbs to gross motor difficulties ) . On the other hand , during early brain development , dynein plays key roles in interkinetic nuclear migration ( INM ) and neuronal migration during development of the neocortex ( Tsai et al . , 2010; Hu et al . , 2013; Tsai et al . , 2007 ) . Nuclear envelope-anchored dyneins move the nucleus tens of microns from the basal to the apical surface of the neuroepithelium ( Hu et al . , 2013 ) . In addition to being a shorter distance traveled compared to axonal transport in a motor neuron , the number of motors engaged with microtubules during a nuclear migration event is likely far greater . Immunofluorescence reveals dynein is present along most of the nuclear envelope surface ( Hu et al . , 2013 ) . Moreover , live cell imaging revealed that the microtubule network is fairly extensive in the proximity of the nucleus ( Tsai et al . , 2010; Tsai et al . , 2007 ) . A similar process may be at play during post-INM neuronal migration , when dynein assists in centrosome advancement of the postmitotic neuron . Evidence suggests that , as in budding yeast , astral microtubules make contacts with teams of cortically anchored dynein motors to move the centrosome ( Tsai et al . , 2007 ) . Thus , for both INM and neuronal migration , many dynein molecules are likely engaging with microtubules to effect neurogenesis . This is in contrast to a vesicle that is being transported along the axon of a motor neuron , which likely possesses far fewer motors ( ~3–7 per vesicle [Hendricks et al . , 2010; Rai et al . , 2013] ) , of which potentially only a subset are engaged due to the geometric constraints associated with a three-dimensional vesicle engaging with a single filament . Thus , the minimal functional requirements for individual dynein motors during INM or neuronal migration are potentially lower due to the large number of motors engaged during a migration event . In such a scenario , teams of motors comprised of a mixture of wild-type and mutant variants ( due to the heterozygous nature of these diseases ) can work together to effect INM , but are less able to effectively transport single vesicular cargoes along the axon . Our data indicate that K1475Q – a mutation in the linker domain – increases dynein run length in vitro , and alters its localization pattern in vivo . The position of this mutation ( at an intermolecular interface that helps mediate formation of an autoinhibited Phi particle conformation of human dynein; Zhang et al . , 2017 ) suggests that the associated phenotypes may be a consequence of altered activity regulation . Given this mutant exhibited dynein-intrinsic enhancements in in vitro activity , the reduction in cellular dynein-dynactin activity is likely a consequence of the altered localization pattern . Specifically , the localization phenotype raises the possibility that a reduction in the number of dynein-dynactin complexes per cortical site ( as apparent by the reduced fluorescence intensity of cortical foci; Figure 5D ) – which would result in fewer motor complexes being engaged for a spindle movement event – is the basis for cellular dysfunction . Given the autoinhibited Phi particle exhibits reduced affinity for dynactin ( Zhang et al . , 2017 ) , the altered localization pattern of the mutant may be due to disruption of a similar autoinhibitory conformation in yeast , consequent increased dynactin binding – which is supported by our ratiometric fluorescence imaging ( Figure 5—figure supplement 1 ) – and thus an increase in the frequency of cortical off-loading events ( Markus and Lee , 2011b ) . Previous studies have suggested that dynein’s interaction with dynactin is a limiting step in the delivery of dynein-dynactin complexes to cortical Num1 sites ( Markus et al . , 2011 ) . We are currently focused on assessing whether yeast dynein indeed adopts such an autoinhibited state , and what role this conformation plays in the regulation of dynein activity . In addition to revealing the potential molecular basis for disease onset or progression in affected patients , our findings also identified the potential structural basis for dynein dysfunction in two of the mutants . In the case of H3639P , our data support a model wherein the proline substitution compromises structural plasticity within an inter-helical loop in AAA5 that ultimately leads to a loss of activity in a large fraction of the motors , potentially as a consequence of protein misfolding . This conclusion is based on the reduced localization phenotype ( Figure 5B and C ) that is rescued by flanking glycines ( Figure 6F ) and proteasome inhibition ( Figure 6G ) , and the large proportion of non-motile microtubule-bound motors we observed in the single molecule assay ( Figure 4D ) . Since nearly all these motors exhibit persistent microtubule binding but no motility ( 96%; see kymograph in Figure 4—figure supplement 1 ) , we propose that a structural defect within the AAA ring is compromising the ability of nucleotide binding or hydrolysis to communicate with the microtubule-binding domain . It is interesting to note that the small fraction of motors that do exhibit processive motility in vitro ( 4 . 5% ) move at velocities that are roughly similar to wild-type motors ( 80 versus 58 nm/sec for wild-type and H3639P , respectively ) . We observed a similar phenomenon during dynein-mediated spindle translocation in vivo ( 42 versus 22 nm/sec ) . These findings are consistent with the notion that a small fraction of the motors are capable of overcoming folding defects to adopt a native , motility-competent conformation . Subsequent to initiating this study , another group published findings describing the in vitro motility properties of a subset of the mutants analyzed here ( Hoang et al . , 2017 ) . In this study , the authors utilized a recombinant human dynein in complex with dynactin and the adaptor BicD2 to assess single molecule motility parameters . With only two exceptions ( E109I and N283R ) , the findings from this highly informative study largely corroborate our own data in that the dynein mutants assessed compromised the processivity of dynein-dynactin complexes ( see Figure 8A , left , ‘D’ ) . For instance , the two mutants with the most severe phenotypic consequences were the same as those observed here ( R1962C and H3822P ) . Although one of the exceptions – E109I – did not exhibit reduced processivity in our assays , the authors’ observation that this mutation leads to a reduction in the number of processive motors ( Hoang et al . , 2017 ) is similar to our observation of reduced cellular dynein-dynactin activity for this mutant ( Figure 8A , left , ‘A’ ) . Interestingly , this same study also noted a similar correlation between degree of dynein dysfunction and disease type ( i . e . , the most severe phenotypes correlated with MCD , while the least severe phenotypes correlated with SMA-LED ) , providing further validation for budding yeast as a model system for dynein studies . In summary , we have established yeast as a medium-throughput model system that can be used to assess the molecular basis for dysfunction of disease-correlated dynein mutants . Our rapid and economical toolbox can be easily applied to understand the underlying basis for dysfunction in newly identified dynein mutants found in patients suffering from neurological diseases . We have also demonstrated the feasibility of rapidly testing hypotheses generated from our battery of assays using the wealth of available structural information that is available for dynein and its regulators .
Strains are derived from either W303 or YEF473A ( Bi and Pringle , 1996 ) and are listed in Supplementary file 1 . We transformed yeast strains using the lithium acetate method ( Knop et al . , 1999 ) . Strains carrying mutations were constructed by PCR product-mediated transformation ( Longtine et al . , 1998 ) or by mating followed by tetrad dissection . Proper tagging and mutagenesis was confirmed by PCR , and in most cases sequencing ( all point mutations were confirmed via sequencing ) . Fluorescent tubulin-expressing yeast strains were generated using plasmids and strategies described previously ( Markus et al . , 2015; Song and Lee , 2001 ) . Yeast synthetic defined ( SD ) media was obtained from Sunrise Science Products ( San Diego , CA ) . For expression and purification of human dynein complex mutants ( or wild-type ) , mutations were engineered into the human dynein heavy chain ( DHC ) -containing plasmid , pbiG1a:6His-ZZ-SNAPf-DHC1 . We used Gibson assembly to engineer point mutations – C1932S , R1962C , or both – into this plasmid . The PmeI-digested gene expression cassette from this plasmid was co-assembled with the PmeI-digested poly-gene cassette from pbiG1b:IC2/LIC2/Tctex1/Robl1/LC8 ( encoding all dynein accessory chains ) into PmeI-digested pbiG2ab using biGBac cloning strategies as previously described ( Weissmann et al . , 2016 ) ( all wild-type plasmids were kind gifts from Andrew Carter ) . The final , sequence-verified plasmids ( wild-type and mutant variants of pbiG2ab:6His-ZZ-SNAPf-DHC1/IC2/LIC2/Tctex1/Robl1/LC8 ) were used to generate recombinant baculoviral genomes by Tn7 transposition into DH10Bac cells ( Life Technologies ) . White , PCR-confirmed colonies were inoculated into LB media supplemented with 7 µg/ml gentamycin , 10 µg/ml tetracycline and 50 µg/ml kanamycin and grown overnight at 37°C . Bacmid preparation was performed as described previously ( Zhang et al . , 2017 ) , stored at 4°C , and used within 2 weeks for subsequent virus production ( see below ) . Purification of yeast dynein ( ZZ-TEV-Dyn1-HALO , under the native DYN1 promoter; or , ZZ-TEV-6His-GFP-3HA-GST-dynein331-HALO , under the control of the galactose-inducible promoter , GAL1p ) was performed as previously described ( Ecklund et al . , 2017; Huang et al . , 2012 ) . Briefly , yeast cultures were grown in YPA supplemented with either 2% glucose ( for full-length dynein ) or 2% galactose ( for GST-dynein331 ) , harvested , washed with cold water , and then resuspended in a small volume of water . The resuspended cell pellet was drop frozen into liquid nitrogen and then lysed in a coffee grinder ( Hamilton Beach ) . After lysis , 0 . 25 vol of 4X dynein lysis buffer ( 1X buffer: 30 mM HEPES , pH 7 . 2 , 50 mM potassium acetate , 2 mM magnesium acetate , 0 . 2 mM EGTA ) supplemented with 1 mM DTT , 0 . 1 mM Mg-ATP , 0 . 5 mM Pefabloc SC ( concentrations for 1X buffer ) was added , and the lysate was clarified at 22 , 000 x g for 20 min . The supernatant was then bound to IgG sepharose six fast flow resin ( GE ) for 1–1 . 5 hr at 4°C , which was subsequently washed three times in 5 ml lysis buffer , and twice in TEV buffer ( 50 mM Tris , pH 8 . 0 , 150 mM potassium acetate , 2 mM magnesium acetate , 1 mM EGTA , 0 . 005% Triton X-100 , 10% glycerol , 1 mM DTT , 0 . 1 mM Mg-ATP , 0 . 5 mM Pefabloc SC ) . To fluorescently label the motors for single molecule analyses , the bead-bound protein was incubated with either 6 . 7 µM HaloTag-AlexaFluor660 ( Promega ) for 10 min at room temperature . The resin was then washed four more times in TEV digest buffer , then incubated in TEV buffer supplemented with TEV protease for 1–1 . 5 hr at 16°C . Following TEV digest , the beads were pelleted , and the resulting supernatant was aliquoted , flash frozen in liquid nitrogen , and stored at −80°C . The human dynein complex was expressed and purified from insect cells ( ExpiSf9 cells; Life Technologies ) as previously described with minor modifications ( Zhang et al . , 2017; Schlager et al . , 2014 ) . Briefly , 4 ml of ExpiSf9 cells at 2 . 5 × 106 cells/ml , which were maintained in ExpiSf CD Medium ( Life Technologies ) , were transfected with 1 µg of bacmid DNA ( see above ) using ExpiFectamine ( Life Technologies ) according to the manufacturer’s instructions . 5 days following transfection , the cells were pelleted , and 1 ml of the resulting supernatant ( P1 ) was used to infect 300 ml of ExpiSf9 cells ( 5 × 106 cells/ml ) . 72 hr later , the cells were harvested ( 2000 x g , 20 min ) , washed with phosphate buffered saline ( pH 7 . 2 ) , pelleted again ( 1810 x g , 20 min ) , and resuspended in an equal volume of human dynein lysis buffer ( 50 mM HEPES , pH 7 . 4 , 100 mM NaCl , 10% glycerol , 1 mM DTT , 0 . 1 mM Mg-ATP , 1 mM PMSF ) . The resulting cell suspension was drop frozen in liquid nitrogen and stored at −80°C . For protein purification , 30 ml of additional human dynein lysis buffer supplemented with cOmplete protease inhibitor cocktail ( Roche ) was added to the frozen cell pellet , which was then rapidly thawed in a 37°C water bath prior to incubation on ice . Cells were lysed in a dounce-type tissue grinder ( Wheaton ) using ≥ 150 strokes ( lysis was monitored by microscopy ) . Subsequent to clarification at 40 , 000 x g , 45 min , the supernatant was applied to 2 ml of IgG sepharose fast flow resin ( GE ) pre-equilibrated in human dynein lysis buffer , and incubated at 4°C for 2–4 hr . Beads were then washed with 50 ml of human dynein lysis buffer , and 50 ml of human dynein TEV buffer ( 50 mM Tris pH 7 . 4 , 150 mM potassium acetate , 2 mM magnesium acetate , 1 mM EGTA , 10% glycerol , 1 mM DTT , 0 . 1 mM Mg-ATP ) . The bead-bound protein was incubated with 3 µM SNAP-Surface Alexa Fluor 647 ( NEB ) for 40–60 min at 4°C ( to fluorescently label the protein ) , washed five times in human dynein TEV buffer , then incubated with TEV protease overnight at 4°C . The next morning , the recovered supernatant was applied to a Superose six gel filtration column ( GE ) equilibrated in GF150 buffer ( 25 mM HEPES pH 7 . 4 , 150 mM KCl , 1 mM MgCl2 , 5 mM DTT , 0 . 1 mM Mg-ATP ) using an AKTA Pure . Peak fractions ( determined by UV 260 nm absorbance and SDS-PAGE ) were pooled , concentrated , aliquoted , flash frozen , then stored at −80°C . Yeast cultures were grown to similar mid-log phase densities ( OD600 ~ 2 ) in 4 ml SD media , and harvested . Cell pellets were resuspended in 0 . 2 ml of 0 . 1 M NaOH and incubated for 10 min at room temperature as described ( Kushnirov , 2000 ) . Following centrifugation , the resulting cell pellet was resuspended in sample buffer . Equal amounts of total cell lysate ( as determined from cell density prior to lysis ) were loaded into each lane , transferred to PVDF and probed with a monoclonal anti-GFP antibody ( at 1:250; Abm ) followed by an HRP-conjugated goat anti-mouse antibody ( at 1:10 , 000; Jackson ImmunoResearch Laboratories ) . Electroblotting to PVDF was performed in 25 mM Tris , 192 mM glycine supplemented with 0 . 05% SDS and 20% methanol . Chemiluminescence signal was acquired with a Chemidoc MP ( BioRad ) . Immunoblots were exposed ( durations ranged from 2 to 5 min ) without saturating the camera’s pixels . The yeast dynein single-molecule motility assay was performed as previously described with minor modifications ( Ecklund et al . , 2017 ) . Briefly , flow chambers constructed using slides and plasma cleaned and silanized coverslips attached with double-sided adhesive tape were coated with anti-tubulin antibody ( 8 μg/ml , YL1/2; Accurate Chemical and Scientific Corporation ) then blocked with 1% Pluronic F-127 ( Fisher Scientific ) . Taxol-stabilized microtubules assembled from unlabeled and fluorescently-labeled porcine tubulin ( 10:1 ratio; Cytoskeleton ) were introduced into the chamber . Following a 5–10 min incubation , the chamber was washed with dynein lysis buffer ( see above ) supplemented with 20 μM taxol , and then purified dynein motors were introduced in the chamber . After a 1 min incubation , motility buffer ( 30 mM HEPES pH 7 . 2 , 50 mM potassium acetate , 2 mM magnesium acetate , 1 mM EGTA , 1 mM DTT , 1 mM Mg-ATP ) supplemented with 0 . 05% Pluronic F-127 , 20 µM taxol , and an oxygen-scavenging system ( 1 . 5% glucose , 1 U/ml glucose oxidase , 125 U/ml catalase ) was added . TIRFM images were collected using a 1 . 49 NA 100X TIRF objective on a Nikon Ti-E inverted microscope equipped with a Ti-S-E motorized stage , piezo Z-control ( Physik Instrumente ) , and an iXon X3 DU897 cooled EM-CCD camera ( Andor ) . 488 nm , 561 nm , and 640 nm lasers ( Coherent ) were used along with a multi-pass quad filter cube set ( C-TIRF for 405/488/561/638 nm; Chroma ) and emission filters mounted in a filter wheel ( 525/50 nm , 600/50 nm and 700/75 nm; Chroma ) . We acquired images at 2 s intervals for 8 min . Velocity and run length values were determined from kymographs generated using the MultipleKymograph plugin for ImageJ ( http://www . embl . de/eamnet/html/body_kymograph . html ) . Human dynein-mediated microtubule gliding assays were performed as previously described ( Zhang et al . , 2017 ) with minor modifications . Briefly , flow chambers were prepared by affixing an ethanol-flamed coverslip to a glass slide using double-stick tape . The chamber was then incubated on an ice block , washed with 1% Pluronic F-127 , following by addition of purified dynein ( five chamber volumes of 60 nM dynein complex ) . Unbound motors were washed out with GF150 buffer . Subsequently , motility buffer ( 30 mM HEPES pH 7 . 0 , 50 mM KCl , 5 mM MgSO4 , 1 mM EGTA , 1 mM DTT , 2 . 5 mM Mg-ATP , 40 µM taxol ) supplemented with 1 . 5% glucose , the oxygen scavenging system ( see above ) , and 150 nM fluorescent microtubules was added to the chamber . Images were acquired every 1 s ( for wild-type ) or 5 s ( for mutants ) , and velocity values were determined from kymographs generated as above . For the single time point spindle position assay , the percentage of cells with a misoriented anaphase spindle was determined after growth overnight ( 12–16 hr ) at a low temperature ( 16°C ) , as previously described ( Li et al . , 2005; Markus et al . , 2009; Sheeman et al . , 2003 ) . A single z-stack of wide-field fluorescence images was acquired for mRuby2-Tub1 . For the spindle dynamics assay , cells were arrested with hydroxyurea ( HU ) for 2 . 5 hr , and then mounted on agarose pads containing HU for fluorescence microscopy . Full Z-stacks ( 23 planes at 0 . 2 µm spacing ) of GFP-labeled microtubules ( GFP-Tub1 ) were acquired every 10 s for 10 min on a stage pre-warmed to 30°C . To image dynein localization in live cells , cells were grown to mid-log phase in SD media supplemented with 2% glucose , and mounted on agarose pads . Images were collected on a Nikon Ti-E microscope equipped with a 1 . 49 NA 100X TIRF objective , a Ti-S-E motorized stage , piezo Z-control ( Physik Instrumente ) , an iXon DU888 cooled EM-CCD camera ( Andor ) , a stage-top incubation system ( Okolab ) , and a spinning disc confocal scanner unit ( CSUX1; Yokogawa ) with an emission filter wheel ( ET525/50M for GFP , and ET632/60M for mRuby2; Chroma ) . Lasers ( 488 nm and 561 nm ) housed in a LU-NV laser unit equipped with AOTF control ( Nikon ) were used to excite GFP and mRuby2 , respectively . The microscope was controlled with NIS Elements software ( Nikon ) . Statistical tests were performed as described in the figure legends . Unpaired Welch’s t tests ( for gaussian distributed velocity data ) and Mann-Whitney test ( for exponentially distributed dispalcement data ) were performed using Graphpad Prism . Z scores , which are a quantitative measure of difference between two proportions , were calculated using the following formula:Z= ( p^1−p^2 ) p^ ( 1−p^ ) ( 1n1+1n2 ) where:p^=y1+y2n1+n2 Z scores were converted to two-tailed P values using an online calculator . To calculate the CDD scores , we used the following approach which permitted a quantitative measure of difference between mean values obtained for wild-type versus those obtained for each mutant . Graphpad Prism was used to calculate q values ( i . e . , the difference between the two means divided by the standard error of that difference ) , whereas Z scores were calculated as described above ( all values are shown in Figure 8—figure supplement 1 ) . We then converted the q values and Z scores for each mutant ( for each assay ) into a ‘normalized relative variance’ score ( nrv ) , which reflects the relative difference between two mean values ( e . g . , between wild-type and mutant 1; as reflected in the Z scores and q values , or ‘v’ ) , where nrv = |v|/vmax for each range of scores ( for each column shown in Figure 8—figure supplement 1A ) . To convert the nrv values into a final CDD score for each mutant , we used the formula shown in Figure 8—figure supplement 1C . Briefly , the nrv values for each assay for a given mutant was added , with the spindle positioning nrv ( nrvSP ) weighed five times that of the others , as described within the Results . In the two cases where a value wasn’t determined ( due to insufficient observations , such as in the case for neck transit success for the H3639P mutant ) , the denominator was reduced from 6 to 5 . In the two instances where the Z score for spindle positioning was negative ( due to a lower number of mispositioned spindles being observed in K540C and D2439K cells than in wild-type cells; see Figure 1B ) , we adjusted the values to 0 so as to avoid them skewing the nrvSP values . | Motor proteins maintain order by transporting biomolecules and various structures within living cells . Dynein is one such motor that moves many types of cargoes along tracks called microtubules , which are spread across the cell’s interior . This motor is particularly important in nerve cells , which can be very long and thus depend heavily on motor proteins to ensure cargoes end up where they are needed . This becomes especially apparent in human diseases that arise as a consequence of mutations in the genes that produce components of the dynein motor . It is assumed that these genetic changes simply prevent dynein from working properly , which ultimately affects the health and survival of cells . However , it is currently unknown what specific effect these mutations have on dynein’s role within the cell , and how these changes lead to particular diseases . Marzo et al . have now used dynein from a budding yeast to closely examine 17 mutations in the dynein gene that are associated with developmental and/or motor neuron diseases in humans . For each mutation , various aspects of how dynein moves ( e . g . average speed , distance travelled ) were measured and quantitatively compared . The results show that the severity of the effect of each mutation can be directly correlated with the type of disease caused by the mutation . In particular , mutations that lead to less severe defects are found in patients that suffer from various motor neuron diseases , while more severe dynein mutations are found in patients with developmental brain disorders . Marzo et al . confirmed the likely structural changes that caused the defects in dynein’s activity in two of the 17 cases , by engineering additional , restorative mutations that lessened the effects of the primary mutation . These findings reveal links between the molecular impact of defects in the dynein gene and human health . They also confirm that budding yeast is a powerful tool for investigating newly discovered dynein mutations that correlate with disease . This study provides a potential system that could be used to screen drugs that might lessen the effects of specific dynein mutations . However , further work is needed to determine how effective this system will be for drug discovery . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2019 | Molecular basis for dyneinopathies reveals insight into dynein regulation and dysfunction |
The nonsense-mediated mRNA decay ( NMD ) pathway degrades mRNAs containing long 3'UTRs to perform dual roles in mRNA quality control and gene expression regulation . However , expansion of vertebrate 3'UTR functions has required a physical expansion of 3'UTR lengths , complicating the process of detecting nonsense mutations . We show that the polypyrimidine tract binding protein 1 ( PTBP1 ) shields specific retroviral and cellular transcripts from NMD . When bound near a stop codon , PTBP1 blocks the NMD protein UPF1 from binding 3'UTRs . PTBP1 can thus mark specific stop codons as genuine , preserving both the ability of NMD to accurately detect aberrant mRNAs and the capacity of long 3'UTRs to regulate gene expression . Illustrating the wide scope of this mechanism , we use RNA-seq and transcriptome-wide analysis of PTBP1 binding sites to show that many human mRNAs are protected by PTBP1 and that PTBP1 enrichment near stop codons correlates with 3'UTR length and resistance to NMD .
Nonsense-mediated mRNA decay ( NMD ) is an evolutionarily conserved co-translational mRNA turnover pathway responsible for degrading diverse eukaryotic mRNAs ( reviewed in Schweingruber et al . , 2013 ) . In addition to its well-known role in detecting aberrant transcripts containing premature termination codons ( PTCs ) , NMD has also been shown to target a broad range of cellular mRNAs under normal conditions ( He et al . , 2003; Hurt et al . , 2013; Lelivelt and Culbertson , 1999; Mendell et al . , 2004; Rehwinkel et al . , 2005; Weischenfeldt et al . , 2008; Weischenfeldt et al . , 2012; Longman et al . , 2013 ) , including those with spliced introns downstream of the TC , upstream open reading frames ( uORFs ) , or long 3’ untranslated regions ( 3’UTRs ) . Together , the pathway is commonly estimated to regulate 5–10% of all human genes . NMD thus has an enormous impact on the human transcriptome in all cells , while also performing a surveillance role that modulates the phenotypic consequences of numerous human genetic diseases arising from nonsense mutations ( Keeling et al . , 2013 ) . A core set of dedicated NMD proteins works in conjunction with general mRNA metabolism and translation factors to accurately select and degrade specific mRNAs . The superfamily I RNA helicase UPF1 is the focal point of the NMD pathway , with characterized functions throughout the process of target discrimination , translational repression , RNase recruitment , and post-decay RNP remodeling ( Czaplinski et al . , 1995; Franks et al . , 2010; Hogg and Goff , 2010; Isken et al . , 2008; Okada-Katsuhata et al . , 2012; Unterholzner and Izaurralde , 2004 ) . UPF2 and UPF3B stimulate UPF1’s ATPase and decay-promoting activities and serve as a physical link between UPF1 and the exon junction complex ( EJC ) , a strong activator of decay in vertebrate NMD ( Chamieh et al . , 2008; Le Hir et al . , 2001 ) . Interaction of UPF1 with release factors eRF1/3 at the terminating ribosome promotes phosphorylation of UPF1 by the SMG1 kinase , which in turn mediates recruitment and/or activity of the SMG6 endonuclease , the SMG5/7 complex , and decapping and deadenylation factors ( Jonas et al . , 2013; Loh et al . , 2013; Chakrabarti et al . , 2014; Glavan et al . , 2006; Huntzinger et al . , 2008; Kashima et al . , 2006; Nicholson et al . , 2014; Ohnishi et al . , 2003; Okada-Katsuhata et al . , 2012; Eberle et al . , 2009; Unterholzner and Izaurralde , 2004 ) . The UPF1 helicase domain forms an extensive surface for high-affinity , sequence-nonspecific RNA binding , allowing the protein to associate with the full diversity of NMD substrates ( Chakrabarti et al . , 2011; Fiorini et al . , 2012; Hurt et al . , 2013; Zund et al . , 2013; Kurosaki et al . , 2014 ) . Elongating ribosomes can disrupt UPF1’s promiscuous RNA binding activity , leading to preferential UPF1 accumulation on 3’UTRs ( Hogg and Goff , 2010; Hurt et al . , 2013; Zund et al . , 2013 ) . We have previously proposed that this property of UPF1 effectively allows it to sense 3’UTR length , predisposing transcripts containing long 3’UTRs to decay ( Hogg , 2011; Hogg and Goff , 2010 ) . This model is supported by recent comprehensive analysis of UPF1 targets demonstrating correlations among mRNA 3’UTR length , UPF1 association , and de-repression of gene expression following NMD inhibition ( Hurt et al . , 2013 ) . Extensive experiments in several eukaryotic model systems have established a tight link between 3’UTR length and decay susceptibility , suggesting a conserved mechanistic basis for NMD target selection ( Amrani et al . , 2004; Behm-Ansmant et al . , 2007; Buhler et al . , 2006; Eberle et al . , 2008; Longman et al . , 2007; Muhlrad and Parker , 1999; Singh et al . , 2008 ) . Directly probing the relationship between 3’UTR length and NMD-sensitivity in human cells , systematic analyses of reporter transcripts revealed a progressive UPF1-dependent decrease in mRNA half-life with increasing 3’UTR length ( Buhler et al . , 2006; Eberle et al . , 2008 ) . While global correlations between 3’UTR length and NMD susceptibility have been described in mammals ( Hansen et al . , 2009; Hurt et al . , 2013; Mendell et al . , 2004; Ramani et al . , 2009; Yepiskoposyan et al . , 2011 ) , these studies have been complicated by the fact that a large number of eukaryotic transcripts with long 3’UTRs appear to evade NMD through as-yet-unknown mechanisms . This is particularly true in human cells , in which the average 3’UTR length has evolved to exceed that necessary to induce NMD of reporter transcripts . Indeed , several TC-proximal human mRNA sequences capable of protecting mRNAs from NMD have been identified ( Toma et al . , 2015 ) . Thus , mechanisms to protect specific mRNAs from NMD are likely to be a major force shaping human gene expression . Retroviruses have long served as excellent model systems to explore the functions of host mRNA processing , translation , and decay pathways . As a consequence of evolutionary pressures to maximize the coding potential of compact RNA genomes , retroviral RNAs possess features that are predicted to be targeted by host cell RNA surveillance machineries , including long 3’UTRs , multiple open reading frames , and retained introns ( Bolinger and Boris-Lawrie , 2009; Withers and Beemon , 2011 ) . Consequently , some retroviruses have developed protein-based protective mechanisms against NMD to ensure RNA integrity and translation ( Mocquet et al . , 2012; Nakano et al . , 2013 ) . Of known cellular or viral mechanisms for NMD evasion , the most extensively characterized is a cis-acting RNA sequence found in the Rous sarcoma virus ( RSV ) , designated the RNA stability element ( RSE; Arrigo and Beemon , 1988; Barker and Beemon , 1991 , 1994 ) . The RSE is a 400-nt element located immediately downstream of the gag TC in the unspliced RSV viral RNA that robustly protects the viral RNA from UPF1-dependent decay in chicken cells ( Quek and Beemon , 2014; Weil and Beemon , 2006; Weil et al . , 2009; Withers and Beemon , 2010 , 2011 ) . Despite thorough studies of the RSE structure and function , its mechanism of action has remained unclear . Here , we elucidate the mechanism underlying the ability of the RSE to protect mRNAs from NMD and show that numerous human transcripts containing long 3’UTRs also exploit this strategy to maintain stability . By affinity purifying endogenously assembled mRNPs containing the RSE , we identify the polypyrimidine tract binding protein 1 ( PTBP1 ) as the key mediator of RSE function . We show that mutations preventing PTBP1 binding to the RSE abolish protection from NMD , while artificial recruitment of PTBP1 immediately downstream of an NMD-triggering TC recapitulates RSE activity . Together , our findings indicate that PTBP1 functions to exclude UPF1 from 3’UTRs , disrupting its ability to accurately discriminate 3’UTR length and induce decay . Furthermore , we performed RNA-seq analysis on human cells depleted of PTBP1 and UPF1 together and in isolation to identify endogenous transcripts with long 3’UTRs protected from NMD by PTBP1 . Transcriptome-wide analysis of PTBP1 interaction sites reveals preferential binding of PTBP1 near TCs , a binding pattern correlated with 3’UTR length and resistance to NMD .
We first set out to test whether the avian retrovirus-derived RSE retains its anti-NMD function in human cells , reasoning that the ability to function in highly divergent vertebrates would imply a conserved mechanism for mRNA stabilization . For these studies , we used tetracycline ( tet ) -regulated reporter mRNAs containing a β-globin mini-gene and the SMG5 3’UTR ( Singh et al . , 2008 ) . The SMG5 3’UTR triggers NMD as part of an extensive program of autoregulation by the NMD pathway , in a manner proposed to be due to its length ( 1342 nt; Huang et al . , 2011; Singh et al . , 2008; Yepiskoposyan et al . , 2011 ) . We inserted the 400 nt RSE sequence or a control sequence of the same length ( the antisense RSE sequence , AS-RSE ) into the reporter mRNAs immediately downstream of the TC , mimicking the natural context of the RSE in the RSV RNA ( Figure 1A , B ) . To assess the specific antagonistic activity of the RSE against NMD , constructs encoding reporter mRNAs were co-transfected with a vector constitutively expressing a control RNA into 293 Tet-off cells treated with control or anti-UPF1 siRNAs . The expression of the tet-regulated mRNAs was induced for 4 hr before transcription was inhibited by addition of doxycycline , and mRNA decay was monitored at the indicated time points ( Figure 1C ) . Transcripts containing only the SMG5 3’UTR exhibited a half-life of ~120 min in cells treated with control siRNAs , in agreement with its previous characterization as an NMD substrate . Transcripts containing the RSE were substantially more stable ( half-life ~400 min ) than transcripts containing the AS-RSE sequence ( <120 min ) , confirming the protective activity of the RSE ( Figure 1C , upper panel ) . In contrast , in cells depleted of UPF1 , all transcripts had half lives of greater than 240 min , indicating that the observed decay in siNT-treated cells was due to the activity of UPF1 ( Figure 1C , lower panel ) . 10 . 7554/eLife . 11155 . 003Figure 1 . The RSE protects reporter mRNA from NMD in mammalian cells . ( A ) Schematic of the Rous sarcoma proviral genome . The RSE is located immediately downstream of the gag stop codon . ( B ) Schematic of tet-regulated β-globin reporter mRNA constructs used in RNA decay assays . The RSE sequence ( middle ) and a control sequence , the antisense RSE ( AS-RSE ) sequence ( bottom ) , were inserted into reporter mRNAs containing the β-globin gene and the human SMG5 3’UTR ( top ) . ( C ) Decay assays of reporter mRNAs containing the wild-type SMG5 3’UTR or variants supplemented with RSE or AS-RSE sequences . 293 Tet-off cells were treated with non-targeting siRNA ( siNT; upper panel ) or UPF1 siRNA ( siUPF1; lower panel ) . Constructs encoding the indicated tet-regulated transcripts were co-transfected with the constitutively expressed wild-type β-globin reporter ( pcβwtβ; bottom bands ) . Remaining RNA levels at indicated time points were normalized to levels of the wild-type β-globin transfection control . Half-lives and 95% confidence intervals were obtained from 3 independent experiments ( ***p<0 . 001; ****p<0 . 0001 in two-tailed ANCOVA analysis when compared to pcTET2-βwt-SMG5 ) . Rapid decay of AS-RSE mRNAs to background levels in siNT samples precluded accurate quantification of decay rate . See also Figure 1—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 00310 . 7554/eLife . 11155 . 004Figure 1—figure supplement 1 . RSE protective activity is position-dependent . ( A ) Schematic of tet-regulated β-globin reporter mRNA constructs used in RNA decay assays . RSE sequence was inserted in reporter mRNAs either upstream , in the middle , or downstream of the SMG5 3’UTR . ( B ) Decay assays of reporter mRNAs containing the RSE at different positions in the 3'UTR . Constructs encoding the tet-regulated transcripts described in A ( pcTET2-βwt-RSE-SMG5 , pcTET2-βwt-MRSE-SMG5 , or pcTET2-βwt-3’RSE-SMG5; top bands ) were co-transfected with the constitutively expressed wild-type β-globin reporter ( pcβwtβ; bottom bands ) in HeLa Tet-off cells for decay analysis . Remaining RNA levels at indicated time points were normalized to levels of the wild-type β-globin transfection control . Half-lives and 95% confidence intervals were obtained from 3 independent experiments ( p-values from two-tailed ANCOVA analysis when compared to pcTET2-βwt-RSE-SMG5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 00410 . 7554/eLife . 11155 . 005Figure 1—figure supplement 2 . The RSE protects reporter mRNAs from EJC-stimulated NMD ( A ) Schematic of tet-regulated β-globin reporter mRNA constructs used in RNA decay assays . RSE or antisense RSE ( AS-RSE ) sequences were inserted in reporter mRNAs upstream of a previously-characterized version of the GAPDH-derived 3’UTR engineered to contain the adenovirus major-late intron ( AdML intron; Singh et al . , 2008 ) , leading to EJC-stimulated NMD . ( B ) Decay of AdML intron-containing mRNAs in HeLa Tet-off cells . Half-lives and 95% confidence intervals were derived from linear regression of semi-log plots of normalized RNA abundances from three independent experiments ( p-values from two-tailed ANCOVA analysis comparing the AS-RSE mRNAs to the RSE mRNAs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 005 Previous studies of the RSE have shown that its activity is strongly position-dependent . Specifically , the RSV viral RNA becomes unstable when a TC is created upstream of the natural gag stop codon , an effect due to the increased distance between the RSE and the PTC ( Barker and Beemon , 1994; Weil and Beemon , 2006; Withers and Beemon , 2011 ) . Here , we confirmed that the RSE also exhibits this property when used to shield reporter transcripts in human cells , as placement of the RSE in the middle or at the end of the SMG5 3’UTR abolished its protective activity ( Figure 1—figure supplement 1 ) . Together , our data suggest that the mechanisms governing RSE-mediated protection from NMD are conserved from birds to humans . While the physiological mRNAs subject to RSE-mediated protection are unspliced retroviral RNAs , we also tested whether the RSE is capable of inhibiting NMD promoted by the presence of a 3’UTR-bound EJC . We inserted the wild-type RSE or antisense RSE sequences upstream of a previously characterized artificial GAPDH-derived 3’UTR engineered to contain the adenovirus major-late intron ( AdML; Singh et al . , 2008; Figure 1—figure supplement 2 ) . As expected , the intron-containing 3’UTR caused rapid decay of tet-regulated reporter constructs . In contrast , mRNAs containing the RSE were highly stabilized , indicating that the RSE is capable of suppressing both EJC-independent and EJC-stimulated NMD . Since UPF1 recruitment is a prerequisite for NMD , we hypothesized that the RSE may employ the simple strategy of preventing UPF1 from associating with the 3’UTR of the protected transcript . To test this idea , we immunopurified endogenous UPF1 and assayed the recovery of reporter mRNAs containing RSE or control sequences placed upstream of the artificial 3’UTR derived from the GAPDH ORF and 3’UTR , in this case lacking additional intronic sequence ( Figure 2A ) . This GAPDH-derived 3’UTR has been previously used for studies of UPF1 association and decay and can be efficiently protected by a TC-proximal RSE ( Figure 2—figure supplement 1; Hogg and Goff , 2010; Singh et al . , 2008 ) . Because UPF1 associates with mRNAs in a 3’UTR length-dependent manner ( Hogg and Goff , 2010; Kurosaki and Maquat , 2013 ) , we used an NMD-permissive 397 nt fragment of the SMG5 3’UTR ( SMG5-397 ) to equalize 3’UTR lengths among the mRNAs studied and thus allow accurate assessment of the influence of the RSE on UPF1 binding . As additional controls , we tested the ability of UPF1 to bind mRNAs in which the RSE or the SMG5-397 fragments were moved to the 3’ end of the mRNA ( 3’-RSE and 3’-SMG5-397 , respectively ) . 10 . 7554/eLife . 11155 . 006Figure 2 . The RSE reduces UPF1 association with 3'UTRs in a position-dependent manner . ( A ) Schematic of reporter mRNA constructs used in immunoprecipitation assays . RNAs containing the GFP ORF and an artificial NMD-inducing 3’UTR comprising a portion of the human GAPDH ORF and the GAPDH 3’UTR ( Singh et al . , 2008 ) were modified to contain the RSE or the SMG5-397 sequence 5’ or 3’ of the GAPDH sequence . ( B ) Upf1 is reduced on 3'UTRs containing a TC-proximal RSE . Plasmids expressing the indicated mRNAs were co-transfected in 293 cells with a construct expressing the GFP ORF followed by the bovine growth hormone ( bGH ) polyadenylation signal . Endogenous UPF1 was immunoprecipitated from transfected cells , and co-purifying mRNAs were analyzed by northern blot . Bulk goat IgG was used as a non-specific interaction control . ( C ) Quantification of relative RNA recovery upon UPF1 immunoprecipitation , normalized to the recovery of co-transfected GFP-bGH mRNAs . The amount of RSE-containing RNA recovered was set to 1 . Error bars indicate ± SD; n = 3 ( *p<0 . 05; **p<0 . 01 in two-tailed Student’s t-tests when compared to RSE recovery ) . See also Figure 2—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 00610 . 7554/eLife . 11155 . 007Figure 2—figure supplement 1 . The RSE protects reporter mRNAs containing a GAPDH-derived NMD-sensitive 3’UTR . ( A ) Schematic of tet-regulated β-globin reporter mRNA constructs used in RNA decay assays . RSE or SMG5-397 sequences were inserted in reporter mRNAs upstream of the artificial GAPDH-derived 3’UTR ( Singh et al . , 2008 ) . ( B ) Decay assays of reporter mRNAs containing the RSE or the first 397 nt of the SMG5 3’UTR ( SMG5-397 ) sequence in cells expressing either WT UPF1 ( left ) or an ATPase-dead UPF1 ( right ) . Constructs encoding the tet-regulated transcripts described in ( A ) were co-transfected with the constitutively expressed wild-type β-globin reporter ( pcβWTβ; bottom bands ) in HeLa Tet-off cells . Half-lives and 95% confidence intervals were derived from linear regression of semi-log plots of normalized RNA abundances from three independent experiments ( p-values from two-tailed ANCOVA analysis comparing the indicated mRNAs to the SMG5-397 control in the presence of wild-type UPF1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 00710 . 7554/eLife . 11155 . 008Figure 2—figure supplement 2 . RSE inhibition of UPF1 binding is independent of translation . ( A ) Schematic of GFP reporter mRNA constructs used in immunoprecipitation assays . RNAs containing the GFP ORF and an artificial NMD-inducing 3’UTR comprising a portion of the human GAPDH ORF and the GAPDH 3’UTR ( Singh et al . , 2008 ) were modified to contain the RSE ( top ) or the SMG5-397 sequence ( middle ) . As a control to illustrate the behavior of mRNAs with short 3’UTRs ( untreated condition ) or shorter overall length ( puromycin treatment ) , we used constructs without added RSE or SMG5-397 sequence in which the GFP termination codon was changed to CAA ( bottom ) , resulting in termination at the downstream GAPDH stop codon . ( B ) UPF1 association is reduced on 3'UTRs containing the RSE . Plasmids expressing the indicated mRNAs were co-transfected in 293 cells with a construct expressing the GFP ORF followed by the bovine growth hormone ( bGH ) polyadenylation signal . Endogenous Upf1 was immunoprecipitated from transfected cells grown in the absence or presence of puromycin ( 100 μg/mL for 3 hr ) as indicated , and co-purifying mRNAs were analyzed by northern blot . ( C ) Quantification of relative RNA recovery upon UPF1 immunoprecipitation , normalized to the recovery of co-transfected GFP-bGH mRNAs . The amount of RSE-containing RNA recovered was set to 1 . Error bars indicate ± SD; n ≥ 3 ( *p<0 . 05 , **p<0 . 01 in two-tailed Student’s t-tests when compared to RSE recovery ) . Under both untreated and puromycin treated conditions , the RSE caused reduced accumulation of UPF1 on mRNAs relative to the SMG5-397 control . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 008 Plasmids encoding the experimental RNAs were co-transfected with a control plasmid producing mRNA containing only the GFP ORF and the bovine growth hormone polyadenylation element ( bGH ) , which associates with UPF1 at low but consistent levels due to its short 3’UTR and serves as an internal standard for RNA recovery throughout immunoprecipitation and RNA isolation ( Hogg and Goff , 2010 ) . The extent of UPF1 binding was then assayed by northern blotting of the reporter mRNAs co-immunopurified with endogenous UPF1 . As shown in Figure 2B and C , we observed a marked reduction in UPF1 association with RSE-containing RNAs relative to RNAs in which the RSE was moved to the 3’ end of the transcript or replaced with the SMG5-397 element . Together , these data indicate that the RSE prevents NMD by reducing UPF1 binding to 3’UTRs in a sequence- and position-dependent manner ( Figure 2B , C ) . While previous studies have shown that UPF1 association with mRNAs precedes commitment to decay and does not require ongoing translation ( Hogg and Goff , 2010; Hurt et al . , 2013; Zund et al . , 2013 ) , it remained possible that a reduction in UPF1 binding to RSE-containing 3’UTRs was a consequence rather than a cause of NMD inhibition . To rule out this possibility , we performed immunoprecipitations using extracts from cells treated with the chain-terminating translation inhibitor puromycin . Indeed , the RSE retained the capacity to antagonize UPF1 binding in the absence of translation termination events ( Figure 2—figure supplement 2 ) . These data suggest that the position of the RSE in the mRNA with respect to the 5’ and 3’ ends may contribute to its ability to disrupt UPF1 binding to mRNAs . To identify possible protein cofactors of the RSE , we isolated mRNPs containing the RSE in the sense or antisense orientations by an RNA-based affinity purification technique ( Hogg and Goff , 2010 ) . Reporter mRNAs tagged with a single copy of the 25 nt Pseudomonas phage 7 coat protein ( PP7CP ) RNA hairpin binding site ( Figure 3—figure supplement 1A ) were expressed in 293T cells , and the resulting endogenously assembled mRNPs were purified from cell extracts using the PP7CP tagged with tandem Staphylococcus aureus protein A domains . The isolated mRNPs containing either the RSE sequence or the control sequence displayed a similar profile of co-purifying proteins , while mock purifications from cell extracts containing no tagged RNA yielded very few contaminating proteins ( Figure 3—figure supplement 1B , additional data not shown ) . Comprehensive analysis of the composition of the purified mRNP complexes by tandem mass spectrometry revealed a large number of proteins that were equally represented in the RSE-containing mRNPs and the control mRNPs , many of which are common RNA binding proteins , ribosomal proteins , and translation factors ( Figure 3—source data 1 ) . UPF1 was enriched in the control mRNP ( Figure 3—figure supplement 1C ) , consistent with our finding that the RSE inhibits UPF1 association with the 3’UTR ( Figure 2 ) . Several proteins were over-represented in mRNPs containing the RSE sequence , including polypyrimidine tract binding protein 1 ( PTBP1 ) , heterogeneous nuclear ribonucleoprotein L ( hnRNP L ) , MATRIN 3 , and splicing factor , arginine and serine-rich 14 ( SFRS14; Figure 3—figure supplement 1C ) . We were able to confirm the preferential interaction between these proteins and the RSE by immunoblotting of purified mRNPs or co-immunoprecipitation experiments ( Figure 3 and additional data not shown ) . 10 . 7554/eLife . 11155 . 009Figure 3 . Accumulation of PTBP1 on the 3‘UTR prevents UPF1 binding . ( A ) Schematic of reporter mRNA constructs used in immunoprecipitation assays . ( B ) PTBP1 is reduced on transcripts containing the RSE mutants lacking the putative PTBP1 binding sites . Immunoprecipitations were performed as in Figure 2B . Goat IgG samples were imaged using identical settings to those used for anti-UPF1 samples . ( C ) Left Panel: Quantification of relative RNA recovery upon PTBP1 immunoprecipitation , normalized to the recovery of co-transfected GFP-bGH mRNAs . Right Panel: Quantification of relative RNA recovery upon UPF1 immunopreciptation , normalized to the recovery of co-transfected GFP-bGH mRNAs . The amount of RSE-containing RNA recovered was set to 1 . Error bars indicate ± SD; n ≥ 3 ( *p<0 . 05 in two-tailed Student’s t-tests when compared to RSE-ΔPTB recovery . See also Figure 3—figure supplement 1 and Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 00910 . 7554/eLife . 11155 . 010Figure 3—source data 1 . Table of raw mass spectrometry data . List of proteins identified in mRNP purifications of PP7-tagged mRNAs containing RSE or antisense ( AS ) sequence , with numbers of peptides and spectral counts derived from each protein indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 01010 . 7554/eLife . 11155 . 011Figure 3—figure supplement 1 . Identification of RSE-interacting proteins by tandem mass spectrometry . ( A ) Schematic of β-globin reporter mRNA constructs used for PP7-based affinity purification . The RSE sequence ( top ) and a control sequence , the antisense RSE sequence ( AS-RSE; bottom ) were inserted into reporter mRNAs containing the β-globin gene and the GAPDH 3’UTR . ( B ) mRNP profiles of the transcripts described in ( A ) . Proteins were separated on a 4–12% SDS-PAGE gel and visualized by Krypton Infrared Protein Stain ( Pierce ) . ( C ) Purified mRNPs were subjected to trypsin digestion and tandem mass spectrometry . Spectral counts from selected proteins enriched in either the RSE-containing sample or the antisense RSE sample are shown . Complete mass spectrometry data are compiled in Figure 3—source data 1 . ( D ) Control for post-lysis reassortment . Extracts from cells in which protein A-tagged UPF1 was co-expressed with the indicated GFP reporter mRNAs or extracts from cells separately expressing protein A-tagged UPF1 and the exogenous mRNAs were used for affinity purification . mRNAs containing the GAPDH artificial 3’UTR were used as recovery controls . Top panels: Northern blots of co-transfected reporter mRNAs . Identical settings were used to image northern blots of input extracts and purified material from co-expressed and mixed samples , respectively . Bottom panels: immunoblotting of mixed extracts with an α-UPF1 antibody indicates equal purification efficiency across all conditions . ( E ) Relative recoveries of the indicated mRNAs with affinity purified UPF1 were determined , normalized to recovery of mRNAs containing the GAPDH 3’UTR . Error bars indicate ± SD; n = 3 ( **p<0 . 01 in two-tailed Student’s t-test comparing the recovery of RSE-GAPDH to RSE-∆PTB-GAPDH ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 011 Among the putative RSE-binding proteins identified , we considered PTBP1 to be a promising putative anti-NMD factor for several reasons . In addition to its established role in alternative splicing , PTBP1 has been implicated as an RNA stability factor ( reviewed in Romanelli et al . , 2013 ) . Moreover , as expected for a protein responsible for protecting specific mRNAs , PTBP1 has been found to be preferentially associated with long 3’UTRs ( Gama-Carvalho et al . , 2006 ) . Finally , a variety of techniques have been used to characterize the interaction specificity of PTBP1 and establish that the protein’s four RNA recognition motifs ( RRMs ) each contribute to binding of degenerate pyrimidine-rich sequences . UV cross-linking and immunoprecipitation ( CLIP ) experiments indicate that PTBP1 binding is driven both by high-affinity heptameric or hexameric motifs and by local enrichment of pyrimidines ( Xue et al . , 2009; 2013; Han et al . , 2014 ) . Indicative of the potential for robust PTBP1 binding , examination of the RSE sequence revealed as many as eleven CU-rich clusters corresponding to putative PTBP1 binding sites spread throughout the 400 nt element . This includes three instances of CU-rich hexameric sequences identified as over-represented in PTB CLIP sequence tags , as well as five additional clusters containing heptamers enriched in CLIP tags ( Xue et al . , 2009; see Materials and methods for sequences ) . Therefore , we predicted that multiple PTBP1 molecules might be recruited to RSE-containing mRNPs , forming an RNP refractory to UPF1 association . To test this idea , we constructed two RSE sequence variants . First , we mutated all putative PTBP1 binding sites in the RSE by replacing a subset of U and C residues in CU-rich clusters with G and A residues , respectively ( RSE-ΔPTB; see Materials and methods for sequences ) . Next , we inserted six previously characterized PTBP1 binding sites into the RSE-ΔPTB sequence at arbitrary positions ( Figure 3A; Xue et al . , 2009 ) , with ~60 nt between each site ( RSE-ΔPTB+6xPTBBS ) . The wild-type RSE and the RSE mutants were inserted into reporter mRNAs upstream of the artificial GAPDH-derived 3’UTR , and the extent of PTBP1 association with the reporter transcripts was assayed by monitoring the co-purification of reporter mRNAs with immunopurified endogenous PTBP1 . As shown in Figures 3B and C , PTBP1 association with the RSE-ΔPTB-containing transcripts was significantly reduced compared to the wild-type RSE-containing transcripts . Reintroducing six PTBP1 binding sites into the RSE-ΔPTB led to recovery of PTBP1 association to levels similar to those found in association with the wild-type RSE . We next investigated whether UPF1 binding was affected by eliminating the putative PTBP1 binding sites in the RSE sequence . To test the possibility that PTBP1 binding causes exclusion of UPF1 from the mRNP , we assayed the co-purification of the same reporter mRNAs with immunopurified endogenous UPF1 . Northern blots of co-purified mRNAs indicated transcripts containing the RSE-ΔPTB sequence displayed substantially increased UPF1 association ( Figures 3B and C ) . The inverse relationship between PTBP1 and UPF1 association is consistent with our hypothesis that PTBP1 prevents UPF1 from associating with the 3’UTRs . Interestingly , although the level of UPF1 association on the transcripts containing RSE-ΔPTB+6xPTBBS is significantly reduced compared to RSE-ΔPTB , it was still higher than that of the wild-type RSE , indicating that the natural arrangement of the PTBP1 binding sites in the RSE may be optimal for antagonizing UPF1 binding . As a control to ensure that the PTBP1-mediated inhibition of UPF1 binding was due to cellular interactions rather than a consequence of protein-RNA reassortment in extracts , we further tested the ability of stably expressed tandem protein A-tagged UPF1 to co-purify mRNAs expressed in distinct cells ( Mili and Steitz , 2004 ) . In these experiments , we observed co-purification of mRNAs when tagged UPF1 and mRNAs were expressed in the same cells , but none when reporter RNA-containing extracts from cells lacking tagged protein were mixed with extracts containing tagged UPF1 ( Figure 3—figure supplement 1D , Figure 3—figure supplement 1E ) . As the physiological role of the RSE is the stabilization of unspliced RSV RNAs , we investigated the role of PTBP1 in stabilization of authentic viral RNAs ( Figure 4A ) . To do so , we replaced the wt RSE sequence in the unspliced RSV RNA with the RSE-ΔPTB sequence or RSE-ΔPTB variants containing three or six artificial PTBP1 binding sites ( RSE-ΔPTB+3xPTBBS or RSE-ΔPTB+6xPTBBS , respectively ) and measured the steady state levels of the viral RNAs in chicken embryonic fibroblasts . In these experiments , the RSE-ΔPTB-containing viral RNAs were present at much lower levels than RNAs protected by the wt RSE sequence . Viral RNAs containing the RSE-ΔPTB+3xPTBBS sequence exhibited substantial rescue due to the inclusion of artificial PTB binding sites ( Figure 4B , C ) , and six PTBP1 binding sites led to almost complete restoration of RNA levels , consistent with the effects of this RSE variant on PTBP1 and UPF1 binding ( Figure 3 ) . 10 . 7554/eLife . 11155 . 012Figure 4 . PTBP1 plays an essential role in the RSE’s protective activity against NMD . ( A ) Schematic of unspliced RSV RNA expression constructs used for RNA accumulation assays in chicken fibroblasts . ( B ) RNase protection assays of RSV unspliced viral RNA containing the wt RSE or RSE variants . Experimental and control constructs were co-transfected into CEFs , and total cellular RNAs were harvested 43–48 hr post-transfection . Top band: protected fragment of the probe corresponding to the unspliced viral RNAs from the experimental constructs . Bottom band: stable viral loading control that protects a different sized fragment of the same probe due to a small in-frame deletion . ( C ) Quantification of RNase protection assays . Levels of the experimental unspliced RSV RNAs ( top band ) were normalized to levels of the transfection control ( bottom band ) . RNA levels are reported as a fraction of RSV RNA containing wt RSE . Error bars indicate ± SD; n ≥ 5 ( *p<0 . 05; ****p<0 . 0001 in two-tailed Student’s t-tests when compared to RSE constructs ) . ( D ) Schematic of reporter mRNAs containing the β-globin gene , RSE variants ( RSE , RSE-ΔPTB , or RSE-ΔPTB+6xPTBBS ) , and the full-length human SMG5 3‘UTR . ( E ) Decay assays of the indicated reporter mRNAs . Tet-regulated transcripts ( upper bands ) were co-transfected with the constitutively expressed wild-type β globin reporter ( pcβwtβ; bottom bands ) in HeLa Tet-off cells . Levels of tet-regulated reporter mRNAs were normalized to levels of the wild-type β-globin transfection control . Half-lives and 95% confidence intervals were obtained from 3 independent experiments ( p-values from two-tailed ANCOVA analyses when compared to pcTET2-βwt-SMG5 ) . See also Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 01210 . 7554/eLife . 11155 . 013Figure 4—figure supplement 1 . PTPB1 protects reporter mRNAs from NMD . Decay assays of pcTET2 -βwt-RSE-ΔPTB-SMG5 in HEK293 Tet-off cells treated with non-targeting or anti-UPF1 siRNAs . The reporter constructs ( upper band ) were co-transfected with the constitutively expressed wild-type β-globin reporter ( pcβwtβ; bottom bands ) into cells 24 hrs after the siRNAs were introduced to the cells . Levels of tet-regulated reporter mRNAs were normalized to levels of the wild-type β-globin transfection control . Relative remaining RNA levels at indicated time points from three independent experiments were used to calculate half-lives and 95% confidence intervals ( ***p<0 . 001 in two-tailed ANCOVA analysis , compared to the half-life of βwt-RSE-ΔPTB-SMG5 in cells treated with non-targeting siRNAs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 013 To test whether PTBP1 binding sites are sufficient to confer stability to NMD targets , we next inserted the wt RSE or the RSE-ΔPTB mutant sequence into reporter mRNAs containing the SMG5 3’UTR ( Figure 4D ) . mRNA decay assays showed a dramatic UPF1-dependent reduction in the stability of the transcripts containing RSE-ΔPTB ( half-life ~120 min ) when compared to transcripts containing the wt RSE ( half life >9 hr; Figure 4E; see Figure 4—figure supplement 1 for UPF1 RNAi experiments ) . In fact , targeted mutation of the putative PTBP1 binding sites in the RSE resulted in transcript stability similar to that of mRNAs containing the inactive antisense RSE sequence . Transcripts containing the RSE-ΔPTB+6xPTBBS variant ( half-life >9 hr ) exhibited stability comparable to transcripts containing the wt RSE , indicating that the RSE regained most or all of its function after restoration of six PTB binding sites at arbitrarily chosen sites ( Figure 4E ) . The results presented above suggest that PTB binding is necessary for the protective activity of the RSE . To determine whether PTB binding downstream of a TC is sufficient to stabilize NMD targets , we constructed a series of reporter transcripts containing three or six model PTBP1 binding sites inserted at the 5’ end of the NMD-inducing SMG5 3’UTR . In these experiments , we tested two arrangements of PTBP1 binding sites: separated by either 8 nt linker sequences ( PTB3 and PTB6 ) or by 100 nt segments of the SMG5 3’UTR to better mimic the arrangement of PTB binding sites in the RSE ( PTB3s and PTB6s; Figure 5A ) . In RNA decay assays , we found that addition of three PTBP1 binding sites significantly improved transcript stability , with the distributed binding sites having a marginally stronger effect ( Figure 5B ) . Importantly , transcripts containing either arrangement of six PTBP1 binding sites were highly stabilized relative to control transcripts , demonstrating that PTBP1 binding is capable of recapitulating RSE function . Consistent with the position-dependence of the full-length RSE ( Figure 1—figure supplement 1 ) , mRNAs in which a cluster of six PTBP1 binding sites were introduced into the SMG5 3’UTR 600 nt or 1200 nt downstream of the TC were highly unstable ( Figure 5—figure supplement 1A , B ) . 10 . 7554/eLife . 11155 . 014Figure 5 . PTBP1 protects NMD-sensitive transcripts from NMD when artificially recruited to the 3‘UTR . ( A ) Schematic of tet-regulated β-globin reporter mRNA constructs used in RNA decay assays . Three or six canonical PTBP1 binding sites were inserted into the SMG5 3‘UTR . PTBP1 binding sites were inserted with short linker sequences between each site ( PTB3 and PTB6 ) , or spaced at 100 nt intervals within the SMG5 3‘UTR ( PTB3s and PTB6s ) . ( B ) Decay assays of reporter mRNAs containing PTBP1 binding sites . Constructs encoding the indicated tet-regulated transcripts ( top bands ) were co-transfected with the constitutively expressed wild-type β-globin reporter ( pcβwtβ; bottom bands ) in HeLa Tet-off cells . Levels of tet-regulated reporter mRNAs were normalized to levels of the wild-type β-globin transfection control . Half-lives and 95% confidence intervals were obtained from 3 independent experiments ( p-values from two-tailed ANCOVA analyses when compared to pcTET2-βwt-SMG5 ) . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 01410 . 7554/eLife . 11155 . 015Figure 5—figure supplement 1 . PTBP1 binding recapitulates the RSE’s position-dependent NMD inhibition . ( A ) Schematic of tet-regulated β-globin reporter mRNA constructs containing six closely spaced PTBP1 binding sites at the 5’ end , in the middle , or at the 3' end of the SMG5 3’UTR . ( B ) Decay assays of reporter mRNAs containing the RSE at different positions in the 3'UTR . Constructs encoding the tet-regulated transcripts described in panel A ( pcTET2-βwt-PTB6-SMG5 , pcTET2-βwt-MPTB6-SMG5 , or pcTET2-βwt-3’PTB6-SMG5; top bands ) were co-transfected with the constitutively expressed wild-type β-globin reporter ( pcβwtβ; bottom bands ) in HeLa Tet-off cells . Remaining RNA levels at indicated time points were normalized to levels of the wild-type β-globin transfection control . Half-lives and 95% confidence intervals were obtained from 3 independent experiments ( p-values from two-tailed ANCOVA analysis when compared to pcTET2-βwt-PTB6-SMG5 ) . ( C ) Schematic of β-globin reporter mRNA constructs containing four MS2 coat protein binding sites at the 5’ end of the SMG5 3’UTR . ( D ) Tethering reporter constructs ( top bands ) were co-transfected with plasmids expressing the indicated MS2 fusion protein ( vector control , PAPBC , or PTBP1 ) and the constitutively expressed wild-type β-globin reporter ( pcβWTβ; bottom bands ) . RNA was harvested 30 min after transcription was halted by addition of doxycycline and at 2 hr intervals thereafter . Remaining RNA levels at indicated time points were normalized to levels of the wild-type β-globin transfection control . Half-lives and 95% confidence intervals were obtained from 3 independent experiments ( p-values from two-tailed ANCOVA analysis when compared to pcTET2-βwt-4xMS2-SMG5 co-transfected with MS2 protein alone ) . ‡ denotes an aberrantly processed mRNA isoform that was excluded from quantification . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 015 To further test the specificity of PTBP1’s role as an inhibitor of NMD , we used a tethering approach to artificially target PTBP1 and other RNA binding proteins to sites immediately downstream from NMD-inducing stop codons . For these assays , we fused proteins of interest with the MS2 bacteriophage coat protein and co-expressed them with reporter mRNAs containing four MS2 hairpins placed at the 5’ end of the SMG5 3’UTR . A PABPC1-MS2 coat protein fusion was included in this assay as a positive control , as PABPC1 has previously been reported to antagonize NMD when tethered to NMD-sensitive mRNAs ( Amrani et al . , 2004 ) . Our results showed that the reporter mRNAs were comparably stabilized by co-expression with MS2-PABC1 or MS2-PTBP1 , relative to co-expression with the MS2 coat protein alone ( Figure 5—figure supplement 1C , D ) . To test the hypothesis that the mechanism used by the RSV viral RNA to evade NMD may be shared by mammalian mRNAs with long 3’UTRs , we performed RNA-seq analysis on 293 Tet-off cells transfected with anti-UPF1 and anti-PTBP1 siRNAs individually or in combination , using a validated non-silencing siRNA as a control . In biological replicates , mRNAs from 495 genes reproducibly decreased in abundance upon depletion of PTBP1 , relative to their expression levels in cells treated with control siRNA ( Figure 6A; see also Figure 6—source data 1 ) , consistent with previous reports implicating PTBP1 in the stabilization of diverse human mRNAs . To determine whether the down-regulated mRNAs were made susceptible to NMD in the absence of PTBP1 , we examined the effects of co-depletion of UPF1 and PTBP1 . Of the transcripts down-regulated upon PTBP1 knockdown , we designated mRNAs from 188 genes observed to reproducibly increase in siPTB/siUPF1 vs siPTB alone as 'rescued . ' On average , concurrent knockdown of UPF1 and PTBP1 caused an increase in mRNA levels among PTBP1-dependent genes ( Figure 6A ) , while the average accumulation of PTBP1-dependent mRNAs was not altered upon UPF1 depletion alone ( Figure 6B ) . Among all of the PTBP1-dependent genes , the extent of mRNA rescue upon co-depletion correlated with the reduction upon PTBP1 knockdown ( Figure 6A , C ) . Consistent with a role for PTBP1 binding near the stop codon , this correlation was dependent on the presence of putative high-affinity PTBP1 hexamer binding sites in the region downstream of the TC ( Figure 6C ) . 10 . 7554/eLife . 11155 . 016Figure 6 . PTBP1 protects many human mRNAs with long 3’UTRs from NMD . ( A ) For mRNAs from 495 genes reproducibly down-regulated by PTBP1 depletion in RNAseq of biological replicates , fold change comparing siPTBP1/siUPF1 to siPTBP1 alone was plotted vs . fold change comparing siPTBP1 to siNT . Mean fold change among all analyzed genes ( dashed line; p<0 . 0001 in two-tailed Student’s t-test of null hypothesis of zero fold change ) , best fit determined by the least squares method ( solid line ) and correlation coefficient ( Spearman’s ρ ) are indicated . See also Figure 6—source data 1 . ( B ) Mean fold change in transcript abundance upon PTBP1 depletion versus siNT treatment was plotted against the mean fold change in abundance between siUPF1 and siNT conditions for the set of mRNAs as described in A . ( C ) Table of Spearman’s correlation coefficients for all mRNAs reproducibly downregulated by siPTBP1 , mRNAs containing one or more putative PTBP1 hexamer binding sites within 400 nt of the termination codon , and mRNAs lacking putative PTBP1 hexamers in that region . ( D ) mRNAs protected from NMD by PTBP1 have long 3’UTRs . Continuous distribution function ( CDF ) plot of annotated 3’UTR lengths among all expressed exemplar mRNAs ( blue ) , mRNAs down-regulated by PTBP1 depletion ( red ) , and mRNAs rescued by co-depletion of PTBP1 and UPF1 in RNA-seq analyses ( green; see Materials and methods for details ) . Statistical significance was evaluated by two-tailed K-S test , comparing the indicated mRNA sets to the distribution of 3’UTR lengths among all exemplar mRNAs . ( E ) qRT-PCR analysis of mRNAs protected from NMD by PTBP1 . Graph of average abundance of mRNAs normalized to housekeeping UBC control mRNAs ( n=3 , error bars indicate SD ) . Statistical significance was determined by two-tailed Student’s t-test , comparing siPTBP1 to siNT and siPTBP1/siUPF1 to siPTBP1 . See also Figure 6—figure supplements 1 , 2 and Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 01610 . 7554/eLife . 11155 . 017Figure 6—source data 1 . Table of mRNAs down-regulated by PTBP1 depletion . Table lists transcript accession numbers , log2 fold changes in gene expression , and p-values from one-way ANOVA analyses when comparing siPTBP1 vs . siNT and siPTBP1/siUPF1 vs . siPTBP1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 01710 . 7554/eLife . 11155 . 018Figure 6—figure supplement 1 . PTBP1-mediated protection does not depend on changes in splicing . Sashimi plots were generated from reads mapping to genes rendered susceptible to NMD by PTBP1 depletion in RNA-seq and qRT-PCR experiments using Integrative Genomics Viewer software ( Broad Institute ) . SELT , VRK3 , and ZNF322 were omitted from this analysis due to insufficient read coverage spanning exons . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 01810 . 7554/eLife . 11155 . 019Figure 6—figure supplement 2 . Analysis of PTBP1 binding to protected mRNAs . Histograms of sequence reads from PTBP1 CLIP experiments ( Xue et al . , 2013 ) mapping to the indicated mRNAs were visualized using the Genomatix genome browser . Schematics of representative RefSeq transcripts are shown , and histograms are autoscaled according to maximum peak height for each gene window . Thin lines indicate introns , thick lines indicate coding regions , and red indicates 3’UTRs . Positions of putative high-affinity hexamer PTBP1 binding sites 3’ of TCs are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 019 As 3’UTR length is a major determinant of NMD , we also asked whether the class of mRNAs protected from NMD by PTBP1 exhibited unusually long 3’UTRs . Indeed , mRNAs down-regulated in PTBP1-depleted cells had a median 3’UTR length of 1577 nt , significantly longer than the median 3’UTR length among all mRNAs expressed in the RNAseq dataset ( 1308 nt; Figure 6D; see Materials and methods for details ) . Most importantly , the population of mRNAs that were both significantly down-regulated by siPTBP1 treatment and rescued by co-depletion of PTBP1 and UPF1 exhibited an even greater median length of 1843 nt . Together , these data suggest that protection of human mRNAs containing long 3’UTRs by PTBP1 is widely exploited to minimize the impact of NMD on the transcriptome . To verify the response of endogenous transcripts to depletion of PTBP1 and UPF1 alone and in combination , we selected a panel of transcripts for follow-up analysis by quantitative reverse transcription-PCR ( qRT-PCR ) . These experiments confirmed that several mRNAs identified in RNA-seq analyses were down-regulated by PTBP1 depletion and fully or partially rescued by co-depletion of PTBP1 and UPF1 ( Figure 6E ) . Importantly , PTBP1 mRNA levels were equal in the PTBP1 and PTBP1/UPF1 double knockdown conditions . Incomplete rescue of PTBP1-dependent mRNAs via UPF1 depletion could either be due to PTBP1 antagonizing multiple RNA decay pathways or due to insufficient UPF1 knockdown for complete abrogation of NMD activity . As alterations in RNA stability could be caused by changes in PTBP1-dependent alternative splicing events , we also examined the RNAseq data for evidence of altered splicing . With the exception of TPM1 , a known target of PTBP1 splicing regulation , we did not observe differences in the splicing patterns of the selected genes ( Figure 6—figure supplement 1 ) . Among nine genes validated by qRT-PCR as protected from NMD by PTB , eight have at least one putative PTB hexamer binding site within 150 nt of the annotated TC , and six of nine have two or more hexamer binding sites in that region ( Figure 6—figure supplement 2 ) In addition , we analyzed previously published whole-genome CLIP data of PTBP1 in HeLa cells to probe a possible relationship between protection from NMD and PTBP1 binding in living cells ( Xue et al . , 2013 ) . This analysis revealed PTBP1 CLIP reads mapping to the 3’UTRs of each of the validated mRNAs listed above ( Figure 6—figure supplement 2; see below for further CLIP analysis ) . In order to understand the global relationship between PTBP1 binding and NMD , we analyzed the RNA-seq data described above with respect to published PTBP1 ( Xue et al . , 2013 ) and UPF1 CLIP-seq datasets ( Zund et al . , 2013 ) . As previously reported , analysis of aggregate binding of UPF1 to all mRNAs showed that UPF1 levels are low in coding sequences and increase downstream of TCs to reach a broad plateau throughout 3’UTRs ( Figure 7A ) . In contrast , PTBP1 CLIP signal derived from exons was preferentially located within 200 nt of TCs , consistent with a role for PTBP1 in marking translation termination events . 10 . 7554/eLife . 11155 . 020Figure 7 . PTBP1 binding near stop codons is correlated with 3’UTR length and NMD evasion . ( A ) Density of PTBP1 ( green , left axis ) and UPF1 ( red , right axis ) CLIP reads derived from peaks called using PIPE-CLIP software , plotted relative to TC position . A bin size of 20 nt was used to determine read density; bin sizes ranging from 5 nt to 20 nt and plots of peak occurrences showed similar patterns . ( B ) CDF plot of annotated 3’UTR lengths among mRNAs containing or lacking PTBP1 CLIP peaks centered in the indicated intervals relative to the TC . P-values were calculated in two-tailed K-S tests of 3’UTR lengths in the indicated mRNA classes compared to mRNAs lacking PTBP1 peaks . Only mRNAs with 3’UTRs greater than 500 nt were included to avoid bias due to selection of transcripts containing CLIP peaks . ( C ) CDF plot of log2 fold changes in mRNA abundance in UPF1 siRNA vs siNT RNAseq . mRNAs with 3’UTR lengths in the middle 50% of the overall distribution ( 484–2251 nt ) were selected to avoid confounding effects of increased 3’UTR lengths among mRNAs with PTBP1 peaks . P-values were determined by two-tailed K-S tests , comparing the indicated mRNA classes to mRNAs lacking PTBP1 peaks from +1–400 nt . ( D ) CDF plot as in C . mRNAs were classified according to the presence of one or more predicted high-affinity PTBP1 hexamer binding sites in the indicated intervals relative to annotated TCs ( see Materials and methods for details; the top 6 hexamers identified in previous PTBP1 CLIP seq , associated with >50% of observed CLIP peaks , were used for this analysis; Xue et al . , 2009 ) . Statistical significance was determined by two-tailed K-S tests comparing mRNAs with putative PTBP1 binding sites in the indicated positions to mRNAs lacking hexamer binding sites at positions from 1 to 50 nt downstream of the TC . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 020 Having observed enrichment of PTBP1 CLIP peaks near TCs , we next asked whether mRNAs with TC-proximal PTBP1 binding exhibited evidence of NMD evasion . In line with a role for PTBP1 in protecting potential NMD substrates , transcripts with PTBP1 CLIP peaks centered within the first 200 nt of the TC had longer 3’UTRs than those lacking peaks near the TC ( Figure 7B ) . As a control , mRNAs with PTBP1 CLIP peaks centered between 200 and 500 nt from the TC exhibited no tendency to contain longer 3’UTRs than normal . Most importantly , mRNAs with PTBP1 peak centers within 200 nt of the TC were significantly less likely to be up-regulated upon UPF1 knockdown than mRNAs without TC-proximal peaks , indicating that this population of PTBP1-bound mRNAs is insensitive to NMD under normal conditions ( Figure 7C ) . To avoid potential confounding effects of differences in 3’UTR lengths among distinct transcript classes , we used mRNAs with 3’UTRs falling between the 25th and 75th percentile of all 3’UTR lengths for these analyses ( mRNAs with 3’UTRS from 484–2251 nt ) . Providing further evidence of the importance of PTBP1 binding position relative to the terminating ribosome , mRNAs with PTBP1 CLIP peaks centered from 200 to 500 nt from the TC showed no significant trend toward protection from NMD ( Figure 7C ) . We further corroborated this analysis by examining the effect of UPF1 depletion on transcripts containing high-affinity PTBP1 binding sites near the TC . Using this approach , we found evidence that PTBP1 hexamer binding sites located in the immediate vicinity of the terminating ribosome are particularly important in determining protection from UPF1-mediated degradation . Specifically , transcripts containing PTBP1 hexamers within 50 nt of the TC were significantly less likely to be up-regulated upon UPF1 depletion than transcripts lacking putative binding sites in this interval , while transcripts with hexamers located 50–100 from the TC showed no apparent protection from NMD ( Figure 7D ) . Together , these data suggest that human mRNAs have evolved to avoid recognition and degradation by the NMD machinery by recruiting PTBP1 to the vicinity of TCs .
Through a combination of functional and biochemical assays , we have shown that the RSE functions through PTBP1 to inhibit NMD . This activity is associated with a reduction in UPF1 binding to transcripts containing long 3’UTRs , an early event in the process of NMD . In our model , the RSE and PTBP1 effectively mask the true length of the downstream 3’UTR from UPF1 , allowing the mRNA to elude surveillance . The ability of PTBP1 to antagonize UPF1 binding depends on its location on the mRNA; PTBP1 binding at the beginning of the 3’UTR causes a loss of UPF1 binding , while relocation to distal sites reduces this function . This position-dependence has the advantage of allowing stabilization in response to translation termination at the proper stop codon while preserving the ability of NMD to detect premature TCs . In contrast , a position-insensitive mechanism would prevent NMD from accurately detecting de novo nonsense mutations , reducing the scope of quality control . In its role as a splicing regulator , PTBP1 functions in a partially redundant manner with its close paralogs , PTBP2 and PTBP3/ROD1 ( Keppetipola et al . , 2012 ) . In many human cell types , including those used for this study , depletion of PTBP1 leads to induction of PTBP2 , which compensates for a subset of PTBP1 functions . As it possesses highly similar biochemical activities , PTBP2 may also stabilize a subset of mRNAs usually protected by PTBP1 , meaning that the population of mRNAs identified here as protected from NMD is likely to underestimate the role of PTB paralogs in NMD inhibition . In addition , we expect that marking genuine TCs with PTBP1 is just one strategy employed by long 3’UTRs to evade NMD . Supporting the existence of diverse solutions to the problem , yeast Pub1p was found to stabilize certain NMD targets by means that remain undefined ( Ruiz-Echevarria and Peltz , 2000 ) , and manipulation of 3’UTR structure to bring the poly-A tail in spatial proximity to the TC has been shown to stabilize reporter mRNAs ( Eberle et al . , 2008 ) . Extensive work on the role of PTB proteins in mediating exon exclusion in pre-mRNA splicing provides a template for a proposed mechanism for PTBP1’s anti-UPF1 activity . PTBP1 contains four RRMs , each with affinity for CU-containing sequences ( Oberstrass et al . , 2005 ) . Of these , RRM3 and RRM4 are oriented back-to-back , resulting in RNA loops of 15 nt or more between sites of PTBP1 contact ( Lamichhane et al . , 2010 ) . These short-range looping events may also be accompanied by capture of distant CU-rich sequences by individual RRMs forming larger looped RNA structures ( Clerte and Hall , 2009; Kafasla et al . , 2009 ) . It has been suggested that long-range interactions involving multiple PTBP1 molcules bound to a single RNA may be responsible for excluding binding of splicing factors , resulting in splicing repression ( Chou et al . , 2000; Amir-Ahmady et al . , 2005; Cherny et al . , 2010; Lamichhane et al . , 2010; Sharma et al . , 2011 ) . Alternatively , PTBP1 may interact with proteins such as MATRIN 3 to promote the assembly of higher-order repressive complexes ( Joshi et al . , 2011 ) . By direct analogy to its role in splicing , PTBP1 assembled on multiple TC-proximal CU-rich sequences could thus result in a zone of UPF1 exclusion , preventing the initial assembly of UPF1 into mRNPs ( Figure 8 ) . In addition to preventing initial UPF1 association with 3’UTRs , PTBP1 could affect UPF1 already bound to an mRNA . In this scenario , PTBP1 may , as has been recently described for cytoplasmic poly-A binding protein , stimulate ATPase-dependent dissociation of UPF1 from the mRNP ( Lee et al . , 2015 ) . 10 . 7554/eLife . 11155 . 021Figure 8 . Model for NMD inhibition by PTBP1 . In the absence of the RSE or other sequences capable of recruiting PTBP1 , UPF1 binds 3’UTRs in a length-dependent manner , potentiating NMD ( top ) . PTBP1 can bind the RSE at multiple sites throughout the 400 nt sequence . This establishes an mRNP structure in the vicinity of the stop codon that is refractory to UPF1 binding , possibly by ( 1 ) blocking 5’-3’ translocation of UPF1 or ( 2 ) preventing initial binding of UPF1 to the mRNA . The NMD pathway thus judges the 3’UTR to be short and fails to degrade the mRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 11155 . 021 The PTB family of proteins is a pioneering example of the coupling of alternative splicing regulation with NMD . A series of studies has elucidated a complex regulatory network in which PTBP1 and PTBP2 auto- and cross-regulate each other to favor production of PTC-containing transcripts that are efficiently degraded by NMD ( Markovtsov et al . , 2000; Romanelli et al . , 2000 , 2005; Rahman et al . , 2002; Wollerton et al . , 2004; Boutz et al . , 2007; Spellman et al . , 2007 ) . In most tissues , PTBP1 regulates the alternative splicing of PTBP2 to favor production of PTC-containing transcripts that are efficiently degraded by NMD . In the brain and testes , PTBP1 is down-regulated , allowing increased expression of PTBP2 and execution of tissue-specific splicing programs . As in their roles in alternative splicing , we expect that PTBP1 and PTBP2 may stabilize partially overlapping sets of potential NMD substrates in distinct cell types ( Keppetipola et al . , 2012 ) .
Wild-type RSE , RSE mutants , anti-sense RSE , and SMG5-397 sequences were amplified by PCR from the E/S wild-type RSV vector ( Withers and Beemon , 2011 ) , pcTET2-βwt-SMG5 ( Jens Lykke-Andersen , UCSD ) or synthesized ( Integrated DNA Technologies , Coralville , IA ) and inserted into reporter mRNAs at the NotI site of pcTET2-βwt-SMG5 , pcTET2-βwt-GAPDH , or pcTET2-βwt-GAPDH-AdML or the BamHI ( 5’ ) or PmeI ( 3’ ) sites of pcDNATPEGFP-GAP ( Hogg and Goff , 2010; Singh et al . , 2008 ) . To create pcTET2-βwt-PTB3-SMG5 and pcTET2-βwt-PTB6-SMG5 or pcTET2-βwt-4xMS2-SMG5 , DNA cassettes containing three or six PTBP1 binding sites ( CTCTCTCTTCTTCTT and TCTTCTTCTTCTTCT; as used in Xue et al . , 2009 ) or four MS2 binding sites , respectively , were introduced into the NotI site in the parental pcTET2-βwt-SMG5 plasmid . For pcTET2-βwt-MPTB6-SMG5 , an AgeI site was engineered into the SMG5 sequence of pcTET2-βwt-SMG5 600 nt downstream of the TC , and DNA cassettes containing six PTBP1 binding sites were introduced into the AgeI site . To construct pcTET2-βwt-3’PTB6-SMG5 , DNA cassettes containing six PTBP1 binding sites were introduced into the XbaI site in the parental pcTET2-βwt-SMG5 plasmid , 1343 nt downstream of the TC . In pcTET2-βwt-PTB3s-SMG5 and pcTET2-βwt-PTB6s-SMG5 , DNA fragments containing the PTBP1 binding sites at desired locations were synthesized ( Life Technologies , Frederick , MD ) and cloned into pcTET2-βwt-SMG5 between NotI and AgeI sites . To generate RSV RSE variants , RSE-∆PTB and RSE-∆PTB+3xPTBBS sequences were amplified by PCR from the respective pcTET2-βwt-SMG5 variant plasmids and cloned into the E/S wild-type RSV vector between EagI and SpeI sites to replace the wild-type RSE sequence ( Withers and Beemon , 2011 ) . For stable expression of tagged UPF1 , pBABE hygro ( Morgenstern and Land , 1990 ) was modified to contain tandem protein A domains and a TEV protease cleavage site between the NaeI and NotI sites , and full-length UPF1 coding sequences was inserted between the NotI and XhoI sites . To construct plasmids expressing PTBP1 or PABPC1 with the MS2 coat protein fused at the N-terminus , the MS2cp V75E; A81G mutant ( LeCuyer et al . , 1995 ) was amplified by PCR and inserted into the NotI site of pQCXIP ( Clontech , Mountain View , CA ) . DNA fragments encoding PTBP1 or PABC1 were amplified from respective cDNAs ( Open Biosystems/Dharmacon , Lafayette , CO ) and cloned into pCQXIP-MS2CP plasmid between BamHI and XhoI sites . Sequences of siRNAs ( Thermo Scientific , Philadelphia , PA; Dharmacon , Lafayette , CO ) used are as follows: UPF1 , GAUGCAGUUCCGCUCCAUUUU; PTPB1 , CUUCCAUCAUUCCAGAGAAUU; non-targeting , UAAGGCUAUGAAGAGAUAC ( Mendell et al . , 2004; Wagner and Garcia-Blanco , 2002 ) . HEK-293T cells maintained in DMEM supplemented with 10% fetal bovine serum , 100U of penicillin/streptomycin , and 0 . 3 mg/ml L-Glutamine ( Life Technologies ) were used for mRNP purification experiments . Cells were transfected with desired constructs by calcium phosphate as described ( Hogg and Collins , 2007 ) . Cells were harvested 48 hr post transfection , resuspended in hypotonic lysis buffer ( HLB; 20 mM HEPES pH 7 . 6 , 2 mM MgCl2 , 10% glycerol , 1 mM DTT , supplemented with protease inhibitors ) , subjected to freeze-thaw lysis , and extracted in 150 mM NaCl as described ( Hogg and Collins , 2007 ) . PP7-based mRNP purification and immunoprecipitation were performed as described , with minor modifications ( Hogg and Goff , 2010 ) . For immunoprecipitation , goat anti-Upf1 antibody ( Santa Cruz ) , goat anti-PTBP1 antibody ( Abcam , Cambridge , MA ) , or nonspecific goat IgG ( Sigma , St . Louis , MO ) pre-bound to protein G Dynabeads ( Life Technologies ) were incubated in extracts for 1 hr , followed by extensive washing with HLB supplemented with 150 mM NaCl and 0 . 1% NP-40 . Bound RNA was eluted from beads in Trizol reagent ( Life Technologies ) . Where indicated , cells were treated with 100 µg/ml puromycin ( Sigma , St . Louis , MO ) for 4 hr prior to cell extract preparation . To quantify the relative recovery of RNA , the amount of RNA in the bound fraction was divided by the amount in the total fraction for experimental and recovery control RNAs , and the recovery of the experimental RNA was normalized using the recovery of the co-expressed control RNA . Average and SD of relative mRNA recoveries from biological replicates performed using extracts from cells transfected separately are reported . Experiments to test for post-lysis rearrangement of protein-RNA interactions were conducted as described above , except that parental 293 Tet-off cells or cells stably transduced with retroviral vectors encoding tandem protein A-tagged UPF1 were transfected with mRNA expression vectors . Equal amounts of extract ( determined by total protein content ) prepared from cells expressing both tagged protein and exogenous mRNA were mixed with extracts from parental cells lacking exogenous RNA expression , or extracts from parental cells expressing only exogenous mRNAs were mixed with extracts from cells expressing only tagged UPF1 . Affinity purifications were carried out using rabbit IgG conjugated with M-270 epoxy Dynabeads ( Life Technologies ) according to the manufacturer’s instructions . For mRNP purification , whole cell extracts of HEK293T cells were adjusted to 5 mg/mL protein in HLB with 150 mM NaCl . Purified ZZ-tev-PP7CP ( 0 . 5 µg ) was mixed with 500 µl of extracts by rotating for 60 min at 4°C . 50 µl of the Rabbit IgG-conjugated M-270 Dynabeads were added to 500 µl of extracts premixed with ZZ-tev-PP7CP . The mixture was rotated for an additional 60 min at 4°C to allow binding . The beads were then washed extensively in HLB supplemented with 150 mM NaCl , 0 . 1% NP-40 , and 1 mM DTT . Bound RNA and protein were eluted from beads in LDS sample buffer ( Life Technologies ) or in HLB supplemented with 1 M NaCl ( in preparation for mass spectrometry ) . Proteins eluted from mRNP purifications were precipitated in cold acetone ( Thermo Scientific ) and subsequently resuspended in 6 M urea ( Thermo Scientific ) . Proteins were reduced in 20 mM DTT ( Thermo Scientific ) , alkylated in 50 mM iodoacetamide ( Thermo Scientific ) , diluted to 2 M urea , and digested with trypsin ( Promega , Madison , WI ) overnight . Peptides were purified using C18 Zip-Tips ( EMD Millipore , Taunton , MA ) , according to the manufacturer’s instructions . Samples were loaded onto a Zorbax C18 trap column ( Agilent Tech , Santa Clara , CA ) using an on-line Eksigent ( Dublin , CA ) nano-LC ultra HPLC system to further desalt the peptide mixture . Peptides were separated on a 10 cm Picofrit C18 column ( New Objective , Woburn , MA ) using a 90 min linear gradient of 5–35% acetonitrile/water containing 0 . 1% formic acid . The samples were ionized via electrospray ionization ( ESI ) in positive mode and analyzed on a LTQ Orbitrap Velos mass spectrometer ( Thermo Electron , San Jose , CA ) . LC MS/MS experiments were carried out using a 'data-dependent' analysis in which the top 6 most intense precursor ions from the full MS precursor scan were selected for fragmentation via collision-induced dissociation ( CID ) . Precursor ions were measured in the orbitrap at a resolution of 30 , 000 ( m/z = 400 ) and all fragment ions were analyzed in the linear ion-trap . All LC MS/MS data was searched against the human Swissprot database ( 20312 sequences ) using MASCOT to obtain peptide and protein identifications . Mass tolerances of ± 20 ppm and ± 0 . 8 Da were selected for precursor and fragment ions , respectively . Trypsin was specified as the digestion enzyme allowing for up to 2 missed cleavages . Carbamidomethylation ( C ) was selected as a static modification and oxidation ( M ) was selected as a variable modification . MASCOT database search results were then imported into Scaffold ( Proteome Software , Portland , OR ) for further validation of peptide and protein identifications . HeLa Tet-off Advance cells ( Clontech , Mountain View , CA ) maintained in DMEM supplemented with 10% FBS and 5 ng/ml doxycycline ( Sigma , St . Louis , MO ) were transfected in 60 mm plates using Turbofect ( Thermo Scientific ) , according to the manufacturer’s instructions . For each plate , 800 ng of the indicated pcTET2-βwt plasmid was co-transfected with 200 ng pcβwtβ and 1000 ng pcDNA3 . 1 empty vector ( Lykke-Andersen et al . , 2000; Singh et al . , 2008 ) . Cells were split into four equal aliquots in 12-well plates 24 hr post-transfection . The next day , cells were washed with 1 ml PBS and incubated in medium without doxycycline for 4 hr . Transcription was quenched by adding doxycycline to a final concentration of 1 µg/ml , and cells were harvested in Trizol ( Life Technologies ) after 30 min and at the indicated intervals . mRNA levels from biological replicates performed using extracts from cells transfected separately are reported for all decay assays . Prism software ( GraphPad ) was used to graph the fraction of remaining RNA at the indicated time points on semi-log plots , obtain the rate of RNA decay ( kdecay ) by fitting to a linear equation using the least squares method , determine 95% confidence intervals , and test the significance of differences among the slopes ( two-tailed ANCOVA ) . The half-lives of the mRNAs were calculated using the equation: t1/2 =ln2/kdecay . To conduct mRNA decay assays in cells depleted for specific factors by RNAi , 293 Tet-off cells were first transfected with 120 pmol ( single siRNA treatment ) or 60 pmol ( dual siRNA treatment ) of each of the indicated siRNAs ( Thermo Scientific; Dharmacon , Lafayette , CO ) in 6-well plates using Lipofectamine RNAiMAX ( Life Technologies ) , according to the manufacturer’s instructions . Cells were split into four equal aliquots in 24-well plates 24 hr post-transfection . The next day , cells were transfected with reporter constructs using Turbofect ( Thermo Scientific ) . For each well , 100 ng of the indicated pcTET2-βwt plasmid was co-transfected with 25 ng pcβwtβ and 125 ng pcDNA3 . 1 empty vector , and mRNA decay assays were performed and analyzed as described above . RNA was isolated using Trizol ( Life Technologies ) and resolved on formaldehyde/agarose gels . 32P-labeled in vitro transcribed probes against β-globin or random-primed DNA probes against GFP were used for detection of mRNAs . Northern blots were imaged on Storm or Typhoon Trio scanners and quantification was performed using ImageQuant software ( GE Healthcare , Pittsburgh , PA ) . Secondary chicken embryo fibroblast ( CEF ) cultures were grown at 39°C and 5% CO2 in medium 199 supplemented with 2% tryptose phosphate broth , 1% chick serum , 1% calf serum and 1% antibiotic-antimycotic . Transient transfections were performed as described ( Paca et al . , 2000 ) , using 100 µg/mL of DEAE-dextran in serum-free medium 199 . Cells were transfected with 3 µg of DNA in 6 cm dishes that were 80% confluent . Probe generation and RNase protection assays were performed as previously described ( LeBlanc and Beemon , 2004 ) . Total cellular RNA was harvested from CEFs 43–48 hr post-transfection using RNA-Bee ( Tel-test ) . RNAs were hybridized overnight with ~250 , 000 cpm of [α-32P] GTP radiolabeled probe , then digested with 10 U/mL of RNase T1 and 5 µg/mL of RNase A at 33°C for 45 min . Sodium dodecyl sulfate and proteinase K were added and samples were incubated at 37°C for 20 min to halt digestion . RNAs were extracted with phenol-chloroform-isoamyl alcohol ( 25:24:1 ) , followed by ethanol precipitation . RNAs were resuspended in 95% formamide loading dye and denatured for 3 min at 95°C . Samples were electrophoresed on 6% acrylamide-8 M urea sequencing gels and RNA levels were quantified using a Typhoon Phosphorimager ( GE Healthcare ) and ImageQuant software ( GE Healthcare ) . mRNA levels from biological replicates performed using extracts from cells transfected separately are reported . 293 Tet-off cells were transfected with siRNAs in 6-well plates using RNAiMAX ( Life Technologies ) , as described above . Cells were harvested 72 hr post-transfection and total RNA was extracted using RNeasy Mini Kit ( Qiagen , Venlo , Netherlands ) . 500 ng of each total RNA sample was used for cDNA synthesis using the Maxima First Strand cDNA Synthesis Kit for RT-qPCR ( Thermo Scientific ) . cDNAs were diluted 1:40 with water and used for qPCR with the FastStart Essential DNA Green Master kit on a LightCycler 96 thermocycler ( Roche , Basel , Switzerland ) . Relative transcript abundances were calculated by the ΔΔCt method , and statistical significance was assessed by two-tailed Student’s t-test ( Prism software , GraphPad ) . mRNA levels from biological replicates performed using extracts from cells transfected separately are reported . For RNA-seq , 1 μg of each total RNA sample was used for library preparation . Ribosomal RNAs ( rRNAs ) were subtracted from total RNA with Ribo-Zero rRNA Removal Kits ( Human/Mouse/Rat; Epicentre ) . Following purification , RNA was fragmented and converted into first strand cDNA using reverse transcriptase and random primers , followed by second strand cDNA synthesis using DNA Polymerase I and RNase H . Following adapter ligation , products were purified and amplified by PCR to create final cDNA libraries . cDNA libraries were validated using an Illumina Miseq and paired-end 50 bp sequencing was performed on an Illumina HiSeq 2000 instrument . Raw data in FASTQ format was aligned to the reference genome hg19 using TopHat2/bowtie2 . Reads mapping to the 3’UTRs of RefSeq genes were counted by Bedtools . Every RefSeq transcript and its 3’UTR expression level were normalized by three steps: calculating raw reads into RPKM ( reads per-kilobase-per million ) , 75th percentile normalization and log2 transformation . The dataset was filtered to remove mRNAs represented by fewer than 25 3’UTR-mapped reads , and differentially expressed genes were selected by one-way ANOVA analysis ( Figure 6; p<0 . 05 ) . The RefSeq transcript with the longest annotated 3’UTR for each gene was used for construction of the reference set of mRNAs used in all subsequent analyses ( designated 'exemplar' RefSeq mRNAs ) . For Figures 6 and 7 , the six most highly enriched hexamers in PTBP1 CLIP experiments ( Xue et al . , 2009; UUCUCU , UCUUCU , UCUCUU , UCUCUG , CUUUCU , and CUUCUC ) , representing over 50% of all PTBP1 CLIP peaks , were used . PTBP1 ( GSE42701; Xue et al . , 2013 ) and UPF1 ( GSE47976; Zund et al . , 2013 ) CLIP datasets were obtained from the NCBI SRA database . For UPF1 data , reads not starting with GTT ( derived from the iCLIP RT primer ) were removed to avoid ambiguous reads , and duplicate reads with identical degenerate barcode sequences were removed with fastx_collapser ( http://hannonlab . cshl . edu/fastx_toolkit/commandline . html ) . The first 7 nucleotides ( containing barcodes ) of collapsed reads were trimmed before alignment . For PTBP1 , sequence duplicates were collapsed , and the first three nucleotides were trimmed before alignment . Novoalign ( Novocraft ) , which allows the detection of small insertions , deletions and substitutions , was used to map CLIP reads to the hg19 human genome . Basic parameters were the same as those previously described ( Moore et al . , 2014 ) : a maximum of two-nucleotide mismatches and a minimum of 25 high quality matches were allowed . Bam files were then input into PIPE-CLIP ( Chen et al . , 2014 ) to identify peaks with default parameters for iCLIP ( UPF1 ) and HITS-CLIP ( PTBP1 ) . The peaks were annotated with Pavis ( Huang et al . , 2013 ) , and HOMER ( Heinz et al . , 2010 ) was used to quantify CLIP tag density around stop codons . Only peaks narrower than 500 nucleotides were used for further analysis . Stop codon chromosomal locations were retrieved from the refGene table in hg19 of UCSC . For Figure 7B , only mRNAs with 3’UTRs greater than 500 nt were analyzed to avoid bias due to the selection of transcripts containing CLIP peaks in the indicated intervals . Sequences used for RSE wild-type and mutant constructs are listed below , with mutated putative PTBP1 binding sites underlined . | Genes are used as templates to create molecules of messenger RNA ( mRNA ) that contain all the information needed to make a protein . This information begins with a 'start site' and ends with a 'stop site . ' The regions of the mRNA outside of the start and stop sites are called untranslated regions . Not all mRNAs are correctly made , and cells combat this problem by detecting and destroying faulty mRNAs before they are translated into protein . One way cells do this is by recognizing and destroying mRNAs that include long untranslated regions , which can indicate that the mRNA might have a stop site too early in its sequence . A key problem with this mechanism , however , is that long untranslated regions also serve important roles in the cell: for example , by determining where and when mRNA molecules are read to make protein . How then do mRNAs with long but important untranslated regions escape detection and degradation ? Ge et al . have now investigated this question using an approach that allows a 'handle' to be attached to particular RNA molecules . This allows the RNA and any proteins bound to it to be purified away from all other RNAs and proteins in the cell , and the proteins can then be identified by a technique called mass spectrometry . Ge at al . found that mRNAs can recruit a protein called PTBP1 to part of the RNA sequence near the stop site . This prevents an RNA decay protein recognizing and triggering the degradation of the mRNA , even if the mRNA has a long untranslated region . Thus , PTBP1 plays a crucial role in protecting human RNAs with long untranslated regions from destruction by the nonsense-mediated decay pathway . Some viral RNAs are also able to evade decay , and so Ge et al . hypothesize that the virus stole this method for maintaining its RNAs from host cells . A future goal is to understand whether this system works the same way in all cell types or protects different RNAs in different cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2016 | Polypyrimidine tract binding protein 1 protects mRNAs from recognition by the nonsense-mediated mRNA decay pathway |
Free heme is cytotoxic as exemplified by hemolytic diseases and genetic deficiencies in heme recycling and detoxifying pathways . Thus , intracellular accumulation of heme has not been observed in mammalian cells to date . Here we show that mice deficient for the heme transporter SLC48A1 ( also known as HRG1 ) accumulate over ten-fold excess heme in reticuloendothelial macrophage lysosomes that are 10 to 100 times larger than normal . Macrophages tolerate these high concentrations of heme by crystallizing them into hemozoin , which heretofore has only been found in blood-feeding organisms . SLC48A1 deficiency results in impaired erythroid maturation and an inability to systemically respond to iron deficiency . Complete heme tolerance requires a fully-operational heme degradation pathway as haplo insufficiency of HMOX1 combined with SLC48A1 inactivation causes perinatal lethality demonstrating synthetic lethal interactions between heme transport and degradation . Our studies establish the formation of hemozoin by mammals as a previously unsuspected heme tolerance pathway .
Hemolysis can arise from genetic mutations ( Larsen et al . , 2012 ) , parasitic and bacterial infections ( Orf and Cunnington , 2015 ) , and drug-induced autoimmune reactions ( Garratty , 2009 ) . Rupturing of red blood cells ( RBCs ) releases high concentrations of heme , which promotes reactive oxygen species ( ROS ) production ( Fibach and Rachmilewitz , 2008 ) and inflammation ( Dutra et al . , 2014 ) causing tissue damage and death . Consistent with the cytotoxicity of heme , genetic ablation of the heme degrading enzyme , heme oxygenase 1 ( HMOX1 ) , results in >90% embryonic lethality ( Kovtunovych et al . , 2010; Poss and Tonegawa , 1997a; Poss and Tonegawa , 1997b ) . The few mice that survive are susceptible to hemolytic infections ( Seixas et al . , 2009 ) , and have little or no reticuloendothelial macrophages ( Kovtunovych et al . , 2010 ) , a phenotype attributed to heme cytotoxicity . The majority of body heme is encapsulated within RBCs , and each RBC contains over one billion heme molecules . As RBCs undergo senescence or become damaged , they are engulfed by macrophages from the reticuloendothelial system ( RES ) through a process termed erythrophagocytosis ( EP ) ( Ganz , 2012; Ganz and Nemeth , 2006; Klei et al . , 2017; Winter et al . , 2014 ) . As RBCs contain both heme and non-heme iron , transporters for each metabolite are recruited to the erythrophagosomal membranes ( Delaby et al . , 2012; Soe-Lin et al . , 2010 ) . The heme responsive gene-1 ( HRG1/SLC48A1 ) transports heme from the erythrophagosome into the cytosol ( Rajagopal et al . , 2008; White et al . , 2013 ) . Heme is enzymatically degraded by HMOX1 ( Kovtunovych et al . , 2010; Kovtunovych et al . , 2014 ) to liberate iron , which can either be stored in ferritin ( FTH1 ) or exported out of the cell by ferroportin ( SLC40A1 ) to be reutilized for new RBC production ( Knutson et al . , 2005; Knutson et al . , 2003 ) , including developing erythroblasts in the bone marrow ( Raja et al . , 1999 ) . Excess heme is exported from the cell by feline leukemia virus subgroup C cellular receptor 1 ( FLVCR1 ) to prevent heme toxicity ( Keel et al . , 2008 ) . Consequently , genetic disruption in HMOX1 , SLC40A1 , FTH1 , and FLVCR1 - steps in the heme-iron recycling pathway - causes embryonic lethality in mice . Here , we show that mice lacking the heme transporter SLC48A1 are viable despite accumulating high concentrations of heme . These animals are heme tolerant because they sequester heme within enlarged lysosomes in the RES macrophages and form crystalline hemozoin , which heretofore has only been found in blood-feeding organisms ( Shio et al . , 2010; Toh et al . , 2010 ) . Our work suggests the existence of a previously unknown pathway for heme detoxification and tolerance in mammals .
To uncover the in vivo function of SLC48A1 in mammals , we generated SLC48A1-deficient mice using CRISPR/Cas9 gene editing . Guide RNAs targeting exon 1 of mouse SLC48A1 ( Figure 1A ) produced seven mutant alleles in C57BL/6J × 129/SvJ F1 animals ( Table 1 in Supplementary file 1 ) which were backcrossed to C57BL/6J mice before intercrossing . We observed similar phenotypes in all mutant alleles and focused on the M6 allele which contains a two base-pair deletion in exon 1 of SLC48A1 ( M6 ) . This deletion causes a frameshift within the thirty-third codon immediately after the first transmembrane domain ( Figure 1B; Figure 1—figure supplement 1A ) . Intercrossing SLC48A1 HET animals produced KO ( knockout ) animals with the expected Mendelian ratio ( Figure 1—figure supplement 1B ) . While SLC48A1 mRNA was still detected ( not shown ) , immunoblots and immunohistochemistry of KO RES tissues showed no detectable SLC48A1 protein , compared to WT ( wildtype ) tissues which express abundant SLC48A1 ( Figure 1C–D; Figure 1—figure supplement 1C–D ) . KO mice had significantly larger spleens and lower hematocrits ( Figure 1E , F ) . Gross morphological examination of six-week old KO mice revealed darkened spleen , bone marrow , and liver ( Figure 1G ) that corresponded with dark intracellular pigments in histochemical tissue sections ( Figure 1D , right panel ) . An enlarged spleen is a common hallmark of ineffective and stress erythropoiesis ( Lenox et al . , 2005; Perry et al . , 2009 ) . We therefore investigated the erythroid compartment of bone marrows and spleens of these mice using Ter-119 and CD44 markers to follow erythroid differentiation and proliferation ( Chen et al . , 2009 ) . Consistent with the reduced hematocrits , KO mice had fewer total Ter-119+ cells in the bone marrow but more basophilic erythroblasts ( population II ) which are symptomatic of an impairment in erythroid maturation ( Figure 2A–C; Figure 2—figure supplement 1A ) . In the spleen , despite similar numbers of total Ter-119+ cells ( Figure 2D , E ) , KO animals showed significantly more immature erythroid precursor cells ( population II+III ) , suggesting compensatory erythropoiesis in the spleen ( Figure 2D–F; Figure 2—figure supplement 1B ) . Since SLC48A1 is primarily expressed in RES macrophages ( White et al . , 2013 ) , we analyzed red pulp macrophages ( RPMs ) , which are the primary iron-recycling macrophages in the spleen ( Beaumont and Delaby , 2009; Ganz and Nemeth , 2012; Haldar et al . , 2014 ) . Significantly fewer mature RPMs ( F4/80hiTreml4+ ) were detected in KO spleens ( Figure 2G , H; Figure 2—figure supplement 1C ) which correlated with increased numbers of immature RPMs by ratiometric quantification of monocytes ( F4/80int: F4/80lo-CD11bhi ) ( Figure 2I , J; Figure 2—figure supplement 1D ) . KO mice on a standard diet ( 380 ppm Fe ) have normal serum iron , total iron-binding capacity ( TIBC ) and transferrin saturation but significantly elevated serum ferritin , an indicator of tissue iron-overload ( Ganz and Nemeth , 2012 ) ( Table 2 in Supplementary file 1 ) . Histological analysis by H and E staining showed dark pigmented inclusions accumulating within RES organs of KO mice ( Figure 3A , right panel ) . However , in situ Perls’ Prussian blue staining did not show significant differences in iron deposition in tissue biopsies ( Figure 3B ) . Because it is possible that the dark pigments masked the visualization of the Prussian blue iron complex , we performed inductively coupled plasma mass spectrometry ( ICP-MS ) to measure total metal content . Significantly more iron was found in the spleens , livers and bone marrows of KO mice ( Figure 3C ) , as compared to copper , zinc , and manganese ( Figure 3—figure supplement 1A–D ) ; a modest 1 . 4-fold change in manganese was observed in the spleens of KO mice in contrast to 2-fold increase in iron ( Figure 3—figure supplement 1D ) . Differences in total metal content were also detected in other organs analyzed ( Figure 3—figure supplement 1A–D ) . These results show that KO mice accumulate tissue iron that is not detectable by Prussian blue staining . Through a combination of ultra-performance liquid chromatography ( UPLC ) ( Sinclair et al . , 2001 ) and ICP-MS to measure heme-iron and total iron , we observed that KO spleens and bone marrows had significantly more total heme ( 10-fold and 3-fold , respectively ) than WT organs ( Figure 3D , Figure 3—figure supplement 1E ) , which indicates that the higher iron content observed in these tissues is due to the accumulation of heme . We next sought to determine the identity of the cells containing the dark pigments . Immunohistochemistry using a macrophage-specific antibody showed dark inclusions in F4/80+ cells in all three RES tissues ( Figure 3E , right panel arrows , Figure 3—figure supplement 1F–H ) . To spatially resolve the metal content at the subcellular level , we employed synchrotron-based X-ray fluorescence microscopy ( XFM ) which provides quantitative information about elemental distribution at the nanoscale . F4/80+ macrophages were isolated from the bone marrow by magnetic bead separation , seeded on an appropriate sample support and visualized using XFM ( Chen et al . , 2015; Vogt , 2003 ) . Overall , KO macrophages contained significantly higher concentrations of cellular iron ( Figure 3F , G ) , copper , sulfur , and calcium ( Figure 3G ) . However , region-of-interest ( ROI ) quantification around the dark pigments in KO macrophages confirmed that iron was the most abundant trace element within this region ( Figure 3H ) . Total body iron is a composite of recycled heme-iron from damaged RBCs and dietary iron absorption . Since heme-iron accumulates within KO macrophages which facilitate iron recycling during RBC turnover , we hypothesized that KO mice might be more susceptible to a low-iron diet . The standard rodent diet contains ≈380 ppm of iron , which is significantly greater than the amount of iron ( 25 to 100 ppm ) required to sustain normal erythropoiesis and the growth of laboratory mice ( Sorbie and Valberg , 1974 ) . We fed weanling WT and KO mice a low-iron diet ( 1 . 8 to 2 ppm Fe ) and measured the impact of limiting dietary iron on their hematological indices . A significant percentage of KO mice , but not WT mice , died within 20 weeks on the low-iron diet ( Figure 4A ) . Importantly , for the first five weeks , the hematocrits of WT and KO mice on the low-iron diet are comparable ( Mazzaccara et al . , 2008 ) ( Figure 4B ) , but by 8 weeks the hematocrits of KO mice were significantly lower ( p<0 . 01 ) , and by 20 weeks they were severely anemic ( 13 . 2 ± 2 . 3 KO versus 39 . 8 ± 2 . 2 WT , p<0 . 0001; Figure 4C ) . These results imply that the defect in heme-iron recycling in KO mice becomes more pronounced after five weeks on an iron-deficient diet . We therefore focused our subsequent analyses on mice that were on low-iron diets for five weeks to detect any molecular and physiological changes in the absence of overt hematological differences . Analyses of serum from iron-deficient animals showed no significant differences in serum iron , TIBC , or transferrin saturation , but serum ferritin levels remained elevated in KO mice on a low-iron diet ( Table 2 in Supplementary file 1 , 2 ppm diet ) . The levels of iron and heme retained within RES organs of KO mice were also significantly higher than WT mice ( Figure 4D , E , Figure 4—figure supplement 1A ) , suggesting that the heme-iron stores in KO mice were not bioavailable despite systemic iron-deficiency . In response to iron deficiency , the spleens of WT mice increased by approximately 120% ( Figure 4F ) ( Lenox et al . , 2005; Perry et al . , 2009; Socolovsky , 2007 ) , with a concomitant increase ( 7 . 5-fold ) in immature Ter119+ cells , a hallmark of stress erythropoiesis ( Figure 4G , H ) . In contrast , the stress erythropoiesis response in iron-deficient KO mice was significantly attenuated ( Figure 4F–H , Figure 4—figure supplement 1B ) , with erythroblast differentiation blocked at an earlier stage ( population I ) in the bone marrow ( Figure 4I , Figure 4—figure supplement 1B–D ) . Iron deficiency fully restored the total numbers of mature RPM population and balance in monocyte populations ( Figure 4J , K ) in KO mice to WT levels ( Figure 4—figure supplement 1E , F ) . This could reflect increased viability of erythrophagocytic RPMs due to reduced heme accumulation in these cells under systemic iron deficiency . To interrogate the responsiveness of iron and heme metabolism genes to iron deficiency in these mice , we assembled a custom qRT-PCR array comprising probes for 90 key iron/heme metabolism genes . Iron-deficient KO mice showed significant dysregulation in 38% and 6% of these mRNAs in the spleen and liver , respectively ( Figure 4L , Figure 4—figure supplement 1G–I , and Table 1 in Supplementary file 1 ) . Under iron-deficient conditions , the levels of 32 iron metabolism mRNAs were significantly reduced in WT spleens ( Figure 4—figure supplement 1H , compare WT Standard vs WT 2 ppm , KO Standard vs KO 2 ppm ) , but remained high in KO spleens . Together , these data suggest that KO mice are unable to respond normally to iron-deficiency , especially at sites where heme-iron accumulates at high levels such as the spleen . To investigate the in vivo function of SLC48A1 in heme-iron recycling , we labeled normal RBCs with 59Fe and followed the labeled iron after injection into WT and KO mice . WT donor mice were treated with phenylhydrazine to induce acute hemolytic anemia and stimulate erythropoiesis , followed by an injection of 59Fe-citrate which is incorporated into heme of newly-formed RBCs . Three days later , ≈80% of 59Fe label was incorporated into heme and the 59Fe-labeled RBCs were opsonized and injected into WT and KO recipient mice ( Franken et al . , 2015; Soe-Lin et al . , 2009 ) that had been maintained on a low-iron diet for six weeks ( Figure 5A ) . The retention and distribution of 59Fe in the tissues of the recipient mice were monitored over a period of 96 hr . The spleens of WT mice accumulated 59Fe within 24 hr followed by a decrease at 96 hr with a concomitant increase of 59Fe in the circulation and kidneys ( Figure 5B ) . By contrast , the spleens of KO mice accumulated more 59Fe over the course of 96 hr with limited amounts of 59Fe in the circulation and kidneys , suggesting that the release of 59Fe from phagocytosed RBCs was impaired ( Figure 5B ) . To discriminate 59Fe from [59Fe]heme , homogenized spleens were subjected to high-speed centrifugation to precipitate proteinase-resistant and detergent-insoluble fractions . The resulting supernatant was differentially extracted by ethyl-acetate to separate heme from iron . While KO spleens accumulated significantly greater levels of 59Fe compared to WT spleens , these differences were primarily due to 59Fe counts within the organic heme fraction ( [59Fe]heme ) and not in the aqueous iron fraction ( 59Fe ) ( Figure 5B , C ) . A large portion of 59Fe was detected in the detergent-insoluble fraction of KO spleens ( Figure 5C ) and livers ( Figure 5—figure supplement 1A ) . A proteinase and detergent resistant fraction enriched in iron is characteristic of hemozoin , a crystallized form of heme generated during the digestion of hemoglobin by blood-feeding organisms such as Plasmodium ( Egan , 2008; Francis et al . , 1997 ) . To determine whether the insoluble fractions from KO tissues had hemozoin-like properties , we treated the fractions with buffers known to dissolve hemozoin ( Figure 5D ) . Spectroscopic analysis of the dissolved detergent-resistant fraction revealed absorption spectra characteristic of heme ( Figure 5E , Figure 5—figure supplement 1B ) . High resolution X-ray powder diffraction of the purified insoluble fraction from the KO spleens confirmed that the dark pigment is identical to malarial hemozoin ( Figure 5F ) ( Coronado et al . , 2014; Slater et al . , 1991 ) . The structure reveals a repeating unit of heme dimers linked together by iron-carboxylate bonds to one propionic acid side chain with the adjacent heme dimers forming stable chains by hydrogen bonding via the free propionate side chain ( Figure 5G , Figure 5—figure supplement 1C ) ( Coronado et al . , 2014 ) . Spectrophotometric analysis of the dissolved hemozoin fractions showed substantial amounts of hemozoin-heme present in the spleens and livers of KO mice , but not in WT mice ( Figure 5H , Figure 5—figure supplement 1D ) . Notably , a significant amount of heme in non-hemozoin form was also found in KO spleens ( 2-fold of WT; Figure 5H ) . Scanning electron microscopy images showed mammalian hemozoin was heterogeneous; purified hemozoin crystals from spleen and liver appeared larger than Plasmodium falciparum hemozoin , while bone marrow-purified hemozoin was similar in size ( Figure 5I ) . Transmission electron microscopy with purified F4/80+ tissue-resident macrophages from the spleen and bone marrow showed that hemozoin was sequestered within a membrane-enclosed compartment ( Figure 5J ) . However , isolated KO monocytes that were differentiated into macrophages in vitro did not contain any hemozoin ( Figure 5—figure supplement 1E ) , indicating that hemozoin formation occurs in vivo after tissue resident macrophages are established . Confocal microscopy using antibodies against LAMP1 , which has been previously demonstrated to be a strong marker of erythrophagosomes in macrophages ( Delaby et al . , 2012; Huynh et al . , 2007 ) , showed that all of the hemozoin in isolated-F4/80+ macrophages from KO bone marrow is enclosed within LAMP1-positive compartments ( Figure 5K , Figure 5—figure supplement 1F ) . Furthermore , these hemozoin-containing LAMP1-positive vesicles were significantly larger ( diameter 570 to 4300 nm ) compared to typical lysosomes which range from 50 nm to 500 nm ( Bandyopadhyay et al . , 2014 ) ( Figure 5K , compare vesicles with white vs yellow arrows; Figure 5—figure supplement 1G ) . Intracellular hemozoin is typically known to be inert and ingested malarial hemozoin can remain unmodified for long periods of time within human monocytes ( Schwarzer et al . , 1993 ) . To test whether mouse hemozoin could be used as a source of iron when macrophages are forced to undergo apoptosis , we administered clodronate liposomes to iron-deficient WT and KO mice . RES macrophages were completely depleted by day three post-injection but KO spleens , livers and bone marrows still retained hemozoin , which were now extruded into the interstitial space ( not shown ) . However , this hemozoin was not bioavailable as an iron source to correct iron deficiency anemia even after seven days ( Figure 5—figure supplement 1H ) . Altogether , these results strongly indicate that in the absence of SLC48A1 , RES macrophages are unable to recycle and degrade heme from ingested RBC resulting in heme accumulation and biocrystallization into hemozoin within enlarged phagolysosomes . To evaluate the impact of SLC48A1 deficiency at the cellular level , we analyzed the ability of bone marrow-derived macrophages ( BMDMs ) to recycle heme-iron derived from phagocytosed RBCs . Although WT and KO BMDMs engulfed similar numbers of opsonized RBCs ( Figure 6—figure supplement 1A ) , SLC48A1-deficient cells accumulated significantly more heme even after 72 hr post-EP ( Figure 6A ) . As heme is cytotoxic and damaging to macrophages ( Kovtunovych et al . , 2010 ) , we measured cellular markers of oxidative stress during the two critical phases of heme-iron recycling after EP - the early ( 4 hr ) and late ( 24 hr ) phase . Reactive oxygen species ( ROS ) production ( Figure 6B ) ( Delaby et al . , 2005; Kovtunovych et al . , 2010 ) and the ratio of oxidized to reduced glutathione levels ( GSSG:GSH ) ( Figure 6C ) was significantly lower in KO cells during the early phase , when active digestion of red cells takes place . Consistent with this early suppression of ROS , lactate dehydrogenase ( LDH ) release into the growth medium ( Rayamajhi et al . , 2013 ) was also significantly reduced in KO cells in the late phase ( Figure 6D ) , with no loss in cell adherence ( Figure 6—figure supplement 1B ) ( Delaby et al . , 2005 ) . Together , these results imply that SLC48A1 deficiency protects the macrophages from the damaging effects of EP ( Delaby et al . , 2005 ) . The in vivo and cell biological studies posit that loss of SLC48A1 confers heme tolerance by confining heme to the phagolysosome and preventing its degradation by HMOX1 . However , EP studies reveal that HMOX1 induction and abundance were comparable between WT and KO BMDMs over 72 hr post-EP ( Figure 6E ) These results raised the possibility that , in the absence of SLC48A1 , an alternate albeit less-efficient heme transport pathway could translocate heme from the phagolyosome for degradation by HMOX1 . To evaluate the genetic contribution of HMOX1 , we generated HMOX1 HET and DKO ( HMOX1 KO; SLC48A1 KO double knockout ) double mutant mice . Analysis of offspring from HMOX1 HET and HMOX1 HET; SLC48A1 KO intercrosses showed reduced numbers of viable HMOX1 KO progeny ( Observed/Expected: HMOX1 KO 9/64 versus DKO 9/86 ) regardless of the presence of SLC48A1 ( Figure 6F , Figure 6—figure supplement 1C ) . Consistent with this observation , HMOX1 KO and DKO mice showed similar reductions in RES macrophages and erythropoietic profiles from the bone marrow and spleen at ten weeks of age ( Figure 6—figure supplement 1D–I ) , and developed anemia as they grew older ( not shown ) . Unexpectedly , heterozygous offspring from HMOX1 HET; SLC48A1 KO intercrosses did not show Mendelian ratio at birth , with almost 40% embryonic lethality ( Observed/Expected: HMOX1 HET 133/128 versus HMOX1 HET; SLC48A1 KO 108/172 ) ( p<0 . 001 by chi-square test ) ( Figure 6F , Figure 6—figure supplement 1C ) . Flow cytometry analyses of HMOX1 HET; SLC48A1 KO mice showed an enhanced reduction in splenic RPMs ( Figure 6G ) and total Ter-119+ cells in the bone marrow ( Figure 6H ) compared to HMOX1 HET mice . The bone marrow subpopulations of Ter-119+ cells were also significantly reduced in populations II to V in HMOX1 HET; SLC48A1 KO mice ( Figure 6I ) . The spleens of HMOX1 HET; SLC48A1 KO mice had fewer Ter-119+ cells with more immature Ter-119+ cells ( Figure 6J , K ) , suggesting ineffective stress erythropoiesis . Together these results reveal that inhibition of HMOX1 in the absence of SLC48A1 has two different outcomes depending on the degree of HMOX1 impairment . Partial reduction or haploinsufficiency of HMOX1 , as observed in HMOX1 HET; SLC48A1 KO mice , increases perinatal lethality and a shift to synthetic lethality . However , complete inhibition of HMOX1 leads to loss in cell viability ( Figure 6—figure supplement 1J , K ) and greater embryonic mortality of DKO and HMOX1 KO mice ( Figure 6—figure supplement 1C ) , demonstrating a delicate balance between heme transport and degradation .
Our results show that ( SLC48A1 ) KO mice are defective in transporting heme across the phagosomal membrane during EP , resulting in large amounts of heme accumulation and hemozoin formation within phagolysosomes of RES macrophages . Heme retention leads to impaired iron recycling for erythrocyte production in the bone marrow , stimulating extramedullary erythropoiesis similar to the effects caused by iron-deficiency ( Figure 6L ) . However , the overall impact on RBC production and anemia is modest on a standard iron diet ( 380 ppm ) . These findings support the conclusion that mice on an iron-rich diet can circumvent a block in heme-iron recycling by relying more heavily on dietary iron for erythropoiesis ( Ganz and Nemeth , 2006 ) . This is in stark contrast to humans where the daily iron absorption does not exceed 10% of the amount of recycled iron utilized for erythropoiesis ( Kautz and Nemeth , 2014; Muñoz et al . , 2009; Winter et al . , 2014 ) . However , when KO mice are fed a low-iron diet , the block in heme-iron recycling is exacerbated . Furthermore , when KO mice encounter prolonged dietary iron limitation ( 8 to 20 weeks ) , they are unable to sustain erythropoiesis , implying that it takes approximately eight weeks for body iron stores to be depleted in the absence of heme-iron recycling . Indeed , results from the in vivo 59Fe-labeled RBC studies show that the relative proportion of 59Fe retained within the spleens of KO mice are 50 . 3% hemozoin: 34 . 4% heme: 15 . 3% non-heme iron after 96 hr ( Figure 5C ) . Assuming that ineffective heme-iron recycling reduces recycled iron according to the law of exponential decay ( Leike , 2002 ) and RBCs have a lifespan of ~40 days , this 59Fe distribution pattern would predict that KO mice would become significantly anemic after six weeks as hemozoin accumulates after RBC turnover . Another potential contributing factor is the mouse genetic background . The KO mice were developed on a mixed genetic background ( 75% C57BL/6J , 25% 129/SvJ ) . Published studies comparing the iron status of C57BL/6J and 129/SvJ mice on an iron-balanced diet have shown that 129/SvJ mice have more hepatic iron stores ( Dupic et al . , 2002; Wang et al . , 2007 ) . We recently generated KO mice on a pure C57BL/6J background and placed these mice on an iron-deficient diet post-weaning ( P21 ) . KO C57BL/6J mice showed severe anemia within 5 . 5 weeks compared to WT littermate controls , which approximates one lifespan of circulating erythrocytes ( not shown ) . This result indicates that genetic strain differences that influence body iron stores are important modifiers of the SLC48A1 mutant phenotype . Taken together , we conclude that KO mice reach a ‘metabolic threshold’ on a low-iron diet , that is a physiological transition from an intermediate iron state to severe iron limitation , which eventually lead to mortality in these animals ( Figure 6L , right panel ) . Even though WT and KO mice had similar hematocrits at five weeks on a low-iron diet , KO mice were unable to undergo stress erythropoiesis and showed dysregulation of over 35 iron and heme metabolism genes in their spleens and livers . Typically under iron-deficient conditions , the production of the liver hormone hepcidin is suppressed , which results in greater mobilization of iron by SLC40A1 to iron-deprived compartments ( Ganz and Nemeth , 2006; Nemeth et al . , 2004 ) . The inability to sustain erythropoiesis , despite a significant suppression in hepatic hepcidin expression in KO iron-deficient mice ( Figure 4—figure supplement 1G ) , suggest that these mice are iron-deprived even though the RES tissues are heme-iron loaded , that is systemic iron status is uncoupled from tissue and cellular iron status in the absence of SLC48A1 . While hepatic hepcidin is considered to be the major facilitator of systemic iron homeostasis , recent studies have uncovered an autocrine role for locally-produced hepcidin in regulating cardiac iron homeostasis ( Lakhal-Littleton et al . , 2016 ) . Indeed , KO mice show elevated splenic hepcidin regardless of iron status ( Figure 4—figure supplement 1H ) , implying a role for splenic hepcidin in stress erythropoiesis . In the absence of SLC48A1 , heme accumulates in the erythrophagosome of RES macrophages in the form of hemozoin biocrystals . Heme-iron recycling of senescent erythrocytes occurs in the RES macrophages - mainly in the splenic RPMs and , to some extent , in liver Kupffer cells and bone marrow macrophages ( Beaumont and Delaby , 2009; Ganz , 2012; Theurl et al . , 2016 ) . Our results establish a direct role for SLC48A1 in transporting heme from the phagolysosomal compartments to the cytosol for degradation by HMOX1 . Unlike ( SLC48A1 ) KO mice , loss of HMOX1 causes over 90% embryonic lethality . Why do ( SLC48A1 ) KO mice survive while HMOX1 KO mice show high embryonic mortality ? Our results show that in the absence of SLC48A1 , high concentrations of heme is sequestered as hemozoin within the acidic phagolysosome , resulting in cellular heme tolerance . By contrast , in the absence of HMOX1 , heme accumulates within the cytosol resulting in cytotoxicity and embryonic lethality . Genetic epistasis would have predicted that the heme sequestration would rescue the embryonic lethality of the DKO mice . Possible explanations for this unexpected finding could be that HMOX1 has additional functions beyond heme degradation , or that the heme degradation products are required for cell differentiation and specification in vivo . Our results support both of these explanations and further reveal that KO mice require a fully-operational heme degradation pathway to confer complete heme tolerance , as partial reduction of HMOX1 in the absence of SLC48A1 show significant embryonic lethality with severe impairment in macrophage and erythroblast maturation in the bone marrow and spleen ( Figure 6F–K ) raising the possibility that functional HMOX1 requires SLC48A1 on the erythrophagosomal membranes . In the absence of SLC48A1 , bioactive HMOX1 drops below the 50% expected in HMOX1 HET animals . Blood-feeding parasites degrade hemoglobin in their digestive vacuole and accumulate heme in the form of hemozoin ( Chen et al . , 2001; Toh et al . , 2010 ) . Although the exact mechanisms of in vivo hemozoin formation are not well understood , it is generally agreed upon that hemozoin formation is a heme detoxification method used to protect the parasite from heme toxicity ( Hempelmann , 2007; Toh et al . , 2010 ) . Hemozoin biocrystals are thought to be well-tolerated by both the parasite and subsequent host cells which ingest it , despite some reports on the inflammatory responses towards hemozoin ingestion ( Basilico et al . , 2003; Schwarzer et al . , 1993 ) . We found hemozoin within LAMP1-positive intracellular vesicles , which are typically between pH 4 and 6 ( Johnson et al . , 2016; Mellman et al . , 1986 ) . Acidic pH appears to be a common requirement for hemozoin formation in parasitic digestive vacuoles , in vitro and in macrophages ( Egan et al . , 2001; Shio et al . , 2010 ) . To our knowledge , hemozoin formation in mammals has not been documented prior to this study . This phenomenon observed in KO spleens , livers and bone marrows suggests that macrophages are able to elicit a protective mechanism against high concentrations of heme by exploiting the low pH environment of lysosomes . It is important to note that KO mice have fewer RPMs , which implies that hemozoin sequestration may be incomplete or insufficient to confer complete heme tolerance , or that hemozoin itself could cause detrimental effects to these cells . Hemozoin in KO RES macrophages leads to tissue heme accumulation in these mice without causing major damages to tissue architecture . We speculate that pharmacologic inhibition of SLC48A1 in humans would lead to tolerance and protection from heme toxicity and iron overload .
All mice used were housed in a 12 hr light-dark cycle . For SLC48A1 pups , genetic segregation was computed on 21 day old ( P21 ) mice pups . SLC48A1 mice were genotyped from tail genomic DNA extracts using a custom ordered TaqMan SNP Genotyping Assays probe ( ThermoFisher Scientific ) on a Bio-rad CFX Connect system . For HMOX1/SLC48A1 mice , genetic segregation was computed on 5 day old ( P6 ) mice pups via toe-clip DNA extracts . HMOX1 was genotyped by PCR using primers HMOX1 KO Forward 5’-GCTTGGGTGGAGAGGCTATTC-3’ , HMOX1 KO Reverse 5’-CAAGGTGAGATGACAGGAGATC-3’ , HMOX1 WT Forward 5’-GTACACTGACTGTGGGTGGGGGAG-3’ , HMOX1 WT Reverse 5’-AGGGCCGAGTAGATATGGTAC-3’ . Mice in all studies were males unless otherwise noted , although initial experiments to exclude gender variation were done using both males and females . All animal protocols were approved by the Institutional Animal Care and Use Committee at the University of Maryland , College Park ( IACUC Animal Study Protocol R-NOV-18–61 ) . Guide and Cas9 RNAs: The guide RNA ( 5'-TAGGGACGGTGGTCTACCGACAACCGG-3' ) was purchased from Sage Laboratories 2033 Westport Center Drive , St Louis , MO . Cas 9 RNA was purchased from Trilink Biotechnologies , San Diego , CA . The guide RNA and Cas9 RNA were combined at a concentration of 5 ng/µl ( each ) in 10 mM Tris , 0 . 25 mM EDTA ( pH 7 . 5 ) for injection . Pronuclear Injection: Pronuclear injection was performed using standard procedures ( Behringer et al . , 2014 ) . Briefly , fertilized eggs were collected from superovulated C57BL/6J × 129/SvJ F1 females approximately 9 hr after mating with C57BL/6J male mice . Pronuclei were injected with a capillary needle with a 1–2 µm opening pulled with a Sutter P-1000 micropipette puller . The RNAs were injected using a FemtoJet 4i ( Eppendorf ) with continuous flow estimated to deposit approximately 2 pl of solution . Injected eggs were surgically transferred to pseudo-pregnant CB6 F1 recipient females . Genotyping: DNA was obtained from founder ( F0 ) animals by tail biopsy , amplified by PCR ( Forward 5’-TGCACCTGTGACTCGGCG-3’ Reverse 5’-TAGGTCCCGCCACGTTCATAA-3’ ) and sequenced to determine the mutation . F0 animals carrying mutations were crossed to C57BL/6J animals and the resulting heterozygous F1 animals were either intercrossed ( F ) to generate homozygous mutant animals or back crossed ( N ) to C57BL/6J mice for propagation . All mice used were N1F2 generation derived from intercrosses between heterozygotes , and speed congenics ( DartMouse , NH ) determined the genetic background to be about 75% C57BL/6J and 25% 129/SvJ . WT mice denote wildtype littermates of KO mice . WT and KO mice obtained from SLC48A1 HET crosses were weaned at 21 days of age ( P21 ) and placed on their respective diets , supplemented with deionized water . Standard rodent diet was obtained from Envigo and the two ppm ( TD . 09127 ) Fe diet was custom ordered from Envigo , Madison , WI . Body weights and feed intake per cage were measured weekly to ensure the conditions of the animals . Blood was collected by retro-orbital bleeding using microcapillary tubes ( Fisher Scientific , cat . number 22–362566 ) . At the end of five weeks , mice were sacrificed by cardiac perfusion using Dulbecco's phosphate-buffered saline ( DPBS ) ( Gibco , cat . number 14190250 ) under anesthesia ( 10% ketamine , 8% xylazine mix ) . Prior to perfusion , whole blood was collected into tubes and allowed to clot at room temperature for 45 min and serum was separated from the sample by centrifugation at 2000 g for 10 min . For the prolonged dietary iron study where the Kaplan Meier survival curve was obtained , whole blood hematocrits were obtained every two weeks until week 14 . All surviving KO mice were sacrificed when the survival rate fell below 50% ( ~week 16 ) , while WT mice were kept on the diets until week 18 . Paraffin-embedded tissue sections were processed for antigen retrieval by heat-induced epitope retrieval in citrate buffer pH 6 ( DAKO , Glostrup , Denmark , cat . number S2367 ) . After epitope retrieval , sections were then incubated with either rat anti-F4/80 ( Invitrogen , cat . number MF48000 ) ( 1:1000 in blocking buffer TBS with 2% FBS ) or rabbit anti-SLC48A1 ( Zhang et al . , 2018 ) ( 1:500 ) overnight at 4°C . Polyclonal SLC48A1 antibody serum was generated in rabbit using the C-terminal 17 amino acid peptide sequence ( YAHRYRADFADIILSDF ) of human SLC48A1 as antigen ( Epitomics , Inc ) . Sections were then incubated with secondary biotinylated anti-rat antibody ( Vector labs , cat . number BA-9400 ) for 30 min at room temperature . Signals were detected by DAB substrate or the alkaline phosphatase red substrate ( Vector labs , cat . number: SK-5100 ) incubation and slides were lightly counterstained with hematoxylin . H and E and Perl’s Prussian blue stainings were conducted by Histoserve , Inc . Tissues were snap frozen in liquid nitrogen and ground using a ceramic pestle and mortar maintained ice-cold . In a dounce homogenizer , powdered tissue samples were added to membrane prep buffer ( 250 mM Sucrose , 1 mM EDTA , 10 mM Tris-HCl pH 7 . 4 , 3X protease inhibitor cocktail ) . Samples were dounce homogenized until no obvious chunks of tissue were observed and the number of strokes used for homogenization was kept consistent across all samples . Homogenates were centrifuged at 800 g for 10 min at 4°C . The supernatant was transferred to ultracentrifugation tubes and spun at 100 , 000 g for 2 hr at 4°C . The pellet from ultracentrifugation was then resuspended in lysis buffer ( 150 mM NaCl , 1 mM EDTA , 20 mM HEPES pH 7 . 4 , 2% Triton-X , 3X protease inhibitor ) and sonicated to ensure complete lysis . The sample was then centrifuged at 11 , 000 g for 30 min at 4°C and the insoluble pelleted debris was discarded . Protein concentration of the supernatant was determined using the BCA assay ( Pierce BCA Protein Assay Kit , ThermoFisher Scientific , cat . number 23225 ) . Samples were mixed with SDS-loading buffer without heating and electrophoretically separated on a 4–20% Criterion TGX Precast Midi Protein Gel ( Bio-rad , cat . number 5671094 ) and transferred to nitrocellulose membrane . Proteins were cross-linked to membranes by UV treatment and stained by ponceau S before incubation in blocking buffer ( 5% nonfat dry milk in 0 . 05% Tris-buffered saline-Tween 20 , TBS-T ) for 1 hr at room temperature . Blots were then incubated overnight at 4°C in blocking buffer containing rabbit anti-SLC48A1 antibody ( 1:300 dilution ) . After three washes in TBS-T , blots were incubated 1 hr with horseradish peroxidase ( HRP ) -conjugated goat anti-rabbit IgG secondary antibody ( 1:20000; Invitrogen cat . Number 31460 ) in blocking buffer . Blots were then washed five times with TBS-T and signals were visualized by using enhanced chemiluminescence ( SuperSignal West Pico , Pierce ) and detected using ChemiDoc Imaging Systems ( Bio-rad ) . Total RNA was isolated from samples using TRIzol Reagent ( ThermoFisher Scientific ) . cDNA synthesis was done using RT2 First Strand Kit ( Qiagen ) . The Qiagen iron metabolism RT2 profiler array was custom built ( Cat . number CLAM25204D ) and used with RT2 SYBR Green Fluor qPCR mastermix ( Qiagen ) on a CFX Connect system ( Bio-rad ) . Analysis of gene expression data was conducted using the online data analysis web portal provided by Qiagen . Briefly , Ct values for each gene in each group ( eg . WT , Standard diet ) were obtained by taking the average across all mice ( n = 9 per group ) . ΔCt values were then obtained by the following formula: ΔCt = Ct ( target gene ) – Ct ( housekeeping gene ) . The housekeeping gene used was RPL13a . Gene expression was then calculated by the formula 2^ ( - ΔCt ) . p values for gene expression were calculated using ΔCt values , the corresponding standard deviations and n = 9 . The gene expression heatmap was generated by the Heatmapper software and clustering was performed using Pearson’s distance measurement method with average clustering ( Babicki et al . , 2016 ) . Hemozoin extraction was performed according to the method of Deroost et al . ( 2012 ) . Approximately 50 to 100 mg perfused mouse liver or spleen were ground in liquid nitrogen with mortar and pestle . The ground powder was resuspended in five to ten volumes of homogenization buffer ( 50 mM Tris/HCl pH 8 . 0 , 5 mM CaCl2 , 50 mM NaCl , 1% Triton X-100% and 1% Proteinase K ) and incubated overnight at 37°C with gentle shaking . For mice younger than three weeks , a whole liver or spleen was homogenized in minimum five volumes of homogenization buffer using FastPrep-24 ( MP Bio ) for 30 s at the 6 . 5 m/s setting , followed by overnight incubation at 37°C with gentle shaking . The proteinase K digested homogenate was then sonicated ( Heat Systems-Ultrasonics , W350 ) for 1 min ( 20 W , pulse 1 s ) and centrifuged at 11 , 000 g for 45 min . The supernatant was collected in a new tube ( Supernatant fraction ) and the pellet was washed three times in washing buffer ( 100 mM NaHCO3 , pH 9 . 0% and 2% SDS ) with 1 min sonication and 30 min centrifugation at 11 , 000 g . All the supernatant from three wash steps was collected and combined ( Washing fraction ) . The remaining pellet ( Pellet fraction , Hz ) was dissolved and sonicated for 1 min in dissolving buffer ( 100 mM NaOH , 2% SDS and 3 mM EDTA ) and centrifuged at 11 , 000 g for 30 min to discard any insolubles . For X-ray powder diffraction analysis , after the third wash , extracted Hz was washed five more times in distilled H2O to remove the salts and detergents . Heme concentrations of all three fractions ( Supernatant , Washing and Pellet ) were determined by the pyridine hemochromogen spectra method ( Barr and Guo , 2015 ) . A pyridine reagent mix was prepared by adding 3 mL 1 M NaOH and 6 mL pyridine to 19 mL H2O in a glass container . For oxidized spectrum , 35 μL sample and 17 μL 15 mM K3Fe ( CN ) 6 was mixed with 1 mL pyridine reagent in a cuvette and the spectrum at 400–675 nm was recorded ( Shimadzu , UV-1601 ) . Two to five mg of powdered Na2S2O4 were added to the mixture and the reduced spectrum was recorded at 400–675 nm . Heme concentrations were calculated by subtracting the absorbance readings at 541 , 556 and 575 nm in the oxidized spectrum from the corresponding readings in the reduced spectrum to get ∆A540 , ∆A556 and ∆A575 , using the extinction coefficients 20 . 7/mM for ∆A540 ( ( ∆A556 - ∆A540 ) /20 . 7 ) and 32 . 4/mM for ∆A575 ( ( ∆A556-∆A575 ) /32 . 4 ) , multiplying by the dilution factor of the sample ( 30 . 06 or ( 1000 + 35 + 17 ) /35 ) and averaging the two results . Total heme ( nmol/mg tissue ) in each fraction was calculated by multiplying heme concentration with corresponding fraction volume , and then divided by the weight of homogenized tissue . Total heme ( nmol/organ ) in each fraction was calculated by multiplying total heme ( nmol/mg tissue ) with the total weight of corresponding organ . Total heme in the Supernatant and Washing fractions was summed up as non-Hz heme . Prior to metal and heme analyses , frozen tissues were added to three volumes of pure water and homogenized in ceramic bead tubes ( Qiagen ) using an Omni Bead Ruptor 24 . For metal analysis , homogenate aliquots were digested overnight in 5:1 HNO3:H2O2 , dried , and resuspended in 2% HNO3 for analysis using an Agilent 7900 ICP-MS . Calibration standard solutions for determination of Fe , Zn , Cu and Mn were prepared from Agilent multi-element calibration standard-2A . Protein concentrations of homogenates were determined by BCA Protein Assay ( Thermo Fisher Scientific ) for normalization . Enough tissue homogenate preparation for ICP-MS was adjusted with water to make 50 µL of 5 mg/mL protein . The resulting suspension was extracted with 200 µL of EA ( a mixture of four volumes of ethyl acetate to one volume glacial acetic acid ) , centrifuged at 13 . 5 K rpm for 0 . 5 min , and the supernatant was removed . The residual was re-extracted with 200 µL of water-saturated EA and centrifuged similarly . The two supernatants were combined to make about 400 µL total volume , and 10 µL of which was injected into the UPLC . 20 µL of 2xNES ( 0 . 2M NaOH , 4% w/v SDS and 6 mM EDTA ) added to the ~20 µL residual . The resulting suspension was slowly mixed with 280 µL EA and then centrifuged at 13 . 5 K rpm for 10 min . 10 µL of the supernatant was injected into the UPLC . In addition , 25 µL of un-extracted homogenate was mixed with 25 µL 2xNES , then with 350 µL EA , centrifuged at 13 . 5 K rpm for 10 min and the supernatant was analyzed by UPLC . About 10 µL of sample extract was injected into a Waters Acquity UPLC system which included a binary solvent manager , sample manager , photodiode array detector ( PDA ) , fluorescence detector ( FLR ) , column heater and an Acquity UPLC BEH C18 , 1 . 7 µM , 2 . 1 × 100 mm column . The PDA was set to measure hemin absorbance at 398 nm and the FLR to measure fluorescence of protoporphyrin IX ( PPIX ) at 404 nm excitation and 630 nm emission . Solvent A was 0 . 2% aqueous formic acid while Solvent B was 0 . 2% formic acid in methanol . The flow rate at 0 . 40 mL per min at 60 ˚C for the total run time of 7 min . The following successive gradient settings for run time in min versus A: 0 . 0 , 80%; 2 . 5 , 1%; 4 . 5 , 1%; 5 , 80% . The solvent composition gradient settings were all linear . For standards , solutions of known concentrations of authentic hemin , PPIX dissolved in NES were extracted and then analyzed by UPLC as the samples . Serum iron , total iron binding capacity ( TIBC ) and transferrin saturation were measured using the Stanbio Iron and TIBC kit ( VWR , cat . number 10152–550 ) . Serum ferritin was quantified using the mouse ferritin ELISA kit ( Abcam , cat . number ab157713 ) . All kits were used according to the manufacturer’s protocol . 59FeCl3 purchased from Perkin Elmer Life Sciences ( cat . number NEZ037001MC ) was mixed with sodium citrate ( 1:50 molar ratio in a total volume of 1 ml ) and incubated for 1 hr at room temperature to make 59Fe-citrate . To generate 59Fe-labeled red blood cells ( RBCs ) , adult donor mice were first injected intraperitoneally once per day for three consecutive days with 50 mg/kg of phenylhydrazine ( Sigma Aldrich , cat . number 114715 ) to induce anemia . Thirty min after the last phenylhydrazine dose , donor mice were injected intraperitoneally with 200 μl of radiolabeled 59Fe-citrate ( 0 . 03 μM , ~12 million cpm ) . Following a 3 day rest , donor mice were anesthesized ( 10% Ketamine with 5% Xylazine ) and whole blood was collected by retro-orbital bleeding using heparinized tubes . Whole blood was mixed with an equal volume of Alsever’s solution ( Sigma Aldrich , cat . number A3551 ) . Mice were then sacrificed by cervical dislocation . 59Fe-RBCs were collected by centrifugation and washed with DPBS before counting . 59Fe-RBCs were opsonized with the mouse red blood cell antibody ( Rockland , cat . number 210–4139 ) ; 20 μl of antibody was used for approximately 109 RBCs . The suspension was diluted to 10 ml with PBS , and incubated at 37°C on a rotating table for 20 min . The opsonized cells were washed twice with DPBS and counted again . The opsonized 59Fe-RBCs were diluted with DPBS and injected intraperitoneally into iron-deficient WT and KO mice ( 250 µl , 860 , 000 cpm per mouse ) . 59Fe-RBCs -injected mice provided with food and water were sacrificed by CO2 asphyxiation at 24 and 96 hr post-injection . Whole blood was collected retro-orbitally for counting prior to sacrifice . Counts for separate tissues and spleen homogenates were collected on a Perkin Elmer ( Packard ) Wizard gamma counter ( Efficiency ~21% ) and whole carcass counts were collected using a Model 2200 scalar ratemeter ( Efficiency ~0 . 3% ) ( LUDLUM measurements , Inc , Sweetwater , TX ) . To count for 59Fe in different extracts in spleens , spleens from the 96 hr time-point were dounce homogenized in lysis buffer ( 1% Triton X-100 , 1% proteinase K , Tris-HCl pH 8 . 0 , NaCl , CaCl2 ) and incubated overnight at 37°C . One-third of the homogenate was centrifuged at 11 , 000 g for 45 min to obtain the insoluble fraction . Ethyl acetate ( EA ) ( one part glacial acetic acid , four parts ethyl acetate ) was added to the supernatant ( one part supernatant , four parts EA ) . The suspension was vortexed for 1 min and centrifuged at 800 g for 2 min to separate the organic and aqueous phase . This extraction was repeated once to the aqueous layer and EA extracts were pooled for counting . A mouse spleen was cut up into 1–3 mm pieces and placed in 5 ml of dissociation buffer ( RPMI , 1X collagenase B , 1X DNase I ) in tubes with a magnetic stir bar . Tubes were placed on a magnetic plate and spleens were dissociated at 37°C for 45 min . Homogenates were passed through 70 μm filters and cells were pelleted by centrifugation at 800 g for 10 min . Bone marrow cells were flushed from the femur and tibia of mice using a syringe and 18G needle with 10 ml of FACS buffer ( DPBS with 2% FBS ) . Cell aggregates were dissociated by pipetting . Cell suspensions were centrifuged at 800 g for 5 min and supernatants were discarded . When red blood cells ( RBCs ) were not needed for analysis , cells were resuspended in 1 ml of RBC lysis buffer ( 150 mM NH4Cl , 10 mM NaHCO3 , 1 . 3 mM EDTA ) and left at room temperature for 3 min . The lysis step was quenched by adding 5 ml of RPMI and cells were collected by centrifugation at 800 g for 10 min and resuspended in appropriate buffers for downstream application . From splenic and bone marrow cell suspensions , anti-F4/80 microbeads were used in conjunction with LS columns ( Miltenyi Biotech , cat . number 130-110-443 and cat . number 130-042-401 ) according to the manufacturer’s protocol . To isolate monocytes from bone marrow suspension , the mouse monocyte isolation kit was used ( Miltenyi Biotech , cat . number 130-100-629 ) . Monocytes and macrophages were cultured as described previously ( White et al . , 2013 ) . Whole blood was collected by retro-orbital bleeding using heparinized tubes from mice and mixed with an equal volume of Alsever’s solution ( Sigma Aldrich , cat . number A3551 ) . RBCs were opsonized with the mouse red blood cell antibody ( Rockland , cat . number 210–4139 ) ; 20 μl of antibody was used for approximately 109 RBCs . For in vitro EP experiments , RBCs were added at a ratio of 1:10 ( macrophage:RBCs ) into cell culture medium and applied to BMDMs . One h later , medium was aspirated and cells were washed with DPBS and RBC lysis buffer ( 150 mM NH4Cl , 10 mM NaHCO3 , 1 . 3 mM EDTA ) before fresh medium was replaced . BMDMs were seeded in 6-well plates at a density of 4 × 106 cells per well and treated with the indicated treatments . Cells were harvested at indicated time points . LDH assays were conducted using a kit purchased from Sigma ( Cat . number TOX7 ) and assay was conducted according to the manufacturer’s protocol . BMDMs were seeded in 24-well plates at a density of 2 × 105 cells per well , treated with opsonized RBCs ( 1:10 ) and intracellular ROS was determined with a fluorometric intracellular ROS kit ( Sigma-Aldrich , MAK142 ) . Stained cells were visualized and captured using Leica DMI6000B with Cy5 filter set . The ImageJ software was applied to analyze regions of interest ( ROIs ) over multiple cells ( n > 50 ) . The fluorescence intensity from all the ROIs was averaged , and the corresponding background average was subtracted to yield the signal intensity for each condition . BMDMs were seeded in 6-well plates at a density of 2 × 106 cells per well and treated with opsonized RBCs ( 1:10 ) . Cells were harvested at indicated time points . LDH assays were conducted using a kit purchased from Abcam ( Cat . number ab138881 ) and assay was conducted according to the manufacturer’s protocol . The amount of GSH and GSSG were normalized to the protein concentration of each sample measured by a BCA protein assay . BMDMs were lysed in a lysis buffer ( 1% Trition X-100% and 2% SDS 62 . 5 mM NaCl , 1 mM EDTA , 20 mM Hepes pH 7 . 4 , 2X protease inhibitor ) and sonicated . Heme in the lysate was quantified by adding 5 µl lysate to 200 µl 2M oxalic acid and heated at 95°C for 30 min . A duplicate set of samples were kept at room temperature . Both sets were then measured for fluorescence at excitation and emission wavelengths of 400 nm and 662 nm , respectively . Values for the room temperature set were subtracted from that of the heated set and all measurements were normalized to the protein concentration of the corresponding samples . Lysates were then subjected to SDS-PAGE and immunoblotting by HMOX1 antibody ( Enzo , cat . number ADI-SPA-896 , 1:1000 dilution ) and SLC48A1 antibody as mentioned previously . Bone marrow isolated F4/80+ macrophages ( BMMs ) were seeded onto coverslips and fixed with 4% PFA pH 7 . 4 for 40 min on ice , then washed twice with DPBS . Quenching was done using 0 . 1 M ethanolamine for 5 min and room temperature , twice . Coverslips were washed twice with DPBS and incubated in a buffer containing DPBS , 3% BSA , 0 . 4% saponin for 20 min at room temperature . Coverslips were then incubated in a buffer containing DPBS , 1% BSA and 0 . 15% saponin for 10 min before incubation with the LAMP1 primary antibody ( Developmental Studies Hybridoma Bank , cat . number 1D4B , 1:100 dilution ) for 1 hr at room temperature . Cells were then washed three times with DPBS and incubated with the secondary antibody in the same buffer ( Invitrogen , cat . number A-2121 , 1:2000 dilution ) for 1 hr at room temperature . Coverslips were then washed three times with DPBS and stained with DAPI ( 1:30000 dilution ) for 1 min before mounting with pro-long antifade ( Thermofisher , cat . number P36930 ) . Images were taken using the DeltaVision Elite Deconvolution microscope . Prior to staining for flow cytometry , cells were resuspended in FACS buffer and counted to ensure that appropriate amounts of antibodies would be added . Cells were stained in 500 µl FACS buffer with the respective antibodies on ice for 30 min . After staining , cells were centrifuged at 800 g for 5 min and washed with FACS buffer before being resuspended in FACS buffer for analysis . For erythroid cell populations in the spleen and bone marrow , T cells , B cells , platelets , megakaryocytes and neutrophils were stained with fluorescein isothiocyanate ( FITC ) -conjugated-CD4 ( eBioscience , cat . number 14-0041-86 ) , CD8a ( eBioscience , cat . number 14-0081-86 ) , B220 ( eBioscience , cat . number 14-0452-86 ) , CD41 ( eBioscience , cat . number 14-0411-85 ) , Gr-1 ( eBioscience , cat . number 11-5931-82 ) and CD11b ( eBioscience , cat . number 14-0112-86 ) and dump gating was used to exclude these cell populations during analysis . In addition , cells were stained with antibodies for allophycocyanin ( APC ) -conjugated Ter-119 ( eBioscience , cat . number 47-5921-82 ) and eFluor-450-conjugated CD44 ( eBioscience , cat . number 48-0441-82 ) . For non-erythroid cell analysis , splenic cells were treated with RBC lysis buffer and stained with the following antibodies: phycoerythrin ( PE ) -conjugated Treml4 ( Biolegend , San Diego , CA , cat . number 143304 ) , APC-conjugated F4/80 ( eBioscience , cat . number 17-4801-80 ) , and PE-Cyanine7 ( PE-Cy7 ) -conjugated CD11b ( eBioscience , cat . number 25-0112-81 ) . For all antibodies , 1 µl was used to stain 1 × 106 cells . Samples were run in a FACS Canto II or FACS Aria system ( BD ) and analysis was performed using FlowJo software ( Tree Star Inc , Ashland , OR ) . The gating strategy for identifying different erythroid populations was as described elsewhere ( Chen et al . , 2009 ) , using CD44 as a marker for different populations . Cells were seeded/cultured onto pre-cleaned Aclar disks as monolayers at a density of 1 × 106 cells per well . For fixation , the growing medium was replaced with cold fixative solution ( 2 . 5% glutaraldehyde , 1% PFA; 0 . 1 M Cacodylate buffer , pH 7 . 2 ) . The cells were incubated at 4°C overnight . The aclar disks were then gently rinsed twice with cacodylate buffer and post-fixed with 2% Osmium tetroxyde for 1 hr at room temperature and pre-stained for 1 hr with saturated filtered uranyl acetate at room temperature . Dehydration and infiltration followed by using ethanol -graded series for 5 min each ( 50%; 70%; 95%−2X; 100%−3X ) and pure acetone ( 3 × 5 min ) . The aclar disks were infiltrated with 50% epon resin:acetone for 1 hr , then with 75% epon resin for overnight and 100% epon resin over 8 hr with three changes . Cells were embedded and polymerized at 60°C for 24 hr . Ultrathin ( 70 nm ) sections were obtained with diamond knife ( Diatome ) and an ultratome Leica UC6 ( Leica Microsystems , Vienna , Austria ) . Grids with sections were post stained with saturated uranyl acetate in dH2O for 20 min and with lead citrate for 10 min and imaged at 120 kV using JEOL-JEM 1400 Plus electron microscope ( Tokyo , Japan ) . Hemozoin crystals were adhered to an aluminum stub with a carbon adhesive . The stubs were then coated in a sputter coater with a layer of gold/palladium to decrease charging of sample . Samples were examined using a F . E . I . Quanta 600 with field emission gun at 20KV and a working distance of approximately 10 nm . The powder samples were mounted in a 0 . 4 mmϕ glass capillary . The X-ray powder diffraction data were measured using a large Debye-Scherrer camera with an imaging-plate as a detector installed at SPring-8 BL02B2 ( Nishibori et al . , 2001 ) . The wavelength of incident X-ray was 0 . 800 Å . The exposure time was 17 min . The X-ray powder diffraction patterns were collected in 0 . 01° steps in 2θ . The data range of the present analysis was from 2 . 0° to 30 . 0° in 2θ , which corresponds to more than 1 . 55 Å in d-spacing range . Peak positions and relative intensities of powder profile were similar to those of β-hematin ( Straasø et al . , 2011 ) . The structure of β-hematin reported by Tine et al was used for the initial model of Rietveld refinement . The rigid-body Rietveld analysis was carried out as an initial stage of refinement using the program SP ( Nishibori et al . , 2007 ) . Overall isotropic thermal factor was used in the analysis . The position and orientation of molecules were refined in the analysis . The reliability factors of final Rietveld refinement were Rwp = 2 . 6% and RI = 5 . 7% , respectively . The refined lattice constants are a = 12 . 244 ( 4 ) Å , b = 14 . 734 ( 3 ) Å , c = 8 . 072 ( 2 ) Å , α = 90 . 45 ( 2 ) ° , β = 96 . 80 ( 3 ) ° , and γ = 97 . 56 ( 2 ) ° with space group P-1 . Sample preparation was performed as described previously ( Bhattacharjee et al . , 2016 ) . Briefly , macrophages were cultured directly onto sterilized silicon nitride membranes ( 1 . 5 × 1 . 5 mm , SiN , Silson Ltd , Northhampton , England ) that had been incubated with sterile 0 . 01% Poly-L-lysine solution ( Sigma-Aldrich , St Louis , MO ) . For the XFM experiments , the cells on the SiN membranes were fixed with 4% paraformaldehyde , rinsed sequentially with PBS , isotonic 100 mM ammonium acetate , DI water and air-dried . XFM data were collected on the Bionanoprobe ( Chen et al . , 2015 ) , beamline 9-ID-B , at the Advanced Photon Source , Argonne National Laboratory , Argonne , IL . The incident X-ray energy was tuned to 10 keV using a Si-monochromator , the monochromatic beam was focused to 80 × 80 nm using a Fresnel zone plate . The sample was placed at 15˚ to the incident X-ray beam and the resulting X-ray fluorescence was collected at 90° using an energy dispersive 4-element detector ( Vortex ME-4 , SII Nanotechnology , Northridge , CA ) . Elemental maps were generated by extracting , background subtracting , and fitting the fluorescence counts for each element at each point using the program MAPS ( Vogt , 2003 ) . The fluorescent photon counts were translated into µg/cm2 using calibrated X-ray standards ( AXO products , Dresden , Germany ) . All data are shown as means ± SEM unless otherwise stated . Means of groups were compared by using Student’s unpaired t test . A p value of < 0 . 05 was considered statistically significant . Analyses were performed using PRISM seven software ( GraphPad ) . | Specialized cells , known as red blood cells , are responsible for transporting oxygen to various organs in the body . Each red blood cell contains over a billion molecules of heme which make up the iron containing portion of the hemoglobin protein that binds and transports oxygen . When red blood cells reach the end of their life , they are degraded , and the heme and iron inside them is recycled to produce new red blood cells . Heme , however , is highly toxic to cells , and can cause severe tissue damage if not properly removed . Scavenger cells called macrophages perform this recycling role in the spleen , liver and bone marrow . Collectively , macrophages can process around five million red blood cells every second or about 100 trillion heme molecules . But , it is unclear how they are able to handle such enormous volumes . Macrophages isolated from human and mice have been shown to transport heme from damaged red blood cells using a protein called HRG1 . To investigate the role HRG1 plays in heme-iron recycling , Pek et al . used a gene editing tool known an CRISPR/Cas9 to remove the gene for HRG1 from the macrophages of mice . If HRG1 is a major part of this process , removing the gene should result in a build-up of toxic heme and eventual death of the mouse . But , rather than dying of heme-iron overload as expected , these mutant mice managed to survive . Pek et al . found that despite being unable to recycle heme , these mice were still able to make new red blood cells as long as they had a diet that was rich in iron . However , the darkening color of the spleen , bone marrow , and liver in these HRG1 deficient mice indicated that these mice were still accumulating high levels of heme . Further experiments revealed that these mice protected themselves from toxicity by converting the excess heme into crystals called hemozoin . This method of detoxification is commonly seen in blood-feeding parasites , and this is the first time it has been observed in a mammal . These crystals invite new questions about how mammals recycle heme and what happens when this process goes wrong . The next step is to ask whether humans also start to make hemozoin if the gene for HRG1 is faulty . If so , this could open a new avenue of exploration into treatments for red blood cell diseases like anemia and iron overload . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2019 | Hemozoin produced by mammals confers heme tolerance |
Sleep is an essential behavioral state . It is induced by conserved sleep-active neurons that express GABA . However , little is known about how sleep neuron function is determined and how sleep neurons change physiology and behavior systemically . Here , we investigated sleep in Caenorhabditis elegans , which is induced by the single sleep-active neuron RIS . We found that the transcription factor LIM-6 , which specifies GABAergic function , in parallel determines sleep neuron function through the expression of APTF-1 , which specifies the expression of FLP-11 neuropeptides . Surprisingly FLP-11 , and not GABA , is the major component that determines the sleep-promoting function of RIS . FLP-11 is constantly expressed in RIS . At sleep onset RIS depolarizes and releases FLP-11 to induce a systemic sleep state .
Sleep is a systemic physiological state that is defined by behavioral criteria such as an increased arousal threshold , reversibility , and homeostatic regulation ( Campbell and Tobler , 1984 ) . Sleep behavior has been found in all animals that have a nervous system and that have been studied thoroughly ( Cirelli and Tononi , 2008 ) . Sleep homeostasis suggests that sleep is vital for animal life , and deprivation of this behavior typically has detrimental consequences ( Rechtschaffen and Bergmann , 1995 ) . Because of its importance , sleep is highly controlled by the nervous system , and failures of the sleep regulatory system cause disorders that are widespread in modern societies ( Panossian and Avidan , 2009 ) . In mammals , wake and sleep are controlled by two antagonistic systems . The ascending system is wake-promoting ( Moruzzi and Magoun , 1949; Starzl et al . , 1951 ) , and the descending system is sleep-promoting ( von Economo , 1930; Nauta , 1946; McGinty and Sterman , 1968; McGinty et al . , 2004; Saper et al . , 2005 ) . Central to the control of sleep are sleep-active sleep-promoting neurons such as those located in the preoptic area ( Ventral Lateral Preoptic Nucleus , VLPO and Median Preoptic Nucleus , MnPO ) , the parafacial zone of the medulla , and the thalamic reticular nucleus ( McGinty et al . , 2004; Anaclet et al . , 2012; 2014; Alam et al . , 2014; Ni et al . , 2016 ) . These neurons typically fire preferentially at the onset of sleep , they actively induce sleep , and they express the neurotransmitter GABA . The VLPO also expresses the neuropeptide Galanin ( Sherin et al . , 1998; Szymusiak et al . , 1998; Gaus et al . , 2002 ) . Interestingly , small brain areas can induce the global state of sleep that affects all areas of the brain and also other organs . It has been proposed that sleep neurons induce sleep through projections to arousal centers ( Saper et al . , 2005; Sherin et al . , 1996 ) . To ensure that sleep and wake are manifested as discrete states , the ascending and descending systems mutually inhibit each other in a flip-flop switch ( Gallopin et al . , 2000; Saper et al . , 2001 ) . Caenorhabditis elegans has become an invaluable model system for molecular dissection of biological processes ( Brenner , 1974 ) . It is amenable to genetics , has a small and invariant nervous system of just 302 neurons , and it is transparent ( Brenner , 1974; White et al . , 1986; Chalfie et al . , 1994 ) . In C . elegans , quiescence behavior can be found in satiated adults , after stress , during dauer diapause , and during larval development ( Cassada and Russell , 1975; You et al . , 2008; Hill et al . , 2014 ) . For some of these types of quiescence , it is yet unclear how they relate to sleep and they are , at least in part , regulated by different mechanisms ( Trojanowski et al . , 2015 ) . Here , we focus on a well-characterized developmentally controlled sleep behavior that can be found in C . elegans larvae prior to each of the four molts ( Cassada and Russell , 1975 ) . Developmentally controlled sleep fulfills the criteria that define sleep in other organisms ( Raizen et al . , 2008; Trojanowski and Raizen , 2016 ) . These criteria are reversibility , an increased arousal threshold , and homeostatic regulation ( Raizen et al . , 2008; Jeon et al . , 1999; Schwarz et al . , 2011; Driver et al . , 2013; Iwanir et al . , 2013; Nagy et al . , 2014 ) . Further analysis has shown that sleep behavior in C . elegans and sleep in other organisms are controlled by homologous genes such as period/lin-42 , Notch signaling , EGF signaling and several other molecules including neurotransmitter systems ( Nagy et al . , 2014; Monsalve et al . , 2011; Van Buskirk and Sternberg , 2007; Singh et al . , 2011; Singh et al . , 2014; Choi et al . , 2013 ) . These molecular similarities suggest that sleep behavior in C . elegans and sleep in other organisms share a common evolutionary origin . Sleep behavior in C . elegans has been shown to profoundly change the activity of neurons and muscles ( Schwarz et al . , 2011; Iwanir et al . , 2013; Cho and Sternberg , 2014; Schwarz et al . , 2012 ) . It requires the activity of the single interneuron RIS ( neuron class of one ring interneuron; White et al . , 1986 ) . This neuron is active at the onset of sleep , it actively induces sleep , and it expresses GABA ( Turek et al . , 2013 ) . Thus , RIS is similar to sleep-active neurons in mammals . In order to be sleep-inducing , RIS requires APTF-1 , a highly conserved transcription factor of the AP2 family . Without APTF-1 , RIS is still sleep-active but can no longer induce sleep ( Turek et al . , 2013 ) . In humans , mutation in the AP2 homolog TFAP2beta causes Char syndrome , which is linked to insomnia or sleepwalking ( Mani et al . , 2005 ) . Together , this supports the view that sleep-neurons and AP2 transcription factors are conserved regulators of sleep . However , the mechanism of how APTF-1 renders RIS sleep promoting is unclear . Here , we identify a gene regulatory system that determines the sleep-inducing function of RIS . In this network , a transcription factor that controls GABAergic function in a subset of neurons , LIM-6 , in parallel controls the expression of the APTF-1 transcription factor . APTF-1 , in turn , specifies the expression of sleep-inducing FLP-11 peptides . FLP-11 is always present in RIS , and thus , this neuron can induce sleep at any time it gets activated . At sleep onset , calcium transient activity of RIS increases and leads to the release of FLP-11 peptides , which induce quiescence . Thus , we show that sleep can be induced systemically by the single RIS neuron through FLP-11 release .
Sleep-active neurons express GABA in both mammals and C . elegans . The LIM homeobox transcription factor LIM-6 is expressed in a subset of GABAergic neurons including RIS and has been shown to be required for some aspects of GABAergic neuron specification , including the expression of the GABA-synthesizing enzyme glutamate decarboxylase UNC-25 ( Hobert et al . , 1999; Jin et al . , 1999 ) . Thus , we tested whether LIM-6 is involved in sleep control . First , we investigated the spontaneous behavior of two lim-6 mutants , which contain large deletions and represent strong loss-of-function mutations . We cultured worms in microfluidic compartments , filmed their activity over the sleep-wake cycle , and quantified their locomotion behavior by tracking nose movement . Similar to the wild type , lim-6 mutants stopped feeding before the molt , allowing the identification of lethargus , the developmental time the larvae should be sleeping . The mutants had strongly reduced or even complete absence of immobility during the non-pumping phase ( Figure 1A ) . We then quantified RIS activation in lim-6 mutant worms by imaging the calcium indicator GCaMP3 expressed in RIS ( Tian et al . , 2009; Butler , 2012 ) . We found that RIS activity increased at sleep onset in wild-type animals and also normally increased in lim-6 mutants during the time the animal should enter sleep ( Figure 1B ) . Because the sleep phenotype of the lim-6 mutant worms was similar to the sleep phenotype of aptf-1 mutants ( Turek et al . , 2013 ) , we tested whether lim-6 controls aptf-1 expression . We crossed a line expressing mKate2 , a red-fluorescent protein ( Shcherbo et al . , 2009 ) , under the control of the aptf-1 promoter into lim-6 mutant worms and quantified the expression of mKate2 in RIS . aptf-1 expression was completely abolished in most individuals and strongly reduced in the remaining ones in all developmental stages ( Figure 1C , Figure 1—figure supplement 1A ) . We also looked at GABAergic function in aptf-1 mutant worms and found that aptf-1 did not control the expression of unc-25 or the vesicular GABA transporter gene unc-47 ( Figure 1D , Figure 1—figure supplement 1B ) . Thus , LIM-6 does not appear to affect sleep primarily through GABAergic function determination . Rather , LIM-6 controls sleep-promoting function primarily through a parallel pathway that depends on APTF-1 . 10 . 7554/eLife . 12499 . 003Figure 1 . The LIM homeobox transcription factor LIM-6 controls sleep by specifying expression of the transcription factor APTF-1 in RIS . ( A ) Probability distribution of nose speeds during wake and sleep for wild type and lim-6 mutants . lim-6 ( nr2073 ) shows substantially reduced and lim-6 ( tm4836 ) shows a complete lack of immobility during the time the animals should be sleeping . ( B ) Averaged RIS calcium activity pattern across time in wild type and lim-6 ( tm4836 ) . RIS is active at the onset of sleep in wild type and in lim-6 ( tm4836 ) . There was no statistically significant difference between wild-type and lim-6 worms ( p > 0 . 05 , Welch test ) . ( C ) Expression of paptf-1::mKate2 in wild type and lim-6 ( nr2073 ) L1 larvae . Expression of mKate2 is absent in RIS in lim-6 ( nr2073 ) showing that LIM-6 controls expression of APTF-1 in RIS . ( D ) Expression of punc-25::GFP in wild type and aptf-1 ( gk794 ) . Reporter GFP expression is normal in RIS in aptf-1 mutant worms indicating that GABAergic function is not controlled by APTF-1 . Statistical tests used were Wilcoxon Signed Paired Ranks test for comparison within the same genotype and Student’s t-test for comparisons between genotypes . Error bars are SEM . ** denotes statistical significance with p<0 . 01 , *** denotes statistical significance with p<0 . 001 . Scale bars are 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12499 . 00310 . 7554/eLife . 12499 . 004Figure 1—figure supplement 1 . LIM-6 controls expression of APTF-1 across development but does not control the expression of the GABA vesicular transporter gene unc-47 . ( A ) Expression of paptf-1::mKate2 in wild type and in lim-6 ( nr2073 ) L4 larvae . Expression of mKate2 is absent in RIS in lim-6 ( nr2073 ) at the L4 stage showing that LIM-6 generally controls expression of APTF-1 in RIS rather than the onset of expression . Scale bar is 10 µm . ( B ) punc-47::GFP is expressed in aptf-1 ( gk794 ) mutant worms . Shown are L1 larvae . Scale bar is 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12499 . 004 How is sleep-promoting function generated by APTF-1 ? AP2 transcription factors are highly conserved regulators of gene expression ( Zhao et al . , 2011; Eckert et al . , 2005 ) suggesting that aptf-1 acts through gene expression in RIS . Thus , we determined the transcriptional profile of aptf-1 mutants . We performed transcriptional profiles of aptf-1 mutants using microarrays using two conditions . First , because aptf-1 is most strongly expressed during late embryogenesis , we analyzed pretzel-stage embryos . Second , because aptf-1 is required for sleep behavior , we analyzed sleeping L4 larvae . The microarray experiment revealed that flp-11 , a neuropeptide gene , was strongly downregulated in aptf-1 mutant worms in both embryos and larvae ( Figure 2A , Supplementary file 1 , Table 1A and 1B ) . We thus further analyzed the role of flp-11 in sleep control . We also analyzed some other genes that were regulated by aptf-1 , but none of them appeared to be important for sleep regulation ( Figure 2—figure supplements 1 and 2 ) . To test whether flp-11 is expressed in RIS and to verify its regulation by aptf-1 , we generated an mKate2 promoter fusion as a reporter line . We then checked the expression in wild-type and aptf-1 mutant backgrounds using fluorescence microscopy . flp-11 was expressed strongly in RIS and faintly in a few additional neurons . Expression was abolished or greatly reduced in RIS in aptf-1 mutant worms ( Figure 2B , Figure 2—figure supplement 3 ) . We followed the expression of flp-11:mKate2 over the sleep-wake rhythm and found that mKate2 was constantly expressed during larval development both during sleep and wake ( Figure 2C ) . Transcription factors are often structurally conserved , and the homolog from a highly divergent species can replace the mutation of the endogenous factor ( Halder et al . , 1995 ) . There are five homologs of APTF-1 in mammals , designated TFAP2-alpha to TFAP2-epsilon ( Eckert et al . , 2005 ) . We tested for structural conservation of AP2 by expressing the mouse homolog of aptf-1 , tfap2beta , in RIS using the aptf-1 promoter . For our experiment , we chose TFAP2beta , because it has been linked to insomnia in humans ( Mani et al . , 2005 ) . Nose speed measurements during sleep showed that tfap2beta expression partially restored immobility ( Figure 2D , E ) . Expression of mouse tfap2beta also partially restored the expression of flp-11 in aptf-1 mutant worms ( Figure 2F , Figure 2—figure supplement 4 ) . The structural conservation of AP2 transcription factors suggests that also the DNA binding site is conserved . In fact , it has been shown that the binding site of AP2 is conserved over 600 million years of bilaterian evolution , and the different mammalian AP2 paralogs have nearly identical binding sites ( Nitta et al . , 2015 ) . Thus , we searched for mammalian AP2 transcription factors binding sites in the promoter region of flp-11 ( Eckert et al . , 2005; Grant et al . , 2011 ) . Indeed , we found a putative AP2-binding site in the flp-11 promoter region , consistent with the regulation of flp-11 by aptf-1 ( Figure 2G ) . These results show that aptf-1 is required for flp-11 expression in RIS . 10 . 7554/eLife . 12499 . 005Figure 2 . The AP2 transcription factor APTF-1 controls FLP-11 expression in RIS . ( A ) Transcriptional analysis of aptf-1 ( gk794 ) mutants revealed genes that are regulated by APTF-1 . Wild-type and aptf-1 ( gk794 ) pretzel-stage embryos and sleeping L4 larvae were used for a transcriptome analysis . In both life stages , expression of the FMRFamide-like neuropeptide FLP-11 was strongly reduced in aptf-1 ( gk794 ) . This suggests transcriptional control of FLP-11 by APTF-1 . Data can be found in Supplementary file 1 , Tables 1A and 1B . ( B ) Expression of pflp-11::mKate2 in wild type and aptf-1 ( gk794 ) . Expression of mKate2 was absent in RIS in aptf-1 ( gk794 ) showing that APTF-1 controls expression of FLP-11 . Expression of flp-11 in RIS was reminiscent to the expression of the flp-11 homolog afp-6 in RIS in Ascaris nematodes ( Yew et al . , 2007 ) . Expression for additional genes can be found in Figure 2—figure supplements 2 and 3 . ( C ) flp-11 expression profile in RIS over the sleep-wake cycle . Expression does not change with the sleep-wake cycle . ( D ) Probability distribution of nose speeds during wake and sleep for wild type , aptf-1 ( gk794 ) and aptf-1 ( gk794 ) ; paptf-1::tfap2β rescue . ( E ) Comparison of immobility during sleep for wild type , aptf-1 ( gk794 ) , and aptf-1 ( gk794 ) ; paptf-1::tfap2β . The mouse TFAP2β partially rescued the aptf-1 ( gk794 ) sleep phenotype . ( F ) Comparison of pflp-11::GFP fluorescence intensity in RIS for wild type , aptf-1 ( gk794 ) , and aptf-1 ( gk794 ) ; paptf-1::tfap2β . The mouse TFAP2β partially rescued the expression of flp-11 in RIS ( 18% of wild-type level ) . ( G ) Analysis of putative AP2-binding sites in the flp-11 promoter region . The flp-11 promoter region was scanned for the primary mouse AP2α-binding site . Overlap was found ( p<0 . 001 , q=0 . 06 ( Grant et al . , 2011 ) ) for one binding site . Statistical test used was Wilcoxon Signed Paired Ranks test . ** denotes statistical significance with p<0 . 01 , *** denotes statistical significance with p<0 . 001 . Scale bar is 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12499 . 00510 . 7554/eLife . 12499 . 006Figure 2—figure supplement 1 . C10C6 . 7 is a putative four transmembrane helix protein that is expressed in RIS and that is controlled by aptf-1 . ( A ) Bioinformatics analysis suggests that C10C6 . 7 is a four transmembrane helix protein ( Krogh et al . , 2001 ) . ( B ) Expression pattern of GFP-tagged fosmids for C10C6 . 7 . Expression is visible in nine cells: interneuron RIS; pharyngeal neurons M1 , M2 , Motor-interneuron ( M-IN ) , an unidentified pair of pharyngeal neurons ( ** ) and an unidentified pair of sensory neurons ( * ) . Expression of C10C6 . 7 protein in RIS is controlled by aptf-1 . ( C ) Probability distribution of nose speeds during wake and sleep for C10C6 . 7 ( goe3 ) and C10C6 . 7 ( goe5 ) shows that C10C6 . 7 does not play a significant role in sleep control . Statistical test used was Wilcoxon Signed Paired Ranks test . Scale bars are 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12499 . 00610 . 7554/eLife . 12499 . 007Figure 2—figure supplement 2 . Expression pattern of sto-3 and H19N07 . 3 . ( A ) Expression pattern of sto-3 promoter fusions . STO-3 is expressed in RIB neuron and additionally in three unidentified non-neuronal cells in the tale ( not shown ) . ( B ) Probability distribution of nose speeds during wake and sleep for sto-3 ( tm1488 ) shows that sto-3 does not play a significant role in sleep control . ( C ) Expression pattern of GFP-tagged fosmids for H19N07 . 3 . The H19N07 . 3 protein is expressed in all somatic cell nuclei , but its levels are not regulated by aptf-1 . Statistical test used was Wilcoxon Signed Paired Ranks test . Scale bars are 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12499 . 00710 . 7554/eLife . 12499 . 008Figure 2—figure supplement 3 . FLP-11 is strongly expressed in RIS and weakly in additional neurons . APTF-1 controls the expression in RIS . ( A ) Expression pattern of ynIs40 ( pflp-11::GFP ) in wild type and aptf-1 ( gk794 ) mutant . The transgene expresses in several neurons including RIS . aptf-1 ( gk794 ) abolishes the expression specifically in RIS . ( B ) Expression pattern of goeIs288 ( pflp-11::mKate2 ) in wild type and the aptf-1 ( gk794 ) mutant . Strong expression is visible only in RIS . By increasing the contrast to the point where RIS is over-saturated several additional neurons becomes visible that may be identical to those seen in ynIs40 ( Kim and Li , 2004 ) . aptf-1 ( gk794 ) strongly reduces the expression specifically in RIS . Scale bars are 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12499 . 00810 . 7554/eLife . 12499 . 009Figure 2—figure supplement 4 . Mouse TFAP2beta partially restores expression of flp-11 neuropeptides in RIS in aptf-1 mutant worms . Expression of pflp-11::GFP in wild type , aptf-1 ( gk794 ) , and aptf-1 ( gk794 ) ; tfap2beta rescue . In aptf-1 ( gk794 ) , expression of flp-11 is strongly reduced but could partially be restored by the mouse TFAP2beta . Expression of flp-11::GFP in the rescue strain varied between 5–50% of wild-type levels . Here , we show a picture of 50% rescue . Scale bar is 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12499 . 009 We next tested whether flp-11 is required for sleep behavior . We analyzed an flp-11 deletion that is predicted to affect all four peptides encoded by the gene and filmed and quantified sleep behavior as before . flp-11 mutant larvae showed a normal pre-molting cessation of pumping . However , locomotion quiescence was strongly reduced in flp-11 mutant larvae . Although nose immobility was 48% in wild type , it was strongly reduced to less than 14% in flp-11 mutants ( Figure 3A ) . To verify that the sleep phenotype observed in flp-11 mutants was caused by flp-11 deletion , we tested whether a wild-type copy of flp-11 would restore locomotion quiescence during sleep . We found that the wild-type transgene could rescue the flp-11 mutant phenotype ( Figure 3A ) . If flp-11 is a major target of aptf-1 , a wild-type copy of flp-11 should also rescue , at least partially , the sleep phenotype of aptf-1 mutants . Because the flp-11 promoter is regulated by aptf-1 , we used the aptf-1 promoter to drive expression of flp-11 in aptf-1 mutants . Whereas aptf-1 mutants without the transgene did not show any detectable immobility , the flp-11 transgene partially restored immobility ( Figure 3A ) . It is likely that aptf-1 acts through additional targets , which may explain why the rescue observed after flp-11 expression in the aptf-1 mutant was partial and small . We next investigated the activation of RIS at sleep onset in flp-11 mutants using GCaMP3 . RIS strongly activated at the onset of the non-pumping period in flp-11 mutant worms ( Figure 3B ) . Thus , flp-11 is not required for activation of RIS at sleep onset . To test whether sleep induction by RIS is impaired in flp-11 mutant worms , we optogenetically activated RIS with Channelrhodopsin-2 during wake and followed the behavioral response of the worms by nose tracking ( Nagel et al . , 2005 ) . Although wild-type animals showed a reduction in movement and became immobile after blue light illumination , flp-11 mutant worms did not decrease their movement but rather increased it ( Figure 3C ) . If FLP-11 peptides are sleep-promoting , then ectopic overexpression during wake may induce anachronistic quiescence ( Singh et al . , 2011; Nelson et al . , 2013; 2014 ) . We overexpressed FLP-11 in adult worms using a heat-shock-inducible promoter that drives broad expression in the nervous system and other tissues ( Jones et al . , 1986 ) . After a 5-min heat shock , we followed the fraction of immobilized worms over time . Adult worms that were expressing FLP-11 driven by the heat shock promoter became immobile 1–2 hr after the heat shock . Control worms that were heat shocked but did not express the transgene or expressed other flp genes did not show any immobilization ( Figure 4A , Figure 4—figure supplement 1 ) . Taken together these data imply that RIS induces sleep through FLP-11 . These peptides are expressed in RIS during both sleep and wake , and optogenetic activation of RIS can induce quiescence during wake . This suggests a model in which RIS can induce sleep at any time . According to this model , RIS depolarizes at sleep onset and it releases FLP-11 , which then induces sleep . 10 . 7554/eLife . 12499 . 010Figure 3 . RIS induces sleep through the sleep-inducing FMRFamide-like neuropeptide FLP-11 . ( A ) Probability distribution of nose speeds during wake and sleep for wild type , flp-11 ( tm2706 ) , flp-11 ( tm2706 ) ; pflp-11::flp-11 rescue and aptf-1 ( gk794 ) ; paptf-1::flp-11 rescue . Immobility during the time the animal should be sleeping was substantially reduced in flp-11 ( tm2706 ) . flp-11 ( tm2706 ) could be rescued by expression of the wild-type flp-11 gene . Furthermore , expression of flp-11 in aptf-1 ( gk794 ) partially rescued sleep behavior . ( B ) Averaged RIS calcium activity pattern across time in wild type and flp-11 ( tm2706 ) . RIS was strongly activated at the onset of sleep in flp-11 ( tm2706 ) ( Student’s t-test ) . ( C ) Channelrhodopsin-2 activation of aptf-1-expressing neurons caused immediate immobility in wild type . In contrast , flp-11 ( tm2706 ) accelerated upon blue light stimulation showing that RIS-dependent immobility is impaired . Statistical tests used were Wilcoxon Signed Paired Ranks test for comparisons within genotypes and Student’s t-test for comparisons between genotypes . Error bars are SEM . ** denotes statistical significance with p<0 . 01 , *** denotes statistical significance with p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12499 . 01010 . 7554/eLife . 12499 . 011Figure 4 . Multiple receptors may be involved in sleep induction . ( A ) Behavioral analysis over time after the heat-shock-induced overexpression of flp-11 in wild-type and the npr-22 ( ok1598 ) ; frpr-3 ( ok3302 ) ; npr-4 ( tm1782 ) triple mutant . To assess the effect of the heat shock on quiescence , wild-type worms without flp-11 overexpression were analyzed at the same time and do not show any behavioral changes . Overexpression of flp-11 caused anachronistic quiescence that was lasting approximately 1 hr in the wild type . Quiescence was significantly reduced by approximately 50% in the triple mutant . ( B ) Probability distribution of nose speeds during wake and sleep for wild type and npr-22 ( ok1598 ) ; frpr-3 ( ok3302 ) ; npr-4 ( tm1782 ) triple mutant . Immobility during sleep was reduced by about 30% in the npr-22 ( ok1598 ) ; frpr-3 ( ok3302 ) ; npr-4 ( tm1782 ) triple mutant . ( C ) Expression patterns of frpr-3 , npr-4 and npr-22 promoter fusions . FRPR-3 is expressed in about 30 neurons , mostly in the head . Expression of NPR-4 was seen in about five neurons . NPR-22 was expressed in several neurons and muscle tissue including pharynx and head muscle . ( D ) Expression patterns of GFP-tagged fosmids for frpr-3 , npr-4 , and npr-22 . FRPR-3 and NPR-4 were mostly expressed around the nerve ring . NPR-22 localized broadly to the plasma membrane in several neurons , pharynx muscle , head muscle , and the anal sphincter muscle . Statistical tests used were Wilcoxon Signed Paired Ranks test for comparisons within genotypes and Student’s t-test for comparisons between genotypes . Error bars are SEM . ** denotes statistical significance with p<0 . 01 , *** denotes statistical significance with p<0 . 001 . Scale bars are 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12499 . 01110 . 7554/eLife . 12499 . 012Figure 4—figure supplement 1 . Heat-shock-induced flp-11 overexpression causes quiescence but heat-shock-induced overexpression of three other flp genes does not , suggesting that quiescence cannot be induced by overexpression of any flp . ( A ) Behavioral analysis of wild-type adult worms over time after 5 min of heat shock at 37°C . ( B-E ) Behavioral analysis of wild-type adult worms over time after heat shock-induced overexpression of flp-11 , flp-1 , flp-10 , and flp-20 . Only heat-shock-induced overexpression of flp-11 induces quiescence . DOI: http://dx . doi . org/10 . 7554/eLife . 12499 . 01210 . 7554/eLife . 12499 . 013Figure 4—figure supplement 2 . Single receptors mutants do not show reduced quiescence during sleep , but do show reduced quiescence upon heat-shock-induced overexpression of flp-11 . ( A ) Heat shock-induced overexpression of flp-11 in single and double mutants of frpr-3 ( ok3302 ) , npr-4 ( tm1782 ) , and npr-22 ( ok1598 ) . The strength of the anachronistic quiescence correlates with the combination of receptors ranging from highest for single mutants to lowest for the double mutants . ( B ) Probability distribution of nose speeds during wake and sleep for wild type and npr-22 ( ok1598 ) , frpr-3 ( ok3302 ) , npr-4 ( tm1782 ) single mutants . Immobility during sleep is not significantly different between wild type and single mutants . Statistical tests used were Wilcoxon Signed Paired Ranks test for comparisons within genotypes and Student’s t-test for comparisons between genotypes . Error bars are SEM . *denotes statistical significance with p<0 . 05 , **denotes statistical significance with p<0 . 01 , *** denotes statistical significance with p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12499 . 013 Many neuropeptides act through G-protein-coupled receptors ( Peymen et al . , 2014; Frooninckx et al . , 2012 ) . To search for effectors through which FLP-11 induces sleep , we investigated three neuropeptide receptors that are activated by FLP-11 peptides in in vitro assays . These receptors are FRPR-3 , NPR-4 , and NPR-22 ( Frooninckx et al . , 2012; Mertens et al . , 2006; Mertens et al . , 2004; Cohen et al . , 2009 ) . We first tested whether deletion of these receptors can suppress FLP-11-induced anachronistic quiescence . We crossed receptor deletions into our FLP-11 overexpressing line and quantified heat-shock-induced quiescence . We found that the maximum quiescence was reduced in each of these mutants , albeit only slightly ( Figure 4—figure supplement 2A ) . Thus , we tested whether these receptors act redundantly by testing all double mutant permutations and a triple mutant containing all three receptor deletions at the same time . The double mutants had further decreased quiescence , and the triple mutant had the strongest reduction in quiescence ( Figure 4A , Figure 4—figure supplement 2A ) . Thus , the quiescence induced by FLP-11 overexpression partly depends on multiple effectors . We next investigated sleep behavior in the receptor mutants . Whereas the single receptor mutants did not show a significant reduction in immobility , the receptor triple mutant showed a small increase in nose speed during sleep and a small reduction of immobility ( Figure 4B , Figure 4—figure supplement 2B ) . We investigated transgenic animals expressing promoter fusions containing the putative FLP-11 receptor and mKate2 . To investigate the subcellular localization , we made transgenic animals expressing GFP-tagged proteins for the three receptors ( Sarov et al . , 2012 ) . The promoter fusions showed that FRPR-3 expressed in approximately 30 neurons , mostly in the head . NPR-4 expressed in five neurons ( Cohen et al . , 2009 ) . NPR-22 expressed in several neurons , muscle tissue in the pharynx and in the head ( Figure 4C ) . Interestingly , the receptors were expressed in neurons that are not postsynaptic to RIS ( White et al . , 1986 ) . The only exception that we found was the AVK neuron , which is postsynaptic to RIS and expressed FRPR-3 . Translational fusions showed that FRPR-3 and NPR-4 were weakly expressed and localized mostly to the nerve ring , whereas NPR-22 was localized broadly to the plasma membrane of neurons , and muscles of the pharynx , head , and muscle ( Figure 4D ) . The phenotypes for the GPCRs that we observed were small and the binding affinities of FLP-11 peptides to these GPCRs in vitro were low which makes it premature to conclude that these receptors act by binding of FLP-11 ( Frooninckx et al . , 2012; Mertens et al . , 2004; 2006; Cohen et al . , 2009 ) . It could be that FLP-11 acts through one main receptor , which remains unidentified . Alternatively , FLP-11 could act through several redundant receptors , which may include FRPR-3 , NPR-4 , and NPR-22 .
GABA has been proposed to play a major role in sleep function . Its conserved expression in sleep-active neurons suggests that it has an important function in these neurons . Enhancers of GABAergic neurotransmission have been used to treat sleeping problems and GABA has been suggested to play a role in sleep induction in several systems including C . elegans ( Singh et al . , 2014; Dabbish and Raizen , 2011 ) . In contrast to the common view that GABA is the major sleep inducer in sleep-active neurons , we did not find evidence that GABA is the major sleep-inducing transmitter in RIS . This is consistent with our previous observation that optogenetic activation of RIS still causes quiescence in unc-25 mutant worms and that unc-25 mutant worms still show sleeping behavior ( Turek et al . , 2013 ) . In addition , here we show that GABAergic function induction can be separated from sleep neuron function downstream of lim-6 . This suggests that GABA plays a rather minor role in sleep-induction in RIS and that we did not detect it . More specific and more sensitive assays may resolve the question of the role of GABA in RIS in the future . Our results show that FLP-11 is a crucial sleep-inducing component in RIS and an important target of aptf-1 . In mammals , sleep neurons of the VLPO express the inhibitory neuropeptide Galanin , and projections extend to the tuberomammillary nucleus , which expresses Galanin receptors ( Sherin et al . , 1998; Gaus et al . , 2002 ) . The locus coeruleus , a wake-promoting brain region , is also innervated by VLPO projections and can be inhibited by Galanin administration ( Seutin et al . , 1989; Pieribone et al . , 1995 ) . Also , Galanin has been shown to have sedating effects on both zebrafish and human subjects ( Woods et al . , 2014; Murck et al . , 2004 ) . However , sleep phenotypes for Galanin knockouts have not been reported , despite being available for several years ( Wynick et al . , 1998; Kerr et al . , 2000 ) . These experiments suggest that Galanin has a modulatory role on sleep , but may not be central to sleep induction . Galanin does not appear to be homologous to FLP-11 , as it belongs to the family of Galanin peptides ( Lang et al . , 2007 ) , whereas flp-11 encodes peptides of the RFamide family ( Li et al . , 1999 ) . Also , unlike FLP-11 , Galanin is expressed widely in the brain and has diverse functions ( Maria Vrontakis , 2002 ) . In Drosophila , sleep requires a neuropeptide called sNPF , which may be functionally similar to FLP-11 as both are inhibitory and are released from sleep-promoting neurons ( Shang et al . , 2013; Vecsey et al . , 2014; Chen et al . , 2013 ) . Thus , inhibitory neuropeptides appear to play important roles in sleep-promoting neurons across species . Taken together , we present a model for how sleep-promoting function is generated and for how sleep is induced ( Figure 5 ) . In this model , the transcription factor LIM-6 separately controls GABAergic and sleep-promoting functions . Sleep-promoting function is mediated by the expression of the APTF-1 transcription factor , which is crucially required for sleep induction by RIS . APTF-1 , in turn , is required for the expression of sleep-inducing FLP-11 peptides . FLP-11 is always present in RIS allowing the induction of sleep at any time the neuron activates . At sleep onset , an unknown signal triggers depolarization and calcium influx in RIS , which then triggers release of FLP-11 peptides to systemically induce sleep behavior . 10 . 7554/eLife . 12499 . 014Figure 5 . Model for generation of sleep-promoting function of RIS and sleep induction by RIS . According to this model , the transcription factor LIM-6 controls GABAergic and peptidergic function in RIS in parallel . To render this neuron sleep-promoting , LIM-6 is required for the expression of the APTF-1 transcription factor . APTF-1 , in turn , is required for the expression of sleep-inducing FLP-11 peptides . FLP-11 is present in RIS at all times . Sleep onset is triggered by an unknown signal , which leads to a depolarization and to calcium influx . This triggers FLP-11 release , which in turn systemically induces sleep behavior . DOI: http://dx . doi . org/10 . 7554/eLife . 12499 . 014
C . elegans worms were grown on Nematode Growth Medium ( NGM ) plates seeded with E . coli OP50 at 25°C as described ( Brenner , 1974 ) . The strains and alleles that were used in this study can be found in Supplementary file 2 . The deletion alleles were backcrossed two to ten times against N2 to generate HBR lines ( the exact number of backcrosses for each strain is indicated in brackets ) . Insertions were backcrossed two times against N2 to remove the unc-119 ( - ) background . Backcrossed strains were the basis for all experiments . During backcrossing , the genotypes were followed by PCR . Primers to detect the deletions using a three primer PCR can be found in Supplementary file 3 . All constructs were cloned using the Multisite Gateway system ( Invitrogen , Waltham , MA , USA ) into pCG150 ( Merritt and Seydoux , 2010 ) . All constructs obtained from LR reactions were sequenced for verification . Plasmids and Fosmid that were used are listed in Supplementary file 4 . The tfap2beta gene was codon-optimized for C . elegans as described ( Redemann et al . , 2011 ) . We generated transgenic strains by microparticle bombardment or by microinjection using unc-119 ( ed3 ) rescue as a selection marker ( Wilm et al . , 1999; Praitis et al . , 2001 ) . For fosmid isolation , we used the FosmidMAX DNA Purification kit ( epicentre ) or Qiagen plasmid midi kit . The deletion alleles C10C6 . 7 ( goe3 ) and C10C6 . 7 ( goe5 ) were created with a CRISPR/Cas9 system as it was described before ( Friedland et al . , 2013 ) . Target sequence for the sgRNA was GTTATGGTGAGAAGGAAAGCtgg . The C10C6 . 7 gene locus was sequenced and the deletions were mapped to the second exon . They are 25 bp and 4 bp long , respectively , and cause a frame shift , thus most likely are molecular null alleles . All long-term imaging experiments were carried out using agarose microchamber imaging as described ( Bringmann , 2011; Turek et al . , 2015 ) . For behavioral analysis , worms were filmed in a burst mode every 10–15 min for 20 s with a frame rate of 2 pictures/second . Nose tracking was performed manually . Mean velocities of nose speed were calculated for sleep and wake , where sleep was defined as the non-pumping phase and wake was defined as a 2-hr period directly before sleep . Calcium imaging was performed similar as described before using GCaMP3 . 35 and co-expression of mKate2 as an expression control ( Schwarz et al . , 2011; 2012; Turek et al . , 2013; 2015; Schwarz and Bringmann , 2013 ) . For calcium imaging , we used an Andor ( UK ) iXon ( 512 x 512 pixels ) EMCCD camera and LED illumination ( CoolLed , UK ) using standard GFP and Texas Red filter sets ( Chroma , Bellow Falls , VT ) . Exposure times were in the range of 5-20 ms and allowed imaging of moving worms without blurring . The EMCCD camera triggered the LED through a TTL 'fire' signal to illuminate only during exposure . LED intensity was in the range of 15–30% . EM gain was between 50 and 250 . All calcium-imaging experiments were done in agarose microchambers . Typically , 4–15 individuals were cultured in individual microchambers that were in close vicinity . Animals were filmed by taking a z-stack every 6 or 10 min or in a continuous mode , which means using a frame rate of 1 picture / 4 s . If more than four animals were filmed in parallel , individual compartments were repeatedly visited by using an automatic stage ( Prior Proscan2/3 , Rockland , MA ) set to low acceleration speeds . Before each fluorescent measurement , we took a brief DIC movie to assess the developmental stage and behavioral state . Larvae that showed pharyngeal pumping were scored as being in the wake-like state . Movies were analyzed using homemade Matlab routines . For fluorescence imaging of reporter lines ( Figure 1C , D , Figure 2B , Figure 4C , D , Figure 1—figure supplements 1A , B , 2B , 3A , B , 4A , B , 5 ) , we used spinning disc imaging with an Andor Revolution spinning disc system using a 488 nm laser and a 565 nm laser , a Yokogawa ( Japan ) X1 spinning disc head , a 100x oil objective and an iXon EMCCD camera . Z stacks were taken and a maximum intensity projection calculated using iQ software . Channelrhodopsin experiments were performed inside agarose microchambers as described ( Turek et al . , 2013 ) . We grew hermaphrodite mother worms on medium that was supplemented with 0 . 2 mM all trans Retinal ( Sigma-Aldrich , St . Louis , MO ) . We then placed eggs from these mothers together with food into microchambers without any further retinal supplementation . We stimulated Channelrhodopsin with an LED of 490 nm with about 0 . 36 mW/mm2 as measured with a light voltmeter . Images were captured with an Andor Neo sCMOS camera ( 2560 x 2160 pixels ) . Worms were filmed every 30 min for 60 s with a frame rate of two pictures / second . Channelrhodopsin stimulation with constant blue light was applied for 20 s starting after 20 s . Nose tracking was performed manually . We calculated mean velocities for wake using a period of 2 hr directly before sleep . We crossed goeIs290 and goeIs285 in the following strains to identify neurons: MU1085 ( Wightman et al . , 2005 , EG1285 ( McIntire et al . , 1997 ) , BZ555 ( Nass et al . , 2002 ) , HBR1213 , OH1422 ( Tsalik et al . , 2003 ) , HBR887 , HBR777 , QW122 ( Donnelly et al . , 2013 ) . For both wild-type and mutant conditions , four biological samples were collected . For transcriptional profiling of pretzel-stage embryos , each sample contained approximately 3000 animals that were picked manually into one ml of Trizol ( Invitrogen ) . For transcriptional profiling of sleeping L4 larvae , each sample contained approximately 200 animals that were picked manually into one ml of Trizol ( Invitrogen ) . Transcriptional profiling and microarray data analysis was done the same way as it was described before ( Turek and Bringmann , 2014 ) except that fold change threshold was 1 . 5 and the GO term analysis was omitted . Microarray data was deposited at the GEO database and can be accessed using the following links: http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE73282 http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE73283 GSM1890106 aptf-1_1 GSM1890107 aptf-1_2 GSM1890108 aptf-1_3 GSM1890109 aptf-1_4 GSM1890110 N2_1 GSM1890111 N2_2 GSM1890112 N2_3 GSM1890113 N2_4 GSM1890140 aptf-1_1 in pretzel-stage embryos GSM1890141 aptf-1_2 in pretzel-stage embryos GSM1890142 aptf-1_3 in pretzel-stage embryos GSM1890143 aptf-1_4 in pretzel-stage embryos GSM1890144 N2_1 in pretzel-stage embryos GSM1890145 N2_2 in pretzel-stage embryos GSM1890146 N2_3 in pretzel-stage embryos GSM1890147 N2_4 in pretzel-stage embryos For heat-shock-induced overexpression of flp neuropeptides , we cultured adult worms on NGM plates seeded with E . coli OP50 and sealed with parafilm . Heat shock was applied using a water bath at 37°C where the plates were placed for 5 min , the agar side facing the water . Worms were scored for moving / pumping behavior directly after heat shock and consecutively in time intervals of 30 min for 4 hr . Statistical tests used were Wilcoxon Signed Paired Ranks test , Student’s t-test or Welch test using Origin software . Error bars are SEM . For statistical analysis of overexpression experiments , a Fisher’s exact test was done in Matlab . | Sleep keeps us healthy and happy , and is essential for all animals . Specialized neurons in the brain become highly active to generate this restful state . There are relatively few of these “sleep-active” neurons in the brain , but they are able to control sleep in the entire animal . Like most other neurons , sleep-active neurons release substances called neurotransmitters . The sleep-active neurons in many different species release a neurotransmitter called GABA , although they also contain other neurotransmitters called neuropeptides that were thought to be less important for triggering sleep . The roundworm Caenorhabditis elegans has become an important model system for studying the molecular biology of sleep as it contains only one sleep-active neuron . Turek et al . have now studied this C . elegans neuron and have discovered transcription factors – proteins that control gene expression – that are required for the sleep-active neuron to induce sleep . Further investigation revealed that the transcription factors specify the production of a neuropeptide called FLP-11 . The sleep-active neuron always contains FLP-11 , but only releases it as sleep begins . Once released , FLP-11 moves onto target cells to induce sleep in the entire organism . Thus , FLP-11 – and not GABA – is the major sleep-inducing neurotransmitter in C . elegans . To induce a sleep state throughout an entire organism , the activities of many different cells must be controlled . A future challenge will be to figure out how FLP-11 does this . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Sleep-active neuron specification and sleep induction require FLP-11 neuropeptides to systemically induce sleep |
Skeletal muscle is highly sensitive to mutations in genes that participate in membrane stability and cellular attachment , which often leads to muscular dystrophy . Here we show that Thrombospondin-4 ( Thbs4 ) regulates skeletal muscle integrity and its susceptibility to muscular dystrophy through organization of membrane attachment complexes . Loss of the Thbs4 gene causes spontaneous dystrophic changes with aging and accelerates disease in 2 mouse models of muscular dystrophy , while overexpression of mouse Thbs4 is protective and mitigates dystrophic disease . In the myofiber , Thbs4 selectively enhances vesicular trafficking of dystrophin-glycoprotein and integrin attachment complexes to stabilize the sarcolemma . In agreement , muscle-specific overexpression of Drosophila Tsp or mouse Thbs4 rescues a Drosophila model of muscular dystrophy with augmented membrane residence of βPS integrin . This functional conservation emphasizes the fundamental importance of Thbs’ as regulators of cellular attachment and membrane stability and identifies Thbs4 as a potential therapeutic target for muscular dystrophy .
Muscle degenerative diseases such as muscular dystrophy ( MD ) are most commonly caused by mutations in genes that are part of the dystrophin-glycoprotein ( DGC ) complex or the integrin complex of proteins ( Grounds et al . , 2005; McNally and Pytel , 2007 ) . In addition , proper post-translational processing and trafficking of these complexes to the sarcolemma are essential to form a molecular attachment network between the myofilament proteins within the myofibers and the basal lamina and extracellular matrix ( ECM ) outside the cell ( Goddeeris et al . , 2013; Liu et al . , 2012; Xu et al . , 2009 ) . This attachment network provides critical structural support to the plasma membrane ( sarcolemma ) to withstand contractile forces ( Burr and Molkentin , 2015; Grounds et al . , 2005; Gumerson and Michele , 2011; Lapidos et al . , 2004 ) . When this attachment network is deficient in MD , membrane ruptures occur leading to intracellular calcium influx that causes myofiber necrosis , an inflammatory response , fibrosis and fatty tissue replacement , and ultimately muscle functional loss and death ( Burr and Molkentin , 2015; Gumerson and Michele , 2011; Lapidos et al . , 2004 ) . Skeletal muscle is perhaps the most sensitive of all tissues to genetic mutations in genes that impact cellular attachment complexes or membrane repair capacity , in part because of the dynamic changes in length that occurs in each myofiber during contraction ( Burr and Molkentin , 2015; Grounds et al . , 2005; McNally and Pytel , 2007 ) . Thrombospondins ( Thbs ) comprise a family of 5 genes in mammals that encode secreted matricellular proteins involved in diverse biologic processes ( Adams and Lawler , 2011 ; Schellings et al . , 2009 ) . The thrombospondin family consists of two subgroups based on their sequence conservation and oligomeric structure . Thbs3 , Thbs4 and Thbs5 form pentamers and are the most similar to Thbs genes found in lower organisms ( Adams and Lawler , 2011 ) . Thbs1 and Thbs2 form trimers and have evolved additional domains such as a type 1 repeat important for transforming growth factor-β binding and a region that affects angiogenesis ( Adams and Lawler , 2011; Schellings et al . , 2009 ) . Drosophila contains a single thrombospondin gene ( Tsp ) that forms pentamers , and when deficient causes developmental lethality due to disruption in muscle and tendon attachment within the body wall segments of the embryo ( Adams and Lawler , 2011; Subramanian et al . , 2007 ) . While traditionally characterized as a secreted ECM or matricellular protein over the past 3 decades , ( Adams and Lawler , 2011; Schellings et al . , 2009 ) Thbs can also function within the cell , and in some systems this appears to be their primary role ( Ambily et al . , 2014; Baek et al . , 2013; Brody et al . , 2016; Duquette et al . , 2014; Lynch et al . , 2012; McKeown-Longo et al . , 1984; Posey et al . , 2014 ) . For example , Thbs4 was recently shown to have a critical cardioprotective function from within the endoplasmic reticulum ( ER ) where it mediates an adaptive ER stress response ( Brody et al . , 2016; Lynch et al . , 2012 ) . The traditional ER stress response involves sensing of calcium and unfolded or damaged proteins within the ER through the calcium binding chaperone protein BiP ( GRP78 ) , which binds/regulates at least 3 distinct stress response pathways initiated by either PKR-like ER kinase ( PERK ) , inositol-requiring enzyme 1α ( IRE1α ) or activating transcription factor 6 ( ATF6 ) , each resident within the ER membrane ( Glembotski , 2007; Mori , 2009 ) . These 3 ER stress response mediators initiate a cascade of signaling that alters protein synthesis and other features of cellular adaptation to stress or protein unfolding and aggregation ( Glembotski , 2007; Mori , 2009 ) . Here , Thbs4 directly binds the ER luminal domain of ATF6α to promote its shuttling to the Golgi and then nucleus , thereby inducing genes underlying adaptive aspects of the ER stress response ( Brody et al . , 2016; Lynch et al . , 2012 ) . Thbs proteins move through the secretory pathway where they appear to facilitate secretion of ECM proteins or perhaps chaperone protein complexes to the cell membrane ( Adams and Lawler , 2011 ) . Once secreted , Thbs proteins transiently or permanently reside in the ECM where they interact with fibronectin , collagens and proteoglycans ( Adams and Lawler , 2011; Frolova et al . , 2014 , 2012; Hauser et al . , 1995; Södersten et al . , 2006 ) . Thbs proteins are also recycled back into the cell through the low-density receptor-related protein ( LRP ) ( Wang et al . , 2004 ) . One critical feature of the Thbs family is that each member is induced following injury events or in response to processes requiring tissue growth , healing and remodeling . Interestingly , Thbs4 is largely restricted to cardiac and skeletal muscle where its expression is induced with injury or disease ( Adams and Lawler , 2011; Chen et al . , 2000; Frolova et al . , 2014 , 2012; Hauser et al . , 1995; Lynch et al . , 2012; Schellings et al . , 2009; Södersten et al . , 2006 ) . In addition , markers of ER stress are upregulated during progression of skeletal muscle disease and MD ( Lavery et al . , 2008; Moorwood and Barton , 2014 ) . Here we observed that in response to MD in skeletal muscle , Thbs4 mRNA and protein are induced . Overexpression of Thbs4 in skeletal muscle of transgenic ( Tg ) mice protected against MD , while mice lacking Thbs4 ( Thbs4-/- ) showed signs of spontaneous MD with aging . Mechanistically , Thbs4 directs a membrane attachment intracellular vesicular trafficking network that promotes greater stability of the DGC and integrin complexes at the sarcolemma of skeletal muscle fibers . This function of Thbs is conserved in Drosophila as overexpression of either mouse Thbs4 or Drosophila Tsp in muscle rescues MD that occurs in Drosophila deficient in its δ-sarcoglycan-related gene ( Allikian et al . , 2007 ) .
In agreement with previous findings , Thbs4 RNA is induced in muscle biopsies from human patients with Becker MD , Duchenne MD , and limb-girdle MD ( LGMD ) ( Figure 1A; Figure 1—figure supplement 1A ) ( Chen et al . , 2000 ) . We next turned to 2 different mouse models of MD . One due to deletion of the δ-sarcoglycan ( Sgcd ) gene to model LGMD2F , and a second due to defective dystrophin expression resulting from the mdx mutation that models Duchenne MD ( Durbeej and Campbell , 2002 ) . Thbs4 protein is induced in skeletal muscle of each mouse model at six weeks and three months of age , along with induction of an ER stress response associated with greater cleaved ATF6α-N ( nuclear form ) and increased total BiP levels ( Figure 1B , C; Figure 1—figure supplement 1B , C ) . 10 . 7554/eLife . 17589 . 003Figure 1 . Thbs4 is induced in dystrophic skeletal muscle and its overexpression augments ER activity and vesicle content . ( A ) Thbs4 mRNA levels in human skeletal muscle biopsies from normal or patients with Becker MD ( BMD; n = 5 ) , Duchenne MD ( DMD; n = 10 ) or 2 different types of limb-girdle MD ( LGMD; n = 10 for both ) . *p<0 . 05 vs . normal ( n = 18 ) by Student’s t test . Data are presented as mean ± SEM . Full analysis including all 11 human muscle diseases is shown in Figure 1—figure supplement 1A . ( B , C ) Western blot for the expression of Thbs4 , ATF6α-N ( 50 kDa , nuclear ) and BiP in the quadriceps of WT , Sgcd-/- and mdx mice at six weeks ( w ) and three months ( mo ) of age ( n = 4 biological replicates ) . ( D ) Schematic diagram showing the skeletal muscle-specific transgene to overexpress Thbs4 and ( lower ) Western blots for Thbs4 or gapdh control from WT and Tg mice at 6 w of age from Quad , quadriceps; Gas , gastrocnemius; Sol , soleus; Diaph , diaphragm; and heart ( n = 2 biological replicates ) . ( E ) Upper micrographs represent co-immunofluorescent labeling of intracellular Thbs4 ( green ) with calreticulin ( red ) on paraffin embedded quadriceps ( Quad . ) of WT , Thbs4-Tg and Sgcd-/- mice at 3 mo of age ( scale bar = 20 μm ) . Arrowheads indicate co-localization of Thbs4 with calreticulin in intracellular vesicles in the myofibers . Lower micrographs represent co- immunofluorescent labeling of Thbs4 ( green ) with collagen I ( red ) on cryo-embedded Quad of WT , Thbs4-Tg and Sgcd-/- mice at 3 mo of age ( scale bar = 50 μm ) . Arrowheads indicate co-localization of Thbs4 with collagen I in the extracellular milieu; the star marks a myofiber with both intra- and extracellular Thbs4 labeling from a diseased muscle . Nuclei are visualized in blue . Representative images of 4 mice per genotype are shown . ( F ) Western blot analysis of Thbs4 and the ER-stress proteins ATF6α-N ( 50 kDa , nuclear ) , BiP , PDI , calreticulin ( Calret . ) , and Armet in 6w old WT , Thbs4-Tg and Sgcd-/- quadriceps ( n = 4 biological replicates ) . ( G ) Transmission electron microscopy in WT versus Thbs4 Tg quadriceps at 3 mo of age showing a massive expansion of intramyofibrillar and subsarcolemmal ER and associated vesicles with Thbs4 overexpression ( arrowheads , scale bar = 2 μm ) . Representative images of 2 mice per genotype are shown . ( H ) Immunogold electron microscopy shows that Thbs4 ( 6 nm gold particles; yellow arrows ) robustly localizes to the expanded sub-sarcolemmal vesicular compartment in Thbs4-Tg quadriceps , compared to endogenously expressed Thbs4 in WT quadriceps . Representative images of 2 mice are shown . Scale bar = 50 nmDOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 00310 . 7554/eLife . 17589 . 004Figure 1—figure supplement 1 . Thbs4 expression levels in human muscle diseases and quantitation of protein levels relative to Gapdh loading control of immunoblots shown in Figure 1B , C and F . ( A ) RelativeThbs4 mRNA levels in human skeletal muscle biopsies from normal , healthy subjects ( n = 18 ) or patients with acute quadriplegic myopathy ( AQM; n = 5 ) , juvenile dermatomyositis ( JDM; n 21 ) , amyotophic lateral sclerosis ( ALS; n = 9 ) , spastic paraplegia ( SPG4; n = 4 ) , fascioscapulohumeral muscular dystrophy ( FSHD; n = 14 ) , Emery Dreifuss muscular dystrophy ( EDMD; n = 7 ) , Becker muscular dystrophy ( BMD; n 5 ) , Duchenne muscular dystrophy ( DMD; n = 10 ) , calpain 3 mutation ( LGMD2A; n = 10 ) , dysferlin ( LGMD2B; n = 10 ) , FKRP ( Homozygous for a missense mutation Fukutin-related protein; n = 7 ) . *p<0 . 05 vs . Healthy by Student’s t test . ( B ) Relative protein levels for immunoblots shown in Figure 1B; ( C ) Relative protein levels for Figure 1C; and ( D ) relative protein levels for Figure 1F . An additional set of immunoblots was included to obtain an n = 4 per genotype . Relative Thbs4 protein levels were determined by quantifying both bands observed for Thbs4 . *p<0 . 05 vs WT; #p<0 . 05 vs Thbs4 Tg by one-way ANOVA with post hoc Tukey’s test . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 00410 . 7554/eLife . 17589 . 005Figure 1—figure supplement 2 . Analysis of Thbs4 glycosylation and vesicular expansion quantitation . ( A ) Glycosylation pattern of Thbs4 protein from quadriceps protein extracts of Thbs4-Tg mice treated with enzymes endoglycosidase H ( Endo H ) and peptide N-glycosidase F ( PNGase F ) and compared with a total deglycosylation enzyme mix ( Deglycos . ) . Gapdh is presented as processing and loading control ( n = 4 biological replicates ) . Endo H cleaves high mannose residues and hybrid oligosaccharides present on proteins in the ER while PNGase cleaves both these and more complex oligosaccharides that result from processing in the Golgi . Here , both enzymes equally reduced the apparent size of Thbs4 , consistent with high mannose residues typical for proteins in the ER lumen , and proteins that are transitioning to the Golgi to be secreted . ( B ) Representative Western blot analysis for Thbs4 from extracellular matrix ( ECM ) enriched protein fractions from WT and Thbs4-Tg quadriceps . Laminin is presented as ECM protein control; Gapdh as cytosolic contamination control ( n = 3 biological replicates ) . The data show that whileoverexpressed Thbs4 in non-diseased muscle cannot be readily observed by immunohistochemistry from muscle sections , a more sensitive analysis with ECM-specific western blotting does show some Thbs4 protein outside the myofibers . ( C ) Micrographs representing co-immunofluorescent labeling of Thbs4 ( green ) with periostin ( red ) on cryo-embedded Quad of WT , Thbs4-Tg and Sgcd-/- mice at three months of age ( scale bar = 50 μm ) . Nuclei are visualized in blue . Arrowheads indicate co-localization of Thbs4 with periostin in the extracellular milieu during disease; the star marks a myofiber with both intra- and extracellular Thbs4 labeling . Representative images of 4 mice per genotype are shown . ( D ) Quantitation of sub-sarcolemmal vesicular expansion relative to the length of the sarcolemma on transmission electron microscopy images of WT and Thbs4-Tg myofibers from quadriceps at approximately three months of age ( n = 22 and 17 myofibers , respectively , from 2 mice per genotype ) . Data are presented as mean ± SEM; *p<0 . 0001 vs WT by Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 00510 . 7554/eLife . 17589 . 006Figure 1—figure supplement 3 . Internalization of Thbs4 by cultured C2C12 myoblasts and myotubes . ( A ) Representative images of C2C12 myoblasts cultured with 1 μg/ml Alexa-488 conjugated recombinant Thbs4 ( rThbs4-488 ) or BSA-488 control for the indicated periods . Bright green vesicles containing internalized rThbs4-488 , but not BSA-488 were detected in the cytoplasm of C2C12 myoblasts . Cells were counterstained with phalloidin-568 ( red ) ; nuclei are visualized in blue . Scale bar = 10 μm . ( B ) Representative images of C2C12 myoblasts cultured with 1 μg/ml rThbs4-488 or BSA-488 control for 24 hr showing partial co-localization of Thbs4 with a late endosome marker ( Rab7 ) but not BSA-488 control . Phalloidin ( red ) was used to visualize cells; nuclei are visualized in blue . Scale bar = 10 μm . ( C ) Representative images of C2C12 myotubes cultured with 1 μg/ml rThbs4-488 or BSA-488 control for 24 hr . The regions marked by dotted boxes are shown on the right in higher magnification . Bright green vesicles containing internalized rThbs4-488 , but not BSA-488 were detected in the cytoplasm of C2C12 myotubes after recombinant Thbs4-488 was added to the media for 24 hr . Phalloidin ( red ) was used to visualize cells; nuclei are visualized in blue . Scale bar = 25 μm . ( D , E ) Representative immunoblots and semi-quantitative analysis of rThbs4 internalization after addition of 1 μg/ml biotin labeled rThbs4 to the cultured media of C2C12 myotubes for indicated times . n = 3 experiments for all experiments presented in this figure . *p<0 . 05 vs untreated myotubes by Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 006 To model the known increase in Thbs4 protein that occurs in dystrophic skeletal muscle we generated Tg mice with Thbs4 protein overexpression specific to skeletal muscle ( Figure 1D ) . High levels of Thbs4 protein overexpression were observed in fast-twitch containing muscles such as the quadriceps , gastrocnemius , with intermediate levels in the diaphragm and very low levels in the soleus , while the heart lacked expression ( Figure 1D ) . Tissue Immunofluorescent analysis revealed that Thbs4 protein was undetectable in uninjured skeletal muscle while the transgene produced abundant expression that co-localized with calreticulin to a vesicular network on the periphery of the myofibers of paraffin-embedded quadriceps and was also clearly inside of collagen I staining that marks the ECM of cryo-embedded quadriceps ( Figure 1E ) . Furthermore , although Thbs4 protein localization appeared slightly different between paraffin- and cryo-embedded skeletal muscle of the Sgcd-/- mouse , induction of endogenous Thbs4 again showed localization within the vesicular network inside the myofibers , and only within limited regions outside of myofibers where fibrotic tissue deposition was prominent ( Figure 1E ) . In agreement with the above observations of Thbs subcellular localization , biochemical analysis revealed that Thbs4 was high in the type of glycosylation that typifies ER resident proteins , but it also contains glycosylation that is observed on proteins that transit through the Golgi in route to be deposited in the ECM ( Figure 1—figure supplement 2A , B ) . Interestingly , in-depth in vitro analysis demonstrated that extracellular Thbs4 is rapidly internalized by both cultured C2C12 myoblasts and myotubes and at least in the case of myoblasts is transported to rab7-positive late endosomes , potentially explaining why Thbs4 protein is low or undectable in the ECM of healthy Thbs4 Tg muscles ( Figure 1—figure supplement 3 ) . Similar to the results observed with collagen I , co-labeling with another ECM/matricellular protein periostin again showed that Thbs4 could co-localize to the ECM region in Sgcd-/- diseased myofibers , although under non-diseased conditions overexpressed Thbs4 was again only appreciably observed intracellularly under the sarcolemma within a peripheral vesicular pattern ( Figure 1E; Figure 1—figure supplement 2C ) . Hence , although our observations do not exclude the possibility that non-muscle cells might also express Thbs4 , our data collectively identify the myofiber as an important cellular source of Thbs4 expression , secretion and re-uptake . Careful analysis of other markers of the ER compartment and ER stress showed that Thbs4 overexpression in skeletal muscle induced a profile very similar to diseased skeletal muscle in Sgcd-/- mice , with increased levels of nuclear ATF6α , BiP , protein disulfide isomerase ( PDI ) , calreticulin and Armet , as compared to WT muscle ( Figure 1F; Figure 1—figure supplement 1D ) . Remarkably , transmission electron microscopy and immunogold detection revealed that Thbs4 overexpression in skeletal muscle caused a dramatic induction of sub-sarcolemmal and intramyofibrillar ER and post-ER vesicles that contained Thbs4 protein ( Figure 1G , H; Figure 1—figure supplement 2D ) . These Thbs4-dependent vesicles were highly uniform in size and more electron dense compared with similar vesicles in subsarcolemmal regions from WT muscle . Future studies will investigate the nature of these Thbs4-expanded vesicles and their composition based on known variables ( Malhotra and Erlmann , 2015; Paczkowski et al . , 2015 ) . To determine if Thbs4 induction in MD was adaptive or maladaptive we first crossed the Thbs4 Tg into both the Sgcd-/- and mdx backgrounds . Importantly , Thbs4 overexpression itself in skeletal muscle caused no histopathology or functional defects compared to WT mice at three and 12 months of age ( Figure 2A–E; Figure 2—figure supplement 1A–F ) . More importantly , Thbs4 overexpression significantly reduced multiple histopathological hallmarks of dystrophic disease , including elevated serum creatine kinase ( CK ) levels , reduced myofiber degeneration/regeneration cycles as marked by reduced centrally nucleated myofibers , reduced fibrotic remodeling and less functional decline in skeletal muscle at both three and 12 months of age in both Sgcd-/- and mdx mice , compared with each dystrophic model alone ( Figure 2A–E; Figure 2—figure supplement 1A–F , Figure 2—figure supplement 2A–E , Figure 2—figure supplement 3A , B ) . 10 . 7554/eLife . 17589 . 007Figure 2 . Thbs4 overexpression in skeletal muscle mitigates MD in mice . ( A ) Masson’s trichrome stained sections of quadriceps ( Quad . ) and diaphragm ( Diaph . ) from WT , Thbs4-Tg , Sgcd-/- and Sgcd-/- Thbs4-Tg mice at 3 mo of age ( Scale bar = 100 μm ) . Blue staining is fibrosis . ( B ) Quantitation of serum CK levels ( units/liter ) in indicated genotypes at 3 mo of age . *p<0 . 05 vs WT; #p<0 . 05 vs Sgcd-/- by one-way ANOVA with post hoc Tukey’s test . N = 8 mice for WT , Thbs4 Tg and Sgcd-/-Thbs4 Tg and n = 7 mice for Sgcd-/- . The legend in this panel also refers to panels C–F . ( C ) Percentage of myofibers with centrally located nuclei in Quad and Diaph from H&E-stained histological sections at 3 mo of age . Representative images are shown in Figure 2—figure supplement 3 . *p<0 . 05 vs WT; #p<0 . 05 vs Sgcd-/- by one-way ANOVA with post hoc Tukey’s test . ( D ) Interstitial fibrosis analyzed in trichrome stained histological sections from Quad , gastrocnemius ( Gas ) and Diaph at three months of age . *p<0 . 05 vs WT; #p<0 . 05 vs Sgcd-/- by one-way ANOVA with post hoc Tukey’s test . N = 6 mice for WT and Thbs4 Tg and n = 8 mice for Sgcd-/-and Sgcd-/-Thbs4 Tg in panel A , C and D . ( E ) Time to fatigue in seconds with forced downhill treadmill running at 3 mo of age in the indicated genotypes of mice . *p<0 . 05 vs WT; #p<0 . 05 vs Sgcd-/- by one-way ANOVA with post hoc Tukey’s test . N = 6 mice per genotype . ( F , G ) Quantitation of total Evan’s blue dye ( EBD ) positive fibers and representative immunofluorescent images of EBD uptake ( red ) in Quad of three month-old mice . Membranes of myofibers are shown in green . *p<0 . 05 vs WT; #p<0 . 05 vs Sgcd-/- by one-way ANOVA with post hoc Tukey’s test . N = 5 mice per genotype in panel F . Scale bar = 100 μm . ( H ) Transmission electron microscopy of quadriceps muscle in Sgcd-/-Thbs4 Tg mice compared to Sgcd-/- mice at 3 mo of age . The arrowheads show subsarcolemmal vesicular expansion due to the Thbs4 Tg . Representative images of 2 mice per genotype are shown . Scale bar = 500 nm . All data are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 00710 . 7554/eLife . 17589 . 008Figure 2—figure supplement 1 . Thbs4 overexpression mitigates muscular dystrophy in Sgcd-/- and mdx mice with aging to one year . ( A ) Masson’s trichrome-stained histological sections of quadriceps ( Quad . ) and diaphragm ( Diaph . ) from WT , Thbs4-Tg , Sgcd-/- and Sgcd-/-Thbs4-Tg mice at one year of age . Blue color represents fibrosis . Representative images of 6 mice per genotype are shown . Scale bars = 100 μm . ( B , C ) Quantitation of serum creatine kinase ( CK ) levels ( units/liter ) and time to fatigue in seconds with forced downhill treadmill running for the indicated genotypes at one year of age . For panel B , n = 9 mice for WT , n = 8 mice for Thbs4 Tg , n = 6 mice for Sgcd-/- and n = 7 mice for Sgcd-/-Thbs4 Tg; and for panel C , n = 5 mice for WT and Thbs4 Tg , and n = 6 mice for Sgcd-/- and Sgcd-/-Thbs4 Tg . Data are presented as mean ± SEM; *p<0 . 05 vs WT; #p<0 . 05 vs . Sgcd-/- by one-way ANOVA with post hoc Tukey’s test . ( D ) Masson’s trichrome stained histological sections of quadriceps ( Quad . ) and diaphragm ( Diaph . ) from WT , Thbs4-Tg , mdx and mdx Thbs4-Tg mice at one year of age . Blue color represents fibrosis . Representative images of 6 mice per genotype are shown . Scale bars = 100 μm . ( E , F ) Quantitation of serum CK levels ( units/liter ) and time to fatigue in seconds with forced downhill treadmill running for the indicated genotypes at one year of age . For panel E , n = 9 mice for WT , n = 8 mice for Thbs4 Tg , n = 10 mice for mdx and n = 6 mice for mdx Thbs4 Tg . For panel F , n = 5 mice for WT and Thbs4 Tg , n = 7 mice for mdx and n = 6 mice for mdx Thbs4 Tg . Data are presented as mean ± SEM; *p<0 . 05 vs WT; #p<0 . 05 vs . mdx by one-way ANOVA with post hoc Tukey’s test . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 00810 . 7554/eLife . 17589 . 009Figure 2—figure supplement 2 . Thbs4 overexpression mitigates MD in mdx mice at three months of age . ( A ) Masson’s trichrome histological staining of mdx and mdx Thbs4 Tg quadriceps at three months of age . Blue areas represent fibrosis . Representative images of 6 mice for mdx and 7 mice for mdx Thbs4 Tg are shown . Scale bars = 100 μm . ( B ) Percentage of myofibers with centrally located nuclei in histological sections from quadriceps ( Quad ) and diaphragm ( Diaph ) in the indicated groups of mice at three months of age . Representative images are shown in Figure 2—figure supplement 3 . N = 6 mice for WT and Thbs4 Tg , n = 7 mice for mdx and n = 8 mdx Thbs4 Tg . *p<0 . 05 vs WT; #p<0 . 05 vs . mdx by one-way ANOVA with post hoc Tukey’s test . Data are presented as mean ± SEM . ( C ) Interstitial fibrosis analyzed in Quad and Diaph from histological sections at three months of age in the indicated groups of mice . N = 7 mice for WT and mdx Thbs4 Tg , n = 6 mice for Thbs4 Tg and mdx . *p<0 . 05 vs WT; #p<0 . 05 vs . mdx by one-way ANOVA with post hoc Tukey’s test . Data are presented as mean ± SEM . ( D ) Quantitation of serum CK levels ( units/liter ) in the indicated genotypes of mice at three months of age . n = 7 mice for mdx and n = 6 mdx Thbs4 Tg . *p<0 . 05 vs WT; #p<0 . 05 vs . mdx by one-way ANOVA with post hoc Tukey’s test . Data are presented as mean ± SEM . ( E ) Time to fatigue in seconds with forced downhill treadmill running at three months of age in the indicated genotypes of mice . N = 6 for indicated genotypes . *p<0 . 05 vs WT; #p<0 . 05 vs . mdx by one-way ANOVA with post hoc Tukey’s test . Data are presented as mean ± SEM . ( F , G ) Representative immunofluorescent images and quantitation of total Evan’s blue dye ( EBD ) uptake and positive fibers ( red ) in Quad in three month-old mice subjected to downhill running for 30 min during a 24 hr period of time when EBD is circulating in the mouse . Membranes of myofibers are shown in green . Scale bars = 75 μm . n = 6 for indicated genotypes . *p<0 . 05 vs WT; #p<0 . 05 vs . mdx by one-way ANOVA with post hoc Tukey’s test . Data are presented as mean ± SEM . ( H ) Transmission electron microscopy of quadriceps showing preservation of muscle architecture with a dramatic sub-sarcolemmal accumulation of ER and vesicles ( arrows ) in mdx Thbs4-Tg mice , compared with noticeable disease and few vesicles in mdx . Representative images of 2 mice per genotype are shown . Scale bar: 500 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 00910 . 7554/eLife . 17589 . 010Figure 2—figure supplement 3 . In depth imaging of H&E sections showing how Thbs4 overexpression reduces myofiber degeneration-regeneration ( central nucleation ) in both Sgcd-/-and mdx dystrophic quadriceps . ( A , B ) representative images and higher magnification inset from H&E-stained histology from quadriceps at three months of age from the indicated genotypes ( Scale bar = 100 μm and 50 μm , respectively ) . Note the significant reduction in myofiber degeneration-regeneration cycles marked by reduced centrally located nuclei upon overexpression of Thbs4 in both Sgcd-/- and mdx quadriceps . Quantification is shown in Figure 2C and Figure 2—figure supplement 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 01010 . 7554/eLife . 17589 . 011Figure 2—figure supplement 4 . Intramuscular AAV9-mediated overexpression of Thbs4 mitigates MD in Sgcd-/- mice . ( A ) Separate three day-old Sgcd-/-mouse pups were injected in the gastrocnemius ( Gastroc . ) with 1E10 viral particles of either adeno-associated virus 9 ( AAV9 ) -Thbs4 or AAV9-eGFP control . The gastroc . was then harvested at six weeks of age for further analysis . ( B ) Representative Western blots showing robust overexpression of eGFP or Thbs4 ( Gapdh as loading control , n = 3 biological replicates ) in Sgcd-/-gastrocnemius treated with AAV9-eGFP or AAV9-Thbs4 , respectively . ( C ) Representative H&E and Masson’s trichrome-stained histological sections from Sgcd-/- gastrocnemius ( Gastroc . ) treated with control AAV9-eGFP or experimental AAV9-Thbs4 at six weeks of age ( separate mice ) . Representative images of 5 mice per genotype are shown . Scale bar = 100 μm . ( D , E ) Percent myofiber central nucleation and fibrosis from histological sections of the gastrocnemius taken from Sgcd-/- mice six weeks after AAV9-Thbs4 or control AAV9-eGFP treatment . n = 5 mice per treatment group . Data are represented as mean ± SEM; *p<0 . 05 vs . eGFP by Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 011 The sarcolemma of dystrophic myofibers are weak and frequently rupture , which can be assessed in vivo by Evans blue dye ( EBD ) uptake into muscle fibers after systemic injection ( Goonasekera et al . , 2011; Lapidos et al . , 2004 ) . Here , the percentage of myofibers with ruptured membranes after forced treadmill running was significantly reduced in Sgcd-/- and mdx mice that contained the Thbs4 Tg , compared with Sgcd-/- and mdx mice alone , while no EBD uptake was observed in WT or Tg muscle ( Figure 2F , G; Figure 2—figure supplement 2F , G ) . Ultrastructural analysis again showed that Thbs4 overexpression resulted in a dramatic induction in sub-sarcolemmal and intramyofibrillar vesicles in skeletal muscle of both Sgcd-/- and mdx mice ( Figure 2H; Figure 2—figure supplement 2H ) . This dramatic protection from MD observed in 2 mouse models of this disease with Tg-mediated overexpression of Thbs4 led us to investigate whether overexpression of Thbs4 would be sufficient to reduce acute dystrophic disease for the first time using a gene therapy-related approach to bypass possible developmental effects of overexpression . Hence , here we performed a study in Sgcd-/- mice with an adeno-associated virus serotype-9 ( AAV9 ) -Thbs4 vector , which was injected into the gastrocnemius of three day-old neonates , followed by harvesting at six weeks of age to assess histopathology ( Figure 2—figure supplement 4A ) . Littermates injected with an eGFP expressing AAV9 were used as a control . In agreement with our previous findings , this approach also resulted in abundant expression of either the control eGFP protein or a 5-fold increase in Thbs4 in the muscle of these mice ( n = 3 mice per group , p<0 . 05 by Student’s t test; Figure 2—figure supplement 4B ) . The results showed that AAV9-mediated Thbs4 overexpression significantly reduced central nucleation and fibrotic remodeling in Sgcd-/- mice with ongoing dystrophic disease ( Figure 2—figure supplement 4C–E ) . Hence , overexpression of Thbs4 is protective and mitigates dystrophic disease even if initiated later in postnatal life . The induction of Thbs4 that normally occurs in skeletal muscle with dystrophic disease was further shown to be an adaptive and physiologic mechanism through the analysis of mice lacking the Thbs4 gene . Here we crossed Sgcd-/- and mdx mice into the Thbs4 null background and performed a full analysis of pathogenesis at three months of age . Thbs4-/- mice alone at three months of age showed minimal or no pathological changes in skeletal muscle , although combinatorial Thbs4-/- Sgcd-/- or Thbs4-/- mdx mice showed a significant worsening of MD , including greater histopathology , greater serum CK levels and reduced treadmill running performance , compared with single null Sgcd and mdx mice ( Figure 3A–C; Figure 3—figure supplement 1A–C ) . Sgcd-/- Thbs4-/- mice also showed significantly greater EBD uptake in skeletal muscle after forced treadmill running versus Sgcd-/- mice alone ( Figure 3D , E ) . 10 . 7554/eLife . 17589 . 012Figure 3 . Loss of Thbs4 induces and exacerbates MD in mice . ( A ) Masson’s trichrome stained histological sections of Quad and Diaph at three months ( mo ) of age in Thbs4-/- , Sgcd-/- and Sgcd-/-Thbs4-/- mice . Representative images of 6 mice for WT and Thbs4 Tg mice and 8 mice for Sgcd-/-and Sgcd-/-Thbs4 Tg . Scale bar = 100 μm . ( B ) Quantitation of serum CK ( units/liter ) in the indicated genotypes shown in the legend , at 3 mo of age . *p<0 . 05 vs . WT; #p<0 . 05 vs Sgcd-/- by one-way ANOVA with post hoc Tukey’s test . N = 5 mice for WT and n = 6 mice for Thbs4-/- , Sgcd-/- and Sgcd-/-Thbs4-/- . ( C ) Time to fatigue in seconds with forced downhill treadmill running in mice at three months of age . *p<0 . 05 vs . WT; #p<0 . 05 vs . Sgcd-/- by one-way ANOVA with post hoc Tukey’s test . N = 5 mice per genotype . The legend above panels B and C applies to the remainder of the figure . ( D , E ) Representative immunofluorescent images of EBD ( red fluorescence ) uptake and quantitation in Quad histological sections from three month-old mice . Membranes of myofibers are shown in green . *p<0 . 05 vs WT; #p<0 . 05 vs . Sgcd-/- by one-way ANOVA with post hoc Tukey’s test . N = 5 and 6 mice for Sgcd-/- and Sgcd-/-Thbs4-/- , respectively . Scale bar = 40 μm . ( F , G ) Time to fatigue in seconds with forced downhill treadmill running and quantitation of serum CK levels in WT and Thbs4-/- mice at the indicated ages; abbreviations , y = year . n = 6 mice per genotype per age for panel F . For panel G , n = 7 mice per genotype at six weeks of age; n = 5 WT and 6 Thbs4-/-mice at three months of age , n = 5 mice per genotype at six months of age and n = 9 WT and 8 Thbs4-/-mice at one year of age . *p<0 . 05 vs WT at the same age by Student’s t test . ( H ) H&E histological staining ( upper ) and transmission electron microscopy ( lower ) of tissue pathology in Quad at one year of age in WT and Thbs4-/-mice . The H&E scale bar = 25 μm . The electron microscopy scale bar = 2 μm . The arrows show myofibers with central nucleation due to loss of the Thbs4 gene . Representative images of 6 mice per genotype for H&E staining and 2 mice per genotype for electron microscopy . ( I ) Percentage of myofibers with centrally located nuclei in Thbs4-/- compared to WT Quad at one year of age . *p<0 . 001 vs WT by Student’s t test . n = 6 mice per genotype . ( J ) Masson’s trichrome stained ( upper ) and EBD uptake ( lower ) histological images in WT and Thbs4-/-mice in the Diaph at one year of age . The EBD images show membranes in green and fibers with EBD uptake produce red fluorescence . Representative images of 6 mice per genotype studied . Scale bars = 50 μm . All data are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 01210 . 7554/eLife . 17589 . 013Figure 3—figure supplement 1 . Thbs4-/- mice show enhanced MD pathology in the mdx background . ( A ) Masson’s trichrome-stained histological sections of quadriceps ( Quad ) and diaphragm ( Diaph ) from mdx and mdx Thbs4-/- mice at three months of age . Blue staining shows fibrosis . Representative images of 6 mice per genotype are shown . Scale bars = 100 μm . ( B ) Time to fatigue in seconds with forced downhill treadmill running of the indicated genotypes of mice at three months of age . n = 6 mice per genotype . *p<0 . 05 vs WT and Thbs4-/-; #p<0 . 05 vs . mdx by one-way ANOVA with post hoc Tukey’s test . Data are presented as mean ± SEM . ( C ) Quantitation of serum CK ( units/liter ) levels in the indicated genotypes of mice at three months of age . n = 9 mice for WT and n = 6 mice for Thbs4-/- , mdx , and mdx Thbs4-/- . *p<0 . 05 vs WT and Thbs4-/-; #p<0 . 05 vs . mdx by one-way ANOVA with post hoc Tukey’s test . Data are presented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 013 The observation that loss of the Thbs4 gene makes dystrophic pathology significantly worse in both Sgcd-/- and mdx mice suggests that induction of this gene product plays an important protective role , and we reasoned that with aging loss of Thbs4 might eventually become pathologic to muscle given that low levels of continuous expression are present . Indeed , we observed that by six months and one year of age , Thbs4-/- mice showed a significant reduction in treadmill running capacity and increased serum CK levels compared to WT muscle ( Figure 3F , G ) . Furthermore , by one year of age Thbs4-/- muscle had increased signs of ongoing myofiber degeneration/regeneration , as marked by centrally nucleated myofibers , as well as noticeable histopathological and ultrastructural changes and greater EBD uptake compared to WT control muscle ( Figure 3H–J ) . Collectively , these results suggest that Thbs4 induction with dystrophic disease and low levels of expression during aging produce an adaptive physiologic response that protects skeletal muscle . To directly evaluate the structural integrity of the sarcolemma we first employed a model of 3 successive lengthening-contraction injury cycles to the tibialis anterior ( TA ) muscle in a whole leg immobilization preparation ( Figure 4A ) . Remarkably , overexpression of Thbs4 significantly protected against lengthening-contraction injuries over all 3 bouts compared to WT mice ( Figure 4B ) . As anticipated , the TA from Sgcd-/- mice showed much greater loss of functional recovery compared to WT after lengthening-contraction injury , but the presence of the Thbs4 Tg provided significant protection , achieving a recovery response now similar to WT levels ( Figure 4B ) . Moreover , loss of Thbs4 resulted in greater injury with all 3 cycles of lengthening-contraction injury , similar to Sgcd-/- ( Figure 4C ) . Interestingly , the passive force of the muscle after simple stretching was greater with the Thbs4 Tg , while it was reduced in Thbs4-/- muscle , although the twitch force itself was not significantly different between any of the groups ( Figure 4—figure supplement 1A , B ) . We also noticed that the tendons were weaker in Thbs4-/- TA muscle , which were more likely to rupture in the isolated lengthening-contraction injury assay ( Figure 4—figure supplement 1C ) . 10 . 7554/eLife . 17589 . 014Figure 4 . Thbs4 regulates skeletal muscle sarcolemma stability . ( A ) Schematic representing the first of 3 consecutive lengthening contraction-induced muscle injury cycles using an in situ tibialis anterior ( TA ) muscle preparation . Briefly , an isometric contraction was performed to determine baseline force generation , followed by 2 consecutive eccentric contractions and finally another isometric contraction ( = injury cycle 1 ) . The force deficit shown in panels B and C was calculated between the first and second isometric contraction in between the two lengthening-contractions , which was repeated 3 cycles total . ( B , C ) Reduction in isometric force generation as a percentage of baseline force after each lengthening contraction injury cycle in the indicated genotypes of mice shown in the legend . *p<0 . 05 vs WT; #p<0 . 05 vs WT and Thbs4-Tg; §P<0 . 05 vs Sgcd-/-and Thbs4-Tg by one-way ANOVA with post hoc Tukey’s test . n = 6 mice for WT , Thbs4 Tg and Sgcd-/-Thbs4 Tg and n = 10 mice for Sgcd-/- for panel B . n = 6 mice for WT , n = 10 mice for Sgcd-/- and n = 5 mice for Thbs4-/- for panel C . ( D ) Representative images before and after laser injury and influx of FM1-43 dye ( green fluorescence ) in FDB myofibers in the presence of 1 . 25 mM Ca2+ isolated from indicated genotypes . The white tag in each image is the position for the laser injury . Scale bars = 10 μm . ( E , F ) Quantitative time course in seconds of FM1-43 fluorescent dye entry in FDB myofibers from the indicated genotypes of mice in the presence of 1 . 25 mM Ca2+ . Laser injury occurred at 60 s . n = 6 fibers per animal from 3 animals per genotype for panels D–F; *p<0 . 05 vs WT; #p<0 . 05 vs Sgcd-/-; §p<0 . 05 vs Thbs4-/- by one-way ANOVA with post hoc Tukey’s test . Data points for Sgcd-/-in panel E and F were derived from a single set of experiments . All data are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 01410 . 7554/eLife . 17589 . 015Figure 4—figure supplement 1 . Thbs4 alters the mechanical and structural properties of muscle and tendons . ( A ) Passive tibialis anterior ( TA ) muscle elasticity measured prior to the lengthening contraction protocol in WT ( n = 6 ) , Thbs4 Tg ( n = 6 ) and Thbs4-/-mice ( n = 12 ) . *p<0 . 05 compared WT by Student’s t test . Data are presented as mean ± SEM . ( B ) Specific isometric force generation averaged over 5 contractions at L0 from in situ TA preparations for the indicated genotypes . n = 6 mice for WT , Thbs4 Tg , and Sgcd-/- Thbs4 Tg , n = 10 mice for Sgcd-/- , n = 5 for Thbs4-/- and n = 8 for Sgcd-/-Thbs4-/- . p-value is not significant by one-way ANOVA with post hoc Tukey’s test . Data are presented as mean ± SEM . ( C ) Percentage of tendon breaks recorded during the during the lengthening contraction protocol in WT , Thbs4-/-and Sgcd-/-Thbs4-/- mice . Tendon breaks were assessed by complete physical rupture of the tendon at the muscle aponeurosis . n = 6 for WT , n = 10 for Sgcd-/- , and n = 12 for Thbs4-/- and for Sgcd-/-Thbs4-/- . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 015 Individual myofibers were isolated from the flexor digitalis brevis ( FDB ) muscle and subjected to laser injury with subsequent measurement of FM1-43 fluorescent dye uptake as a direct measure of membrane stability and repair ( Figure 4D ) . FDB myofibers from Thbs4 Tg mice showed less dye uptake compared with WT myofibers , suggesting that Thbs4 overexpression was inherently protective to the membrane ( Figure 4D , E ) . As expected , FDB myofibers from Sgcd-/- mice showed greater dye uptake suggesting greater injury with less efficient repair , while the presence of the Thbs4 Tg was protective in Sgcd-/-myofibers , bringing dye uptake levels back to WT ( Figure 4D , E ) . FDB myofibers from Thbs4-/- mice also showed greater dye uptake compared with WT myofibers after laser injury , while double Thbs4-/- Sgcd-/- myofibers displayed an even greater injury response than either Thbs4 or Sgcd single deletions alone ( Figure 4F ) . Taken together these results indicate that Thbs4 overexpression provides greater stability to the sarcolemma , while its loss renders the sarcolemma less stable . More importantly , since this assay uses isolated myofibers devoid of ECM attachments , it suggests that part of Thbs4-dependent protection occurs from within the myofiber . To investigate a molecular mechanism whereby Thbs4 might directly regulate the stability of the sarcolemma in skeletal muscle we examined intracellular vesicular trafficking and membrane attachment complex formation . Two important clues that directed our investigation were the known augmentation in the adaptive ER stress response pathway and the dramatic increase in sub-sarcolemmal vesicles observed in skeletal muscle with Thbs4 overexpression . Hence , we instituted two in vitro assays to assess ER-to-Golgi and Golgi-to-sarcolemma vesicular trafficking in response to Thbs4 activity in primary neonatal rat ventricular myocytes ( Figure 5; Figure 5—figure supplement 1A , B , technical issues prevented such studies in myotubes or myofibers ) . To assess ER-to-Golgi trafficking a red fluorescent protein ( RFP ) -labeled Golgi resident enzyme , GalNacT2-RFP , was used with and without Thbs4 overexpression by fluorescent recovery after photobleaching ( FRAP , Figure 5—figure supplement 1A ) . In parallel , ATF6α activity was also modulated since Thbs4 is known to directly regulate this ER-stress transcription factor ( Brody et al . , 2016; Lynch et al . , 2012 ) . Here , Thbs4 overexpression significantly accelerated ER-to-Golgi vesicular trafficking , which was fully inhibited by co-overexpression of a dominant negative ( dn ) ATF6α construct ( Figure 5A ) . Similarly , Golgi-to-sarcolemmal trafficking rates , measured with VSVG-enhanced green floursecent protein ( eGFP ) after inverse ( i ) FRAP , were significantly accelerated upon Thbs4 overexpression , which was again inhibited with ATF6α-dn ( Figure 5B; Figure 5—figure supplement 1B ) . Furthermore , accelerated trafficking with Thbs4 was mimicked by overexpression of a constitutively nuclear ( cn ) ATF6α construct ( Figure 5C , D ) . 10 . 7554/eLife . 17589 . 016Figure 5 . Thbs4 enhances intracellular vesicular trafficking through its ATF6α interacting region . ( A–D ) Time course of GalNac-T2-RFP or VSVG-eGFP fluorescence changes in cultured neonatal ventricular myocytes infected with the indicated adenoviruses to overexpress Thbs4 ( maroon line , n = 18 ) , a dominant negative ( dn ) ATF6α ( turquoise line , n = 8 ) or in combination with Thbs4 ( purple line , n = 13 ) , a constitutively nuclear ( cn ) ATF6α ( green line , n = 12 ) , or βgal expressing control ( black line , n = 14 ) . Change in fluorescence was after FRAP or iFRAP to measure ER-to-Golgi or Golgi to the membrane trafficking , respectively . ( E , G , I ) Quantitative time course of GalNac-T2-RFP recovery in the Golgi network after FRAP to measure ER to Golgi vesicular trafficking in primary neonatal rat ventricular myocytes infected with adenoviral Thbs4 ( green line , n = 18 ) , Nell2 ( red line , n = 9 ) , the Type III repeat ( AdT3R; red line , n = 8 ) domain of Thbs4 , the N-terminal Laminin G ( AdLamG , red line , n = 7 ) domain of Thbs4 , or βgal control ( blue line , n = 14 ) . Shown above these graphs is a schematic diagram depicting the domain structure of Thbs4 , Nell2 and T3R or LamG of Thbs4 . All data are represented as mean ± SEM . ( F , H , J ) Quantitative time course of loss of VSVG-eGFP fluorescence in the Golgi after iFRAP as a measurement for Golgi-to-membrane ( Golgi exit ) vesicular trafficking in primary neonatal rat ventricular myocytes infected with adenoviral Thbs4 ( green line , n = 18 ) , Nell2 ( red line , n = 8 ) , the Type III repeat ( AdT3R; red line , n = 8 ) domain of Thbs4 , the N-terminal Laminin G ( AdLamG; red line , n = 7 ) , or βgal control ( blue line , n = 14 ) . All data are represented as mean ± SEM . Each cell imaged per experimental condition represents an independent experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 01610 . 7554/eLife . 17589 . 017Figure 5—figure supplement 1 . Thbs4 enhances intracellular vesicular trafficking . ( A ) Representative images showing the recovery of GalNacT2-RFP fluorescence ( red ) into the photobleached Golgi network ( FRAP; outlined green area ) over time in primary neonatal rat ventricular myocytes infected with an adenovirus expressing Thbs4 ( AdThbs4 ) or a βgal control . Recovery of Golgi fluorescence was monitored as a measure for ER-to-Golgi vesicular trafficking . Scale bar = 10 μm . ( B ) Representative images of VSVG-eGFP fluorescence ( green ) before and after inverse FRAP ( iFRAP , outlined area , red ) in primary neonatal rat ventricular myocytes . Adenovirus expressing Thbs4 or βgal ( control ) were used . Reduction in Golgi fluorescence ( outlined in yellow ) was monitored as a measure for Golgi-to-membrane vesicular trafficking . Scale bar = 25 μm . ( C ) Western blots from quadriceps muscle of WT and Thbs4 Tg mice for vesicular markers associated with ER to Golgi trafficking: Sar1 , Rab24 and Rab6 , and vesicular markers associated with Golgi-to-membrane trafficking ( secretory vesicles ) : Rab3 and Rab8 ( n = 3 biological replicates ) . Gapdh is presented as processing and loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 017 To examine this effect of Thbs4 in greater molecular detail we also employed 2 domain-specific Thbs4 constructs and a related oligomeric glycoprotein Nell2 ( Figure 5E–J ) ( Brody et al . , 2016; Kuroda et al . , 1999; Lynch et al . , 2012 ) . Like Thbs4 , Nell2 contains an N-terminal laminin-G like ( LamG ) domain and an epidermal growth factor ( EGF ) -like repeat domain but lacks the ATF6α interacting Type III repeat ( T3R ) and TSP-C domains from Thbs4 . Importantly , unlike full-length Thbs4 , overexpression of Nell2 or the LamG domain of Thbs4 did not increase ER-to-Golgi or post-Golgi trafficking rates ( Figure 5E , F , I , J ) . However , overexpression of just the T3R domain of Thbs4 , which functions as the ATF6α interacting region , was sufficient to accelerate ER to Golgi and post-Golgi trafficking ( Figure 5G , H ) . Thbs4 overexpression in skeletal muscle also augmented the levels of trafficking regulatory proteins such as Sar1 , Rab24 , Rab6 , Rab3 and Rab8 , which control ER-to-Golgi and post-Golgi vesicular trafficking ( Figure 5—figure supplement 1C ) ( Brandizzi and Barlowe , 2013; Stenmark , 2009 ) . Collectively , these results indicate that Thbs4 accelerates intracellular vesicular trafficking in an ATF6α-dependent manner , thereby suggesting at least one molecular mechanism whereby Thbs4 might enhance membrane stability through greater fluxing of vesicles to the sarcolemma . To more definitively investigate if greater vesicular trafficking rates associated with Thbs4 activity might augment the residency of membrane attachment complex proteins , we generated membrane-specific protein preparations for Western blotting , as well as performed immunohistochemistry on skeletal muscle for direct visualization of the membranes in cross-section . As previously observed , loss of δ-sarcoglycan in skeletal muscle of Sgcd-/- mice resulted in the near complete absence of the other sarcoglycans at the membrane ( as they all form a complex; Figure 6A ) ( Durbeej and Campbell , 2002 ) . Interestingly , Thbs4 Tg alone displayed increased sarcolemmal levels of β-dystroglycan , dystrophin and β1D-integrin ( Figure 6A ) . More importantly , the Thbs4 Tg in the Sgcd deficient background resulted in greater membrane localization of α- , β- , and γ-sarcoglycan and β-dystroglycan , as well as in dystrophin , utrophin and β1D-integrin ( Figure 6A ) . More quantitative assessment of this effect by Western blotting of membrane-specific protein extracts showed that even the Thbs4 Tg alone gave increased membrane levels of utrophin , α- and β-dystroglycan , as well as β1D- , α7 and α5-integrins compared to WT controls ( Figure 6B; Figure 6—figure supplement 1A , B; red boxes ) , which likely explains our earlier observations whereby Thbs4 Tg skeletal muscle was protected from lengthening-contraction injury and laser injury in isolated myofibers versus WT . The Thbs4 Tg also augmented the membrane residency of these same proteins in the Sgcd-/- and mdx background , as well as increased membrane levels of α-sarcoglycan and β-sarcoglycan in skeletal muscle of Sgcd-/- mice ( Figure 6B; Figure 6—figure supplement 1A , B , for replicate samples ) . Importantly , previous work has shown that overexpression of DGC components augment the assembly of the entire complex and are inherently protective to MD ( Allikian et al . , 2004; Grounds et al . , 2005; Gumerson and Michele , 2011; Tinsley et al . , 1998 ) . Furthermore , we observed that expression of β1- and α7-integrin from total cytoplasmic protein extracts was not increased , suggesting that Thbs4 overexpression directly augmented membrane trafficking and localization of these critical attachment proteins to the surface ( Figure 6B , lower panel ) . Finally , we also observed that loss of Thbs4 in skeletal muscle partially reduced membrane residency of α- and β-dystroglycan , β1D- and α7-integrins ( Figure 6C , burgundy boxes ) . Collectively , these observations solidify a mechanism whereby Thbs4 regulates membrane stability in skeletal muscle by augmenting the trafficking of membrane attachment protein complexes to the sarcolemma . 10 . 7554/eLife . 17589 . 018Figure 6 . Thbs4 enhances stabilizing proteins at the sarcolemma . ( A ) Immunofluoresence ( green ) detection of δ- , α- , β- , and γ-sarcoclycan ( SGC ) , β-dystroglycan ( β-DG ) , utrophin ( Utro . ) , dystrophin ( Dystro . ) and β1D-integrin in littermates of three month-old WT , Thbs4-Tg , Sgcd-/- and Sgcd-/-Thbs4-Tg quadriceps . Representative images of 4 mice per genotype are shown . Scale bar = 25 μm . ( B ) Representative Western blots of sarcolemmal protein extracts ( upper ) or total cytoplasmic protein extracts ( lower ) from the quadriceps of the indicated groups of mice for the indicated proteins ( n = 4–5 biological replicates ) . Abbreviations: Utro , utrophin; Dystro , dystrophin; Dysfer , dysferlin; α-DG , α-dystroglycan; β-DG , β-dystroglycan; δ-SCG , δ-sarcoglycan; α-SCG , α-sarcoglycan; β-SGC , β-sarcoglycan; β1D- , α7- and α5- integrin . The red boxes show increased protein levels . Also see Figure 6—figure supplement 1 for replicates . ( C ) Representative immunoblotting for structural components of the DGC- and integrin-associated protein complexes in sarcolemmal preparations from Thbs4-/- and WT quadriceps at four months of age ( n = 4 biological replicates ) . The burgundy-boxed areas show reduced protein levels . Ponceau staining of a nonspecific band and dihydropyridine receptor α1 ( Cav1 . 1 ) were used as loading controls for sarcolemmal protein extracts; Gapdh was used as loading control for total cell protein extracts . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 01810 . 7554/eLife . 17589 . 019Figure 6—figure supplement 1 . Thbs4 enhances stabilizing proteins at the sarcolemma and directly interacts with integrins . ( A , B ) Representative Western blots of sarcolemmal protein extracts from the quadriceps of the indicated groups of mice at three months of age ( n = 4–5 biological replicates ) . Sgcd-/- Tg and mdx Tg indicate Sgcd-/- and mdx with skeletal muscle specific Thbs4 overexpression , respectively . Ponceau staining of a nonspecific band and dihydropyridine receptor α1 ( Cav1 . 1 ) were used as loading controls . Abbreviations: Utro , utrophin; Dystro , dystrophin; Dysfer , dysferlin; α-DG , α-dystroglycan; β-DG , β-dystroglycan; δ-SCG , δ-sarcoglycan; α-SCG , α-sarcoglycan; β-SGC , β-sarcoglycan; β1D- , α7- and α5-itg ( integrin ) . ( C ) Immunoblots for β1D- and α7-Integrin ( Itg ) , β-dystroglycan ( DG ) and Thbs4 ( Flag ) from neonatal rat ventricular myocyte extracts immunoprecipitated with a Flag antibody ( Thbs4 ) . Adβgal was used as a control infection ( n = 3 biological replicates ) . An adenovirus expressing a Flag-tagged Thbs4 protein was used to achieve high level of this protein to identify the interaction . ( D , E ) Representative Western blots for Thbs4 , α5- and β1D-integrin ( itg ) from intracellular vesicular isolates from WT and Thbs4 Tg quadriceps ( D ) or WT and Sgcd-/-quadriceps ( E ) that were immunoprecipitated with an antibody raised against the cytoplasmic domain of β1D-integrin ( n = 3 biological replicates ) , showing that Thbs4 and α5-integrin localize to β1D-integrin-positive intracellular vesicles . α-tubulin and Gapdh are presented as loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 019 As previously reported with other Thbs proteins ( Adams and Lawler , 2011 ) , we observed that Thbs4 could directly bind intracellular β1D-integrin , but not α7-integrin , in primary neonatal rat cardiomyocytes ( Figure 6—figure supplement 1C ) . In addition , both Thbs4 and α5-integrin localized to β1D-integrin-positive intracellular vesicles in WT , Thbs4 Tg and Sgcd-/- quadriceps muscle , placing Thbs4 within the same vesicles and intracellular compartment as the attachment proteins themselves ( Figure 6—figure supplement 1D , E ) . Given the results presented to this point , we hypothesized that ATF6α induced ER and post-ER vesicular expansion was responsible for increased intracellular trafficking of membrane attachment protein complexes to the sarcolemma and hence , protection from MD . Indeed , ATF6α was previously shown to expand the ER and post-ER vesicular compartment when activated or overexpressed ( Brody et al . , 2016; Lynch et al . , 2012 ) . Thus , to directly test this hypothesis we generated skeletal muscle-specific Tg mice overexpressing ATF6α , which showed high levels of nuclear ATF6α protein and induction of BiP , PDI and calreticulin as compared to WT levels , all without influencing Thbs4 protein expression levels ( Figure 7A , B; Figure 7—figure supplement 1 ) . Moreover , except for PDI we noted significantly higher levels of these ATF6α-dependent ER stress responsive factors in ATF6α Tg quadriceps as compared to our Thbs4 Tg ( Figure 7A , B; Figure 7—figure supplement 1 ) . ATF6α skeletal muscle-specific Tg mice appeared overtly normal and showed no histopathology of skeletal muscles ( Figure 7C; Figure 7—figure supplement 2A , B ) . Similar to Thbs4 Tg mice , ultrastructural analysis of skeletal muscle from ATF6α Tg mice showed a remarkable expansion of ER and the sub-sarcolemmal vesicular compartment , although these ATF6α-dependent vesicles appeared less dense in comparison to those observed in Thbs4 Tg mice ( Figure 7D versus Figure 1G ) . 10 . 7554/eLife . 17589 . 020Figure 7 . Skeletal muscle specific ATF6α overexpression drives ER stress and intracellular vesicular expansion , but not protection against MD . ( A ) Schematic diagram of the transgene ( Tg ) used to overexpress ATF6α in skeletal muscle . ( B ) Western blot analysis for Thbs4 , ATF6α , BiP , PDI and calreticulin ( Calret . ) expression in quadriceps ( Quad ) from WT , Thbs4 Tg and ATF6α Tg mice at six weeks of age . Gapdh is a processing and loading control ( n = 6 biological replicates ) . ( C ) Masson’s trichrome-stained histological sections from Quad of WT and ATF6α-Tg littermates at six weeks of age . Representative images of 5 mice per genotype are shown . Scale bar = 100 μm . ( D ) Transmission electron micrographs in Quad from WT and ATF6α-Tg mice at six weeks of age . The white arrows show dramatic expansion of ER and associated vesicles throughout the cell and especially in the sub-sarcolemmal region . Representative images of 2 mice per genotype are shown . Scale bar = 2 μm . ( E ) Masson’s trichrome stained histological sections of Quad from Sgcd-/-and Sgcd-/-ATF6α Tg mice at six weeks of age . Representative images of 5 mice per genotype are shown . Scale bar = 100 μm . ( F ) Quantitation of serum CK levels ( units/liter ) in the indicated genotypes of mice shown in the legend below the graph at six weeks of age . n = 10 mice for WT and n = 8 mice for the remaining genotypes . *p<0 . 05 versus WT by one-way ANOVA with post hoc Tukey’s test . ( G , H ) Histological analysis of the Quad showing percentage of myofibers with centrally located nuclei ( n = 5 mice for WT and ATF6α Tg , and n = 6 mice for Sgcd-/-and Sgcd-/-ATF6α Tg ) and interstitial fibrosis ( n = 5 mice per genotype ) at six weeks of age in WT , ATF6α Tg , Sgcd-/-and Sgcd-/-ATF6α Tg mice . *p<0 . 05 versus WT by one-way ANOVA with post hoc Tukey’s test . ( I ) Time to fatigue in seconds with forced downhill treadmill running in the indicated genotypes of mice shown in the legend . n = 6 mice per genotype . *p<0 . 05 versus WT by one-way ANOVA with post hoc Tukey’s test . ( J ) Representative immunofluorescent images of EBD ( red ) uptake in myofibers in the Quad of six week-old mice of the indicated genotypes . Membranes of myofibers are shown in green . Scale bars = 75 μm . Percent EBD-positive myofibers is indicated . Six mice per genotype were analyzed for EDB uptake . ( K ) Western for structural components of the DGC and integrin-associated protein complexes in sarcolemmal protein preparations from Quad of WT , ATF6α Tg , Sgcd-/-and Sgcd-/-ATF6α Tg littermates at six weeks of age . Ponceau staining of a nonspecific band and dihydropyridine receptor α1 ( Cav1 . 1 ) were used as loading controls ( n = 3 biological replicates ) . Abbreviations: Utro , utrophin; Dystro , dystrophin; α-DG , α-dystroglycan; β-DG , β-dystroglycan; δ-SCG , δ-sarcoglycan; α-SCG , α-sarcoglycan; β-SGC , β-sarcoglycan; β1D- , α7- and α5-Itg ( integrin ) . All data are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 02010 . 7554/eLife . 17589 . 021Figure 7—figure supplement 1 . Relative protein levels for immunoblots shown in Figure 7B . An additional set of immunoblots was included to obtain an n = 6 per genotype . *p<0 . 05 vs WT; #p<0 . 05 vs Thbs4 Tg by one-way ANOVA with post hoc Tukey’s test . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 02110 . 7554/eLife . 17589 . 022Figure 7—figure supplement 2 . ATF6α skeletal muscle-specific Tg mice are not protected from MD in the mdx genetic background . ( A ) H&E and Masson’s trichrome-stained histological sections of quadriceps from WT , ATF6α Tg , mdx and mdx ATF6α Tg at six weeks of age . Representative images of 6 mice per genotype are shown for H&E , whereas n = 6 mice for WT , ATF6α Tg , and mdx ATF6α Tg , and n = 7 mice for mdx are shown for Masson’s trichrome staining . Scale bar = 100 μm . ( B ) Quantitation of serum CK levels ( units/liter ) in WT , ATF6α Tg , mdx and mdx ATF6α Tg at six weeks of age . n = 10 mice for WT , n = 8 mice for ATF6α Tg , n = 7 mice for mdx , n = 6 mice for mdx ATF6α Tg . *p<0 . 05 vs WT by one-way ANOVA with post hoc Tukey’s test . Data are represented as mean ± SEM . ( C , D ) Percent central nucleated myofibers and interstitial fibrosis in histological sections from the quadriceps of the indicated genotypes of mice at six weeks of age . n = 6 mice for indicated genotypes in panel C; and n = 7 and 6 mice for mdx and mdx ATF6α Tg , respectively , in panel D . *p<0 . 05 vs WT by one-way ANOVA with post hoc Tukey’s test . Data are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 02210 . 7554/eLife . 17589 . 023Figure 7—figure supplement 3 . Complete membrane trafficking of Thbs4 is essential for its membrane stabilizing function . ( A ) Schematic of the experimental protocol used . Briefly , purified adenoviral ( Ad ) constructs encoding Flag-tagged full-length Thbs4 ( AdThbs4 ) , Flag-tagged Thbs4 calcium-binding mutant ( AdThbs4-mCa2+ , which induces ER-stress but isretained in the ER ) or βgal control were injected into the gastrocnemius ( Gastroc . ) muscles of postnatal day one rat pups , followed by a second ‘boost’ injection 48 hr later ( 108 viral particles per injection ) . Muscle tissue was harvested at postnatal day eight , cryo-sectioned and co-immunostained for Thbs4 and β1-integrin ( Itg ) . ( B ) Immunofluorescence detection of Thbs4 and Thbs4-mCa2+ ( both Flag tagged; green ) and β1-integrin ( red ) from muscles infected with indicated Thbs4 constructs or βgal control . Representative images of 3 muscles per condition are shown . Scale bar = 10 μm . ( C ) Quantitative analysis of sarcolemma localized β1-integrin positive immunofluorescence . Data are presented as % β1-integrin positive area from total muscle area infected with the indicated Thbs4 constructs ( Flag-positive ) or from total muscle area in the case of βgal control . N = 3 per treatment condition . *p<0 . 05 vs . Adβgal and AdThbs4-mCa2+ by Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 023 Next , ATF6α mice were crossed with both Sgcd-/- and mdx mice to directly examine the hypothesis that the adaptive ER stress response and sub-sarcolemmal expansion of vesicles induced by ATF6α was a protective mechanism underlying Thbs4 action . However , ATF6α overexpression in the Sgcd-/- or mdx dystrophic background provided no protection whatsoever ( Figure 7E–J; Figure 7—figure supplement 2A–D ) . There was no reduction in histopathology or serum CK levels or membrane rupture as assessed with EBD , nor was treadmill running improved by the ATF6α Tg in either the Sgcd-/- or mdx backgrounds . More importantly , ATF6α overexpression did not increase the sarcolemmal localization of any of the membrane attachment proteins observed with Thbs4 overexpression in skeletal muscle ( Figure 7K ) . Thus , while ATF6α overexpression activated an adaptive ER stress response in skeletal muscle with a dramatic induction of intracellular and sub-sarcolemmal vesicles to an even higher level than observed in our Thbs4 overexpressing mice , it did not augment the membrane residency of membrane stabilizing proteins in muscle . Hence , our data indicate that ATF6α is only one part of a more integrated mechanism whereby Thbs4 regulates membrane stability of skeletal muscle . Taken together , our data so far indicate that increased levels of Thbs4 itself and its trafficking through the secretory pathway are essential to increase membrane attachment protein complexes at the sarcolemma . To test this hypothesis , we took advantage of a previously established adenoviral construct encoding a Thbs4 calcium-binding containing mutant that is retained in the ER ( Ad-Thbs4-mCa2+ ) ( Brody et al . , 2016 ) . Importantly , the mutant still induces an ATF6α mediated ER-stress response , both in neonatal rat cardiomyocytes and when expressed in the gastrocnemius muscle of early postnatal rat pups ( Brody et al . , 2016 ) . Utilizing an identical in vivo approach with adenoviral gene transfer into the gastrocnemius of early neonatal rat pups , we compared βgal control with full-length Thbs4 versus the Thbs4-mCa2+ for effects on β1 Integrin membrane levels . The data showed that only the secretion competent full-length Thbs4 , but not the full-length ER-retained Thbs4 mutant , promoted greater β1 integrin membrane occupancy ( Figure 7—figure supplement 3 ) . Hence , Thbs4 must move through the secretory pathway to chaperone at least β1 integrin to the sarcolemma . Our results in mice were reminiscent of data from Drosophila , which have a single Tsp gene that when deficient causes embryonic lethality due to ruptures in tendon/muscle attachments ( Subramanian et al . , 2007 ) . Moreover , Tsp in Drosophila was also shown to interact with αPS2/βPS integrin ( Chanana et al . , 2007; Subramanian et al . , 2007 ) . Thus to investigate a potential conservation of Thbs4’s function , we generated Tg Drosophila in which the mouse Thbs4 cDNA or the Drosophila Tsp cDNA were driven with a muscle-specific myocyte enhancer factor 2 ( MEF2 ) regulatory region to achieve overexpression of either protein in Drosophila muscle . These lines were subsequently crossed with a Drosophila model of MD that lacks the δ-sarcoglycan-like gene ( Sgcd840 ) , a line that is marked by muscle disease with shortened life-span , loss of muscle function and overt rupture of the muscles due to structural weakness ( Allikian et al . , 2007 ) . Remarkably , both Thbs4 and Tsp rescued the reduced life span in the Sgcd840 MD Drosophila line , and normalized muscle function as assessed with a negative geotaxis assay ( Figure 8A , B; Figure 8—figure supplement 1A , B ) . The rupture of the dorsal median indirect flight muscles and their loss of proper ectoskeletal attachment were also rescued by muscle-specific overexpression of either mouse Thbs4 or Drosophila Tsp ( Figure 8C; Figure 8—figure supplement 1C ) . 10 . 7554/eLife . 17589 . 024Figure 8 . Thbs4 regulates muscle membrane integrity in Drosophila . ( A ) Survival of the Drosophila lines shown over a period of 40 days . p<0 . 001 for Drosophila line 840 lacking the δ-sarcoglycan homologue gene ( Sgcd840 ) versus Sgcd840;mThbs4 line that express the mouse Thbs4 protein in muscle . N = 358 for WT; 163 for mThbs4; 395 for Sgcd840; 300 for Sgcd840;mThbs4 . Statistical analysis performed with log rank , Mantel-Cox test . ( B ) Physical performance with a negative geotaxis assay of WT ( n = 208 ) , mThbs4 ( n = 190 ) , Sgcd840 ( n = 203 ) , and Sgcd840;mThbs4 Drosophila ( n = 205 ) at the indicated ages . All data are represented as mean ± SEM . *p<0 . 05 vs WT and mThbs4; #p<0 . 05 vsSgcd840 by one-way ANOVA with post hoc Tukey’s test . ( C ) Representative H&E stained histological sections of the dorsal median indirect flight muscles at 30 days of age in the indicated genotypes of Drosophila . Representative images of 14 Drosophila per genotype studied . The arrow shows a prominent area of muscle rupture in the Sgcd840 MD Drosophila line . Scale bar = 200 μm . ( D ) Immunohistochemistry showing increased intracellular mThbs4 and calreticulin ( Calret . ; both green ) in longitudinal sections of the dorsal median indirect flight muscle of 30 day-old mThbs4 and Sgcd840; mThbs4 expressing flies compared to WT and Sgcd840 lines . Membranes are shown in yellow , nuclei in blue . Representative images of 14 Drosophila per genotype studied . Scale bars = 10 μm . ( E ) Immunofluorescence detection of fly βPS integrin ( green ) in cross-sections the dorsal median indirect flight muscle of 30 day-old Drosophila of the indicated genotypes . Representative images of 10 Drosophila per genotype studied . Scale bars = 10 μm . ( F ) Representative Western blot of mThbs4 and ER-stress proteins BiP , PDI , calreticulin ( Calret . ) , as well as βPS integrin in 15 day-old WT , mThbs4 , Sgcd840 , and Sgcd840;mThbs4 lines ( n = 4 biological replicates ) . β actin was used as loading control . The red boxes show upregulation of βPS integrin with muscle specific mThbs4 overexpression in the WT and Sgcd840 mutant lines . ( G ) Transmission electron microscopy of the dorsal median indirect flight muscle of WT , mThbs4 , Sgcd840 , and Sgcd840;mThbs4 Drosophila . Arrows indicate the characteristic sarcomeric tears and disorganization present in the Sgcd840 line . Representative images of 4 Drosophila studied for the WT , mThbs4 , and Sgcd840;mThbs4 lines , and 5 Drosophila for the Sgcd840 line . Scale bar = 2 μm . ( H ) Immunogold-transmission electron microscopy of mThbs4 in longitudinal section of the dorsal median indirect flight muscle of mThbs4 Drosophila . The arrows indicate mThbs4 ( 6 nm gold particles ) localized to intracellular vesicles . Representative images of 4 Drosophila are shown . Scale bar = 50 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 02410 . 7554/eLife . 17589 . 025Figure 8—figure supplement 1 . Tsp expression in muscle of Drosophila rescues MD due to deletion of the δ-sarcoglycan-like gene ( Sgcd840 ) . ( A ) Fly survival was compared over a period of 40 days . p<0 . 001 for the muscular dystrophy Drosophila line lacking the δ-sarcoglycan homologue gene ( Sgcd840 ) compared to Sgcd840 mutant Drosophila line expressing the single Drosophila Tsp gene in a muscle specific manner using the MEF2 driver ( Sgcd840;Tsp ) with the GAL4-UAS system . N = 379 for WT; 118 for Tsp; 478 for Sgcd840; 189 for Sgcd840;Tsp . Results were compared by Kaplan–Meier statistical analysis with log rank , Mantel-Cox test . The data show that Tsp overexpression in muscle rescues foreshortened life span in the Sgcd840 line , back to that of WT flies . ( B ) Physical performance with a negative geotaxis assay of WT , Tsp , Sgcd840 mutant , and Sgcd840;Tsp Drosophila at the indicated ages . n = 208 for WT , n = 55 for Tsp , n= 203 for Sgcd840 mutant and n = 145 for Sgcd840;Tsp Drosophila . All data are represented as mean ± SEM . *p<0 . 05 vs WT and Tsp; #p<0 . 05 vsSgcd840 mutants by one-way ANOVA with post hoc Tukey’s test . ( C ) Representative H&E staining of histological sections from the dorsal median indirect flight muscle of 30 day-old flies of the indicated genotypes . The arrow shows a prominent area of muscle rupture in the Sgcd840 mutant line , which is rescued by overexpression of the Drosophila Tsp gene in muscle . Representative images of 14 Drosophila per genotype studied . Scale bar = 200 μm . ( D ) Immunohistochemistry showing increased intracellular Tsp protein ( green ) in longitudinal sections of the dorsal median indirect flight muscle of 30 day-old WT , Tsp , Sgcd840 mutant , and Sgcd840;Tsp Drosophila . Membranes are shown in yellow , nuclei in blue . Representative images of 14 Drosophila per genotype studied . Scale bars = 10 μm . ( E ) Immunofluorescence detection of Drosophila βPS integrin ( green ) in histological cross-sections of the dorsal median indirect flight muscle of 30 day-old lines is shown . Representative images of 10 Drosophila per genotype studied . Scale bars = 10 μm . ( F ) Representative Western blot of Drosophila Tsp and ER-stress proteins BiP , PDI , calreticulin ( Calret . ) , as well as βPS integrin in 15-day-old WT , Tsp , Sgcd840 , and Sgcd840;Tsp lines ( n = 4 biological replicates ) . β-actin was used as loading control . The red boxes show upregulation of βPS integrin with muscle-specific Tsp overexpression in the WT and Sgcd840 mutant lines . ( G ) Transmission electron microscopy of the dorsal median indirect flight muscle of 25 day-old WT , Tsp , Sgcd840 , and Sgcd840;Tsp Drosophila . Arrows indicate the characteristic sarcomeric tears and disorganization present in the Sgcd840 line . Representative images of 4 Drosophila studied for the WT , mThbs4 , and Sgcd840;mThbs4 lines , and 5 Drosophila for the Sgcd840 line . Scale bar = 2 μm . ( H , I ) Quantitative transmission electron microscopy analysis of the number of sarcomeric tears per myofiber in the dorsal median indirect flight muscle of WT ( n = 14 fibers from 4 Drosophila ) , Tsp ( n = 17 fibers from 4 Drosophila ) , mThbs4 ( n = 14 fibers from 4 Drosophila ) , Sgcd840 ( n = 23 fibers from 5 Drosophila ) , Sgcd840;Tsp ( n = 16 fibers from 4 Drosophila ) , and Sgcd840;mThbs4 ( n = 21 fibers from 4 Drosophila ) . All data are represented as mean ± SEM . *p<0 . 001 vs all other lines; #p<0 . 001 vs Sgcd840 mutants by one-way ANOVA with post hoc Tukey’s test . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 025 Mechanistically , Thbs4 or Tsp overexpression was exclusively restricted to a vesicular compartment within the indirect flight muscles of these overexpressing Tg lines , but not outside the myofibers ( Figure 8D , H; Figure 8—figure supplement 1D ) . More provocatively , both mouse Thbs4 and Drosophila Tsp overexpression produced noticeably greater levels of membrane βPS integrin localization in Drosophila muscle ( Figure 8E , F; Figure 8—figure supplement 1E , F ) . Ultrastructural analysis revealed a remarkable rescue of the sarcomeric tears in Sgcd840Drosophila with the mouse Thbs4 or Drosophila Tsp Tg , although without inducing ER-stress nor ER and post-ER vesicular expansion in these muscles ( Figure 8F , G; Figure 8—figure supplement 1G–I ) . Collectively , these results indicate that Thbs proteins underlie an ancient program for membrane stabilization through regulation of intracellular attachment protein complexes and their content at the surface membrane , although ER expansion through ATF6α appears to have evolved phylogenetically after Drosophila .
A vast majority of the Thbs literature over the past 3 decades have invoked or interpreted data consistent with a primary extracellular function for these proteins , while only a handful have shown a direct intracellular function ( Adams and Lawler , 2011; Ambily et al . , 2014; Baek et al . , 2013; Brody et al . , 2016; Christopherson et al . , 2005; Duquette et al . , 2014; Frolova et al . , 2014 , 2010 , 2012; Hauser et al . , 1995; Lynch et al . , 2012; McKeown-Longo et al . , 1984; Posey et al . , 2014; Schellings et al . , 2009; Södersten et al . , 2006 ) . Here , we identify a fundamental yet previously unrecognized intracellular role for Thbs4 in skeletal muscle , where it directly augments selective vesicular trafficking and chaperones DGC and integrin attachment complexes to the membrane , leading to greater stability and levels of select complexes at the sarcolemma , and thereby enhancing the mechanical stability of the myofiber ( Figure 9 ) . In fact , our findings identify Thbs4 as a crucial component to maintain muscle fiber integrity as loss of Thbs4 results in sarcolemma weakness that causes spontaneous dystrophic changes with aging . Importantly , the Thbs protective effect holds true in both mouse and Drosophila skeletal muscle . 10 . 7554/eLife . 17589 . 026Figure 9 . Model of how Thbs4 functions as an intracellular regulator of muscle cellular attachment and membrane stability . As a matricellular protein , thrombospondin-4 ( Thbs4 ) pentamers are synthesized in the ER lumen and then transported to the Golgi , where after they traverse the secretory pathway to fuse with the plasma membrane for secretion . Thbs4 can then reside within the extracellular matrix ( ECM ) or be actively endocytosed and returned to the intracellular compartment by a recycling receptor . In addition to its established extracellular functions , combined studies in the heart and skeletal muscle now reveal that while in the ER , Thbs4 can compete with BiP ( GRP78 ) for binding to the ER-resident transcription factor ATF6α , thereby facilitating ATF6α translocation to the Golgi for processing and subsequent shuttling to the nucleus where it regulates expression of ER stress responsive genes that are also part of the unfolded protein response ( UPR ) . ATF6α induction in cardiac and skeletal muscle , or by overexpression in Tg mice ( lower panel ) causes a dramatic expansion of the ER and post-ER vesicles , as well as increased vesicular trafficking to the membrane . The ability of Thbs4 to induce ATF6α processing and nuclear trafficking also causes this same ER expansion and augmentation of intracellular vesicular trafficking to the membrane . However , Thbs4 uniquely regulates trafficking of selected integrins and dystrophin-associated glycoprotein complexes ( DGC ) members to the sarcolemma , thereby enhancing the mechanical stability of the myofiber , such as observed in Thbs4 Tg muscle ( middle panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17589 . 026 As secreted matricellular proteins , Thbs’ are first produced in the ER where they are glycosylated and transit to the Golgi for additional modifications , where after they traverse the remainder of the secretory pathway ( Adams and Lawler , 2011 ) . Interestingly , we observed that skeletal muscle-specific overexpression of Thbs4 did not result in accumulation within the ECM , but predominantly produced an intracellular protein localization pattern within the ER and post-ER vesicular network . In fact , we have also since generated cardiac-specific Tg mice overexpressing Thbs1 , 2 , 3 , 4 or 5 in the heart ( [Lynch et al . , 2012] and data not shown ) . In these hearts all five Thbs’ reside mainly within the intracellular vesicular network and ER with only limited detectable protein accumulation outside cardiomyocytes . This is in dramatic contrast to the overexpression of an array of other matricellular proteins , such as periostin , which saturates the ECM when overexpressed ( Oka et al . , 2007 ) . In contrast , dystrophic skeletal muscle did reveal occasional Thbs4 protein accumulation in fibrotic regions around myofibers , confirming that this protein can reside for a period of time in the ECM . These dynamic localization differences at baseline versus during fibrotic disease could be attributed to Thbs recycling at the cell surface ( Adams and Lawler , 2011; Wang et al . , 2004 ) . Indeed , we observed rapid up take of recombinant Thbs4 when given exogenously to cultured C2C12 myoblasts or myotubes . It is possible that extracellular Thbs4 reuptake could occur through its designated receptor or through the integrin and DGC complexes . Indeed , other matricellular proteins such as SPARC were shown to be actively taken up back into myofibers through such a process with integrin associated endocytosis , sorting and recycling ( Chlenski et al . , 2011; De Franceschi et al . , 2015; Nakamura et al . , 2014 ) . Our data identify ATF6α as a transcriptional regulator of secretory pathway activity in muscle , a function that was previously established for the inositol-requiring enzyme 1α / X-box binding protein ( IRE1α/XBP-1 ) axis of the canonical ER stress response pathway during plasma cell differentiation ( Shaffer et al . , 2004 ) . Importantly , although enhancement of the ATF6α-mediated adaptive ER stress pathway was sufficient to drive the dramatic expansion of the ER and post-ER vesicular content and increase vesicular trafficking to the membrane , it did not augment membrane residency of DGC and integrin attachment complexes at the sarcolemma , nor was it sufficient to protect against MD . Thus , ATF6α is only part of a more complex intracellular mechanism whereby Thbs4 regulates mechanical stability of the muscle fiber and its sarcolemma , in coordination with greater vesicle formation and trafficking ( Figure 9 ) . Data from both Drosophila and zebrafish show that thrombospondin proteins localize outside of cells within the tendinous junctions ( Chanana et al . , 2007; Subramanian and Schilling , 2014; Subramanian et al . , 2007 ) . In addition , vertebrate Thbs1 and Thbs2 genes have evolved domains that are tailored to affecting processes outside the cell , such as altering transforming growth factor-β activity and the angiogenic response ( Adams and Lawler , 2011; Bornstein , 2001; Carlson et al . , 2008 ) . Hence , the simplest interpretation of our data and that reported in the literature is that Thbs proteins are complex , multifactorial proteins that function both inside and outside the cell . However , our working hypothesis is that Thbs4 appears more tailored to intracellular functionality in cardiac and skeletal muscle ( Figure 9 ) , and this same paradigm appears to hold true for the other Thbs family members in other tissues . For example , Thbs1 was previously shown to localize to the intracellular side of membrane attachment sites by immunogold-electron microscopy in endothelial cells ( Hiscott et al . , 1997 ) . Furthermore , Thbs proteins are known to strongly interact with many different integrin heterodimers , and singular proteomic analysis of the α5β1 integrin complex identified Thbs2 as a core element of its 'interactome' ( Adams and Lawler , 2011; Bouvard et al . , 2013; De Franceschi et al . , 2015; Plow et al . , 2000; Schiller et al . , 2013 ) . Our ultrastructural and biochemical analyses showed that the Thbs-integrin complex resides within the lumen of vesicles . Although further , in-depth analyses will suggest whether Thbs4 co-regulates the signaling competence of this integrin complex and the exact preassembly stage that is influenced by Thbs4 on the way to the cell surface . Finally , Thbs1 silencing or overexpression in human cancer cells was shown to decrease or enhance integrin protein levels , respectively , in the intracellular and plasma membrane compartment and thereby modulate cellular adhesion ( Duquette et al . , 2013; John et al . , 2010 ) . Taken together , various lines of evidence indicate that Thbs proteins stabilize integrins and the DGC at membrane attachment complexes of the sarcolemma through an intracellular function . Importantly , this observation holds true in both mouse and Drosophila skeletal muscle where integrin protein content at the cell membrane was increased with Thbs4 overexpression , yet very little Thbs protein was observed outside the cell in either species at baseline . While previously published data showed a large concentration of Drosophila Tsp protein to the myotendinous junction in stage 16 embryos , this is also the very same region where the integrins are highly concentrated in identical foci and the data do not distinguish if Tsp in inside or outside the cell ( Subramanian et al . , 2007 ) . At earlier embryonic stages ( stage 12–13 ) however , the Tsp protein is intracellular with a diffuse pattern similar to that of integrins that have yet to be deposited at the cell membrane ( Subramanian et al . , 2007 ) . Moreover , in later stage Drosophila larvae Tsp is no longer detected in the myotendinous junctions or in tendons . Rather , a network of Tsp positive staining is observed within the muscle of larvae ( unpublished observations , Talila Volk ) . However , in mammals Thbs4 can oligomerize with Thbs5 and be deposited in the tendon ( Hauser et al . , 1995; Södersten et al . , 2006 ) , loss of Thbs4-/- in mice showed altered ECM composition and weakened muscle-tendons ( Frolova et al . , 2014 ) , and altered inflammatory responses associated with arteriogenesis ( Frolova et al . , 2010 ) , whereas Thbs1 and Thbs2 were shown to function from outside the cell in augmenting developmental synaptogenesis ( Christopherson et al . , 2005 ) . Hence , Thbs proteins are clearly complex regulatory proteins with intra- and extracellular functions . One aspect of the biology that was not conserved in Drosophila was the ability of Thbs or Tsp overexpression to expand the intracellular vesicular compartment in this lower organism , likely because Drosophila does not rely on an ATF6-like mechanism for the ER stress response , and the Thbs interacting domain within ATF6α is not contained in the Drosophila homologue of this gene ( Mori , 2009 ) . Thus , the ability of Thbs proteins to activate ATF6α to augment ER protein production and secretory pathway activity is a later evolutionary adaptation beyond just stabilizing membrane attachment complexes , which in higher organisms more effectively coordinates tissue remodeling and healing through the Thbs proteins . Taken together , the unique aspects of muscle membrane biology allowed us to uncover a previously unknown and possibly dominant intracellular function of the Thbs proteins that is evolutionarily conserved ( Figure 9 ) . This new model for Thbs protein function has many disease ramifications , especially in skeletal muscle , a tissue that is highly sensitive to mutations in genes that cause weaknesses in cellular attachment and membrane stability . Indeed , there is a lack of therapeutic strategies to effectively treat MD and our study suggests that this protein may provide a universal approach to strengthen the sarcolemma in skeletal muscle if employed in a gene therapy approach . Importantly , this protein is already present in muscle and it would not be perceived as a neo-antigen by viral-mediated overexpression , hence it should be well tolerated and best used in patients where the genetic basis of their MD disease is due to a loss of structural support of the sarcolemma ( majority of cases ) .
Skeletal muscle-specific transgenic mice for Thbs4 and ATF6α were generated using the modified human skeletal α-actin ( Ska ) promoter construct as previously described ( Goonasekera et al . , 2011; Lynch et al . , 2012 ) . Briefly , full-length mouse Thbs4 cDNA was obtained from Open Biosystems ( Accession number: BC139414 ) and amplified by PCR and cloned into the BamHI and EcoRV sites of the Ska-promotor expressing vector ( forward: 5'-CGCGGATCCATGCCGGCCCCACGCGCG-3’ , and reverse: 5'-ATCTCAATTATCCAAGCGGTC AAAACTCTGGG-3’ ) . Full-length mouse ATF6α cDNA was obtained from a previously generated pcDNA1-ATF6α plasmid ( Lynch et al . , 2012 ) . Mouse ATF6α cDNA was amplified by PCR and subsequently cloned by PCR into the KpnI and NotI sites of the Ska-promotor expressing vector ( forward: 5'- GGGGTACCATGGAGTCGCCTTTTAGTCC-3’ , and reverse: 5'-ATAAGAATGCGGCCGCCTACTGCAACGACTCAGGGAT-3’ ) . All constructs were confirmed by DNA sequencing . To make Tg mice , the Ska-plasmid backbone was removed and the Ska-Thbs4 and Ska-ATF6α fragments were gel purified followed by Elutip-D column purification ( Schleicher and Schuell Bioscience; Dassel , Germany , Cat . 10462617 ) for newly fertilized oocyte injection at the Cincinnati Children’s Hospital Transgenic Animal and Genome Editing Core Facility . All transgenic mice were produced in the FVB/N background . Mice deficient for Thbs4 ( Thbs4-/-; Strain: B6 . 129P2-Thbs4tm1Dgen/J ) and mdx mice ( Strain: C57BL/10ScSn-Dmdmdx/J ) were purchased from Jackson Laboratories ( Bar Harbor , Maine ) . Sgcd-/- mice were previously described ( Hack et al . , 2000 ) . Next , Ska-Thbs4-Tg , ska-ATF6α-Tg and Thbs4-/- mice were backcrossed for at least 6 generations into the Sgcd-/- background to generate Sgcd-/- Thbs4-Tg mice , Sgcd-/- ATF6α-Tg mice and Sgcd-/- Thbs4-/- mice , as well as their littermate controls . An identical breeding strategy was used to generate mdx Thbs4-/- mice . In addition , males from each transgenic line were crossed to mdx heterozygous females to generate mdx-Tg and mdx non-Tg male littermates and their appropriate controls . All animal experiments were approved by the Institutional Animal Care and Use Committee of the Cincinnati Children’s Hospital Medical Center ( Protocol# IACUC2013-0013 ) . No human subjects or human tissue was directly used in experiments in this study . A search of the ‘National Center for Biotechnology Information Gene Expression Omnibus ( NCBI GEO ) ’ database ( Barrett and Edgar , 2006 ) revealed that Bakay and colleagues recently performed microarray experiments on human muscle biopsies of various muscle diseases ( GEO accession GDS1956/204776 , [Bakay et al . , 2006] ) . Available data included 11 different muscle diseases , with a total of 121 human muscle biopsy specimens tested on Affymetrix U133A microarrays . Individual Thbs4 mRNA levels from samples with Becker Muscular Dystrophy ( BMD , n = 5 ) ; Duchenne Muscular Dystrophy ( DMD , n = 10 ) ; dystrophy due to calpain-3 mutations ( LGMD2A , n = 10 ) ; and dystrophy due to a paucity of dysferlin ( LGMD2B , n = 10 ) were averaged and compared to those from healthy muscle biopsies ( n = 18 ) using an unpaired two-tailed t-test . Mouse Thbs4 or an eGFP cDNA was amplified by PCR and inserted into the BamHI and XhoI sites of pAAV-MCS vector . AAV9-CMV-eGFP and AAV9-CMV-Thbs4 were produced using the triple transfection method in HEK293 cells as previously described and stored at −80°C until commencing the in vivo experiments ( Gray et al . , 2011; Zincarelli et al . , 2008 ) . Next , both left and right gastrocnemius muscles of three-day-old Sgcd-/- mice were injected with either AAV9-Thbs4 or AAV9-eGFP ( both 1E10 viral particles in 30 μl isotonic saline; [Goonasekera et al . , 2011] ) . Mice were sacrificed at six weeks of age . The left gastrocnemius of each mouse was fixed , processed , paraffin embedded , sectioned , and stained with H&E and Masson’s trichrome , whereas the right muscles were snap-frozen in liquid nitrogen for storage in −80°C . A subset of muscles were embedded in Optimal Cutting Temperature Compound ( O . C . T , Tissue-Tek , Sakura Americas , Torrance , CA , Cat #4583 ) , frozen , and 7 μm cryosections were generated to confirm eGFP expression by direct fluorescence ( not shown ) . Indicated muscles were fixed overnight in 4% paraformaldehyde , dehydrated in ethanol and paraffin embedded . Global muscle architecture and pathological indices were determined from 5 μm thick transverse sections at the center of the muscle stained with either H&E or Masson’s trichrome ( Goonasekera et al . , 2011 ) . Approximately 1000 fibers per mouse for each muscle group were counted for analysis of percentage central nucleation using ImageJ software . Interstitial fibrosis was quantified using ImageJ software as percentage of blue area in Masson’s trichrome stained paraffin-embedded sections ( Goonasekera et al . , 2011 ) . For co-labeling of tissue sections with Thbs4 and ER marker calreticulin on paraffin-embedded muscles , 5 μm sections were rehydrated and heated in 1x antigen retrieval CITRA ( BioGenex , Fremont , CA , Cat# HK086-9K ) . Muscle sections were permeabilized for 10 min in 0 . 3% triton/PBS and then in a blocking buffer ( 0 . 1% triton/PBS , 5% goat serum , 2% BSA ) for 1 hr at room temperature . Primary antibody incubations were overnight at 4°C ( Thbs4: AF2390 , 1:150 dilution , R&D Systems , Minneapolis , MN; and ER-marker calreticulin: Abcam , ab2907 , Cambridge , MA , 1:100; all in blocking buffer ) . Appropriate Alexa Fluor-488 ( green ) and Alexa Fluor-568 ( red ) secondary antibodies ( Invitrogen , Waltham MA , 1:400 in blocking buffer ) were applied for 2 hr at room temperature and subsequently for 10 min with DAPI nuclear DNA stain ( Invitrogen , 1:10 . 000 ) . For co-labeling of tissue sections with Thbs4 and collagen I or periostin , freshly harvested quadriceps were fixed for 4 hr in 4% PFA at 4°C , rinsed with PBS and cryoprotected in 30% sucrose/PBS overnight before embedding in O . C . T . Afterwards , 10 μm cryosections were collected , rinsed in PBS and blocked for 30 min at room temperature in blocking solution ( PBS with 5% goat serum , 2% bovine serum albumin , 0 . 1% Triton X-100 ) . Primary antibody incubations were overnight at 4°C ( Thbs4: AF2390 , 1:150 dilution , R&D Systems , Minneapolis , MN; collagen type I , 1:300 dilution , Abcam , ab34710; periostin , NBP1-30042 , Novus Biologicals , Littleton , CO; 1:200 dilution; all in blocking buffer ) . Next , Alexa Fluor-488 ( green ) and Alexa Fluor-568 ( red ) secondary antibodies ( Invitrogen , Waltham MA , 1:400 in blocking buffer ) were applied for 2 hr at room temperature and subsequently for 10 min with DAPI nuclear DNA stain ( Invitrogen , 1:10 . 000 ) . For IHC of the dystrophin-glycoprotein complex ( DGC ) -associated proteins , freshly harvested quadriceps from three month-old mice were embedded in O . C . T and frozen in liquid nitrogen . Transverse tissue sections were cut at a thickness of 7 μm and stored in −80°C until further use . Here sections were acclimated to RT for 15 min , post-fixed in ice-cold methanol for 10 min , washed in PBS and then blocked for 30 min with either 0 . 1% triton/PBS , 5% goat serum , 2% BSA or with Mouse on Mouse ( M . O . M . ) blocking reagent ( Vector Laboratories , Burlingame , CA , BMK-2202 ) as described by the manufacturer’s protocol for mouse primary antibodies . Sections were incubated with primary antibody in blocking solution at 4°C overnight . Primary antibodies included: δ-sarcoglycan ( Abcam , ab92896 , 1:100 ) , α-sarcoglycan , β-sarcoglycan , γ-sarcoglycan ( NovaCastra , Buffalo Grove , IL , NCL-a-sarc , NCL-b-sarc and NCL-g-sarc , all 1:250 ) , β-dystroglycan ( Development Studies Hybridoma Bank , Iowa City IA , MANDAG2; 1:50 ) , utrophin ( Santa Cruz Biotechnology , MANCHO7 , sc-81557; 1:50 ) , dystrophin ( Abcam , ab15277 , 1:200 ) and β1D-integrin ( Millipore , Billerica , MA , MAB1900 , 1:250 ) . Primary antibodies were detected by applying Alexa Fluor-488 conjugated goat-anti-rabbit ( Invitrogen , 1:400 in blocking buffer ) for 2 hr at RT or biotinylated anti–mouse ( Vector Laboratories , M . O . M kit , 1:500 ) followed by Alexa Fluor-488 streptavidin conjugate ( Invitrogen , 1:200 ) , both for 45 min at RT . For each immunostain , sections were incubated with secondary and tertiary antibodies alone as a control for specificity ( not shown ) . All sections were mounted in Vectashield Hard Set ( Vector Laboratories , H-1400 ) to prevent photobleaching and visualized using a Nikon A1 confocal laser microscope system equipped with 40x H2O objective ( NA = 1 . 15 ) . All imaging was done under identical conditions using NIS Elements Advanced Research ( AR ) microscope imaging software ( Nikon Instruments Inc . Melville , NY ) . Quadriceps , gastrocnemius , soleus , diaphragm and hearts were harvested and immediately frozen in liquid nitrogen for storage at −80°C . To evaluate ER-stress and Thbs4 protein expression , muscles were homogenized ( Fisher Scientific , Waltham , MA , TissueMiser ) in ice-cold RIPA buffer containing Halt Protease Inhibitor cocktail ( ThermoScientific , Waltham , MA , #78430 ) . Next , samples were sonicated ( SP Scientific , Warminster , PA , VirSonic 60 , power setting 3 for 3 times 10 s ) , lysates were cleared by centrifugation at 14 , 000 rpm for 14 min at 4°C and stored at −80°C . To evaluate glycosylation pattern , quadriceps protein extracts were treated with Endoglycosidase H ( Endo H; New England Biolabs Inc . , Ipswich , MA , P07P2 ) , peptide N-glycosidase F ( PNGase F , New England Biolabs Inc . , P0704 ) or protein deglycosylation mix ( New England Biolabs Inc . , P6039 ) prior to SDS-PAGE , according to the manufacturer’s instructions . Endo H cleaves high mannose residues at hybrid oligosaccharides present on proteins in the ER , whereas PNGase cleaves both these and more complex oligosaccharides that result from processing in the Golgi ( Hewett et al . , 2004 ) . Control samples were treated the same way without addition of enzymes . To evaluate DGC-associated proteins , fresh quadriceps muscle was harvested and crude sarcolemmal isolates were prepared as previously described ( Kobayashi et al . , 2008 ) . Briefly , freshly harvested quadriceps was homogenized in 7 . 5x volumes of ice-cold lysis buffer ( 20 mM Na4P2O7 , 20 mM NaH2PO4 , 1 mM MgCl2 , 0 . 303 M sucrose , 0 . 5 mM EDTA , pH 7 . 1 with 5 μg/ml aprotinin and leupeptin , 0 . 5 μg/ml pepstatin A , 0 . 23 mM PMSF , 0 . 64 mM benzamidine , and 2 μM calpain inhibitor I and calpeptin ) , then centrifuged 14 , 000 g for 20 min at 4°C; the pellet was re-suspended , re-homogenized and both supernatants were centrifuged 30 , 000 g for 30 min at 4°C after which the pellet was re-suspended in 100 μl lysis buffer and stored at −80°C until further use . Extracellular protein fractionation from quadriceps muscle was essentially performed as previously described ( Tjondrokoesoemo et al . , 2016 ) . In summary , freshly harvested quadriceps muscle was minced , washed with PBS and subjected a 1 hr washing step in 0 . 5 M NaCl , 10 mM Tris-HCl , 25 mM EDTA ( PH 7 . 5 ) . Next , samples were decellularized overnight in 0 . 1% SDS , 25 mM EDTA , followed by extracellular matrix extraction with 4 M guanidine hydrochloride , 50 mM C2H3NaO2 , 25 mM EDTA ( pH 5 . 8 ) . Finally , proteins were precipitated overnight in 80% EtOH , air dried and treated with protein deglycoylation mix ( New England Biolabs Inc . , P6039 ) . Immunoprecipitations were performed as described ( Brody et al . , 2016 ) . Briefly , rat neonatal cardiomyocytes were transduced with adenovirus to overexpress Thbs4 with a C-terminal Flag tag or a β-galactosidase ( βgal ) expressing construct as a control . Two days later , cardiomyocyte lysates were harvested and immunoprecipitated with anti-Flag magnetic beads ( Sigma-Aldrich , M8823 ) . Immunoprecipitates were resolved by SDS-PAGE , transferred to PVDF membranes , and immunoblotted for Flag ( Cell Signaling Technology , 2368; 1:1000 ) ; β1D-integrin ( Millipore , MAB1900 , 1:500 ) ; α7-Integrin ( Santa Cruz Biotechnology , sc-27706; 1:200 ) ; and β-dystroglycan ( Development Studies Hybridoma Bank , MANDAG2 clone 7D11; 1:100 ) . β1D-integrin positive intracellular vesicles were isolated from quadriceps using an endoplasmic reticulum isolation kit ( Sigma Aldrich , ER0100 ) , according to the manufacturer’s instructions . Briefly , tissues were homogenized in isotonic extraction buffer using a 2 ml Dounce homogenizer . Homogenates were cleared by centrifugation at 12 , 000 g for 15 min at 4°C . Two mg of vesicles was incubated with for 12 hr at 4°C with an antibody raised against the cytoplasmic domain of β1D-integrin ( Millipore , MAB1900 ) , immunoprecipitated using A/G magnetic beats ( ThermoFisher Scientific , #88803 ) at 4°C for 1hr , and subsequently subjected to SDS-PAGE . All protein concentrations were determined using DC Protein Assay Kit ( Bio-Rad , Hercules , CA , #5000111 ) . Then , 5X Laemmli buffer was added to protein preparations , which were then heated to 95°C for 5 min , and equal quantities were subjected to SDS-PAGE . In all instances , the wet transfer method was utilized with PVDF membranes ( Millipore , IPVH00010 ) . Staining of non-specific bands on PVDF membranes with Ponceau S solution ( Sigma-Aldrich , P7170 ) was used as a loading control for sarcolemmal isolates . Approximately 5% blotto ( nonfat dry milk [Carnation] in TBS with 0 . 2% Tween 20 [ThermoFisher Scientific] ) was used to block membranes for 45 min at RT and to incubate the membranes in primary antibodies overnight at 4°C . Two primary antibodies used to detect Thbs4 ( R&D Systems , AF2390 shown in Figure 1D , or Santa Cruz Biotechnology , sc-7657-R shown in all other figure panels as a doublet for Thbs4; 1:1000 dilution for both ) in our transgenic and gene deleted muscles . For ATF6α ( Abcam , ab37149 at 1:1000 ) , the 50 kDa cleaved active form , which resides primarily in the nucleus , was shown in all the western blots . Other primary antibodies used in this study included: α-actinin ( sarcomeric , A7811; 1/1000 ) ; Armet ( Abcam , ab67271; 1:1000 ) ; BiP ( Cell Signaling Technology , Danvers , MA , 3177; 1:1000 ) ; calreticulin ( Cell Signaling Technology , 2891; 1:1000 ) ; dihydropyridine receptor α1 ( Cav1 . 1; Thermo Fisher Scientific , MA3-920; 1:1000 ) ; dysferlin ( Abcam , ab124684 [JAI-1-49-3]; 1:1000 ) ; CLIC3 ( Santa Cruz , sc-390006; 1:200 ) ; α-dystroglycan ( EMD Millipore , 05–593; 1:500 ) ; β-dystroglycan ( Development Studies Hybridoma Bank , MANDAG2 clone 7D11; 1:100 ) ; dystrophin ( Sigma-Aldrich , D8043; 1:1000 ) ; gadph ( Fitzgerald , Acton , MA , 10R-G109A; 1:10000 ) ; GFP ( Abcam , Ab290; 1:1000 ) ; Flag ( Cell Signaling Technology , 2368; 1:1000 ) ; α5-integrin ( EMD Millipore , Billerica , MA , AB1928; 1/1000 ) ; α7-integrin ( Abacm , ab203254; 1:1000 ) ; β1D-integrin ( Millipore , MAB1900 , 1:500 ) ; Laminin ( Abcam , ab11575; 1/1000 ) ; Pdi ( Cell Signaling Technology , 2446; 1:1000 ) ; rab3 ( Abcam , ab3336; 1:1000 ) ; rab4 ( Cell Signaling Technology , 2167; 1:1000 ) ; rab5 ( Sigma-Aldrich , R7904; 1:500 ) ; rab6 ( Cell Signaling Technology , 4879; 1:1000 ) ; rab7 ( Cell Signaling Technology , 9367; 1:1000 ) ; rab8 ( Abcam , ab188574; 1:1000 ) ; rab11 ( Abcam , ab95375; 1:1000 ) ; rab24 ( BD Biosciences , Franklin Lakes , NJ , 612174; 1:1000 ) ; Sar1 ( Abcam , ab125871; 1:1000 ) ; α-sarcoglycan ( Development Studies Hybridoma Bank , IVD3 ( 1 ) A9; 1:100 ) ; β-sarcoglycan ( Novus Biologicals , Littleton , CO , NBP1-90300; 1:1000 ) ; δ-sarcoglycan ( Abcam , ab137101; 1:500 ) ; sarcospan ( Santa Cruz Biotechnology , sc-393187; 1:200 ) ; α-tubulin ( Santa Cruz Biotechnology , sc-8035; 1:1000 dilution ) ; utrophin ( Development Studies Hybridoma Bank , MANCHO3 ( 8A4 ) ; 1:50 ) and Wash1 ( Abcam , ab157592; 1:1000 ) . In all instances , appropriate IgG-AP conjugated secondary antibodies ( Santa Cruz Biotechnology , 1:2500 ) were utilized and membranes were either exposed to ECF substrate ( Amersham Biosciences , GE Healthcare , Buckinghamshire , England , RPN5785 ) and visualized using a Gel-Doc XR+ system with Image Lab Software ( Bio-Rad Laboratories ) or probed with the appropriate secondary antibodies and visualized using the Odyssey CLx Imaging System ( both Li-COR Biosciences , Lincoln , NE ) . Semi-quantitative analysis presented in Figure 1—figure supplement 1 and Figure 7—figure supplement 1 was performed using ImageJ software . To assess the exercise capacity of mice and sarcolemmal stability , mice were subjected to forced treadmill running in the presence of EBD as previously described ( Goonasekera et al . , 2011 ) . Briefly , adult mice were intraperitoneally injected with EBD ( 10 mg/ml; 0 . 1 ml per 10g body weight ) and 24 hr later subjected to forced downhill treadmill running to measure membrane rupture events . For the exercise protocol , exhaustion of the mice was assessed as greater than 10 consecutive seconds on the shock grid without attempting to re-engage running on the treadmill . Mice were then sacrificed and quadriceps , gastrocnemius and diaphragm was embedded in O . C . T . and frozen in liquid nitrogen . Tissue sections were cut at a thickness of 7 μm , air-dried , washed in PBS and stained with wheat germ agglutinin conjugated to FITC ( Sigma-Aldrich , green ) for 1 hr at RT to visualize the membranes . Images were taken on a Nikon Eclipse Ti-S inverted microscope system equipped with NIS Elements Advanced Research ( AR ) microscope imaging software ( Nikon Instruments Inc . Melville , NY ) to determine the percentage of EBD-positive fibers . In addition , blood was taken from a separate cohort of un-exercised mice of each genotype to evaluate their baseline serum CK levels as previously described at the clinical laboratory of Cincinnati Children’s Hospital Medical Center by an observer blinded to the genotypes ( Kobayashi et al . , 2008 ) . Mice were anesthetized with an intraperitoneal injection of pentabarbitol and placed supine on the muscle testing apparatus ( Aurora Scientific , Aurora , ON , Canada ) . A midline incision running from the ankle to the thigh was created and the skin and fascia was gently removed leaving the tibialis anterior ( TA ) muscle exposed . The leg was immobilized by securing it in a custom jig ( Aurora Scientific ) with thumbscrews at the distal femur . A 4–0 nylon suture was tied to the distal TA securing it with a small plastic ring at the muscle tendon junction . The distal tendon was transected and the TA elevated to remove its contact with the tibia , and the muscle was mounted to a servomotor ( Aurora Scientific , 305C ) using the plastic ring . Two intramuscular electrodes were placed on either side of the peroneal nerve and stimulation voltages and optimal muscle length ( L0 ) were determined and then adjusted to produce maximal isometric force ( P0 ) at a stimulation frequency of 200 Hz . Five consecutive isometric contractions were averaged as a measure of the maximal specific tension . An additional lengthening contraction injury protocol was added in which an isometric contraction was performed to determine baseline force generation , followed by 2 consecutive 20% L0 lengthening contractions , and finally another isometric contraction performed at L0 . The force deficit was calculated between the first isometric contraction and those following the lengthening injury cycles ( See Figure 4A for schematic representation ) . This contraction injury protocol was repeated 2 more times and the force deficit was calculated relative to the pre-injury isometric contraction . In a subset of experiments passive tension was measured prior to the lengthening contraction protocol . Here the TA was set to L0 and passively stretched to 5 , 10 , 15 , and 20% of L0 . Each stretch was held for 2 min before the length was returned to its starting position . Maximal passive tension was recorded at the peak of the stretch . A 2-min rest period occurred between each contraction for all experiments and force values were normalized to the muscle’s physiologic cross-sectional area . In some cases we observed tendon breaks during the lengthening contraction protocol . Percentage of tendon breaks , assessed by complete physical rupture of the tendon at the muscle aponeurosis , was recorded . C2C12 mouse myoblasts ( ATCC , ATCC-1772 ) were plated in either Ibidi μ-slide 8-well dishes ( Ibidi USA , Cat# 80826 ) or in 6-well plates ( Fisher Scientific , Cat# 353046 ) and maintained in DMEM/high glucose ( Fisher Scientific , Cat# SH30022 . 01 ) supplemented with 10% bovine growth serum ( Fisher Scientific , Cat# SH3054103 ) and 1x penicillin-streptomycin ( Cellgro 30-0002-CI , Mediatech , Corning Life Sciences ) at 37°C in 5% CO2 . Cells were kept to a maximum of eight passages . For differentiation , cells were grown to confluence and then switched to DMEM/high glucose supplemented with 2% donor equine serum ( Life Technologies , Cat# 26050088 ) and 1x penicillin-streptomycin for five days . Alexa Fluor-488 Microscale Protein Labeling Kit ( Life Technologies , Cat#30006 ) and EZ-Link Sulfo-NHS-Biotin ( Thermo Fisher Scientific , Cat# 21217 ) were used to label recombinant mouse Thbs4 ( rThbs4 , R&D Systems , Cat# 7860-TH-050 ) according to the manufacturer’s instructions . Equal amounts of Alexa-488 labeled bovine serum albumin ( BSA , Fisher Scientific , Cat# BP1605-100G ) were used as control . To assess internalization of rThbs4 by cultured C2C12 myoblasts and myotubes , established approaches were used as previously described ( Chlenski et al . , 2011; Nakamura et al . , 2014 ) . First , cells plated in Ibidi μ-slide 8-well dishes were treated with either 1 μg/ml Alexa-488 labeled rThbs4 or equal amounts of Alexa-488 labeled BSA control for the indicated periods . Next , cells were rinsed with sterile PBS and fixed with 4% paraformaldehyde for 10 min . After fixation cells were washed three times with PBS , followed by blocking with 3% normal goat serum/PBS/0 . 1% Triton for 20 min at room temperature and subsequently incubated with Alexa Fluor-568 labeled phalloidin ( Life Sciences , Cat# A12380; 1/100 in blocking buffer ) at room temperature to visualize the F-actin cytoskeleton or incubated with anti-Rab7 ( late endosomes; Cell Signaling , #9367; 1/100 in blocking buffer ) overnight at 4°C , followed by an Alexa fluor-568 ( red ) secondary antibody for 45 min ( Invitrogen , 1:400 in blocking buffer ) . In both conditions nuclei were counterstained with DAPI nuclear DNA stain ( Invitrogen , 1:10 . 000 ) , mounted in Ibidi mounting medium for fluorescent microscopy ( Ibidi USA , Cat# 50001 ) and visualized using a Nikon A1 confocal laser microscope system ( Nikon Instruments Inc . Melville , NY ) as described above . In parallel , cells plated in 6-well plates were treated with either 1 μg/ml Biotin labeled rThbs4 . Next , cells were rinsed with PBS , lysates were prepared as described above and intracellular biotin labeled proteins were visualized and quantified by western blot analysis using Streptavidin DyLight 650 conjugate ( ThermoFisher Scientific , Cat# 84547 ) at a 1:1000 dilution on the Odyssey CLx Imaging System ( Li-COR Biosciences , Lincoln , NE ) . All experiments described above were performed in triplicate . Flexor digitorum brevis ( FDB ) muscle fibers were isolated from male age-matched mice of each genotype as previously described and plated onto 35 mm glass-bottomed MatTek dishes ( MatTek Corp . , Ashland MA , P35G-0-10-C ) in isotonic Tyrode buffer containing 1 . 25 mM Ca2+ ( Cai et al . , 2009 ) . Membrane damage was induced in the presence of 2 . 5 μM FM1-43 dye ( Molecular Probes , Eugene OR ) using a Nikon A1 confocal laser microscope through a Plan Apo 60x H2O immersion objective . To induce damage , a 5x5 pixel area of the sarcolemma on the surface of the muscle fiber was irradiated using a UV laser at full power ( 80 mW , 351/364 ) for 10 s at t = 60 s ( Cai et al . , 2009 ) . Images were captured 5 min after irradiation at 5-second intervals ( Cai et al . , 2009 ) . For each image , fluorescence intensity in an area of about 200 μm2 directly adjacent to the injury site was measured using ImageJ software . To allow for statistical analysis from different experiments , data are presented as fluorescence intensity relative to the value before injury ( ΔF/F0 ) . Recombinant adenoviruses harboring Thbs4 , the N-terminal Laminin G ( Ad-LamG ) domain of Thbs4 , the Type III repeat ( Ad-T3R ) domain of Thbs4 , a Thbs4 Ca2+-binding mutant containing mutations in six DXDXDG calcium-binding sites within the T3R domain of mouse Thbs4 ( Ad-Thbs4-mCa2+ ) , constitutively-nuclear ATF6α ( Ad-ATF6α-CN , amino acids 1–364 ) , the ER luminal domain of ATF6α ( amino acids 448–570 ) with a C-terminal KDEL ER retention signal ( Ad-ATF6α-DN ) and β-gal control were previously generated and validated ( Brody et al . , 2016; Lynch et al . , 2012 ) . Adenoviral LamG , T3R and Thbs4-mCa2+ were engineered to contain the N-terminal signal peptide and the coiled-coil domain of Thbs4 to ensure proper intracellular trafficking and assembly and oligomerization in the ER ( Brody et al . , 2016 ) . The cDNAs of human Nell2 ( Harvard Plasmids , HsCD00331038 ) and eGFP-tagged VSV-G-ts045 ( VSVG-eGFP , Addgene: Plasmid #11912 deposited by Dr . Jennifer Lippincott-Schwartz ) were amplified by PCR for insertion into the pAdenoX-CMV vector ( Clontech , Mountain View , CA ) and transfected into HEK cells to generate recombinant adenovirus following manufacturer’s instructions ( Brody et al . , 2016; Patterson et al . , 2008 ) . Experiments were performed as previously described ( Brody et al . , 2016 ) . Briefly , either purified AdThbs4-Flag , AdThbs4-mCa2+ -Flag or Adβgal control were injected into the left and right gastrocnemius muscle of individual one-day-old Sprague Dawley rat pups ( Envigo , Indianapolis IN , USA ) , followed by an additional injection 48 hr later ( 108 viral particles for each injection ) . Rat pups were sacrificed at eight days of age and muscles were embedded in O . C . T . frozen , and 10 μm cryosections were generated . To visualize adenoviral transduction ( Flag ) and the effects of our constructs on the membrane residency of β1 integrin , tissue sections washed three times with PBS , followed by blocking with 3% normal goat serum/PBS/0 . 1% Triton for 30 min at room temperature and then incubated with anti-flag and anti-β1 integrin primary antibodies ( Cell Signaling , #2368; 1:500 and EMD Millipore , MAB1997 , 1:100 , respectively , in blocking buffer ) overnight at 4°C . Primary antibodies were detected by applying Alexa Fluor-488 conjugated goat-anti-rabbit and biotinylated anti–mouse ( Vector Laboratories , M . O . M kit , 1:500 ) followed by Alexa Fluor-568 streptavidin conjugate ( Invitrogen , 1:200 ) for 45 min at RT . Sections were mounted in Vectashield Hard Set ( Vector Laboratories , H-1400 ) to prevent photobleaching and imaged as described above . Sarcolemmal localized β1 integrin was quantified as percentage of red area per square surface of adenoviral transduced ( Flag-positive ) myofibers using ImageJ software . Low transducibility with adenovirus prevented us from conducting trafficking assays in C2C12 myoblasts or myotubes . Hence , all live cell imaging was performed using primary neonatal rat ventricular myocytes ( NRVMs ) . NRVMs were prepared from 1- to 2-day-old Sprague-Dawley rat pups as previously described ( Lynch et al . , 2012 ) . Exactly 50x103 cells were plated in either Ibidi μ-slide 8-well dishes ( Ibidi USA , Inc . Madison , WI , Cat# 80826 ) or in 35 mm glass-bottomed laminin-coated MatTek culture dishes ( MatTek Corp . , P35G-0-10-C ) and cultured in HyClone Medium 199/EBSS ( ThermoScientific , SH30253FS ) supplemented with 2 . 5% fetal bovine serum ( Sigma-Aldrich , F2442 ) and 1x penicillin-streptomycin ( Cellgro 30-0002-CI , Mediatech , Corning Life Sciences , Tewksbury , MA ) . The next day , NRVMs were infected with adenoviruses harboring the protein of interest ( see section adenoviruses , and Figure 5 ) for 3 hr in serum-free media after which they were switched back to culture media supplemented with 2 . 5% fetal bovine serum . Life cell quantitative imaging and photobleaching to evaluate ER to Golgi ( FRAP ) and Golgi to membrane vesicular trafficking ( iFRAP ) was performed using a Nikon A1 confocal laser microscope system and equipped with Plan Apo 40x oil immersion objective ( NA = 1 . 0 ) , an INU-TIZ-F1 stage top incubator ( Tokai hit CO , Ltd , Shizuoka-ken , Japan ) and NIS Elements AR microscope imaging software ( Nikon Instruments Inc . ) as previously described with modifications ( Hirschberg et al . , 1998; Patterson et al . , 2008; Zaal et al . , 1999 ) . For ER-to-Golgi protein trafficking experiments , 24 hr after adenoviral infection , Ibidi dishes were infected with CellLight Golgi-RFP Bacmam 2 . 0 ( ThermoFisher Scientific , c10593 ) , a baculovirus containing a fusion construct of human Golgi resident enzyme N-acetylgalactosaminyltransferase and TagRFP ( GalNacT2-RFP ) , according to manufacturer’s instructions and incubated overnight . The next day , 100 μg/ml cycloheximide ( Sigma-Aldrich , C4859 ) was added to the NRVMs to block new protein synthesis , 30 min prior to imaging . After acquiring a few baseline images , fluorescence from the Golgi pool of RFP was bleached by irradiating a region of interest ( ROI ) that encompasses the juxtanuclear Golgi region with a high intensity laser at 561 nm ( 100% laser power; Figure 5—figure supplement 1A ) . Next , recovery of GalNacT2-RFP was monitored by time-lapse imaging ( 5% laser power ) at 5-min intervals for 2 hr as a measure of ER to Golgi protein trafficking . For Golgi to membrane protein trafficking experiments , 24 hr after initial adenoviral infection , NRVMs in MatTek dishes were infected with adenovirus harboring the temperature sensitive VSVG-eGFP and incubated at 40°C for 24 hr to retain the VSVG-eGFP in the ER ( Hirschberg et al . , 1998; Patterson et al . , 2008 ) . Approximately 60 min prior to imaging , 100 μg/ml cycloheximide was added to the cells . Thirty minutes prior to imaging MatTek dishes were shifted to 32°C allowing the VSVG-eGFP to traffic to the Golgi . After acquiring a few baseline images , the cargo pool in the Golgi was selectively highlighted by photobleaching VSVG-eGFP from the entire cell excluding the perinucelear Golgi network using a high intensity laser at 488 nm ( 100% laser power; iFRAP; Figure 5—figure supplement 1B ) . Then , time-lapse imaging ( 5% laser power ) at 1-minute intervals for 2 hr was performed to monitor export of VSVG-eGFP molecules from the Golgi using a second area that encompasses the Golgi network as a measurement for Golgi to membrane protein trafficking . The combination of low energy , high attenuation , and the less concentrated excitation laser beam caused by the low NA objective resulted in negligible photobleaching during repetitive imaging in all experiments . As such , control experiments performing either time-lapsed imaging for 2 hr or the above described FRAP experiment in Golgi-RFP expressing cells in the presence of cycloheximide and brefeldin A ( Sigma-Aldrich , B5936; 5 μg/ml ) and iFRAP experiment in VSVG-eGFP expressing cells in the presence of cycloheximide and AlF ( AlCl3 , 60 μM and NaF , 20 μM; 30 min after shift to 32°C ) showed no difference in Golgi fluorescence intensity ( data not shown ) ( Hirschberg et al . , 1998; Patterson et al . , 2008 ) . Analysis of FRAP and loss of fluorescence after inverse FRAP ( iFRAP ) experiments was performed as previously described ( Patterson et al . , 2008; Zaal et al . , 1999 ) . The Golgi fluorescent values were normalized to the average baseline Golgi fluorescence prior to FRAP or to the first data point after iFRAP . Transgenic UAS-Thbs4 Drosophila were generated by P-element-mediated insertion ( Rainbow Transgenic Flies , Inc . , Camarillo , CA ) . Briefly , the mouse Thbs4 cDNA was cloned into the GAL4-responsive pUAST expression vector at the EcoRI site to yield the UAS-mThbs4 clone . UAS-mThbs4 transgenic Drosophila were generated by microinjection into y[1]w[1118] embryos . The MEF2-Gal4 driver line was obtained from the Bloomington Stock Center ( Indiana University , Bloomington , IN ) . The Drosophila model of MD that lacks the δ-sarcoglycan-like gene ( Sgcd840 ) and UAS-Tsp Drosophila were described previously ( Allikian et al . , 2007; Subramanian et al . , 2007 ) . The following genotypes were created for the current study: Sgcd[+];UAS-mThbs4/+;MEF2-Gal4/+ , Sgcd840 ;UAS-mThbs4/+;MEF2-Gal4/+ , Sgcd[+];UAS-Tsp/+;MEF2-Gal4/+ , Sgcd840 ;UAS-Tsp/+;MEF2-Gal4/+ , Sgcd[+];UAS-mThbs4/+ , Sgcd[+];UAS-Tsp/+ and Sgcd[+];MEF2-Gal4/+ . All stocks were raised and maintained on a standard cornmeal-molasses-yeast medium and kept at 25°C on 12:12 light–dark cycle , with 20–40% relative humidity . Male Drosophila of each genotype were collected at one-day posteclosion . Throughout the study , Drosophila were aged at 25°C with a maximum of 12 flies in 25x95 mm polystyrene vials ( Fischer Scientific , AS515 ) and transferred to new vials containing fresh food every three to four days without the use of anesthesia . Survival was recorded every day until 40 days of age . Kaplan–Meier statistical analysis was performed and significance determined by log-rank ( Mantel-Cox ) tests . The negative geotaxis assay was performed as previously described ( Allikian et al . , 2007 ) . For each genotyped tested , male Drosophila were collected and kept at no more than 12 flies per vial . At 2 , 5 , 10 , 15 and 20 days of age , they were immobilized using CO2 , and 12 groups of the various genotypes were transferred into empty polystyrene vials with a line drawn at 80 mm from the base of the vial . ( Fisher Scientific , AS515 ) . Drosophila were allowed to recover for 1 hr before testing . Each vial was assayed by gently tapping the flies down to the bottom of the vial , thereby engaging their negative geotactic response . The number of Drosophila able to climb across the 80-mm against gravity in 10 s was recorded . Four separate trials were performed with a 1-minute resting period in between . Percentage of Drosophila across were averaged and expressed as ‘physical performance’ . Genotypes were assayed simultaneously to eliminate variability attributed to RT and room humidity . To obtain longitudinal sections of the dorsal median indirect flight muscles , 30-day old Drosophila of each genotype were positioned into mounting collars ( Genesee Scientific , San Diego , CA 48–100 ) and subsequently fixed overnight at 4°C in Carnoy’s solution ( 6:3:1 ethanol:chlorophorm:acetic acid ) , dehydrated , and infiltrated with paraffin . Longitudinal sections were cut at 7 µm thickness and stained for H&E to evaluate dystrophic pathology in the dorsal median indirect flight muscle . To obtain cryosections of the dorsal median indirect flight muscles , the same mounting collars were positioned into O . C . T . freezing medium and placed into liquid nitrogen cooled-isopentane . For Thbs4 , Tsp and calreticulin immunohistochemistry , paraffin sections were cleared of paraffin in xylene ( 2 times for 4 min ) , rehydrated in ethanol ( 100% EtOH 2 X 4’ , 95% EtOH 1 X 3’ , 70%EtOH 1 X 2’ , H2O 1’ ) , incubated for 15 min in PBT ( PBS/0 . 2% Triton X-100 ) and blocked in PBTB ( PBT , 2% BSA ) for one hour at RT . The sections were incubated with primary antibody overnight at 4°C ( Thbs4: Santa Cruz , Sc-7657-R , 1:50 dilution in PBTB; Tsp , 1/50 [Subramanian et al . , 2007]; calreticulin: Abcam , ab2907; 1:100 dilution in PBTB ) , rinsed with PBT and incubated with the appropriate secondary antibody ( goat-anti-rabbit-Alexa Fluor-488 , Invitrogen; 1:400 in PBTB ) for 2 hr at RT along with counterstains . Counterstains included membrane marker WGA conjugated to Alexa Fluor-647 ( Invitrogen , W32466; 5 μg/ml , Far Red ) and/or a nuclear marker DAPI ( Invitrogen , D3571; 1 μg/ml in H2O ) . For Drosophila integrin subunit βPS immunostaining , cryosections were post-fixed in ice-cold methanol for 10 min , washed in PBT , blocked for one hour at room temperature with PBTB and incubated with primary antibody overnight at 4°C ( DSHB: CF . 6G11; 1:50 in PBTB ) . Then , tissue was then washed in PBS and subsequently incubated with rabbit-anti-mouse-Alexa Fluor-488 ( 1:300 in PBTB; Invitrogen ) for 2 hr . For each immunostain , consecutive sections were incubated with secondary antibodies alone as a control for specificity ( not shown ) . Images were obtained using a Nikon A1 confocal laser microscope system equipped with 40x H2O immersion objective ( NA = 1 . 1 ) and NIS Elements Advanced Research ( AR ) microscope imaging software ( Nikon Instruments Inc . ) . For Drosophila immunoblotting , twenty 30-day old flies of each genotype were pooled and snap-frozen prior to protein extraction . Total Drosophila protein extraction and immunoblotting was essentially performed as described above . Primary antibodies used included: Thbs4 ( Santa Cruz , Sc-7657-R; 1:1000 ) ; Tsp; 1/100 [Subramanian et al . , 2007] , BiP ( GeneTex , GTX48663; 1/1000 ) , calreticulin ( Abcam , ab2907; 1:100 ) , PDI ( Abcam , ab190883; 1:1000 ) , βPS integrin ( DSHB: CF . 6G11; 1:100 ) and β-actin ( Abcam , ab8227; 1:2000 ) . Fresh quadriceps was harvested and immediately immersed in relaxing buffer ( 0 . 15% sucrose , 5% dextrose , 100 mM KCl in PBS ) , subsequently fixed overnight ( 3 . 5% glutaraldehyde , 0 . 15% sucrose in 0 . 1 M sodium cacodylate ph 7 . 4 ) and post-fixed in 1% OsO4 ( in water ) for 2 hr at RT . Next they were washed , dehydrated and embedded using Epoxy resin . Tissue processing for electron microscopy of the dorsal median inferior flight muscle was performed as previously described ( Allikian et al . , 2007 ) . Briefly , 25 day-old flies were positioned dorsal side up on a spot of OTC freezing medium , dipped in liquid nitrogen , bisected sagittally using a pre-cooled razor and then fixed overnight in 2 . 5% glutaraldehyde in NaH2PO4 0 . 1 M pH 7 . 4 at 48°C . Next , they were washed and post-fixed in 2% osmic acid in phosphate buffer for 2 hr at room temperature , dehydrated and embedded using Epon resin . Ultrathin sections of all tissues were counterstained with 1 . 5% uranyl acetate , 70% ethanol and lead nitrate/Na citrate . Images were obtained using a Hitachi 7600 transmission electron microscope connected to an AMT digital camera . Sub-sarcolemmal vesicular expansion in myofibers of mouse quadriceps relative to the length of the sarcolemma was determined from 8 to 10 images per myofiber at 2000X of non-overlapping longitudinal regions that were randomly collected . In each image , the size of the vesicular content perpendicular to the sarcolemma -between the sarcolemma and the first sarcomere , and the length of the sarcolemma was determined using ImageJ software . The number of sarcomeric tears per myofiber in Drosophila dorsal median inferior flight muscle was determined by imaging complete myofibers at 1500X from longitudinal sections . Clear disruptions within the sarcomeric structure as indicated in Figure 8G and the associated Figure 8—figure supplement 1G . For immunogold labeling of Thbs4 , mouse quadriceps and Drosophila dorsal median inferior flight muscle were fixed with 4% paraformaldehyde in 0 . 1 M phosphate buffer ( pH 7 . 4 ) overnight . After 2 buffer washes , samples were post-fixed with 0 . 1% OsO4 in the same buffer for 30 min , dehydrated and embedded in hydrophilic acrylic resin following manufacture instruction ( L . R . White , #14380; Electron Microscopy Sciences , Hatfield , PA ) . Ultrathin sections were cut using a Leica UltraCut S or UC6rt ultramicrotome at a thickness of 90 nm and placed on Formvar and carbon coated 200-mesh nickel for immunogold labeling of Thbs4 . Briefly , ultrathin sections on grids were first treated with 1% sodium metaperiodate for 60 s to quench OsO4 . After several washes with distilled water , grids were placed on drops of PBS containing 5% BSA and 0 . 1% cold-water fish gelatin to block nonspecific binding . Sections were then incubated overnight at 4°C with goat anti-thrombospondin4 polyclonal primary antibody ( R&D Systems , AF2390 ) at a final concentration of 5 µg/ml . Following several washes , sections were incubated in 6 nm colloidal gold particles conjugated rabbit anti-goat ( Electron Microscopy Sciences; #25223 ) at a concentration of 10–20 µg/ml for 2 hr . After additional washes , all ultrathin sections were stained with 5% uranyl acetate for 2 min and 2% lead citrate for 15 min . For each experiment , both non-transgenic and transgenic sections were labeled and consecutive sections were incubated with secondary antibody alone as a control for specificity ( not shown ) . Immunogold labeling was imaged on a JEOL JEM-1400 transmission electron microscope ( JEOL Ltd , Japan ) equipped with a Gatan US1000 CCD camera ( Gatan , Pleasanton , CA ) . All results are presented as mean ± SEM . All data was normally distributed . Statistical analysis was performed with unpaired two-tailed Student’s t test for two independent groups or one-way ANOVA with post hoc Tukey’s test for multiple comparisons of 3 or more independent groups , as indicated in the individual figure legends . For survival analysis , Kaplan–Meier statistical analysis was performed and significance was determined by log-rank ( Mantel-Cox ) tests . All statistics were performed using GraphPad Prism 5 . 0 for Mac OS X and values were considered statistical significant when p<0 . 05 . No statistical analysis was used to predetermine sample size . The experiments were not randomized and no animals were excluded from analysis . The investigators were not blinded to allocation during experiments and outcome assessment , except for data displayed in Figure 2B , E; Figure 2—figure supplement 1B , C , E , F; Figure 2—figure supplement 2D , E , Figure 3B , C , F , G; Figure 3—figure supplement 1B , C; Figure 7F , I; Figure 7—figure supplement 2B; Figure 8A , B; Figure 8—figure supplement 1A , B . The exact number of animals or biological replicates for each experiment is indicated in the figure legends . Sample size for the mouse and Drosophila experiments were estimated based on previous experiments with similar procedures but also based on past power calculations for appropriate group sizes , and based on this all data reported here were based on adequate sampling . No outlier data were excluded . | Muscle cells , also known as myofibers , need to be robust in order to withstand the physical stresses of contracting and relaxing . As a result , the cell surface membrane that surrounds myofibers is more strongly anchored to its surroundings than that of other cells . Muscular dystrophies are a group of muscle-wasting disorders that usually arise when this surface membrane becomes less stable . For example , mutations that affect a protein called dystrophin-glycoprotein or integrin protein complexes can cause muscular dystrophy since these proteins normally keep the membrane anchored and stable when the muscle contracts and relaxes . When myofibers in mammals become injured , as is the case during muscular dystrophy , they produce more proteins called thrombospondins – with thrombospondin-4 being the most common . However , until now it was not clear what these proteins did in muscle cells . Vanhoutte et al . hypothesized that thrombospondin-4 may protect injured myofibers and tested their theory by first deleting the gene for thrombospondin-4 from mutant mice that were predisposed to develop muscular dystrophy . This worsened the muscle wasting in the mutant mice , and furthermore , deleting the gene for thrombospondin-4 also caused otherwise normal mice to develop muscular dystrophy in their old age . Conversely , when Vanhoutte et al . artificially increased the levels of thrombospondin-4 in the myofibers , it protected the mice against muscular dystrophy . Additional experiments conducted in fruit flies demonstrated that the protective effects of thrombospondin are conserved or similar in insects too . Lastly , biochemical experiments in mouse and rat cells showed that thrombospondin-4 aids dystrophin-glycoproteins and integrins in getting to the cell surface membrane , increasing its stability . Overall these findings provide a clearer picture of the molecular underpinnings of muscular dystrophies . In the future , more experiments will have to focus on exactly how thrombospondins stabilize and direct dystrophin-glycoproteins and integrins to the cell surface membrane . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2016 | Thrombospondin expression in myofibers stabilizes muscle membranes |
Allostery is an inherent feature of proteins , but it remains challenging to reveal the mechanisms by which allosteric signals propagate . A clearer understanding of this intrinsic circuitry would afford new opportunities to modulate protein function . Here , we have identified allosteric sites in protein tyrosine phosphatase 1B ( PTP1B ) by combining multiple-temperature X-ray crystallography experiments and structure determination from hundreds of individual small-molecule fragment soaks . New modeling approaches reveal 'hidden' low-occupancy conformational states for protein and ligands . Our results converge on allosteric sites that are conformationally coupled to the active-site WPD loop and are hotspots for fragment binding . Targeting one of these sites with covalently tethered molecules or mutations allosterically inhibits enzyme activity . Overall , this work demonstrates how the ensemble nature of macromolecular structure , revealed here by multitemperature crystallography , can elucidate allosteric mechanisms and open new doors for long-range control of protein function .
Proteins are collections of atoms that are mechanically coupled to one another , which gives rise to coordinated motions within the constraints of the folded structure . These motions are critical for many processes in molecular biology , including small-molecule and protein:protein binding interactions , catalytic cycles in enzymes , and allosteric communication between active sites and distal regulatory sites . Allostery in particular is now recognized to occur not only in classical oligomeric proteins like hemoglobin but also in monomers -- and indeed may be inherent to nearly all protein structures ( Motlagh et al . , 2014; Gunasekaran et al . , 2004 ) . However , we do not yet understand at a fundamental level how mechanically coupled atoms underlie communication through protein structures , which prevents us from mapping their intrinsic allosteric ‘circuitry’ . Moreover , because protein surfaces are large and intermolecular interactions are complex , it is difficult to predict which surface sites can bind an effector ( such as a small molecule ) that will allosterically communicate with the active site . These gaps severely limit our ability to elucidate natural allosteric regulatory mechanisms in biology , and to exploit allosteric circuitry in proteins for therapeutic intervention with perturbations such as small molecules . One system that would benefit immensely from an improved mechanistic understanding of allostery is the archetypal protein tyrosine phosphatase , PTP1B ( also known as PTPN1 ) . PTP1B is highly validated as a therapeutic target for diabetes ( Elchebly et al . , 1999 ) and cancer ( Krishnan et al . , 2014 ) and has also been linked to Rett syndrome ( Krishnan et al . , 2015 ) . Extensive efforts have been made to develop active-site inhibitors for PTP1B . Unfortunately , active-site inhibitors in general often bind non-specifically to homologous proteins , leading to off-target cellular effects ( DeDecker , 2000 ) . Moreover , the active sites of many enzymes , including phosphatases , are highly polar , and the polar inhibitors which bind to them often suffer from poor bioavailability ( Hardy and Wells , 2004; Zhang , 2001 ) . Attempts have been made to circumvent these limitations and selectively target the active site of PTP1B -- for example , by linking non-hydrolyzable phosphotyrosine ( pTyr ) analogs that bind the active site with small-molecule fragments that bind in nearby , less conserved sites ( Zhang , 2017 ) . Nevertheless , no active-site inhibitors for PTP1B have reached clinical use , leading some to label PTP1B ‘undruggable’ . By contrast , an allosteric inhibitor that binds to a less-conserved and less-polar surface site could bypass the limitations of active-site inhibitors . Two classes of compounds have been identified that allosterically inhibit PTP1B , although each has limitations . The first class of compounds are based on a benzbromarone ( BB ) scaffold and inhibit allosterically by binding to the space normally occupied by the regulatory C-terminal α7 helix ( Wiesmann et al . , 2004 ) . Recent work combining mutagenesis , X-ray crystallography , NMR spectroscopy , and molecular dynamics simulations revealed how rotations of the α3 helix and a discrete switch of the catalytic WPD loop are impacted by these BB allosteric inhibitors ( Choy et al . , 2017 ) . Unfortunately , the BB molecules were not successfully translated to the clinic . The second class are natural products , including a molecule called MSI-1436 , that bind to multiple sites that primarily involve the disordered C-terminus ( Krishnan et al . , 2014 ) . However , the binding poses were not structurally resolved , limiting our ability to understand the molecules’ allosteric mechanism and rationally improve their potency . For example , a variant of MSI-1436 had improved inhibition but a different response to mutations at the putative binding sites , suggesting an unknown change in mechanism ( Krishnan et al . , 2018 ) . MSI-1436 passed Phase I clinical trials but was not advanced to Phase II ( Ghattas et al . , 2016 ) . A new approach to revealing the intrinsic allosteric circuitry of proteins would reveal different opportunities to develop allosteric inhibitors for PTP1B that could potentially overcome the limitations of these existing molecules . Such an approach would additionally set the stage for efforts to dissect allosteric regulatory strategies in other biologically important phospho-signaling proteins . Here , we have addressed the challenge of discovering unique opportunities for allosteric inhibition of PTP1B by taking advantage of two new techniques in X-ray crystallography that reveal minor conformational states of protein and ligands . First , multitemperature crystallography ( Keedy et al . , 2015b ) can reveal previously hidden alternative conformations that enable biological functions . Here , we use this approach in PTP1B to reveal alternative conformations that are coupled to each other , forming an allosteric network . Our findings provide support for the previously hypothesized allosteric network in PTP1B that responds to BB inhibitors ( Choy et al . , 2017 ) . Moreover , they reveal extensions of this network , including additional allosteric binding sites that are distinct from the BB site ( Figure 1 ) . Similar regions of PTP1B have been implicated as allosteric sites based on mutagenesis coupled with traditional cryogenic X-ray crystallography , molecular dynamics simulations , and NMR spectroscopy ( Choy et al . , 2017; Cui et al . , 2017 ) ; here , we complement those studies by using multitemperature crystallography to reveal in atomic detail the conformational heterogeneity that allosterically links these sites to the active site . Second , high-throughput small-molecule fragment soaking and structure determination ( Collins et al . , 2017 ) has enabled new algorithms for revealing low-occupancy ligands ( Pearce et al . , 2017 ) . We use this approach to comprehensively canvas the PTP1B surface with 1627 small-molecule fragments , 110 of which were structurally resolved in complex with PTP1B . The fragments cluster into 11 fragment-binding hotspots outside of the active site . To prioritize putative allosteric sites rather than benign binding sites , we focused on the subset of fragment-binding sites that were also conformationally coupled to the active site based on multitemperature crystallography of apo PTP1B . Strikingly , the sites chosen in this way bound more fragments than did any other sites -- suggesting that conformational heterogeneity may be important for both allostery and ligand binding . Our work builds on previous studies of these sites in PTP1B ( Choy et al . , 2017; Cui et al . , 2017 ) , which did not report chemical matter that binds to them . Finally , we use covalently tethered small molecules ( Erlanson et al . , 2004 ) at one of these sites to confirm that it is functionally linked to enzyme activity , thereby supporting our predictions from multitemperature crystallography of the apo protein . Overall , by highlighting promising allosteric sites and ligands that bind to them , our work may aid future development of potent non-covalent small-molecule allosteric inhibitors for PTP1B . More broadly , we illustrate a generalizable approach to characterizing and exploiting coupled conformational heterogeneity to enable long-range control of protein function .
To identify allosteric sites in PTP1B that can communicate with the active site , we searched for regions of the protein whose conformational heterogeneity is coupled to that of the active site . We began by examining the conformational heterogeneity of the active-site WPD loop . Transition of this loop from the open to the closed state is rate-limiting for catalysis ( Whittier et al . , 2013 ) . In the only available apo crystal structure of PTP1B in which the WPD loop is free from crystal-lattice contacts ( PDB ID 1sug ) ( Pedersen et al . , 2004 ) , the loop is modeled in the closed state . However , low-contour electron density can reveal hidden alternative conformations in protein crystal structures ( Lang et al . , 2010; Fraser et al . , 2011 ) . We therefore investigated the electron density near the WPD loop in the apo structure more closely ( Figure 2B ) . Surprisingly , upon closer inspection , the electron density strongly suggests a significant population for the open state as well ( Figure 2C , left ) . Our re-refined model with both open and closed states as alternative conformations visually accounts for the electron density around this loop much better than the original model ( Figure 2C , left ) . By contrast , when we re-refined 36 other available crystal structures of PTP1B complexed with active-site inhibitors using both open and closed loop states as putative alternative conformations , Fo-Fc difference electron density and the bimodal distribution of refined occupancies indicated the single-state models were a better fit ( Figure 2—figure supplement 1 ) . These results suggest that , even in the crystal , apo PTP1B samples both WPD loop states and that active-site inhibitors then lock the loop either fully open or fully closed . To better characterize the conformational heterogeneity of the WPD loop in apo PTP1B , we collected X-ray datasets at several elevated temperatures including 180 K , 240 K , and 278 K ( ‘room temperature’ ) in addition to the 100 K ( ‘cryogenic’ ) model from the PDB , all at better than 2 Å resolution ( Table 1 ) . Each complete dataset was obtained from a single crystal , and crystallographic statistics indicated that radiation damage was not a concern even at the elevated temperatures ( Diederichs , 2006 ) ( Figure 2—figure supplement 3 ) . We built an initial multiconformer model for each temperature using the automated algorithm qFit ( Keedy et al . , 2015a ) . These models are parsimonious in that each atom has alternative positions only if justified by the experimental data , and a single position otherwise . Such models are equally good and usually better explanations of the experimental X-ray data ( Keedy et al . , 2015a; van den Bedem et al . , 2009 ) , and have been used to understand many biologically relevant phenomena at protein:water interfaces ( Keedy et al . , 2014 ) , dynamic enzyme active sites ( Keedy et al . , 2015b; Fraser et al . , 2009 ) , and allosteric networks perturbed by mutations ( van den Bedem et al . , 2013 ) . We then manually refined alternative conformations for protein , buffer components , and solvent . In particular , we took advantage of the wealth of available structures of PTP1B in the PDB ( Berman et al . , 2000 ) to sample coordinates for putative alternative conformations; in many cases , these conformations explained missing regions with positive Fo-Fc electron density that would have otherwise been difficult to model . Removing the alternative conformations and re-refining the resulting single-conformer models , either with or without automated solvent placement , yields deteriorated statistics ( Table 1— source data 1 ) , which confirms that the multiconformer models are appropriate explanations of the experimental data at each temperature . The WPD loop adopts both the open and closed conformations across this range ( Figure 2C ) and the population of the open vs . closed states was sensitive to temperature ( Figure 2D ) . The loop is approximately 67% closed at 100 K , but 65% open at 278 K . These occupancies evolve non-linearly ( Keedy et al . , 2015b ) at intermediate temperatures . Overall , we also observed temperature-dependent conformational heterogeneity for several other regions of PTP1B , including the previously characterized BB allosteric site , plus additional sites we refer to as the ‘197 site’ and the ‘loop 16 ( L16 ) site’ . These regions are all contiguous in the structure ( Figure 2E ) , suggesting that they together constitute an expanded collective allosteric network in PTP1B that is coupled to the WPD loop . The manner in which they are connected is described in detail in the following sections . To connect these multitemperature structures to known allosteric regulatory mechanisms , we first turned to a benzbromarone derivative compound ( here referred to as BB2 ) that binds to an allosteric site >12 Å away from the active site and inhibits enzyme activity ( Wiesmann et al . , 2004 ) . The authors of the study reporting BB2 described a series of induced conformational changes that begins with BB2 directly displacing Trp291 to disorder the entire C-terminal α7 helix , and ends with Phe191 χ2 dihedral-angle rotations clashing with the WPD loop anchor residue Trp179 to stabilize the open state . We tested the hypothesis that these allosterically inhibited conformations pre-exist in apo PTP1B by examining these regions in our multitemperature apo crystal structures . Indeed , in apo PTP1B the α7 helix is more ordered at lower temperatures but more disordered at higher temperatures ( Figure 3A ) . Also , Trp179 and Phe191 adopt dual conformations at higher temperatures ( Figure 3B ) that coincide well with the apo and allosterically inhibited conformations ( Figure 3C ) . We also see alternative conformations at high temperatures for several residues within and directly flanking the WPD loop ( Arg221 , Pro185 , Trp179 , Phe269 ) which have been implicated as being important for a CH/π switch during WPD loop opening/closing ( Choy et al . , 2017 ) ( Figure 3—figure supplement 1 ) . Multiple conformations for Leu192 were more difficult to detect at higher temperatures in apo PTP1B . This is likely because Leu192 shifts more subtly between the 100 K apo and allosterically inhibited conformations , which is also consistent with a recent report that Leu192 is a relatively static inter-helical ‘wedge’ ( Choy et al . , 2017 ) . Taken together , these results suggest that BB2 stabilizes a subset of pre-existing conformations in apo PTP1B . We additionally solved a high-resolution ( 1 . 80 Å , Table 1 ) structure of PTP1B in complex with BB3 ( which differs from BB2 only by an extra terminal aminothiazole group ) at 273 K and found it to be very similar to the 100 K structures with BB3 ( PDB ID 1t4j ) and with BB2 ( PDB ID 1t49 ) despite the difference in temperature ( Figure 3—figure supplement 2 ) . However , two interesting features are evident at 273 K . First , at 273 K but not at 100 K , modeling BB3 with a single conformer leads to Fo-Fc difference electron density peaks at both ends of the molecule ( Figure 3—figure supplement 3A ) . To account for these peaks in the map , it is necessary to add a second alternative conformer to the model , which includes a translation at one end and dihedral-angle changes at the other end ( Figure 3—figure supplement 3B ) . Chemical changes to BB3 designed to eliminate this remaining heterogeneity could potentially improve affinity and inhibition . Second , at 273 K , we observe significant electron density just above BB3 ( Figure 3—figure supplement 3C ) . Modeling a reordered , non-helical conformation of α7 explains this density well , and places Trp291 in good position for aromatic stacking interactions with BB3 and other interactions with nearby sidechains on the α3 helix ( Figure 3—figure supplement 3D ) . Trp291 is displaced by BB3 or BB2 binding in a striking example of molecular mimicry ( Wiesmann et al . , 2004 ) ( Figure 3C ) . Our 273 K data suggest that a subsequent reordering of the α7 polypeptide occurs , which may contribute to the affinity of BB3 for PTP1B . In contrast to our 273 K data , electron density in this region is weak in the 100 K structures with BB3 and BB2 . However , in the 100 K structure with BB1 , a different derivative of the BB scaffold , α7 also reorders -- but adopts a significantly different conformation than we observe at 273 K with BB3 ( Figure 3—figure supplement 3E , G ) . Together , these results suggest that in addition to being a major allosteric hub when ordered ( Choy et al . , 2017 ) , α7 is also quite malleable when disordered , and may interact in diverse ways with bound ligands -- behavior which is similar to the mechanism proposed for inhibitors that bind via the disordered C-terminus beyond α7 ( Krishnan et al . , 2014 ) . We also observed temperature-dependent ordering in a loop ( loop 16 , L16; residues 237–243 ) that sits underneath the α6-α7 junction just beyond the BB binding site . By contrast to lower temperature ( Figure 4A ) , the electron density for L16 at higher temperature ( Figure 4B ) clearly reveals an alternative conformation with its backbone shifted by >5 Å from the primary conformation ( Figure 4D ) . Modeling this alternative loop conformation back into the lower-temperature models and refining its occupancy reveals a temperature dependence ( Figure 4E , Figure 4—figure supplement 1 ) that is qualitatively similar to the temperature dependences of the WPD loop . Remarkably , this L16 alternative conformation sampled by apo PTP1B matches the L16 conformation when PTP1B is allosterically inhibited by BB2 ( Figure 4C ) . This rearrangement provides further evidence that BB2 selects pre-existing , globally dispersed conformations rather than inducing new ones . The L16 site is seemingly coupled to the α6 helix: Lys239 from L16 H-bonds with Ile281 from α6 in the global closed state , but not in the global open state in which L16 adopts its alternative conformation . Because α6 is directly coupled to the α7 order-disorder transition , we therefore propose that the L16 site is a component of the collective allosteric network in PTP1B . The L16 site was not identified as part of the allosteric network in PTP1B based on a study using mutagenesis , NMR , and MD ( Choy et al . , 2017 ) . However , in a more recent study , several residues lining what we call the L16 site ( including Met3 , Lys237 , and Ser242 ) were included in a region called ‘Cluster II’ , which was suggested to be a previously unidentified allosteric site based on reciprocal NMR chemical shift perturbations upon mutation of this site or the WPD loop ( Cui et al . , 2017 ) . Our work here using multitemperature crystallography complements these findings by independently identifying this allosteric site using a new methodology , and by revealing in atomic detail how multiple conformational states at the L16 site may aid communication with the active site . Interestingly , a separate approach combining molecular dynamics and machine learning also recently pointed to this area as a potential ‘cryptic’ binding site ( Cimermancic et al . , 2016a ) . Therefore , the L16 site may be not only energetically coupled to the active site , but also capable of forming an under-appreciated small-molecule binding pocket via the conformational heterogeneity we observe . In addition to the temperature-dependent conformational heterogeneity observed at the BB site and L16 site , we observed residues with temperature-sensitive conformational heterogeneity in the ‘197 site’ ( Figure 5 ) . Moreover , the alternative conformations of several residues in this region have a pattern of steric incompatibility with multiple states of the WPD loop and α7 helix , suggesting that the 197 site may be mechanistically linked to the active site in a similar way as the BB binding site . A major link between the WPD loop and the 197 site is Tyr152 . When the WPD loop is closed and the α7 helix is ordered , Tyr152 adopts a ‘down rotamer’ ( Figure 5—figure supplement 1 , red ) . By contrast , when the WPD loop is open and the α7 helix is disordered , the 278 K electron density suggests that Tyr152 adopts an ‘up rotamer’ ( Figure 5—figure supplement 1C , orange ) . However , difference electron density peaks remain ( Figure 5—figure supplement 1C ) that indicate the presence of the down rotamer as an alternative conformation . Consistent with this interpretation , modeling just the additional down rotamer is insufficient to explain the density ( Figure 5—figure supplement 1D ) . These two rotamers are accommodated in the WPD-loop-open state by a shift of the L11 backbone ( Figure 5—figure supplement 1 ) . The down rotamer is sterically incompatible with phosphorylation of Tyr152 , which occurs in vivo ( Bandyopadhyay et al . , 1997; Rhee et al . , 2001 ) , suggesting that the up rotamer may have additional regulatory roles . Tyr152 in the L11 backbone conformation with just the down rotamer ( red in Figure 5—figure supplement 1 ) is sterically incompatible with the open WPD loop conformation ( Figure 5—figure supplement 1E ) . Similarly , the Tyr152 up rotamer is sterically incompatible with the ordered α7 conformation ( Figure 5—figure supplement 1E ) . In turn , α7 is conformationally synchronized with the WPD loop ( Figure 3A and Figure 2D ) and is a key hub connecting loop 11 and the WPD loop ( Choy et al . , 2017 ) . These results together suggest that the allosteric circuitry of PTP1B involving Tyr152 is complex and multibody . Tyr152 likely exemplifies a population shuffling mechanism whereby mixtures of microstates ( rotameric state of Tyr152 ) exchange on a fast timescale as the protein transitions between macrostates ( WPD loop state , α7 ordering , and L11 backbone shifting ) on a slower timescale ( Smith et al . , 2015 ) . Our findings thus shed additional light on the mechanism by which loop 11 allosterically communicates with the active site , thus complementing other recent studies using mutagenesis , MD , and NMR to map allostery in PTP1B ( Choy et al . , 2017; Cui et al . , 2017 ) . In our datasets at temperatures above 100 K , the electron density suggests a complex interplay between alternative conformations for Asn193 on the α3 helix and Tyr152 on loop 11 ( L11 ) ( Figure 5—figure supplement 1 ) . Asn193 is part of the α3 helix ( residues 187–202 ) , which immediately follows the WPD loop in sequence . The N-terminal region of this helix ( through Phe196 ) rotates by 2–20° , resulting in shifts of 0 . 2–0 . 7 Å for some Cα atoms , based on 100 K crystal structures of apo ( WPD-open ) vs . active-site-inhibitor-bound ( WPD-closed ) PTP1B ( Choy et al . , 2017 ) . Similarly , the multiconformer model for our 278 K apo dataset includes alternative backbone conformations for the WPD loop and the beginning of α3 , through Asn193 plus Phe196-Lys197 ( this is a conservative interpretation of which residues in the helix have alternative backbone conformations ) . Our results suggest that α3 inherently shifts as the protein transitions betweens its global macrostates , even in the apo state . Strikingly , several residues propagating down L11 from Tyr152 , and down α3 from Asn193 , also adopt multiple conformations at higher temperatures ( Figure 5 ) . These residues colocalize in a shallow pocket nestled between loop 11 , the β4 and β5 strands , and the α3 and α7 helices . We refer to this area here as the ‘197 site’ because the sidechain of Lys197 extends into the pocket . Our analysis indicates a complex , interconnected network involving multiple aromatic stacking , hydrogen-bonding , van der Waals , and electrostatic interactions . To complement this model-based assessment with a map-based approach , for several residues in the pocket we quantified electron density as a function of temperature for atom positions that are unique to the minor conformation ( i . e . do not overlap with any atoms in the major conformation ) , reasoning that residues which respond to temperature similarly may be conformationally coupled ( Keedy et al . , 2015b ) . The population of each minor conformation increases non-linearly with temperature ( Figure 5A ) in a similar fashion as the open state of the WPD loop ( Figure 2D ) and the disordered state of the α7 helix ( Figure 3A ) , in support of the idea that these various regions of the protein are mutually conformationally coupled . We next discuss several similarities and a few differences between what we refer to as the 197 site and similar regions implicated by other recent studies of allostery in PTP1B . First , in addition to predicting the L16 site ( see above ) , reciprocal chemical shift changes upon mutation suggested that several residues at both ends of the 197 site ( Tyr152 , Tyr153 , Lys150 , Arg105 ) are part of a region referred to as ‘Cluster I’ that is allosterically linked to the active site ( Cui et al . , 2017 ) . However , that study did not implicate additional key residues on the α3 helix , for example Asn193 and Lys197 . Second , mutagenesis , NMR , and cryogenic crystallography implicated several elements of our 197 site as being part of the larger allosteric network in PTP1B: loop 11 ( including Tyr152 and Tyr153 ) , the α3 helix ( especially Asn193 ) , and the α7 helix ( Choy et al . , 2017 ) . Chemical-shift-restrained molecular dynamics simulations further suggested that Tyr152 on loop 11 and Asn193 on the α3 helix have mutually coupled alternative conformations ( Choy et al . , 2017 ) . However , here we highlight additional residues ( e . g . Asp148 and Glu157 on the β strands on either end of loop 11 ) as being conformationally coupled to each other and to the rest of the allosteric network and the active site , and which may collectively form a binding pocket . Therefore , our work accomplishes two things with regard to these existing studies . First , we add support to their findings by reaching similar conclusions using orthogonal methods . Second , we complement the other studies by revealing additional amino acid residues that may play roles in binding and allosteric communication at the 197 site . We also emphasize that the 197 site is structurally distinct from the two allosteric sites that have previously been targeted with small molecules to achieve inhibition ( Krishnan et al . , 2014; Wiesmann et al . , 2004 ) , so any small molecules that bind to the 197 site would represent a distinct strategy for inhibiting PTP1B . Surprisingly , in all of our multitemperature apo structures , ordered glycerols are present not only in the active site as mentioned above but also in the 197 site ( Figure 5—figure supplement 2 ) , and MPD also binds here in another published structure ( PDB ID 2cm2 ) . These observations suggest that the 197 site may be bindable by other small molecules . We therefore hypothesized that binding of a small molecule to the 197 site could propagate changes in conformational heterogeneity to the WPD loop to interfere with catalysis . To test whether more directed perturbations to the 197 site can allosterically modulate enzyme function , we introduced ‘dynamically destructive’ mutations ( Y152G , Y153A , K197A ) that were predicted to preserve the protein’s general structure , yet interfere with the conformational heterogeneity along the putative allosteric pathway lining the 197 site by removing interactions between alternative conformations . For Y152 we chose a mutation to glycine instead of alanine to more fully disengage residue 152 from the WPD loop , given that the Cβ and Hβ atoms of Y152 sterically engage with the WPD loop ( Figure 5—figure supplement 1E ) . All three mutations indeed reduce catalytic efficiency , to varying extents: the mutation nearest to the WPD loop ( Y152G ) reduced kcat/KM the most , and the mutation farthest from the WPD loop ( K197A ) reduced kcat/KM the least ( Figure 5—figure supplement 3 ) . Our results are generally in line with reported effects for the Y152A + Y153A ( ‘YAYA’ ) double mutation ( Choy et al . , 2017 ) and for the Y153A single mutation ( Cui et al . , 2017 ) ; small discrepancies may be due in part to differences in the length of the protein construct being used . Overall , our results illustrate that local perturbations in the vicinity of the allosteric 197 site can impact catalysis . Overall , we describe a large , collectively coupled allosteric network on one contiguous face of the protein ( Figure 2E ) . This network is interconnected not only on the surface , but also within the hydrophobic core . For example , Tyr176 adopts alternative sidechain conformations at higher temperatures that differ by a small rotation of the relatively non-rotameric χ2 dihedral angle ( Lovell et al . , 2000 ) ( Figure 5—figure supplement 4 ) . The two conformations of Tyr176 are structurally compatible with different conformations of the surface-exposed Tyr152 ( Choy et al . , 2017; Cui et al . , 2017 ) in one direction , and of the buried Trp179 in the WPD loop and BB allosteric pathway ( Wiesmann et al . , 2004 ) in the other direction ( Figure 5—figure supplement 4 ) . Thus , surface residues such as Tyr152 may be conformationally coupled to the buried underside of the WPD loop via a similar mechanism as BB binding -- remotely modulating the Trp179 anchor via coordinated hydrophobic shifts -- but from a different angle of attack , via Tyr176 . Overall , such coordinated local shifts within the hydrophobic core likely ‘lubricate’ the transition between discrete global states of PTP1B . Although the results described above establish a conformationally coupled network within the structure of PTP1B , allosteric inhibition also requires binding sites for small molecules that can conformationally bias this network to modulate function . To identify potential allosteric ligand-binding sites in PTP1B , we mapped the small-molecule binding potential or ‘ligandability’ of the entire protein surface . Specifically , we used small-molecule fragments , which by virtue of their small size provide a relatively large sampling of drug-like chemical space ( Murray and Blundell , 2010 ) . Astex Pharmaceuticals has previously explored fragment-based drug design for PTP1B ( Hartshorn et al . , 2005 ) ; however , that screen used molecules pre-selected to enrich for binders to phosphatase active sites , which contrasts with our goal of exploring the surface outside of the active site . To determine cocrystal structures of hundreds of fragments with PTP1B , we used the high-throughput fragment-soaking and crystallographic pipeline available at Diamond Light Source ( Collins et al . , 2017 ) to individually soak 1918 apo PTP1B crystals with small-molecule fragments in DMSO from several curated libraries , and another 48 with just DMSO . We then used robotic sample handling to automatically collect complete X-ray datasets at 100 K ( Figure 6—source data 1 ) . Of the 1966 total soaks , 1774 yielded diffraction data that could be successfully processed . The data were generally high-resolution: the average resolution was 2 . 1 Å , 65% of resolutions were better than 2 . 0 Å , and 87% were better than 2 . 5 Å ( Figure 6A , Figure 6—source data 1 ) . The large number of datasets enabled us to use the new Pan-Dataset Density Analysis ( PanDDA ) algorithm ( Pearce et al . , 2017 ) to reveal bound fragments . PanDDA performs weighted subtractions of the ‘background’ electron density ( computed from apo and unbound datasets ) from each electron density map ( Figure 6B–C ) . The optimal subtraction , chosen by a heuristic , yields electron density corresponding to the ligand-bound fraction of unit cells in the crystal . Our PanDDA analysis of 1774 datasets revealed 381 putative binding events . We manually inspected each putative binding event , and were able to confidently model the fragment in atomic detail for 110 hits ( Figure 6D ) . Overall , 12 different sites in PTP1B were observed to bind fragments ( Figure 6E ) . These sites are structurally distinct from one another -- that is , they share no residues in common , and fragments bound within different sites do not overlap with each other . They are also widely distributed across the protein surface . Twenty-five fragments bind to multiple sites , but promiscuous binding is not unexpected from such small fragments , and still provides valuable information about favorable binding poses in each site . PanDDA initially identified >80 putative binding events in the active site . Many of these can be attributed to movements of the WPD loop ( Figure 2 ) , often induced by oxidation of the catalytic Cys215 , which is a natural regulatory mechanism ( van Montfort et al . , 2003 ) . Apart from these protein events and other false positives , we observe four fragments bound in the active site . This number is relatively low likely because our libraries were not customized to bind to the highly charged active site of PTP1B , as was the case in the Astex study ( Hartshorn et al . , 2005 ) . To identify allosteric binders , we examined sites outside of the active site . Strikingly , we observed 24 bound fragments in the BB allosteric site ( Figure 7A ) . The poses of many of these fragments overlap portions of the BB scaffold ( Figure 7A , Figure 7—figure supplement 1 ) . However , many of them also contain chemical groups that suggestively protrude in new directions from the BB scaffold ( Figure 7—figure supplement 1 ) . This retrospective result validates the idea that fragment screening identifies binding sites , and specific fragment poses in those sites , that can be fruitfully exploited for allosteric inhibition . Interestingly , in one structure with a fragment bound in the BB site , the α7 helix adopts a reordered conformation that covers the binding site ( Figure 7A ) , reminiscent of other examples in published structures and in our high-temperature datasets ( Figure 3—figure supplement 3 ) . These compounds could also inspire design of modified BB2 derivatives that may overcome the low affinity that limited the development of that series . We also examined fragments bound to the L16 site and the 197 site , which were suggested to be allosteric sites by our multitemperature analysis of apo PTP1B . Excitingly , both sites are fragment-binding hotspots: 17 fragments bind to the L16 site ( Figure 7B ) and 30 fragments bind to the 197 site ( Figure 7C ) . Thus , independent methods to assess allosteric coupling and ligandability converge on the same sites in PTP1B . Our results agree with previous studies , based on mutagenesis and NMR , which implicated several residues in the L16 site ( Cui et al . , 2017 ) and in the 197 site ( Choy et al . , 2017; Cui et al . , 2017 ) as participating in an active-site-linked allosteric network . We also add value to those studies in another way: by reporting the binding poses of a few dozen small-molecule ligands that bind to these sites in atomic detail . Because these two sites are both conformationally coupled to the active site and capable of binding a variety of small molecules , they may be promising sites to explore for small-molecule allosteric inhibition of PTP1B activity . The L16 site is between loop 16 ( L16 ) , the beginning of α1 , and the end of α6 . Most of the 17 fragments that bind here appear to ‘pry apart’ these elements ( Figure 7B ) to create a cryptic binding site ( Cimermancic et al . , 2016a ) . Because the end of α6 is coupled to the beginning of α7 , which is perhaps the central allosteric hub of PTP1B ( Choy et al . , 2017 ) , this site seems promising for allosteric inhibition . The fragments that bind here are diverse but have some common features: aromatic moieties sandwich between Pro239 ( of L16 ) and Met282 ( α6 ) , and carboxyl groups hydrogen-bond to the backbone amide of Glu2 ( α6 ) . These fragments do not spatially overlap with any fragments in the nearby BB site , confirming that the L16 site is genuinely distinct from the previously explored allosteric site . The 197 site is on the opposite side of the BB site , near α3 ( including Lys197 ) and L11 ( including Tyr152 ) . Thirty fragments bind in the 197 site , with 14 in the primary subsite near Lys197 , and 17 in a nearby but distinct subsite separated by a ‘ridge’ formed by the Gln157 and Glu170 sidechains ( Figure 7C ) ( one fragment binds in both the primary subsite and the secondary subsite ) . These fragments are characterized by packing of aromatic moieties above Leu172 , with additional aromatic or polar extensions in various directions . As with the L16 site , fragments in this site do not overlap with any fragments in the nearby BB site . However , several of the fragments in the 197 site do overlap with the positions of ordered glycerols from our multitemperature structures ( which were absent from all fragment-soaked structures to avoid competition for binding ) ( Figure 7—figure supplement 2 ) . Similarly , glycerol in PDB ID 3qkp and β-octylglucoside in PDB ID 2cmc ( among other examples ) bind to sites that are occupied by fragments in our structures . These findings emphasize that fortuitous binding of buffer components and other miscellaneous compounds can in some cases provide useful information about binding sites ( Mattos and Ringe , 1996 ) . It may be possible to link fragments in the primary subsite and secondary subsite to increase binding affinity . Although some fragments in the secondary subsite are largely stabilized by crystal-lattice contacts , they still enjoy favorable interactions with the protein that could potentially be useful for fragment extension . By contrast , the primary subsite is generally free from crystal-lattice contacts . To assess the effect of the bound fragments on the structure of PTP1B more globally , for each dataset we built an ensemble structure consisting of both the ground state and the bound state . Each dataset was modeled with an innovative PDB format as a multiconformer structure that represents both a heterogeneous apo state and a heterogeneous holo state . Due to limitations in the PDB model format and in the ability of conventional refinement programs to interpret and create reasonable restraints for this model type , either one conformation or four alternative conformations were used to describe each residue , often when only two were necessary . Due to this forced degeneracy , refinement of coordinates , occupancy , and B-factors must be highly restrained . We interpret the resulting occupancies as a good approximation of the fraction of unit cells that have a ligand present . Refining these ensemble structures using restraints that avoid overfitting allowed for some structural differences between the two states to emerge . In principle , these structural differences could give some prediction of the functional effects one might expect upon developing a higher affinity version of the molecule . The refined ensemble structures were of high quality ( Figure 7—source data 1 ) . However , generally speaking , the structural differences were subtle: the global backbone RMSD ( N , Cα , C atoms ) between the ground state and bound state ranged from 0 . 7 to 1 . 7 Å . Cases with larger RMSD ( >1 . 25 Å ) generally involved either active-site fragments that directly shift the WPD loop , or fortuitous oxidation of the active-site Cys215 ( van Montfort et al . , 2003 ) . Thus , fragment binding did not dramatically shift PTP1B from the open to the closed state in many of these structures . Many of these fragments are certainly benign binders that bind to non-allosteric sites . However , the strong preference for the open state even with fragments that bind to allosteric sites is likely due to the absence of glycerol , which is present in our multitemperature structures ( see Materials and methods ) . It is likely that weak fragments do not overcome this energetic preference , and instead elicit conformational changes primarily in their immediate vicinity . Including glycerol to place the protein in a regime in which the open vs . closed states are more nearly isoenergetic during fragment soaks could potentially interfere with fragment binding to the 197 site , since ordered glycerols also fortuitously bind there ( Figure 7—figure supplement 2 ) . The small-molecule fragments described above were identified by a naive screen and are not optimized for high-affinity binding to the 197 site or L16 site of PTP1B . Nevertheless , we selected 20 fragments that were deemed to bind in either site during early rounds of iterative PanDDA analysis ( see Materials and methods ) ( Figure 7—source data 2 ) and tested whether they have allosteric effects using enzyme activity assays . Unsurprisingly , we did not observe inhibition of enzyme activity by the fragments up to the maximum concentrations we were able to assay due to solubility of the fragments . It is important to note that this is not surprising due to the fragments’ small sizes , relatively simple chemical structures , and low affinities ( soaking with fragments at 30–150 mM concentrations resulted in observed occupancies of only 10–30% in the crystal structures ) . However , looking ahead , the dozens of cocrystal structures with small-molecule fragments bound at these promising allosteric sites ( and at the previously explored BB2 site ) that we have reported here offer a foothold for future medicinal chemistry efforts to design allosteric inhibitors for PTP1B . Instead , here we focus on an alternative strategy to validate the concept of allosteric inhibition at the 197 site: covalent binding to enhance ligand occupancy . Specifically , we used ‘Tethering’ ( Erlanson et al . , 2000; Erlanson et al . , 2004 ) , in which a residue near the site of interest is mutated to cysteine , then the mutant is mixed with disulfide fragments under partially reducing conditions . Affinity of the fragments for the site of interest drives the formation of a disulfide bond between the fragment and the adjacent cysteine . The extent of cysteine labeling can be measured using whole-protein mass spectrometry , and serves as a metric to rank the affinity of fragments for a given site . One major advantage of Tethering over other fragment-based approaches is that it can leverage low-affinity binding events into quantitatively labeled protein species , whose enzymatic activity can then be assayed . Here , we used Tethering to successfully identify a covalent allosteric inhibitor at the 197 site of PTP1B ( Figure 8 ) . For the allosteric 197 site in PTP1B , we chose to tether to a K197C mutant for several reasons . First , K197 is on the α3 helix , which is a key allosteric element in PTP1B ( Choy et al . , 2017 ) . We predicted that small-molecule tethering to our site could could perturb the helix via K197C , perhaps mimicking the effects of a free molecule binding to the WT protein and altering the K197 conformational distribution . Second , K197 and E200 are the two residues on α3 whose Cα-Cβ vectors point in roughly the correct direction toward the allosteric 197 site we describe; however , E200 engages in crystal-lattice contacts which would interfere with tethering in our P3121 space group , so we focused on K197C instead . To efficiently explore the chemical space of covalent small molecules for the 197 site , we used a library of 1600 disulfide-capped fragments designed for covalent tethering experiments ( Kathman et al . , 2014; Burlingame et al . , 2011a ) . From our initial screen , we identified 50 fragments that tethered to K197C > 3 standard deviations above the average percent tethering for all 1600 compounds ( Figure 8—figure supplement 1A ) . We next measured the ability of these top fragments to modulate PTP1B’s phosphatase activity ( Figure 8—figure supplement 1B ) . Formation of the tethered complex followed by a pNPP assay identified only one fragment , 1 ( Figure 8—figure supplement 1C ) , that appeared to inhibit PTP1B at a percentage comparable to the percentage of tethered complex ( Figure 8—figure supplement 1B ) , suggesting a direct relationship between labeling and inhibition . While 1 thus showed the behavior we desired , the percent labeling and inhibition were relatively low . We hypothesized that altering the linker between the fragment core and the disulfide bond may lead to improved interactions between the protein and small molecule . For this reason , we designed and synthesized 2 ( Figure 8A ) , which has the orientation of the amide bond reversed , allowing for one less carbon in the disulfide linker ( Figure 8—figure supplement 1C ) . When assayed , 2 showed improved tethering and inhibition of K197C relative to 1 . 2 exhibited dose-dependent tethering and partial noncompetitive allosteric inhibition of K197C with a tethering EC50 of 7 . 8 ± 1 . 1 µM and a Ki for pNPP activity of 7 . 1 ± 1 . 1 µM ( maximum inhibition of ~60% ) ( Figure 8B and Figure 8—figure supplement 2A ) . Importantly , 2 appeared to show little to no tethering of WT* and minimal inhibition , supporting that 2’s activity is specific to the 197 site and not due to tethering of the active-site cysteine found in both K197C and WT/WT* ( Figure 8C and Figure 8—figure supplement 2B ) . In fact , the inhibition that is observed for WT* does not correlate with tethering , suggesting the inhibition may be from nonspecific factors , such as aggregation , at higher concentrations of 2 ( ≥50 µM ) . Michaelis-Menten kinetic analysis of K197C in the presence of 2 ( 50 µM ) showed a statistically significant ~50% reduction in Vmax relative to DMSO treatment , but no significant effect on KM for the pNPP substrate ( Figure 8—figure supplement 2C , E , F ) . This supports a noncompetitive allosteric mechanism of inhibition . The effect on WT* kinetics was similar to the nonspecific inhibition observed in the dose titration experiment ( Figure 8—figure supplement 2D ) , once again supporting that the activity of 2 is specific for the K197 site on PTP1B . To further profile the inhibitory effect of tethering of 2 on K197C , we assayed the ability of the tethered complex to dephosphorylate the alternative substrate DiFMUP ( Welte et al . , 2005 ) . As with pNPP , the tethered complex was inhibited , with kinetic analysis showing a dramatic reduction in Vmax , but no significant effect on KM ( Figure 8—figure supplement 3 ) . These results once again support a partial noncompetitive allosteric mechanism of inhibition . To futher validate that 2 acts specifically through the K197 site and to explore the mechanism of inhibition by 2 , we solved a high-resolution ( 1 . 95 Å , Table 1 ) crystal structure of K197C tethered with 2 . The structure confirms that 2 tethers to K197C rather than to the active-site catalytic Cys215 , and also that tethered 2 resides in the 197 site ( Figure 9A ) rather than in the relatively nearby BB site , which is also theoretically within reach of the tethering linker on the other side of the α3 helix . We modeled 2 as partially populated and , indeed , the 83% refined occupancy in the crystal structure was very similar to the ~85% conjugation measured after tethering in solution prior to crystallization . 2 adopts a conformation in which the two rings are nearly coplanar . This interpretation is further validated by a ‘polder map’ , in which both the ligand and bulk solvent are omitted ( Liebschner et al . , 2017 ) ( Figure 9—figure supplement 1 ) . While coplanar biphenyl rings are typically believed to be disfavored due to steric clashes , it is possible that hydrogen bonding of D148 with the phenol combined with the electronegativity of the para-fluoro leads to delocalization of the rings’ electrons and promotes a coplanar conformation . Additionally , 2 packs against the hydrophobic floor , centered on Leu172 , of the relatively shallow binding pocket in the 197 site . Trapping of coplanar biphenyl rings covalently attached to a protein has previously been reported ( Pearson et al . , 2015 ) . To elucidate confomational changes induced by 2 , we also solved a high-resolution ( 1 . 95 Å , Table 1 ) crystal structure of apo K197C in the same crystal form for comparison . As mentioned previously , PTP1B remains in the open state without glycerol; glycerol was absent from the tethered K197C structure ( to avoid competition for binding in the 197 site ) and from the K197C apo structure ( for consistency ) , so we are unable to see any dramatic shifts in the global open-closed equilibrium that 2 may induce . Tyr153 shifts its position slightly and Tyr152 responds by shifting fully to its up rotamer , but this is likely due to the loss of interactions with the WT K197 upon mutation . Beyond these mutation-induced effects , we see some conformational changes associated with tethering of 2 . The key residue Asn193 ( Choy et al . , 2017 ) changes rotamers , the sidechain of Phe196 on α3 ‘slides’ to change its aromatic stacking arrangement with Phe280 on α6 ( Figure 9C ) , and Glu276 -- which contacts the wedge residue Leu192 ( Choy et al . , 2017 ) -- rotates sidechain dihedral angles . These sidechain movements appear to couple to subtle , more distributed backbone shifts ( Deis et al . , 2014 ) of the α3 helix , several residues of which move up toward the WPD loop by ~0 . 5 Å ( Figure 9C ) . Interestingly , these sidechain and particularly backbone movements are somewhat similar to those between the two macrostates of apo PTP1B at high temperature ( Figure 9D ) . Thus , although the mechanistic details remain unclear , allosteric inhibition by 2 may involve conformational changes , especially of α3 , that are similar to those that occur during the global transition from the open to the closed state ( Choy et al . , 2017 ) . This interpretation is consistent with a recent report that mutations ( Y153A , M282A ) in what we here recognize as the 197 site and L16 site alter the equilibrium between the WPD loop’s open and closed states ( Cui et al . , 2017 ) . We note that the non-competitive allosteric mechanism observed suggests that tethering 2 to K197C may shift the protein’s energy landscape in such a way as to alter the kinetics of WPD loop motions . Future work to explore this issue would nicely complement the crystallographic and functional analysis we provide here . Interestingly , several of the other noncovalent fragments bound to WT* overlay well with the aromatic rings of 2 tethered to K197C ( Figure 9B ) . This structural convergence suggests promising new avenues for future medicinal chemistry efforts . First , more conservatively , portions of specific fragments could be added to 2 to yield improved covalent allosteric inhibitors for K197C PTP1B . Second , perhaps more promisingly , portions of 2 could be combined with ( portions of ) specific fragments to create potent new non-covalent allosteric inhibitors for WT PTP1B .
Our analysis of PTP1B paints a portrait of an inherently allosteric system . Allostery is fundamentally tied to protein functions such as catalysis via the theme of conformational motions ( Goodey and Benkovic , 2008 ) . Here , we have harnessed new approaches in X-ray crystallography to map coordinated conformational redistributions that underlie allostery in the dynamic enzyme PTP1B . Metaphorically , we were able to use this map of PTP1B’s ‘intramolecular nervous system’ to reveal allosteric ‘pressure points’ that enable long-range modulation of its function . Proteins sample many conformations from a complex energy landscape ( Frauenfelder et al . , 1991 ) , many of which are accessible and represented among the millions to trillions of molecules in a protein crystal . However , an X-ray crystallographic dataset provides only ensemble-averaged information -- so it is difficult to decipher individual minor conformations from a single dataset . A key to our work was harnessing the power of en masse structural analysis , which let us reveal minor conformations and the shifts between them that allow a dynamic protein to function . We exploited families of structures in two different ways . First , we contrasted structures at several different temperatures ( Keedy et al . , 2015b ) for PTP1B to track coordinated conformational shifts which underlie allosteric communication . Second , we used hundreds of structures of PTP1B with different small-molecule fragments to calculate a statistical ‘background’ electron density map representing the unbound state , which we could subtract to reveal fragment-bound conformations ( Pearce et al . , 2017 ) for many allosteric sites . This requires using the PDB format of alternative locations to encode both compositional and conformational heterogeneity within a single model . Our multi-structure equilibrium X-ray approaches complement other methods for breaking the degeneracy of ensemble-averaged data to resolve multiple conformations of macromolecules . For example , 3D classification algorithms in cryo-electron microscopy enable in silico purification of different compositional and conformational states ( Scheres , 2016 ) . Time-resolved X-ray experiments , for example with free-electron lasers , offer great promise for mapping conformational changes with both spatial and temporal resolution , although general experimental strategies are still forthcoming for the vast majority of proteins that are non-photoactivatable ( Hekstra et al . , 2016 ) . More generally , integrative modeling algorithms can synthesize data from disparate sources at different resolutions , including solution NMR or small-angle X-ray scattering , to build ensembles of structures that are consistent with all the experimental data ( van den Bedem and Fraser , 2015; Russel et al . , 2012 ) . By exploiting a new multitemperature multiconformer X-ray approach , we have identified a collective allosteric network that is contiguous on the ‘back side’ of the protein , centered around the quasi-ordered α7 helix ( Figure 2E , Figure 9E ) . This network includes the BB site , which was previously targeted with a small-molecule allosteric inhibitor ( Wiesmann et al . , 2004 ) . It also includes adjacent sites ( the 197 site and the L16 site ) in either direction from the BB site , which have not been targeted previously with small-molecule inhibitors . Several residues in these additional sites were implicated as being part of putative allosteric sites by recent work using mutagenesis and NMR chemical shift and dynamics information ( Choy et al . , 2017; Cui et al . , 2017 ) . Our work agrees with those studies in identifying the 197 site and L16 site as potentially important players in PTP1B’s collective allosteric network . We additionally complement them by revealing , in atomic detail , alternative conformations that these sites natively populate . Our work suggests that allosteric perturbations do not necessarily induce conformational changes in PTP1B -- instead , the alternative conformations are already latently sampled by the apo protein and are simply stabilized by the allosteric perturbations . Portions of the allosteric network we observe here in PTP1B -- in particular the series of aromatic and hydrophobic residues linking the allosteric BB site to the active site ( Figure 3 ) and the cluster of aromatic residues behind and beneath the WPD loop ( Figure 5—figure supplement 4 ) -- are analogous to the dynamic hydrophobic spines that play central roles in allosteric activation of protein kinases ( Taylor and Kornev , 2011; Kim et al . , 2017 ) . Our analysis suggests that allostery in PTP1B is characterized by interdependent conformational changes spanning length scales: helical order-disorder transitions , hydrophobic shifts , local sidechain rotamer changes , subtle helical twists , and large discrete active-site loop motions . Our work reveals new opportunities for long-range control of PTP1B function by impinging upon tendrils of this expanded allosteric network with small molecules . Specifically , we have used two different small-molecule techniques with complementary strengths to reveal new footholds for developing allosteric inhibitors for PTP1B in the future . First , we used a new high-throughput method with small-molecule fragments to map the ligandability of the entire surface of PTP1B with high structural resolution . Although individually these fragments have low affinity , collectively the 110 protein:fragment structures we report reveal overlapping poses with a multitude of precise binding interactions at specific sites in PTP1B ( Figure 7A , B , C ) . Many of the 11 binding sites outside the active site are likely to be benign . Importantly , our multitemperature X-ray analysis of the apo protein provides a way to predict which binding sites are instead likely to be allosterically coupled to function . Based on the result that the same small number of sites are both ( a ) implicated as allosteric by multitemperature X-ray analysis of conformational changes in apo PTP1B and ( b ) the most ligandable sites from the fragment screen across the entire surface , we conclude that conformational changes may be important for allosteric ligand binding in this protein . Second , we used covalent tethering to probe the functional effects of a high-occupancy ligand at one promising allosteric site . Our work takes advantage of a synergy between fragments and tethering . Fragments bind weakly , but allow for the visualization of hundreds of chemical entities bound to distinct sites in a protein . By contrast , tethering does not provide high-throughput structural information , but allows for targeted perturbations to functionally probe a specific site , and additionally can provide lower-throughput structural information about chemical matter that allosterically inhibits a protein variant . Importantly , the structural and functional information from these techniques can be combined to open doors for future structure-based drug design efforts ( Figure 9B ) . Together , our results illuminate a promising new region of chemical space that may be fruitfully explored by future medicinal chemistry efforts to develop potent non-covalent allosteric inhibitors for WT PTP1B . Although our work helps set the stage for such efforts , generating potent and selective ligands based on linking of fragment-screening hits is a difficult and time-consuming process that is beyond the scope of this current study . While none of the fragments identified to bind at the 197 site in our work have the potency or selectivity for proper assaying of their effect on enzyme activity , let alone for use as a cell-active ligand , their identification supports the ligandability of the 197 site , and motivates future efforts to identify potent ligands at this site . Our use of covalent tethering at the 197 site allowed us to drive occupancy of fragment 2 at this site and functionally validates the allosteric effect of ligand binding at this site on PTP1B phosphatase activity . The percent tethering observed for 2 is correlated with the percent inhibition observed ( Figure 8B ) , indicating that the tethered compound drives inhibition . However , the maximum inhibition is only ~60% ( Figure 8B and Figure 8—figure supplement 3A ) , which is consistent with a partial noncompetitive allosteric mechanism of inhibition . This mechanism is supported by our kinetic analysis of 2-bound K197C ( Figure 8—figure supplement 2C and Figure 8—figure supplement 3B ) where the reduction in enzyme rate is driven by a reduction in Vmax , not a change in KM ( Figure 8—figure supplement 2E , F and Figure 8—figure supplement 3C , D ) . Partial inhibition is a common paradigm in noncompetitive allosteric inhibition , where ligand occupancy does not perfectly correlate with inhibition ( Ramsay and Tipton , 2017; Whiteley , 2000 ) . The ‘druggability’ of the 197 site in WT PTP1B is further supported by recent work using structure-based simulations and virtual screening , which found that the 197 site bound several diverse molecules in silico , and was the most responsive to allosteric perturbations among several potential allosteric pockets identified in PTP1B ( Kumar et al . , 2018 ) . Together with our findings , these data imply that while the development of potent non-covalent allosteric inhibitors targeting the 197 site may be arduous , a concerted effort toward this goal may ultimately be fruitful . Future efforts to develop potent non-covalent allosteric inhibitors for PTP1B can explore an additional feature of the allosteric 197 site that may enhance its druggability . As stated above , the quasi-disordered α7 helix is capable of reordering into different conformations , some of which cover the α3-α6 region including the 197 site . We observe several reordered α7 conformations under different conditions: with BB3 at 273 K , with BB1 at 100 K , in the S295F ( α7 ) mutant , in the L192A ( α3 ) mutant with an active-site inhibitor ( Figure 3—figure supplement 3C ) , and with a fragment in the BB site ( Figure 7A ) . Similarly , the disordered C-terminus , including α7 , accommodates a recently reported allosteric inhibitor ( Krishnan et al . , 2014 ) . For the BB site , the malleability of α7 is likely a double-edged sword . On one hand , reordered conformations of α7 may contribute binding energy to allosteric inhibitors by forming a ‘lid’ over what is partially a ‘cryptic site’ ( Cimermancic et al . , 2016 ) . On the other hand , they may also accommodate different small-molecule variants equally well , such that it is difficult to improve upon inhibition -- that is , the structure-activity relationship ( SAR ) is flat . However , the 197 site is structurally distinct from the BB site , and is accessible to different , more C-terminal portions of α7 . Future work will test the hypothesis that the 197 site can yield improved allosteric inhibitors that take advantage of these unique structural features . Notably , it is possible that the flat SAR observed at the BB site is due to an inhibition mechanism that is unrelated to the particular binding site , which is not uncommon in PTP1B drug discovery efforts -- if this is the case , targeting the 197 site may face similar hurdles . Our work motivates specific future efforts to allosterically inhibit PTP1B activity for therapeutic purposes . However , the allosteric network illuminated here by probing PTP1B with non-biological perturbations ( temperature and small molecules ) in vitro may also be relevant to how the enzyme is regulated in vivo , in at least three ways . First , in addition to directing subcellular localization ( Frangioni et al . , 1992 ) , the quasi-disordered C-terminal region of PTP1B reorders to interact with the ‘back side’ of the catalytic domain in different ways , is phosphorylated at disordered serine residues in vivo to regulate function ( Brautigan and Pinault , 1993 ) , and mediates allosteric inhibition by natural product molecules ( Krishnan et al . , 2014 ) . Moreover , removing α7 reduces activity in PTP1B ( Choy et al . , 2017 ) , and removing the disordered C-terminus in the related phosphatase TCPTP reduces activity even more dramatically ( Hao et al . , 1997 ) . Second , Tyr152 ( Figure 5 ) is phosphorylated in vivo , which contributes to binding to the insulin receptor kinase ( IRK ) ( Bandyopadhyay et al . , 1997; Rhee et al . , 2001 ) and is required for binding to N-cadherin ( Bandyopadhyay et al . , 1997; Rhee et al . , 2001 ) . Tyr152 phosphorylation is sterically compatible only with the ‘up’ rotamer , which is correlated with the global open state of PTP1B; it is therefore possible that Tyr152 phosphorylation and/or subsequent IRK or cadherin binding events directly affect PTP1B activity via the allosteric network we report here . Third , in several other protein tyrosine phosphatases ( PTPs ) , protein-protein interactions occur on the ‘back side’ of the catalytic domain -- coinciding with where we observe three major allosteric sites in PTP1B . For example , in RPTPγ and RPTPε , the non-catalytic D2 domain binds to the catalytic D1 domain at an interface coinciding with the allosteric network we report on the back of PTP1B ( Barr et al . , 2009 ) . Additionally , the N-terminus of PTPL1 wraps around the area coinciding with the L16 site in PTP1B , and docks in the area coinciding with the ɑ7 helix in PTP1B ( Villa et al . , 2005 ) . Together , these observations suggest that the allosteric network we establish here within the catalytic domain of PTP1B may function as a ‘receiver’ for allosteric inputs from the C-terminus in cells . If so , this strategy operates in parallel with other mechanisms such as active-site oxidation ( van Montfort et al . , 2003 ) and phosphorylation of other sites in the catalytic domain ( Ravichandran et al . , 2001 ) as part of a complex , multifaceted regulatory scheme . Finally , it is interesting to note that different protein tyrosine phosphatases ( PTPs ) share a structurally conserved catalytic domain with PTP1B -- but have different variants of the α7 helix or even entirely different N- or C-terminal domains ( Alonso et al . , 2004 ) that can be trapped in inhibitory conformations for allosteric inhibition , as recently realized for SHP2 ( Chen et al . , 2016 ) . Similarly , regulatory domains or subdomains were recently targeted for allosteric inhibition of the serine/threonine phosphatases Wip1 ( Gilmartin et al . , 2014 ) and PP1 ( Carrara et al . , 2017 ) . It will be exciting to dissect the mechanisms by which different PTPs are allosterically controlled by their specific regulatory domains -- both to unravel these proteins’ unique cellular roles , and to reveal new opportunities to correct their dysregulation in disease .
For all ‘wild-type’ PTP1B experiments here , we used what we refer to as the WT* construct: residues 1–321 , with the C32S/C92V double mutation ( Erlanson et al . , 2003 ) to prevent off-target tethering reactions , in a pET24b vector with a kanamycin resistance gene . K197C , K197A , Y152G , and Y153A were created using site-directed mutagenesis from the WT* construct . Protein was expressed and purified as previously reported ( Pedersen et al . , 2004 ) , with some minor variations . For expression , we transformed BL21 E . coli cells with plasmid , grew cells on LB + kanamycin plates overnight at 37°C , inoculated 5 mL starter cultures of LB + kanamycin with individual colonies , grew shaking overnight at 37°C , inoculated larger 1 L cultures of LB + kanamycin , grew shaking at 37°C until optical density at 600 nm was approximately 0 . 6–0 . 8 , induced with 100 mM IPTG , and grew shaking either for 4 hr at 37°C or overnight at 18°C . Cell pellets ( ‘cellets’ ) were harvested by centrifugation and stored at −80°C in 50 mL conical tubes . For purification , we first performed cation exchange with an SP FF 16/10 cation exchange column ( GE Healthcare Life Sciences ) in lysis buffer ( 100 mM MES pH 6 . 5 , 1 mM EDTA , 1 mM DTT ) with a multi-stage 0–1 M NaCl gradient ( shallow at first for elution of PTP1B , then steeper ) ; PTP1B eluted around 200 mM NaCl . We then performed size exclusion with a Superdex 75 size exclusion column ( GE Healthcare Life Sciences ) in size exclusion buffer ( 100 mM MES pH 6 . 5 , 1 mM EDTA , 1 mM DTT , 200 mM NaCl ) . PTP1B appeared highly pure in SDS-PAGE gels . WT* PTP1B was dialyzed into crystallization buffer ( 10 mM Tris pH 7 . 5 , 0 . 2 mM EDTA , 25 mM NaCl , 3 mM DTT ) with at least a 200x volume ratio overnight at 4°C . We were unable to grow apo WT* PTP1B crystals initially , so we synthesized the active-site inhibitors OBA and OTP as in ( Andersen et al . , 2002 ) ( OBA = compound 3a , OTP = compound 12h ) . We were unable to solubilize OTP as used in ( Pedersen et al . , 2004 ) . Instead , we co-crystallized PTP1B with OBA ( Andersen et al . , 2000 ) . We first solubilized OBA to 250 mM in DMSO , then created a 10:1 molar ratio of PTP1B:OBA . Crystallization drops were set in 96-well sitting- or hanging-drop format at 4°C with 10–15 mg/mL protein with 1 μL of protein solution + 1 μL of well solution ( 0 . 2–0 . 4 M magnesium acetate , 0 . 1 M HEPES pH 7 . 3–7 . 6 , 12–17% PEG 8000 ) , then trays were incubated at 4°C . Crystals several hundred μm long grew within a few days , and often continued to grow bigger for several more days . We created seed stocks from these crystals by pipetting the entire drop into 50 μL of well solution , iterating between vortexing for 30 s and sitting on ice for 30 s several times , and performing serial 10-fold dilutions in well solution . Apo crystals were grown by introducing seed stock ( 0x , 10x , or 100x diluted ) into freshly set drops , either by streaking with a cat whisker or pipetting a small amount ( e . g . 0 . 1 μL into a 2 μL drop ) . Serial seeding using new apo crystals successively improved crystal quality . We also added ethanol to the well solution based on an additive screen ( Hampton Research ) , and added glycerol to mimic the previously published apo structure protocol ( Pedersen et al . , 2004 ) , resulting in the following final WT* PTP1B crystallization well solution: 0 . 3 M magnesium acetate , 0 . 1 M HEPES pH 7 . 5 , 0 . 1% β-mercaptoethanol , 16% PEG 8000 , 2% ethanol , 10% glycerol . The resulting crystals were used for our WT* PTP1B multitemperature analysis . We also crystallized WT* PTP1B in MRC SwissCi 3-well sitting-drop trays . Protein was at 30–50 mg/mL protein in the same crystallization buffer . The well solution was very similar except for having a slightly lower precipitant concentration ( 13–14% PEG 8000 ) and no glycerol . Drops were set at room temperature with 135 nL protein solution + 135 nL well solution + 30 nL seed stock , then trays were incubated at 4°C . Crystals appeared within a few days . The best seed stocks had been diluted 10-100x . These crystals were used for the BB3-soaking and fragment-soaking experiments . We crystallized apo K197C in the microbatch format with Al’s oil covering all wells . Protein was at 5–30 mg/mL in the same crystallization buffer as WT* but without DTT . The well solution was 0 . 3 M magnesium acetate , 0 . 1 M HEPES pH 7 . 5 , 0 . 1% β-mercaptoethanol , 10–26% PEG 8000 , 2% ethanol . Drops were set on ice with 1 μL protein solution + 1 μL well solution , then trays were incubated at 4°C . Crystals appeared within a few days . We also grew apo K197C crystals in a few other similar conditions . We crystallized K197C tethered to 2 in the 96-well hanging-drop format . Protein was at 15 mg/mL in the same crystallization buffer as WT* but without DTT . The well solution was 0 . 2 M magnesium acetate tetrahydrate , 20% PEG 3350 . Drops were set at room temperature with 100 nL protein solution + 100 nL well solution , then trays were incubated at 4°C . Crystals appeared within a few days . We used PDB ID 1sug for the apo WT 100 K dataset . The apo WT* 278 K dataset was collected at Stanford Synchrotron Radiation Lightsource ( SSRL ) beamline 12–2 . All fragment-soaked datasets were collected at Diamond Light Source beamline I04-1 . All other datasets were collected at Advanced Light Source ( ALS ) beamline 8 . 3 . 1 . For apo WT* 180 , 240 , and 278 K , crystals had been grown in 2 μL drops with 10% glycerol in the mother liquor , then 1 . 5 μL of 50% glycerol was added several hours before data collection , resulting in a final concentration of ~27% glycerol . Some crystals were also dabbed into more 50% glycerol just before mounting . For BB3-complexed WT* , crystals were soaked with 125 nL of 10 mM BB3 ( in DMSO ) . No glycerol was present in these crystals . For apo WT* and BB3-complexed WT* , crystals were looped and placed in a plastic capillary with ~70% mother liquor , ~30% water to prevent dehydration during data collection , regardless of temperature; datasets were obtained at different temperatures simply by adjusting the cryojet temperature before placing the crystal on the goniometer . Helical data collection ( translation along the crystal coupled to goniometer rotations ) was used to expose fresh regions with each shot , to minimize radiation damage . For apo and 2-tethered K197C , crystals were simply looped and directly mounted on the goniometer in front of the cryojet . For apo K197C , a small amount of ice was likely present on the crystal . For fragment-soaked PTP1B , WT* PTP1B crystals in MRC SwissCi 3-well sitting-drop trays were soaked with small-molecule fragments using acoustic droplet ejection technology and a database mapping individual fragments to individual crystals as described ( Collins et al . , 2017 ) . PTP1B crystals were quite tolerant to DMSO , so we were able to achieve high fragment concentrations and long incubation times: we soaked overnight for >8 hr at final concentrations of 30% DMSO and 30–150 mM fragment ( depending on the concentration of the fragments in the source library ) . Additionally we collected X-ray data for 48 ‘apo’ datasets ( soaked with DMSO ) , 42 of which gave high-resolution datasets , to better establish the unbound background electron density for PanDDA analysis . Despite the high DMSO concentrations , we did not observe difference electron density consistent with any ordered DMSO molecules bound to PTP1B . Some fragments were soaked into additional crystals if good datasets were not obtained from the initial soak; however , only 2 of the 110 fragment-bound datasets contain the same fragment . We also collected a relatively small number of trial datasets ( 28 ) near room temperature instead of cryogenic temperature , but they were generally low-resolution , and none revealed bound fragments . Most crystals stuck to the bottoms of wells regardless of construct and tray format , but it was often possible to gently dislodge them , or to physically break them off then expose the unperturbed portion of the crystal to the X-ray beam . Each dataset in this study was collected from a single crystal . To process the multitemperature and tethered datasets , we used XDS ( Kabsch , 2010 ) . In each case we chose a resolution cutoff for which CC1/2 ( Karplus and Diederichs , 2012 ) was statistically significant at the 0 . 1% level ( above 0 . 4 ) . We created a new set of Rfree flags for the 278 K WT* apo dataset , then transferred them to the MTZ file of every other dataset with the reflection file editor in PHENIX ( Adams et al . , 2010 ) ( for PDB ID 1sug , we first deleted the existing Rfree flags ) . We solved each structure by molecular replacement with Phaser ( McCoy et al . , 2007 ) . One solution was obtained for each dataset . For WT* , we used PDB ID 1c83 with all waters and the WPD loop removed . For K197C , we used a refined WT* PTP1B model for molecular replacement . For fragment-soaked datasets , we used XDS and a custom script [80; copy archived at https://github . com/elifesciences-publications/xds_iter] to automatically determine resolution cutoffs for all datasets . The resolution cutoff was initialized at 1 . 4 Å and incremented until the following criteria were met for the highest-resolution bin: at least 1 . 0 I/σ ( I ) , at least 50% CC1/2 ( Karplus and Diederichs , 2012 ) , and at least 90% completeness . Rfree flags were created for the highest-resolution dataset by transferring and extending the flags from the 278 K WT* apo dataset using the PHENIX reflection file editor . These Rfree flags were then transferred from that highest-resolution dataset to every other dataset in the fragment-soaking experiment . For PanDDA to accept the MTZ files as inputs , it was necessary to modify each file so that all columns ( H , K , L , structure factors , map coefficients , and R-free flags ) had the same number of indices; no observations were omitted in this step . We then phased each dataset with Phaser using a reference model that was created by interpreting a high-resolution DMSO-soaked apo dataset . Next , we refined each initial model from Phaser using phenix . refine with the following flags to prevent excessive coordinate drift: ‘reference_coordinate_restraints . enabled=True ‘reference_coordinate_restraints . sigma=0 . 1’ . Structure factors were then dropped from MTZ files , leaving map coefficients as inputs to PanDDA . Filled map coefficients ( from PHENIX ) were used to avoid Fourier series truncation effects in PanDDA maps . The resulting models were used as input to PanDDA ( see below ) . For Figure 2—figure supplement 1 , we re-refined the following 36 structures of PTP1B from the PDB either as-is , or with a dual-conformation WPD loop: 1bzj 1kak 1oem 1oeo 1sug 1t48 1t49 1t4j 2azr 2b07 2bgd 2f6f 2f6t 2f6v 2f6w 2f6y 2f6z 2f70 2f71 2h4k 2hb1 2qbp 2qbq 2qbr 2qbs 2zmm 2zn7 3cwe 3d9c 3eax 3eb1 3eu0 3i7z 3i80 3sme 4i8n . For the dual-conformation refinements , we constrained occupancies of the open + closed conformations of the WPD loop to 1 . For multitemperature WT/WT* , we refined the initial model using phenix . refine for 10 macrocycles with automated water picking turned off . Next , we inserted preliminary open and closed alternative conformations for the WPD loop , and refined for another 10 macrocycles with automated water picking turned on . Finally , we performed several rounds of manual rebuilding , including manual addition and deletion of protein , solvent , and glycerol conformations and refinement with automated water picking turned off . Anisotropic B-factors were not used in refinement . The α7 helix was modeled as alternative conformation A only , with the unmodeled B conformation presumed to correspond to the disordered state; this allowed the occupancy of the ordered state to be refined . It was necessary to provide explicit occupancy parameter files to phenix . refine in some cases . For many residues , conformations obtained from PDB ID 1sug or 1t49 or another of our datasets ( usually higher temperatures ) were useful for ‘filling in’ missing density . Often the missing conformations would not have been obvious based on the map alone , but once inserted and refined they seemed to fit well . This cross-dataset conformational-sampling approach also had the effect of emphasizing differences between models from different temperatures while minimizing differences due to chance or arbitrary modeling choices . Nevertheless , we encourage future users of these datasets to compare across different temperatures based at least in part on the electron density , and not just our models . The building process was guided by all-atom structure validation with MolProbity ( Chen et al . , 2010 ) . The 100 K WT model ( 1sug ) is truly WT , whereas our new WT* datasets are all C32S/C92V , as noted above; however , both cysteine-scrubbing mutations are structurally conservative , distal to the active site , and apparently uncoupled from the WPD loop and all allosteric regions . Glycerol ( ~20% final ) was present in WT* crystals during each multitemperature data collection to maintain consistency with the 100 K WT structure ( PDB ID 1sug ) , in which glycerol was used as a cryoprotectant . Several ordered glycerol molecules , including those in contact with the closed WPD loop and at the allosteric 197 site , were evident from electron density at all temperatures . However , in some cases , it was difficult to differentiate between ordered waters , glycerols , or simply noise in the map . For example , the electron density was uncertain at some of the elevated temperatures for some glycerols originally modeled in PDB ID 1sug . When glycerol was omitted from crystals , the WPD loop was entirely in the open conformation regardless of temperature , from cryogenic temperature to near room temperature ( data not shown ) . Our interpretation is that ordered glycerols in the active site , which are evident from the electron density at all temperatures , make weak contacts with the WPD loop’s closed conformation , and thus shift the protein’s energy landscape to a regime in which the open vs . closed conformations are near enough to isoenergetic that temperature can modulate their populations . This interpretation is strengthened by the fact that these glycerol molecules align well with a bound mimic of the pTyr substrate ( PDB ID 1pty ) , which causes the loop to close during the catalytic cycle . For K197C , we used a similar refinement procedure , including many manual tweaks of alternative conformations for protein and water atoms . For the 2-tethered K197C structure , we omitted 2 for many rounds of refinement , allowing the electron density for the missing molecule to become extremely convincing before we finally added it to the model . The distance between the sulfur atoms in K197C and the ligand was restrained to 2 . 15 Å with a σ of 0 . 1 Å for refinement . For fragment-soaked datasets , we used the PanDDA approach ( Pearce et al . , 2017 ) in a few stages . First , using PanDDA version 0 . 1 , we ran pandda . analyse , and interpreted and modeled events . Next , using the new PanDDA version 0 . 2 , we ran pandda . analyse again . During this second run , datasets which were events in the first run were excluded from background density calculation , and datasets that had substantial map artifacts or very noisy/low-quality maps in the first run were excluded entirely . Some events for which we modeled bound fragments in the earlier PanDDA version 0 . 1 runs were not detected as events in the final PanDDA version 0 . 2 run . In these cases , we manually created event maps based on the 1-BDC background subtraction values from the earlier PanDDA runs . In most cases , visual inspection confirmed that these early events likely correspond to bound fragments . Many PanDDA ‘events’ in the active site corresponded not to ligand binding , but rather to protein conformational changes of the WPD loop , P-loop , and Y46 loop that are caused by oxidation of the catalytic Cys215 , a natural PTP1B regulatory mechanism ( van Montfort et al . , 2003 ) . Some other active-site events were difficult to interpret , perhaps due to active-site dynamics or differences in the appropriate background model for the open vs . closed state of the protein; future methodological improvements may clarify modeling in such cases . We built a generic unbound-state model by interpreting both an average map for the highest-resolution bin and one of the best individual apo datasets . The WPD loop was modeled as open , Tyr152 was modeled with two alternative sidechain conformations on the loop 11 backbone that is compatible with the open WPD loop , and the N-terminus ( start of α1 ) and C-terminus ( end of α6 , since α7 was disordered ) were fit as well as possible . Ordered waters were also manually positioned . This generic unbound-state model was superposed onto each PanDDA input model in the correct reference frame , then refined , to create an unbound-state model for each dataset . For each fragment-bound state , we inspected the fragment binding site , plus the several interesting regions of the protein mentioned above , in detail interactively . Waters were copied over from the unbound-state model , then moved or deleted where they conflicted with the bound fragment and/or the PanDDA event map . For a small number of planar fragments , several copies of the fragment bind in a parallel stack bridging the 197 site and a symmetry-related copy of the WPD-loop residue Phe182 via a crystal-lattice contact . Some areas such as Tyr152 were modeled with alternative conformations in the bound state only when they were well justified in the event map; otherwise we generally adhered to the unbound-state model . The correct modeling choice for the termini was uncertain in some cases . To refine structures for the 110 datasets with one or more modeled fragments , first we created restraints files for the ligands with eLBOW ( Moriarty et al . , 2009 ) . For a small number of ligands , we additionally used AceDRG ( Long et al . , 2017 ) and found that AceDRG generated more realistic restraints . Next , the pandda . export method in PanDDA version 0 . 2 was used to create an ‘ensemble structure’ containing both the unbound state ( including alternative conformations ) and the bound state ( including alternative conformations ) in one multiconformer model . In pandda . export , the parameter ‘options . prune_duplicates_rms = 0 . 2’ was used to merge alternative conformations that were highly similar , and the parameter ‘duplicates . rmsd_cutoff = 0 . 4’ was used to restrain the coordinates of somewhat similar alternative conformations to be identical . These parameter values were chosen to effectively merge residues with very similar coordinates , while still allowing residues we evaluated as having genuine alternative conformations to remain separate and unrestrained . The resulting geometry restraint files from pandda . export are necessary to minimize overfitting or coordinate drift during refinement of this model type . For refinement of the ensembles representing multiconformer models of the apo and bound states , we first refined each ensemble structure with phenix . refine to obtain water positions . The first stage of PHENIX refinement was 10 macrocycles with no removal or addition of waters ( ‘ordered_solvent = False’ ) to let the existing waters relax into local minima . The second stage of PHENIX refinement was another 10 macrocycles with automated removal and addition of waters ( ‘ordered_solvent = True’ ) to remove waters that were unable to reach local minima and add waters that were clearly missing . Adding and removing waters , when compared to only removing them , generally had negligible effect on MolProbity scores , but improved Rwork and Rfree . During this first PHENIX refinement stage to obtain water positions , occupancies were fixed to the original PanDDA BDC value for the ground state and 1-BDC for the bound state; occupancy was distributed evenly between substates when the ground state or the bound state had alternative conformations for some residues . We observed coordinate drift and unstable B-factors for the protein with PHENIX refinement . Therefore , we copied the water positions obtained from PHENIX into the initial ensemble models , and refined with Refmac ( Murshudov et al . , 2011 ) . To do so , we first set all B-factors to 40 Å2 , set bound-state occupancies to 2* ( 1-BDC ) and unbound-state occupancies to 1–2* ( 1-BDC ) ( with occupancy evenly distributed across alternative conformations within each state ) , and generated new restraints files that included the water molecules by running the PanDDA utility giant . make_restraints with the same RMSD parameter as with giant . merge_conformations: ‘duplicates . rmsd_cutoff = 0 . 4’ . Then , each ensemble model was refined with Refmac using giant . quick_refine with the ligand CIF and custom giant . make_restraints restraint parameter files using the protocol herein . First , each model was refined for the default 10 cycles , with the extra arguments to Refmac ‘MAKE HOUT Yes’ , to preserve hydrogens , and ‘HOLD 0 . 001 100 100’ to restrain XYZ coordinates but still allow for some geometry regularization and encourage B-factor and occupancy convergence . Next , the output from that refinement was fed into a loop of Refmac refinement with the default 10 cycles per run , and the ‘HOLD 0 . 0001 100 100’ argument , essentially fixing the XYZ coordinates , while letting occupancies and B-factors refine . Our output was the result of the 4th round ( 1 round + 3 rounds ) of refinement in Refmac . However , the occupancies refined with these refinements did not converge to the correct occupancy ( as seen by huge difference peaks describing the ligand ) . We then refined these structures with PHENIX , fixing the XYZ coordinates and manually scanning across possible occupancies while refining B-factors with the following settings: refinement . refine . strategy = individual_adp , hydrogens . optimize_scattering_contribution = False , main . number_of_macro_cycles = 10 , optimize_mask = True , optimize_adp_weight = True . While in principle one could interpret the difference density to pick an optimal refined occupancy , no other statistics calculated provided a clear choice of occupancy . We ultimately chose to deposit occupancies of the bound state at 2 . 2 times the event map occupancy ( 1-BDC ) . This occupancy choice was motivated by the trend previously found by Pearce et al . ( 2017 ) . In cases where the total bound occupancy was 50% or higher , the models were manually inspected , and a few dropped to low occupancies that minimized difference features of the ligand . The resulting final ensemble structures of the unbound state plus the fragment-bound state were converted from PDB to mmCIF format and deposited in the PDB using the new multimodel submission procedure . Coot ( Emsley et al . , 2010 ) was instrumental to visualizing and interactively modeling all structures . PyMOL ( Schrödinger , 2016 ) was used for all molecular graphics after initial modeling . We frequently used the volume rendering feature for low-contour electron density alongside the traditional mesh for higher-contour electron density . We screened K197C against a previously synthesized library of 1600 disulfide fragments made available by the UCSF Small Molecule Discovery Center ( SMDC ) ( Kathman et al . , 2014; Burlingame et al . , 2011b ) . For the screen , tethering reactions were performed using the following conditions: 1x tethering buffer ( 25 mM Tris pH 7 . 5 , 100 mM NaCl ) , with 500 nM of K197C , 1 mM β-mercaptoethanol , and 100 µM of fragment ( 0 . 2% DMSO ) , 1 hr at rt . Unless otherwise noted , tethering reactions for follow-up experiments and activity assays were performed using the following conditions: 1x tethering buffer , 1 µM of K197C , 0 . 1 mM β-mercaptoethanol , and 50 µM of fragment ( 2% DMSO ) , 1 hr at rt . For DiFMUP assays 100 µM of fragment ( 0 . 2% DMSO ) was used during tethering . For crystallography , tethering reactions were performed using the following conditions: 1x tethering buffer , 0 . 76 mg/mL of K197C , 0 . 1 mM β-mercaptoethanol , 500 µM of TCS401 , and 250 µM of fragment ( 2% DMSO ) , 2 hr at rt . A total reaction size of 3 . 5 mL was used for preparation of crystallography samples . Following labeling , the reaction was dialyzed into crystallization buffer overnight to remove TCS401 and unbound fragment . In all cases , the percent of tethering was measured using a Waters Xevo G2-XS Mass Spectrometer , and calculated by comparing the relative peak heights of the unmodified and modified protein . Tethering EC50 values were calculated using nonlinear fitting in Prism 7 ( Graphpad ) , n = 3 . For activity assays of WT* PTP1B vs . allosteric mutants ( Figure 5—figure supplement 3 ) , protein was diluted to 269 nM ( WT* ) or 200 nM ( mutants ) in a variant of pNPP activity assay buffer ( 50 mM HEPES pH 7 , 100 mM NaCl , 1 mM EDTA , and 1 mM DTT ) . WT* assays were performed at 269 nM protein and mutant assays were performed at 200 nM , so WT* data is normalized to 200 nM in both panels in Figure 5—figure supplement 3 . Enzyme activity assays were performed across 10 p-nitrophenyl phosphate ( pNPP ) concentrations obtained by serial two-fold dilutions starting from 20 mM . A no-enzyme well was also assayed . Absorbance at 405 nm for each reaction was monitored every 30 s for 5 min using a Tecan Infinite M200 Pro . The rate ( mAU/min ) of each reaction was calculated over the 5 min . Michaelis-Menten parameters were then calculated using Prism 7 ( Graphpad ) . kcat values were calculated using an pNPP extinction coefficient of 18 , 000 M−1 cm−1 and a path length of 0 . 29 cm . These parameters for WT* PTP1B were similar to those reported previously for WT ( Choy et al . , 2017 ) ; small discrepancies may be due in part to differences in the length of the protein construct being used . For activity inhibition assays of WT* PTP1B with small-molecule fragments , 20 fragments were chosen early in the iterative PanDDA analysis process ( see ‘Structure modeling’ ) . Protein was diluted to 200 nM in a variant of pNPP activity assay buffer ( 50 mM HEPES pH 7 , 100 mM NaCl , 1 mM EDTA , 0 . 05% Tween-20 , and 100 mM β-mercaptoethanol ) . Enzyme activity assays were performed with 0 . 15 or 1 mM fragment in 2% DMSO ( final ) or with 2% DMSO without fragment as a control , with 5 mM pNPP . A no-enzyme well was also assayed . Absorbance at 405 nm for each reaction was monitored every 30 s for 5 min . The rate ( mAU/s ) of each reaction was calculated over the 5 min . These rates were compared with fragment vs . with DMSO . For single-point assays of tethered K197C , completed tethering reactions ( post 1 hr incubation ) were diluted to a final concentration of 200 nM K197C with a variant of pNPP activity assay buffer ( 50 mM HEPES pH 7 , 100 mM NaCl , 1 mM EDTA , 0 . 05% Tween-20 , and 100 mM β-mercaptoethanol ) and pNPP ( 5 mM final ) . A no-enzyme well and a DMSO-only well were also assayed . Absorbance at 405 for each reaction was monitored every 30 s for 5 min using a Tecan Infinite M200 Pro . Percent inhibition was calculated using the following equation: 100 ( 1- ( ( RateFragment-RateNo Enzyme ) / ( RateDMSO-RateNo Enzyme ) ) ) . For 2 titration assays of tethered K197C and WT* , tethering reactions were performed at nine different concentrations of 2 obtained by serial three-fold dilutions starting at 50 µM . After 1 hr incubation , the reactions were diluted to a final concentration of 100 nM K197C and WT* with a variant of pNPP activity assay buffer ( 50 mM HEPES pH 7 , 100 mM NaCl , 1 mM EDTA , 0 . 05% Tween-20 , and 100 mM β-mercaptoethanol ) , the same concentration of 2 as used during tethering , and pNPP ( 5 mM final ) . A no-enzyme well and a DMSO-only well were also assayed . Absorbance at 405 nm for each reaction was monitored every 30 s for 5 min using a Tecan Infinite M200 Pro . The rate ( mAU/s ) of each reaction was calculated over the 5 min . Percent inhibition was calculated using the following equation: 100 ( 1- ( ( RateFragment-RateNo Enzyme ) / ( RateDMSO-RateNo Enzyme ) ) ) . Ki values were calculated using nonlinear fitting in Prism 7 ( Graphpad ) , n = 3 . For pNPP kinetics experiments with tethered complexes , completed K197C and WT* tethering reactions ( post 1 hr incubation ) were diluted to a final concentration of 200 nM K197C and WT* with a variant of pNPP activity assay buffer ( 50 mM HEPES pH 7 , 100 mM NaCl , 1 mM EDTA , 0 . 05% Tween-20 , and 100 mM β-mercaptoethanol ) , 2 ( 50 µM ) , and 12 pNPP concentrations obtained by serial two-fold dilutions starting from 20 mM . Absorbance at 405 nm for each reaction was monitored every 30 s for 5 min using a Tecan Infinite M200 Pro . The rate ( mAU/min ) of each reaction was calculated over the 5 min . Data was plotted and fit using Prism 7 ( Graphpad ) , n = 3 . Vmax and KM values were calculated using the Michaelis-Menten nonlinear fit in Prism 7 ( Graphpad ) . All p-values and significances were calculated using a FDR-Approach Multiple t-test in Prism 7 ( Graphpad ) . For DiFMUP inhibition experiments with tethered complexes , completed K197C and WT* tethering reactions ( post 1 hr incubation ) were diluted to a final concentration of 2 nM K197C and WT* with a variant of DiFMUP activity assay buffer ( 50 mM HEPES pH 7 , 100 mM NaCl , 1 mM EDTA , 0 . 05% Tween-20 ) , and 9 . 4 µM of DiFMUP . Fluorescence at 450 nm with 358 nm excitation for each reaction was monitored every 30 s for 3 min using a Tecan Infinite M200 Pro . The rate ( ΔF/min ) of each reaction was calculated over the 3 min . Percent inhibition was calculated using the following equation: 100 ( 1- ( ( RateFragment-RateNo Enzyme ) / ( RateDMSO-RateNo Enzyme ) ) ) , n = 3 . For DiFMUP kinetics experiments with tethered complexes , completed K197C tethering reactions ( post 1 hr incubation ) were diluted to a final concentration of 2 nM K197C with a variant of DiFMUP activity assay buffer ( 50 mM HEPES pH 7 , 100 mM NaCl , 1 mM EDTA , 0 . 05% Tween-20 ) , and 11 DiFMUP concentrations obtained by serial two-fold dilutions starting from 300 µM . Fluorescence at 450 nm with 358 nm excitation for each reaction was monitored every 30 s for 3 min using a Tecan Infinite M200 Pro . The rate ( ΔF/min ) of each reaction was calculated over the 3 min . Data was plotted and fit using Prism 7 ( Graphpad ) , n = 3 . Vmax and KM values were calculated using the Michaelis-Menten nonlinear fit in Prism 7 ( Graphpad ) . All p-values and significances were calculated using a FDR-Approach Multiple t-test in Prism 7 ( Graphpad ) . Multiconformer models and structure factors for the multitemperature WT and WT* ( 6B90 , 6B8E , 6B8T , 6B8X ) , BB3-bound ( 6B8Z ) , K197C apo ( 6BAI ) and tethered ( 6B95 ) datasets have been deposited in the Protein Data Bank ( Berman et al . , 2000 ) . We have made publicly available several files that document our PanDDA analysis of all WT* fragment-soaked datasets . For each dataset , we provide a model of the unbound state , structure factors , an average map for the corresponding resolution bin , a PanDDA Z-map , and one or more PanDDA event map ( s ) as applicable . For fragment-bound datasets , we also provide the refined ground state model and the bound state model ( before they were merged into an ensemble and refined ) as separate PDB files , along with PHENIX , Refmac , and ligand restraint files used in the ensemble refinement . Additionally , we provide an overall PanDDA log file . These files are hosted at Zenodo at the following DOI: 10 . 5281/zenodo . 1044103 . Finally , using a new deposition procedure , refined ensemble structures for the 110 WT* fragment-bound datasets have been deposited to the PDB ( 5QDE , 5QDF , 5QDG , 5QDH , 5QDI , 5QDJ , 5QDK , 5QDL , 5QDM , 5QDN , 5QDO , 5QDP , 5QDQ , 5QDR , 5QDS , 5QDT , 5QDU , 5QDV , 5QDW , 5QDX , 5QDY , 5QDZ , 5QE0 , 5QE1 , 5QE2 , 5QE3 , 5QE4 , 5QE5 , 5QE6 , 5QE7 , 5QE8 , 5QE9 , 5QEA , 5QEB , 5QEC , 5QED , 5QEE , 5QEF , 5QEG , 5QEH , 5QEI , 5QEJ , 5QEK , 5QEL , 5QEM , 5QEN , 5QEO , 5QEP , 5QEQ , 5QER , 5QES , 5QET , 5QEU , 5QEV , 5QEW , 5QEX , 5QEY , 5QEZ , 5QF0 , 5QF1 , 5QF2 , 5QF3 , 5QF4 , 5QF5 , 5QF6 , 5QF7 , 5QF8 , 5QF9 , 5QFA , 5QFB , 5QFC , 5QFD , 5QFE , 5QFF , 5QFG , 5QFH , 5QFI , 5QFJ , 5QFK , 5QFL , 5QFM , 5QFN , 5QFO , 5QFP , 5QFQ , 5QFR , 5QFS , 5QFT , 5QFU , 5QFV , 5QFW , 5QFX , 5QFY , 5QFZ , 5QG0 , 5QG1 , 5QG2 , 5QG3 , 5QG4 , 5QG5 , 5QG6 , 5QG7 , 5QG8 , 5QG9 , 5QGA , 5QGB , 5QGC , 5QGD , 5QGE , 5QGF ) . | Proteins perform many important jobs in each of the cells in our bodies , such as transporting other molecules and helping chemical reactions to occur . The part of the protein directly involved in these tasks is called the active site . Other areas of the protein can communicate with the active site to switch the protein on or off . This method of control is known as allostery . Switching proteins on and off could help us to develop treatments for certain diseases . For example , a protein called PTP1B reduces how well cells can respond to insulin . Switching this protein off could therefore help to treat diabetes . However , much like it’s hard to guess how a light switch is wired to a light bulb without seeing behind the walls , it is hard to predict which remote areas of a protein are ‘wired’ to the active site . Keedy , Hill et al . have now used two complementary methods to examine the structure of PTP1B and find new allosteric sites . The first method captured a series of X-ray images from crystallized molecules of the protein held at different temperatures . This revealed areas of PTP1B that can move like windshield wipers to communicate with each other . The second method soaked PTP1B crystals in trays with hundreds of drug-sized molecules and assessed which sites on the protein the molecules bound to . The molecules generally bound to just a few sites of the protein . Further tests on one of these sites showed that it can communicate with the active site to turn the protein on or off . Further work will be needed to develop drugs that could treat diabetes by binding to the newly identified allosteric sites in PTP1B . More generally , the methods developed by Keedy , Hill et al . could be used to study allostery in other important proteins . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2018 | An expanded allosteric network in PTP1B by multitemperature crystallography, fragment screening, and covalent tethering |
In most well-studied rod-shaped bacteria , peptidoglycan is primarily crosslinked by penicillin-binding proteins ( PBPs ) . However , in mycobacteria , crosslinks formed by L , D-transpeptidases ( LDTs ) are highly abundant . To elucidate the role of these unusual crosslinks , we characterized Mycobacterium smegmatis cells lacking all LDTs . We find that crosslinks generate by LDTs are required for rod shape maintenance specifically at sites of aging cell wall , a byproduct of polar elongation . Asymmetric polar growth leads to a non-uniform distribution of these two types of crosslinks in a single cell . Consequently , in the absence of LDT-mediated crosslinks , PBP-catalyzed crosslinks become more important . Because of this , Mycobacterium tuberculosis ( Mtb ) is more rapidly killed using a combination of drugs capable of PBP- and LDT- inhibition . Thus , knowledge about the spatial and genetic relationship between drug targets can be exploited to more effectively treat this pathogen .
Peptidoglycan ( PG ) is an essential component of all bacterial cells ( Vollmer et al . , 2008a ) , and the target of many antibiotics . PG consists of linear glycan strands crosslinked by short peptides to form a continuous molecular cage surrounding the plasma membrane . This structure maintains cell shape and protects the plasma membrane from rupture . Our understanding of PG is largely derived from studies on laterally growing model rod-shaped bacteria like Escherichia coli and Bacillus subtilis ( Figure 1—figure supplement 1A ) . In these organisms , new PG is constructed along the lateral side wall by the concerted effort of glycosyltransferases , which connect the glycan of a new PG subunit to the existing mesh , and transpeptidases , which link peptide side chains . An actin-like protein , MreB , positions this multi-protein complex along the short axis of the cell so that glycan strands are inserted circumferentially , creating discontinuous hoops of PG around the cell ( Domínguez-Escobar et al . , 2011; Garner et al . , 2011 ) . This orientation of PG creates a mechanical anisotropy that is responsible for rod shape ( Hussain et al . , 2018 ) . However , not all rod-shaped bacteria encode MreB . In fact , there are important differences between model bacteria and Actinobacteria like mycobacteria , a genus of rod-shaped bacteria that includes the major human pathogen Mycobacterium tuberculosis ( Mtb ) . In mycobacteria , new PG is inserted at the cell poles ( at unequal amounts based on pole age ) , rather than along the lateral walls ( Figure 1A ) . Additionally , mycobacteria are missing several factors , including MreB , that are important for cell elongation ( Kieser and Rubin , 2014 ) . Furthermore , in E . coli and B . subtilis the vast majority ( >90% ) of the peptide linkages are created by D , D-transpeptidases known as penicillin-binding proteins ( PBPs ) ( Pisabarro et al . , 1985 ) . PBPs , the targets of most β-lactams , link the fourth amino acid of one peptide side chain to the third amino acid of another , forming 4–3 crosslinks . Peptidoglycan crosslinks can also be catalyzed by L , D-transpeptidases ( LDTs ) , which link peptide side chains by the third amino acid forming 3–3 linkages ( Figure 1—figure supplement 1B ) . In mycobacteria , these 3–3 crosslinks , are highly abundant , accounting for at least 60% of linkages ( Kumar et al . , 2012; Lavollay et al . , 2008; Wietzerbin et al . , 1974 ) . Although there has been extensive characterization of LDTs in vitro ( Cordillot et al . , 2013; Dubée et al . , 2012; Lavollay et al . , 2008; Magnet et al . , 2007; Mainardi et al . , 2005; Mainardi et al . , 2007; Triboulet et al . , 2013 ) , because PG has been most well studied in bacteria where 3–3 crosslinks are rare , the cellular role of these enzymes and the linkages they create is poorly understood . As is the case with PBPs , there exists many copies of LDTs in the cell - there are five LDTs in Mtb and six in Mycobacterium smegmatis ( Msm ) , a non-pathogenic relative of Mtb ( Sanders et al . , 2014 ) , making genetic characterization challenging . Also similarly to PBPs , LDT homologues do not appear to functionally overlap completely ( Cordillot et al . , 2013; Kumar et al . , 2017; Schoonmaker et al . , 2014 ) . Tuberculosis remains an enormous global health problem , in part , because treating even drug susceptible disease is difficult . The standard regimen includes a cocktail of four drugs given over six months . Treatment of drug-resistant Mtb is substantially longer and includes combinations of up to seven drugs ( Global Tuberculosis Report , 2017 ) . While some of the most important anti-mycobacterials target cell wall synthesis , surprisingly , drugs that target PG are not part of the core treatment for either drug-susceptible or drug-resistant disease . However , carbapenems , β-lactam antibiotics that potently inhibit LDTs in vitro ( Cordillot et al . , 2013; Dubée et al . , 2012; Lavollay et al . , 2008; Mainardi et al . , 2007; Triboulet et al . , 2013 ) , are also effective against drug resistant Mtb in vitro and drug-sensitive Mtb in patients ( Diacon et al . , 2016; Hugonnet et al . , 2009 ) . But , why are LDTs important in mycobacteria ? To explore this , we constructed a strain of Msm that lacks the ability to form 3–3 crosslinks . We find that 3–3 crosslinks are formed in maturing peptidoglycan and that they are necessary to stabilize the cell wall and prevent lysis . Cells that lose the ability to synthesize 3–3 crosslinks have increased dependence on 4–3 crosslinking . Thus , simultaneous inhibition of both processes results in rapid cell death .
PG uniquely contains D-amino acids , which can be conjugated to fluorescent probes ( fluorescent D-amino acids , FDAAs ) to visualize PG synthesis in live bacterial cells ( Kuru et al . , 2012 ) . When we incubated Msmwith FDAAs for a short 2 min pulse ( < 2% of Msm’s generation time ) we observed incorporation at both poles , the sites of new PG insertion in mycobacteria ( Figure 1A , B ) ( Aldridge et al . , 2012 ) . However , we also saw a gradient of fluorescence along the sidewalls , extending from the old pole ( the previously established growth pole ) that fades to a minimum at roughly mid-cell as it reaches the new pole ( the pole formed at the last cell division ) ( Figure 1B , Figure 1—figure supplement 2 ) . To identify the enzymes responsible for this unexpected pattern of lateral cell wall FDAA incorporation , we performed a fluorescence-activated cell sorting ( FACS ) -based transposon screen ( Figure 1C ) . Briefly , we stained an Msm transposon library with FDAA and repeatedly sorted the least fluorescent 12 . 5% of the population by FACS . After each sort we regrew cells , extracted gDNA and used deep sequencing to map the location of the transposons found in the low-staining population . From this screen , we identified three LDTs ( ldtA - MSMEG_3528 , ldtB - MSMEG_4745 , ldtE - MSMEG_0233 ) ( Figure 1D ) that appeared primarily responsible for FDAA incorporation . Deleting these three LDTs significantly reduced FDAA incorporation and this defect in incorporation could be partially complemented with constitutive expression of LdtE alone ( ldtE-mRFP , Figure 1—figure supplement 3A ) . To further investigate the physiological role of LDTs , we constructed a strain lacking all 6 LDTs ( ΔldtAEBCGF , hereafter ΔLDT ) . Whole genome sequencing verified all six deletions and did not detect crossover events or chromosomal duplications ( see supplemental methods ) . FDAA incorporation and 3–3 crosslinking are both nearly abolished in ΔLDT cells and can be partially restored by complementation with a single LDT ( ldtE-mRFP; ΔLDTcomp ) ( Figure 1E , F , Figure 1—figure supplements 3B and 4 ) . Thus , as might be the case in Bdellovibrio ( Kuru et al . , 2017 ) , FDAA incorporation in Msm is primarily LDT-dependent . LDTs have previously been shown to exchange non-canonical D-amino acids onto PG tetrapeptides in Vibrio cholera ( Cava et al . , 2011 ) . As deletion of a subset of LDTs in Msm produces morphologic changes ( Sanders et al . , 2014 ) , we visualized ΔLDT cells by time-lapse microscopy . We observed that a subpopulation of cells loses rod shape progressively over time , resulting in localized spherical blebs ( Figure 2A – top row , Figure 2—figure supplement 1A , Figure 2—video 1 ) . Complemented cells are able to maintain rod shape ( Figure 2—figure supplement 1B ) . We reasoned that localized loss of rod shape may occur for two reasons: ( 1 ) spatially-specific loss of cell wall integrity and/or ( 2 ) cell wall deformation due to uncontrolled , local PG synthesis . If the first hypothesis were true , high osmolarity should protect cells against forming blebs . Indeed , switching cells from iso- to high- osmolarity prevented bleb formation over time ( Figure 2A – bottom row , Figure 2—video 2 ) . To test the second hypothesis , we stained ΔLDT or WT cells with an amine-reactive dye , and observed outgrowth of new , unstained material ( Figure 2B ) . Blebs that formed in the ΔLDT cells retained stain , indicating a lack of new cell wall synthesis in the region . WT cells maintained rod shape over time at the stained portion of the bacillus . Collectively , these results indicate that 3–3 crosslinks are required to counteract turgor pressure and maintain rod shape in Msm . This led us to hypothesize that bleb formation is a result of a local defect in cell wall rigidity . To directly measure cell wall rigidity , we used atomic force microscopy ( AFM ) on live ΔLDT and WT cells . We measured the rigidity of cells in relation to their height . Generally , WT cells are stiffer than ΔLDT cells ( Figure 2C ) . Blebs in ΔLDT cells can be identified by a sharp increase in height ( Figure 2D , pink shaded ) . Since circumferential stress of the rod measured by AFM is proportional to the radius of the cell , and inversely proportional to the thickness of the cell wall ( an immeasurable quantity by AFM ) , we used cell height , a proxy for radius , to normalize the stiffness measurement . We found that stiffness drops in the area of blebs ( Figure 2D , pink shaded ) . Why does loss of rod shape occur locally and only in a subpopulation of cells ? Mycobacterial polar growth and division results in daughter cells with phenotypic differences ( Aldridge et al . , 2012 ) . For example , the oldest cell wall is specifically inherited by the new pole daughter ( Figure 2—figure supplement 2A , Aldridge et al . , 2012 ) . We hypothesized that the loss of rod shape might occur in specific progeny generated by cell division . Indeed , the daughter which inherited the new pole from the previous round of division , and the oldest cell wall , consistently lost rod shape over time , while the old pole daughter maintained rod shape ( Figure 2E , Figure 2—figure supplement 2B ) . In addition , blebs localized to the oldest cell wall ( Figure 2B ) , as visualized by pulse-chase labeling of the cell wall . Thus , 3–3 crosslinking is likely occurring in the oldest cell wall , which is non-uniformly distributed along a single cell and in the population via asymmetric polar growth and division . Taken together , these data suggest that LDTs act locally to reinforce aging PG and to maintain rod shape in a subpopulation of Msm cells - specifically , new pole daughters ( Figure 2F ) . Our observations lead to the following model: 4–3 crosslinks made by PBPs are formed at the poles where new PG is inserted and where pentapeptide substrates reside . These newly synthesized 4–3 crosslinks can then be gradually cleaved ( by D , D-endopeptidases ) as PG ages and moves toward the middle of the cell , leaving tetrapeptide substrates for LDTs to create 3–3 crosslinks . This is consistent with the FDAA incorporation pattern , which reflects the abundance of tetrapeptide substrates available for LDT exchange . Specifically , there are more available tetrapeptides near the poles and fewer near mid-cell , the site of older PG ( Figure 1F ) . In the absence of LDTs to catalyze 3–3 crosslinks , old cell wall loses integrity and turgor pressure causes bleb formation . This model predicts that ΔLDT cells should be even more dependent on 4–3 crosslinking than wild-type cells . To test this hypothesis , we used TnSeq ( Long et al . , 2015 ) to identify genes required for growth in cells lacking LDTs ( Figure 3A ) . We found that mutants of two PBPs , pbpA ( MSMEG_0031 c ) and ponA2 ( MSMEG_6201 ) , were recovered at significantly lower frequencies in ΔLDT cells ( Figure 3B ) . Likewise , using allele swapping ( Kieser et al . , 2015b ) ( Figure 3C , Figure 3—figure supplement 1 ) , a technique that tests the ability of various alleles to support viability , we found that the transpeptidase ( TP ) activity of PonA1 , which is non-essential in WT cells ( Kieser et al . , 2015b ) , becomes essential in ΔLDT cells ( Figure 3D ) . Thus , cells that lack 3–3 crosslinks are more dependent on 4–3 crosslinking enzymes . Given our model , we hypothesized that enzymes catalyzing and processing different types of crosslinks should be differentially localized along the length of the cell . Specifically , we postulated that 4–3 generating PBPs would localize at sites of new PG , while 4–3 cleaving D , D-endopeptidases and 3–3 crosslinking LDTs would localize to sites of older PG . Polar growth segregates newer PG to the poles , and , as growth occurs , older PG migrates towards the middle of the cell . To test whether 4–3 and 3–3 crosslinking enzymes localize differently , we visualized fluorescent fusions of a PBP ( PonA1 ) , and an LDT ( LdtE ) , ( Figure 4A ) . Intriguingly , both enzymes localized in a gradient pattern along the long axis of the cell , not unlike the pattern observed for FDAA incorporation . We found that the distribution pattern of PonA1-RFP was highest at the old and new poles , where new PG is inserted ( Figure 4A , B , Figure 4—video 1 , Figure 4—figure supplement 1A ) . Compared to PonA1-RFP , the LdtE-mRFP localization is highest farther from the poles , more inward from the ends of the bacillus ( albeit in a similar gradient pattern ) , at the sites of older PG ( Figure 4A , B , Figure 4—video 2 , Figure 4—figure supplement 1B ) . Thus , enzymes responsible for 4–3 and 3–3 crosslinks show distinctive subcellular localizations with respect to the site of new PG synthesis . This is consistent with the model that these enzymes act on differentially aged PG . We next sought to localize a D , D-endopeptidase . As no D , D-endopeptidase has been clearly identified in mycobacteria , we used HHPRED ( Zimmermann et al . , 2018 ) to find candidates . By homology to the E . coli protein AmpH , an enzyme with both D , D- carboxy- and endopeptidase activity ( González-Leiza et al . , 2011 ) , we identified DacB2 ( MSMEG_2433 ) , a protein previously shown to have D , D-carboxypeptidase activity in Msm ( Bansal et al . , 2015 ) , as a candidate to also harbor D , D-endopeptidase capability . We expressed and purified DacB2 and found that it , like AmpH , had both D , D-carboxypeptidase and D , D-endopeptidase activity on peptidoglycan substrates generated in vitro ( Figure 4C , Figure 4—figure supplement 1A–C ) . We used a recently developed CRISPRi system for mycobacteria to knockdown dacB2 expression in ΔLDT cells ( Rock et al . , 2017 ) . Induction of the sgRNA and dCas9 by anhydro-tetracycline ( aTc ) led to smaller blebs ( Figure 4—figure supplement 3 ) . Furthermore , DacB2-mRFP localized closer to LDT-mRFP , farther from the poles , at sites of older PG ( Figure 4A , B , Figure 4—video 3 , Figure 4—figure supplement 1C ) . Taken together , these data are consistent with a model in which blebs are formed in ΔLDT cells due to unchecked D , D-endopeptidase activity . Given that bleb formation is not completely rescued by knockdown of dacB2 , we speculate that there are additional D , D-endopeptidases in M . smegmatis . The importance of 3–3 crosslinks in mycobacteria suggests a unique vulnerability . While Mtb can be killed by most non-carbapenem ( N-C ) β-lactams like amoxicillin , which largely target the PBPs , carbapenem β-lactams , which target both PBPs and LDTs ( Kumar et al . , 2017; Mainardi et al . , 2007; Papp-Wallace et al . , 2011 ) are also effective against Mtb ( Diacon et al . , 2016; Hugonnet et al . , 2009 ) . It has been previously proposed ( Gonzalo and Drobniewski , 2013; Gupta et al . , 2010; Kumar et al . , 2017; Mainardi et al . , 2007 ) that more rapid killing of Mtb could be achieved with drug combinations that target both PBPs and LDTs . Msm Tnseq data suggests that typically dispensable 4–3 transpeptidase activity becomes essential in cells lacking LDTs ( Figure 3 ) , supporting the notion that inhibition of both PBPs and LDTs could kill mycobacteria very successfully . Interestingly , while we could create a strain of Msm lacking all LDTs , previously published Mtb Tnseq data suggests that LDTs may be essential in the pathogen ( Kieser et al . , 2015a ) . We utilized Msm and Mtb strains expressing the luxABCDE operon from Photorhabdus luminescens ( Andreu et al . , 2012; Andreu et al . , 2010 ) , where light production can be correlated to growth ( Figure 5—figure supplement 1 ) , to test if the combination of amoxicillin ( a penam ) and meropenem ( a carbapenem ) killed Msm or Mtb more rapidly than either drug alone . We found that these drugs together kill both Msm and Mtb faster than either alone ( Figure 5A , B ) . Furthermore , this combination exhibits synergism in minimal inhibitory concentration in Mtb but , not against Msm ( where synergism is defined as Σ Fractional Inhibitory Concentration <0 . 5 ( ‘Synergism Testing: Broth Microdilution Checkerboard and Broth Macrodilution Materials and methods , ’ 2016 ) , Figure 5B , Figure 5—figure supplement 2 ) . This may reflect a difference in LDT expression or essentiality between Msm and Mtb .
The success of antibiotics that target PG , like β-lactams , has led to decades of research on this critical bacterial polymer . Recently developed fluorescent probes ( FDAAs ) have been used extensively to study PG synthesis in live cells of numerous bacterial species ( Kuru et al . , 2012; Kuru et al . , 2017; Liechti et al . , 2014 ) . Intriguingly , these probes can be incorporated through diverse pathways in different bacteria and thus , their pattern can mark distinct processes ( Kuru et al . , 2012 ) . We find that in mycobacteria , FDAA incorporation is primarily LDT-dependent . FDAA incorporation in Msm shows an unusual gradient pattern ( Botella et al . , 2017 ) , suggesting an asymmetric distribution of tetrapeptide substrate for the LDT-dependent exchange reaction . In addition to their ability to exchange D-amino acids onto tetrapeptides , LDTs also catalyze non-canonical 3–3 crosslinks . Crosslinks catalyzed by LDTs are rare in model rod-shaped bacteria like E . coli and B . subtilis but , are abundant in polar growing bacteria like mycobacteria , Agrobacterium tumefaciens and Sinorhizobium meliloti ( Brown et al . , 2012; Cameron et al . , 2015; Kumar et al . , 2012; Lavollay et al . , 2008; Pisabarro et al . , 1985 ) . Here , we find that Msm cells lacking 3–3 crosslinks cannot maintain rod shape at sites of aging cell wall . 4–3 crosslinks made by PBPs appear able to maintain rod shape near the poles , the sites of newer cell wall ( Figure 6A ) . Over time , as older cell wall moves toward the middle of the cell , it loses structural stability , and begins to bleb . The gradual manner in which rod shape is lost in cells lacking 3–3 crosslinks suggests that cell wall processing must occur to de-stabilize this portion of the rod . Consistent with this idea , we find that an enzyme that cleaves 4–3 crosslinks , the D , D-endopeptidase/D , D-carboxypeptidase DacB2 , also localizes to sites of old cell wall and knockdown of this enzyme leads to smaller blebs . Why would Msm cells create 4–3 crosslinks to eventually cleave them ? There are many possibilities . For example , perhaps in the absence of lateral cell wall synthesis , the creation of substrate for LDTs through the destruction of 4–3 crosslinks allows the cell to engage the PG along the lateral cell body . This could be important for altering the thickness of the PG layer or anchoring it to the membrane at sites of otherwise ‘inert’ cell wall . Additionally , it may be that as PG ages , it is being manicured or marked for septal synthesis . Supporting this idea , we find that the gradient localization patterns of fluorescently tagged PonA1 , LdtE and DacB2 , ( as well as FDAAs ) all have local minima at mid-cell closer to the new pole- a location that correlates with the asymmetric site of division in mycobacteria ( Aldridge et al . , 2012; Santi et al . , 2013; Eskandarian et al . , 2017 ) . The lack of localization of PG synthesis enzymes and FDAAs suggests a lack of penta- and tetra- peptide substrates . This implies that this region of the cell may be more abundantly crosslinked , as crosslinking utilizes these peptide species . Could 3–3 crosslinking be a signal for septal placement ? Mycobacteria are missing known molecular septal placement mechanisms like the Noc and the Min system ( Hett and Rubin , 2008 ) . The major septal PG hydrolase is RipA , a D , L-endopeptidase which cleaves the bond between the second and third amino acid of PG side chains , a substrate available on 3–3 crosslinked material ( Böth et al . , 2011; Vollmer et al . , 2008b ) . While LdtE-mRFP does not itself strongly localize to the site of division , the crosslinks it synthesizes could migrate toward the mid-cell through polar elongation . Transmitting information from the tip to mid-cell through polar growth was recently described in mycobacteria: atomic force microscopy revealed cell-envelope deformations formed at the pole of Msm travel to mid-cell through polar growth , marking the future site of division ( Eskandarian et al . , 2017 ) . Thus , it is intriguing to speculate that 3–3 crosslinks found at aging cell wall could be important for localizing cell division machinery . In well-studied rod-shaped bacteria like E . coli and B . subtilis , shape is maintained by MreB-directed PG synthesis along the lateral cell body ( Garner et al . , 2011; Hussain et al . , 2018; Ursell et al . , 2014 ) . On the other hand , mycobacteria maintain shape in the absence of an obvious MreB homolog , and in the absence of lateral cell wall elongation . Furthermore , in contrast to lateral-elongating bacteria , in which new and old cell wall are constantly intermingled during growth , polar growth segregates new and old cell wall ( Figure 6B ) . We find that mycobacteria appear to utilize 3–3 crosslinks at asymmetrically distributed aging cell wall to provide stability along the lateral body , something that may not be required in the presence of MreB-directed PG synthesis . New drug combinations for TB are desperately needed . There has been a renewed interest in repurposing FDA-approved drugs for TB treatment ( Diacon et al . , 2016 ) . Some of that interest has focused on β-lactams , the oldest class of antibiotics which are the therapeutic bedrock for most other infections . We find that the protein targets of two different classes of β-lactams – enzymes which do very similar chemistry – PG crosslinking – are distributed differentially in a single cell and across the population . In the absence of 3–3 crosslinks , 4–3 crosslinks become more important for cell viability . These data predict that a drug combination which inhibits both PBPs and LDTs will work synergistically to more quickly kill Mtb , a prediction we verified in vitro . Interestingly , meropenem combined with amoxicillin/clavulanate resulted in early clearance of Mtb from patient sputum ( Diacon et al . , 2016 ) . In fact , the combination might be key to accelerated killing of Mtb ( Gonzalo and Drobniewski , 2013 ) .
Unless otherwise stated , M . smegmatis ( mc2155 ) was grown shaking at 37°C in liquid 7H9 media consisting of Middlebrook 7H9 salts with 0 . 2% glycerol , 0 . 85 g/L NaCl , ADC ( 5 g/L albumin , 2 g/L dextrose , 0 . 003 g/L catalase ) , and 0 . 05% Tween 80 and plated on LB agar . M . tuberculosis ( H37Rv ) was grown in liquid 7H9 with OADC ( oleic acid , albumin , dextrose , catalase ) with 0 . 2% glycerol and 0 . 05% Tween 80 . Antibiotic selection for M . smegmatis and M . tuberculosis were done at the following concentrations in broth and on agar: 25 μg/mL kanamycin , 50 μg/mL hygromycin , 20 μg/mL zeocin and 20 μg/mL nourseothricin and , twice those concentrations for cloning in E . coli ( TOP10 , XL1-Blue and DH5α ) . Whole genome sequencing was performed on wild-type mc2155 as well as the ΔLDT mutant . Sequencing was done on an Illumina HiSeq 4000 ( RRID:SCR_016386 ) with 150 bp paired-end reads . There was a mean depth of coverage of 148x . All 6 LDT genes were verified as deleted . Furthermore , there was no evidence of any duplications or cross-over events based on a coverage plot . The sequencing has been uploaded to NCBI’s SRA ( details for sample identifiers are provided below ) . STUDY: PRJNA451029 ( SRP141343 ) ΔLDT SAMPLE: deltaLdtAEBCGF ( SRS3442031 ) ΔLDT EXPERIMENT: deltaLdtAEBCGF ( SRX4275943 ) ΔLDT RUN: deltaLdtAEBCGF_R2 . fastq ( SRR7403831 ) WT SAMPLE: Msmeg-KB ( SRS3442032 ) WT EXPERIMENT: Msmeg-KB ( SRX4275944 ) WT RUN: Msmeg-KB_R2 . fastq ( SRR7403830 ) Mtb or Msm Lux was grown to log phase and diluted to an OD600 = 0 . 006 in each well of non-treated 96-well plates ( Genesee Scientific ) containing 100 μL of meropenem ( Sigma Aldrich ) and/or amoxicillin ( Sigma Aldrich ) diluted in 7H9 + OADC + 5 μg/mL clavulanate ( Sigma Aldrich ) . Msm media contained ADC rather than OADC . Cells were incubated in drug at 37°C shaking for 7 days ( Mtb ) or 1 day ( Msm ) , 0 . 002% resazurin ( Sigma Aldrich ) was added to each well , and the plates were incubated for 24 hr before MICs were determined . Pink wells signify metabolic activity and blue signify no metabolic activity . ( Kieser et al . , 2015a ) Checkerboard MIC plates and fractional inhibitory concentrations were calculated as described in ( Synergism Testing: Broth Microdilution Checkerboard and Broth Macrodilution Methods , 2016 ) . Mtb Lux was grown to log phase ( kanamycin 25 μg/mL ) and diluted in 30 mL inkwells ( Corning Lifesciences ) to an OD600 = 0 . 05 in 7H9 + OADC + 5 μg/mL clavulanate with varying concentrations of amoxicillin , meropenem , or both . 100 μL of these cultures were pipetted in duplicate into a white 96-well polystyrene plate ( Greiner Bio-One ) and luminescence was measured in a Synergy H1 microplate reader from BioTek Instruments , Inc . using the Gen5 Software ( 2 . 02 . 11 Installation version ) . The correlation between relative light units ( RLU ) and CFU is shown in Msm in Figure 5—figure supplement 1 . Msm Lux was grown to log phase and diluted into white 96-well polystyrene plates to an OD600 = 0 . 05 . Plates were sealed with 4titude Moisture Barrier Seals and shaken continuously at 37°C . Luminescence measurements ( RLU ) were taken at 15-min intervals integrated over 1000 ms in a TECAN Spark 10M plate reader for 18 hr . NADA ( 3-[7-nitrobenzofurazan]-carboxamide-D-Alanine ) , HADA ( 3-[7-hydroxycoumarin]-carboxamide-D-Alanine ) and TADA ( 3-[5-carboxytetramethylrhodamine]-carboxamide-D-Alanine ) were synthesized by Tocris following the published protocol ( Kuru et al . , 2015 ) . To 1 mL of exponentially growing cells 0 . 1 mM of FDAA final was added and incubated for 2 min before washing in 7H9 twice . For still imaging , after the second wash , cells were fixed in 7H9 + 1% paraformaldehyde before imaging . For pulse chase experiments , cells were stained , washed with 7H9 and allowed to grow out for 40 min before being stained with a second dye and imaged . An M . smegmatis transposon library was grown to mid-log phase , and stained with 2 µg/mL NADA for 2 min . Cells were centrifuged and half of the supernatant was discarded . The pellet was resuspended in the remaining supernatant , passed through a 10 µm filter and taken to be sorted ( FACSAria; Excitation: 488 nm; Emission filter: 530/30; RRID:SCR_009839 ) . Two bins were drawn at the lowest and highest staining end of the population , representing 12 . 5% of the population . 600 , 000 cells from these bins were sorted into 7H9 medium . Half of this was directly plated onto LB agar supplemented with kanamycin to select for cells harboring the transposon . The remaining 300 , 000 cells were grown out in 7H9 to log phase , stained with FDAA and sorted again to enrich the populations . Genomic DNA ( gDNA ) was harvested from the sorted transposon library colonies and transposon-gDNA junction libraries were constructed and sequenced using the Illumina Hi-Seq platform ( Long et al . , 2015 ) . Reads were mapped on the M . smegmatis genome , tallied and reads at each TA site for the bins ( low/high incorporating sort 1 and 2 ) were imported into MATLAB and processed by a custom scripts as described in Rego et al . ( 2017 ) . Source code for this analysis can be found on GitHub at: https://github . com/hesperrego/baranowski_2018 ( copy archived at https://github . com/elifesciences-publications/baranowski_2018 ) . Sequencing data are available in NCBI’s SRA with accession number SRP141343 . Both still imaging and time-lapse microscopy were performed on an inverted Nikon TI-E microscope at 60x magnification . Time-lapse was done using a CellASIC ONIX2 Microfluidic System ( Millipore Sigma , B04A plate ) with constant liquid 7H9 flow in a 37°C chamber . For turgor experiment ( Figure 2A ) , cells were grown in either 7H9 or 7H9 500 mM sorbitol overnight , and then switched to either 7H9 with 150 mM sorbitol ( high osmolar ) or to 7H9 alone ( iso-osmolar ) . AFM experimentation was conducted as previously ( Eskandarian et al . , 2017 ) . In short , polydimethylsiloxane ( PDMS ) – coated coverslips were prepared by spin-coating a mixture of PDMS at a ratio of 15:1 ( elastomer:curing agent ) with hexane ( Sigma 296090 ) at a ratio of 1:10 ( PDMS:hexane ) ( Koschwanez et al . , 2009; Thangawng et al . , 2007 ) . A 50 µl filtered ( 0 . 5 µm pore size PVDF filter – Millipore ) aliquot of bacteria grown to mid-exponential phase and concentrated from 2 to 5 ml of culture was deposited onto the hydrophobic surface of a PDMS-coated coverslip and incubated for ~20 min to increase surface interactions between bacteria and the coverslip . 7H9 medium ( ~3 ml ) was supplied to the sample so as to immerse the bacterial sample and the AFM cantilever in fluid . The AFM imaging mode , Peak Force QNM , was used to image bacteria with a Nanoscope five controller ( Veeco Metrology ) at a scan rate of 0 . 5 Hz and a maximum Z-range of 12 µm . A ScanAsyst fluid cantilever ( Bruker ) was used . Height , peak force error , DMT modulus , and log DMT modulus were recorded for all scanned images in the trace and retrace directions . Images were processed using Gwyddion ( Department of Nanometrology , Czech Metrology Institute ) . ImageJ was used for extracting bacterial cell profiles in a tabular form . Correlated optical fluorescence and AFM images were acquired as described ( Eskandarian et al . , 2017 ) . Briefly , optical fluorescence images were acquired with an electron-multiplying charge-coupled device ( EMCCD ) iXon Ultra 897 camera ( Andor ) mounted on an IX81 inverted optical microscope ( Olympus ) equipped with an UPLFLN100XO2PH x100 oil immersion objective ( Olympus ) . Transmitted light illumination was provided by a 12V/100W AHS-LAMP halogen lamp . An U-MGFPHQ fluorescence filter cube for GFP with HQ-Ion-coated filters was used to detect GFP fluorescence . The AFM was mounted on top of the inverted microscope , and images were acquired with a Dimension Icon scan head ( Bruker ) using ScanAsyst fluid cantilevers ( Bruker ) with a nominal spring constant of 0 . 7 N m−1 in Peak Force QNM mode at a force setpoint ~1 nN and typical scan rates of 0 . 5 Hz . Indentation on the cell surface was estimated to be ~10 nm in the Z-axis . Optical fluorescence microscopy was used to identify Wag31-GFP puncta expressed in a wild-type background ( Santi et al . , 2013 ) in order to distinguish them from cells of the ∆LDT mutant strains . A cell profile was extracted from AFM Height and DMT Modulus image channels as sequentially connected linear segments following the midline of an individual cell . A background correction was conducted to by dividing the DMT modulus values of the cell surface by the mean value of the PDMS surface and rescaled to compare the cell surface rigidity between individual cells from different experiments . The DMT modulus reflects the elastic modulus ( stress-strain relationship ) for each cross-sectional increment along the cell length . Using a segmented line , profiles of cells from new to old pole were created at the frame ‘pre-division’ based on physical cell separation of the phase image . A custom FIJI ( Schindelin et al . , 2012 ) script was run to extract fluorescence line profiles of each cell and save them as . csv files . These . csv files were imported to Matlab where a custom script was applied to normalize the fluorescence line profile to fractional cell length and to interpolate the fluorescence values to allow for averaging . Source code for this analysis can be found on GitHub at:https://github . com/hesperrego/baranowski_2018 Cells were stained with AlexaFluor 488 NHS ester ( ThermoFisher Scientific ) as described previously ( Aldridge et al . , 2012 ) and followed via time-lapse microscopy in the CellASIC device . Briefly , 1 mL of log phase cells was pelleted at 8000 rpm for 1 min and washed with 1 mL PBST . The pellet was resuspended in 100 uL of PBST and 10 uL Alexa Fluor 488 carboxylic acid succinimidyl ester was added for a final concentration of 0 . 05 mg/mL . This was incubated for 3 min at room temperature . Stained cells were pelleted for 1 min at 13 , 000 rpm and washed with 500 μL PBST . They were spun again and resuspended in 7H9 for outgrowth observation over time in the CellASIC device . Images were analyzed using a combination of Oufti ( Paintdakhi et al . , 2016 ) ( RRID:SCR_016244 ) for cell selection followed by custom coded Matlab scripts to plot FDAA fluorescence over normalized cell length , calculate cell length and bin cells by existence of an FDAA labeled septum . This code and a manual for its use has been included as a source code file with this manuscript ( Source Code-Instructions and code for FDAA image analysis in Figure 1 ) . M . smegmatis cells were transduced at ( OD6001 . 1–1 . 7 ) with φMycoMarT7 phage ( temperature sensitive ) that has a Kanamycin marked Mariner transposon as previously described ( Long et al . , 2015 ) . Briefly , mutagenized cells were plated at 37°C on LB plates supplemented with Kanamycin to select for phage transduced cells . Roughly 100 , 000 colonies per library were scraped , and genomic DNA was extracted . Sequencing libraries were generated specifically containing transposon disrupted DNA . Libraries were sequenced on the Illumina platform . Data were analyzed using the TRANSIT pipeline ( DeJesus et al . , 2015 ) ( RRID:SCR_016492 ) . Sequencing data are available in NCBI’s SRA with accession number SRP141343 . 600 mL of wild-type and ΔLDT cells were grown to log phase and collected via centrifugation at 5000 x g for 10 min at 4°C . The resulting pellet was resuspended in PBS and cells were lysed using a cell disruptor at 35 , 000 psi twice . Lysed cells were boiled in 10% SDS ( sodium dodecyl sulfate ) for 30 min and peptidoglycan was collected via centrifugation at 17 , 000 x g . Pellets were washed with 0 . 01% DDM ( n-Dodecyl β-D-maltoside ) to remove SDS and resuspended in 1XPBS + 0 . 01% DDM . PG was digested with alpha amylase ( Sigma A-6380 ) and alpha chymotrypsin ( Amersco 0164 ) overnight . The samples were again boiled in 10% SDS and washed in 0 . 01% DDM . The resulting pellet was resuspended in 400 μL 25 mM sodium phosphate pH6 , 0 . 5 mM MgCl2 , 0 . 01% DDM . 20 μL of lysozyme ( 10 mg/mL ) and 20 μL 5 U/μL mutanolysin ( Sigma M9901 ) were added and incubated overnight at 37°C . Samples were heated at 100°C and centrifuged at 100 , 000 x g . 128 μL of ammonium hydroxide was added and incubated for 5 hr at 37°C . This reaction was neutralized with 122 μL of glacial acetic acid . Samples were lyophilized , resuspended in 300 μL 0 . 1% formic acid and subjected to analysis by LC-MS/MS . Peptide fragments were separated with an Agilent Technologies 1200 series HPLC on a Nucleosil C18 column ( 5 μm 100A 4 . 6 × 250 mm ) at 0 . 5 mL/min flow rate with the following method: Buffer A = 0 . 1% Formic Acid; Buffer B = 0 . 1% Formic Acid in acetonitrile; 0% B from 0 to 10 min , 0–20% B from 10 to 100 min , 20% B from 100 to 120 min , 20–80% B from 120 to 130 min , 80% B from 130 to 140 min , 80–0% B from 140 to 150 min , 0% B from 150 to 170 min . MS/MS was conducted in positive ion mode using electrospray ionization on an Agilent Q-TOF ( 6520 ) . MSMEG_2433 was expressed and purified using a modified method for purification of low-molecular-weight PBPs that was previously published ( Qiao et al . , 2014 ) . An N-terminally truncated MSMEG_2433 ( 29-296 ) was cloned into the pET28b vector for isopropyl β-D-1-thiogalactopyranoside ( IPTG ) inducible expression in E . coli BL21 ( DE3 ) ( see strain construction notes above ) . 10mLs of overnight culture grown in LB with Kanamycin ( 50 μg/mL ) were diluted 1:100 into 1 L of LB with Kanamycin ( 50 μg/mL ) and grown at 37°C until an OD600 of 0 . 5 . The culture was cooled to room temperature , induced with 0 . 5 mM IPTG , and shaken at 16°C overnight . Cells were pelleted via centrifugation at 4000 rpm for 20 min at 4°C . The pellet was suspended in 20 mL binding buffer ( 20 mM Tris pH 8 , 10 mM MgCl2 , 160 mM NaCl , 20 mM imidazole ) with 1 mM phenylmethylsulfonylfluoride ( PMSF ) and 500 μg/mL DNase . Cells were lysed via three passage through a cell disrupter at ≥10 , 000 psi . Lysate was pelleted by ultracentrifugation ( 90 , 000 × g , 30 min , 4°C ) . To the supernatant , 1 . 0 mL washed Ni-NTA resin ( Qiagen ) was added and the mixture rocked at 4°C for 40 min . After loading onto a gravity column , the resin was washed twice with 10 mL wash buffer ( 20 mM Tris pH 8 , 500 mM NaCl , 20 mM imidazole , 0 . 1% Triton X-100 ) . The protein was eluted in 10 mL of elution buffer ( 20 mM Tris pH8 , 150 mM NaCl , 300 mM imidazole , 0 . 1% reduced Triton X-100 ) and was concentrated to 1 mL with a 10kD MWCO Amicon Ultra Centrifuge Filter . The final protein concentration was measured by reading absorbance at 280 nm and using the estimated extinction coefficient ( 29459 M−1cm−1 ) calculate concentration . The protein was diluted to 200 μM in elution buffer with 10% glycerol , aliquoted , and stored at −80°C . Proper folding of purified MSMEG_2433 ( 29-296 ) was tested via Bocillin-FL binding . Briefly , 20 μM of purified protein was added to penicillin G ( 100 , 1000 U/mL in 20 mM K2HPO4 , 140 mM NaCl , pH7 . 5 ) in a 9 μL reaction . The reaction was incubated at 37°C for 1 hr . 10 μM Bocillin-FL was added and incubated at 37°C for 30 min . SDS loading dye was added the quench the reaction and samples were loaded onto a 4–20% gel . MSMEG_2433 ( 29-296 ) bound by Bocillin-FL was imaged using a Typhoon 9400 Variable Mode Imager ( GE Healthcare ) ( Alexa Excitation-488nm Emission-526nm ) . B . subtilis Lipid II was extracted as previously published ( Qiao et al . , 2017 ) . S . aureus SgtB was purified as previously published ( Rebets et al . , 2014 ) . Purification of B . subtilis PBP1 was carried out as previously described ( Lebar et al . , 2014 ) . 20 μM purified BS Lipid II was incubated in reaction buffer ( 50 mM HEPES pH 7 . 5 , 10 mM CaCl2 ) with either 5 μM PBP1 or 0 . 33 μM SgtB for 1 hr at room temperature . The enzymes were heat denatured at 95°C for 5 min . Purified MSMEG_2433 ( 29-296 ) was added ( 20 uM , final ) and the reaction was incubated at room temperature for 1 hr . Mutanolysin ( 1 μL of a 4000 U/mL stock ) was added and incubated for 1 . 5 hr at 37°C ( twice ) . The resulting muropeptides were reduced with 30 μL of NaBH4 ( 10 mg/mL ) for 20 min at room temperature with tube flicking every 5 min to mix . The pH was adjusted to ~4 using with 20% H3PO4 and the resulting product was lyophilized to dryness . The residue was resuspended in 18 μL of water and analyzed via LC-MS as previously reported ( Welsh et al . , 2017 ) . Biological replicates – independent cultures; Technical replicates – the same culture in replicate . Microscopy for Figure 1E , F and Figure 1—figure supplement 2 was done once and analyzed . The data shown in Figure 1—figure supplement 3 was done once in technical triplicate . The graph shows one replicate . Time-lapse experiment in Figure 2A was done twice ( biological duplicate on separate days ) . Included for this figure are videos of full fields of view of the time-lapse experiments ( Figure 2—video 1- full field; Figure 2—video 2- full field ) . Microscopy for Figure 2B was done in biological triplicate on three separate days . The time-lapse phenotype highlighted Figure 2—figure supplement 1 was observed in biological triplicate on 3 independent days . AFM data in Figure 2D and E was derived from two independent experiments done on separate days . Allele swapping experiment in Figure 3C was done once . Time-lapse microscopy for Figure 4A was performed in biological duplicate . The graph in Figure 4B represents data from one experiment . Figure 4C is representative data from two technical replicates ( the same protein and substrate preparations were used ) . Microscopy and quantification of bleb size in dacB2 CRISPRi knock-down ( Figure 4—figure supplement 3 ) was done twice ( biological duplicate on separate days ) . Luciferase Msm data in Figure 5A was performed once . Luciferase Mtb survival data in Figure 5B was done in biological triplicate and technical triplicate . Biological triplicates are plotted . Minimum inhibitory concentrations ( MIC ) were determined in biological duplicate ( two separate cultures on two separate days ) and technical duplicate for Figure 5B . Combination MIC for Figure 5—figure supplement 2 was determined once for Mtb and twice for Msm strains . Fluorescent D-amino acid pulse-chase for Figure 6B was done on two independent days ( biological duplicate ) . | Most bacteria have a cell wall that protects them and maintains their shape . Many of these organisms make their cell walls from fibers of proteins and sugars , called peptidoglycan . As bacteria grow , peptidoglycan is constantly broken down and reassembled , and in many species , new units of peptidoglycan are added into the sidewall . However , in a group of bacteria called mycobacteria , which cause tuberculosis and other diseases , the units are added at the tips . The peptidoglycan layer is often a successful target for antibiotic treatments . But , drugs that treat tuberculosis do not attack this layer , partly because we know very little about the cell walls of mycobacteria . Here , Baranowski et al . used genetic manipulation and microscopy to study how mycobacteria build their cell wall . The results showed that these bacteria link peptidoglycan units together in an unusual way . In most bacteria , peptidoglycan units are connected by chemical links known as 4-3 crosslinks . This is initially the same in mycobacteria , but as the cell grows and the cell wall expands , these bonds break and so-called 3-3 crosslinks form . In genetically modified bacteria that could not form these 3-3 bonds , the cell wall became brittle and weak , and the bacteria eventually died . These findings could be important for developing new drugs that treat infections caused by mycobacteria . Baranowski et al . demonstrate that a combination of drugs blocking both 4-3 and 3-3 crosslinks is particularly effective at killing the bacterium that causes tuberculosis . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2018 | Maturing Mycobacterium smegmatis peptidoglycan requires non-canonical crosslinks to maintain shape |
Understanding genome to phenotype linkages has been greatly enabled by genomic sequencing . However , most genome analysis is typically confined to the nuclear genome . We conducted a metabolomic QTL analysis on a reciprocal RIL population structured to examine how variation in the organelle genomes affects phenotypic variation . This showed that the cytoplasmic variation had effects similar to , if not larger than , the largest individual nuclear locus . Inclusion of cytoplasmic variation into the genetic model greatly increased the explained phenotypic variation . Cytoplasmic genetic variation was a central hub in the epistatic network controlling the plant metabolome . This epistatic influence manifested such that the cytoplasmic background could alter or hide pairwise epistasis between nuclear loci . Thus , cytoplasmic genetic variation plays a central role in controlling natural variation in metabolomic networks . This suggests that cytoplasmic genomes must be included in any future analysis of natural variation .
A central goal of modern biology is to understand how the integration of gene functions across a genome lead to the individual’s specific phenotype . A key facet to this effort is to develop models that would allow directly inferring a species phenotypic variation from its genetic variation . This goal of mathematically linking genetic to phenotypic variation is central to all studies of quantitative genetics ranging from human genetics to plant breeding to ecology and has led to massive genome resequencing projects focused on developing the genomic databases to allow these studies ( Liti et al . , 2009; Altshuler et al . , 2010; Cao et al . , 2011 ) . However , in most quantitative genomics studies , genome analysis is largely confined to the nuclear genome with much less attention paid to the organellar genome . This is in contrast to the central role that the organellar genome plays in controlling organismal metabolism and function and the developing body of literature suggesting that organellar genomic variation can modulate the effects of nuclear genomic variation . Genomic variation in human organelles has been linked to several severe diseases ( Wallace et al . , 1988; Taylor and Turnbull , 2005; Schon et al . , 2012 ) . However , these organellar variants are typically rare with large phenotypic consequences such that they can be followed using simple maternal inheritance studies without consideration for quantitative variation in the nuclear genome ( Schon et al . , 2012 ) . More recently , quantitative studies on human diseases suggest that genetic variation in organellar genomes modify the quantitative effect of nuclear loci and disease phenotypes ( Battersby and Shoubridge , 2001; McRae et al . , 2008; Schon and Przedborski , 2011 ) . Additionally , the use of structured populations in mice , yeast , and birds have shown that cytoplasmic genome variation can influence high order phenotypes including fitness , cognition , and biomass ( Roubertoux et al . , 2003; Zeyl et al . , 2005; Park et al . , 2006; Dimitrov et al . , 2009 ) . Within ecological studies , mitochondrial variation has also been shown to alter fitness and create hybrid isolation in a variety of invertebrate species ( Willett and Burton , 2004; Wade and Goodnight , 2006; Dowling et al . , 2007 , 2010; Wolf , 2009; Willett , 2012 ) . However , these studies didn’t directly interrogate the interaction of the nuclear and cytoplasmic genomes for quantitative traits or test the breadth of phenotypes affected . In plants , genetic variation in mitochondria is also linked with large qualitative phenotypes , such as cytoplasmic male sterility ( Hanson , 1991; Schnable and Wise , 1998 ) . Plant breeding has a long history of using diallele crosses to test for the presence of maternal effects ( cytoplasmic genetic variation ) on a phenotype and this was recently extended to a small reciprocal F2 family structure to show that the cytoplasmic effects could have significant impacts on plant height in maize ( Tang et al . , 2013 ) . In rice , agronomic traits have also been shown to be influenced by interactions between cytoplasmic and nuclear genomes but the specific loci were not identified ( Tao et al . , 2004 ) . However , these reciprocal F2 populations typically have generated the impression that cytoplasmic effects on phenotypic variation are quite small in plants possibly because of an inability to account for interactions between the nuclear and cytoplasmic genomes ( Singh , 1965; Crane and Nyquist , 1967; Eenink and Garretsen , 1980; Miura et al . , 1997; Primomo et al . , 2002 ) . In contrast to the previous estimates of small effects , genomic sequencing within Arabidopsis has shown the presence of considerable genetic polymorphism in both the plastidic and mitochondrial genomes suggesting the potential for broad phenotypic consequences ( Moison et al . , 2010 ) . The above studies have shown that cytoplasmic genome variation can influence phenotypic variation . Additionally , genes underlying qualitative interactions between the nucleus and cytoplasm leading to cytoplasmic male sterility and interspecific isolation have been identified . There are , however , numerous open questions remaining about how cytoplasmic variation influences quantitative phenotypic variation . What is the level of phenotypic variation influenced by genetic variation in the cytoplasm in comparison to individual nuclear loci ? How much epistatic interaction is there between genetic variation in nuclear loci and the cytoplasmic genomes ? What is the breadth of phenotypes that might be influenced by cytoplasmic variation ? To begin answering these questions , we utilized metabolomics to investigate how genetic variation in the cytoplasmic and nuclear genomes interacts to control metabolome variation in the reciprocal Arabidopsis Kas × Tsu recombinant inbred line ( RIL ) population ( McKay et al . , 2008; Juenger et al . , 2010 ) . We focused on metabolomics because it is cost effective for large sample numbers and because plant organelles , mitochondria and plastids , are central to the function of plant primary metabolism , as well as many specialized metabolites in both energy generation and biosynthesis ( Fiehn et al . , 2000; Roessner et al . , 2001; Fiehn , 2002 ) . Within Arabidopsis , there are nearly 13 , 000 genes in the nuclear genome predicted to be involved in metabolic processes of which ∼3000 may be targeted to the mitochondria and the plastid ( Arabidopsis genome initiative , 2000 ) . This included complete biosynthetic pathways for numerous amino acids and key energy production processes like respiration and photosynthesis . Aiding these processes are 88 total predicted genes in the plastidic genome and 121 in the mitochondrial genome; most of which function to facilitate the metabolic processes that occur in organelles using both nuclear and cytoplasmic-encoded proteins . Thus , variation in the organellar genomes could directly influence the function of any of nuclear gene functioning within the organelle ( Etterson et al . , 2007; Tang et al . , 2007; Wolf , 2009 ) . This suggests that metabolism is an ideal phenotype to use for testing how cytoplasmic variation can influence phenotypic variation . We measured the metabolome within the Kas × Tsu RIL population to test how cytoplasmic variation can affect quantitative variation within the metabolome . This population was generated from a reciprocal cross and approximately half of the resulting lines carry Kas organelles while the other half carry Tsu organelles allowing for explicit analysis of the influence of cytoplasmic genetic variation ( McKay et al . , 2008; Juenger et al . , 2010 ) . Analysis of the Kas × Tsu RIL population metabolome showed that genetic variation in the organelles influenced the accumulation of over 80% of the detectable metabolites . In contrast to previous observations suggesting that the cytoplasm has only small effects , phenotypic changes associated with cytoplasmic variation were as large and often larger than that found for individual nuclear loci ( Singh , 1965; Crane and Nyquist , 1967; Eenink and Garretsen , 1980; Miura et al . , 1997; Primomo et al . , 2002 ) . In addition , the cytoplasm was found to be a central hub in the epistatic network controlling natural variation in plant metabolism . This centrality led to the cytoplasmic background displaying the unexpected ability to hide or alter epistatic interactions between nuclear loci . Thus , genetic variation in the cytoplasmic organelles has widespread and large quantitative effects on natural phenotypic variation and can influence the link between nuclear loci .
To partition phenotypic variance between the effects of the nuclear and cytoplasmic genomes upon the A . thaliana metabolome , we identified and measured metabolite levels , using non-targeted GC-TOF-MS , in leaf tissues of 316 lines of the Kas × Tsu A . thaliana RIL population harvested with fourfold replication across two separate experiments . Within the 316 lines for this population , 136 have the Kas cytoplasm and 180 have the Tsu cytoplasm . A total of 2435 metabolites were identified in over 25% of the RILs with 215 of these being known compounds and 2220 being unidentified compounds . Of these 2435 metabolites , 559 were identified in both experiments ( 161 known and 398 unknown ) while the rest were specific to one of the two experiments . Classical breeding studies utilize reciprocal crosses with linear modeling to partition heritability between the nuclear and cytoplasmic genomes ( Singh , 1965; Crane and Nyquist , 1967; Eenink and Garretsen , 1980; Miura et al . , 1997; Primomo et al . , 2002 ) . In line with the standard heritability calculations for classical breeding experiments with reciprocal crosses , we used a linear modeling approach to approximate the level of heritability in this population that can be statistically ascribed to the nuclear and cytoplasmic genome . The line heritability model simply partitions the RILs into two subpopulations based on the maternal parent and tests the level of reproducibility that is controlled by differences between the lines and between the subpopulations . Using the line heritability model , we focused on the 559 metabolites found in both experiments to estimate the broad sense heritability of the metabolome . This analysis showed that 361 metabolites had a significant ( p<0 . 01 ) line effect , of which 334 metabolites showed significant heritability based on the maternal parent subpopulation . This maternal effect was quite small with an average of 1 . 6 ± 0 . 1% ( range 15–0% ) ( Figure 1 and Figure 1—source data 1 ) . In comparison , 77 metabolites had significant nuclear broad sense heritability with an average of 21 . 3 ± 0 . 3% ( range 58–2% ) ( Figure 1 and Figure 1—source data 1 ) . This is a similar level of variation to that found in a previous analysis of metabolomic variation in Arabidopsis ( Rowe et al . , 2008; Chan et al . , 2010 ) . The combined variance of the genetic components , both nuclear and cytoplasmic , explained approximately 19 times as much of the variance as the combination of experiment and block ( Figure 1—source data 1 ) ( Rowe et al . , 2008; Chan et al . , 2010 ) . This raises an apparent disparity between the nuclear genome having high heritability that is significant for a low fraction of metabolites ( Figure 1—source data 1 ) . In contrast , the cytoplasmic effects have relatively low average heritability but significantly impact most metabolites ( Figure 1—source data 1 ) . The most likely explanation for this disparity is the large difference in the degrees of freedom for the two factors , nuclear vs cytoplasmic used for the significance test . When conducting the ANOVA test , there are only two maternal subpopulations , Kas or Tsu , whereas there are 316 different nuclear genomes/RILs , and as such , the F value for the same level of variance will be very different in the two tests ( Rowe and Kliebenstein , 2008 ) . Thus , variation in most metabolites identifies a significant effect of cytoplasmic genetic variation , even though this effect appears to be smaller than that for the nuclear genome in the standard line heritability estimation approach . 10 . 7554/eLife . 00776 . 003Figure 1 . Line estimation of heritability in nuclear and organellar genomes . We compared the estimated metabolite heritability’s due to nuclear ( solid line ) and organelle ( dashed line ) variation across the lines of the Kas × Tsu ( black ) RIL populations . Shown are frequency plots of heritability and for each class , the bin size is 5% for the frequency plots . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 00310 . 7554/eLife . 00776 . 004Figure 1—source data 1 . Heritability . Results of the ANOVA model ( Model I ) for metabolic variance in the experiment are shown by metabolite . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 00410 . 7554/eLife . 00776 . 005Figure 1—source data 2 . Means . All metabolic mean accumulation is shown by line and subpopulation . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 005 We next moved beyond the classical line heritability approach and directly test if cytoplasmic variation may have similar effects to individual nuclear loci . We mapped QTLs in the Kas × Tsu RILs controlling the mean accumulation of all 2435 measured metabolites ( 215 known and 2220 unidentified compounds ) using 1069 markers with a median spacing of 0 . 35 cM ( Figure 1—source data 2 ) ( Jansen , 1994; Zeng et al . , 1999; Broman et al . , 2003 ) . This identified 2974 QTLs effecting the accumulation of 1822 metabolites with an average of 1 . 22 ± 0 . 02 QTLs per metabolite ( Figure 2—source data 1 ) . These QTLs predominantly partitioned into 14 metabolomic QTL hotspots that were unequally distributed across the genome ( Figure 2 ) . Chromosome IV had five detectable hotspots while chromosomes I and II had only a single significant hotspot ( Figure 2 ) . The hotspots had an average LOD interval size of approximately 5 cM that on average contains a similar number of genes as found within the combined mitochondrial and plastidic genomes . 10 . 7554/eLife . 00776 . 006Figure 2 . Genetic architecture of metabolite QTLs across the Kas × Tsu genome . ( A ) The number of metabolites for which a QTL was detected within a 5 cM sliding window is plotted against the genetic location of the metabolite QTLs in cM . The permuted threshold ( p=0 . 05 ) for detection of a significant metabolite hotspot is 54 QTLs . The graph is scaled to match part ( B ) . Hotspots are labeled above the respective locus with the chromosome and cM . ( B ) Heat map showing the location and effect of significant QTLs detected for average metabolite accumulation across the five chromosomes . Red indicates a positive effect of the Kas allele , while green indicates a positive effect of the Tsu allele . Vertical white lines separate the chromosomes ( I to V from left to right ) . Clustering on the left is based on the absolute Pearson correlation of QTL effects across all significant loci for each metabolite . Only metabolites with two or more QTLs were plotted . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 00610 . 7554/eLife . 00776 . 007Figure 2—source data 1 . QTL Lists . The location and effect of all metabolic QTLs identified from the CIM model are presented . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 007 To compare nuclear and cytoplasmic genetic variation effects upon metabolism , we utilized these metabolomic QTL hotspots to develop an additive model . This model uses the genetic marker at the center of each metabolomic QTL hotspot as separate terms and also incorporates the cytoplasmic variation as an additional term equivalent to each nuclear locus . We used this additive model to directly test all QTL hotspot-metabolite linkages and obtain the mean effect of variation at each hotspot on the metabolome ( Figure 3—source data 1 , Figure 3—Figure supplements 1–12 ) . For all hotspots , the primary metabolites altered by variation at each hotspot were distributed across the primary metabolism pathways with no obvious strong qualitative network specific loci as previously found ( Figure 3 ) ( Rowe et al . , 2008 ) . 10 . 7554/eLife . 00776 . 008Figure 3 . Metabolomic consequence of variation at nuclear loci . A map of central metabolism was created in cytoscape and this was used to plot the estimated allele effect of genetic variation at nuclear loci . A red box shows increased metabolite accumulation when the line contains the Kas allele at the nuclear locus while green shows increased metabolite accumulation when the line contains the Tsu allele at the nuclear locus . White boxes are metabolites that were detected but not significantly influenced by the cytoplasmic genome and gray boxes are metabolites that were not detected . The two loci shown are those that had the most metabolites affected within the metabolic map . All other nuclear loci are plotted in Figure 3—figure supplements 1–12 . ( A ) Estimated allele effects of the M . V . 59 hotspot . ( B ) Estimated allele effects of the M . I . 83 hotspot . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 00810 . 7554/eLife . 00776 . 009Figure 3—source data 1 . Single marker ANOVA . p values for main effect model of metabolic QTL variance using hotspot markersDOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 00910 . 7554/eLife . 00776 . 010Figure 3—figure supplement 1 . Metabolomic consequence of variation at nuclear locus M . II . 91 . A map of central metabolism was created in cytoscape and this was used to plot the estimated allele effect of genetic variation at each nuclear hotspot locus . A red box shows increased metabolite accumulation when the line contains the Kas allele at the nuclear locus while green shows increased metabolite accumulation when the line contains the Tsu allele at the nuclear locus . White boxes are metabolites that were detected but not significantly influenced by the locus and gray boxes are metabolites that were not detected . The specific hotspot metabolomic QTL locus is listed at the top of the page using the nomenclature shown in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 01010 . 7554/eLife . 00776 . 011Figure 3—figure supplement 2 . Metabolomic consequence of variation at nuclear locus M . III . 51 . A map of central metabolism was created in cytoscape and this was used to plot the estimated allele effect of genetic variation at each nuclear hotspot locus . A red box shows increased metabolite accumulation when the line contains the Kas allele at the nuclear locus while green shows increased metabolite accumulation when the line contains the Tsu allele at the nuclear locus . White boxes are metabolites that were detected but not significantly influenced by the locus and gray boxes are metabolites that were not detected . The specific hotspot metabolomic QTL locus is listed at the top of the page using the nomenclature shown in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 01110 . 7554/eLife . 00776 . 012Figure 3—figure supplement 3 . Metabolomic consequence of variation at nuclear locus M . III . 64 . A map of central metabolism was created in cytoscape and this was used to plot the estimated allele effect of genetic variation at each nuclear hotspot locus . A red box shows increased metabolite accumulation when the line contains the Kas allele at the nuclear locus while green shows increased metabolite accumulation when the line contains the Tsu allele at the nuclear locus . White boxes are metabolites that were detected but not significantly influenced by the locus and gray boxes are metabolites that were not detected . The specific hotspot metabolomic QTL locus is listed at the top of the page using the nomenclature shown in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 01210 . 7554/eLife . 00776 . 013Figure 3—figure supplement 4 . Metabolomic consequence of variation at nuclear locus M . IV . 3 . A map of central metabolism was created in cytoscape and this was used to plot the estimated allele effect of genetic variation at each nuclear hotspot locus . A red box shows increased metabolite accumulation when the line contains the Kas allele at the nuclear locus while green shows increased metabolite accumulation when the line contains the Tsu allele at the nuclear locus . White boxes are metabolites that were detected but not significantly influenced by the locus and gray boxes are metabolites that were not detected . The specific hotspot metabolomic QTL locus is listed at the top of the page using the nomenclature shown in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 01310 . 7554/eLife . 00776 . 014Figure 3—figure supplement 5 . Metabolomic consequence of variation at nuclear locus M . IV . 17 . A map of central metabolism was created in cytoscape and this was used to plot the estimated allele effect of genetic variation at each nuclear hotspot locus . A red box shows increased metabolite accumulation when the line contains the Kas allele at the nuclear locus while green shows increased metabolite accumulation when the line contains the Tsu allele at the nuclear locus . White boxes are metabolites that were detected but not significantly influenced by the locus and gray boxes are metabolites that were not detected . The specific hotspot metabolomic QTL locus is listed at the top of the page using the nomenclature shown in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 01410 . 7554/eLife . 00776 . 015Figure 3—figure supplement 6 . Metabolomic consequence of variation at nuclear locus M . IV . 23 . A map of central metabolism was created in cytoscape and this was used to plot the estimated allele effect of genetic variation at each nuclear hotspot locus . A red box shows increased metabolite accumulation when the line contains the Kas allele at the nuclear locus while green shows increased metabolite accumulation when the line contains the Tsu allele at the nuclear locus . White boxes are metabolites that were detected but not significantly influenced by the locus and gray boxes are metabolites that were not detected . The specific hotspot metabolomic QTL locus is listed at the top of the page using the nomenclature shown in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 01510 . 7554/eLife . 00776 . 016Figure 3—figure supplement 7 . Metabolomic consequence of variation at nuclear locus M . IV . 51 . A map of central metabolism was created in cytoscape and this was used to plot the estimated allele effect of genetic variation at each nuclear hotspot locus . A red box shows increased metabolite accumulation when the line contains the Kas allele at the nuclear locus while green shows increased metabolite accumulation when the line contains the Tsu allele at the nuclear locus . White boxes are metabolites that were detected but not significantly influenced by the locus and gray boxes are metabolites that were not detected . The specific hotspot metabolomic QTL locus is listed at the top of the page using the nomenclature shown in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 01610 . 7554/eLife . 00776 . 017Figure 3—figure supplement 8 . Metabolomic consequence of variation at nuclear locus M . IV . 72 . A map of central metabolism was created in cytoscape and this was used to plot the estimated allele effect of genetic variation at each nuclear hotspot locus . A red box shows increased metabolite accumulation when the line contains the Kas allele at the nuclear locus while green shows increased metabolite accumulation when the line contains the Tsu allele at the nuclear locus . White boxes are metabolites that were detected but not significantly influenced by the locus and gray boxes are metabolites that were not detected . The specific hotspot metabolomic QTL locus is listed at the top of the page using the nomenclature shown in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 01710 . 7554/eLife . 00776 . 018Figure 3—figure supplement 9 . Metabolomic consequence of variation at nuclear locus M . IV . 82 . A map of central metabolism was created in cytoscape and this was used to plot the estimated allele effect of genetic variation at each nuclear hotspot locus . A red box shows increased metabolite accumulation when the line contains the Kas allele at the nuclear locus while green shows increased metabolite accumulation when the line contains the Tsu allele at the nuclear locus . White boxes are metabolites that were detected but not significantly influenced by the locus and gray boxes are metabolites that were not detected . The specific hotspot metabolomic QTL locus is listed at the top of the page using the nomenclature shown in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 01810 . 7554/eLife . 00776 . 019Figure 3—figure supplement 10 . Metabolomic consequence of variation at nuclear locus M . V . 45 . A map of central metabolism was created in cytoscape and this was used to plot the estimated allele effect of genetic variation at each nuclear hotspot locus . A red box shows increased metabolite accumulation when the line contains the Kas allele at the nuclear locus while green shows increased metabolite accumulation when the line contains the Tsu allele at the nuclear locus . White boxes are metabolites that were detected but not significantly influenced by the locus and gray boxes are metabolites that were not detected . The specific hotspot metabolomic QTL locus is listed at the top of the page using the nomenclature shown in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 01910 . 7554/eLife . 00776 . 020Figure 3—figure supplement 11 . Metabolomic consequence of variation at nuclear locus M . V . 82 . A map of central metabolism was created in cytoscape and this was used to plot the estimated allele effect of genetic variation at each nuclear hotspot locus . A red box shows increased metabolite accumulation when the line contains the Kas allele at the nuclear locus while green shows increased metabolite accumulation when the line contains the Tsu allele at the nuclear locus . White boxes are metabolites that were detected but not significantly influenced by the locus and gray boxes are metabolites that were not detected . The specific hotspot metabolomic QTL locus is listed at the top of the page using the nomenclature shown in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 02010 . 7554/eLife . 00776 . 021Figure 3—figure supplement 12 . Metabolomic consequence of variation at nuclear locus M . V . 94 . A map of central metabolism was created in cytoscape and this was used to plot the estimated allele effect of genetic variation at each nuclear hotspot locus . A red box shows increased metabolite accumulation when the line contains the Kas allele at the nuclear locus while green shows increased metabolite accumulation when the line contains the Tsu allele at the nuclear locus . White boxes are metabolites that were detected but not significantly influenced by the locus and gray boxes are metabolites that were not detected . The specific hotspot metabolomic QTL locus is listed at the top of the page using the nomenclature shown in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 021 We next proceeded to use the additive model to directly compare the role of cytoplasmic genome variation to that of individual nuclear loci in controlling metabolome variation ( Figure 3—source data 1 ) . In comparison to the broader question of cytoplasmic heritability tested in the line heritability model ( Figure 1—source data 1 ) , this model allows us to ask the more specific question of how the cytoplasmic variation compares to individual nuclear loci in altering metabolite accumulation . The additive model showed that variation in the cytoplasmic genome significantly altered variation in 1755 out of the 2435 detected metabolites . This is in comparison the nuclear loci that only affected 298 metabolites on average ( range 438–158 ) . In addition to affecting more metabolites than nuclear loci , the average effect size of the cytoplasmic genome on metabolite variation was higher than that for the nuclear loci ( Figure 4 ) . 10 . 7554/eLife . 00776 . 022Figure 4 . Comparison of estimated QTL allele effects between nuclear and cytoplasmic genetic variation . The distribution of percent allele effects are shown for all metabolite/loci combinations with positive being the Tsu allele increases the metabolite concentration in comparison to the Kas allele . Solid black line shows the allele effects for all nuclear genomic loci while the dashed line shows cytoplasmic genetic variation allele effects . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 02210 . 7554/eLife . 00776 . 023Figure 4—figure supplement 1 . Main effect estimations . Using pairwise epistasis model run for separately for all metabolites , we identified all significant main effect markers and , epistatic interaction terms . For each metabolite , we identified each significant term and estimated the genetic variance controlled by this term by taking the ratio of the Type III sums of squares ( SS ) to the total sums of squares ( total model SS + Residual SS ) for that term . In this figure , we present the distribution of the per marker and per interaction heritability for each marker/interaction based on metabolites that are significant for that marker/interaction . The two plots titled ‘main’ and ‘main trim’ that show the heritability solely attributable to the individual main effect nuclear and cytoplasmic loci . ‘main’ shows the full plot while ‘main_trim’ shows the distribution trimmed to improve the resolution on the nuclear markers . The plot entitled ‘cytonuclear’ shows the distribution of the heritability attributed to all significant interactions between the cytoplasmic genome and the listed nuclear locus . For instance , the plot of M . I . 83 in this graph shows the distribution of heritability across all metabolites significant for the cytoplasm × M . I . 83 cytonuclear interaction . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 02310 . 7554/eLife . 00776 . 024Figure 4—figure supplement 2 . Distribution of heritability of pairwise interaction of nuclear loci M . II . 91 and M . I . 83 . Using the pairwise epistasis model analysis separately for all metabolites , we identified all significant main effect markers , individual genetic loci , and interactions , epistatic terms . For each metabolite , we identified each significant term and estimated the genetic variance controlled by this term by taking the ratio of the Type III sums of squares ( SS ) to the total sums of squares ( total model SS + Residual SS ) for that term from Model III . In this figure , we present the distribution of the per marker and per interaction heritability for each marker/interaction for the metabolites that are significant for that marker/interaction . These plots are titled with specific nuclear locus show the distribution of heritability controlled by interactions between the titled marker and all the other loci in the model . Again this is only for interaction terms that were significant in the individual metabolite models . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 02410 . 7554/eLife . 00776 . 025Figure 4—figure supplement 3 . Distribution of heritability of pairwise interaction of nuclear loci M . III . 64 and M . III . 51 . Using the pairwise epistasis model analysis separately for all metabolites , we identified all significant main effect markers , individual genetic loci , and interactions , epistatic terms . For each metabolite , we identified each significant term and estimated the genetic variance controlled by this term by taking the ratio of the Type III sums of squares ( SS ) to the total sums of squares ( total model SS + Residual SS ) for that term from Model III . In this figure , we present the distribution of the per marker and per interaction heritability for each marker/interaction for the metabolites that are significant for that marker/interaction . These plots are titled with specific nuclear locus show the distribution of heritability controlled by interactions between the titled marker and all the other loci in the model . Again this is only for interaction terms that were significant in the individual metabolite models . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 02510 . 7554/eLife . 00776 . 026Figure 4—figure supplement 4 . Distribution of heritability of pairwise interaction of nuclear loci M . IV . 23 and M . IV . 17 . Using the pairwise epistasis model analysis separately for all metabolites , we identified all significant main effect markers , individual genetic loci , and interactions , epistatic terms . For each metabolite , we identified each significant term and estimated the genetic variance controlled by this term by taking the ratio of the Type III sums of squares ( SS ) to the total sums of squares ( total model SS + Residual SS ) for that term from Model III . In this figure , we present the distribution of the per marker and per interaction heritability for each marker/interaction for the metabolites that are significant for that marker/interaction . These plots are titled with specific nuclear locus show the distribution of heritability controlled by interactions between the titled marker and all the other loci in the model . Again this is only for interaction terms that were significant in the individual metabolite models . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 02610 . 7554/eLife . 00776 . 027Figure 4—figure supplement 5 . Distribution of heritability of pairwise interaction of nuclear loci M . IV . 51 and M . IV . 23 . Using the pairwise epistasis model analysis separately for all metabolites , we identified all significant main effect markers , individual genetic loci , and interactions , epistatic terms . For each metabolite , we identified each significant term and estimated the genetic variance controlled by this term by taking the ratio of the Type III sums of squares ( SS ) to the total sums of squares ( total model SS + Residual SS ) for that term from Model III . In this figure , we present the distribution of the per marker and per interaction heritability for each marker/interaction for the metabolites that are significant for that marker/interaction . These plots are titled with specific nuclear locus show the distribution of heritability controlled by interactions between the titled marker and all the other loci in the model . Again this is only for interaction terms that were significant in the individual metabolite models . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 02710 . 7554/eLife . 00776 . 028Figure 4—figure supplement 6 . Distribution of heritability of pairwise interaction of nuclear loci M . IV . 82 and M . IV . 72 . Using the pairwise epistasis model analysis separately for all metabolites , we identified all significant main effect markers , individual genetic loci , and interactions , epistatic terms . For each metabolite , we identified each significant term and estimated the genetic variance controlled by this term by taking the ratio of the Type III sums of squares ( SS ) to the total sums of squares ( total model SS + Residual SS ) for that term from Model III . In this figure , we present the distribution of the per marker and per interaction heritability for each marker/interaction for the metabolites that are significant for that marker/interaction . These plots are titled with specific nuclear locus show the distribution of heritability controlled by interactions between the titled marker and all the other loci in the model . Again this is only for interaction terms that were significant in the individual metabolite models . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 02810 . 7554/eLife . 00776 . 029Figure 4—figure supplement 7 . Distribution of heritability of pairwise interaction of nuclear loci M . V . 59 and M . V . 45 . Using the pairwise epistasis model analysis separately for all metabolites , we identified all significant main effect markers , individual genetic loci , and interactions , epistatic terms . For each metabolite , we identified each significant term and estimated the genetic variance controlled by this term by taking the ratio of the Type III sums of squares ( SS ) to the total sums of squares ( total model SS + Residual SS ) for that term from Model III . In this figure , we present the distribution of the per marker and per interaction heritability for each marker/interaction for the metabolites that are significant for that marker/interaction . These plots are titled with specific nuclear locus show the distribution of heritability controlled by interactions between the titled marker and all the other loci in the model . Again this is only for interaction terms that were significant in the individual metabolite models . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 02910 . 7554/eLife . 00776 . 030Figure 4—figure supplement 8 . Distribution of heritability of pairwise interaction of nuclear loci M . V . 94 and M . V . 82 . Using the pairwise epistasis model analysis separately for all metabolites , we identified all significant main effect markers , individual genetic loci , and interactions , epistatic terms . For each metabolite , we identified each significant term and estimated the genetic variance controlled by this term by taking the ratio of the Type III sums of squares ( SS ) to the total sums of squares ( total model SS + Residual SS ) for that term from Model III . In this figure , we present the distribution of the per marker and per interaction heritability for each marker/interaction for the metabolites that are significant for that marker/interaction . These plots are titled with specific nuclear locus show the distribution of heritability controlled by interactions between the titled marker and all the other loci in the model . Again this is only for interaction terms that were significant in the individual metabolite models . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 030 Using the additive model , we generated per locus estimates of heritability that showed the cytoplasmic genome explained twice the variance as any individual nuclear locus on average ( Figure 4—figure supplements 1–8 ) . Interestingly , the number of genes within a 5 cM region surrounding the nuclear hotspots is comparable to the number of genes present within the combined organellar genomes , which suggests that this is not simply a matter of genetic potential . Additionally , the Kas allele of both the nuclear and cytoplasmic loci were more frequently associated with increasing metabolite concentration leading to a change in the entire metabolic network within this population based on the cytoplasmic genome ( Figures 4 and 5 ) . Thus , the line heritability approach to partitioning genetic variation dramatically underestimated the effect of cytoplasmic genetic variation and cytoplasmic variation has larger effects than any individual nuclear locus even though there are a similar potential number of causative genes . 10 . 7554/eLife . 00776 . 031Figure 5 . Metabolomic consequences of cytoplasmic genomic variation . A map of central metabolism was created in cytoscape and this was used to plot the estimated allele effect of genetic variation in the cytoplasmic genomes using the reciprocal sub-populations . Colors are as given in Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 03110 . 7554/eLife . 00776 . 032Figure 5—source data 1 . Genetic polymorphisms between the Kas and Tsu Organelles . AGI represents the code for each gene with position being from the +1 of transcription for that gene . The specific base at each site in Col-0 , Kas and Tsu are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 032 To identify potential causal polymorphisms between the Kas and Tsu organelles , we obtained short-read sequencing data and identified SNPs in the organellar genomes . This analysis showed that there were polymorphisms spread out across a large number of genes including regulatory genes , such as rpoC2 in the plastid , several unknown genes , and key energy genes , such as Rubisco large subunit . The most striking changes were the large number of polymorphisms within the NADH deyhdrogenases in both the mitochondria and the plastid ( Figure 5—source data 1 ) . Specifically in the mitochondria , 32 of the 96 polymorphisms were within the NADH dehydrogenase 7 and a further 13 polymorphisms in four other NADH dehydrogenase complex genes . This is vastly more polymorphisms than would be expected by random chance suggesting a change in NADH and related metabolism between the Kas and Tsu organelles ( hyper-geometric test , p<0 . 001 ) . This agrees with the widespread metabolic consequences of the cytoplasmic genetic variation on processes that require NADH like glutamine synthesis , lipid metabolism , and any other process that utilizes NADH requiring cytochromes P450 . However , within this population the mitochondrial and plastidic genomes are perfectly co-inherited making it impossible to resolve the individual effects of the mitochondrial and plastidic genomes ( Joseph et al . , 2013 ) . Additionally , it isn’t possible to ascribe metabolites to specific subcellular compartments using a whole tissue extract as most metabolites , even those synthesized in specific compartments , often accumulate in most compartments within a plant cell ( Krueger et al . , 2011 ) . Thus , genetic variation in the cytoplasm can have significant effects on modulating global plant metabolism requiring further work to identify the specific mechanistic causes . Previous work on natural variation in the plant metabolome had shown extensive two-way and three-way epistasis between nuclear loci but could not test for the presence of epistatic interactions between nuclear and cytoplasmic loci ( Rowe et al . , 2008 ) . To test for cytonuclear epistatic interactions , we used a pairwise epistasis model that directly tested all pairwise combinations of the 14 nuclear QTL and cytoplasm against all metabolites while including all terms from the additive model ( Figure 6—source data 1 , 2 ) . This pairwise epistasis model only uses 120 of the 314 available degrees of freedom ( 38% ) leaving the majority of the degrees of freedom for the residual variance , which suggests that we are not over fitting the model . We did not extend this analysis to a full genome survey of all possible loci because these surveys do not account for existing main effect loci . After multiple testing adjustments , we only plotted pairwise interactions that significantly affected at least 10% of the metabolites to present a conservative image of the interaction network ( Figure 6 ) . 10 . 7554/eLife . 00776 . 033Figure 6 . Epistatic networks of metabolism . All epistatic interactions between the cytoplasmic genomic variation and nuclear loci ( as labeled in Figure 2 ) are plotted as lines connecting the main effect loci as nodes . The size of the main effect nodes represents the fraction of total metabolites affected by the given locus . The color of the lines show the fraction of metabolites linked with this interaction , light blue 10–15% of metabolites , dark blue 15–20% of metabolites . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 03310 . 7554/eLife . 00776 . 034Figure 6—source data 1 . Pairwise ANOVA p values . p values for main effect and interaction terms using hotspot markers . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 03410 . 7554/eLife . 00776 . 035Figure 6—source data 2 . Pairwise ANOVA sums of squares . The type III sums of squares from the model run in Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 035 Importantly , the cytoplasmic locus was one of the two nodes with the highest number of significant pairwise epistatic interactions connecting to 6 of the 14 nuclear loci . This level of interaction is higher than the average of the network degree distribution ( 3 . 5 ± 0 . 46 SE ) ( Figure 6 ) . The M . I . 83 nuclear locus had the highest level of interactions involving 7 of the other 13 nuclear loci but did not interact directly with the cytoplasmic variation ( Figure 6 ) . Thus , there is strong pairwise epistasis between the cytoplasmic and nuclear genetic loci . Further , the percent of genetic variation controlled by interactions of two nuclear loci was of similar scale to that involving a cytonuclear interaction ( Figure 4—figure supplements 1–8 ) . The large number of significant epistatic interactions between nuclear and cytoplasmic genetic variation forms a cohesive network but the network approach does not let us visualize or directly quantify the effects of each locus in an epistatic pair . Thus , we wanted to develop a better approach to visualize and compare epistatic effects ( Carlborg and Haley , 2004; Carlborg et al . , 2006; Alvarez-Castro and Carlborg , 2007 ) . We utilized the center of mass concept from physics to generate a description of epistatic effect ( Gartenhaus and Schwartz , 1957 ) . In center of mass calculations across two dimensions , each individual objects mass is described in its x , y coordinates . For describing epistasis , we converted the x , y-cartography such that the alleles for one QTL with Tsu being -1 and Kas being 1 are along the x-axis . Similarly , along the y-axis are the alleles at the second QTL thus creating a calculable genotypic cartography ( Figure 7A ) . This effectively replaces the physical distance metric in the center of mass calculation with a genetic distance metric such that they represent a scaled allelic effect . We then replace the individual objects mass the in center of mass calculations with the unscaled average phenotypic value for each of the homozygous genotypic combinations of the two QTLs allowing a genetic centroid of the phenotype to be estimated for each metabolite and plotted ( Figure 7A ) . This allows a single plot to compare the magnitude of epistasis and the allelic direction across multiple metabolites ( Please see the materials and methods for a more detailed description of this approach and Figure 7—figure supplement 1 for a representative model plot ) . 10 . 7554/eLife . 00776 . 036Figure 7 . Different epistatic patterns across genotypic combinations . The center of mass calculations were used to estimate the phenotypic center for each metabolite that was significantly affected by the given combination of loci . The hexbin plots show the distribution of phenotypic centers for all significant metabolites . The number of metabolites per hexbin are shown in the legend to the right of each graph . All other significant epistatic pair combinations are plotted in Figure 7—figure supplement 2 . ( A ) Metabolites that are additive for both the cytoplasmic loci and M . IV . 23 . ( B ) Metabolites that are epistatically affected by an interaction between M . IV . 23 and the cytoplasmic genotypes . ( C ) Metabolites that are epistatically affected by an interaction between M . I . 83 and the cytoplasmic genotypes . ( D ) Metabolites that are epistatically affected by an interaction between M . I . 83 and M . IV . 23 genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 03610 . 7554/eLife . 00776 . 037Figure 7—figure supplement 1 . Descriptive model of the epistatic center plot . This is a representative plot showing where the specific modeled metabolites , as presented in the table , are position on the center of phenotype plot . Letters label each modeled metabolite on the plot . The allele at each locus is positioned as the Kas allele at 1 and Tsu allele at −1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 03710 . 7554/eLife . 00776 . 038Figure 7—figure supplement 2 . Hexbin plots show the distribution of phenotypic centers for cytoplasmic and nuclear pairwise epistatic interactions . Only interactions significantly affecting 10% or more of the metabolites are shown . Each interaction is listed at the top of the page and the number of metabolite per hexbin is shown in the legend to the right of each graph . Center of mass calculations were used to estimate the phenotypic center for each metabolite that was significantly affected by the given epistasis interaction . The hexbin plots show the distribution of phenotypic centers for all significant metabolites . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 038 To compare the role of each locus in a pairwise epistatic interaction , we plotted the phenotypic center for only the metabolites that showed significant epistasis for each pairwise interaction in our pairwise marker model analysis ( Figure 7 and Figure 7—figure supplement 2 ) . For all cytonuclear epistatic interactions , the axis representing cytoplasmic genetic variation showed a wider phenotypic range than the axis representing variation in the nuclear locus ( Figure 7B , C and Figure 7—figure supplement 2 ) . Additionally , the range of phenotypic effects in epistasis between two nuclear loci was typically smaller than that between nuclear and cytoplasmic genome ( Figure 7—figure supplement 2 ) . Thus , variation within the cytoplasmic locus has a larger epistatic effect upon metabolomic variation than genetic variation in the nuclear locus further suggesting that genetic variation in the cytoplasm has larger than recognized phenotypic consequences . The above analysis showed that the cytoplasmic genetic variation plays an important role in mediating variation within the plant metabolome . To quantify how much the inclusion of the cytoplasm improved the genetic models in describing metabolomic variation , we analyzed our additive and pairwise epistasis model models with and without the cytoplasmic terms . For each metabolite for each model , we estimated the percent of variance explained by the significant terms in the model ( Figure 8 ) . This showed that including the cytoplasmic genome in the additive model nearly doubled the average variance compared to additive model containing only the nuclear loci . Adding the pairwise epistatic terms to the additive model without the cytoplasmic genome nearly tripled the fraction of metabolite variance explained by the genetic model without over-fitting the model ( Figure 8 ) . Including the cytoplasmic genome into the pairwise model shifted the average model genetic variance from 38% to 53% ( Figure 8 ) . Thus , the inclusion of the cytoplasmic term into the pairwise model explained as much ( 15% ) , if not more , variance than could be explained from the simple additive model built with only nuclear loci ( 13% ) ( Figure 8 ) . This shows that the cytoplasmic genetic variation is critical to being able to fully describe metabolomic variation within Arabidopsis . 10 . 7554/eLife . 00776 . 039Figure 8 . Distribution of estimated variance between main effects and epistatic interactions . The cytoplasmic term was added or dropped from the different statistical models to compare the total variance explained by each model for each metabolite . Dotted lines show estimated variance using solely the main effect loci without interactions ( additive model ) , solid lines show the distribution of estimated variance across the metabolites using the pairwise epistasis model while the dashed lines are the results for the three-way epistasis model including the most prevalent three-way interactions as indicated by the epistatic network . The gray lines show the models with only the nuclear loci while the black lines show the model with the nuclear and cytoplasmic loci . For the frequency plot , bin size is 0 . 025 r2 . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 039 The epistatic network diagram showed that there were several instances of triangles where two nuclear loci showing pairwise epistasis also interacted with the cytoplasmic variation ( Figure 6 ) . This connectivity between the cytoplasmic genetic variation and nuclear QTL epistatic pairs led us to question if genetic variation in the cytoplasmic genomes could alter epistatic interactions between nuclear loci . To test if cytoplasmic genetic variation could alter nuclear epistasis , we identified all pairs of nuclear loci that interacted with each other and the cytoplasmic genome ( Figure 6 ) . These 10 three-way interactions were then included in the pairwise epistatic model to generate a three-way interaction model ( Figure 6—source data 1 , 2 ) . We found that including these 10 new three-way interactions significantly increased the total variance explained by the model for the distribution of metabolites while only using less than 42% of the available degrees of freedom ( Figure 8 , t-test p<0 . 001 ) . To examine how cytoplasmic genetic variation influences nuclear epistasis we visualized all significant three-way interactions using the phenotypic center approach , ( Figure 9—figure supplements 1–3 ) . This showed that the cytoplasmic variation enhanced or even changed nuclear epistasis . Within the significant Cytoplasm × M . I . 83 × M . IV . 23 epistatic interaction , we identified larger effects of the nuclear loci than had been possible in the pairwise model ( Figures 7 and 9 , Figure 9—figure supplements 1–3 ) . For example pyruvate and metabolite 227710 have quite large nuclear epistatic interactions that are only visible in one or the other cytoplasmic background ( Figure 9B , C ) . Neither metabolite had a large nuclear epistasis due to comparisons averaging across the different cytoplasmic backgrounds ( Figure 7 vs Figure 9 ) . In contrast , salicylic acid had different patterns of nuclear epistasis in the two cytoplasmic genetic backgrounds ( Figure 9D ) . A similar pattern of the cytoplasmic variation altering nuclear epistasis was seen for other three-way interactions ( Figure 9—figure supplements 4–7 ) . 10 . 7554/eLife . 00776 . 040Figure 9 . The cytoplasmic background alters nuclear epistatic interactions . The analysis of the M . IV . 23 × M . I . 83 × cytoplasmic three-way epistatic interaction is shown for all significantly affected metabolites . All other metabolite distributions for three-way epistatic combinations are plotted in Figure 9—figure supplements 1–3 . ( A ) The center of mass calculations were used to estimate the phenotypic center for each metabolite that shows a significant three-way epistasis with the M . IV . 23 , M . I . 83 and cytoplasmic genetic variation . All significant metabolites are plotted as unique points . The specific metabolites boxed and labeled show the location of the metabolites shown in parts , B , C and D respectively . For each locus , the Kas allele is plotted at 1 while Tsu is −1 . ( B ) Effect of the M . IV . 23 × M . I . 83 × cytoplasmic epistasis upon the accumulation of unknown 227710 . Average and standard error are shown . ( C ) Effect of the M . IV . 23 × M . I . 83 × cytoplasmic epistasis upon the accumulation of pyruvate . Average and standard error are shown . ( D ) Effect of the M . IV . 23 × M . I . 83 × cytoplasmic epistasis upon the accumulation of salicylic acid . Average and standard error are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 04010 . 7554/eLife . 00776 . 041Figure 9—figure supplement 1 . Distribution of phenotypic centers of metabolites significantly affected by M . I . 83 × M . IV . 3 × cytoplasmic three-way epistatic interaction . The page shows the distribution of phenotypic centers of metabolites significantly affected by three way epistatic interactions . Of the 10 three-way interactions involving cytoplasmic loci that were tested , only the interactions showing significant epistasis for 10% or more metabolites are shown . All significant metabolites are plotted as unique points . For each locus , the Kas allele is plotted at 1 while Tsu is −1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 04110 . 7554/eLife . 00776 . 042Figure 9—figure supplement 2 . Distribution of phenotypic centers of metabolites significantly affected by M . I . 83 × M . IV . 17 × cytoplasmic three-way epistatic interaction . The page shows the distribution of phenotypic centers of metabolites significantly affected by three-way epistatic interactions . Of the 10 three-way interactions involving cytoplasmic loci that were tested , only the interactions showing significant epistasis for 10% or more metabolites are shown . All significant metabolites are plotted as unique points . For each locus , the Kas allele is plotted at 1 while Tsu is −1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 04210 . 7554/eLife . 00776 . 043Figure 9—figure supplement 3 . Distribution of phenotypic centers of metabolites significantly affected by M . I . 83 × M . V . 82 × cytoplasmic three-way epistatic interaction . For all metabolites that were epistatically affected by M . I . 83 × M . IV . 23 × cytoplasmic three-way interaction , hexbin plot showing change in the distribution of phenotypic centers corresponding interaction of nuclear loci M . I . 83 × M . IV . 23 as compared to the distribution shown in Figure 7D . The number of metabolite per hexbin is shown in the legend to the right of the graph . Center of mass calculations were used to estimate the phenotypic center for each metabolite that was significantly affected by the given interaction . The hexbin plots show the distribution of phenotypic centers for all significant metabolites . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 04310 . 7554/eLife . 00776 . 044Figure 9—figure supplement 4 . Distribution of phenotypic centers of metabolites significantly affected by M . I . 83 × M . IV . 23 × cytoplasmic three-way epistatic interaction . For all metabolites that were epistatically affected by M . I . 83 × M . IV . 23 × cytoplasmic three-way interaction , hexbin plot showing change in the distribution of phenotypic centers corresponding interaction of nuclear loci M . I . 83 × M . IV . 23 as compared to the distribution shown in Figure 7D . The number of metabolite per hexbin is shown in the legend to the right of the graph . Center of mass calculations were used to estimate the phenotypic center for each metabolite that was significantly affected by the given interaction . The hexbin plots show the distribution of phenotypic centers for all significant metabolites . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 04410 . 7554/eLife . 00776 . 045Figure 9—figure supplement 5 . Movement of phenotypic centers of metabolites significantly affected by M . I . 83 × M . IV . 23 × cytoplasmic three-way epistatic interaction . For metabolites epistatically affected by M . I . 83 × M . IV . 23 × cytoplasmic three-way interaction , we plotted the center of phenotypes corresponding to interaction of nuclear loci M . I . 83 × M . IV . 23 as estimated using the three-way epistasis and pairwise epistasis models . For each metabolite , open diamonds indicate the center of phenotype estimated using the pairwise epistasis model while closed diamonds indicate the estimated location using the three-way epistasis model . For each metabolite the pairwise and three-way center of phenotypes are connected by a line to indicate the change in location . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 04510 . 7554/eLife . 00776 . 046Figure 9—figure supplement 6 . Long range movement of phenotypic centers of metabolites significantly affected by M . I . 83 × M . IV . 23 × cytoplasmic three-way epistatic interaction . For metabolites epistatically affected by M . I . 83 × M . IV . 23 × cytoplasmic three-way interaction , we plotted the center of phenotypes corresponding to interaction of nuclear loci M . I . 83 × M . IV . 23 as estimated using the three-way epistasis and pairwise epistasis models . For each metabolite , open diamonds indicate the center of phenotype estimated using the pairwise model while closed diamonds indicate the estimated location using the three-way model . For each metabolite the pairwise and three-way center of phenotypes are connected by a line to indicate the change in location . Only those metabolites whose centers of phenotypes are located farther than 0 . 1 from the origin are provided to increase visual resolution . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 04610 . 7554/eLife . 00776 . 047Figure 9—figure supplement 7 . Distribution of phenotypic center of metabolites significantly affected by M . I . 83 × M . IV . 23 × cytoplasmic three-way epistatic interaction . For metabolites epistatically affected by M . I . 83 × M . IV . 23 × cytoplasmic three-way interaction , we measured the change in distance from the ( 1 , 1 ) point to test if the inclusion of the cytoplasmic genotype has a uniform shift upon the phenotype of the affected metabolites . The histogram with the gray dashed border shows the frequency distribution of metabolites that decreased the distance from the point ( 1 , 1 ) when comparing the three-way to pairwise position . The histogram with black solid border shows the frequency distribution of metabolites that increased the distance . The respective mean and median for each class are shown in the respective lines . DOI: http://dx . doi . org/10 . 7554/eLife . 00776 . 047
Even though maternal contribution to complex phenotypes has long been suspected , most previous genomic surveys of natural variation using transcriptomics or metabolomics have either not had the capacity to directly assess the influence of cytoplasmic genetic variation or just ignore it completely . Thus , we used the reciprocal Kas × Tsu Arabidopsis RIL population to directly quantify the role of cytoplasmic genetic variation in quantitative variation of metabolomics traits . Our analysis showed that the cytoplasmic genome variation affected the phenotypic variation for over 80% of metabolites . The affected metabolites included key parts of central metabolism and some of the most specialized metabolites ( Figures 1 , 5 and 6 ) . Interestingly , the combined variance of the nuclear genome was larger than the variance due to cytoplasmic genome ( Figure 1 ) . In contrast , cytoplasmic genetic variation affected more metabolites than any of the individual nuclear loci typically with larger effects ( Figures 3–5 and Figure 4—figure supplements 1–5 ) . This is true even though the average nuclear locus spanned approximately the same number of genes as the organellar genomes combined . We also identified a high level of significant epistatic interaction between cytoplasmic genetic variation and nuclear genetic variation , as cytonuclear epistasis ( Etterson et al . , 2007; Tang et al . , 2007; Wolf , 2009 ) . This cytonuclear epistasis explained as much , if not more , metabolic variation than the combined additive effects of the nuclear loci . In addition , the cytoplasmic genome controlled larger phenotypic changes than the nuclear locus within a cytonuclear interaction ( Figure 7 ) . This interaction between the cytoplasmic and nuclear loci was further extended to three-way interactions and showed that the cytoplasmic background could hide or alter the interaction between two nuclear loci ( Figure 9 ) . Thus , the cytoplasmic genetic background plays a key role in determining how natural variation within nuclear loci will function . The effect of the cytoplasmic genetic variation was spread out across nearly all of primary metabolism making it impossible to ascribe a causal link with solely plastidic or mitochondrial genes ( Figure 5 ) . Similarly analysis of genomic variation in the two organelles found a large number of SNPs that could be affecting numerous genes within the organelles with an observed bias towards genes involved in the NADH dehydrogenase complex ( Figure 5—source data 1 ) . Thus , organellar genomes could be playing a role in the cytoplasmic genetic influence on metabolite variation , but given their coinheritance , it is not possible to resolve the direct causal polymorphisms . Separating the two genomes to resolve their relative roles would require identifying rare individuals where the two organelles show bi-parental inheritance instead of the uni-parental/maternal mode of inheritance typical for Arabidopsis ( Azhagiri and Maliga , 2007 ) . The plant cytoplasmic genomes contain only about 1% of the number of genes as found within the nuclear genome ( Arabidopsis genome initiative , 2000 ) . Yet genetic variation in this small fraction of genes has a large consequence upon the plants metabolic variation ( Figure 5 ) . This could be from genetic polymorphisms in organellar genes that are central to plant metabolism , such as those central to photosynthesis or NADH synthesis ( Figure 5—source data 1 ) . Alternatively , the polymorphisms could be within the organellar genes that are required to facilitate the function of the several thousand nuclear genes whose protein products are transported into the organelles and function there ( Ajjawi et al . , 2010 ) . Finally , it is possible that the polymorphisms in the organelle impact retrograde signaling pathways , thus altering the function of the nuclear regulatory mechanisms ( Vinti et al . , 2000; Larkin et al . , 2003; Estavillo et al . , 2011; Xiao et al . , 2012 ) . Identifying the specific causal polymorphisms within the cytoplasmic genomes will require direct whole genome sequencing of the organellar genomes from all available genotypes to query the extent of natural variation in the organellar genome ( Cao et al . , 2011 ) . An alternative may be to identify the causal genes underlying the nuclear loci involved in three-way interactions with the cytoplasm to triangulate the identity of the cytoplasmic gene . In this work , we report a first genomic survey of how genetic variation within cytoplasmic genomes influences metabolomic variation . Cytoplasmic genetic variation alters metabolite variation with effects that are equivalent to , if not greater than , individual nuclear loci . More importantly , the cytoplasmic background significantly influences the ability to detect epistasis between nuclear loci . Thus cytoplasmic genomes should be included in any future analysis of natural variation , either by being included as a genotype in GWA studies or by designing future populations as reciprocal populations to allow for direct analysis of the cytoplasmic genomic variation in controlling the phenotype . This inclusion will allow direct assessment of how cytoplasmic genomic variation influences other phenotypic classes , such as transcriptomics or broader physiological phenotypes . Natural genetic variation in the organellar genomes while frequently ignored will have to be kept at the front of future experimental approaches designed to understand the evolution and genetic architecture of organismal phenotypes .
Seeds of the 341 lines of the Kas × Tsu recombinant inbred population were obtained from the Arabidopsis Biological Resource Center ( ABRC , Columbus OH , USA ) ( Juenger et al . , 2006 , 2010; McKay et al . , 2008 ) . We grew a total of four to five plants per line , split into two randomized complete blocks per experiment with two independent experiments separated by approximately 3 months . This provides four independent replicates per genotype . We grew plants in large planting trays with 156 individual wells ( b × w × h: 30 × 25 × 100 mm ) , filled with standard potting soil ( Sunshine Mix #1 , Sun Gro Horticulture , Bellevue WA ) . Prior to sowing , we imbibed seeds in distilled water and cold stratified them at 4°C for 4 days . We placed approximately 3–5 seeds of a single genotype in the center of a well and covered the trays with a transparent plastic hood to retain humidity during germination . Plants from both reciprocal subpopulations where intermixed in the block design to allow direct statistical testing of cytoplasmic effects . After 1 week , we removed the transparent hoods and surplus plants to leave one seedling per well . We watered plants twice a week with nutrient-enriched water ( 0 . 5% N-P-K fertilizer in a 2-1-2 ratio , Grow More 4-18-38 , Grow More Inc . , Gardena CA ) and kept them in a climate-controlled chamber at 22°C and a day/night cycle of 10 hr/14 hr . These plants and experiment were the same as previously described ( Joseph et al . , 2013 ) . 31 days after sowing , we harvested plants for metabolomics analysis . One leaf from the first fully mature adult leaf pair of each plant was removed and ground in extraction solution as previously described ( Rowe et al . , 2008; Chan et al . , 2010 ) . Metabolite identity was determined by comparing retention time and mass to the 2007 UC Davis Genome Center Metabolomics Facility metabolites database ( http://fiehnlab . ucdavis . edu/Metabolite-Library-2007; Fiehn et al . , 2005 ) . Mixed samples were run approximately every 20 samples to optimize the peak identification and quantification algorithms and to control for variation in the detection as previously described ( Fiehn et al . , 2008; Fernie et al . , 2011 ) . The ion count values were used as a surrogate for metabolite abundance . Metabolite abundance was median normalized prior to analysis to account for any technical variation between samples . A separate leaf was extracted for glucosinolates and analyzed by HPLC according to previously described methods with the results reported elsewhere ( Kliebenstein et al . , 2001a , b ) . From the analysis of glucosinolates , a set of 10 lines with aberrant or genetically impossible glucosinolate profiles based on the known parentage were removed from the analysis , leaving a total 316 lines . All RIL lines were represented in every block in both experiments creating a perfectly balanced randomized complete block design . All phenotypic data was used to calculate estimates of broad-sense heritability ( H ) for each phenotype as H = σ2g/σ2p , where σ2g was estimated for both the RIL genotypes and cytoplasmic genotypes and σ2p was the total phenotypic variance for a trait ( Liu , 1998 ) . The ANOVA model ( Line heritability Model ) for each metabolite phenotype in each line ( ygmeb ) was: ygceb= μ+Cc+Gg ( Cc ) +Ee+Bb ( Ee ) +Cc× Ee+ εgceb , where c = the Kas or Tsu cytoplasm; g = the 1…316 for the 316 RILs , e = experiment 1 or 2 and b = block 1…8 nested within experiment . This allowed cytoplasmic effects to be directly tested in the C term and each RIL genotype ( G ) nested within the appropriate cytoplasmic class , either Kas or Tsu . Experiment and block nested within experiment were treated as random terms within the model to better parse the variation . All resulting variance estimates , p values and heritability terms are presented ( Figure 1—source data 1 ) . σ2g for RIL was pulled from the Gg ( Cc ) term while σ2g for cytoplasmic variation was pulled directly from the CcMm term . We used mean values for the RILs for further analysis as we had a randomized complete block design with no missing lines . Additionally , means and LSmeans were correlated with an average r2 of 0 . 96 for the 159 known metabolites present in both experiments . We the previously reported genetic map for these lines of the Kas × Tsu RIL population ( McKay et al . , 2008; Joseph et al . , 2013 ) . To detect metabolite QTLs , we used the average phenotype per RIL across all experiments ( Figure 1—source data 2 ) ( Basten et al . , 1999; Zeng et al . , 1999; Wang et al . , 2006 ) . For QTL detection , composite interval mapping ( CIM ) was implemented using cim function in R/qtl package with a 10 cM window . Forward regression was used to identify three cofactors per trait . The declaration of statistically significant QTLs was based on permutation-derived empirical thresholds using 1000 permutations for each mapped trait . QTLs with a LOD score above 2 were considered significant for further analysis ( Churchill and Doerge , 1994; Doerge and Churchill , 1996 ) . Composite interval mapping to assign significance based on the underlying trait distribution is robust at handling normal or near normal trait distributions ( Rebai , 1997 ) , as found for most of our phenotypes . The define . peak function implemented in R/eqtl package was used to identify the peak location and one-LOD interval of each significant QTL for each trait ( Wang et al . , 2006 ) . The effectscan function in R/qtl package was used to estimate the QTL additive effect ( R Development Core Team , 2012 ) . Allelic effects for each significant QTL are presented as percent effect , by estimating [x¯Tsu−x¯Kas]/x¯RIL for each significant main effect marker . QTL clusters were identified using a QTL summation approach where the position of each QTL for each trait was plotted on the chromosome by placing a 1 at the peak of the QTL . This was then used to sum the number of traits that had a detected QTL at a given position using a 5 cM sliding window across the genome ( Kliebenstein et al . , 2006 ) . The QTL clusters identified defined genetic positions that were named respective to their phenotypic class and genetic positions with a prefix indicating the phenotype followed by the chromosome number and the cM position . For example , M . I . 83 indicates a metabolomics QTL hotspot on chromosome I at 83 cM . The QTLs detected at the previously characterized and cloned glucosinolate AOP locus lies underneath the M . IV . 17 metabolomics hotspot ( Magrath et al . , 1994; Kliebenstein et al . , 2001c , a; Kroymann et al . , 2003 ) . To directly test the additive effect of each identified QTL cluster we used an ANOVA model containing the markers most closely associated with each of the significant QTL clusters as individual main effect terms . For each metabolite the average accumulation in lines of genotype g at marker m was shown as ygm . The model ( Additive Model ) for each metabolite in each line ( ygm ) was: ygm= μ+ ∑g=12∑m=1mMgm+ εgm , where g = Kas ( 1 ) or Tsu ( 2 ) ; m = 1 , … , 14 . The main effect of the markers was denoted as M involving 15 markers ( m ) . The cytoplasmic genome was included as an additional marker to test for cytoplasmic genome effects . We tested all metabolites with the appropriate model using lm function implemented in the R/car package , which returned all p values , Type III sums-of-squares for the complete model and each main effect . QTL main-effect estimates ( in terms of allelic substitution values ) were estimated for each marker ( Fox and Weisberg , 2011; R Development Core Team , 2012 ) . There is no significant single marker or pairwise segregation distortion in this population indicating that the model is balanced for all markers ( McKay et al . , 2008 ) . To test directly for epistatic interactions between the detected QTLs , we conducted an ANOVA using the pairwise epistasis model . We used this pairwise epistasis model per metabolite because we had previous evidence that RIL populations have a significant false negative QTL detection issue and wanted to be inclusive of all possible significant loci ( Chan et al . , 2011 ) . Within the model , we tested all possible pairwise interactions between the markers . For each phenotype , the average value in the RILs of genotype g at marker m was shown as ygm . The model ( Pairwise epistasis model ) for each metabolite in each line ( ygm ) was: ygm= μ+ ∑g=12∑m=1mMgm+ ∑g=12∑m=1m∑n=m+1mMgmMgm+ εgmn where g = Kas ( 1 ) or Tsu ( 2 ) ; m = 1 , … , 14 and n was the identity of the second marker for an interaction . The main effect of the markers was denoted as M having a model involving 15 markers . The cytoplasmic genome was included as an additional single-locus marker to test for interactions between the cytoplasmic and nuclear genomes . p values , Type III sums-of-squares for the complete model and each individual term and QTL pairwise-effect estimates in terms of allelic substitution values were obtained as described for additive model ANOVA ( Fox and Weisberg , 2011; R Development Core Team , 2012 ) . Significance values were corrected for multiple testing within a model using FDR ( <0 . 05 ) . The main effect and epistatic interactions of the loci were visualized using cytoscape . v2 . 8 . 3 with interactions significant for less than 10% of the phenotypes were excluded from the network analysis ( Rowe et al . , 2008; Smoot et al . , 2011 ) . The 10% threshold was chosen as an additional multiple testing correction to provide a more conservative image of the network . There are no pairwise locus segregation distortions within this population showing that the genotypes in this analysis are balanced ( McKay et al . , 2008 ) . The same style of model was run to test for specific three-way interactions by including specific three-way terms as indicated ( three-way epistasis model ) . To plot the epistatic effect of QTLs upon a set of metabolites , we utilized the center of mass calculations . To do this we transitioned the physical distance metric in center of mass to a genetic distance metric for mapping the center of phenotype . In this , we classify one locus of each interaction as being on the x-axis and the other locus being positioned on the y-axis . On each axis , the allelic value of each specific genetic locus is plotted in relation to the heterozygote . For instance , in a two QTL situation the x-axis would be the alleles of QTL1 with the Kas having an allelic value of 1 and the Tsu having an allelic value of −1 while the other axis would be the same for QTL2 there by positioning each of the four homozygous genotypic class in one of the four quadrants . For each metabolite , the center of phenotype was calculated using center of mass calculations as xgeno=∑i=1Npixi∑i=1Npi and ygeno=∑i=1Npiyi∑i=1Npi where p is the average un-scaled phenotypic value for each of the four homozygous genotypic class and x is the x-coordinate of the corresponding genotypic class and y is the y-coordinate of corresponding genotypic class . The center of phenotype ( xgeno , ygeno ) of all the metabolites significant for an interaction were plotted to visualize the distribution centers of phenotype for each pairwise interaction . Kas-1 and Tsu-1 reads were obtained from NCBI ( www . ncbi . nlm . nih . gov/ , accessions SRX246466 and SRX246442 respectively ) and aligned using Bowtie2 ( Langmead et al . , 2009; Langmead and Salzberg , 2012 ) to the Col-0 mitochondrial and chloroplast references obtained from TAIR ( www . arabidopsis . org/ ) ( Rhee et al . , 2003 ) . Aligned sequence reads were subsequently processed using SAMtools ( Li et al . , 2009 ) , Picard ( http://picard . sourceforge . net ) and the Genome Analysis Toolkit ( GATK ) ( McKenna et al . , 2010 ) . SNP discovery between Kas-1 and Tsu-1 was carried out using the UnifiedGenotyper package of GATK ( Figure 5—source data 1 ) . | The vast majority of genes in plant and animal cells are located on chromosomes within the nucleus . However , cells also contain a small number of genes outside the nucleus in cellular organelles such as the mitochondria , which generate energy , and the chloroplasts , which carry out photosynthesis . All these non-nuclear genes comprise the organellar genome . When trying to explain how variation in genes leads to differences in the characteristics of animals and plants , geneticists have historically paid most attention to the genes inside the nucleus . However , more recent work has shown that variation in the organellar genome can also contribute to differences between individuals , although the relative contribution of organellar genes versus nuclear genes remains unclear . Now , Joseph et al . have performed the first large-scale analysis of how variation in the organellar genome affects the characteristics ( or phenotype ) of the plant model organism , Arabidopsis . The study examined the degree to which variation in each of roughly 13 , 000 nuclear genes and 200 organellar genes affected the levels of thousands of metabolites inside cells . This metabolomics analysis revealed that variation in the organellar genome contributed to variation in the levels of more than 80% of the metabolites studied . Organellar genes also helped to regulate the effect of nuclear genes . This combination of direct and indirect influences helps to explain how a small number of organellar genes can have a disproportionately large effect on phenotype . The work of Joseph et al . suggests that the role of the organellar genome has been significantly underestimated to date , and that geneticists should consider variation in both the nuclear and organellar genome when attempting to determine how genes affect phenotype . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology",
"genetics",
"and",
"genomics"
] | 2013 | Cytoplasmic genetic variation and extensive cytonuclear interactions influence natural variation in the metabolome |
Neuronal circuit mapping using electron microscopy demands laborious proofreading or reconciliation of multiple independent reconstructions . Here , we describe new methods to apply quantitative arbor and network context to iteratively proofread and reconstruct circuits and create anatomically enriched wiring diagrams . We measured the morphological underpinnings of connectivity in new and existing reconstructions of Drosophila sensorimotor ( larva ) and visual ( adult ) systems . Synaptic inputs were preferentially located on numerous small , microtubule-free 'twigs' which branch off a single microtubule-containing 'backbone' . Omission of individual twigs accounted for 96% of errors . However , the synapses of highly connected neurons were distributed across multiple twigs . Thus , the robustness of a strong connection to detailed twig anatomy was associated with robustness to reconstruction error . By comparing iterative reconstruction to the consensus of multiple reconstructions , we show that our method overcomes the need for redundant effort through the discovery and application of relationships between cellular neuroanatomy and synaptic connectivity .
Mapping neuronal circuits from electron microscopy ( EM ) volumes is hard ( Helmstaedter , 2013 ) . Manually working through large volumes is slow and prone to attentional errors ( Kreshuk et al . , 2011; Helmstaedter et al . , 2011 ) . Combining multiple independent reconstructions of the same neuron can reduce errors ( Helmstaedter et al . , 2011; Kim et al . , 2014 ) at the cost of multiplying the required labor . Current computational approaches operate only with 'local' information , that is , the EM micrographs and algorithmically detected fine structures such as cell membranes and mitochondria . They are therefore sensitive to noise ( Jain et al . , 2010 ) , particularly in anisotropic EM data where the smallest neurites may be thinner than the thickness of individual serial sections ( Veeraraghavan et al . , 2010; Helmstaedter , 2013 ) . Machine-generated neuron reconstructions are therefore proof-read by humans ( Chklovskii et al . , 2010; Haehn et al . , 2014 ) . Expert neuroanatomists are able to resolve ambiguities that novices and current algorithmic approaches cannot by using large-scale features of neurons to inform decisions made at the level of nanometer-scale image data . For example in Drosophila , where neurons are highly stereotyped , large branches in an EM reconstruction of a given cell can be confirmed by comparing the observed anatomy to that of homologous cells from light microscopy data or other reconstructions ( Takemura et al . , 2013; Ohyama et al . , 2015 ) . This suggests that one way to improve the toolkit for neuron reconstruction and circuit mapping is to facilitate the use of cell- and circuit-level features to find and resolve errors and ambiguities . Crucially , different errors do not alter the wiring diagram equally . Missing small dendrites can be acceptable . Useful and reproducible wiring diagrams can be created even when omitting 56% of all postsynaptic sites ( Takemura et al . , 2013 ) , but missing a single large branch hosting all the synapses in one neuropil region could omit connectivity to entire populations of partners . Prioritizing proofreading time toward those errors that most significantly affect the interpretation of the data improves reconstruction efficiency ( Plaza et al . , 2012; Kim et al . , 2014 ) . To understand the effect of reconstruction errors on measured synaptic connectivity , we need to understand the relationship between synaptic connectivity and cellular neuroanatomy . Mesoscale anatomy , particularly the placement of large branches , is a key component of circuit structure ( Zlatić et al . , 2003 , 2009; Wu et al . , 2011; Couton et al . , 2015 ) . Similarly , the connectivity graph of a stereotyped circuit can relate back to anatomy by consideration of the location of the synaptic sites between pairs of neurons . However , little is known about the smallest scales of synaptic connectivity , the distribution of individual synapses on a neuron . Microtubule-free and actin-rich structures have been identified as key sites of excitatory input in the adult Drosophila visual system ( Scott et al . , 2003; Leiss et al . , 2009 ) , but it is unclear how ubiquitous these are in the nervous system . Here , we describe a collection of quantitative anatomical and connectivity features across scales , from fine dendritic branches to multi-neuron graphs , and tools for measuring them to swiftly and accurately map a wiring diagram from EM . We implemented the calculation and visualization of such features on-demand as an extension of the web-based large image data viewer CATMAID ( Saalfeld et al . , 2009 ) . We propose a novel method for interactively using these features to reconstruct neuronal circuits through iterative proofreading at the level of both EM images and higher level features . We validated this approach by comparing the speed and accuracy of our iterative method to a consensus method , where multiple independent reconstructions are used to calculate regions of agreement across individuals ( Helmstaedter et al . , 2013 ) . Because the detection of high-impact errors can occur concurrently with reconstruction via interactive analysis , our tool removes the need for time-consuming repeated reconstructions ( Helmstaedter et al . , 2013; Kim et al . , 2014 ) . Moreover , because reconstructed neurons did not need to be hidden to ensure independence between repeated reconstructions , our method facilitates concurrent , synergistic collaboration between expert neuroanatomists who , for example , map circuits in different brain regions that happen to spatially overlap or synaptically interact . We demonstrate our methods by mapping a sensorimotor circuit in the Drosophila larva from proprioceptive sensory neurons to motor neurons .
We extended the web-based image data viewer CATMAID ( Saalfeld et al . , 2009 ) to enable a geographically distributed group of researchers to map neuronal circuitry . A neuron is reconstructed with a skeleton , a directed tree graph with one or more nodes in every cross-section of neurite in an EM volume ( Helmstaedter et al . , 2011; Cardona et al . , 2012 ) . Nodes have a spatial coordinate , as well as metadata including authorship , timestamp , review status , and optional annotations such as a radius value , text labels . Importantly , nodes also have a confidence value that can be lowered to indicate uncertainty in following a branch . Where possible , we root skeletons at the soma to model the anatomical notions of proximal and distal in the data structure . Synapses ( Figure 1A and Figure 1—figure supplement 1 ) are annotated as a relation from a node on the presynaptic neuron skeleton to an intermediate 'connector node' and then to a node of a postsynaptic neuron skeleton . To express the polyadic nature of insect synapses ( Meinertzhagen and O’Neil , 1991 ) , connector nodes can have multiple postsynaptic 'targets' , but only one presynaptic 'source' . Reconstructions are immediately synchronized across all collaborators to avoid duplicate or conflicting work , and to take advantage of existing reconstructions to aid further reconstruction and circuit discovery . 10 . 7554/eLife . 12059 . 003Figure 1 . EM ultrastructure shows synapses and microtubule cytoskeleton . ( A ) EM micrograph of a typical Drosophila synapse with a single presynaptic site ( red asterisk ) and multiple postsynaptic sites ( blue asterisks ) . Scale bar is 200 nm . ( B ) Microtubules in neural processes are visible in EM sections whether cut transverse ( top inset , red arrowheads ) or obliquely ( bottom inset , red arrowheads ) . ( C ) Microtubules in a given neuronal process span several sections ( three shown here; microtubules were traced over 16 sections ) and maintain their relative orientations . Microtubules are color coded as in the processes in B and were traced and visualized in TrakEM2 . ( D ) Synaptic distribution ( red , presynaptic site; blue , postsynaptic site ) across the arbor of larval neuron A23a . ( E ) Microtubule distribution of larval neuron A23a . Black indicates the microtubule-containing backbone continuous with the soma , green are microtubule-free twigs . See Video 1 for both microtubules and synapses shown together . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 00310 . 7554/eLife . 12059 . 004Figure 1—figure supplement 1 . Synapses of neurons with different neurotransmitters . For all examples , white arrow points at the most prominent slice of the T-bar . Each panel measures 1024 nanometers on the side . ( A ) Two examples of synapses in the glutamatergic neuron A02b ( a 'looper' or 'PMSI' neuron; Kohsaka et al . 2014 ) spanning twelve and six 50 nm sections respectively . Glutamatergic synapses vary considerably in size and number of postsynaptic partners ( from 3 to over 15 ) . ( B ) An example of a cholinergic synapse ( a sensory axon , dbd; Yasuyama and Salvaterra , 1999 ) spanning 12 sections . ( C ) An example of a synapse from a GABA immunoreactive cell type , A31k , spanning 9 sections . The arrowhead annotates the T-bar at panel 6 for the GABAergic synapse . The black thick line crossing panel 9 in A31k is the shadow of a fold in the support film . ( D ) Two examples of the typically small , dyadic synapses found in serotonergic neurons like A26d ( Chen and Condron , 2008 ) . Synapses in the serotonergic neurons typically span only 2 sections ( 100 nm ) and contact 2 or occasionally 3 postsynaptic partners . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 00410 . 7554/eLife . 12059 . 005Video 1 . Rotation of the A23a neuron showing both synapses ( red , presynaptic sites; blue , postsynaptic sites ) and presence of microtubules ( black , with microtubules; green , without microtubules ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 005 As a case study of our method , we focused on sensorimotor circuits in an abdominal segment of the first instar Drosophila larval central nervous system ( CNS ) using an EM volume covering one and a half abdominal segments ( Ohyama et al . , 2015 ) . In total for this work , nine lab members reconstructed and proofread 425 neuronal arbors spanning 51 . 8 mm of cable , with 24 , 068 presynaptic and 50 , 927 postsynaptic relations , ( see 'Materials and methods' for details ) . Reconstruction time was 469 hours for reconstruction with synapse annotations plus 240 hours for review ( see below ) , for an average rate of ∼73 microns of proofread arbor with synapses per hour . To be able to use neuronal anatomy to guide circuit reconstruction , it was crucial to better understand the distribution of synaptic input onto Drosophila neurons . We started by looking in detail at the relationship between the synaptic inputs ( Figure 1A–B ) and microtubule cytoskeleton ( Figure 1C–E ) in EM reconstructions of neurons from different regions of the nervous system and life stages . For a diverse collection of neurons , we marked all locations where the arbor continued distal to a microtubule-containing process ( Figure 1E , Figure 2A ) . We call such a terminal branch a 'twig' . By definition , all twigs have their base on a microtubule-containing backbone shaft . Following the classification in Leiss et al . ( Leiss et al . , 2009 ) , a spine is a twig with a maximal depth of less than 3 µm and that is not a presynaptic varicosity ( Figure 2A ) . 10 . 7554/eLife . 12059 . 006Figure 2 . Twigs , small microtubule-free neurites , are the primary site of input in Drosophila neurons . ( A ) Twigs less than 3 µm are considered spine-like , while those longer or primarily presynaptic are not . ( B–F ) EM reconstructions ( middle ) of Drosophila neurons from different parts of the nervous system ( left ) showing backbone ( black ) and twigs ( green ) . At right , the fraction of all synaptic inputs onto short spine-like twigs , longer twigs , and backbone . Data sets are indicated by marks: no asterisk: 1 . 5 segment volume . *: Whole CNS volume . **: 3rd instar abdominal segment volume . ***: Adult medulla skeletons and images , generously provided by Janelia FlyEM [9] . Neurons are individually scaled to show complete arbors . ( B ) motor neurons in 1st instar larva . ( C ) Premotor interneurons of 1st instar larva . ( D ) Interneurons in the brain of the 1st instar larva . ( E ) A somatosensory interneuron cell type across life stages , 1st instar and 3rd instar larvae . ( F ) Tm3 cells in the adult fly medulla . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 006 We found twigs in all neurons investigated , across multiple CNS regions and life stages of Drosophila , and in all cases , they were the dominant sites of synaptic input ( Figure 2B–F ) . We first considered larval motor neurons aCC and RP2 ( Landgraf et al . , 1997 ) , which have functional and structural similarities to vertebrate neurons ( Sánchez-Soriano et al . , 2005; Nicolï et al . , 2010; Günay et al . , 2015 ) . In the first instar CNS , we find aCC and RP2 have numerous twigs , which together host more than 80% of their total number of postsynaptic sites ( Figure 2B ) . We found a similar majority of inputs onto twigs in three hemisegmental pairs of premotor interneurons ( Figure 2C ) and brain neurons ( Ohyama et al . , 2015 ) in the first instar larva ( Figure 2D ) . We tested whether the observed distribution of postsynaptic sites onto twigs generalizes across larval stages by comparing a somatosensory interneuron in the first instar to its homologue in late third instar ( Figure 2E ) . At both life stages , we find more than 80% of inputs were onto twigs , suggesting that twigs are not a temporary developmental structure . In the adult fly , light microscopy-level analysis of lobula plate tangential cells of the visual system suggests a similar distribution of postsynaptic sites onto twigs ( Leiss et al . , 2009; Scott et al . , 2003 ) . We annotated EM skeletonizations of medullar Tm3 neurons reconstructed by Takemura et al . ( 2013 ) in the adult visual system neuropil and found that nearly all their inputs were onto twigs ( Figure 2F ) . Our findings suggest that microtubule-free twigs are a general feature of Drosophila neurons and constitute the primary anatomical location of synaptic input . Spine-like twigs are found in all cell types , but host a variable , typically non-majority , amount of synaptic input ( Figure 2C–F ) . We consider all twigs for the remainder of our analysis . For a given presynaptic partner , a postsynaptic neuron could concentrate its input synapses onto a single region or distribute them widely . The spatial distribution of synaptic inputs has implications for dendritic processing ( Polsky et al . , 2004 ) , developmental robustness ( Couton et al . , 2015 ) , and as we show , reconstruction accuracy . To study the relationship between presynaptic neurons and the anatomical locations of post-synaptic sites , we reconstructed all neurons synaptically connected to motor neurons aCC and RP2 in the third abdominal segment of a first instar larva ( Figure 3A–F ) . 10 . 7554/eLife . 12059 . 007Figure 3 . Twigs are crucial to larval motor circuitry . ( A ) The EM volume covers one abdominal segment ( blue box ) of the ventral nerve cord . ( B ) Sagital view of the EM volume . Note segmentally repeated features . ( C ) Dorsal projections of genetically labeled motor neurons RP2 ( top , from 1st instar ) and aCC ( bottom , from 3rd instar ) . Each cell type has characteristic dendritic arbors . Midline indicated by gray arrowhead . ( D ) EM reconstructions of each of four motor neurons aCC and RP2 in the 1st instar larva match the left and right homologs of aCC and RP2 . Backbone is indicated by black , twigs by colors . Midline is shown as dashed line . ( E ) True spatial relationship of the four motor neurons in ( D ) , shown dorsally ( left ) and in cross-section ( right ) . Note that the boundary of the EM volume is limited . ( F ) All arbors presynaptic to aCC and RP2 . Colors indicate if neuron is presynaptic to one or both motor neuron cell types . See Video 2 for rotated views of the arbors . ( G ) Histograms of premotor partners connected via number of synapses . ( H ) Colored lines: the cumulative fraction of total inputs as a function of ranked presynaptic partn ers for each motor neuron are extremely similar . Black dashed line: simultaneous fit for all four motor neurons to 1 - exp ( -r/ρ ) for rank r gives ρ = 22 . 34 . ( I ) Scatterplot and histogram of the total length and number of synapses on each of the 305 twigs for each of the four motor neurons ( colors as previous ) . ( J ) Number of twigs contacted by motor neuron partners as a function of the number of synapses in the connection . Crosses are median , boxes the interquartile range , whiskers the 10th to 90th percentiles . Outliers shown . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 00710 . 7554/eLife . 12059 . 008Figure 3—figure supplement 1 . Counts of reconstructed neuronal arbors . We reconstructed 425 arbors which fall into 4 groups: 'identified neurons' ( arbors that could be associated with a single neuron name or at least with a lineage of origin , even if the whole arbor is not present within the imaged EM volume ) , 'intersegmental neurites' ( neurites that cross the volume from anterior to posterior , the majority of which are an unbranched axon that synapses onto motorneurons ) , 'spillovers' ( partial arbors that are not recognizable and which originate in neurons in the anterior or posterior segments ) and 'ambiguous fragments' ( very small arbors with few inputs or outputs and which terminate within the boundaries of the volume; most of them terminate at a three-section gap between sections 348 and 352 ) . Identified neurons are either synaptic partners of dbd , aCC or RP2 , or are any of the other 16 proprioceptive axons ( the left and right ddaD in segments 3 and 4; ddaE in segments 2 and 3; dmd1 and vbd in se gment 3; and dbd in segments 2 and 4 ) or 9 motorneurons ( the left and right RP5 , U1 , U2 and the unpaired VUMs in abdominal segment 3 ) . While arbors in the intersegmental group could not be identified , in numerous occasions the left and right homologs are recognizable , given idiosyncratic characteristics ( not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 00810 . 7554/eLife . 12059 . 009Figure 3—figure supplement 2 . Bundles of premotor axons that run the length of the imaged volume . These axons originate in neurons whose somas are located in areas of the central nervous system beyond the limits of the imaged EM volume . We distinguish five bundles: dorsolateral ( blue ) , low intermediate bundle ( yellow ) , middle intermediate bundle ( green ) , dorsal intermediate bundle ( magenta; only present on the left side and weakly connected to motor neurons ) , and medial bundle ( cyan ) . The dotted lines delimit the imaged volume , which had a tilt of about 8 degrees relative to the transversal plane . Arbors , particularly the most dorsal ones ( magenta ) may appear outside the plane or short of the imaged limits due to the perspective projection . Below , wiring diagram bundle-wise onto the motor neurons RP2 , RP5 , and aCC . The name of each bundle includes the number of member neurons in brackets; asymmetries between left and right originate in the fact that some members contribute a single synapse onto mot or neurons and may not appear on the other side . Intrabundle edges are not shown . Notice how the only large source of inputs onto RP5 is from the dorsolateral bundle , which also places many synapses onto RP2 . Note that we did not reconstruct all synaptic partners of RP5 . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 00910 . 7554/eLife . 12059 . 010Figure 3—figure supplement 3 . Numerically high synapse edges are distributed over many twigs in adult Tm3 neurons . Each input onto the Tm3s analyzed in Figure 2 is a data point ( see legend ) . The x-axis is the number of synapses this edge is comprised of , the y-axis is the number of distinct twigs this edge spans . As in the larval motor neuron data , edges with multiple synapses are almost always distributed across multiple twigs . Points are jittered to avoid overlap . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 01010 . 7554/eLife . 12059 . 011Video 2 . Rotation of all arbors ( colored skeletons ) presynaptic to RP2 motor neurons ( black skeletons ) . ( Red dots are presynaptic sites , cyan are postsynaptic sites ) . Dorsal is up . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 011 A dynamically generated and interactive table of synaptic connectivity in CATMAID enabled users to systematically trace all connected arbors . We found 198 identifiable neurons ( Figure 3—figure supplement 1 ) and named them according to a developmental lineage-based nomenclature ( Ohyama et al . , 2015 ) , classified 107 other arbors spanning the full segment into eight distinct intersegmental bundles ( Figure 3—figure supplement 2 ) , and classified 120 small fragments that could not be joined into larger arbors . We refer to the connection between a pre- and postsynaptic neuron as an ‘edge’ in the connectivity network , where each edge has a weight equal to the number of synapses between the two neurons . Motor neurons each received between 1 and 28 synaptic inputs from individual presynaptic neurons , with a maximum of 7 . 3% of all inputs coming from a single neuron ( Figure 3G ) . The fraction of synapses accounted for by their presynaptic partners , rank-ordered by number of synapses , is well-fit by an exponential survival function , with a decay indicating that approximately the top 22 presynaptic partners of one motor neuron contribute 63% of all its synaptic inputs ( Figure 3H ) . We next asked how the synaptic input onto aCC and RP2 is distributed across independent twigs . Most individual twigs were small , with the median twig measuring 1 µm in cable and hosting 1 postsynaptic site . The largest typical twig had 16 µm of cable and 20 postsynaptic sites ( Figure 3I ) . We find that presynaptic neurons connect to between 0 ( backbone only ) and 13 twigs , with nearly all connections with 3 or more synapses per edge being distributed across multiple twigs ( Figure 3J ) . Similarly , numerically strong edges spanned multiple twigs in the adult visual system Tm3 neurons ( Figure 3—figure supplement 3 ) . Different neuronal compartments have different metabolic requirements , such as vesicle recycling at presynaptic sites or restoring resting ion concentrations after postsynaptic response to neurotransmitter signaling ( Attwell and Laughlin , 2001; Perkins et al . , 2010 ) . To investigate whether the spatial distribution of mitochondria is a signature of different arbor compartments , we annotated the location of all mitochondria in the four motor neurons and the six premotor interneurons from Figure 3F ( Figure 4A–C ) . Most mitochondria ( 348/425 ) were associated with backbone across motor neurons ( Figure 4D ) and interneurons ( Figure 4E ) . Surprisingly , we found that 97% of central presynaptic sites were located within 3 µm of a mitochondrion ( Figure 4F ) , although only 47% of cable was located within the same distance . A similar rule did not hold with postsynaptic sites , which were more broadly distributed ( Figure 4G ) . This suggests that presynaptic sites and mitochondria are kept near one another , making mitochondrial proximity a useful constraint for validating synapse annotation . 10 . 7554/eLife . 12059 . 012Figure 4 . Mitochondria are associated with presynaptic sites and cytoskeleton . ( A ) EM micrograph shows clear mitochondria ( labeled with M ) and a nearby presynaptic site ( red arrowhead ) . ( B ) Dorsal view of motor neuron RP2 with locations of mitochondria indicated ( top , circles ) and synaptic sites ( bottom ) . ( C ) Dorsal view of interneuron A31k with locations of mitochondria indicated ( top , circles ) and synaptic sites ( bottom ) . See Video 3 for both mitochondria and synapses shown together . ( D ) Number of mitochondria associated with backbone and twig locations on selected motor neurons . ( E ) Number of mitochondria associated with backbone and twig locations on selected interneurons . ( F ) Histogram of the distance between presynaptic sites and their nearest mitochondrion along the arbor for the interneurons in E . Cumulative distribution indicated as a line . ( H ) Histogram of the distance between presynaptic sites and the nearest backbone along the arbor for the interneurons in E . Cumulative distribution indicated as a line . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 01210 . 7554/eLife . 12059 . 013Video 3 . Rotation of A31k showing both synapses ( red , presynaptic sites; cyan , postsynaptic sites ) and mitochondria ( blue dots ) . Anterior is up . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 013 Consistent with this , presynaptic sites were typically also directly associated with microtubules ( Figure 4H ) . Approximately 50% of presynaptic sites were located on the backbone and 90% were within 3 µm . We next looked at the cell and circuit level for regularities that could inform proofreading . In the Drosophila larva , developmentally homologous neurons are strongly stereotyped ( Li et al . , 2014 ) , making quantitative analysis of their properties useful for identifying irregularities between homologous cells . Most cell types are represented in the fly nervous system by at least one homologous bilateral pair of individual cells . Bilateral homology suggests that both arbor morphology and synaptic wiring are mirrored , up to developmental noise ( Ohyama et al . , 2015 ) . To let morphology guide proofreading , we developed a collection of neuroanatomical measurements that were independent of absolute location . These metrics , combined with 3d visualization , quickly summarize the structure of complex neurons to help identify and localize inconsistencies ( Figure 5 ) . 10 . 7554/eLife . 12059 . 014Figure 5 . CATMAID presents multiple , interactive views on EM imagery and quantitative features . ( A–F ) An example of a CATMAID session in the Chrome web browser ( Google , Inc . ) . Different aspects of a pair of connected neurons , A02k and RP2 , are shown across each pane . The number , quantity , location , and neurons in each panel are controllable . ( A ) An image pane shows the EM data , all reconstructed nodes in the view ( purple dots ) , synapse connector nodes ( orange dots ) , and the active node ( green dot , indicated by thin white arrowhead ) . The current active node belongs to an RP2 motor neuron and is postsynaptic to a synapse on interneuron A02k , indicated by the thick white arrowhead . ( B ) Graph representation of a collection of six neurons , including the selected pair indicated as above . Edge labels indicate the number of associated synapse ( red arrowhead ) . ( C ) The pair of neurons indicated in ( A ) , shown in a 3d viewer ( orange , RP2; blue , A02k , indicated as above ) . The active node in the image pane is shown by a green dot in the viewer ( indicated by red arrowhead , also the location of the synapse shown at left ) . ( D ) List of synapses between A02k and RP2 , represented in the graph pane by an edge ( red arrowhead in B ) . Each row is clickable , letting the contributor jump to that location to permit fast reviewing of specific connections . ( E ) Plot of quantitative morphological or network measurements of the six neurons in ( B ) . ( F ) Connectivity list shows neurons synaptically connected to selected neurons ( here , RP2 ) and counts the total number of synapses . The row for the presynaptic neuron A02k is offscreen . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 014 As a case study , we applied our tools to describe a complete sensorimotor circuit . During forward crawling , a peristaltic wave of muscle contraction travels from posterior to anterior segments ( Hughes and Thomas , 2007; Heckscher et al . , 2012 ) . Signals from the segmentally repeated proprioceptive neurons dbd have been suggested to act via a polysynaptic inhibitory pathway to stop motor neuron activity after successful contraction of a given segment ( Hughes and Thomas , 2007 ) . To find pathways between proprioceptive and motor neurons , we further reconstructed axons for proprioceptive sensory neurons dbd , vbd , dmd1 , ddaD , and ddaD ( Hughes and Thomas , 2007; Grueber et al . , 2007 ) . Because of its implication in proprioceptive feedback ( Hughes and Thomas , 2007 ) , we further reconstructed all partners of the left and right dbd ( Figure 6A , B ) . 10 . 7554/eLife . 12059 . 015Figure 6 . Graph search to identify consistent networks . ( A ) The motor neuron RP2 and proprioceptive sensory neuron dbd , shown in transverse . ( B ) All synaptic partners of RP2 and dbd in ( A ) . ( C ) Five symmetric pairs of identified neurons link the two cell types with three or fewer hops of at least three synapses each , as found by search in CATMAID . All edges are observed in both the left and right hemisegments , except for a single edge outside the volume boundary ( red dashed line , see Figure 6—figure supplement 1 ) . Line thickness increases with number of synapses ( maximum and minimum values shown ) . In this and all network diagrams , single synapse edges are not shown for clarity . ( D ) All identified cells in EM ( left ) could be matched to confocal maximum intensity projections of single neurons found in sparsely labeled GAL4 lines ( right , see 'Materials and methods' for details ) . For neuroglian staining , an approximate neuropile boundary is shown; for nc82 staining , the blue region is a profile of neuropile . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 01510 . 7554/eLife . 12059 . 016Figure 6—figure supplement 1 . Four pairs of left and right homologs in posterior view , where one of the pairs ( canonical ) conforms with the arbor shape found in light microscopy ( not shown ) and the other presents deviations ( A , ’A’: A02d , B , B’: A10a , C , C’: A23a , D , D’: dbd ) . There are two cases of an early split of the axon ( A02d a3l and dbd a3r; dotted circle marks the split ) ; normally the split would occur at the proprioceptive domain ( where the output synapses are , in red ) . A10a a3r presents a misrouted axon that reaches the correct target area ( and connects to the same neuron types as the canonical homolog does ) but sprouts an ectopic , supernumerary dendrite along the way ( which accounts for 14 input synapses out of 178 total ) . A23a a3l presents a correct arbor but the path to the soma is different . Connectivity-wise , deviant neurons are very similar to their canonical homologs . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 01610 . 7554/eLife . 12059 . 017Figure 6—figure supplement 2 . Sections were cut approximately 8 degrees from transverse . Due to the volume limits , observed circuitry could be asymmetric , as in the case of A02l synapsing onto 31k ( see Figure 6C ) . The region where the contact occurs ( red circle ) is present on one side of the data but not the other . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 01710 . 7554/eLife . 12059 . 018Figure 6—figure supplement 3 . GABA immunolabeling of proprio-motor interneurons . ( A-O ) Dorsal views . ( A , D , G , J , M ) Projections of fluorescently labeled single-cell clones of identified neurons ( courtesy of James W . Truman , HHMI Janelia Research Campus ) . ( B , E , H , K , N ) Dorsal views of projections of parent lines used to generate single-cell clones , expressing myr::GFP . ( C , F , I , L , O ) Single z-plane at high magnification of cell indicated by arrowhead in ( B , E , H , K , N ) showing immunoreactivity to anti-GABA ( magenta ) and GFP ( green ) . ( C’ , F’ , I’ , L’ , O’ ) Same view as ( C , F , I , L , O ) only showing the anti-GABA channel in grayscale . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 018 Using a graph search within CATMAID , we identified all 1–3 hop pathways from dbd to motor neuron RP2 . Comparison of the identifiable intermediate neurons revealed five pairs of homologous neurons with consistent shape , connectivity , and quantitative morphological properties ( Figure 6C , D ) . Inconsistencies in any property led to further review to determine if they were due to reconstruction error , true developmental variability ( Figure 6—figure supplement 1 ) , or limitations of the raw data . For example , one strong inconsistency in this network , the connection from A02l to A31k ( Figure 6C ) , was due to the expected synapse locations being outside the imaged volume on one side but not the other ( Figure 6—figure supplement 2 ) . The five pairs of identified neurons could also be matched to light-level images of neurons identified through sparse stochastic labeling ( Nern et al . , 2015 ) of neurons within a GAL4 expression pattern ( Figure 6D ) . Of these , two directly premotor interneurons ( A27j and A31k ) are immunoreactive to anti-GABA ( Figure 6—figure supplement 3 ) , whereas the others , all from A02 lineage , are members of the glutamatergic neuron class described in Kohsaka et al . ( 2014 ) . These novel , putatively inhibitory sensorimotor pathways are well-positioned to mediate a hypothesized 'mission accomplished' signal ( Hughes and Thomas , 2007 ) . This map also could explain why genetic silencing of A02 neurons was shown to slow peristalsis ( Kohsaka et al . , 2014 ) , as doing so removes a major channel for proprioceptive feedback which is necessary for normal rates of persitaltic waves ( Suster and Bate , 2002 ) . The physiology of synaptic input and output can differ across neuronal compartments . For example , presynaptic inhibition ( inhibitory synaptic input onto axon terminals ) is important for gain control in fly sensory circuits in a fundamentally distinct manner than dendritic inhibition ( Clarac and Cattaert , 1996 ) . This suggests that connectivity can be stereotyped at the compartmental level and therefore useful for proofreading . We thus sought a graph representation of a circuit that could faithfully distinguish distinct types of connections ( Figure 7 ) . 10 . 7554/eLife . 12059 . 019Figure 7 . Enriching graphs with anatomical compartments . ( A ) Cartoon example of splitting neurons using synapse flow centrality ( SFC ) . ( B ) Examples of two premotor interneurons split into axonal ( darker ) and dendritic ( lighter ) regions with this method . Split point is indicated by the arrowhead . See Video 4 ( A03a1 ) and Video 5 ( A02b ) for rotated views of synapses and splits . ( C ) Splitting interneurons into axonal and dendritic compartments in a proprio-motor circuit reveals stereotypic pre- and post-synaptic connectivity to premotor interneuron A03a1 and differential contributions from proprioceptor dbd relative to other proprioceptors dmd1 , ddaD , and vbd . Note that the axo-axonic connection from dbd to A03a1 is only 2 synapses , and thus would not appear in Figure 6A . ( D ) Splitting interneurons A27j , A27e , and A27h reveals GABAergic pre- and post-synaptic input to a premotor connection , as well as dendro-dendritic coupling between interneurons that connect to synergistic motor neurons aCC and RP2 . ( E ) Dorsal projections of A27h and A27e from EM ( left ) and light ( right ) , as in C . Midline indicated by arrowheads . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 01910 . 7554/eLife . 12059 . 020Figure 7—figure supplement 1 . Neurons are distributed throughout the complete range of possible segregation indices . Top , plot of segregation index vs . cable length for the 39 pairs of identified neurons . Neurons span the full range of the segregation index . Some neurons present completely unsegregated arbors like the serotonergic neurons A26d , with large arbors that present a mixture of inputs and outputs throughout; or compact neurons that are also fully mixed like A27e . Other neurons present intermediate segregation index , which generally takes one of two forms: purely postsynaptic dendrites with mixed axons ( i . e . axons that receive a number of inputs ) like A08a , and neurons that additionally present a small amount of dendrodendritic output synapses ( e . g . A23a ) . Finally , some neurons present purely segregated arbors , with dendrites with only postsynaptic sites and axons with only presynaptic sites , like A19l . At bottom left of the chart , the majority of neurons are intersegmental premotor neu rons ( partial axonal arbors from neurons present in segments not in the imaged EM volume ) . Note that neurons A27e ( both sides ) and A08 a3r have their soma outside the imaged EM volume . SI , segregation index; red dots: neurons presynaptic to aCC or RP2; cyan dots: neurons postsynaptic to dbd; violet dots: neurons both postsynaptic to dbd and presynaptic to aCC or RP2; yellow dots: motor neurons , proprioceptive axons or serotonergic neurons not downstream of dbd . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 02010 . 7554/eLife . 12059 . 021Video 4 . Rotation of A03a1 showing both synapses ( red , presynaptic sites; cyan , postsynaptic sites ) and axon/dendrite split ( magenta skeleton , axon; black skeleton , dendrite ) . Dorsal is up . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 02110 . 7554/eLife . 12059 . 022Video 5 . Rotation of A02b showing both synapses ( red , presynaptic sites; cyan , postsynaptic sites ) and axon/dendrite split ( orange skeleton , axon; black skeleton , dendrite ) . Dorsal is up . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 022 In Drosophila , many neuronal cell types have distinct input and output compartments , while a few have entirely intermingled inputs and outputs . Our approach assumes that the neuron can be split into distinct compartments , and at the end checks to see if the split was successful . First , we calculate all paths along the skeleton from each of the neuron’s input synapses to each of its output synapses and for each node of the skeleton compute the number of centripetal ( toward soma ) and centrifugal ( away from soma ) paths that pass through it ( Figure 7A–B ) . This quantity , which we call “synapse flow centrality” ( SFC ) , is analogous to a synapse-specific version of betweenness centrality ( Newman , 2010 ) . For most neuronal arbors , we find that the most proximal skeleton node with the highest centrifugal SFC corresponds to an intuitive generalization of the locations of spike initiation zones in known polarized neurons of Drosophila ( Gouwens and Wilson , 2009; Günay et al . , 2015 ) and other insects ( Gabbiani et al . , 2002 ) . We quantify how completely input and output are separated on a neuron with a 'segregation index , ' an entropy-based measure of the amount of input/output mixing in each compartment , normalized by that of the whole arbor ( see 'Materials and methods' ) . A very low segregation index means that pre- and post-synaptic sites are intermingled and an axon/dendrite compartmentalization is inappropriate ( Figure 7—figure supplement 1 ) . Using this approach , we classified all identifiable neurons found in both the left and right hemisegments of the proprio-motor circuitry described above . Of the 3834 synapses between these cells , we found 79% were axo-dendritic ( 3033 ) , 11% axo-axonic ( 424 ) , 9% dendro-dendritic ( 334 ) and 1% dendro-axonic ( 43 ) . We consider two examples of how compartment-enriched graphs add important anatomical detail to small microcircuits . First , we analyzed how different proprioceptive inputs converge onto motor neuron RP2 ( Figure 7C–E ) . By splitting interneuron A02b into axon and dendrite , we observed that its dendrites receive bilateral proprioceptive input , while its axon synapses both onto the ipsilateral RP2 and axo-axonically onto its strong premotor partner , A03a1 in both hemisegments ( Figure 7C ) . In contrast , while dbd only connects indirectly with A02b ( Figure 6C ) , it synapses exclusively ipsilaterally and axo-axonicaly onto A03a1 ( Figure 7C ) . This suggests that the role of dbd in modulating motor patterns could be qualitatively different than the other proprioceptive sensory neurons , since its direct pathways are typically longer or involve connections types other than axo-dendritic . Second , we analyzed interactions between the premotor neurons of aCC and RP2 ( Figure 7D , E ) . We found that a neuron presynaptic to the aCC motor neuron on both sides targets dendro-dendritically a pre-RP2 neuron ( A27h ) , potentially coordinating the joint excitation of their targets ( Figure 7D ) . We also found a premotor interneuron ( A27e ) with reciprocal connections with a GABAergic premotor interneuron ( A27j; see Figure 6—figure supplement 3 ) that receives convergent inputs from dorsal proprioceptive neurons ( dmd1 , ddaD , ddaE; Figure 7D ) . This suggests that A27j might not only act as an inhibitory premotor input in response to proprioceptive activity , but also have subtler modulatory effects onto other sources of motor input . Specific connections can also be allocated to specific arbor compartments , which could be used to localize proofreading guided by inconsistencies in connectivity . We thus extended the concept of splitting a neuron into two arbor compartments to an arbitrary number , by defining a compartment as a cluster of synapses near each other along the arbor cable ( see 'Materials and methods' ) . As an example , we consider the axon terminal of dbd , which enters at the interface between two segments and extends symmetric arbors toward the anterior and posterior segments ( Figure 8A ) . The synapses form multiple well-separated clusters that we can visualize as a group of graph nodes ( Figure 8B–C ) , revealing that the anterior and posterior branches synapse onto homologous interneurons ( A08a ) for their respective segments ( Figure 8D–E ) . This pattern suggests that each A08a cell gets convergent input from the dbd of two consecutive segments , which could reflect that adjacent pairs of segments move together during locomotion ( Heckscher et al . , 2012 ) . 10 . 7554/eLife . 12059 . 023Figure 8 . Enriching graphs with compartments defined by synaptic density . ( A ) Dorsal view of the axon terminal of dbd . Dashed line indicates , segmental boundary between A2 and A3 , numbers indicate clusters of synapses . ( B ) Synapse density mapped onto the arbor . Regions were given a weight as the sum of Gaussian functions ( σ = 3 μm ) of the distance to each synaptic site . Colormap is a log scale , arbitrary units . ( C ) Resulting neuron with four nodes , one for the basin of each density peak . Note that the topological structure between clusters ( as defined by the peak location ) is preserved . ( D ) Transverse view of interneuron A08a , shown here in segment A3 . ( E ) Dorsal view of the overlap between dbd ( blue ) and the A08a in segment A3 ( green ) and segment A2 ( yellow ) . ( F ) Network of the dbd extended by synapse clustering and A08a . Different clusters have different synaptic regions with the segmentally repeated interneuron . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 023 Based on the suite of features described above , we developed a two-step iterative method of proofreading after an initial reconstruction ( Figure 9 ) . An initial systematic review consists in traversing a whole arbor from distal to proximal to follow the expected gradual changes in anatomical properties ( e . g . caliber tapering and cytoskeletal changes from microtubule-free to increasing number of microtubules ) . By freeing mental attention from complex spatial navigation , we found that the systematic review leads to the quick discovery of attentional errors , such as missed synapses , or anatomical inconsistencies , such as a non-contiguous microtubule cytoskeleton . The systematic review status is stored on a per-skeleton node and per-contributor basis ( see 'Materials and methods' for details ) . To allow contributors to incorporate the level of proofreading for each neuron into their evaluation of the neurons and circuits , review status of a neuron is displayed throughout CATMAID , as measured by fraction of the skeleton nodes reviewed . For example , when listing synaptic partners for neurons of interest , the review status of all partners is shown alongside information such as neuron name and number of associated synapses . 10 . 7554/eLife . 12059 . 024Figure 9 . The typical reconstruction and proofreading workflow . While reconstruction decisions occur only based on the image data , feature-based comparisons inform specific areas of interest for further proofreading . Each stage in this process can take advantage of the work of other collaborators . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 024 Next , contributors reconstruct the same set of neurons in a different hemisegment ( such as the contralateral side ) and then inspect high-level quantitative anatomical and connectivity measurements for inconsistencies ( Figure 9 ) . In most cases , these inconsistencies can be associated with specific compartments of a neuronal arbor , which are then subjected to focused review . This approach helps ensure that the broad structure of the neuron is consistent and that the large branches are correct , as errors in them would significantly alter the reconstructed circuit . CATMAID provides tools for transitioning from a potential error identified at a high level , to the images supporting the reconstructed skeletons involved . The key property enabling this is the interactivity of CATMAID’s built-in analytical tools that allow for navigating from graphs , 3d views and plots to lists of synapses and spatial locations , and ultimately to the original skeletonizations overlaid on the images ( Figures 5 and 9 ) . Irregularities noticed in higher level features only serve to guide attention , not determine correctness . Any error correction is performed manually on the basis of local information as contained within the EM images ( e . g . microtubules , texture , or consistency with neighboring neurites ) . Despite strong stereotypy in general , developmental variability is present even at the level of high-order branches although often in ways that do not affect connectivity ( Figure 6—figure supplement 1 ) . Our approach to circuit mapping consists of a single initial arbor reconstruction , followed by edits by the same or different collaborators during proofreading or incidental discovery of errors during subsequent work . Small arbor pieces , left over from pruning when proofreading other neurons or from explorative reconstructions in search of specific neurons , are merged in . We refer to this as “iterative , ” as compared to consensus methods that combine multiple independent reconstructions ( Helmstaedter et al . , 2011; Takemura et al . , 2013; Kim et al . , 2014 ) . We evaluated the accuracy of our method for Drosophila circuits by comparing our results to the those of a consensus approach . We selected six interconnected neurons from the premotor network for independent reconstruction by four individuals . Each individual skeletonized and reviewed his or her reconstructions . Consensus skeletons were then computed for each neuron using RESCOP ( Helmstaedter et al . , 2011 ) . Both methods resulted in extremely similar arbors , although each method found branches not seen in the other ( Figure 10A , Figure 10—figure supplement 1 ) . All sites of disagreement between the two methods were validated by an expert to determine a gold-standard morphology . Reconstruction and review of these six neurons in the iterative approach took a total of 26 . 37 hours , while the redundant method by four people took a total of 107 . 73 hours , almost exactly four times as long . 10 . 7554/eLife . 12059 . 025Figure 10 . Comparison of iterative reconstruction to a consensus method . ( A ) Dorsal view of two of six neurons for which we compared our iterative reconstruction method to a RESCOP-generated consensus of four independent reconstructions . Arbor found in both , dark lines; iterative only , blue; consensus only , orange . ( B ) The adjacency matrix produced by our iterative method has an identical set of edges as that of the consensus method , with variability only in the amount of synapses per edge . ( C ) The weights of each edge ( the amount of synapses ) are similar between methods . ( D ) Point errors in iterative reconstructions are not distributed equally across the cable of neuronal arbors , instead falling overwhelmingly on twigs . ( E-G ) Branches missed by our iterative method but observed in the consensus method are typically very small and lightly connected as seen from histograms of their ( E ) cable length , ( F ) synaptic inputs , and ( G ) number of branch points . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 02510 . 7554/eLife . 12059 . 026Figure 10—figure supplement 1 . Four independent reconstructions of a six neuron circuit . ( A ) Morphology of the six neurons used for the comparison . The arbors of these neurons were reconstructed independently four times , and used for generating a consensus skeleton for each arbor using the RESCOP method . Branches found only on the consensus and only in the CATMAID approaches are indicated . Neurons 1 and 2 are A02k , 3 and 4 are A31k , and 5 and 6 are A27l . ( B ) Graphs of each of the four independent reconstructions . Notice that all four individuals agree on almost all edges with similar amounts of synapses per edge , except in a missing edge for two tracers between A02k a3l and A31k a3r . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 026 Existing consensus approaches only calculate neuronal morphology , not synaptic connectivity . Each chunk of the consensus skeleton is associated with the subset of the independent skeletons that are mutually consistent at that location . For some branches , all four individuals agreed , while in others the consensus was based on fewer skeletons . We estimated the connectivity between consensus skeletons by adding each postsynaptic site from each independent skeleton in the consensus , normalized by the number of skeletons contributing to the consensus at that location . Therefore , a given synapse would have a weight of one , the typical value , if it were annotated in all independent skeletons . We found that both methods recover an identical set of edges in the wiring diagram , with similar number of synapses per edge ( Figure 10B , C ) . We next considered the fine differences between consensus skeletons and skeletons reconstructed with our method . The six gold-standard neurons had a total of 1341 postsynaptic sites , with 111 on neurites only present in the consensus skeletons , 229 on neurites only in our method’s reconstructions , and 1001 in the arbor found by both . We located 91 missed or incomplete branches ( false negatives ) in our method , 89 in twigs and 2 in backbones; and 7 incorrect continuations ( false positives ) , 6 in twigs and 1 in backbone . False positives added 30 incorrect postsynaptic inputs in total . Individual missed branches were small in size , complexity , and number of synapses ( Figure 10E–G ) , with more than 40 missed or truncated twigs having no influence on connectivity ( Figure 10B , C ) . The 3 errors in backbones occurred in small distal dendritic shafts containing one single microtubule , resulting in 7 missed and 4 false postsynaptic sites . Error rates for synaptic output were even lower . The gold-standard neurons had a total of 510 presynaptic sites , of which 509 were found by our iterative reconstructions . Our data suggest that anatomical structure strongly influences the rate of reconstruction errors in our iterative method . Our total error rate is dominated by false negatives and is much higher for twigs ( 16 . 2 µm/error ) than for backbone ( 375 . 8 µm/error ) . While attentional errors seemed to dominate missed branches , data ambiguities were often associated with backbone errors . One backbone false merge happened across two adjacent sections in poor registration with one another , while an erroneous truncation occurred across a section where electron-dense precipitate occluded the neurite and its surrounding area . Neuroanatomy strongly constrains the effect of a typical error on the wiring diagram because , as shown above , the most likely error is to miss a twig and an individual twig hosts few or no synapses . To estimate the probability of omitting a true edge in the wiring diagram , we analyzed the distribution of synaptic contacts across twigs as a function of the total number of synapses per edge . Edges comprising multiple synaptic contacts were found to be distributed across multiple twigs ( Figure 3J ) . With the RESCOP-based validation , we found that our method identified 88% ( 672/761 ) of twigs , containing 91 . 7% of synapses ( 1230/1341 ) . From these two observations , we estimated the probability of completely missing a true edge as a function of the number of morphological synapses per edge . We found that our method recovers more than 99% of the wiring diagram edges that have at least 3 synapses ( Figure 11A ) , assuming twigs are missed uniformly at random ( see Figure 11—figure supplement 1 ) . 10 . 7554/eLife . 12059 . 027Figure 11 . Estimating errors that affect graph topology . ( A ) Estimated probability of fully omitting an edge as a function of how many synapses were on the edge based on omitting random twigs with the frequency observed in the validation data . ( B ) Cartoon of dendritic overlap between RP2 and aCC , U1 , and U2 . On average , 91 axons put at least two synapses on any motor neuron ( denoted N in the false positive estimate model , see text for details ) , of which 33 are not connected to RP2 ( denoted N0 ) . ( C ) Probability that , given a pair of homologous postsynaptic neurons , introducing m false inputs randomly distributed across N presynaptic neurons yields at least one pair of false edges of kθ or more synapses each . The number of axons were estimated in b , and false input counts are shown estimated from the validation data ( m = 5 ) , as well as if they came from adding a rare but large twig ( m = 20 ) , and the largest observed twig ( m = 37 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 02710 . 7554/eLife . 12059 . 028Figure 11—figure supplement 1 . We look for clustering in the spatial distribution of errors found by comparison with multiple independent reconstructions . For each of the 89 missed branches , we computed the distance to the nearest other error . The cumulative distribution of such distances is shown in red . We compared this distribution to that obtained by randomly sampling 89 nodes from twigs across all six neurons , disallowing two nodes from the same twig . Distributions for each of the 1000 samples are shown in gray . Only 65/1000 differ from the observed distribution in a two-sample Komologorov-Smirnov test with a p-value < 0 . 05 . We thus conclude that spatial clustering of errors is minimal . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 028 In Drosophila , we are primarily interested in the most reliable edges between cell types , as those are most likely to generalize across individual animals . Moreover , we are concerned less about adding extra synapses to true connections and more about adding false edges that would be interpreted as pathways that are not actually present . To estimate the likelihood of introducing a false edge between cell types not just once , but twice ( e . g . in a left and right pair of homologs ) , we simulated adding false twigs to a neuron . The probability of adding a false edge depends both on the probability of adding a false twig ( observed false positive error rate: 7 errors in 605 twigs ) and the number of nearby but unconnected neurons with presynaptic sites . This will vary depending on the circuit in question . For example , a neuropile with all-to-all connectivity will have no opportunity for false positive edges , while in an array of rigorously separated labeled lines any false positive synapse would be a false positive edge . Further , larger neurons offer more opportunities for false positives than smaller neurons . For a concrete and realistic example , we consider the motor neuron RP2 ( a large neuron ) . We estimated the number of proximate but unconnected neurons by considering all axons presynaptic to all motor neuron dendritic fields that overlap RP2’s dendrites ( Figure 11B ) . We assume that a false-positive reconstruction error distributes m synapses across all available axons at random . Even if we assume that m is always among the largest observed ( m = 20 , which is far larger than the average; Figure 3I ) , our model suggests that for the RP2 wiring diagram we can trust symmetric connections of at least 2 synapses ( Figure 11C ) . We further note that the small size of individual twigs and the ability in CATMAID to jump directly to the image data associated with synapses comprising an edge make review of a suspect false positive edge extremely fast , on the order of seconds . Since most errors were of omission and took the form of truncated twigs , we also measured the effect of omitting the distal ends of twigs . Considering again aCC and RP2 , we looked at the connectivity observed by considering only synapses located at a given depth into the twig relative to its base on the backbone ( Figure 12A ) . With all twigs cropped to zero depth , only inputs onto the backbone remain . More than 90% of postsynaptic sites lay within 5 µm of the backbone ( Figure 12B ) . We observed that the first 2 µm already yields at least two synapses , recovering ∼90% of the most connected partners . The first 4 µm similarly detects ∼90% of all partners with 2 or more synapses and 27/28 pairs of homologous edges ( Figure 12C ) . These results indicate that , given the observed distribution of synapses over multiple twigs , wiring diagram edges with many synapses are robust to errors of omission such as truncated twigs . Considering the marginal time involved in reconstructing the full extent of twigs ( Figure 12D ) , this robustness could be intentionally exploited towards discovering strong synaptic partners in a time-efficient manner . 10 . 7554/eLife . 12059 . 029Figure 12 . Proximal regions of twigs reflect final wiring ( A ) Cartoon of the proximal depth ( red ) into of a twig ( green ) measured from from the backbone ( black ) . ( B ) The fraction of two or more synapse edges onto aCC and RP2 that would be found when considering only synapses onto the backbone and twigs cropped at a maximum depth . From light to dark gray are those edges whose final measured connectivity has more than two , four , and eight synapses . Blue dashed line indicates fraction of all input synapses . ( C ) The fraction of pairs of homologous edges from identified neurons ( N=28 edge pairs ) that would be identified using synapses up to a given depth . ( D ) Fraction of total reconstruction time for each of the four motor neurons ( see legend ) as a function of cropping twigs at a maximum depth . Note that 0µm depth cropping corresponds to backbone reconstruction only . DOI: http://dx . doi . org/10 . 7554/eLife . 12059 . 029
Neurons are highly structured cells . A human expert’s success at circuit mapping from EM volumes stems from the ability to use this structure and apply cell and circuit-level context to interpret nanometer-scale image data . Here , we presented our approach to circuit mapping in EM by building tools in CATMAID that ease and emphasize the use of high level features concurrent with image-level reconstruction . While every reconstruction and editing decision is performed manually , it is informed by a host of quantitative neuroanatomical and connectivity measures computed on demand and , where possible , tightly linked to specific locations in the EM image volumes . In addition to applying existing metrics , we also devised novel algorithms and measures to describe the distribution of synapses across neurons , a feature uniquely well-measurable by EM . Because this method is based extensively on existing information , contributors iterate reconstructions towards more and more correct states . We showed that this more efficiently produces data at least as accurate as computing the consensus of multiple independent reconstructions . Central to our approach is the observation that Drosophila neurons contain a contiguous microtubule-rich backbone and numerous small microtubule-free distal twigs . We found that small twigs are the primary site of synaptic input for Drosophila neurons and that numerically strong connections between neurons are spread across many distinct twigs . If , contrary to observations , neurons were to only contact each other via a single twig that hosts many postsynaptic sites , then this connection would be fragile with respect to developmental noise ( Couton et al . , 2015 ) . Backbones define the spatial extent and stereotyped shape of a neuron , and we found that most presynaptic sites are located on or very near the backbone’s microtubules and mitochondria . Our findings are consistent with the notion that metabolic needs and microtubule-based trafficking are limiting factors for the structure of synaptic output . These different biological requirements for different neuronal compartments are reflected in the rate of reconstruction errors . The large calibers and relatively gradual turns associated with microtubules made errors on backbone less frequent by a factor of nearly 20 relative to on smaller and tortuous twigs . However , we propose that the circuit’s resilience to developmental noise , achieved in part by connecting via multiple twigs , underlies the resilience of wiring diagrams to the omission of small dendritic branches , the most typical error observed both here and in reconstructions in the fly visual system ( Takemura et al . , 2013 ) . Irregularities within a cell type guide review toward small fractions of specific neuronal arbors that could be responsible for a potential error . While reconstructing a neuron , a user can quickly pull up its complete anatomy and connectivity to compare to homologous cells or inspect for irregularities and , crucially , return immediately to the locations in the image data necessary to make the appropriate decisions . We find that this smooth flow from image data to high level features and back to image data—without post hoc or offline analysis—is possibly the most important feature in our EM reconstruction technique . Dispensing with repeated reconstruction without reducing accuracy enables our method to support concurrent neuron reconstruction by many collaborators . This setup prevents duplicated work while ensuring that important locations are visited multiple times . For example , synaptic relations are inspected at least twice in different ways , once each from the pre- and postsynaptic side . The presence of existing and correct skeletons in complicated areas , such as registration errors between consecutive sections or gaps , reduces the time necessary for resolving possible ambiguities and effectively provides an extra step of proof-reading by not allowing contradictory reconstructions . Further savings originate in the reuse of data , for example exploratory reconstruction of backbones in search of specific neurons or branches pruned during proofreading are merged into the arbor currently being reconstructed . In summary , in a collaborative environment , the more neurons that are reconstructed , the faster new ones can be added , and the fewer errors existing reconstructions will contain . Automated methods will be necessary to map circuits with more than a few thousand neurons ( Helmstaedter , 2013 ) , but they require extensive proof-reading ( Chklovskii et al . , 2010; Plaza et al . , 2012; Haehn et al . , 2014 ) . Our methods for analysis of arbor morphology , synaptic distribution and circuit structure and reliability , and their application in proof-reading , apply equally to manually and automatically reconstructed neurons . Neuroanatomical measurements suggest mixed strategies for leveraging both automated algorithms and human effort . For example , mitochondria can be reliably located automatically ( Lucchi et al . , 2011; Funke et al . , 2014 ) which , together with our finding of a distance constraint between mitochondria and presynaptic sites , could assist in automated synapse detection ( Kreshuk et al . , 2011; Becker et al . , 2012; Kreshuk et al . , 2014 ) . Similarly , the properties of neuronal backbone and twigs suggest that algorithms for the automatic detection of microtubules in serial section EM would be a profitable source of constraints for automated reconstruction of neurites across consecutive sections ( Vazquez-Reina et al . , 2011; Funke et al . , 2012 ) . While we only considered the relationship between error rate and the presence or absence of microtubules , with the use of automated detection methods it will be important to look at more detailed measures of arbors such as the number of microtubules , curvature , or caliber . Our iterative reconstruction method explicitly uses fundamental properties of dendritic arborizations to achieve circuit reconstruction accuracy without sacrificing speed . Both in insect and in vertebrates , dendritic arbors present two structural compartments: one with microtubules–the shafts–and one without– all the spines and twigs ( Rolls and Jegla , 2015 ) . Shaft synapses are less likely to be missed because both sides of the synapse present microtubules . Our finding that the connection between two neurons is resilient to errors of omission of spines or twigs is based on the redundancy afforded by the distribution of postsynaptic sites across the receiving arbor . Instances of this pattern have been observed in mammals in multiple brain regions ( Hamos et al . , 1987; Bock et al . , 2011; Kasthuri et al . , 2015 ) . However , the particulars of the distribution of postsynaptic sites over the postsynaptic arbor will be specific to each species—even to brain regions or cell types within each species—and therefore must be measured for the tissue of interest . We have shown that this distribution enables estimating an acceptable false negative rate , the amount of missed synaptic connections that still allow the recovery of strong synaptic connections in the wiring diagram , and thus determines the minimal effort that must be dedicated to twigs or spines . A fundamental difference between insect and mammalian neurons is that in mammals , the axons can be of small caliber , but represent a substantial fraction of all cable ( Helmstaedter et al . , 2013 ) . Nonetheless , axons need transport of cytosolic elements such as vesicles and mitochondria , which are delivered primarily via microtubules , as well as the anchoring of such elements onto the microtubule cytoskeleton ( Sheng and Cai , 2012 ) . As we have shown , microtubules provide a strong signal that enables overcoming ambiguities in serial section EM . Therefore cytoskeletal detail , as visible in rich EM images , can greatly facilitate the reconstruction of mammalian axons despite their small calibers . Catastrophic reconstruction errors are those that affect the backbone of a neuron , dramatically altering the observed circuit wiring diagram . These are generally false positives that originate at ambiguous regions of the image data , and which result in the addition of a large incorrect branch . In Drosophila , we exploited the strong morphological stereotypy and unique identity of every neuron to swiftly detect these kind of errors , which are rare , by comparing the overall arbor morphology of homologous neurons across individuals or bilaterally . In nervous systems without uniquely identifiable neurons , these kind of structural errors can be detected either by comparing a reconstructed arbor with prior light-microscopy imaging of the same sample ( Bock et al . , 2011; Briggman et al . , 2011 ) , or by compiling statistical descriptions of cell type morphology and connectivity ( Sümbül et al . , 2014; Jonas and Kording , 2015 ) . Larval locomotion , like many motor patterns , results from rhythmic activation of motor neurons ( Heckscher et al . , 2012 ) , but few central components of the underlying premotor circuitry had been identified ( Kohsaka et al . , 2014; Couton et al . , 2015 ) . Our reconstruction of propriomotor circuitry revealed a complex network comprised of numerous cell types , including a subset of those previously described ( Kohsaka et al . , 2014 ) . We identified a rich collection of local neurons , including neurons anatomically well-suited to provide common drive to synergistic muscles ( Schaefer et al . , 2010 ) and thus likely a key motor network components . Using anatomically faithful simplifications of neuronal structure , we found several premotor microcircuits employing dendro-dendritic and axo-axonic synapses in parallel with conventional axo-dendritic synaptic connections . For example we found a GABAergic input pre- and post-synaptic to motor neuron input , a motif also observed in mammalian motor circuits ( Fyffe and Light , 1984 ) . Although the motor system is rhythmically active in the absence of sensory input ( Suster and Bate , 2002 ) , proprioceptive sensory feedback is required for natural coordination and timing ( Hughes and Thomas , 2007; Song et al . , 2007 ) . We found diverse and complex circuitry for relaying proprioceptive information , including GABAergic and glutamatergic neurons directly relaying proprioceptive input to motor neurons . This motif is well-posed to provide an inhibitory 'mission accomplished' signal to suppress motor neuron activity after a successful contraction during forward locomotion ( Hughes and Thomas , 2007 ) . However , we also observed that many synaptic partners of dbd were themselves presynaptic to neurons downstream of the other proprioceptive axons , suggesting other , more complex roles for proprioceptive feedback in modulating motor activity . Surprisingly , the axon terminals of proprioceptive neurons themselves almost entirely lacked presynaptic input . This suggests that proprioceptive input is privileged by the larval nervous system and not under fast , dynamic modulation by central circuitry ( Clarac and Cattaert , 1996 ) , unlike proprioceptive afferents in the locust leg ( Burrows and Matheson , 1994 ) and other somatosensory modalities in the larva ( Ohyama et al . , 2015 ) . Wiring diagrams have been deemed necessary , yet not sufficient , for understanding neural circuits ( Bargmann , 2012 ) and a fast approach for discarding hypotheses of circuit function ( Denk et al . , 2012; Takemura et al . , 2013 ) . The neuronal wiring reconstructed here offers insights into the structure of an insect motor circuit and its control by sensory feedback , and serves as a complementary resource for detailed functional studies . With the circuit mapping tools and methods demonstrated here , fast , accurate , and targeted reconstruction of circuits in Drosophila larva ( Ohyama et al . , 2015 ) and adult , and other species ( e . g . Platynereis , Randel et al . , 2015 ) is possible .
We rewrote and greatly developed the Collaborative Annotation Toolkit for Massive Amounts of Image Data , CATMAID ( Saalfeld et al . , 2009 ) ( GPL ) , to implement our methods for neural circuit reconstruction , visualization and analysis , and with a user and group management system with sophisticated permissions for graded access . The toolkit consists of four parts: ( 1 ) the client ( a web page ) , and three types of servers , namely ( 2 ) an application server based on the Django web framework ( https://www . djangoproject . com ) , ( 3 ) one or more image volume servers , and ( 4 ) an instance of the relational database PostgreSQL ( http://www . postgresql . org ) with all non-image data , which includes metadata such as the spatial information of skeletons , the location of which types of synaptic relations , the text annotations , timestamps and provenance of every action . The original web client accesses , in constant time , arbitrary fields of view of remote stored image volumes . We have greatly extended this capability to include 3-way views ( XY , XZ and ZY ) and a number of color overlays for multi-channel data such as light-microscopy images or computed derivative data such as membrane probability maps . Analysis of neurons and circuits is performed primarily in the client using the programming language JavaScript , relying on a large number of open source libraries for numerical processing , data management and visualization ( D3 . js , Numeric Javascript , Cytoscape . js , three . js , jsNetworkX , Raphaël , jQuery , SVGKit ) . Offline analysis for validation and probability calculations was performed by custom scripts in MATLAB ( Mathworks ) . Documentation and installation instructions are available at http://catmaid . org and code is available at https://github . com/catmaid/CATMAID . Wild-type Drosophila first instar larval central nervous systems were manually dissected by mechanical separation of the anterior tip of the larva from the posterior portion in PBS , and immediately transferred to 2% glutaraldehyde in 0 . 1 M Na-cacodylate , pH 7 . 4 buffer . Samples were post-fixed in 1% OsO4 in the same buffer and stained en bloc with 1% aqueous uranyl acetate before subsequent dehydration in ethanol and propylene oxide , and embedding in Epon . Serial 45 nm sections were cut with a Leica UC6 ultramicrotome using a Diatome diamond knife , and picked up on Synaptek slot grids with Pioloform support films . Sections were stained with uranyl acetate followed by Sato’s lead ( Sato , 1968 ) . Sections were then imaged at 4 . 4 nm × 4 . 4 nm resolution using Leginon ( Suloway et al . , 2005 ) to drive an FEI Tecnai 20 transmission electron microscope . The resulting 77 , 000 image tiles were contrast-corrected , montaged and registered with TrakEM2 ( Cardona et al . , 2012 ) using the nonlinear elastic method ( Saalfeld et al . , 2012 ) . The generated data volume of 22 , 775×18 , 326×462 voxels corresponds to a volume of 91×73×21 µm3 . The data set covers approximately the posterior half of abdominal segment A2 , and a nearly complete abdominal segment A3 . For display in CATMAID , we Gaussian-smoothed montages of registered EM images ( sigma=0 . 7 pixels , sufficient to remove high-frequency noise to increase the effectiveness of JPEG compression without sacrificing perceptual image quality ) and then generated an image pyramid with five zoom levels and diced it to collections of 256 × 256 pixel tiles ( 512 × 512 and larger can work better for fast Internet connections ) , stored in JPEG format ( 75% compression and stripped of headers with jpeg-optim ) . This approach reduced data storage from over 700 to 90 gigabytes , which were served from a fast seek time solid-state hard drive . We setup a single server machine ( Intel Xeon X5660 with 12 cores , 48 GB of RAM , 10 Gb network card ) running Ubuntu 12 . 04 to host the PostgreSQL database , the image server and the Django server . LDAP id caching was enabled for best performance . Images were stored on high-performance solid-state drives mounted with noatime flag or as read-only , and served via proxy with in-RAM varnishd for caching . The database was configured with large shared buffers ( 4 GB ) and autovacuum on ( naptime: 8642 min; scale factor 0 . 4; analyze scale factor 0 . 2; cost delay -1; cost limit -1 ) for optimal performance . We chose to serve pages with Nginx , running 8 processes , with epoll on , 768 worker connections , disabled logs and gzip on ( except for JPEG image tiles ) for best performance , and with public caching and no-expire settings for images . Django was run via Gunicorn with python 2 . 7 using 8 processes . Reconstruction in CATMAID , as presented here , is based on manual annotation of neurons from image stacks . Where possible , we’ve honed the interface to reduce the amount of interaction users need to perform while reconstructing neurons . Humans skeletonize a neuronal arbor by placing nodes within a window showing the image stack . A new node , generated by a key press or mouse click , becomes the topological ‘child’ of an explicitly selected ‘active node’ . The active node updates to the most recently created node , and branch points are generated by selecting a node that already has child nodes and generating an additional one . When branches are observed , we place a small stub down that branch to ease later follow-up . Since neurons are typically linear at short distances , nodes do not have be manually placed in every image section . In image sections between those containing manually placed skeleton nodes , CATMAID produces intermediate , linearly interpolated virtual nodes that can be interacted with like a real skeleton node . If edited ( e . g . moved , attached to a synapse , or used as a branch point ) , the virtual node becomes a real node manifested in the database . When a continuation for a neuron has already been reconstructed , the two arbors can be merged with a single click . This is particularly common in the case of synapses , where pre- or post-synaptic placeholder , isolated skeleton nodes ( i . e . a skeleton consisting of a single node ) are placed to fully describe the synapse . If two arbors both have multiple skeleton nodes , an attempted merge of the two must be confirmed visually in an interactive 3d display , to avoid obvious inconsistencies like multiple soma or an unexpected arbor structure ( e . g . spatially intertwined dendritic branches ) . Other , optional metadata can also be associated with nodes . A neurite radius can be measured for each skeleton node , with cylinders modeling arbor segments in 3d and spheres modeling the somas . Each skeleton node may also be tagged with any number of arbitrary text snippets to express metadata . Search tools enable finding tags in a specific skeleton or across the data set . Free text can be used to denote structures of interest ( e . g . 'Golgi apparatus' ) or as a personal or team communication convention ( e . g . 'check this synapse' ) . Tags used frequently or to guide analysis ( e . g . ‘microtubules end’ ) can be mapped to a single key press . Synapses are made by creating a special class of node ( a ‘connector’ ) that can be related to a single presynaptic node and multiple postsynaptic nodes . We opt to place the connector node within the presynaptic neuron near the visible characteristics of the synapse , the presynaptic density and nearby vesicle cloud . Each relation is annotated with a confidence value from 1 to 5 , with 5 being the default and highest confidence . Again , efficient user interaction simplifies manual annotation . A synaptic connector node can be created with a single click either pre- or post-synaptic to the active skeleton node , based on holding different modifier keys . When the connector node is active , clicking nodes while holding modifier keys produces a pre- or post-synaptic link to it . If no node is already present when the user does this , a new node is generated with the desired link type . Alternative types of connectors with their own constraints ( for example abutment or gap junctions ) are straight-forward to add . To guide the flow of reconstruction , we rely on special text tags that describe a user’s decisions about how to continue a neurite . By default , a node without any child ( i . e . a ‘leaf node’ in the topological tree structure ) is expected to need further continuation . When a user decides they have reached the end of a neurite , they can hit the ‘K’ key to label that node with the text tag ‘ends’ , indicating that no more continuation is necessary . In cases of ambiguity in how to continue , other tags are used to indicate no further continuation is currently possible: ‘uncertain end’ , to be used if the user cannot decide if a neurite ends , and ‘uncertain continuation’ , to be used if there is an expectation of a continuation due to , for example , microtubules , but the specific continuation is unclear . In effect , then , the process of finishing a neuronal reconstruction becomes a task of continuing every unlabeled leaf node until no more exist . Eventually , all open leaves have been explicitly declared finished and the first draft of the arbor is complete . To ease this process , a single key press ( ‘R’ ) will position the field of view on the closest unlabeled leaf node of a selected neuron , making it simple and efficient to jump to to the next part of the arbor in need of reconstruction . Systematic review of a skeleton consists in visualizing each of its skeleton nodes in sequence , adding or editing nodes and synaptic relations as necessary . For this purpose , we partition the arbor to generate the smallest possible set of the largest possible sequences of continuous nodes to minimize the number of times that the reviewer has to switch to a different arbor path . We sort leaf nodes by path length ( in number of intermediate nodes ) to the root node in descending order . Starting from the most distal leaf node , we generate a sequence of nodes all the way to the root . Then we pick the second most distal node and generate another sequence of nodes until reaching a branch point that has already been assigned to a sequence , and so on for each remaining leaf node . When done , sequences of nodes are sorted by length . The reviewer then iterates each sequence , automatically marking each node as reviewed upon visiting it ( using ‘q’ and ‘w’ key bindings to go forward and backward in a sequence , and ‘e’ to jump to the beginning of the longest unreviewed sequence ) . As a visual aid , each node is centered in the screen , facilitating the detection of changes in the contour of the sectioned neurite , as well as drastic shifts of the field of view that indicate an error ( e . g . a jump to an adjacent neurite ) . The enforcement of a unique directionality and simple one-dimensional path—from distal ends towards inner parts of the arbor or the soma—facilitates noticing glaring inconsistencies such as a path starting off large and microtubule-rich , then reducing to small and microtubule-free , then becoming again large and microtubule-rich . In other words , a review approach coherent with the expected tapering out of neurite caliber and cytoskeleton from soma to distal ends adds context that helps the reviewer . The total fraction of nodes of a skeleton that have been reviewed is indicated in most tools that can display lists of neurons ( e . g . selection table , connectivity tables , connectivity graph ) , as well as a skeleton coloring mode in the 3d viewer . This enables simple visualization of the current status of review of e . g . all upstream and downstream synaptic partners of one or more neurons in the connectivity widget , of all neurons in a wiring diagram in the connectivity graph , or the review status of a specific branch in the 3d viewer . Given that one or more users may review any node of a skeleton , and the different proficiency of each user , settings allow users to create of a team of other users whose reviews are trusted . Review status visualization can thus be limited to only a user’s own reviews , to the union of everyone’s reviews , or to the union of all reviews performed by the team of trusted reviewers . Synapses are effectively reviewed multiple times , given that they are seen from at least two arbors ( the pre- and the postsynaptic ) ; more in the case of polyadic synapses , as nearly all synapses in the Drosophila larva are . We consider synapses as two elements: the presynaptic relation between a skeleton node and a connector node and the postsynaptic relation between a skeleton node of a neuron and the connector node . Reviewing the associated skeleton node tacitly marks its part of the synapse as reviewed . To further facilitate systematic review , a 'Skeleton Analytics' tool automatically detects and lists in an interactive table some potential issues that must be addressed in a neuron or collection of neurons . The listing is interactive , allowing jumping to the associated field of view in the image data to determine whether or not the issue describes a genuine error . Electron microscopy image volumes of neuropils contain noise . For serial section transmission EM , noise originates during fixation ( e . g . broken membranes and reduced extracellular space ) , serial-sectioning ( e . g . folds , cracks , missing sections , thick sections ) , counter-staining ( e . g . precipitated heavy metals , dust particles , or absence of staining due to microscopic air bubbles ) , and imaging ( e . g . locally uneven illumination , tile-wise constant noise originating in improper correction of the camera’s dark- and brightfields ) ; for examples see Supplemental Figure 2 in Saalfeld et al . , 2012 . The most common form of noise consists in missing data either as a partial occlusion of a section , or by the loss of one or more sections . When reconstructing a neuronal arbor , upon reaching an area with missing data ( a gap ) , we use both global and local cues to identify the correct continuation , labeling the skeleton edge that crosses the gap with an appropriate confidence value to express our degree of certainty in the decision . These low-confidence skeleton edges enter into the visualization and analytical tools for further evaluation . Generally , the direction , caliber , and cytoplasmic characteristics of the neuron and its neighboring neurons suffices to identify the corresponding continuation on the other side of the gap . The larger the gap and smaller the neurite , typically the lower the confidence in the identification of the correct continuation . Locally , gaps up to 500 nm ( e . g . 10 serial sections ) are crossable using microtubules . The number , direction , and spatial arrangement of microtubules in a neurite are constant over lengths of micrometers , making them reliable structures over many sections ( Figure 1 ) . Similarly , mitochondria take tubular shapes inside neurites , and their sparseness and relatively constant dimensions identify a neurite across consecutive serial sections ( Figure 1 ) . Other cues can include the smooth endoplasmatic reticulum that lines large and mid-size neurites; the presence of vesicles of a specific kind ( e . g . dark , 50-nm diameter neuropeptide vesicles , or clear large unevenly shaped vesicles , or small , packed clear-core cholinergic vesicles , and others ) ; or other distinctive characteristics such as the presence of microtubules on a specific side of the neurite , or membrane-associated structures , or distinctive cytoplasm texture , such as relative darkness compared to neighboring neurites . Globally , the properties of a neuronal arbor help to identify continuations across gaps . For example , an axonal neurite tends to continue being axonal in nature within the gap-sized span of a few hundred nanometers; same for dendrites . An obvious feature is that differentiated neurons present a single soma; continuations that lead to a second soma are therefore most likely incorrect . As described above , to identify a neuron quickly in the larva , the first few minutes are best spent skeletonizing the largest structures on the backbone and tracing them to the soma . This minimal representation generally suffices to identify the neuronal lineage and the overall span of the arbor . When the correct neuron has been found , reconstructed in full and reviewed , we begin to map its synaptic partners . To find out the strongly connected partners of a neuron , we use the connectivity table that aggregates all synaptic relationships , whether with fully reconstructed neurons or single-node skeletons used as placeholders to indicate synaptic partners . Starting at each single-node skeleton , we reconstruct the arbor all the way to the soma by choosing , at every branch point , the larger caliber ( may require jumping back to the last branch node occasionally ) , momentarily ignoring the rest of the arbor . This partial reconstruction suffices to obtain a minimum of information about the partner arbor , such as the lineage . Partner neurons that receive more than one synapse from the neuron of interest will quickly accumulate further fractions of their arbors . These preferred partners—those with many synapses with the arbor of interest—can then be selected for full-arbor reconstruction , while the completion of single-synapse partners ( of which a neuron has many , and which in the Drosophila larva may play a lesser role in understanding the circuit role of a neuron ) can be postponed . Interactive , partial wiring diagrams calculated on demand during neuron reconstruction guide circuit mapping and the identification of errors . Connectivity-dependent coloring schemes highlight desired features of the circuit , sorting neurons into groups . A common case is the inspection one or more orders of synaptic neighborhoods . Given one or more neurons of interest ( such as RP2 ) , we load all synaptic partners into the graph . For small circuits , visual comparisons between the neighborhoods of left and right homologs can identify similar neurons ( e . g . by coloring by stereotyped properties such as the ratio of inputs and outputs , or by their graph centrality; see below ) and highlights missing or differently connected neurons , prompting focused review . Coloring the circuit graph relative to a central neuron highlights the relative synaptic order of all other neurons . Given two neurons , an important circuit question is if there are any paths between them and , if so , through what neurons . This can be queried and added to the graph from within the graph widget , with filters for how many synapses an edge must be . Other coloring modes include by betweenness centrality ( Brandes , 2001 ) of the wiring diagram ( calculated as a directed graph ) , which stresses the role of a neuron within a circuit; and by the percentage of review of the neuronal arbor , indicating at a glance the approximate level of completeness within a group . When reconstructing neuronal arbors with skeletons , the nodes of the skeletons are annotated with a confidence value signifying the degree of certainty in the continuation of the axon or dendrite . We carry on this confidence into the dynamic wiring diagram representation by splitting the skeleton that models a neuronal arbor at the low-confidence edges , resulting in independent graph nodes . The connectivity of these fragments aids in evaluating their impact on the wiring diagram and their potential correctness . Neural circuits targeted for reconstruction must be imaged in volumes large enough to encapsulate the complete neuronal arbors of interest ( Helmstaedter , 2013 ) . Finding specific neurons in unreconstructed data demands prior knowledge , for example using image volumes of genetically labeled neurons , and reference markers like neuroglian or fasciclin II tracts ( Landgraf et al . , 2003 ) that have anatomical correlates that are conspicuous in EM . The ability to navigate vast volumes at intermediate or low resolution aids in identifying large features such as nuclei , nerves , trachea , or neuropil boundaries , helpful for crossing the resolution gap between light-level microscopy and EM . Although not used in the project described here , to further facilitate finding specific neurons of interest in vast EM volumes , CATMAID supports overlaying other volumetric image data , such as registered confocal stacks . Given a good guess of the approximate location , finding a neuron of interest involves reconstructing partial backbones ( the low-order , microtubule-rich processes ) . This typically consumes only 10–20 minutes per arbor , due to the large caliber , presence of continuous microtubules , and the paucity of synapses on backbone . In larval Drosophila , even partially reconstructed backbones suffice to identify individual neurons by comparing with high resolution single neuron images at light level , given that the lowest-level branches are typically sufficient for unique identifiability of individual neurons . In our experience , the best starting points are stereotyped features like the main branch points , tracts , commissural crossings , or the neuropil entry point of the primary neurite bundle of all sibling neurons of the same lineage ( Cardona et al . , 2010 ) . Unfinished backbones remain for other contributors to expand or merge into full arbors later , if desired . In an environment where multiple contributors simultaneously reconstruct neuronal arbors , eventually an ongoing reconstruction reaches that of another contributor . Attempted edits are resolved according to predefined permission rules for who can edit whose work . These rules are implemented as permissions granted to a contributor to alter another contributor’s work . The status of “superuser” enables a trusted expert neuroanatomist to edit at will . Our system operates at two levels: locked and unlocked skeletons . Skeletons that are deemed complete are locked by the contributor , and by default cannot be edited by others unless they have been granted permission to do so . Unlocked skeletons , such as partial reconstructions produced when searching for a specific neuron or when pruning away incorrect branches , can be merged or split by others at will . Neurons are unlocked by default and locking is only to be used upon completion , which prevents sudden and unexpected changes in established results . Individual skeleton nodes and their relations to connectors ( which express synapses ) can only be edited by the original author , or by others that have been explicitly granted permission to edit the contributions of the original author . In case of conflict or insufficient permissions , a notification system delivers the request to the contributor who can review and effect the change . The result of the work of multiple contributors can be visualized in the 3d viewer , with each node of the skeleton colored according to the identity of the contributor . Collaborative reconstructions require that contributors be able to trust the work of others . It is therefore important for a project manager to be able to track the work of each contributor . To estimate an individual’s speed and quality , we consider only contributions that have been reviewed by others , within a specific time period . We quantify the number of edits performed by the reviewer , in particular splits ( cutting away an incorrect branch ) , merges ( appending a missing branch ) and the addition or removal of synapses . While speed and quality are independent , we typically see that better contributors are also faster . After an initial period , lasting anywhere from a couple of days to about 2 weeks of continuous work , a contributor typically becomes acquainted with the reconstruction task and stops adding erroneous synapses or merging branches from different neurons into one . Remaining errors are typically missing branches or synapses , which are far easier to resolve and have a less significant impact on interpretation of the wiring diagram . We observe that different areas of the nervous system exhibit profound differences in arbor and synapse morphology , from extensively branching trees in some ventral nerve cord neurons to cloistered self-contacting axons like A02l or in the olfactory lobes ( data not shown ) . Subjectively , contributors that reconstructed neurons in diverse areas of the nervous system experienced a larger variety of shapes and morphologies , which correlated with the acquisition of greater skill . With many expert contributors come many points of view on how to describe neurons . Instead of enforcing a specific ontology , we allow the annotation of any neuron with arbitrary text snippets . These annotations can express a variety of potentially overlapping concepts , from body regions to cell types , gene expression patterns , genetic driver lines and neurotransmitter profiles , among others . The flexibility afforded by the annotation system supports uses from long-term , contributor-centric publication-ready naming schemes to single-use lists helping personal data organization or team collaboration . Our tools allow queries for one or combinations of annotations , as well as metadata such as time or user associated with an annotation . To make annotations discoverable , we construct a hierarchical tree structure that starts off with three entries: the list of all annotations , the list of all neurons , and the list of all contributors , with each paginated list reducible by regular expression matching . For each annotation , we display five lists: neurons annotated with it , annotations annotated with it ( which act as meta-annotations ) , annotations that it annotates ( acting itself as a meta-annotation ) , the list of contributors that have used it to annotate an entity ( a neuron or an annotation ) , and the list of co-annotations ( other annotations onto the neurons that it annotates ) . Each annotation , neuron and contributor is expandable , letting the user navigate a graph of relations . For co-annotations , further expansions constrain the listing of neurons to those that share all chosen annotations . For example , starting at annotation ‘segment A3’ , continuing with the co-annotation ‘left’ , and then the co-annotation ‘GABA’ , leads to the listing of all GABAergic neurons on the left hemisegment of abdominal segment A3 . Similarly , starting from ‘GABA’ could lead to ‘A3’ and ‘left’ as well , resulting in the same list of neurons . This approach enables the co-existence of many contributor-centric representations of the same neuronal circuits . Annotations also enable the co-existence of multiple nomenclatures for naming neurons . These could be for example by GAL4 line , by developmental grouping ( a name composed of region , segment , lineage and birth order ) , or by gene expression . In CATMAID , many widgets lists neurons by name . These displayed names are customizable , so that each contributor can see his or her own names , even if the neurons in question were created by others . Each contributor chooses a setting for neuron display names among multiple possibilities , including skeleton IDs , own annotations , all annotations , or most usefully , annotations that are themselves annotated with , for example , 'Janelia GAL4 line' or 'Developmental nomenclature' to indicate naming schemes . In order to associate synaptic connectivity not to whole neurons , but to regions of neurons , we adopt an approach where we cluster nearby synapses . Mean shift clustering has been shown to be an effective approach to finding synapse clusters in 3d without assuming a particular number of groups a priori ( Binzegger et al . , 2007 ) . This approach involves convolving synapse locations with a Gaussian kernel to estimate the density of synapses in space . A cluster is then the set of synapses for which , starting at their location , gradient ascent reaches the same density peak . However , locations on one neuron that are close in space can be very far apart along the neuron . Here , instead of considering the density of a neuron’s synapses in 3d space , we use a similar procedure to estimate the density of synapses at every point on the arbor ( following the cable ) and define synapse clusters in the same manner . The only parameter in both approaches is the width of the Gaussian kernel , a physically meaningful parameter . For these purposes , we define the skeletonization of a neuron to be a graph with a set of nodes N with locations Xi for i ∈ N and skeleton edges ε ( note that a ‘skeleton edge’ is between nodes in the skeleton of a single neuron and does relate to synapses ) . Because the neuron’s graph is tree-like , there is a unique non-overlapping path on the graph between any two points i , j∈N with distance δij . All synapses ( both inputs and outputs ) associated with the neuron are represented by the set of their associated nodes , S ⊂ N , noting that the same node can be associated with multiple synapses and thus appear multiple times in S . For every node in the neuron graph , we compute the synapse densityd ( i ) =∑j∈Sexp ( −δij22λ2 ) where λ is a bandwidth parameter that effectively determines the size of clusters , and presynaptic sites of polyadic synapses are counted as many times as they have postsynaptic partners . Note that due to branches , a single synapse close to a branch point may contribute more total density than one that is very distant , reflecting its greater within-graph proximity to more of the arbor . We then look for all maxima in the synapse density and the basins of attraction that flow to them via gradient ascent ( i . e . starting at a given node , moving along the maximally positive difference in density between adjacent nodes ) . A cluster of synapses is then all synapses associated with nodes found within a single basin of attraction of the density function . For neurons found in the 1st instar larva , with ≈500–2000 µm of cable , bandwidths around 8–30 µm provide clusterings that match the subjective description of either 'dendritic arbor' or 'axon' . Smaller bandwidth values result in more granular breakdowns of dendritic and axonal trees ( e . g . dbd axons in Figure 8F ) . We define synapse flow centrality ( SFC ) as the number of possible paths between input synapses and output synapses at each point in the arbor . We compute the flow centrality in three flavors: ( 1 ) centrifugal , which counts paths from proximal inputs to distal outputs; ( 2 ) centripetal , which counts paths from distal inputs to proximal outputs; and ( 3 ) the sum of both . We use flow centrality for four purposes . First , to split an arbor into axon and dendrite at the maximum centrifugal SFC , which is a preliminary step for computing the segregation index , for expressing all kinds of connectivity edges ( e . g . axo-axonic , dendro-dendritic ) in the wiring diagram , or for rendering the arbor in 3d with differently colored regions . Second , to quantitatively estimate the cable distance between the axon terminals and dendritic arbor by measuring the amount of cable with the maximum centrifugal SFC value . Third , to measure the cable length of the main dendritic shafts using centripetal SFC , which applies only to insect neurons with at least one output synapse in their dendritic arbor . And fourth , to weigh the color of each skeleton node in a 3d view , providing a characteristic signature of the arbor that enables subjective evaluation of its identity . A textbook neuron has a purely dendritic arbor and a purely axonal arbor , that is , one neuronal compartment that only receives inputs and another that only delivers outputs onto other neurons . In reality , dendro-dendritic and axo-axonic synapses are present in both invertebrates ( Wilson and Mainen , 2006; Olsen and Wilson , 2008 ) and vertebrates ( Rudomin and Schmidt , 1999; Wilson and Mainen , 2006; Pinault et al . , 1997 ) . We have observed that homologous neurons ( e . g . identifiable neurons in the left and right hemisegments ) have a similar synaptic distribution , which differs from that of other neurons . In Drosophila , we find neurons with highly separated input and output ( e . g . motor neurons and many types of projection neurons ) , neurons with entirely intermingled inputs and outputs ( possibly non-spiking interneurons [Burrows , 1992] ) , and everything in between . Having clustered synapses into groups ( either by synapse clustering or by splitting the neuron by centrifugal synapse flow centrality ) , we ask how neuronal inputs and outputs are distributed among the clusters . If the clustering can adequately separate axon from dendrite , then a highly polar neuron will have most of its outputs on the 'axon' cluster and most of its inputs on the 'dendrite' cluster . Motor neurons in the abdominal segments of the Drosophila larva are examples of completely polarized neurons . Conversely , highly non-polar neurons can have inputs and outputs distributed homogeneously throughout their arbor . An example of the latter are non-spiking neurons that perform extremely local computations , such as GABAergic antennal lobe interneurons ( Wilson and Laurent , 2005 ) . A measure of synaptic sign distribution in a neuronal arbor has the potential to distinguish similar yet uniquely different neurons , as well as to suggest broad functional roles of the neuron . Here , we describe an algorithm to quantify the degree of segregation between input and outputs in a neuronal arbor . For each synapse cluster i on a neuron with Ni synapses , compute the fraction pi that are postsynaptic . We measure the uniformity of the distribution of inputs and outputs within cluster i by computing its entropy , for which we consider synapses as entities with two possible states: input or output . At the limits , when all synapses of the cluster are either inputs or outputs , its entropy is zero . When half of the synapses are inputs and the other half are outputs , the entropy is maximal . The contribution of each cluster i to the total entropy is weighted by its fraction of the total synapses ( either inputs or outputs ) . The entropy of the input/output distribution for each cluster is thenSi=− ( pilogpi+ ( 1−pi ) log ( 1−pi ) ) . The total entropy for the arbor is then justS=1∑iNi∑iNiSi . However , for reference we need to compare this to an unstructured arbor ( i . e . non-clustered ) with the same total number of inputs and outputs; for this , we consider the whole arbor as one cluster , and we computeSnorm=∑ipiNi∑Nilog ( ∑ipiNi∑Ni ) +1−∑ipiNi∑Nilog ( 1−∑ipiNi∑Ni ) ( where ∑ipiNi∑Ni is just the total fraction of synapses that are inputs ) . We define the segregation index asH=1−SSnorm so that H = 0 corresponds to a totally unsegregated neuron , while H = 1 corresponds to a totally segregated neuron . Note that even a modest amount of mixture ( e . g . axo-axonic inputs ) corresponds to values near H = 0 . 5–0 . 6 ( Figure 7—figure supplement 1 ) . We consider an unsegregated neuron ( H ¡ 0 . 05 ) to be purely dendritic due to their anatomical similarity with the dendritic domains of those segregated neurons that have dendritic outputs . We validated our iterative reconstruction method , where users’ actions are not independent of one another , by comparing our results to that of an established consensus methods involving multiple independent reconstructions . Six neurons ( three hemilateral pairs with a total of 2387 µm of cable in the iterative method ) were chosen for validation on the criteria that their morphology was well-contained within the EM volume and the entire group formed a connected network . Four contributors ( authors CMSM and SG , as well as Ingrid Andrade and Javier Valdés-Alemán ) who had little to no prior involvement with the selected neurons were given six seed nodes at the soma of each selected neuron . The number of contributors was chosen based on available , trained users at the time . Each contributor reconstructed the six neurons ( skeletons plus synapses ) in an otherwise completely unannotated volume that only he or she was working in and then did skeleton-centered review of their own neurons . To determine a consensus skeleton from these four reconstructions , we re-implemented the RESCOP algorithm Helmstaedter et al . ( 2011 ) in MATLAB ( Mathworks , Inc ) with slight variations due to differences in the details of skeleton-annotation tools . CATMAID skeletons were resampled so that adjacent nodes were no further than 80 nm apart . Nodes in independent reconstructions were considered consistent if they were within 600 nm of one another , a value chosen because smaller values resulted in correct reconstructions of low-order branch points to be inconsistent . We developed a minimal method to estimate the consensus connectivity because existing consensus skeleton methods are purely morphological . Any point on the consensus skeleton consists of chunks of one or more skeletons from the individual contributors . We opted to sum all synapses in the consensus skeleton , but to weight each so that if every user annotated the same synapse it would have a total weight of 1 . For example , if three contributors reconstructed a given dendritic branch in the consensus skeleton , but only two annotated a postsynaptic site associated with a specific active zone , the consensus synapse would have weight 2/3 . To estimate the probability of completely missing an edge as a function of the number of synapses in the edge , we combine the twig distribution with the error rates obtained from multi-user reconstruction . We found that our reconstruction identified 672 out of 761 twigs , giving our method a recall rate for complete twigs of q = 0 . 88 . Let the distribution of nb twigs across edges with m synapses be p ( nb; m ) . Assuming each branch has a probability q of being correctly observed , the probability of not observing a specific connection across all nb twigs is ( 1−q ) nb . The probability of omitting an edge with m synapses is thus given byPlossm= . ∑nb=1mp ( nb;m ) ( 1−q ) nb In our reconstruction method , we emphasize connections that are found consistently between cells of the same type , typically hemisegmental homologs of a presynaptic and postsynaptic neuron . Using a simple model , we approximate the likelihood of introducing a symmetric , but false , edge between cell types in our wiring diagram due to reconstruction mistakes . Consider two neurons , j = 1 , 2 , of the same cell type , with the dendrites of each sufficiently close to N axons to physically permit connections . To add an incorrect edge to the connectivity graph and not just reweight an existing one , any added branches must have synapses from otherwise unconnected neurons . Let the number of axons with zero true connectivity be N0 . Assuming symmetry , the number of axons for both neurons should be similar . We then suppose that errors add m synapses to each neuron , with each synapse assigned uniformly at random to axon i G ( 1 , 2 , . . , N ) , with the final added edge count onto neuro n j from axon i given by ki , j . For clarity , we order the axons such that i ≤ N0 designates an axon with no true connectivity . We then ask what is the probability that both ki , 1 , ki , 2 > kθ for any i ≤ N0 . The parameters of this model will vary depending on the properties of the neuron and neuropil in question . For example , larger neurons will have more opportunities for error than smaller ones , while neurons with more stringent synaptic specificity have more true zero edges than broadly synaps-ing neurons . To estimate a realistic set of values for the neurons here , we consider the validation data . Because nearly all false positives occur on the terminal arbors , the number of synapses added by error m can be expressed as m = rLtk¯ , the product of the rate of incorrect branches per length of twig r , the total length of twigs Lt , and the expected synapses per added twig k¯ . Based on the independent reconstructions , we estimate r as 6 false-positive errors per 1 . 63 × 103 µm , k¯ = 5 synapses , and a typical Lt = 257 µm , making m = 5 . Determining N and N0 is difficult , as it requires knowledge of axons that would not be in the connectivity-driven reconstruction . We estimate reasonable values using the aCC and RP2 network , since aCC dendrites strongly overlap RP2 dendrites , but have several presynaptic neurons not shared with RP2 ( Figure 11B ) . In addition to the axons presynaptic to RP2 , we find a mean of N0 = 36 inputs exclusive to aCC , so we estimate N = 87 . We simulated the 106 iterations of the model for kθ = 1-4 . To investigate more extreme errors than the ones measured , we also considered m = 37 synapses , the largest twin twig observed across all neurons looked at , and m = 20 synapses , a more typical value for the largest twig of a single neuron . Skeletons are chimeras , with multiple contributors creating various parts at different points in time . We estimate the total amount of time spent skeletonizing an arbor—including its synapses—by counting the number of 20-second intervals that contain at least one skeleton node or connector node related to the skeleton . This approach is robust to the discontinuity in time and authorship of adjacent skeleton nodes , but tends to overestimate slightly reconstruction time , given the contribution of 20-second intervals for single nodes that were created earlier in time as pre- or postsynaptic placeholder skeletons with a single node , and which were subsequently merged into the growing skeleton . If the latter were each counted as contributing 6 seconds only , reconstruction times per skeleton typically shrink between 15 and 25% . We estimate the time for the systematic review of a neuron similarly , with the added caveat that parts of the same arbor may have been reviewed more than once . We count the number of minutes for which at least one skeleton node was reviewed , for every contributor that reviewed a fraction of the arbor , and then add up all the minutes of each contributor . The data volume used was described in Ohyama et al . ( 2015 ) . It is comprised of 462 sections , each 45 nm thick and imaged at 4x4 nm per pixel resolution . It is bounded anteriorly approximately at the intersegmental nerve entry point in segment A2 and posteriorly near the segmental nerve entry of segment A3 . Sections were cut approximately 8° angle relative to true transverse , resulting in a slightly skewed volume with the left side posterior to the right . Using their characteristic morphology , we identified and reconstructed motor neurons U1 , U2 , the three VUM motor neurons , aCC , RP5 and RP2 and sensory neurons dbd , dmd1 , ddaD , ddaE , and vbd for segment A3 . Because dbd , ddaD and ddaE axon terminals also project into anterior and posterior segments , we used segmental repetition to identify the projections of these neurons from adjacent segments that participate in the local circuitry of A3 . We chose to focus on dbd , aCC , and RP2 and continued to reconstruct all arbors synaptically connected to the pair of sensory axons and two pairs of motor neuron dendrites for these cells in A3 . We found 425 arbors spanning 51 . 8 millimeters of cable , with a total of 24 , 068 presynaptic and 50 , 927 postsynaptic relations . Nine people contributed data for the finished product: Albert Cardona ( 200 , 773/641 , 740 nodes ) , Casey Schneider-Mizell ( 171 , 718/641 , 740 nodes ) , Julie Tran ( 64 , 362/641 , 740 nodes ) , Stephan Gerhard ( 34 , 837/641 , 740 nodes ) , John Patton ( 10 , 385/641 , 740 nodes ) , Ingrid Andrade ( 8 , 667/641 , 740 nodes ) , Chris Doe ( 1 , 505/641 , 740 nodes ) , Mark Longair ( 1022/641 , 740 nodes ) , and Akira Fushiki ( 675/641 , 740 nodes ) . Some reconstructions ( 147 , 554/641 , 740 nodes ) were imported into CATMAID from prior work in the same volume in TrakEM2 by Albert Cardona , Casey Schneider-Mizell , Mark Longair , Alexander Berthold van der Bourg , and Kenny Floria . All reconstructions were reviewed in CATMAID . Each arbor was named and described as an identifiable neuron ( 198 arbors ) , an ascending or descending projection that spans the full anteroposterior dimension of the imaged volume ( 107 arbors ) , a neuron spilling over from adjacent segments beyond the imaged volume ( 84 arbors ) , or an unresolvable fragment ( 36 arbors ) ( see Figure 3—figure supplement 1 , Figure 3—figure supplement 2 ) . The 198 identifiable neurons amount to 83% of all cable , 88% of all inputs and 62% of all outputs , with ascending or descending projections contributing 29% of all outputs . The anatomy and connectivity of all arbors can be found in the Supplemental Data . CNS of Drosophila larva were dissected in PBG ( 10% NGS [Normal Goat Serum] in 1% PBS ) with tweezers under a scope and fixed with 4% paraformaldehyde in 1% PBS for 30 min , washed 3×10 min in PBT ( 1% Triton-X100 in 1% PBS ) , blocked for 1 hr in PBG , then washed 3×10 min in PBGT ( 1% Triton-X100 in PBG ) , and incubated with primary antibodies ( rabbit anti-GABA: Sigma A2053 at 1/1000; chick anti-GFP: Abcam ab13970 at 1/2000 ) in PBGT for 24 hr at 4°C on small Eppendorf tubes laid on a gentle horizontal shaker . They were then washed 4×15 min in PBT , and incubated with secondary antibodies ( goat anti-chick 488: Invitrogen , at 1/500; goat anti-rabbit 568: Invitrogen , at 1/500 ) in PBGT at 4°C in Eppendorf tubes wrapped in aluminum foil on a horizontal shaker for 24 hr , subsequently washed in PBT 4×15 min ( wrapped in foil ) , and mounted on poly-lysine coated glass slides . Then samples were dehy drated by dipping the slides in an alcohol series ( 30% , 50% , 70% , 90% in distilled water , then twice 100% ) and then in 100% xylene ( 3 times ) using Columbia glass jars with slits for slides; then mounted on glass slides in DePeX ( Li et al . , 2014 ) using spacer coverslips on the sides . Glass slides were left to dry in a large Petri dish with a lid , wrapped in foil and at 4°C for 3 days . Imaging was done with a Zeiss 710 confocal laser-scanning microscope . Positive immunoreactivity was confirmed by consistent labeling across multiple GFP-labeled cells per imaged nervous system in two or more nervous systems . | The nervous system contains cells called neurons , which connect to each other to form circuits that send and process information . Each neuron receives and transmits signals to other neurons via very small junctions called synapses . Neurons are shaped a bit like trees , and most input synapses are located in the tiniest branches . Understanding the architecture of a neuron’s branches is important to understand the role that a particular neuron plays in processing information . Therefore , neuroscientists strive to reconstruct the architecture of these branches and how they connect to one another using imaging techniques . One imaging technique known as serial electron microscopy generates highly detailed images of neural circuits . However , reconstructing neural circuits from such images is notoriously time consuming and error prone . These errors could result in the reconstructed circuit being very different than the real-life circuit . For example , an error that leads to missing out a large branch could result in researchers failing to notice many important connections in the circuit . On the other hand , some errors may not matter much because the neurons share other synapses that are included in the reconstruction . To understand what effect errors have on the reconstructed circuits , neuroscientists need to have a more detailed understanding of the relationship between the shape of a neuron , its synaptic connections to other neurons , and where errors commonly occur . Here , Schneider-Mizell , Gerhard et al . study this relationship in detail and then devise a faster reconstruction method that uses the shape and other properties of neurons without sacrificing accuracy . The method includes a way to include data from the shape of neurons in the circuit wiring diagrams , revealing circuit patterns that would otherwise go unnoticed . The experiments use serial electron microscopy images of neurons from fruit flies and show that , from the tiniest larva to the adult fly , neurons form synapses with each other in a similar way . Most errors in the reconstruction only affect the tips of the smallest branches , which generally only host a single synapse . Such omissions do not have a big effect on the reconstructed circuit because strongly connected neurons make multiple synapses onto each other . Schneider-Mizell , Gerhard et al . 's approach will help researchers to reconstruct neural circuits and analyze them more effectively than was possible before . The algorithms and tools developed in this study are available in an open source software package so that they can be used by other researchers in the future . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Quantitative neuroanatomy for connectomics in Drosophila |
Aberrant alternative pre-mRNA splicing ( AS ) events have been associated with several disorders . However , it is unclear whether deregulated AS directly contributes to disease . Here , we reveal a critical role of the AS regulator epithelial splicing regulator protein 1 ( ESRP1 ) for intestinal homeostasis and pathogenesis . In mice , reduced ESRP1 function leads to impaired intestinal barrier integrity , increased susceptibility to colitis and altered colorectal cancer ( CRC ) development . Mechanistically , these defects are produced in part by modified expression of ESRP1-specific Gpr137 isoforms differently activating the Wnt pathway . In humans , ESRP1 is downregulated in inflamed biopsies from inflammatory bowel disease patients . ESRP1 loss is an adverse prognostic factor in CRC . Furthermore , generation of ESRP1-dependent GPR137 isoforms is altered in CRC and expression of a specific GPR137 isoform predicts CRC patient survival . These findings indicate a central role of ESRP1-regulated AS for intestinal barrier integrity . Alterations in ESRP1 function or expression contribute to intestinal pathology .
The single-layered intestinal epithelium provides an important physical barrier that critically contributes to intestinal homeostasis ( Peterson and Artis , 2014 ) . Dysfunction of intestinal epithelial cells ( IECs ) leading to increased epithelial permeability is associated with intestinal diseases such as inflammatory bowel disease ( IBD ) and colorectal cancer ( CRC ) ( Van der Sluis et al . , 2006; Schmitz et al . , 1999; Grivennikov et al . , 2012 ) . IBD is related to polymorphisms in various IBD susceptibility genes ( Lees et al . , 2011; Van Limbergen et al . , 2014 ) , and numerous genetic alterations in key cellular pathways that underlie CRC have been identified ( Fearon , 2011 ) . However , the role of post-transcriptional modifications in the regulation of IBD and CRC development is still poorly understood . Alternative splicing of pre-mRNAs ( AS ) is a common posttranscriptional modification that is estimated to occur in 92–94% of human genes ( Wang et al . , 2008; Pan et al . , 2008 ) . AS permits generation of protein isoforms with related , distinct or sometimes even opposing functions ( Vorlová et al . , 2011 ) . Moreover , certain isoforms influence cancer progression ( Brown et al . , 2011 ) or are associated with autoimmune and inflammatory disorders ( Ueda et al . , 2003; Laitinen et al . , 2004 ) . While alterations of mRNA splicing have been reported in human colorectal cancer ( CRC ) ( Freund et al . , 2015; Zhou et al . , 2014 ) and IBD ( Häsler et al . , 2011; Mailer et al . , 2015 ) , the consequences of deregulated AS for intestinal homeostasis and disease development are largely unexplored . Epithelial splicing regulatory protein 1 ( ESRP1 ) , which is exclusively expressed in epithelial cells , was identified in a cDNA expression screen for factors that promote the epithelial pattern of FGFR2 ( Fibroblast Growth Factor Receptor 2 ) splicing . ESRP1 was also found to regulate the AS of CD44 and other genes ( Warzecha et al . , 2009 ) . ESRP1 is negatively regulated by mesenchymal transcription factors such as SNAIL , ZEB1 and ZEB2 ( Preca et al . , 2015; Saitoh , 2015; Reinke et al . , 2012 ) . Recently , ESRP1 has been reported to act as a tumor suppressor by negatively regulating epithelial-to-mesenchymal transition ( EMT ) and the metastatic potential of human breast cancer cell lines , via the splicing of different isoforms of CD44 or EXO70 ( Lu et al . , 2013 ) . ESRP1 can suppress cancer cell motility in head and neck carcinoma cell lines ( Ishii et al . , 2014 ) and ESRP1 protein expression is a favorable prognostic factor in pancreatic cancer ( Ueda et al . , 2014 ) . However , ESRP1 may also promote lung metastasis of orthotopically transplanted breast cancer cells by generating CD44 isoforms independently of EMT ( Yae et al . , 2012 ) . Yet , the contribution of ESRP1 to intestinal integrity and function is poorly investigated . Importantly , the lack of viable animal models so far has precluded analyzing the role of ESRP1-mediated AS for intestinal disease in vivo . Here , we used a novel mutant allele of Esrp1 called Triaka to investigate the function of ESRP1 in the intestine . We found that Esrp1Triaka ( later referred to as Triaka ) leads to reduced ESRP1 function causing distinct alterations in the mRNA splicing pattern in colonic IECs ( cIECs ) of Triaka compared with wild-type ( WT ) animals . These changes in several transcript isoforms do not alter intestinal histomorphology , yet they are associated with increased intestinal permeability in Triaka mice . In addition , Triaka mice show alterations in distinct models of intestinal disease . Mechanistically , this phenotype can be ascribed , in part , to changes in the relative frequency of specific Gpr137 splicing isoforms in Triaka cIECs . This affects the survival and function of cIEC by altering the Wnt signaling pathway in Triaka mice . In humans , ESRP1 transcript levels are downregulated in inflamed compared with non-inflamed biopsies from IBD patients . Furthermore , ESRP1 expression is gradually lost during the adenoma to carcinoma sequence in CRC , and loss of ESRP1 protein expression in CRC tumors negatively correlates with patient survival . Moreover , the ratio of specific GPR137 isoforms is different in tumor versus normal intestinal tissue , and expression of a specific ESRP1-dependent GPR137 isoform predicts CRC patient survival . Together , these data indicate an important role for ESRP1 in intestinal disease in humans and mice .
To study the role of ESRP1 in the intestinal epithelium , we used a novel mutant allele of Esrp1 called Triaka that was identified in an N-ethyl-N-nitrosourea genetic screen . The Triaka point mutation results in a methionine-to-valine substitution at the amino acid position 161 ( M161V ) of ESRP1 , a residue conserved in several species ( Figure 1—figure supplement 1 ) . Triaka animals develop overtly normal ( see also Materials and methods ) , which allows investigation of the physiological role of Esrp1 in adults . This is in contrast to the recently published Esrp1-/- or skin epithelial-specific Esrp1-deficient animals that are neonatal lethal ( Bebee et al . , 2015 ) . We first utilized a previously described in vitro reporter system to characterize the effect of Esrp1Triaka on the splicing of known ESRP1-regulated transcripts ( Brown et al . , 2011 ) . Using this system based on luciferase expression , we found that WT ESRP1 protein led to a 2 . 4 and 10 . 9 fold increase of Cd44 variable exon v5 ( Cd44v5 ) and Fgfr2 variable exon IIIb ( Fgfr2-IIIb ) inclusion , respectively , compared to control . This was in line with earlier reports ( Brown et al . , 2011; Warzecha et al . , 2009 ) . ESRP1Triaka however showed reduced levels of Cd44v5 and Fgfr2-IIIb inclusion , with 1 . 8 and 3 . 3 fold induction , respectively ( Figure 1A and B ) . This variation in the extent of in vitro splicing of Cd44 versus Fgfr2 by ESRP1Triaka likely related to the fact that Esrp1-regulated splicing events show distinct sensitivity to Esrp1 loss ( Warzecha et al . , 2009; Bebee et al . , 2015 ) . We then transduced CMT-93 cells with inducible lentiviral encoding Esrp1WT or Esrp1Triaka to assess the effect of the mutation on epithelial cell function . At the same level of overexpression as Esrp1WT , Esrp1Triaka led to diminished inclusion of Cd44v4/5 in CMT-93 cells and reduced cell proliferation ( Figure 1C–E ) . Of note , Esrp1WT transcript levels directly correlated with frequency of Cd44v4/5 splicing events , suggesting a dose-dependent effect of Esrp1 expression on Esrp1-mediated splicing activity ( Figure 1F ) . Taken together , these in vitro data indicate that Esrp1Triaka reduces ESRP1 function , and thereby decreases the proliferative capacity of epithelial cells . Esrp1 has been shown to be epithelial cell-restricted and to be highly expressed in the murine large intestine ( Warzecha et al . , 2009 ) . Thus , we next investigated the impact of Esrp1Triaka on the mRNA splicing pattern of cIECs . RNA sequencing was performed on cIECs isolated from naïve mice . Computational analysis revealed 35 genes for which the relative frequency of splicing isoforms differed in Esrp1Triaka versus Esrp1WT cIECs ( Figure 2A and B , Figure 2—source data 1 ) . These findings were validated using quantitative reverse-transcription polymerase chain reaction ( qPCR ) for selected , previously reported ESRP1 target genes , including Cd44 and Magi1 ( Brown et al . , 2011; Warzecha et al . , 2009; Warzecha et al . , 2010 ) , and for novel candidate targets , including Uap1 and Gpr137 ( Figure 2C ) . Functional alterations induced by ESRP1Triaka were also apparent on the protein level , as expression of CD44 variant 4 ( CD44v4 ) -containing isoforms , but not total CD44 , was reduced in colonic sections of Triaka versus WT mice ( Figure 2D and Figure 2—figure supplement 1 ) . Therefore , these results suggested Esrp1Triaka-dependent functional changes in mutant mice . We next compared the gene expression profiling of Triaka versus WT cIECs to assess potential functional effects of Esrp1Triaka . Pathway analysis of the RNA sequencing data showed that several pathways involved in cell cycle and proliferation were affected in Triaka cIECs ( Figure 3—figure supplement 1 and Figure 3—source data 1 ) . Yet , these transcriptional changes did not overtly alter intestinal histomorphology and epithelial proliferation , as crypt depth , number of goblet cells and Ki-67 expression were similar in Triaka versus WT mice ( Figure 3—figure supplement 2A–F ) . Furthermore , Triaka and WT mice showed similar colon length at steady-state ( Figure 3—figure supplement 2G ) . However , E-cadherin ( CDH1 ) surface expression was reduced on Triaka cIECs ( Figure 3A and B ) . Given the central role of E-cadherin for epithelial cell function , we hypothesized a possible defect in intestinal epithelial barrier integrity in Triaka mice . Indeed , ex vivo measurement of the intestinal electrical resistance indicated an increased ion permeability of Triaka versus WT colonic mucosa ( Figure 3C ) , although there was no such difference in the small intestine ( Figure 3—figure supplement 3A ) . This was accompanied by the presence of bacterial 16S rRNA in intestinal crypts and in the inner mucus layer of Triaka but not WT colons ( Figure 3D and Table 1 ) . As further indirect evidence of reduced barrier integrity , we also detected systemic anti-commensal IgG1 and IgG2b antibody reactivity towards autologous intestinal bacteria in most Triaka mice , but rarely in control animals ( Figure 3E and Table 1 ) . However , levels of fecal albumin and lipocalin-2 – markers for intestinal lesions and inflammation – were similar in fecal pellets of Triaka and WT mice ( Figure 3—figure supplement 3B and C ) . Furthermore , there was no difference in the resorption of macromolecules between Triaka and WT mice ( 22 ± 1 versus 24 ± 3 µg/ml serum FITC-dextran-4000 levels , respectively , p=0 . 6 ) . In summary , these data suggest that reduced Esrp1-dependent mRNA splicing in Triaka animals results in decreased integrity of the colonic epithelial barrier and intestinal penetration of bacterial products . These epithelial defects are however not sufficient to induce intestinal immunopathology in naïve Triaka mice . Next , we addressed whether the Esrp1Triaka-dependent alterations observed at steady-state in Triaka mice may affect intestinal disease . Compared with WT controls , Triaka animals treated with dextran sodium sulfate ( DSS ) in drinking water for 7 days , a disease model of intestinal inflammation and damage , showed increased weight loss and pronounced shortening of the colon ( Figure 4A and B ) . This was also reflected by higher disease scores , both clinically and histologically ( Figure 4—figure supplement 1A and B ) . Notably , this phenotype was not due to an altered drinking behavior of Triaka animals ( Figure 4—figure supplement 1C ) . In addition , Triaka mice were also more susceptible to chronic DSS-induced colitis ( Figure 4—figure supplement 1D ) . To assess a possible functional consequence of the differential expression of proliferation-associated genes in Triaka versus WT cIECs , we applied an in vivo wound-healing model and followed the repair of experimentally-induced intestinal lesions in the two groups of mice . In these settings , we observed diminished wound-healing in Triaka compared with WT mice ( Figure 4C ) . In contrast to what we observed at steady-state , these data imply that Esrp1Triaka impairs the proliferative or regenerative capacity of cIECs during intestinal pathology . Next , we tested whether the increased susceptibility to DSS-colitis and the lower repair ability of Triaka mice may influence cell proliferation and pathogenesis in a CRC model . We found that in azoxymethane ( AOM ) and DSS-treated Triaka mice , colorectal tumors were reduced in number and size compared with WT controls , thus providing additional evidence for a defective proliferation of Triaka IECs ( Figure 4D and E ) . The tumor grade was not different between the two groups ( Figure 4F ) . However , molecular analysis revealed more pronounced upregulation of matrix metalloproteinase-3 , granulocyte-colony stimulating factor , transforming growth factor β1 and interleukin-1α in Triaka versus WT tumors ( Figure 4—figure supplement 2 ) . These proteins are all established drivers of CRC progression , EMT and metastasis , thus indicating a more aggressive phenotype of Triaka tumors ( Li et al . , 2016; Mroczko et al . , 2006; Morris et al . , 2014; Calon et al . , 2012; Matsuo et al . , 2009 ) . These features of Esrp1Triaka CRC lesions likely resulted from a partial EMT signature expressed by Triaka cIECs , prior to transformation ( Figure 4—figure supplement 3 ) . To address the hypothesis that the above-observed phenotypes were indeed due to impaired cIECs proliferation in Triaka mice , we last analyzed intestinal Ki-67 expression in the colonic mucosa , after a 3 day treatment with DSS . Under these conditions of mild intestinal inflammation , which are not sufficient to induce epithelial erosions , Triaka cIECs showed reduced Ki-67 expression compared with WT cIECs ( Figure 4G and H ) . Taken together , these results indicate that altered Esrp1-regulated mRNA splicing deregulates cIEC function and affects the development of intestinal disease in Triaka mice , through a mechanism reducing IEC proliferation . We next investigated the molecular mechanisms downstream of Esrp1Triaka that lead to diminished proliferation of Triaka IECs . Among the 35 genes showing splicing isoforms with different relative frequency in Triaka versus WT cIECs , we chose genes for which only two isoforms were differently expressed . Of those , Gpr137 emerged as a prominent candidate since it is involved in IEC proliferation ( Zhang et al . , 2014 ) , although the mode of action of this orphan G protein-coupled receptor ( GPCR ) or of its isoforms is unknown . Thus , the two Gpr137 splicing isoforms with different relative frequency in Triaka versus WT cIECs , Gpr137_ENSMUST00000166115 and Gpr137_ENSMUST00000099776 ( referred hereafter as Gpr137_Long and Gpr137_Short , respectively ) , were selected for further functional studies . By using our in vitro reporter system , we could validate that Gpr137 is a splicing target of ESRP1 ( Figure 5—figure supplement 1 ) . Our RNA sequencing analysis indicated that Gpr137_Long is preferentially expressed in Triaka cIECs , whereas Gpr137_Short is predominant in WT cIECs . To address a potentially distinct role of Gpr137_Long versus Gpr137_Short in cIECs , CMT-93 cells were transduced with lentiviral vectors encoding these two Gpr137 isoforms . We found a decreased cell proliferation and diminished epithelial monolayer tightness , as indicated by lower electrical resistance , in Gpr137_Long- versus Gpr137_Short-expressing cells . Importantly , in this particular assay the barrier function appeared to be distinct from the IEC proliferation . Indeed , although control vector-transduced cells proliferated less than cells transduced with vectors encoding Gpr137 isoforms , they formed a tighter barrier ( Figure 5A and B , Figure 5—figure supplement 2A ) . These effects of the two Gpr137 isoforms on IEC proliferation and barrier integrity were independent of their relative expression levels and of possible cytotoxic effects ( Figure 5—figure supplement 2B and C ) . Transduction with Gpr137 isoforms also distinctively enhanced the proliferation of murine MC-38 and human Caco-2 IEC lines , thus validating our findings in CMT-93 cells ( Figure 5—figure supplement 2D and E ) . To dissect the differential effect of the two Gpr137 isoforms on CMT-93 cells , we then investigated the activation of signaling pathways known to regulate epithelial cell proliferation or barrier function . Among the pathways examined , differences were only observed for Wnt/β-catenin signaling . Indeed , we found a one-third reduction in levels of active β-catenin protein in Gpr137_Long- versus Gpr137_Short-transduced cells ( Figure 5C ) , which was validated in Esrp1Triaka- versus Esrp1WT-transduced cells ( Figure 5—figure supplement 3 ) . Furthermore , pharmacological inhibition of Wnt/β-catenin signaling abrogated the proliferative advantage of Gpr137_Short- over Gpr137_Long-transduced cells ( Figure 5D ) . This suggested that Gpr137 isoforms modify epithelial function via regulation of Wnt/β-catenin signaling . In line with these in vitro data , we measured ca . 40% reduced levels of active β-catenin , as well as diminished expression of Wnt target genes ( Herbst et al . , 2014 ) in cIECs of Triaka compared with WT mice ( Figure 5E and F ) . Collectively , these results show that distinct Gpr137 isoforms can differently modulate the Wnt/β-catenin pathway and IEC function , which may partially underlie the intestinal phenotype of Triaka mice . To address the general relevance of the findings from our in vitro and in vivo studies , we next assessed ESRP1 expression in intestinal biopsies from Crohn’s disease ( CD ) patients , after normalization to an epithelial-specific marker . Compared with non-inflamed paired biopsies , ESRP1 levels were downregulated in the inflamed biopsies from CD intestines ( Figure 6A ) . This correlated with reduced nuclear ESRP1 expression in the inflamed CD intestine , as measured by high-throughput automated immunohistochemistry quantification ( Figure 6B ) . We also performed a tissue microarray ( TMA ) based analysis of matched intestinal tissues from a cohort of 185 CRC patients , which revealed a gradual decrease in nuclear expression of ESRP1 protein during cancer progression ( Table 2 ) . Furthermore , low nuclear ESRP1 expression was associated with reduced patient survival ( p=0 . 0456 ) , larger tumors ( p=0 . 0034 ) , lymphatic invasion ( p=0 . 0466 ) , advanced pT-stage ( p=0 . 02 ) and presence of nodal metastasis ( p=0 . 016 ) ( Figure 6C and Figure 6—source data 1 ) . ESRP1 expression was determined to be an independent prognostic factor , after adjusting for the confounding effects of pT , pN , pM , tumor budding , and lymphatic invasion ( Figure 6—source data 2 ) . This indicates a possible tumor-suppressive role of ESRP1 in CRC . Finally , we analyzed in The Cancer Genome Atlas ( TCGA ) RNA sequencing data of intestinal tissue from CRC patients to evaluate the relative frequency of GPR137 isoforms in the human intestine . We found a positive correlation between the expression of ESRP1 and GPR137_ENST00000539833 ( referred hereafter as hGPR137_Short ) in normal intestine tissue , which we further validated using independent samples ( Figure 6D and Figure 6—figure supplement 1A ) . Moreover , there was an association between hGPR137_Short expression and transcript levels of Wnt target genes , suggesting a similar signaling downstream of GPR137 in humans and mice ( Figure 6—source data 3 ) . In human CRC , the ratio between hGPR137_Short and GPR137_ENST00000377702 ( referred hereafter as hGPR137_Long ) isoforms was altered compared with normal intestinal tissue ( Figure 6E ) . Similarly , transcript levels of Gpr137_Long were increased in murine CRC tissue ( Figure 6—figure supplement 1B ) . Importantly , a higher ratio of hGPR137_Short to hGPR137_Long transcripts predicted enhanced CRC patient survival , thereby indicating a protective role of the ESRP1-dependent hGPR137_Short isoform ( Figure 6F ) . Indeed , higher expression levels of hGPR137_Short alone was also associated with better prognosis ( Figure 6—figure supplement 1C ) . These findings extend our in vivo data in Esrp1Triaka mice and substantiate their relevance for the human intestine . They also suggest an involvement of hGPR137 isoforms in human CRC tumorigenesis and progression . Collectively , our data in mice and humans indicate that downregulation or loss of ESRP1 function results in dysregulation of alternative mRNA splicing in IECs . This leads to impaired epithelial cell integrity and contributes to intestinal pathology , possibly via altered Gpr137/GPR137 isoform ratios ( Figure 6—figure supplement 2 ) .
Here , we describe a mouse model with hypomorphic ESRP1 function that shows impaired intestinal epithelial barrier integrity . This was associated with changes in the relative frequency of distinct mRNA isoforms in IECs and accompanied by an altered epithelial phenotype , as indicated by diminished surface E-cadherin expression . Of the 35 genes with splicing isoforms present at different ratios in Triaka versus WT primary cIECs , we identified two isoforms of Gpr137 that differently affected IEC function via modulation of the Wnt signaling pathway . In support of these data from mouse models demonstrating a role for Esrp1 in intestinal integrity , ESRP1 was downregulated in colonic tissue from CRC patients . Furthermore , the ratio of specific GPR137 isoforms varied in tumor versus normal intestinal human samples , and expression of a specific ESRP1-dependent GPR137 isoform predicted CRC patient survival . Little is known about GPR137 , an orphan GPCR whose knockdown leads to reduced proliferation of several cancer cell lines , including colon cancer cells ( Zhang et al . , 2014 ) . In particular , the function of distinct Gpr137/GPR137 isoforms in IECs has not been addressed so far . GPR137 is ubiquitously and abundantly expressed in mouse tissues and is most closely related to OR51E2/PSGR , a prostate-specific GPCR ( Vanti et al . , 2003; Regard et al . , 2008 ) . Interestingly , recent data suggest that GPR137 may be involved in EMT . Indeed , silencing of GPR137 resulted in a downregulation of SNAI1/SNAIL and SNAI2/SLUG and a corresponding increase in E-cadherin expression in human prostate cancer cells ( Ren et al . , 2016 ) . Gpr137_Short and hGPR137_Short are predicted to encode a protein with five transmembrane domains , whereas the products of Gpr137_Long and hGPR137_Long have a seven-transmembrane-spanning motif . Human and mouse GPR137 proteins have a 78% identity , yet the respective short and long isoforms we report here do not directly match between the two species . The molecular mechanisms by which distinct Gpr137/GPR137 isoforms differently regulate Wnt/β-catenin signaling or possibly determine CRC survival have yet to be investigated . Our data indicate a reduction of active β-catenin levels in primary Esrp1Triaka IECs and a subsequent downregulation of the transcription of some – but not all – Wnt target genes . Active β-catenin was similarly decreased in CMT-93 cells transduced with an Esrp1Triaka- or a Gpr137_Long-expressing lentiviral vector , compared to controls . The Wnt/β-catenin pathway is central to maintain tissue renewal and stem cell activity in IECs as well as epithelial cell proliferation in intestinal crypts . Genetic ablation of β-catenin in adult mice leads to intestinal epithelial stem cell ( IESC ) differentiation , thereby resulting in impaired intestinal homeostasis and fatal loss of intestinal function ( Fevr et al . , 2007 ) . Reduced Wnt/β-catenin signaling likely underlies the diminished healing capacity of Triaka IECs and their reduced proliferation after transformation in the AOM/DSS model . In contrast , the Wnt/β-catenin pathway is aberrantly activated in most CRC , often caused by APC loss-of-function ( Clevers and Nusse , 2012 ) . Therefore , our study suggests a link between ESRP1 and Wnt/β-catenin signaling in the intestine . Besides Gpr137 , we also revealed other genes with altered splicing patterns in Triaka compared with WT cIECs , which likely also contribute to the intestinal phenotype in Triaka mice . Among those ESRP1 target genes , Cd44 is probably the most studied for its intestinal function . CD44 and its isoforms negatively regulate IEC apoptosis ( Zeilstra et al . , 2008; Lakshman et al . , 2004 ) and participate in various cellular functions , such as proliferation , adhesion , and migration ( Ponta et al . , 2003 ) . Furthermore , CD44v4 and CD44v6 isoforms are overexpressed in neoplastic IECs of patients with familial adenomatous polyposis ( Zeilstra et al . , 2014 ) . The expression of Cd44v4 and Cd44v6 isoforms in murine IESCs positively correlates with expression of LGR5 , a marker of adult IESCs , and CD44 variant isoforms promote CRC in ApcMin/+ mice ( Zeilstra et al . , 2014 ) . Therefore , the decreased expression of CD44v4 protein in Esrp1Triaka IECs may likewise explain the reduced intestinal tumorigenesis observed in Triaka compared with WT mice . Although our data indicate reduced splicing activity from the Triaka mutation , a possible neomorphic effect of ESRP1Triaka cannot be fully excluded , and further investigation using other genetic models is required to examine this aspect . ESRP1 has also been reported to restrict pluripotency in embryonic stem cells ( Fagoonee et al . , 2013 ) , and it may as well regulate the differentiation of adult stem cells , including IESCs . Similarly , ESRP1 downregulation promotes cancer cell stemness and metastasis ( Preca et al . , 2015 ) . Accordingly , frameshift mutations of ESRP1 are found in 25% of microsatellite instability-positive CRC tumors ( Ivanov et al . , 2007 ) . Furthermore , our results from a CRC cohort show that ESRP1 protein is gradually downregulated during the adenoma to carcinoma sequence in intestinal tumors and that few cells express it in lymph node metastases . These findings strengthen previous data suggesting a tumor-suppressive role of ESRP1 in CRC cells ( Leontieva and Ionov , 2009 ) . Remarkably , ESRP1 appears to be involved early during CRC tumorigenesis , since it is already downregulated in adenoma . Reduction of ESRP1 function in adenomatous lesions may compromise the intestinal barrier , thereby facilitating the penetration of bacterial products and promoting inflammation-mediated CRC development ( Grivennikov et al . , 2012 ) . Our data in mice suggest that while AOM/DSS-induced Triaka tumors are smaller , they show a more aggressive molecular signature . We do not have a definitive explanation for these seemingly opposing results . However , ESRP1 is a key regulator of EMT – it maintains the epithelial phenotype of a cell – and ESRP1 downregulation leads to a shift towards a mesenchymal phenotype . Indeed , we find higher transcript levels of the EMT-promoting transcription factors Zeb1 and Zeb2 in Triaka compared with WT cIECs . As EMT is associated with reduced proliferation as well as increased motility and invasiveness of malignant cells , this may account for the phenotype of Triaka tumors ( Kalluri and Weinberg , 2009; Evdokimova et al . , 2009; Hur et al . , 2013 ) . Although we investigated ESRP1 expression and not ESRP1 mutations in human CRC , our in vitro data indicate a direct correlation between ESRP1 expression and splicing activity mediated by ESRP1 . Therefore , we believe that Triaka mice , characterized by decreased ESRP1 activity , represent a valid model to address the contribution of Esrp1-dependent mRNA splicing to intestinal function . In summary , our results demonstrate a role for Esrp1 in intestinal homeostasis and disease in mice . Our findings also suggest that loss of ESRP1 may contribute to CRC and IBD development in humans . Moreover , we identified GPR137 as a novel splicing target of ESRP1 and found that distinct isoforms of GPR137 regulate intestinal homeostasis by modulating IEC function through the Wnt/β-catenin pathway . Additional studies investigating the role of GPR137 and ESRP1 in the intestinal epithelium are warranted to understand how distinct GPR137 isoforms impact on CRC patient survival and to reveal strategies to target the ESRP1-GPR137 axis .
A multi-punch TMA of matched tumor , adenoma and normal tissue from 220 CRC well-characterized patients surgically treated from 2004 to 2007 at the Aretaieion University Hospital , University of Athens , Greece was used to evaluate changes in ESRP1 expression during tumor progression and impact on clinicopathological features ( see also Figure 6—source data 1 ) . Additionally , RNAlater ( Thermo Fisher Scientific , Waltham , MA ) -stored and formalin-fixed paraffin-embedded biopsies from IBD patients were obtained from the Swiss IBD Cohort ( http://www . ibdcohort . ch/ ) and from the Institute of Pathology of the University of Bern ( Switzerland ) , respectively . Alternatively , CRC tissues were provided by the Tissue Bank Bern . The use of patient data and samples was approved by the Ethics Committee at the University of Athens , Greece and the Cantonal Ethics Committee of Bern . Nuclear ESRP1 expression in CRC tissue was detected by immunohistochemistry ( IHC ) and scored by a pathologist ( V . H . K . ) . In detail , each spot was evaluated for the percentage of epithelial cells showing ESRP1 expression ( 0–100% ) . Additionally , the intensity of staining was evaluated on a four point scale ( 0‐3 ) . Final scores were formed by multiplication of these two values ( range 0–300 ) and evaluated in correlation with clinicopathological features and survival . For the Kaplan-Meier curves , Receiver Operating Characteristic curve analysis was performed to identify the optimal immunohistochemical cut-off score with lymph node metastasis as the endpoint . Based on this analysis , a cutoff value of 20% ESRP1 expression in tumors was chosen to separate ESRP1-high from -low tumors . Univariate and multivariate survival analyses were carried out using Cox proportional hazards regression . All significant variables in univariate analysis were entered into multivariate analysis . Hazard ratios and 95% CI were used to determine the effect size for the outcome overall survival . Nuclear ESRP1 expression in intestinal epithelial cells ( IECs ) was quantified using Oncotopix Discovery ( RRID:SCR_015690; Visiopharm , Hoersholm , Denmark ) and applying a custom-made algorithm developed by the authors . In brief , inflammation grade was assessed for each biopsy on H&E-stained tissue sections by a board-certified pathologist ( V . G . ) , in a single-blinded manner , and classified into no , mild , moderate and severe inflammation . Sequential sections were then stained for ESRP1 and analyzed . The analysis algorithm was first designed to automatically distinguish IEC areas from stroma and background . Nuclear ESRP1 expression in IECs was then stratified into no , low , moderate and high expression and the area of these different intensity values was quantified for each biopsy . Thereafter , the area-corrected mean brown stain intensity value was calculated for each biopsy . Finally , intensity values were normalized to mildly inflamed biopsies on each slide to control for potential staining variability between slides . IHC was performed on sections ( 2 . 5 μm ) of IBD biopsies and on the TMA using the BondMax system ( Leica , Wetzlar , Germany ) in Bond Epitope Retrieval Solution 2 ( citrate , pH 9 . 0 ) for 60 min at 95°C . Tissues were then stained for ESRP1 ( 1:50 , Thermo Fisher Scientific , Waltham , MA ) . The Bond Polymer Refine Detection kit ( Leica ) was used for the detection of ESRP1-positive cells . For IHC on formalin-fixed paraffin-embedded murine colon tissue , a heat-induced epitope retrieval was first performed for 18 min in a steam cooker in 1 mM Tris/1 mM EDTA ( pH 9 . 0 ) . A primary antibody detecting total CD44 ( 1:100 , rat anti-mouse , Becton Dickinson , Franklin Lakes , NJ ) , CD44v4-containing isoforms ( 1:100 , rat anti-mouse , eBioscience , Santa Clara , CA ) or Ki-67 ( 1:200 , Dako , Santa Clara , CA ) were then used . Specific binding of primary antibodies was visualized using a secondary antibody ( 1:200 , goat anti-rat , Dako ) followed by a tertiary antibody with linked horseradish peroxidase ( HRP ) as the enzyme ( EnVision+ , anti-goat , Dako ) , and 3 , 3'-diaminobenzidine as the chromogen . Staining intensity of positive cells was quantified using an automated scanner and software ( Aperio , Sausalito , CA ) . Bacterial 16S rRNA was detected with the eubacterial probe EUB338 ( 5’-GCTGCCTCCCGTAGGAGT-3’ ) ( Lücker et al . , 2007 ) labeled with Alexa647 ( Eurofins ) . Sections were counterstained with DAPI ( 1:2000 , BioLegend , San Diego , CA ) and placed in mounting medium ( Dako ) . Fluorescence signal was detected using an Olympus IX81 confocal microscope combined with a FluoView FV1000 device ( Olympus , Tokyo , Japan ) . All animal experiments were performed in accordance with institutional and federal regulations governing animal care and use and were approved by The Scripps Research Institute ( TSRI ) Institutional Animal Care and Use Committee ( La Jolla , CA , USA ) ( IACUC protocols 07–0057 and 09–0079 ) and the Cantonal Veterinary Office of Bern ( Switzerland ) ( protocols BE76-11 and BE130/14 ) . Animal experiments were carried out in compliance with the ARRIVE reporting guidelines . All strains used were on a C57BL/6 background . C57BL/6J mice ( RRID:IMSR_JAX:000664 ) were purchased from Jackson Laboratories and thereafter bred in-house . Esrp1Triaka mice ( RRID:MGI:5515349 ) were generated at TSRI using N-ethyl-N-nitrosourea . For all experiments , non-randomized groups of 8–12 week old Esrp1Triaka/Triaka and C57BL/6J animals were either co-housed ( for females ) or soiled bedding was exchanged weekly ( for males ) 3–4 weeks prior to and during experiments . The Esrp1Triaka allele was generated using N-ethyl-N-nitrosourea , as previously described ( Hoebe et al . , 2003 ) . It was initially identified to have a defect in natural killer cell ( NK ) cytotoxicity as well as a hyperactivity and circling behavior . These phenotypes segregated in subsequent breeding and the name Triaka was retained for the neurobehavioral phenotype . Initial confinement of the mutation was made by outcrossing of Esrp1Triaka/Triaka mutation mice to C3H/HeN mice , followed by backcrossing of F1 hybrids to the Esrp1Triaka/Triaka stock . The mutation was mapped to proximal Chromosome 4 , with a peak LOD of 2 . 71 at D4mit235 . Fine mapping further defined a critical region from the centromere to 13 . 7 mega base pairs . Whole genome sequencing of a homozygous Triaka mouse using SOLiDTM ( Thermo Fisher Scientific , Waltham , MA ) revealed an A to G transition at position 828 of Esrp1 , for which 86 . 3% , 74 . 9% , and 62 . 3% of coding/splicing sequence was covered at least 1× , 2× , or 3× , respectively . Validation sequencing of the critical region covered all nucleotides for which discrepancies were observed and confirmed the Esrp1Triaka mutation , which is located in exon 4 of 16 total exons of Esrp1 . The circling behavior , which is incompletely penetrant among homozygous Triaka mice , was investigated in correlation with an auditory or neurodevelopmental defect . However , Triaka mice displayed normal hearing in tests for auditory brain stem response and distortion product otoacoustic emissions . In addition , there were no obvious abnormalities on histological sections of the neocortex , cerebellum , kidney , liver , lung , heart , pancreas , thymus and spleen . Adult ( 8 to 12 week old ) Esrp1Triaka were slightly leaner than WT mice ( 19 ± 1 g versus 21 ± 1 g for females and 22 ± 2 g versus 26 ± 2 g for males , respectively ) , likely due to the hyperactive behavior . Esrp1Triaka mice were later found to be susceptible to acute challenge with 2% dextran sodium sulfate ( DSS ) . Increased susceptibility to DSS was also observed in homozygous versus WT mice on a mixed C57BL/6J-C57BL/10J background . The Triaka strain is further described at the Southwestern Medical Center Mutagenetix database ( http://mutagenetix . utsouthwestern . edu ) ( Krebs et al . , 2016 ) . Colonic crypt depth was assessed in longitudinally-sectioned crypts on H&E-stained sections . Slides were scanned and crypt depth was digitally measured in 20 crypts per mouse and averaged using the Pannoramic Viewer software ( RRID:SCR_014424; 3DHISTECH Ltd . , Budapest , Hungary ) . Colonic goblet cell numbers were assessed in longitudinally-sectioned crypts on PAS-stained sections using the Pannoramic Viewer software . The number of PAS-positive cells per crypt was averaged from 10 crypts per mouse . Fecal pellets were collected , weighed and resuspended in PBS containing 1% FBS . A sandwich ELISA for fecal albumin and lipocalin-2 was performed on 96-flat bottom Nunc MaxiSorp plates ( eBioscience ) . A purified anti-mouse albumin capture antibody ( Bethyl Laboratories , Montgomery , TX ) and a secondary HRP-conjugated anti-mouse albumin antibody ( Bethyl Laboratories ) were used for the capture and detection of fecal albumin . A Substrate Reagent Pack ( R&D Systems , Minneapolis , MN ) was used for the color reaction and 1 M sulfuric acid was added to stop color development . Colorimetric signals were measured on a SpectraMax M2e device ( Bucher Biotec AG , Basel , Switzerland ) . Fecal lipocalin-2 was measured using the mouse Lipocalin-2/NGAL MAb ELISA kit ( R&D systems ) , according to the manufacturer’s instructions . Colonic tissues were harvested and thoroughly washed in ice-cold calcium- and magnesium-free Hank’s Balanced Salt Solution ( HBSS ) . Colons were cut into 0 . 5–1 cm small pieces and digested in pre-warmed calcium- and magnesium-free HBSS containing 25 mM HEPES ( Sigma-Aldrich , St . Louis , MO ) , 2 mM DTT ( Sigma-Aldrich ) and 5 mM EDTA ( Merck , Darmstadt , Germany ) at 37°C for 20 min . Supernatants were filtered through 70 µm cell strainers ( BD Biosciences , San Jose , CA ) . The purity of colonic IECs ( cIECs ) was assessed by flow cytometry and was generally ≥95% of cIECs ( defined as EpCAM+ , CD45− cells ) in Esrp1Triaka and WT mice . Frequencies of major immune cell populations among CD45+ cells were analyzed in IEC preparations using the markers CD3 , CD19 and CD11b and were comparable in both groups of mice . Isolated cIECs were pelleted and suspended in TRI-reagent ( Sigma-Aldrich ) . For RNA sequencing , mRNA was purified and RNA concentration and integrity were assessed using a Bioanalyzer 2100 ( Agilent , Santa Clara , CA ) prior to cDNA synthesis and library preparation ( TruSeq Stranded mRNA Sample Preparation , Illumina , San Diego , CA ) . The libraries were sequenced on an Illumina HiSeq2500 sequencer by the Next Generation Sequencing Platform of the University of Bern . Between 45 . 7 and 59 . 1 million read pairs were obtained per sample . The reads were mapped to the mouse reference genome ( Ensembl m38 , build 75 ) using TopHat v . 2 . 0 . 11 ( RRID:SCR_013035 ) ( Kim et al . , 2013 ) . We then used HTseq-count v . 0 . 6 . 1 ( RRID:SCR_011867 ) ( Anders et al . , 2015 ) to count the number of reads per gene , and DESeq2 v . 1 . 4 . 5 ( RRID:SCR_015687 ) ( Love et al . , 2014 ) to test for differential expression between groups of samples . To investigate alternative pre-mRNA splicing , we identified transcripts and quantified their expression levels as described in detail elsewhere ( Trapnell et al . , 2012 ) . Specifically , TopHat v . 2 . 0 . 11 ( Kim et al . , 2013 ) was used to map the reads to the iGenomes reference ( Ensembl NCBIM37 available from Illumina ) for which an annotation file with Cufflinks-specific attributes is available . For each sample taken separately , transcripts were assembled with Cufflinks ( RRID:SCR_014597 ) and then combined into a single assembly with Cuffmerge ( RRID:SCR_015688 ) , resulting in approximately 145 , 000 transcripts . We tested for differential expression between Esrp1Triaka and WT mice using Cuffdiff ( RRID:SCR_001647 ) and visualized the results with CummeRbund ( RRID:SCR_014568 ) . In particular , we identified genes where the relative frequency of the isoforms from a given transcription start site ( TSS ) differed between Esrp1Triaka and WT mice ( false discovery rate ( FDR ) -adjusted p-value<0 . 05 ) , consistent with alternative pre-mRNA splicing . The outcome of the DESeq2 analysis was taken to perform gene set enrichment analysis ( GSEA ) using the SetRank method ( RRID:SCR_015689 ) ( Simillion et al . , 2017 ) . This algorithm first calculates the p‐value of a gene set utilizing the ranking of its genes in the ordered list of p‐value as determined by DESeq2 . Next , it discards gene sets that have been initially flagged as significant , if their significance is merely due to the overlap with another gene set . Gene sets were derived from the following databases: REACTOME ( RRID:SCR_003485 ) ( Croft et al . , 2014 ) , Gene Ontology ( RRID:SCR_006447 ) ( Ashburner et al . , 2000 ) , LIPID MAPS ( RRID:SCR_006579 ) ( Fahy et al . , 2009 ) , PhosphoSitePlus ( RRID:SCR_001837 ) ( Hornbeck et al . , 2012 ) , KEGG ( RRID:SCR_012773 ) ( Kanehisa et al . , 2014 ) , BIOCYC ( RRID:SCR_002298 ) ( Karp et al . , 2005 ) , ITFP ( RRID:SCR_008119 ) ( Zheng et al . , 2008 ) and WikiPathways ( RRID:SCR_002134 ) ( Kelder et al . , 2012 ) . Counts indicated in Figure 3—source data 1 were normalized for differences in total sequencing depth using size factor normalization in DESeq2 . Intestinal barrier integrity was measured ex vivo using an Ussing chamber ( Dipl . -Ing . K . Mussler , Scientific Instruments , Aachen , Germany ) , as previously described ( Clarke , 2009 ) . In brief , colon tissue was isolated from Triaka and WT mice , washed in ice-cold calcium- and magnesium-free HBSS and then mounted in a 37°C warm , oxygen saturated and HBSS-containing Ussing chamber . Electrical resistance was measured after 10 to 15 min of equilibration time . To assess systemic antibodies against intestinal commensals , fecal pellets were collected and intestinal bacteria from pellets were cultured in brain-heart infusion medium for 24 hours . Autologous fecal bacterial antibody binding was measured by bacterial flow cytometry . Complement in the serum was heat-inactivated and serum was titrated on sodium azide-inactivated bacteria . Specific binding of IgG1 and IgG2b to bacteria was detected with fluorochrome-labeled antibodies against mouse IgG1 or IgG2b ( BioLegend , San Diego , CA ) on a FACSArray device ( BD Biosciences ) . For the induction of acute colitis , animals were given 2% DSS ( MP Biomedicals , Santa Ana , CA ) in the drinking water for 7 days , followed by regular water . For chronic colitis , DSS was given in 3 cycles with 5 days of DSS followed by 7 days of regular water . Animal weight was measured throughout the procedure . Intestinal inflammation was assessed in a single blinded manner on H&E-stained tissue sections by a board-certified pathologist ( V . G . ) , as previously described ( Brasseit et al . , 2016 ) . Clinical disease activity was quantified as follows: Weight loss , compared to initial weight ( Score 0 , <1% weight loss; score 1 , 1–5% weight loss; score 2 , 5–15% weight loss; score 4 , 15–20% or more weight loss ) ; Stool consistency ( Score 0 , normal stool; score 2 , loose stool; score 4 , diarrhea ) ; Intestinal bleeding ( Score 0 , negative; score 2 , positive ) . Animals were anesthetized with isoflurane prior and during the experimental procedure . A miniature flexible biopsy forceps was used to inflict 2–3 mucosal injuries per mouse . The procedure was visualized using a miniature rigid endoscope system ( Karl Storz , Tuttlingen , Germany ) . Wound-healing was continuously monitored by colonoscopy 3 , 4 , 5 , and 7 days after the biopsy . To quantify the surface area of the excised mucosa and its subsequent regeneration , lesions were photographed and wound areas were measured using Photoshop ( RRID:SCR_014199; Adobe Systems , San Jose , CA ) and normalized to the forceps diameter . The azoxymethane ( AOM ) ( Sigma-Aldrich ) /DSS model of CRC was used to induce tumors , as previously described ( De Robertis et al . , 2011; Mertz et al . , 2016 ) . In short , mice were injected intraperitoneally ( i . p . ) with AOM ( 10 mg/kg of body weight ) . After 7 days , mice were given 1% DSS in the drinking water for 5 days , followed by 7 days of regular water . Thereafter , animals received a second injection of AOM , which was followed by two further cycles of DSS and regular water . Mice were sacrificed 70 days after the first injection of AOM . Tumors were counted macroscopically and measured with a caliper by two independent observers . Tumor size was measured as previously described ( Neufert et al . , 2007 ) . A board-certified pathologist evaluated the tumor grade in a single blinded manner . Grade was scored as follows: Grade 0: no tumor , healthy intestinal epithelium; Grade 1: adenoma with mild dysplasia; Grade 2: adenoma with moderate dysplasia; Grade 3: adenoma with severe dysplasia; Grade 4: adenoma with high grade dysplasia and infiltration of the lamina propria ( carcinoma in situ ) ; Grade 5: invasive carcinoma . Tumors or adjacent tumor-free colonic tissue were suspended in RIPA buffer ( Sigma-Aldrich , St . Louis , MO ) and homogenized using a TissueLyzer II ( Qiagen , Venlo , Netherlands ) . Samples were then centrifuged and the protein concentration in the supernatant was measured with a Bradford Protein Assay ( Bio-Rad , Hercules , CA ) . Protein concentrations were adjusted to 1500 µg/ml for each sample . Cytokines were analyzed by Multiplexing LASER Bead Technology ( Eve Technologies , Calgary , Canada ) . IECs or intraepithelial lymphocytes were filtered through a 70 µm cell strainer ( BD Biosciences ) to obtain a single-cell suspension . Live/dead cell discrimination was performed using DAPI ( BioLegend ) . Intracellular staining was performed after fixation of cells with 4% paraformaldehyde for 5 min and permeabilization with a 90% methanol solution at 4°C for 30 min . Antibodies used for flow cytometry are listed in Supplementary file 1 . RNA from isolated colonic IECs was extracted using TRI-reagent ( Sigma-Aldrich ) . RNA was then reverse-transcribed into cDNA using an M-MLV Reverse Transcriptase ( Promega , Fitchburg , WI ) . FastStart SYBR Green Master ( Roche , Basel , Switzerland ) and commercial primers specific for Cd44t , Cd44v4/v5 , Gapdh , Jun , Nedd9 , Fgf18 , ESRP1 , and EPCAM ( Qiagen , Venlo , Netherlands ) , as well as self-designed primers for specific isoforms of Uap1 , Magi1 , and Gpr137 ( Supplementary file 2 ) were used for the qPCR reaction . Reactions were performed and analyzed on a StepOnePlus Real-Time PCR System ( Life Technologies , Carlsbad , CA ) . Unless indicated , transcript levels were normalized to Gapdh expression . For tissue biopsies from Crohn’s disease patients , ESRP1 transcript levels were normalized to GAPDH and EPCAM expression . To generate a system to assess splicing events in vitro , an EcoRV restriction site was introduced into a commercially-available exon trap vector ( MoBiTec , Göttingen , Germany ) using the following primers: Fwd: 5’-GAGGCCCGATATCTTCAGACC-3’; Rev: 5’-GGTCTGAAGATATCGGGCCT-3’ . A Firefly luciferase gene ( Promega ) lacking the ATG start codon was then cloned into this EcoRV site . The variable exon 5 of Cd44 along with parts of the flanking introns were amplified using the following primers; Fwd: 5’-CGCGGGCTCGAGCATTGCAACAGATATAGAGACAGAATC-3’; Rev: 5’-CGCGGGGCGGCCGCCCTCTTTCAGGCTCTGCAGA-3’ . Similarly , a region of Fgfr2 containing exon IIIb and IIIc , the intron in-between , as well as parts of the flanking introns were amplified using the following primers; Fwd: 5’- CGCGGGCTCGAGGTCTGTTCTAGCACTACGGGGAT-3’; Rev: 5’- CGCGGGGCGGCCGCGCAGTATGTACCTGGCGAAC-3’ . For the cloning of the region of Gpr137 containing exons 2 and 3 , we performed a first amplification using the following primers; Fwd: 5'-GTGATGGGGTATCTCTGCTCC-3'; Rev: 5'-CCTTGATGTAGCACCCTTGGG-3' . This amplicon was then used to perform a nested PCR reaction using the following primers; Fwd: 5'-CTCCCAGGTGGTGTTCAATAGGCC-3'; Rev: 5'-CCAAGGTAGAGCCACC AACC-3' . Amplicons were then cloned into the multiple cloning site of the modified exon trap vector . To generate vectors encoding Triaka or WT Esrp1 , cDNA of Esrp1WT and Esrp1Triaka were amplified using the following primers: Fwd: 5’-CGCGGGGGATCCGCCACCATGACGGCGTCTCCGGATTA-3’; Rev: 5’-CCCGCGGTCGACCTTAAATACAAACCCATTCTTTGGG-3’ . Amplicons were then cloned into a pBRIT-HA/FLAG vector ( Addgene , Cambridge , MA ) . A third vector containing a Renilla luciferase gene was also used to control for variations in the transfection efficiency ( Promega ) . HEK-293 cells ( RRID:CVCL_0045 ) were transiently transfected by calcium phosphate co-precipitation with the modified exon trap vector containing the cloned Cd44 or Fgfr2 regions , the Esrp1WT- or the Esrp1Triaka-encoding vector , and the Renilla luciferase-encoding vector . Firefly and Renilla luciferase activities were measured 24 hours after transfection in an Infinite 200 PRO reader ( Tecan , Männedorf , Switzerland ) , using the Dual-Luciferase Reporter Assay ( Promega ) . All cell lines were routinely tested negative for mycoplasma contamination . CMT-93 cells ( RRID:CVCL_1986 ) were cultured as recommended by ATCC . Cell proliferation was either assessed by manual cell counting or using a WST-1 assay ( Roche ) . In brief , 8 × 104 cells were seeded into 6-well plates and cells were counted daily using a Neubauer chamber . Alternatively , 2–8 × 103 cells were seeded into a 96-well plate and proliferation was measured using a WST-1 assay , according to the manufacturer’s instructions . To measure the trans-epithelial electrical resistance ( TEER ) of transduced CMT-93 cells , 4 × 104 cells were seeded onto rat tail collagen ( Sigma Aldrich ) coated 24-well plate ThinCertsTM inserts ( Greiner Bio-One , Kremsmünster , Austria ) . These inserts were then placed into a cellZscope device ( nanoAnalytics , Münster , Germany ) and TEER was measured every hour for up to 72 hours . Patient information on gender , age at diagnosis , tumor diameter , tumor location , post-operative therapy and disease-specific survival time was extracted from clinical records . The UICC TNM Classification 7th edition was used to assess the T- ( pT ) , N- ( pN ) , and M-stage ( pM ) , lymphatic invasion ( L ) , venous invasion ( V ) , perineural invasion ( Pn ) , tumor grade ( G ) , histological subtype as well as tumor growth pattern . For each case , two TMA spots ( diameter 0 . 6 mm ) each of tumor center and tumor front and one spot of normal mucosa were stained for ESRP1 by immunohistochemistry ( IHC ) . We excluded 35 cases based on insufficient material remaining on the tissue block . Final number of patients with invasive CRC was 185 . From these patients , additional TMAs containing matched samples of normal colonic mucosa ( n = 26 ) , adjacent adenomas ( n = 42 ) and lymph node metastases ( n = 68 ) were constructed . Patient characteristics are indicated in Figure 6—source data 1 . Lentiviruses encoding Gpr137_Long or Gpr137_Short were generated using HEK-293 cells as previously described ( Tschan et al . , 2003 ) . In brief , HEK-293 cells were transfected with pMD2G , pMDLg/pRRE , pRSV-Rev and pLV ( Cyagen Biosciences , Santa Clara , CA ) encoding EGFP with either Gpr137_Long , Gpr137_Short , or a control sequence . HEK-293 cells were authenticated by short tandem repeat ( STR ) profiling and confirmed to be mycoplasma-negative monthly . Lentiviruses were then harvested to infect CMT-93 cells . Transduced CMT-93 cells were sort-purified based on EGFP expression and kept under selection pressure using 5 µg/ml Puromycin ( Sigma-Aldrich ) . CMT-93 cells were originally purchased from the European Collection of Cell Cultures ( ECACC no . 89111413 ) and confirmed to be mycoplasma-negative monthly . A 4-hydroxytamoxifen inducible system previously described elsewhere ( Vince et al . , 2007; Wang et al . , 2006 ) was modified to overexpress Esrp1Triaka , Esrp1WT or EGFP . To this aim , we exchanged the Hoxb8 cassette of the original vector with cDNA of Esrp1Triaka or Esrp1WT . CMT-93 cells were then co-transduced with lentiviral vectors containing these different inserts and a lentiviral vector encoding a 4-hydroxytamoxifen responsive element . Co-transduced cells were kept under selective pressure using 5 µg/ml Puromycin and 200 µg/ml Hygromycin-B ( Sigma-Aldrich ) . Expression of Esrp1Triaka , Esrp1WT or EGFP was induced with 100 nM 4-hydroxytamoxifen . RNA sequencing data form Triaka and WT IECs can be downloaded from the European Nucleotide Archive ( ENA ) ( http://www . ebi . ac . uk/ena/data/search ? query=PRJEB14221 ) . All RNASeqV2 data for datasets COAD and READ ( Cancer Genome Atlas Network , 2012 ) where downloaded though the TCGA data portal ( COAD: tumor: n = 480; normal tissue: n = 41 . READ: tumor: n = 167; normal tissue: n = 10 ) ( https://tcga-data . nci . nih . gov/docs/publications/coadread_2012/; RRID:SCR_003193 ) . From the RSEM results Tables , we extracted gene expression levels for ESRP1 , for all eight reported GPR137 isoforms , for Wnt target genes specific for CRC ( Herbst et al . , 2014 ) , as well as survival data . All results are shown as transcripts per million obtained by multiplying the provided scaled estimates by 106 . The sequences of all GPR137 isoforms were obtained from the FASTA file in the GAF bundle ( hg19 , June 2011 ) . For analysis of the survival data from the TCGA dataset , the X-tile software ( RRID:SCR_005602 ) ( Camp et al . , 2004 ) was used to separate GPR137_Short-high from -low tumors . CRC Patients without survival data or with undetectable gene expression were excluded from the survival analysis . Sample size for in vivo studies was estimated by power analysis and adjusted for β = 0 . 1 , with the assumption that differences between the groups were 1 . 5–2 fold . Statistical tests are two-sided and indicated in the figure legends . All statistical tests were performed with GraphPad Prism v . 5 . 04 for Windows ( RRID:SCR_002798; GraphPad Software , La Jolla , CA ) . Statistical tests were chosen based on the variation in each data group and on whether multiple comparisons were made . Groups with similar variance were compared using parametric tests , and groups with significantly different variations were analyzed using non-parametric tests . No mouse was excluded from the analysis . Only statistically significant differences are indicated in the figures . For all statistical analyses: *p<0 . 05; **p<0 . 01; ***p<0 . 001; ****p<0 . 0001 . | The lining of the intestine is just one cell thick , and yet it can act as an effective barrier between the inside of the body and the contents of the digestive system . This lining is often disturbed during bowel cancer , inflammatory bowel disease and other intestinal diseases , causing the barrier to fail and the gut to become leaky . These diseases often reduce patient life expectancy and quality of life . Intestinal epithelial cells make up the lining of the intestine and the normal activities of these cells are often disturbed during intestinal disease . In the intestine , a protein called ESRP1 is only found in epithelial cells , but its role in maintaining a healthy intestinal lining was not clear . Here , Mager et al . studied the intestines of mice that had been genetically engineered to produce a form of ESRP1 that is less active than normal . The experiments show that lower levels of ESRP1 activity leads to a broken intestinal barrier . The genetically engineered mice were more likely to develop inflammatory bowel disease and more aggressive forms of cancer . ESRP1 controls a gene that encodes another protein called GPR137 , which helps to relay signals to the epithelial cells . Lower levels of ESRP1 resulted in a longer form of the GPR137 protein being produced . This in turn affected the protein’s signaling role and disturbed the activities of intestinal epithelial cells . Further experiments on biopsies taken from patients with inflammatory bowel disease or colorectal cancer revealed that these patients had lower levels of ESRP1 compared to healthy individuals . Furthermore , low levels of ESRP1 and increased levels of the long version of GPR137 were associated with poorer outcomes for cancer patients . Together , these findings may help us to better understand how the intestinal barrier fails in mice and humans and could lead to new ways to monitor and treat intestinal disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"cancer",
"biology"
] | 2017 | The ESRP1-GPR137 axis contributes to intestinal pathogenesis |
The orbitofrontal cortex ( OFC ) has been described as signaling outcome expectancies or value . Evidence for the latter comes from the studies showing that neural signals in the OFC correlate with value across features . Yet features can co-vary with value , and individual units may participate in multiple ensembles coding different features . Here we used unblocking to test whether OFC neurons would respond to a predictive cue signaling a ‘valueless’ change in outcome flavor . Neurons were recorded as the rats learned about cues that signaled either an increase in reward number or a valueless change in flavor . We found that OFC neurons acquired responses to both predictive cues . This activity exceeded that exhibited to a ‘blocked’ cue and was correlated with activity to the actual outcome . These results show that OFC neurons fire to cues with no value independent of what can be inferred through features of the predicted outcome .
The orbitofrontal cortex ( OFC ) is often described as signaling either an outcome expectancy , implying a knowledge of the features of the impending outcome ( Schoenbaum et al . , 1998; Delamater , 2007; Ostlund and Balleine , 2007; Steiner and Redish , 2012; Luk and Wallis , 2013 ) , or a value that exists independent of those features ( Padoa-Schioppa , 2011; Levy and Glimcher , 2012 ) . Support for such pure or abstract value encoding comes largely from reports that single unit activity and the blood-oxygen level dependent ( BOLD ) response in the OFC tracks value , independent of outcome features such as identity or location or even the response required to obtain the outcome ( Padoa-Schioppa and Assad , 2006; Plassmann et al . , 2007; Levy and Glimcher , 2011 ) . Yet in some cases , outcome features could still be the underlying basis of apparent abstract value signals . This is conceivable even if the signal correlates with value across different outcomes , since some features or feature combinations might vary with value across the limited number of outcomes used in any particular session ( but see Padoa-Schioppa and Assad , 2006 , supplemental , and our correction notice ) . Further , any neural element ( voxel or single unit ) may participate in ensembles responding to more than one feature , so it is also possible that a particular element that appears not to distinguish specific features of different outcomes is in fact coding independent features that co-vary with each outcome's value . So how can we address whether the OFC signals features of impending outcomes vs value independent of those features ? One way is to strip away or ‘block’ the value portion of the outcome during learning , while leaving unblocked—free to enter into associations—the outcome's unique sensory and other features . This can be done by pairing a ‘target’ cue with a rewarding outcome in the presence of a cue that has been previously trained to predict a differently-flavored , but similarly-valued outcome . When this is done , the previously conditioned cue predicts the general value that is common to the two outcomes , but does not predict the unique features that distinguish the new outcome ( note features are not limited to sensory properties , but might include the outcome timing , location , temperature , size , number , etc ) . As a result , the target cue acquires associations with the unique features of the new outcome but not with its general or common currency value ( Rescorla , 1999; Burke et al . , 2008 ) . If OFC neurons represent only a general or common currency value , divorced from features , then they should respond no more to such a target cue than to a completely blocked cue ( Kamin , 1969 ) . However , if OFC neurons represent outcome features , independent of value , then they should respond to the target cue just as they do to a cue that has been explicitly unblocked by increasing the amount of the outcome delivered . Indeed , both pure value and outcome expectancy accounts of OFC function would predict neural activity to such an unblocked cue , but only an outcome expectancy account predicts encoding of the target cue signaling a valueless change in outcome flavor .
We recorded single–unit activity in the OFC in six rats during an odor-based unblocking task ( Figure 1A ) . Prior to implantation with electrodes rats were trained to sample an odor in a central port following house light illumination and then respond to a reward well below for two drops of Nestlé's flavored milk ( chocolate or vanilla , counterbalanced ) . This training was meant to establish the initial odor as a reliable predictor of a specific flavor and number of drops of milk . Each rat had extensive experience with both flavors , thus neither flavor was novel . Following initial training rats were implanted with microelectrodes in the OFC . When recovered from surgery , rats were retrained on the initial odor; after retraining , each rat underwent 7–9 rounds of unblocking . 10 . 7554/eLife . 02653 . 003Figure 1 . Experimental outline , behavior summary and recording sites . ( A ) Thirsty rats were initially trained to enter an odor port following illumination of a house light and respond at reward well below for two drops of flavored milk . Unblocking sessions consisted of four trial types . The first was a reminder of initial training . On the remaining three trial types , the originally trained odor was briefly presented followed by 1 of 3 novel odors . The reward following the novel odors was either unchanged ( black; blocked trials ) , increased in number ( blue; number trials ) , or its flavor was altered ( green; flavor trials ) . Learning was assessed in a probe test in which the novel odors were presented in isolation , without reward . ( B ) 10-min consumption testing between chocolate and water , and vanilla and water on non-training days found that both were significantly and equally preferred to water ( ANOVA , F1 , 5 > 5 , p's < 0 . 05 ) . ( C ) 2-min preference testing between chocolate and vanilla immediately following unblocking sessions found no flavor preference ( t test , p > 0 . 1 ) . Scatter plot ( right ) shows preference for the trained flavor on each individual test ( n = 13 ) . ( D ) 20-min consumption testing from a separate group of rats ( n = 8 ) that received selective devaluation of one of the flavors found a significant difference in consumption between the non-devalued ( Con ) and devalued ( Dev ) flavors ( t test , p < 0 . 01 ) . This was true for every rat tested ( right ) . ( E ) Time in the reward well is plotted for the probe test trials . ANOVA for time spent in the reward well with odor ( blocked , number and flavor ) and trial ( 1-15 ) as factors found a significant odor x trial interaction ( F1 , 47 = 3 . 45 , p < 0 . 05 ) . Planned comparisons confirmed that on the first three trials rats spent significantly more time in the reward well following number and flavor odors compared to blocked ( p's < 0 . 05 ) but responding to number and flavor did not differ ( p > 0 . 1 ) . ( F ) Single unit activity was recorded from the lateral orbital and agranular insular cortices at roughly 3 . 2 mm anterior to bregma . *p < 0 . 05; ns = not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 02653 . 00310 . 7554/eLife . 02653 . 004Figure 1—figure supplement 1 . Selective conditioned flavor aversion preference data . Preference for the devalued milk flavor ( devalued ) / ( devalued +non-devalued ) is shown for the 2 days of pre-exposure ( P1-2 ) , the 5 days of conditioning ( C1-5 , red background ) and the final choice test ( T ) . A value of 0 . 5 indicates equal preference while a value of 0 . 0 indicates an aversion to the devalued flavor . ANOVA for consumption over the 5 days of conditioning with session , flavor ( choc vs van ) and treatment ( control vs devalue ) found a significant session × treatment interaction ( F4 , 28 = 10 . 91 , p < 0 . 05 ) but no effect of or interaction with flavor . Asterisks indicate significance of a one-sample t test comparing % Devalue Preference to 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 02653 . 004 Each round of unblocking began with 2 days of training and consisted of four trial types . One type was a reminder; the initially trained odor was followed by the expected outcome . On the other three trial types ( blocked , number , flavor ) , rats were presented with the initially trained odor , followed immediately by one of three novel odors . On ‘blocked’ trials , the novel odor was followed by the expected two drops of the same flavor used in initial training . This outcome is fully predicted by the initial odor , thus the novel odor should be blocked from acquiring associative significance ( Kamin , 1969 ) . On ‘flavor’ trials , the novel odor was followed by two drops of the flavor not used in initial training ( i . e . , chocolate or vanilla ) . Here the value is unchanged , since the two flavors are equally preferred and the same amount is delivered , but the features of the outcome are different . Thus the novel target odor should enter into associations with the unique features of the outcome ( Rescorla , 1999; Burke et al . , 2008 ) . On ‘number’ trials , the novel odor was followed by an additional drop of the flavor used in the initial training . Since the initial odor does not predict anything after the second drop , the novel odor should enter into associations with both the features and the additional value of the outcome presented in the third drop ( Holland , 1984 ) . To ensure that learning in the flavor condition did not result from an explicit shift in value , two types of preference tests were administered in conjunction with the unblocking training procedures . In one test , given on days separate from unblocking , preference for each flavor over water was assessed . Both flavors were highly preferred to water ( Figure 1B ) , demonstrating both are highly palatable . The more important preference test came just following unblocking sessions , from which the critical neural data came . In these tests , the two milk flavors were pitted directly against one another . This test is critical to demonstrate that specific satiety to one flavor did not develop over the course of the unblocking session , a finding that would strongly suggest different valuation of the two flavors . Indeed , we found no evidence of a preference in these tests ( Figure 1C ) . In both types of tests the locations of the bottles were swapped every 20–30 s , meaning that rats were required to switch locations if the solution did in fact differ in value . This pattern was present when either milk flavor was compared to water but was absent when the two flavors were directly compared . The consumption data demonstrate that both flavors were highly palatable yet of equivalent value . Finally , to ensure the two flavors were discriminable we subjected another set of rats to a selective conditioned flavor aversion procedure . After initial exposure to both milk flavors , consumption of one flavor ( fully counterbalanced ) was devalued by pairing with LiCl-induced nausea while consumption of the other was paired with saline injection that is of minimal consequence . At no point did rats show a preference for the chocolate or vanilla flavor but all rats selectively reduced consumption of the devalued flavor . This was apparent both in conditioning ( Figure 1—figure supplement 1 ) and in the final choice test ( Figure 1D ) . Thus extensive consumption testing indicated that the flavors used were of equivalent value but readily discriminable . In the unblocking sessions rats were sensitive to presentation of the novel odors , exhibiting longer latencies to respond at the reward well following odor sampling on these three trial types . Longer latencies to the novel odors were most apparent on the very first trial of each session , particularly on day 1 . In support , ANOVA revealed a main effect of trial ( F1 , 47 > 2 , p's < 0 . 01 ) and a trial × day interaction ( F1 , 47 = 34 . 73 , p < 0 . 01 ) . However the rats also learned that the two novel odors that predicted changes in the outcome were meaningful . This was evident in the extinction probe test in which they initially spent more time in the fluid well following sampling of the flavor and number odors than following the blocked odor ( Figure 1E ) . We recorded 240 single units during the first day of unblocking and 220 units on the second unblocking day in 48 rounds of training across all six rats ( Figure 1F ) . To address our hypothesis , units from both unblocking days were screened for phasic responses to one of the four odors using a t test , which compared firing rates during the ITI and novel odor period ( significance level = p < 0 . 0125; Bonferroni correction ) . This screen found 135 units ( Day 1 = 79 , Day 2 = 56 ) that showed a significant increase in firing to at least one of the odor cues . The majority–98/135 or 73% of the neurons within this population–exhibited activity that fell into one of two categories ( see Figure 2—figure supplement 1 for analysis of other neurons ) . We will consider each category in turn below . The first major category , not directly anticipated by our hypothesis , consisted of neurons ( 55/135 , Figure 2A ) that showed a significant phasic response to each of the four odor cues . Since these neurons fired to all of the odors , even the blocked odor , their firing cannot be easily explained as signaling information about the predicted outcomes . However they might be signaling information about the cues themselves , such as their shared sensory features or intrinsic salience . Unlike shared sensory features , salience should be higher for the novel cues than for the pre-trained , initial odor in our design , and this pattern should be most noticeable early in training when the novel cues were first presented . As a population , these neurons did show greater activity to the blocked , number and flavor odors than to the more familiar , initial odor ( Figure 2B ) . This effect was present despite the fact that we did not select based on this criterion . Further analyses of the individual units showed that many exhibited significantly higher firing to the novel odors than to the initial odor ( Figure 2C ) , and few exhibited differences in firing among the three novel odors ( Figure 2D , E ) . Moreover , when we examined the firing of neurons in this population on the first 10 trials of unblocking ( 31/79 odor-responsive neurons; Figure 2—figure supplement 2 ) , analyzing the difference in firing between the novel and initial odor cues in a sliding , 300-ms window across each trial , we found that activity in this population was maximal at the onset of the novel odors and on the first exposure and then declined rapidly on subsequent trials ( Figure 2F ) . This same pattern held when each novel odor was analyzed separately ( Figure 2—figure supplement 3 ) . Further , this pattern was only seen on the first day of unblocking ( Figure 2—figure supplement 4 ) . This pattern of activity is consistent with signaling of the salience of these cues . 10 . 7554/eLife . 02653 . 005Figure 2 . Single unit and population firing of putative salience neurons . ( A ) Raster plots for firing of a single unit are shown for all initial ( red ) , blocked ( black ) , number ( blue ) and flavor trials ( green ) . Odor on ( On ) is indicated by the first vertical line , onset of novel odor ( Nov ) by the second vertical line and odor offset ( Off ) by the third . Each tick represents a spike . Average activity across all trials for each odor is plotted ( bottom ) . ( B ) Mean neural activity ( novel odor epoch–ITI ) for the putative salience neurons ( n = 55 ) is plotted . Line color as indicated in raster plots; shaded areas indicate standard error of mean . ANOVA with bin and odor as factors found significant effects of bin , odor and the bin × odor interaction ( F1 , 54 > 2 . 0 , p's < 0 . 01 ) . ANOVA restricted to the novel odor period with odor and time ( first 500 ms vs second 500 ms , shown in upper right inset ) as factors found only a main effect of odor ( F1 , 54 = 13 . 0 , p < 0 . 01 ) . Significant firing to the novel odors over the initial odor was observed throughout the novel odor period . ( C ) A scatter plot of novel odor firing vs initial odor firing is shown for putative salience neurons ( n = 55 ) . A signed square root transformation of firing was used to best visualize population spread; all statistics were performed on non-transformed firing rates . Individual neurons showing significant differences in firing between the odors are outlined in black ( t test , p < 0 . 05 ) . A non-parametric sign test found significant , preferential firing to the novel odors ( Z = 3 . 24 , p < 0 . 01 ) . The population bias towards novel odor firing is apparent in the bar histogram aligned to the diagonal axis; on which the difference score for each neuron is plotted . Light gray bars represent units showing no differential firing; dark gray bars represent units showing significant differential firing . ( D ) A scatter plot of predictive vs blocked odor firing is shown . A sign test found no differential firing to the predictive and blocked odors by the putative salience population ( Z = 0 . 54 , p > 0 . 1 ) . ( E ) A scatter plot of flavor and number odor firing is shown . A sign test found no differential firing to the number and flavor odors ( Z = 0 . 27 , p > 0 . 1 ) . ( F ) Differential firing to the novel odors vs the initial odor on the first 10 trials of the first unblocking day was calculated and plotted for the putative salience population ( n = 31 ) . Differential firing was calculated in a 300-ms sliding window for each 50-ms bin moving away from novel odor onset: ( mean [ ( blocked odor–ITI ) + ( number odor–ITI ) + ( flavor odor–ITI ) ]–initial odor–ITI ) . The difference score for each bin was then plotted , with dark red bins indicating maximal differential firing to the novel odors and dark blue indicating the opposite pattern ( y-axis shown on right of heat plot ) . ( G ) The significance of the increased firing to the novel odors was determined by performing a one-tailed t test , comparing increases in differential firing to 0 , using a significance of p < 0 . 05 and a sliding window as in ( F ) . Red bins indicate significant elevations in firing to the novel odors over the initial odor . DOI: http://dx . doi . org/10 . 7554/eLife . 02653 . 00510 . 7554/eLife . 02653 . 006Figure 2—figure supplement 1 . Odor-responsive units not included in primary analyses . Of the 37 units not analyzed in the main text , ( A ) nine showed selective responding to the blocked odor and ( B ) nine showed selective responses to the initial odor . The remaining neurons showed responses to different combinations of odors . ( C ) Neuron fired maximally to the number and blocked odors , possibly signaling predicted reward flavor , independent of number . ( D ) Neuron fired maximally to the flavor and blocked odors , possibly signaling predicted reward number , independent of flavor . However , these patterns were rare and were not reflective of the odor-responsive population . DOI: http://dx . doi . org/10 . 7554/eLife . 02653 . 00610 . 7554/eLife . 02653 . 007Figure 2—figure supplement 2 . Firing of putative salience neurons on unblocking day 1 . An identical analysis of only day 1 salience neurons ( n = 31 ) revealed nearly identical patterns as those reported when both day 1 and 2 neurons were analyzed . ( A ) ANOVA and subsequent post-hoc tests revealed identical significant results as in Figure 2B of the main text . ( B ) Sign test for novel vs initial odor firing differed slightly in it only approached significance ( p = 0 . 07 ) . ( C and D ) All other comparisons for salience neurons were identical to those reported in the main text . DOI: http://dx . doi . org/10 . 7554/eLife . 02653 . 00710 . 7554/eLife . 02653 . 008Figure 2—figure supplement 3 . Heat plots for putative salience neurons for each novel odor . Analyses identical to those performed in Figure 2F , G of the main text , in which mean firing to all three novel odors was analyzed , were performed for each individual novel , odor . Separate analyses of temporal firing to ( A and B ) blocked vs initial odor firing ( C and D ) number vs initial odor firing and ( E and F ) flavor vs initial odor firing revealed nearly identical patterns . Maximal responding is observed early on trial 1 and no or diminished responding is observed on subsequent trials . DOI: http://dx . doi . org/10 . 7554/eLife . 02653 . 00810 . 7554/eLife . 02653 . 009Figure 2—figure supplement 4 . Heat plot for putative salience neurons on unblocking day 2 . Analyses identical to those in Figure 2F , G of the main text were performed for salience neurons on unblocking day 2 . ( A ) Salience neurons on unblocking day 2 did not show the same temporal response to the first presentation of novel odors . The white arrow indicates where maximum response was observed in unblocking day 1 neurons . ( B ) This description is confirmed by statistical analysis which found no significant increase in firing on the first presentation of day 2; but did find significance later in the odor period on subsequent trials . DOI: http://dx . doi . org/10 . 7554/eLife . 02653 . 009 The second major category , of greater relevance to our hypothesis , consisted of neurons ( 43/135 , Figure 3A–C ) that showed a significant phasic response the flavor and/or number odors ( but did not fire to all four odors ) . Activity across this population was greater in response to the two predictive odor cues than to either the blocked or initial odors ( Figure 3D ) , and an analysis of individual units showed that nearly all of these neurons ( 38/43 ) fired more to the predictive odors than to the blocked one ( Figure 3E ) . This result marks these neurons as candidates for encoding of associative information or meaning , since this is the primary feature that distinguishes the two odor cues from the blocked odor . 10 . 7554/eLife . 02653 . 010Figure 3 . Single unit and population firing of putative predictive neurons . Single units plotted exactly as in Figure 2A showing ( A ) selective firing to the number odor ( B ) selective firing to the flavor odor ( C ) firing to both number and flavor odors . ( D ) Mean neural activity ( novel odor epoch–ITI ) for the putative predictive neurons ( n = 43 ) is plotted . Meaning of line colors and shading is maintained . ANOVA with bin and odor as factors found significant effects of bin , odor and the bin × odor interaction ( F1 , 42 > 2 . 0 , p's < 0 . 01 ) . ANOVA restricted to the novel odor period with odor and time ( first 500 ms vs second 500 ms , shown in upper right inset ) as factors found a main effect of odor ( F1 , 42 = 58 . 5 , p < 0 . 01 ) and an odor × time interaction ( F1 , 42 = 3 . 9 , p < 0 . 05 ) . At both times firing to the predictive odors was significantly greater than the blocked and initial odors; blocked firing was greater than initial firing only in the first half . ( E ) A scatter plot comparing firing to the predictive odors ( signed square-root transform ) vs the blocked odor is shown for the predictive population ( n = 43 ) . Within this population there were three kinds of neurons based on firing vs ITI: number only ( blue ) , flavor only ( green ) or flavor and number ( purple ) . A sign test found significant , preferential firing to the predictive odors ( Z = 4 . 88 , p ( s ) < 0 . 01 ) . Across all neurons there was zero correlation between predictive odor firing and blocked odor firing ( R2 = −0 . 01 , p ( r ) = 0 . 39 ) . ( F ) A scatter plot comparing firing to the flavor and number odors is shown for the predictive population ( n = 43 ) . While some neurons did show differential firing ( outlined in black ) to either the number ( n = 14 ) or flavor ( n = 4 ) odor a sign test found no bias in firing to the number or flavor odor across the entire population . ( Z = 0 . 60 , p > 0 . 1 ) . Across all neurons there was a highly significant , positive relationship between number and flavor odor firing ( R2 = 0 . 88 , p < 0 . 01 ) . ( G ) Differential firing to the number odor vs the blocked odor on the first 10 trials of the first unblocking day was calculated and plotted for the number-responsive units within the predictive population ( n = 21 ) . This was done in as in ( Figure 2F ) except that the difference score was calculated as: ( number odor–ITI ) — ( blocked odor–ITI ) . ( H ) Significance for increased firing to the number odor over the blocked odor was calculated as in ( G ) . Red bins indicate significant elevations in firing to the number odor over the blocked odor . Blue bins would indicate significant decreases in firing to the number odor below the blocked odor . ( I ) Differential firing to the flavor odor vs blocked odor on the first 10 trials of the first unblocking day was calculated and plotted for flavor-responsive units within the predictive population ( n = 18 ) as was done in ( G ) . ( J ) Significance of differential firing calculated and displayed as in ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02653 . 01010 . 7554/eLife . 02653 . 011Figure 3—figure supplement 1 . Firing of putative predictive neurons on unblocking day 1 . An identical analysis of only day 1 predictive neurons revealed identical patterns as those reported when both day 1 and 2 neurons were analyzed . ( A ) ANOVA , and subsequent post-hoc tests , revealed identical significant results as those reported in Figure 3D . ( B ) Sign test comparing predictive odor firing to blocked odor firing revealed a significant bias towards the predictive odors ( p < 0 . 05 ) . ( C ) Sign test comparing flavor and number odor firing found no population bias towards either odor ( p > 0 . 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02653 . 011 Interestingly , these neurons did not appear to distinguish , at least as a population , between the flavor and number odors . For example , they showed similar levels of activity in response to both the flavor and the number odor ( Figure 3D , F ) , and when we examined the firing of neurons in this population on the first 10 trials of unblocking ( 25/79 odor-responsive neurons , Figure 3—figure supplement 1 ) , we found that differential firing to each cue developed at a similar rate during training ( Figure 3G–J ) . The acquisition of differential firing to the flavor and number odors demonstrates that selective odor encoding was not driven by physical properties of the odors . If neurons were encoding the odor itself , independent of its outcome signaling , this would have been apparent on the very first trial . Thus , as a population , these neurons responded more strongly to the unblocked ‘flavor’ and ‘number’ odors than to the ‘blocked’ odor . Even more striking , the population responded similarly to a cue signaling a valueless change in the outcome flavor as they did to a cue that signaling that more of the outcome would be delivered . Similar numbers of neurons fired to the flavor and number cues ( flavor: 31 , number: 37; χ2 = 0 . 3 , p = 0 . 47 ) . In our design , firing to the flavor cue cannot be readily explained as signaling general or common value . Thus these data confirm that many OFC neurons signal associative meaning independent of at least a general value . In support of this , odor firing in the flavor population , as well as the number population , was positively correlated with firing to the actual outcome delivered on each of these trials ( Figure 4 ) . This relationship supports the idea that cue-evoked activity in the flavor population is signaling features of the new outcome . 10 . 7554/eLife . 02653 . 012Figure 4 . Outcome selectivity of predictive neurons . ( A ) For each predictive neuron that significantly increased firing to the flavor odor ( total n = 31; flavor-only n = 6 [green]; number and flavor n = 24 [purple] ) we plotted its selective firing to the flavor odor ( x-axis; [normalized flavor odor firing – mean ( normalized initial odor firing , normalized blocked odor firing , normalized number odor firing ) ] ) against its selective firing to the flavor outcome ( y-axis; [normalized flavor outcome firing – mean ( normalized initial outcome firing , normalized blocked outcome firing , normalized number outcome firing ) ] ) . Comparison of odor firing of single neurons to the population found a single outlier ( neuron firing was 3 stdev > population firing ) . The outlier was omitted from this analysis . There was a significant , positive relationship such that greater selective firing to the flavor odor was associated with greater selective firing the flavor outcome ( R2 = 0 . 41 , p < 0 . 01 ) . ( B ) This relationship was restricted to the flavor outcome; plotting selective flavor odor firing against selective number outcome firing revealed zero correlation ( R2 = 0 . 01 , p = 0 . 70; calculation identical to A only mean ( blocked , initial and flavor outcome firing ) was subtracted from number outcome firing . ( C ) For each predictive neuron that significantly increased firing to the number odor ( total n = 37; number only n = 12 [blue]; number and flavor n = 23 [purple] ) we plotted it's selective firing to the number odor ( x-axis; [normalized number odor firing – mean ( normalized initial odor firing , normalized blocked odor firing , normalized flavor odor firing ) ] ) against its selective firing to the number outcome ( y-axis; [normalized number outcome firing – mean ( normalized initial outcome firing , normalized blocked outcome firing , normalized flavor outcome firing ) ] ) . Two neurons showed selective odor firing 3 stdev above the population mean and were excluded from analysis . There was a significant , positive relationship such that greater selective firing to the number odor was associated with greater selective firing the number outcome ( R2 = 0 . 27 , p < 0 . 01 ) . ( D ) This relationship was restricted to the number outcome; plotting selective number odor firing against selective flavor outcome firing revealed zero correlation ( R2 = 0 . 04 , p = 0 . 26 ) calculation identical to C only mean ( blocked , initial and number outcome firing ) was subtracted from flavor outcome firing . Finally , these statistical patterns were maintained if the flavor-only and number-only neurons were analyzed in isolation: ( A ) R2 = 0 . 71 , p = 0 . 03 , ( B ) R2 = 0 . 01 , p = 0 . 82 , ( C ) R2 = 0 . 48 , p = 0 . 01 and ( D ) R2 = 0 . 18 , p = 0 . 17 . Flav = flavor , Num = number , N&F = number and flavor . DOI: http://dx . doi . org/10 . 7554/eLife . 02653 . 012
Neural signals in the OFC are often described as representing either outcome expectancies or abstract value . Although many studies have argued for one or the other , few have used behavioral designs that clearly dissociate predictions of these two hypotheses . Here we tried to address this question by using an unblocking procedure to strip away or ‘block’ the abstract value of the outcome during learning , while leaving unblocked—free to enter into associations—the outcome's sensory and other unique features . This approach revealed two distinct populations of OFC neurons . One population consisted of neurons that fired immediately on initial presentation of all three target cues , perhaps reflecting these cues' novelty or salience . While unexpected and not directly relevant to the question motivating this study , this finding is consistent with reports of neural correlates of salience in OFC ( Kahnt and Tobler , 2013; Ogawa et al . , 2013 ) and with studies implicating the OFC in phenomena such as latent inhibition , set formation , and even auto-shaping ( Chudasama et al . , 2003; Schiller and Weiner , 2004; Chase et al . , 2012 ) which depend in part on the appropriate attribution of salience to cues . Together these results point to a largely unappreciated role for this area in the modulation of attention for the purposes of learning ( Esber et al . , 2012 ) . Of course novelty is just one instance of ‘salience’ . Modern learning theories describe salience as a function of both intrinsic and acquired properties . Within this framework , this population was correlated with intrinsic salience . The second population , of more direct relevance to our hypothesis , consisted of neurons that fired preferentially to the target cues that predicted changes in the outcome flavor . Nearly all of these neurons fired more to these cues than to the similarly trained but fully blocked cue , and this activity was acquired with learning , a pattern consistent with signaling of the associative significance or meaning of these cues ( or possibly their acquired salience ) . Importantly , the acquired neural activity was observed to both the explicitly unblocked ‘number’ cue as well as to the ‘flavor’ cue , which was unblocked by shifting the features of the expected outcome while holding the value constant . This was accomplished by using two differently-flavored but similarly-preferred outcomes ( Figure 1C , D ) . The lack of any flavor preference makes it unlikely that the cue added prior to this manipulation acquired what might be termed a general or cached value . Indeed , in prior work we have found that responding to this target cue in the probe test is dissociable from even the smallest animal-by-animal differences in conditioned or unconditioned responding to the two outcomes used ( Schoenbaum et al . , 2011 ) , indicating that it is not driven by any sort of shift in value that might accrue to the cue . Instead conditioned responding to a target cue unblocked by shifting the identity of the outcome seems to be particularly dependent upon the unexpected outcome's unique features , at least in comparison to an explicitly unblocked cue . As evidence of this , it has been shown that cues trained in this manner support behavior that is more sensitive to ( in fact completely dependent upon ) the features of the predicted outcome—or value inferred through those features—than similar behaviors supported by cues directly paired with reward in isolation . For example , conditioned reinforcement supported by a normally trained cue is insensitive to devaluation of the predicted outcome ( Parkinson et al . , 2005 ) ; however if the cue is trained like the ‘flavor’ cue in the current experiment , then devaluation of the predicted outcome completely abolishes the ability of the cue to serve as a conditioned reinforcer ( Burke et al . , 2008 ) . This result indicates that a cue trained in this manner has little or no intrinsic , cached or acquired value except what can be inferred through knowledge of the features of the outcome . This cue's special link to the sensory features of the outcome is also apparent in that such cues retain the ability to support Pavlovian-to-instrumental transfer when that transfer is specific to the outcome ( Rescorla , 1999 ) . Interestingly outcome-specific transfer is both insensitive to devaluation ( Holland , 2004 ) and disrupted by OFC lesions ( Ostlund and Balleine , 2007 ) , results which are difficult to reconcile with the view that the OFC is directly involved in the representation of value . That OFC neurons developed robust responses to such a valueless cue indicates that many OFC neurons—more than half of the population responsive to the acquired significance of the cues—represent associative features that must be , strictly speaking , independent of general or common value . Notably this population included neurons that fired only to the flavor cue as well as neurons that participated in both the flavor and number ensembles . Such dual encoding would be expected if , as suggested earlier , individual neurons are not labeled lines but participate in ensembles coding more than one outcome feature ( flavor , number , temperature , location , timing , etc ) . Of course some neurons fired preferentially to the valued , number cue . The firing of these neurons could reflect the general value that accrued to this cue during training . However such firing could equally well reflect associations with features of the additional outcome delivered on these trials , known to develop when additional rewards are delivered during unblocking ( Holland , 1984 ) . While it is impossible to say for sure , the similarities in the overall level of neural activity to the flavor and number cues , their similar rates of development with learning , and the finding that neurons do develop activity to a valueless cue make the latter explanation the most parsimonious . If OFC neurons signal associations between cues and specific outcome features , this would accord well with results showing that the OFC is not necessary when behavior—or learning—can be accomplished using general value alone . For example , the OFC is not required for simple Pavlovian or instrumental conditioning ( Gallagher et al . , 1999; Izquierdo et al . , 2004; Ostlund and Balleine , 2007; Gremel and Costa , 2013 ) discrimination learning ( Schoenbaum et al . , 2002; Izquierdo et al . , 2004; McDannald et al . , 2005; Walton et al . , 2010 ) , extinction by reward omission ( Takahashi et al . , 2009 ) , transfer ( Ostlund and Balleine , 2007 ) , and even perhaps reversal learning ( Rudebeck et al . , 2013 ) , all of which can be accomplished without reference to specific information about predicted outcomes . Similarly both blocking and unblocking—when it can be accounted for by value—do not require the OFC ( Burke et al . , 2008; McDannald et al . , 2011 ) . While the preserved function in these studies could reflect compensation by other areas , it must at least call into question the idea that OFC , writ large , is fundamental to all behavior that reflects value , instead highlighting suggestions that common value representation may at least be limited to the medial subregion ( Noonan et al . , 2010 , 2012 ) . Indeed recent work in humans has shown that OFC represents specific outcome features and that more lateral orbital areas represent those outcomes in a way that is dependent upon prior cues ( Klein-Flugge et al . , 2013 ) . Thus far from signaling general value about outcomes without regard to their features and attendant events , this work shows that the OFC maintains highly specific representations . Such highly specific representations are consistent with observations that the OFC is necessary for superficially similar behaviors ( Pavlovian or instrumental responding , discriminations , even learning ) when they require knowledge of the outcome features in order to recognize errors or to derive or infer a value ( Gallagher et al . , 1999; Izquierdo et al . , 2004; Ostlund and Balleine , 2007; McDannald et al . , 2011; Gremel and Costa , 2013 ) . This is even true in the current paradigm , where we have shown that the OFC is required for the development of conditioned responding to the target cue paired with a shift in outcome identity but not to the target cue paired with additional outcome ( McDannald et al . , 2011 ) . Our present finding of robust encoding of a valueless Pavlovian cue that exceeds that of a blocked cue , and is equivalent to a cue paired with additional outcome , provides further support for outcome expectancy theories of OFC function .
Male Long-Evans rats were obtained at 200–250 g from Charles River Labs , ( Wilmington , MA ) . Rats were tested at the University of Maryland School of Medicine and the NIDA-IRP in accordance with SOM and NIH guidelines ( 12-CNRB-108 ) . Using aseptic , stereotaxic surgical techniques , a drivable bundle of 16 , 25 µm diameter FeNiCr wires ( Stablohm 675 , California Fine Wire , Grover Beach , CA ) was chronically implanted dorsal to OFC in the left hemisphere at 3 . 0 mm anterior to bregma , 3 . 2 mm laterally , and 4 . 0 mm ventral to the surface of the brain in each rat . Immediately prior to implantation , these wires were freshly cut with surgical scissors to extend ∼1 mm beyond the cannula and electroplated with platinum ( H2PtCl6; Aldrich , Milwaukee , WI ) to an impedance of ∼300 kΩ . At the end of the study , the final electrode position was marked by passing a 15 µA current through each electrode . The rats were then perfused , and their brains removed and processed for histology using standard techniques . Recording was conducted in grounded aluminum chambers approximately 18″ on each side with sloping walls narrowing to an area of 12″ × 12″ at the bottom . A central odor port was located above a fluid well on a panel in the right wall of each chamber . Two lights were located above the panel . The odor port was connected to an airflow dilution olfactometer to allow the rapid delivery of olfactory cues to the odor port; odors where chosen from compounds obtained from International Flavors and Fragrances ( New York , NY ) . The fluid well was connected to lines controlling the independent delivery of the fluid rewards . Task control was implemented via computer running a behavioral program written in C++ . Prior to implantation with microelectrodes , rats were water deprived by restricting daily access to 1 hr following each training session . Water-deprived rats were progressively shaped to hold in the odor port for 1 s to receive two drops of water at the well . After shaping , rats received further training until they were proficiently responding for the initial odor in order to receive two boli of milk ( vanilla or chocolate-flavored , counterbalanced ) ; this involved as many as 15 sessions , with a maximum of 240 trials in each session . Proficient responding was characterized as correctly completing ∼200 trials per session . Each trial began with house light illumination after which rats had 3 s to enter the odor port . Failure to enter the odor port resulted in restart of the trial . Once in the odor port rats were required to hold for 1 s and upon exit had 3 s to enter the reward well . Again , failure to hold for 1 s or enter the reward well within 3 s resulted in restart of the trial . On alternate days , rats were given 20-min ad libitum exposure to the untrained milk flavor . Following implantation , rats were retrained on the initial odor and once single units were isolated the unblocking procedure began . On the 2 learning days , rats received four trial types . The first was a reminder of initial training . The remaining trial types began with a 200 ms presentation of the initial odor and were followed by 1 of three 800 ms , novel yet distinguishable odors . The behavioral requirements of each of these trial types were exactly as in initial training . Rats completed between 30–60 trials involving each novel odor per session . On the subsequent probe test day , rats received a brief reminder of each trial type , ∼10 total trials and then were presented the novel odors alone without reward , interleaved with trials in which the initial odor was presented with reward , in order to maintain responding . On the unrewarded , novel-odor extinction trials , the requirement to sample the odor for 1-s and respond to the reward well was lifted . The unblocking procedure was repeated seven to nine times per rat using a new set of blocked , number and flavor odors each time; in some cases , a new initial odor was also trained prior to repeating unblocking . When the initial odor was changed rats were trained on the new initial odor for 4–5 sessions prior to the first unblocking procedure with that odor . Consumption tests were given in a housing cage separate from their home cage and experimental chamber . Two varieties of two-bottle tests were given . In the first , consumption of one of the flavored milks ( chocolate or vanilla ) was compared to consumption of water . These tests were 10-min in duration and occurred on days when unblocking training was not performed . The second test directly compared consumption of the two flavored milks . These tests were 2-min in duration and occurred immediately following unblocking sessions . This second test was critical for showing that rats did not become selectively sated to the flavored milk they experienced more in the unblocking sessions . Further , these tests occurred immediately after unblocking sessions while rats were still in the recording setting , informing any preference that would have developed over the course of the unblocking session . For all tests the location of the bottles was swapped roughly every 20–30 s to equate time on each side . Naïve rats were exposed to the vanilla and chocolate-flavored milk twice each in 1-hr sessions . During pre-exposure sessions intraperitoneal saline injections ( 0 . 9% , 5 ml/kg i . p . ) were given to habituate rats to the injection procedure . Conditioning consisted of five , 1-hr exposures to each of the two solutions spaced over 10 days . Exposure to the devalued flavor was followed by injection of lithium chloride ( 0 . 3 M LiCl , 5 ml/kg i . p . ) . Exposure to the control flavor was followed by injection of saline . Consumption of each flavor was measured daily . A final 20-min choice test was given in which both flavors were present . At the 10-min mark the locations of the bottles were swapped . All factors ( identity of devalued flavor , order of flavor presentation , side of flavor during choice test ) were fully counterbalanced . Neural activity was recorded using two identical Plexon Multichannel Acquisition Processor systems ( Dallas , TX ) , interfaced with odor discrimination training chambers described above . After recovery from surgery , electrodes were advanced daily until activity was obtained on a majority of wires . During this process , rats received reminder training using the pre-trained initial odor , as described above . Once the electrode was in a suitable location in OFC , single units were isolated and rats showed proficient responding , the rat began unblocking . During this 3-day procedure , the electrode was generally left in the same position . Thus , although we will not attempt to track neurons across sessions , the general population should be similar across each 3-day period . Following the completion of each 3-day unblocking procedure , the electrode was advanced 40–80 µm , and the process was repeated using new odor cues to test neurons in a new location in OFC . Units were sorted using Offline Sorter software from Plexon Inc ( Dallas , TX ) using a template matching algorithm . Sorted files were then processed in Neuroexplorer to extract unit timestamps and relevant event markers . These data were subsequently analyzed in Matlab ( Natick , MA ) . To examine activity to the novel odors , we examined activity from 300–1300 ms after initial odor onset , which corresponded approximately to the time during which the novel odors were delivered to the odor port . To examine activity to the flavor outcome , we examined activity for 2000 ms starting with the first drop delivery . To examine activity to the number outcome , we examined activity for 2000 ms starting 1000 ms following first drop delivery , coinciding with the third drop delivery . The inter-trial interval was defined as the 2 s prior to illumination of the house light . Normalized firing was calculated by taking the firing rate during a period of interest minus firing rate during the ITI: Normalized firing = ( Period spikes/s ) − ( ITI spikes/s ) . Neurons were identified as being odor responsive with a Bonferroni-corrected t test ( four separate tests , 0 . 05/4 = 0 . 0125; p < 0 . 0125 ) comparing elevations in firing during each of the four odor epochs ( initial , blocked , number and flavor ) from their respective ITIs . Neurons were classified as putative salience neurons if they significantly increased firing to all odors; putative predictive neurons were classified by significantly increasing firing to either or both the number and flavor odors , but not to all four odors . Single-unit and population activity was plotted in 50-ms bins; population activity was analyzed with repeated measures ANOVA with bin ( 50 ms ) and odor trial ( initial , blocked , number and flavor ) as factors . Heat plots were constructed by calculating difference scores between normalized firing to Novel and Initial odors ( Figure 2F ) , Number and Blocked odors ( Figure 3G ) and Flavor and Blocked odors , in 200-ms sliding windows moving away from the novel odor onset in 50 ms increments . Warmer colors ( dark red ) indicated positive difference scores while cooler colors ( dark blue ) indicated negative difference scores . This was done for the first 10 trials of the identified population . Significance of differential firing to Novel and Initial odors ( Figure 2F ) , Number and Blocked odors ( Figure 3G ) and Flavor and Blocked odors was determined by performing a one-tailed t test comparing differential firing to zero in the exact same 200-ms sliding windows for each of the 10 trials . Finally , for purposes of visualization only; single unit and population firing , as well as heat plot comparisons were smoothed by taking a four-bin average moving in 50 ms increments . This applied to Figure 2 ( A–bottom , B , and F ) , Figure 3 ( A–bottom , B–bottom , C–bottom , D , G , I ) and Figure 4A . All statistical analyses were performed on unsmoothed data . | Imagine you are at a restaurant and the waiter offers you a choice of cheesecake or fruit salad for dessert . When making your choice it is likely that you will consider the features of these desserts , such as their taste , their sweetness or how healthy they are . However , when you decide which dessert to have , you will pick the one that you judge to have the highest value for you at that moment in time . In this sense , ‘value’ is a subjective concept that varies from person to person , while ‘features’ remain relatively static . It is generally agreed that the orbitofrontal cortex ( OFC ) is involved in making these sorts of decisions , but its role is still a topic of debate . According to one theory the neurons in the OFC signal the subjective value of an outcome , whereas a rival theory suggests that they signal the features of the expected outcome . However , it has proved challenging to test these theories in experiments because it is difficult to say for certain that a given decision was clearly due to the value or a feature . Now , McDannald et al . have devised an approach that can tell the difference between neurons signaling value and neurons signaling features . They trained thirsty rats to associate different odours with either an increase in the amount of milk they were given ( a change in both value and a feature ) , or a change in the flavor of the milk ( a change in a feature without a change in value ) . Extensive testing showed that the rats did not value one flavor over the other . McDannald et al . then examined how the neurons in the OFC responded . If these neurons signal only value , they should only fire when the value of the outcome changes . On the other hand , if they signal features , they should fire when a feature changes , even if the value does not . It turned out that the neurons in the OFC responded whenever the features changed , irrespective of whether or not the value changed . These findings present a challenge to popular conceptions of the role of the neurons in the OFC . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2014 | Orbitofrontal neurons acquire responses to ‘valueless’ Pavlovian cues during unblocking |
Vascular endothelial growth factor-C ( VEGF-C ) acts primarily on endothelial cells , but also on non-vascular targets , for example in the CNS and immune system . Here we describe a novel , unique VEGF-C form in the human reproductive system produced via cleavage by kallikrein-related peptidase 3 ( KLK3 ) , aka prostate-specific antigen ( PSA ) . KLK3 activated VEGF-C specifically and efficiently through cleavage at a novel N-terminal site . We detected VEGF-C in seminal plasma , and sperm liquefaction occurred concurrently with VEGF-C activation , which was enhanced by collagen and calcium binding EGF domains 1 ( CCBE1 ) . After plasmin and ADAMTS3 , KLK3 is the third protease shown to activate VEGF-C . Since differently activated VEGF-Cs are characterized by successively shorter N-terminal helices , we created an even shorter hypothetical form , which showed preferential binding to VEGFR-3 . Using mass spectrometric analysis of the isolated VEGF-C-cleaving activity from human saliva , we identified cathepsin D as a protease that can activate VEGF-C as well as VEGF-D .
Vascular endothelial growth factor VEGF-A is essential for early embryonic development and for successful implantation of the embryo into the uterus ( Binder et al . , 2014 ) . VEGF-A acts in this function on both vascular and non-vascular targets ( Hannan et al . , 2011 ) . The primary function of the closely related growth factor VEGF-C is stimulation of growth of the lymphatic vasculature ( Rauniyar et al . , 2018 ) . VEGF-C is required for ovarian follicle growth and maturation and endometrial lymphangiogenesis ( Rogers , 2008; Rutkowski et al . , 2013 ) . Unlike VEGF-A , which is secreted as an active growth factor ( Leung et al . , 1989 ) , VEGF-C is secreted as an inactive precursor ( pro-VEGF-C ) , which requires two proteolytic cleavages for activation ( Jeltsch et al . , 2014; Joukov et al . , 1997 ) . The first C-terminal cleavage resulting in pro-VEGF-C occurs constitutively in the endoplasmic reticulum and is mediated by proprotein convertases ( Siegfried et al . , 2003 ) . The second cleavage takes place in the extracellular environment , is highly regulated and requires the assembly of a trimeric complex consisting of VEGF-C , the ADAMTS3 metalloproteinase and the ‘cofactor’ CCBE1 ( Bui et al . , 2016; Jeltsch et al . , 2014 ) . Alternative activation by plasmin has been shown in vitro , but its significance under physiological settings is unknown ( McColl et al . , 2003 ) . VEGF-D is the closest paralog of VEGF-C ( Achen et al . , 1998 ) . Similar to VEGF-C , it is lymphangiogenic ( Veikkola et al . , 2001 ) , but appears to have a higher angiogenic potential than VEGF-C ( Byzova et al . , 2002; Rissanen et al . , 2003 ) . The proteolytic activation of VEGF-D is very similar to that of VEGF-C ( Stacker et al . , 1999a ) , but it deploys distinct , so far unknown proteases ( Bui et al . , 2016 ) . Many kallikrein-related peptidases are highly expressed in the prostate , and some prostate-derived cell lines , such as the immortalized human normal prostate epithelial ( NPrEC ) or PC-3 cells — from which VEGF-C was originally cloned — express high amounts of VEGF-C ( Grennan , 2006; Joukov et al . , 1996 ) . In a peptide library scan , Matsumura et al . ( 2005 ) identified VEGF-C as a potential substrate for KLK4 . Based on these observations , we tested human kallikrein-related peptidases for their ability to activate VEGF-C . In this study , we show that KLK3 , the major protease in human semen , is able to specifically activate VEGF-C and VEGF-D . We further show that cathepsin D cleavage of VEGF-C results in a novel , predominantly VEGFR-3-binding form of VEGF-C , and that cathepsin D cleavage of VEGF-D at the homologous site results in a VEGFR-2-specific ( minor mature ) form of VEGF-D .
We could not demonstrate robust VEGF-C activation by KLK4 as predicted by Matsumura et al . ( 2005 ) ( data not shown ) , but purified KLK3 cleaved pro-VEGF-C , resulting in a mature protein that migrated at about 20 kDa in Western blotting analysis ( Figure 1A , lane 2 ) . To confirm that KLK3 was responsible for the cleavage , we inhibited its protease activity by using the monoclonal antibody 5C7 ( Stenman et al . , 1999 ) in 2-fold molar excess ( Figure 1A , lane 1 versus lane 2 ) . We probed the polypeptide bands resulting from the cleavage with rabbit antiserum 6 and antiserum 3/4 , which were raised against full-length and mature VEGF-C , respectively ( Figure 1A and B , compare the second lanes ) . Probing with antiserum 3/4 , which recognizes both pro-VEGF-C and mature VEGF-C , showed that the majority of pro-VEGF-C had been cleaved by KLK3 . We tested the KLK3-processed VEGF-C for its biological activity in Ba/F3 cells , which had been stably transfected with VEGFR/EpoR chimeras and found that it promoted the survival of both VEGFR-2/EpoR ( Figure 1C ) and VEGFR-3/EpoR cells ( Figure 1D ) . Edman degradation of the KLK3-processed VEGF-C revealed the amino-terminal sequence NTEIL ( Figure 2—figure supplement 1 ) . Thus , KLK3 cleaves VEGF-C between Tyr-114 and Asn-115 , targeting a sequence similar to most of its cleavage sites in the seminogelins , which are the primary target proteins of KLK3 ( Malm et al . , 2000 ) . The KLK3-cleaved VEGF-C is three N-terminal amino acid residues shorter than the mature VEGF-C generated by ADAMTS3 ( Jeltsch et al . , 2014 ) and 12 amino acid residues shorter than the mature VEGF-C produced by PC-3 cells ( Figure 2 ) ( Joukov et al . , 1997 ) . We analyzed the VEGF-C amino acid sequences of 40 vertebrate species and found that residues −7 to +1 relative to the KLK3 cleavage site and −4 to +4 relative to the ADAMTS3 cleavage site ( KFAA↓AHY↓N ) are 100% conserved among all mammals and birds that were included in the analysis . However , we found significant differences in this area in all fish species analyzed ( Figure 2—figure supplement 2 ) . To evaluate the biological significance of VEGF-C activation by KLK3 , we first analyzed the VEGF-C content of human seminal plasma . Because of difficulties in detecting VEGF-C at low ng/ml-range concentrations in a high-protein sample ( ~50 mg/ml ) , such as semen ( Owen and Katz , 2005 ) , we first compared the ability of different anti-VEGF-C antibodies to detect VEGF-C ( Figure 3—figure supplement 1 , Supplementary file 1 ) . VEGF-C was detected in Western blots of sperm liquefied for approximately 20–30 min at room temperature by using antibody sc-374628 after VEGF-C precipitation with soluble forms of its receptors VEGFR-2 ( VEGFR-2/Fc ) and VEGFR-3 ( VEGFR-3/Fc ) or anti-VEGF-C antiserum 882 ( Figure 3A ) . The affinity of seminal plasma VEGF-C towards VEGFR-2 appeared to be much weaker than towards VEGFR-3 in the VEGF-C pull down assay ( Figure 3A , compare lanes 4 and 6 ) . The mobilities of the VEGF-C polypeptides indicated that it is composed of inactive pro-VEGF-C and active mature VEGF-C . Stimulation of VEGFR-3-transfected porcine aortic endothelial ( PAE ) cells with seminal plasma resulted in VEGFR-3 phosphorylation ( Figure 3B , compare lanes 2 and 4 ) . VEGF-C stimulation of PAE cells stably expressing VEGFR-2 led to an even stronger phosphorylation than the recombinant VEGF-C control . We reasoned that this could indicate the presence of VEGF-A , whose concentrations in seminal plasma have been reported to range from less than 2 ng/ml to more than 100 ng/ml ( Obermair et al . , 1999 ) . Indeed , most of the VEGFR-2 phosphorylation was blocked when incubated with soluble VEGFR-2/Fc , but not by incubation with VEGFR-3/Fc ( Figure 3—figure supplement 2 ) . In contrast , VEGF-D was not detected in seminal plasma ( Figure 3—figure supplement 3 ) . When fresh ejaculates were immediately mixed with protease inhibitors , placed on ice and analyzed , less active VEGF-C was detected than in ejaculates that had been liquefied , indicating that pro-VEGF-C is converted into mature VEGF-C after ejaculation ( Figure 3C ) , concurrently with sperm liquefaction . Lowering the pH with citric acid tended to increase slightly the yield of mature VEGF-C ( Figure 3C , lane 5 ) , but ion chelation with CHELEX 100 or EDTA had no effect ( Figure 3C , lanes 2 and 4 , respectively ) . We have shown that CCBE1 enhances the proteolytic activation of VEGF-C by ADAMTS3 , but not by plasmin ( Jeltsch et al . , 2014 ) . Therefore , we tested whether CCBE1 would accelerate KLK3 activation of VEGF-C . We found that KLK3-mediated cleavage of VEGF-C was enhanced by CCBE1 when CCBE1 or KLK3 amounts were titrated down so that only little VEGF-C processing occurred ( Figure 4 ) . Substantial amounts of CCBE1 were detected in seminal plasma by Western blotting ( Figure 4—figure supplement 1 ) , confirming published proteomics results ( Jodar et al . , 2016 ) . This indicates that VEGF-C cleavage could be increased by CCBE1 also in semen . Activated VEGF-C binds to VEGFR-2 ( Joukov et al . , 1997 ) , but in our assays with seminal plasma , VEGFR-2 binding was very weak ( Figure 3A , lane 4 ) . To explain this finding , we focused on the cleavage of the N-terminal helix , because its partial removal in VEGF-D decreases selectively VEGFR-3 binding while leaving VEGFR-2 binding intact ( Leppänen et al . , 2011 ) . Since complete proteolytic removal of the N-terminal helix of VEGF-C abolishes all receptor binding and phosphorylation-stimulating activity ( Jeltsch et al . , 2014 ) , we first tested if a partial removal of the N-terminal helix ( cutting between Leu-118 and Lys-119 , corresponding to the proteolytic cleavage site between Leu-114 and Lys-115 of VEGF-D ) would result in a selective loss of VEGF-C binding to its receptors . The protease that cleaves between Leu-114 and Lys-115 of VEGF-D ( and hypothetically between the homologous Leu-118 and Lys-119 of VEGF-C ) is unknown . Therefore , we generated this form of VEGF-C by truncating the VEGF-C cDNA , which was then expressed in S2 cells . Interestingly , unlike the corresponding VEGF-D form , this ‘VEGF-CDMH’ ( for ‘D Minor Homology’ ) bound to VEGFR-3 , but only weakly or not at all to VEGFR-2 ( Figure 5A ) . Because of this unexpected result , we performed the experiment using proteins produced in 293 T cells and found that in conditions where all other mature forms of VEGF-C interacted with their receptors as predicted , VEGF-CDMH did not bind to VEGFR-2 or VEGFR-3 ( Figure 5B ) . Since the loss of binding compared to the S2 cell-produced VEGF-CDMH is not associated with a loss of receptor-interacting amino acid residues , we attributed the loss of binding to the extra N-terminal four amino acid residues of the mammalian linker ( see also Figure 5—figure supplement 1 ) . We then used equimolar amounts of truncated VEGF-Cs expressed in transiently transfected CHO cells ( Figure 5—figure supplement 1 ) to test the bioactivity of different N-terminally truncated VEGF-Cs in VEGFR-2 and VEGFR-3 phosphorylation assays . The receptor phosphorylation results mirrored the binding results . The longest mature VEGF-C resulted in the strongest stimulation , and progressive shortening of the N-terminus resulted in gradually decreased stimulation of the receptor phosphorylation ( Figure 5C ) . We also tested the activity of purified VEGF-CDMH expressed in S2 cells . In agreement with the binding results , 100 ng/ml VEGF-CDMH did stimulate the phosphorylation of VEGFR-3 but not or only very weakly of VEGFR-2 ( Figure 5D ) . A comparison of the sizes of VEGF-C polypeptides produced by S2 cells transfected with N-terminally truncated cDNAs encoding the polypeptide resulting from cleavage by ADAMTS3 and the ( longer ) form generated by the 1st plasmin cleavage revealed bands of identical size , indicating additional proteolytic processing ( Figure 5—figure supplement 1 ) . N-terminal sequencing of the form produced from the longer cDNA revealed that about ⅔ had the KSIDNE… N-terminus , and about ⅓ had AAAHYN . . . as N-terminus . Hence , the DMH-form of mature VEGF-C can also be produced by proteolytic processing of a longer mature form of VEGF-C by a yet unknown protease . We refer to such cleavage on top of an existing activation in the following as secondary activation ( irrespectively of the receptor activation ability of the resulting protein species ) . The presence of a VEGF-C-cleaving protease in seminal fluid prompted us to search for such a protease also in other body fluids . We enriched the VEGF-C activating component of human saliva by cation exchange chromatography ( Figure 6—figure supplement 1 ) and subjected the fractions containing the peak activity to mass spectrometric analysis . Among the highest scoring proteases ( Supplementary file 2 ) , cathepsin D was identified as the most likely candidate due to the cleavage context of the DMH-form of VEGF-C ( Leu-118↓Lys-119 ) . Using purified recombinant proteins , we confirmed that cathepsin D cleaves pro-VEGF-C into active VEGF-C ( Figure 6A and Figure 7AB ) and performs a secondary activation of the minor , mature form of VEGF-C ( Figure 6A and Figure 7F ) . Because the sequence contexts of the cleavage sites of cathepsin D and KLK3 are conserved between VEGF-C and VEGF-D ( see Figure 2 ) , we investigated , if also cathepsin D and KLK3 could activate pro-VEGF-D . Indeed , cathepsin D activated pro-VEGF-D and performed a secondary activation of the longer , mature form of VEGF-D . The cleavage of mature VEGF-D was rapid and complete ( Figure 6C ) , whereas the cleavage of both pro-VEGF-C and mature VEGF-C was slower and incomplete even after 16 hr ( Figure 6AB ) . As expected , the cathepsin D processing of mature VEGF-D abolished most of its activity in the Ba/F3-VEGFR-3/EpoR assay ( Figure 7C ) and reduced , but did not abolish its activity in the Ba/F3-VEGFR-2/EpoR assay ( Figure 7D ) , while processing of pro-VEGF-D stimulated the phosphorylation of VEGFR-2 ( Figure 7—figure supplement 1 ) . KLK3 also activated pro-VEGF-D ( Figure 6D and Figure 7E ) . When VEGF-D was produced from a full-length cDNA using the baculovirus system , a significant fraction of the protein did not undergo processing by proprotein convertases . This allowed us to observe two additional KLK3 cleavage sites in pro-VEGF-D . One of these cleavages has been reported previously ( Stacker et al . , 1999a ) ; the other cleavage mimics the C-terminal cleavage catalyzed by proprotein convertases that cleave between the VHD and the N-terminal propeptide ( Figure 6D and Figure 7—figure supplement 2 ) . To confirm that the two new forms of VEGF-C have also an effect in-vivo , we transduced mouse skeletal muscle ( tibialis anterior ) with recombinant adeno-associated viruses serotype 9 encoding the KLK3- or the cathepsin D- ( CATD ) form of VEGF-C . Both vectors stimulated lymphangiogenesis and angiogenesis ( Figure 8 ) . As expected on the basis of the binding and receptor phosphorylation results , the response to the KLK3-form was stronger compared to the cathepsin D-form . Both forms appeared to give a stronger response compared to the positive control ( the ADAMTS3-form of VEGF-C ) , but the higher expression level of the shorter VEGF-C forms likely explains most of this difference ( Figure 8—figure supplement 1 ) .
The angiogenic effect of VEGF-A is required for example for implantation ( Torry et al . , 2007 ) and corpus luteum formation ( Reynolds et al . , 2000 ) . VEGF-A levels in human seminal plasma are variable , typically between 10–20 ng/ml ( Brown et al . , 1995; Obermair et al . , 1999 ) , and VEGF-A has been implicated as a fertility factor that acts on sperm cells ( Obermair et al . , 1999 ) . Sperm motility has been reported to increase slightly as a response to VEGF-A ( Iyibozkurt et al . , 2009 ) , and overexpression of a testis-specific VEGF-A transgene resulted in infertility ( Korpelainen et al . , 1998 ) . VEGF-C is the lymphangiogenic counterpart of VEGF-A , and lymphangiogenesis is required for ovarian follicle maturation ( Rutkowski et al . , 2013 ) , corpus luteum formation ( Abe et al . , 2014; Nitta et al . , 2011 ) and uterine implantation ( Red-Horse , 2008 ) . Furthermore , VEGF-C and VEGF-D are hormonally regulated in the reproductive system ( Nitta et al . , 2011 ) . The prostate produces KLK3 and contributes active KLK3 to semen . KLK3 is the major protease in semen and participates in seminal clot liquefaction . KLK3 from human seminal plasma cleaved VEGF-C between its N-terminal propeptide and the VEGF homology domain . Compared to the major form of mature VEGF-C , the form produced by KLK3 lacks three amino acid residues from the N-terminus , but it still activated both VEGFR-2 and VEGFR-3 . In vitro , both sperm liquefaction and VEGF-C exposure to KLK3 resulted in efficient cleavage of VEGF-C . However , in natural insemination , several factors , such as the vaginal environment or the absent mixing of the early prostatic fraction with the seminal vesicular fluid fraction ( Björndahl and Kvist , 2003 ) may interfere with VEGF-C activation . Apart from KLK3 , seminal plasma also contains many other proteases involved in the proteolytic liquefaction cascade ( Emami and Diamandis , 2013 ) , which might contribute to VEGF-C activation ( and inactivation ) , including cathepsin D ( this study ) and plasmin ( Jeltsch et al . , 2014; Stief , 2007 ) . Similar to seminal plasma TGF-β ( Robertson et al . , 2002 ) , which is also activated during liquefaction ( Emami and Diamandis , 2010 ) , VEGF-C might also contribute to the impregnation-associated immunomodulation . Several types of immune cells express VEGF-C receptors ( Hamrah et al . , 2003; Krebs et al . , 2012; Li et al . , 2016 ) , and VEGF-C may be responsible for the immune tolerance of uterine NK cells during pregnancy ( Kalkunte et al . , 2009 ) . However , since KLK3 exists only in higher primates ( Pavlopoulou et al . , 2010 ) , any function of KLK3-mediated VEGF-C activation in seminal fluid is difficult to address experimentally . On the other hand , mice have many kallikrein-related peptidases that have no human counterparts ( Pavlopoulou et al . , 2010 ) , and one of these might functionally replace KLK3 as an activator of VEGF-C . Unlike in mice , KLK3 prevents copulatory plug formation in humans , where sperm liquefaction is thought to be a physical requirement for sperm movement ( Mann and Lutwak-Mann , 2012 ) . VEGF-A inhibition marked a conceptual breakthrough in antiangiogenic cancer treatment ( Ferrara et al . , 2005 ) . Although every single VEGF paralog in humans ( PlGF , VEGF-B , VEGF-C , VEGF-D ) has been proposed to mediate the tumor escape under anti-VEGF-A treatment ( Li et al . , 2014; Lieu et al . , 2013 ) , only VEGF-C and VEGF-D activate VEGFR-2 and VEGFR-3 ( Achen et al . , 1998; Joukov et al . , 1996 ) and are therefore prime suspects ( Kubota , 2012; Li et al . , 2014; Stacker et al . , 2001; Wang and Tsai , 2015 ) . VEGF-A is able to activate VEGFR-2 immediately after secretion , but VEGF-C and VEGF-D need to be proteolytically processed to gain angiogenic ( Joukov et al . , 1997; Stacker et al . , 1999a ) or lymphangiogenic activity ( Jeltsch et al . , 2014 ) . The involvement of KLK3/PSA for tumor progression is still debated with studies arguing both in favor or against a tumor-promoting function of KLK3 ( Fortier et al . , 1999; Ishii et al . , 2004; LeBeau et al . , 2010; Mattsson et al . , 2008; Webber et al . , 1995 ) . KLK3 expression is largely restricted to the male prostate ( MediSapiens Ltd , 2019; Shaw and Diamandis , 2007 ) , but small amounts can be found in other tissues , such as Skene's gland , the female homolog to the prostate ( Zaviacic and Ablin , 2000 ) . In pathological settings , the highest expression levels are found in prostate cancers ( MediSapiens Ltd , 2019 ) . We have hypothesized that KLK3 may facilitate early development of prostate cancer , but at later stages slow down cancer growth ( Koistinen and Stenman , 2012 ) . VEGF-C expression , which overlaps in the prostate with KLK3 expression ( Joory et al . , 2006 ) , is similarly controversial , with some studies supporting ( Jennbacken et al . , 2005; Yang et al . , 2014 ) and others refuting ( Mori et al . , 2010 ) its predictive ability for prostate cancer progression . Most experimental animal models confirm the role of VEGF-C for metastatic spread ( Brakenhielm et al . , 2007; Burton et al . , 2008 ) , and potential mechanisms have been identified in cell culture models ( Rinaldo et al . , 2007 ) . This study shows that , at least in principle , KLK3 could contribute to the activation of tumor-derived VEGF-C or VEGF-D and thus to a ( lymph ) angiogenic tumor phenotype . KLK3 is a serine protease , but like ADAMTS3 , its activity towards VEGF-C was increased by CCBE1 . This reinforces the view that CCBE1 interacts with VEGF-C in the trimeric VEGF-C/ADAMTS3/CCBE1 complex , removing the masking of the cleavage site by the C-terminal domain of VEGF-C ( Joukov et al . , 1997 ) . This idea is supported by the ability of the isolated C-terminal domain of VEGF-C to competitively inhibit CCBE1-accelerated VEGF-C activation by ADAMTS3 ( Jeltsch et al . , 2014; Jha et al . , 2017 ) . It would also explain why VEGF-C activation by plasmin is not controlled by CCBE1 , as the plasmin cleavage site is located ~10 amino acids residues further away from the receptor binding epitopes than the ADAMTS3 and KLK3 cleavage sites ( see Figure 2 ) . VEGF-CDMH was a designed variant with an N-terminal cleavage resembling that in the minor ( VEGFR-2-specific ) form of VEGF-D . After we had established that VEGF-CDMH-like form is produced by cathepsin D via proteolytic cleavage of a longer VEGF-C polypeptide , we confirmed that also the VEGFR-2-specific form of VEGF-D ( minor , mature form ) is indeed produced by cathepsin D cleavage . However , cathepsin D cleavage affects VEGF-C and VEGF-D activities differently . While VEGF-D loses practically all binding affinity towards VEGFR-3 , VEGF-C seems to lose preferentially its affinity towards VEGFR-2 . The minor mature form of VEGF-D was identified in the supernatant of VEGF-D-producing 293 cells ( Stacker et al . , 1999a ) , where VEGF-CDMH was not detected ( Joukov et al . , 1997 ) , presumably because cathepsin D-cleavage of VEGF-C is inefficient . Alternatively , the ADAMTS3-cleavage of VEGF-C in 293 cells may have preemptively removed the recognition epitope required for cathepsin D cleavage . Our data show that the longest forms of mature VEGF-C and VEGF-D can undergo secondary activation ( i . e . N-terminally cleaved on top of a prior , activating cleavage ) . This introduces an additional layer of complexity into the regulation of VEGF-C and VEGF-D signaling since the cathepsin D-cleavage abolishes the VEGFR-3 binding of VEGF-D and reduces the VEGFR-2 binding of VEGF-C . Cathepsin D is ubiquitously expressed , and although it is involved predominantly in lysosomal protein degradation ( Benes et al . , 2008 ) , it can be secreted and soluble cathepsin D is found in saliva ( our present findings ) and in seminal plasma ( Jodar et al . , 2016 ) . Secondary activation by cathepsin D may explain why we saw only weak VEGF-C-VEGFR-2 interaction when analyzing seminal plasma . It should be noted that cathepsin D has also been implicated in cancer metastasis ( Benes et al . , 2008; Spyratos et al . , 1989 ) , where VEGF-C can also play a role ( Karpanen et al . , 2001; Mandriota et al . , 2001; Skobe et al . , 2001 ) . However , the cathepsin D-mediated secondary activation of the major , mature form of VEGF-D was very rapid , when compared to the very slow activation of VEGF-C ( compare Figure 6A and C ) . Therefore , VEGF-D activation appears to be a more relevant function of cathepsin D than VEGF-C activation . The cathepsin D-processed minor form of mature VEGF-D showed a lower potency to activate VEGFR-2 than the major form of mature VEGF-D , likely reflecting the corresponding KD values ( Leppänen et al . , 2011 ) . Despite this , as a net effect , cathepsin D cleavage of VEGF-D may result in increased angiogenic activity . In VEGF-A , the N-terminal helix in the VEGF homology domain appears essential for the receptor dimerization and activity ( Siemeister et al . , 1998 ) , whereas the platelet derived growth factor does not need an N-terminal helix for receptor binding ( Muller et al . , 1997; Shim et al . , 2010 ) . The Leu119↓Lys120 ( cathepsin D ) cleavage of VEGF-C happens within the N-terminal helix , which contains binding epitopes for VEGFR-2 ( Leppänen et al . , 2010 ) . The N-terminal helix also interacts with VEGFR-3 . However , mutating the contacting amino acid residues Asp123 and Gln130 only ameliorates binding of VEGF-C to VEGFR-3 ( Leppänen et al . , 2013 ) . The present receptor phosphorylation data strongly suggest that shortening of the helix leads to decreased activation of both VEGFR-2 and VEGFR-3 , whereas a complete or near-complete removal of the N-terminal alpha helix- for example by extended plasmin exposure - abolishes all receptor binding . Inline with this , both the KLK3- and the cathepsin D-forms of VEGF-C induced lymphangiogenesis and angiogenesis in skeletal muscle . The N-terminal helix of VEGF-C is largely conserved among vertebrates , but the C-terminal end of the N-terminal propeptide and linker preceding the VEGF homology domain represent the most diverse sequences among VEGF-Cs in different species . These differences are especially noticeable between fish and the rest of the vertebrate clade ( Figure 2—figure supplement 2 ) , indicating potential differences in the VEGF-C activation . Although ADAMTS3 appears to be responsible only for developmental lymphangiogenesis , our study indicates that other proteases may activate VEGF-C for specific niche functions , for example KLK3 in the reproductive system . The possible involvement of cathepsin D and KLK3 in tumor metastasis could be addressed in the appropriate gene-targeted mouse models . The possible other niche functions of VEGF-C , for example in the central nervous system ( Mackenzie and Ruhrberg , 2012 ) , in osmoregulation ( Machnik et al . , 2009 ) or in the immune system ( Loffredo et al . , 2014 ) , may also be controlled by differentially regulated proteases .
KLK3 ( isoform B ) was purified by immunoaffinity chromatography from pooled seminal plasma ( Wu et al . , 2004 ) . The separation of the different isoforms by anion-exchange chromatography was performed as described ( Zhang et al . , 1995 ) . For the production of untagged pro-VEGF-C , full-length human VEGF-C cDNA was cloned into the Drosophila expression vector pMT-BiP ( Invitrogen/Thermo Fisher Scientific , Waltham , MA ) . The protein was expressed in stably transfected S2 cells in Insect-Xpress medium ( Lonza , Basel , Switzerland ) supplemented with 250 µg/ml hygromycin at 26°C . The cells were induced with 1 mM CuSO4 and the conditioned medium was harvested 4 days post-induction . VEGF-C was purified from the conditioned medium by Heparin affinity chromatography ( HiTrap Heparin HP , GE Healthcare , Chicago , IL ) at pH 6 . 7 , followed by cation exchange chromatography over a MonoS or HiTrap SP HP column ( GE Healthcare ) at the same pH and gel filtration on a Superdex 200 Increase ( GE Healthcare ) column in PBS . C-terminally his-tagged pro-VEGF-D was produced with the baculovirus system as described ( Achen et al . , 1998 ) . Purification was performed by affinity chromatography over Excel sepharose ( GE Healthcare ) , followed by gel filtration as described for pro-VEGF-C . C-terminally his-tagged mature VEGF-D was produced from a truncated cDNA analogous to pro-VEGF-D . CCBE1 protein was produced and purified as described ( Jeltsch et al . , 2014 ) . Similarly , human VEGFR-3/Fc ( containing extracellular domains 1–7 ) and VEGFR-2/Fc ( containing extracellular domains 1–3 ) were purified as described ( Jeltsch et al . , 2006; Leppänen et al . , 2013; Leppänen et al . , 2010 ) . Recombinant Adeno-Associated Viruses ( AAVs ) were produced as previously described ( Jeltsch et al . , 2014 ) . Stably transfected cell lines were obtained directly from the generating laboratories ( indicated by the reference in the key resource table ) and other cell lines were obtained from the indicated vendors , who authenticate and monitor for mycoplasma status of these products according to applicable regulations . All anti-VEGF-C antibodies are listed in the Supplementary file 1 . We used further the following antibodies: anti-phosphotyrosine antibody 4G10 ( Merck/Millipore ) , anti-VEGFR-3 antibody sc-321 ( Santa Cruz Biotechnology , Dallas , TX ) , anti-VEGFR-2 ( AF357 , R and D Systems , Minneapolis , MN ) , anti-VEGF-D ( AF286 , R and D Systems ) , anti-CCBE1 ( HPA041374 , Atlas Antibodies/Sigma-Aldrich/Merck ) , and Penta-His antibody ( #34660 , Qiagen , Hilden , Germany ) . For the immunofluorescence , the primary antibodies anti-CD31 ( BD Biosciences ) and anti-Lyve-1 ( Karkkainen et al . , 2004 ) were detected using the appropriate Alexa Fluor 488 and 594 secondary antibody conjugates ( Molecular Probes/Invitrogen ) . Antiserum ( AS ) no . 3/4 , AS 885 and AS 890 were generated like AS no . 6 ( Baluk et al . , 2005 ) , except that mature VEGF-C ( Kärpänen et al . , 2006 ) was used as the antigen instead of pro-VEGF-C for AS no . 3/4 , and peptide antigens ( see Supplementary file 1 for details ) for AS 885 and AS 890 . 0 . 94 μg of purified KLK3 was incubated with 1 . 7 μg of recombinant growth factor in TBS pH 7 . 7 at 37°C for 24 hr , if not differently indicated . For blocking , the monoclonal antibody against KLK3 , 5C7 ( Stenman et al . , 1999 ) was used in 2-fold molar excess and the cleavage was analyzed by SDS-PAGE/Western using antiserum 6 and 3/4 ( VEGF-C ) and AF286 ( VEGF-D , R and D Systems ) . For CCBE1-enhanced cleavage experiments , 10 μl CCBE1-StrepIII ( equal to the amount of CCBE1 purified from 12 . 5 ml of conditioned 293T medium ) were included in the reaction . 80 µg of pro-VEGF-C/pro-VEGF-D in 240 µl PBS or ΔNΔC-VEGF-C/ΔNΔC-VEGF-D in 60 µl PBS were incubated with the same volume of human , recombinant cathepsin D , which had been activated and diluted according to the instructions of the manufacturer ( 1014-AS , R and D Systems ) . Incubation was performed at 37°C , and aliquots were taken at 15 min , 1 hr , 4 hr and 16 hr and frozen at −80°C until analysis . Samples were resolved by reducing SDS-PAGE and proteins were visualized by Coomassie Blue staining . The activation of pro-VEGF-D and ΔNΔC-VEGF-D was visualized by Western blotting . 293T and CHO cell transfections and procedures were performed as described ( Jeltsch et al . , 2014 ) . For the N-terminal sequence analysis , the digestion mixture of purified KLK3 and recombinant pro-VEGF-C or purified protein was resolved by SDS-PAGE and blotted to a PVDF membrane using 1xCAPS buffer/10% methanol . The membrane was Coomassie-stained and the band at 20 kDa was excised after destaining with 50% methanol . Edman degradation was performed using a Procise 494 HT sequencer ( Applied Biosystems/Thermo Fisher Scientific ) and data analyzed with the Sequence Pro software . Multiple N-termini were disambiguated by a fuzzpro search ( Rice et al . , 2000 ) of the major peaks against the VEGF-C and KLK3 sequences and eliminating results incompatible with the molecular weight observed on the gel . The Ba/F3-hVEGFR-3/EpoR ( Achen et al . , 2000 ) and Ba/F3-mVEGFR-2/EpoR ( Stacker et al . , 1999b ) bioassays were performed with recombinant proteins as described ( Mäkinen et al . , 2001 ) . Fresh ejaculates , showing normal sperm parameters ( Cooper et al . , 2010 ) , were collected from healthy volunteers among the authors ( three different individuals ) in full agreement with local regulations and institutional oversight . For analysis by SDS-PAGE/Western blotting , seminal plasma was separated from the cellular fraction and debris after approximately 30 min of liquefaction at RT by centrifuging twice for 10 min ( at 1000 g and 10000 g ) . Seminal plasma was stored at −80°C until further analyses . Prior to analysis , thawed seminal plasma samples were sonicated and centrifuged again for 10 min at 16000 g at 4°C . The upper white layer was discarded and the clear fraction was collected for analyses . To test the effects of divalent cation concentration and pH on the cleavage of VEGF-C in seminal plasma , 50 mg of Chelex 100 Resin ( Bio-Rad , Hercules , CA ) , 10 µl 0 . 5M EDTA , or 25 µl 0 . 1M citric acid were added during the initial liquefaction to each ml of seminal fluid , and samples were incubated for 24 hr at 37°C before continuing with the centrifugation steps . To slow the proteolytic liquefaction cascade , fresh ejaculates were immediately transferred to ice and a protease inhibitor cocktail ( cOmplete , Roche ) pre-dissolved in PBS was added at twice the recommended final concentration . Two centrifugation steps of 10000 g were performed at 4°C to separate the cellular and gel fraction from the liquid phase and samples were stored at −80°C until further analysis . Before the gel fraction was loaded , it was incubated at 37°C until liquefaction . For precipitation with antibodies or soluble receptors , the seminal plasma samples ( processed as described above ) were diluted 1 + 1 with PBS and incubated with 30 µl protein A-Sepharose-4B beads and the respective antibody or soluble receptor overnight at 4°C . The beads were washed three times with PBS/0 . 05% Tween-20 and the bound proteins were eluted by adding 30 µl of 2X Laemmli standard buffer ( LSB ) followed by heating at 95°C for 10 min . For direct loading of proteins ( digestion analysis of VEGF-C and VEGF-D , CCBE1 from seminal plasma ) , 2x or 5x LSB was added to the samples prior to boiling . For Western blotting , proteins were resolved on SDS-PAGE , transferred to PVDF membranes , blocked with 5% BSA in TBS-T for 1 hr and probed overnight with the relevant primary antibodies . The membranes were incubated with the appropriate HRP-conjugated secondary antibodies ( Jackson Immuno Research , Cambridgeshire , UK , anti-rabbit IgG ( 111-035-003 ) , anti-mouse IgG ( 115-035-003 ) or anti-goat IgG ( 705-035-003 ) , 1:2500 in 5% skimmed milk in TBS-T ) for 1 hr at RT and bands were visualized with ECL plus Western Blotting Substrate ( Pierce/Thermo Fisher Scientific , Waltham , MA ) or SuperSignal West Femto Maximum Sensitivity Substrate ( Pierce/Thermo Fisher Scientific ) using the LI-COR Odyssey Fc or cDigit Imaging System ( Li_COR , Lincoln , NE ) . Direct visualization of proteins in the PAGE gels was performed by Coomassie Blue or silver staining . The level of VEGF-C in seminal plasma ( processed as described above ) was estimated using the Human VEGF-C Quantikine ELISA Kit ( DVEC00 , R and D Systems ) following the manufacturer’s instructions . Near confluence , PAE cells expressing strep-tagged VEGFR-3 ( Leppänen et al . , 2013 ) or VEGFR-2 ( Anisimov et al . , 2013 ) were washed with PBS and starved for 4–5 hr in DMEM . PAE cells expressing untagged VEGFR-3 or VEGFR-2 starved for 16 hr in DMEM/0 . 1% BSA were used to analyze N-terminally truncated VEGF-Cs . The cells were stimulated for 10 min with sonicated centrifugation-cleared seminal plasma diluted 1 + 1 with PBS ( as described above ) , 20 ng/ml ΔNΔC-VEGF-C ( Kärpänen et al . , 2006 ) or equimolar amounts of N-terminally truncated VEGF-Cs ( adjusted after quantification of VEGF-C levels in conditioned supernatant after transient transfection of CHO cells ) to detect phosphorylation of VEGFR-3 and VEGFR-2 . Then , the cells were washed twice with ice-cold PBS , lysed with modified RIPA buffer ( 50 mM Tris-HCl pH 8 , 0 . 5% NP-40 , 0 . 5% Triton X-100 , EDTA-free protease inhibitor cocktail ( cOmplete , Roche , Pleasanton , CA ) , 0 . 1 mM PMSF , 1 mM Na3VO4 , and 1 mM NaF ) . VEGFR-3 and VEGFR-2 were precipitated from the cell lysate using Strep-Tactin Sepharose ( IBA , Göttingen , Germany ) for strep-tagged VEGFR-2/–3 or immunoprecipitated using protein A Sepharose ( PAS ) and anti-VEGFR-3 ( clone 9D9F9 , Dumont et al . , 1998 ) or anti-VEGFR-2 ( AF357 , R and D Systems ) , washed three times with PBS/0 . 05% Tween-20/1 mM Na3VO4 and eluted with 2x Laemmli buffer and analyzed by SDS-PAGE/Western blot using the phospho-tyrosine-specific antibody 4G10 ( Merck/Millipore , Darmstadt , Germany , 1:5000 ) . Membranes were stripped using Re-Blot plus strong solution ( Merck/Millipore ) and re-probed with HRP-conjugated Strep-Tactin ( IBA , 1:100000 ) , anti-VEGFR-3 ( 9D9F9 ) or anti-VEGFR-2 ( AF357 ) to verify equal loading . 7 ml of filter-sterilized saliva collected from volunteers among the authors in agreement with local regulations were diluted 1 + 2 with running buffer ( 20 mM sodium acetate , pH 4 . 67 ) and loaded onto a MonoS column ( GE Healthcare ) . After washing with running buffer , elution was performed with a linear 0–1M NaCl gradient and 1 ml fractions were collected . 20 µl of each fraction were diluted 1 + 4 with running buffer and 1 . 3 µg of pro-VEGF-C was added . After 36 hr incubation at 37°C , a Ba/F3-VEGFR-3/EpoR assay was performed with the samples . Amino acid sequences of 40 VEGF-C orthologs representing all major vertebrate groups ( fish , amphibians , reptiles , birds , mammals ) were retrieved via a blastp search against human VEGF-C ( UniProtKB P49767 ) . To analyze clade-specific differences in the sequence context of the VEGF-C-activating cleavage , the sequences were truncated to include only sequences corresponding to human VEGF-C amino acids 55 to 228 ( i . e . from the center of the N-terminal propeptide to the end of the VEGF homology domain ) . Alignment was performed with m_coffee ( Wallace et al . , 2006 ) and the sequences attached to the tip nodes of a phylogenetic species tree generated by opentree ( Hinchliff et al . , 2015 ) . The results were rendered with the ETE toolkit ( Huerta-Cepas et al . , 2016 ) . A Python script of the complete workflow is available from GitHub ( Jeltsch , 2018; copy archived at https://github . com/elifesciences-publications/VEGFC ) . Six bands , with identical replicates , were cut from a Coomassie-stained SDS-PAGE gel . Samples were in-gel digested according to the standard protocols and analyzed by LC-ESI-MS/MS using the LTQ Orbitrap Velos Pro mass spectrometer ( Thermo Fisher Scientific ) . The data files were searched for protein identification using Proteome Discoverer 1 . 4 software ( Thermo Fisher Scientific ) connected to a server running Mascot 2 . 4 . 1 ( Matrix Science , Boston , MA ) . Data were searched against the SwissProt database ( release 2014_01 ) . The following search parameters were used: type of search - MS/MS Ion Search , taxonomy - human , enzyme - trypsin , fixed modifications - carbamidomethyl ( C ) , variable modifications - oxidation ( M ) , mass values - monoisotopic , peptide mass tolerance - ± 5 ppm , fragment mass tolerance - ± 0 . 5 Da , max missed cleavages - 1 , instrument type - ESI-TRAP . Only proteins assigned at least with two unique peptides were accepted . The pMX-hCCBE1-StrIII construct has been described before ( Jeltsch et al . , 2014 ) . The S2 cell-expression vector pMT-Ex-VEGF-C-DMH was generated by deleting the 51 nucleotides coding for amino acids 103 to 119 of VEGF-C from pMT-Ex-ΔNΔC-VEGF-C-H6 , a modified pMT/BiP/V5-His C vector ( Invitrogen/Thermo Fisher Scientific ) , expressing mature VEGF-C ( Kärpänen et al . , 2006 ) . pMT-hygro-BiPSP-hVEGF-C-FL ( for the production of untagged pro-VEGF-C ) was generated by PCR-amplification of sequences corresponding to amino acids 32–419 of VEGF-C and cloning of the product into BglII-opened pMT-BiPV5HisC-hygro , another derivative of pMT/BiP/V5-His C , in which the 260 bp SapI-AccI fragment had been replaced by the SapI-AccI hygromycin expression cassette from pCoHygro ( Invitrogen/Thermo Fisher Scientific ) . pSecTagI-IgKSP-ΔNΔC-hVEGF-C-H6 ( the mammalian vector expressing mature VEGF-C corresponding to VEGF-C activated by plasmin cleavage between VEGF-C amino acids 102 and 103 ) was constructed by inserting the BamHI/BclI-cut VEGF-C PCR amplification product of primers 5’-GATGCTCGAGGATCCGACAGAAGAGACTATAAAATTTGC-3’ and 5’-GCATGATCACAGTTTAGACATGC-3’ into the BamHI-opened pMosaic vector ( Jeltsch et al . , 2006 ) . The cDNAs coding for N-terminally truncated VEGF-C ( corresponding to mature VEGF-C forms as activated by KLK3 cleavage , ADAMTS3 cleavage , and plasmin cleavage between amino acid residues 127 and 128 ) were PCR amplified from pSecTagI-IgKSP-ΔNΔC-hVEGF-C-H6 using specific forward primers ( 5’-TCCG GATCCGGATCCAAATACAGAGATCTTGAAAAGTATTGATAATGAGTGG-3’; 5’-TC CGGATCCGGATCCAGCACATTATAATACAGAGATCTTGAAAAGTATTG-3’; and 5’-TCCGGATCCGGATCCAAAGACTCAATGCATGCCACG-3’ ) and the same reverse primer ( 5’-ACCTACTCAGACAATGCGATGC-3’ ) , and subcloned into pSecTagI-IgKSP-ΔNΔC-hVEGF-C-H6 as BamHI-EcoRI fragments . The DMH/CatD form and C156S mutant of VEGF-C were subcloned in the same fashion from pMT-Ex-VEGF-C-DMH and pREP7-VEGF-C-C156S ( Joukov et al . , 1998 ) into the same vector ( using forward primers 5’-CGGATCCAAAAAGTATTGATAATGAGTGGAGA-3’ and 5’-GCGGATCCGACAGAAGAGACTATAAAA-3’ and reverse primer 5’-GGAATTCAATGATGATGATGGTGATGCAGTTTAGACATGC-3’ ) . The shuttle vectors to produce pro-VEGF-D ( pFB1-melSP-hVEGF-D-FL-H6 ) and a mature form of VEGF-D ( pFB1-melSP-ΔNΔC-hVEGF-D-H6 ) with the baculovirus system were generated by restriction-cloning the BamHI/HindIII-fragments of the PCR products of primers 5’-TGCGGATCCCTCCAGTAATGAACATGGACCAGTGAAGCGATC-3’ and 5’-GACAAGCTTAATGATGATGATGGTGATGAGGATTCTTTCGGCTGTGGGGC-3’ ( for pro-VEGF-D ) and 5’-TGCGGATCCGTCAGCATCCCATCGGTCCACTAGGTTTG-3’ and 5’-GACAAGCTTAATGATGATGATGGTGATGGGGGGCTGTTGGCAAGCACTTAC-3’ ( for mature VEGF-D ) into a modified pFASTBAC1 vector ( Gerhardt et al . , 2003 ) . The sequence coding for the Immunoglobulin Kappa signal peptide was amplified using forward primer 5’-CTAAAAGCTGCGGAATTGTACCCGCGGCCGCTAGCGCCACCATGGAGACAGAC-3’ and reverse primer 5’-GTCACCAGTGGAACCTGG-3’ and the VEGF-C CDS was amplified using forward primer 5’-CTGCTCTGGGTTCCAGGTTCCACTGGTGACAAAAGTATTGATAATGAGTGGAGAAAGAC-3’ and reverse primer 5’-AAATTTTGTAATCCAGAGGTTGATTATCGACGCGTTCAACGTCTAATAATGGAATGAACT-3’ . Both fragments were assembled into a MluI- and NheI-opened and CIPped psubCAG-WPRE vector ( Weltner et al . , 2012 ) resulting in psubCAG-WPRE-IgKSP-ΔNΔC-hVEGF-C-CATD . psubCAG-WPRE-IgKSP-ΔNΔC-hVEGF-C-KLK3 was assembled as above , but the reverse primer for the Immunoglobulin Kappa signal peptide CDS amplification was replaced by 5’-TTATCAATACTTTTCAAGATCTCTGTATTGTCACCAGTGGAACCTGG-3’ and the forward primer for VEGF-C CDS amplification by 5’-CAATACAGAGATCTTGAAAAGTATTGATAATG-3’ . AAV9s ( dose of 4 × 1010 in 40 µl ) encoding negative control , positive control ( ADAMTS3-cleaved form of VEGF-C ) , KLK3-cleaved form of VEGF-C ( KLK3-form ) and Cathepsin D-cleaved form of VEGF-C ( CATD-form ) were injected into the Tibialis anterior ( TA ) muscles of C57Bl/6JRccHsd ( Envigo Harlan ) female mice . Mice were sacrificed 3 weeks after transduction and the tibialis muscles were harvested . All animal experiments carried out in this study were performed according to guidelines and regulations approved by the National Board for Animal Experiments of the Provincial State Office of Southern Finland . Mouse tibialis anterior muscle samples were embedded into Tissue-Tek OCT and frozen in liquid nitrogen-cooled isopentane . 10µm-sections were stained for the lymphatic marker Lyve-1 ( Karkkainen et al . , 2004 , 1:1000 ) and blood vascular marker CD31 ( BD Biosciences , San Jose , CA , 1:500 ) , followed by Alexa-conjugated secondary antibodies ( Molecular Probes/Thermo Fisher Scientific ) . Nuclei were stained with DAPI with VECTASHIELD ( Vector Laboratories , Burlingame , CA ) . Fluorescent images were obtained with an Axio Imager Z2 upright epifluorescence microscope ( Carl Zeiss AG , Oberkochen , Germany ) . Images were processed and analysed with Fiji ImageJ ( NIH ) . Muscle tissues were lysed using Trisure reagent ( Bioline , London , UK ) and the RNA was extracted with Nucleospin RNA II kit ( Macherey-Nagel , Düren , Germany ) . cDNA was synthesized with High-Capacity cDNA Reverse Transcription Kits ( Applied Biosystems/Thermo Fisher Scientific ) using 1 µg RNA . qRT-PCR was performed with SensiFast SYBR No-ROX Kit ( Bioline ) . All data were normalized to GAPDH . Relative gene expression levels were calculated using the 2-∆∆Ct method . VEGF-C ( fwd 5’-TGAACACCAGCACGAGCTAC-3’ , rev 5’-TCGGCAGGAAGTGTGATTGG-3’ ) and mGAPDH ( fwd 5’-ACAACTTTGGCATTGTGGAA-3’ , rev 5’-GATGCAGGGATGATGTTCTG-3’ ) primers were used for the real time PCR . Data are presented as mean ± SD or mean ± SEM . Data were analysed using GraphPad Prism statistical analysis software ( Version 8 ) . Data analysis details are mentioned in the respective figure legends . | The lymphatic system is composed of networks of vessels that drain fluids from the body’s tissues and filter it back into the blood . Growing these vessels depends on a factor known as VEGF-C , which is released in an inactive form and must be cut by enzymes before it can work . One enzyme that is known to activate the VEGF-C signal when the early embryo is developing is ADAMTS3 . If this signal fails to switch on this can result in a condition known as lymphedema – whereby problems in the lymphatic system cause tissues to swell due to insufficient drainage . However , it is unknown whether the VEGF-C signal can be activated by enzymes other than ADAMTS3 . To investigate this Jha , Rauniyar et al . tested a specific family of proteins commonly found in the human prostate , which have previously been predicted to act on VEGF-C . This revealed that the lymphatic vessel growth factor can also be activated by an enzyme found in seminal fluid called prostate specific antigen , or PSA for short . To see if enzymes in other bodily fluids could switch on VEGF-C , different components of human saliva were separated and tested to see which could cut inactive VEGF-C . This showed that VEGF-C could be converted to an active form by another enzyme called cathepsin D . Unexpectedly , Jha , Rauniyar et al . found that VEGF-C was also present in semen . For conception to occur PSA must liquify the semen following ejaculation . It was discovered that PSA activates VEGF-C just as the semen starts to liquify , suggesting that the lymphatic vessel growth factor might also play an important role in reproduction . In addition to VEGF-C , both PSA and cathepsin D were found to activate another growth factor called VEGF-D , which has an unknown role in the human body . VEGF-C helps the spread of tumors , and blocking the two enzymes that activate this growth factor may be a new therapeutic approach for cancer . However , more work is needed to validate which types of tumor , if any , use these enzymes to activate VEGF-C . In addition , understanding the relationship between PSA and VEGF-C could help improve our knowledge of human reproduction . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cancer",
"biology"
] | 2019 | KLK3/PSA and cathepsin D activate VEGF-C and VEGF-D |
While more than 70 genes have been linked to deafness , most of which are expressed in mechanosensory hair cells of the inner ear , a challenge has been to link these genes into molecular pathways . One example is Myo7a ( myosin VIIA ) , in which deafness mutations affect the development and function of the mechanically sensitive stereocilia of hair cells . We describe here a procedure for the isolation of low-abundance protein complexes from stereocilia membrane fractions . Using this procedure , combined with identification and quantitation of proteins with mass spectrometry , we demonstrate that MYO7A forms a complex with PDZD7 , a paralog of USH1C and DFNB31 . MYO7A and PDZD7 interact in tissue-culture cells , and co-localize to the ankle-link region of stereocilia in wild-type but not Myo7a mutant mice . Our data thus describe a new paradigm for the interrogation of low-abundance protein complexes in hair cell stereocilia and establish an unanticipated link between MYO7A and PDZD7 .
Hair cells are the sensory cells of the inner ear; they transduce mechanical signals evoked by sound and head movement and relay these signals to the central nervous system . The mechanically sensitive organelle responsible for mechanotransduction is the hair bundle , a cluster of ~100 actin-filled stereocilia that protrude apically from a hair cell ( Gillespie and Müller , 2009 ) . Because stereocilia bend at their bases yet are coupled together with elastic linkages , the bundle pivots as a whole when exposed to mechanical stimuli . One special type of linkage , the tip link , couples bundle pivoting to electrical excitation of the hair cell . The tip link is composed of a dimer of CDH23 ( cadherin 23 ) molecules interacting with a dimer of PCDH15 ( protocadherin 15 ) molecules ( Kazmierczak et al . , 2007 ) , and is coupled to the mechanically sensitive transduction channel . Another class of linkage , the ankle link , couples stereocilia together at their bases , transiently in auditory hair cells but persistently in vestibular hair cells ( Goodyear and Richardson , 1999; Adato et al . , 2005; McGee et al . , 2006; Michalski et al . , 2007 ) . Because many genes encoding proteins known to be associated with these links are affected by mutations that cause deafness , it is of great interest how these links couple with proteins within the stereocilia . Myosin VIIA ( MYO7A ) was the first protein involved in hair-cell mechanotransduction to be identified by genetics ( Gibson et al . , 1995; Weil et al . , 1996 ) . While MYO7A has been implicated in ankle-link positioning by its location in frog stereocilia ( Hasson et al . , 1997 ) and by mislocalization of ankle-link proteins in Myo7a-null mice ( Michalski et al . , 2007 ) , more precise localization experiments suggested that MYO7A also participates in mechanotransduction ( Grati and Kachar , 2011 ) . These differing roles could be reconciled if MYO7A forms several different protein complexes . The tail of MYO7A has two tandem FERM-MyTH4 domains and one SH3 domain , each of which might interact with stereocilia proteins . Biochemical and structural studies have indicated that MYO7A interacts both with PCDH15 ( Senften et al . , 2006 ) and with a complex including USH1C , USH1G , and CDH23 ( Boëda et al . , 2002; Wu et al . , 2011 ) . Preliminary evidence suggests that MYO7A interacts with ADGRV1 , which was formerly known as GPR98 and VLGR1 ( Hamann et al . , 2015 ) , USH2A ( usherin ) , and DFNB31 ( whirlin ) ( Delprat et al . , 2005; Michalski et al . , 2007 ) . No interactions of MYO7A have been reported with PDZD7 , another member of the ankle-link complex ( Grati et al . , 2012; Zou et al . , 2014 ) , and how MYO7A positions the ankle-link complex remains unclear . While genetic methods have been successful in identifying essential stereocilia proteins , the approach can miss proteins that are more widely expressed or that can be compensated for by the expression of a paralog . To determine all the proteins present in hair bundles , we have developed biochemical methods for their characterization , coupling purification of bundles using the twist-off method with detection of proteins by mass spectrometry ( Gillespie and Hudspeth , 1991; Shin et al . , 2007 , 2013 ) . Because of the fine dissection and tissue manipulation required , however , the twist-off method is limited in its throughput and is less ideal for biochemical experiments examining rare protein-protein complexes in bundles . We therefore developed a high-throughput method for stereocilia-membrane enrichment using the monoclonal antibody D10 , which allowed us to prepare large amounts of solubilized inner-ear material that is suitable for immunoaffinity purification of protein complexes from stereocilia detergent extracts . We initially used this approach to determine which proteins interact with MYO7A in the inner ear , using the monoclonal antibody 138-1 to show that MYO7A interacts tightly with several scaffolding proteins , notably PDZD7 . By immunoprecipitating expressed proteins in tissue-culture cells , we confirmed that the interaction of PDZD7 and MYO7A is direct . Moreover , MYO7A is also in a protein complex including ADGRV1 , which likely also includes PDZD7 . We also show that USH1C , USH1G , and CDH23 co-precipitate from chick stereocilia with MYO7A , confirming the presence in the tissue of a complex that had only been suggested by genetics and recombinant-protein expression . Because other scaffolding proteins co-precipitate with MYO7A , these experiments show not only that MYO7A contributes to the tip-link and ankle-link complexes , but also suggest that MYO7A participates in other stereocilia protein complexes . Finally , the experiments indicate that the coupled stereocilia-membrane purification and immunoaffinity isolation methods will be useful for isolating even very rare protein complexes from stereocilia .
We enriched stereocilia membranes from chicks using the D10 preparation , named for the monoclonal antibody that is critical for the procedure ( Figure 1 ) . D10 recognizes PTPRQ ( Goodyear et al . , 2003 ) , the most abundant transmembrane protein of hair bundles ( Shin et al . , 2013 ) . As in our previous experiments coupling stereocilia isolation with mass spectrometry ( Shin et al . , 2007 , 2013 ) , we used E19-E21 chicks because of chicks’ availability , moderate cost , sequenced genome , and near-mature inner ear . In most cases , we treated the exposed surfaces of dissected inner ears with DTSSP , a membrane-impermeable , reversible crosslinker with two N-hydroxysuccinimide groups coupled with an 12 Å spacer that includes a disulfide bond ( Staros , 1982 , 1988 ) ; this treatment stabilizes complexes of membrane proteins ( Corgiat et al . , 2014 ) , presumably including ankle links and tip links . DTSSP treatment should not affect intracellular protein complexes , however , and is reversed by reduction . If their binding affinity is sufficiently high , cytoplasmic proteins in association with membrane proteins should still be detected , however , even without covalent crosslinking . Because one investigator can comfortably dissect only ~200 ears per day , to allow larger preparations ( ≥1000 ears ) , we froze tissues daily after quenching the DTSSP reaction . 10 . 7554/eLife . 18312 . 003Figure 1 . D10 chick stereocilia-membrane enrichment procedure . ( A ) Flow chart for stereocilia membrane enrichment . Bold red lettering highlights the principal steps of the procedure . ( B ) Imaging of M3 fraction of D10 beads; samples were labeled with phalloidin ( magenta , for actin ) , DAPI ( green , for nuclei ) , and anti-mouse IgG ( white , for D10 antibody on beads ) . Aggregation of beads and binding of nuclei ( and other contaminants ) limits the enrichment gained with this step . ( C ) SDS-PAGE analysis of enrichment fractions; gel stained with silver . Note that fraction S7 was loaded in a second lane with 4x more material . ( D ) Immunoblot analysis of purification fractions . For actin ( ACT ) , lanes 1–10 had 0 . 01 ear-equivalents loaded and lane 11 had 0 . 2 ears . For all other immunoblots , lanes 1–10 had 0 . 05 ear-equivalents loaded and lane 11 had 0 . 2 ears . Other than actin , all proteins are referred to by their official gene symbols . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 00310 . 7554/eLife . 18312 . 004Figure 1—figure supplement 1 . Protein recovery with D10 and twist-off methods . Samples ( 0 . 3 and 1 . 0 ear-equivalents ) of hair bundles purified by twist-off and stereocilia membranes purified by the D10 purification were subjected to immunoblotting with antibodies against actin , ATP2B2 , and ATP1A1 . Aliquots of whole utricle ( 0 . 05–1 . 2 utricles ) were run on the same gel for calibration . Fiji was used to measure the immunoblot intensity for each protein in each lane , and the calibration curve was used to determine the amount of each protein recovered in twist-offs or D10 preps relative to the amount in whole utricle . Linear regression was used to determine the recovery ( in utricle-equivalents ) per ear of starting material . Because seven sensory epithelia were sampled , actin and ATP2B2 were present in the D10 prep at levels considerably higher per ear than was seen in the whole utricles . ATP1A1 was not detected in twist-offs , and while substantial amounts were detected in the D10 prep , they were relatively low compared to the amount present in a whole utricle . Mean ± standard error for the regression fit are plotted . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 004 After thawing the inner-ear tissue , we used conventional biochemical procedures to enrich stereocilia membranes . We used antibodies to ATP2B2 , the plasma-membrane Ca2+ pump of stereocilia ( Dumont et al . , 2001 ) , to locate fractions containing stereocilia membranes ( Figure 1D ) . Tissues were homogenized , subjected to differential centrifugation to remove nuclei ( yielding fraction S1; Figure 1A ) , and dense organelles were sedimented on top of a sucrose pad ) . In pilot experiments , we subjected the post-nuclear supernatant to fractionation using discontinuous equilibrium sucrose gradients; most of the ATP2B2 and hence the stereocilia membranes sedimented to the top of the 1 . 8 M sucrose layer . In large-scale experiments , we layered the post-nuclear supernatant on a cushion of 2 . 2 M sucrose , centrifuged to equilibrium , and collected material at the interface ( fraction P2 ) . Stereocilia membranes were captured using magnetic beads coupled with D10 ( fraction M3 ) . We labeled beads with the antibody at high density , which facilitated efficient binding to stereocilia and enrichment with a magnet . Because stereocilia were present at extremely low concentrations in the extract , however , there was a large excess of antibody-coupled beads over stereocilia membranes . As a consequence , substantial amounts of non-stereocilia protein bound nonspecifically to the beads . While phalloidin staining showed that many of the D10 beads were decorated with stereocilia ( Figure 1B ) , nuclei and other contaminants were also associated with the beads , presumably nonspecifically . Membranes were eluted using sonication ( fraction S4 ) ; they were then washed several times prior to detergent extraction , yielding enriched stereocilia membranes ( fraction P6 ) . We solubilized protein complexes using RIPA buffer , a stringent buffer that contains both SDS and deoxycholate; the final solution , fraction S7 , was used for subsequent immunoaffinity purification steps . We examined the purification fractions using silver-stained SDS-PAGE gels ( Figure 1C ) and immunoblotting of marker proteins ( Figure 1D ) . Very little total protein bound to the D10 beads ( M3 ) , and cytosolic ( HSPA5 , PITPNA1 ) , nuclear ( LMNB1 ) , mitochondrial ( MDH2 ) , and plasma-membrane ( ATP1A1 ) proteins were largely absent from D10 eluates ( S4 ) . A modest amount of the endoplasmic reticulum marker CANX was visible in the final RIPA-solubilized fraction ( S7 ) , however , as was the supporting-cell antigen PTPRJ . Nearly all of the ATP2B2 was detected in S7 , however , showing that stereocilia membranes were efficiently solubilized . In quantitative immunoblotting experiments , we found that the total amount of actin and ATP2B2 per ear , as compared to the amount in a single utricle , was much higher in the D10 stereocilia membrane preparation than in the twist-off hair-bundle preparation ( Figure 1—figure supplement 1 ) . This elevated level of stereocilia protein derived in part from sampling seven organs ( cochlea , lagena , utricle , saccule , and three semicircular canals ) rather than just one . Contaminating ATP1A1 from the basolateral membrane was detectable in the D10 preparation , but at modest levels compared to that in one utricle ( Figure 1—figure supplement 1 ) . The D10 and twist-off preparations were compared in Table 1; note that a major advantage of the D10 preparation is its high throughput . 10 . 7554/eLife . 18312 . 005Table 1 . Comparison of stereocilia purification methods . The D10 preparation enriches stereocilia membranes , and therefore is not directly comparable to the twist-off method , which isolates the entire hair bundle . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 005ParameterTwist-offD10Throughput<40 ears/person/day>200 ears/person/dayStereocilia purity>90%-Recovery~1/3>1/2Organs sampled17Stereocilia protein recovered1x10xStereocilia protein per day1x50xSubsequent purificationChallengingEasy We used shotgun ( data-dependent acquisition , DDA ) and targeted ( parallel reaction monitoring , PRM ) mass spectrometry to determine enrichment of key proteins in the stereocilia membrane preparation ( Figure 2A–E ) . In fractions from two independent preparations , each of 1000 chick ears , the total amount of protein ( Table 2 ) decreased from ~110 mg in S1 ( post-nuclear supernatant ) to ~1 mg in S7 ( RIPA solubilized membranes ) , consistent with the silver-stain analysis . We analyzed two technical replicates each of the two preparations by shotgun mass spectrometry ( four runs per fraction ) , and used Andromeda peptide searching and MaxQuant protein assembly and quantitation to determine the protein composition of each fraction . Equal amounts of protein were analyzed , and over 3000 proteins were detected in each fraction . Proteins were quantified using the intensity-based absolute quantification ( iBAQ ) method ( Schwanhäusser et al . , 2011 ) . To estimate the molar fraction of each protein in each sample , we converted iBAQ to relative iBAQ ( riBAQ ) , which is the iBAQ for a given protein divided by the summed iBAQ for all the proteins of the fraction after exclusion of contaminants ( Shin et al . , 2013; Krey et al . , 2014 ) . All data are deposited at ProteomeXchange ( http://www . proteomexchange . org ) with the identifier PXD004222 , and the analysis is tabulated in Figure 2—source data 1 . 10 . 7554/eLife . 18312 . 006Figure 2 . Mass spectrometry analysis of D10 stereocilia-membrane enrichment . ( A ) Shotgun mass-spectrometry quantitation of 3313 proteins detected in at least three of five purification fractions where enrichment of stereocilia membranes was expected . The slope of the purification fraction ( assuming an interval of 1 ) vs . log riBAQ was calculated for each protein , and the top 25 proteins with the steepest positive slopes ( 'enrichment slope' ) were highlighted . Color indicates the slope steepness . For this analysis , ATP2B2 was split out of the group also containing ATP2B1 and ATP2B4 group; data corresponding both ATP2B2 and PTPRQ were emphasized to illustrate the effectiveness of the stereocilia-membrane enrichment . The number of proteins detected in each fraction was also indicated . ( B ) Gene ontology analysis with DAVID of the S7 ( RIPA-soluble ) fraction using S1 ( PNS ) as background . The top ten cellular component terms enriched are indicated; the log of the P-value ( adjusted for false-discovery rate , FDR ) is plotted . The inset shows a magnification of the 'plasma membrane' term , which had a substantial increase in significance in the S7 fraction . ( C ) Relationship between enrichment slope and frequency of Ensembl-annotated transmembrane helicies . Each protein identified was assigned 1 if they were annotated as having a transmembrane helix , and 0 if not . All proteins were sorted according to enrichment slope , and an average was calculated every 50 proteins . ( D ) Stereocilia membrane protein enrichment . Displayed are bundle-enriched ( >three-fold ) proteins from Wilmarth et al . ( 2014 ) that were annotated as containing transmembrane domains , plus SLC34A2 , SLC9A9 , and SLC9A6 . ND , not detected . ( E ) Clustering analysis of shotgun mass-spectrometry data using mclust . After data were log-transformed , they were normalized within a row , subtracting the row mean and dividing by the row standard deviation ( thus the resulting mean was 0 and the standard deviation was 1 ) . The best solution returned 25 clusters , which were ordered here by the enrichment slope ( only the five fractions used in A were used to calculate the slope ) . PTPRQ , ATP2B2 , and XIRP2 were all in the cluster with the steepest enrichment slope . ( F–I ) Targeted proteomics analysis of enrichment fractions . ( F ) Chromatograms of SAPLHILTDEDAPSSPPESLSVK peptide from PTPRQ detected in S1 ( dark gray ) and S7 ( orange ) fractions; inset shows magnified S1 chromatogram . Sum of 12 daughter ions . Arrows indicate where multiple MS2 spectra matched to PTPRQ were acquired . ( G ) Example MS2 spectrum from S7 sample; multiple peaks matched to predicted PTPRQ peptide fragment ions . ( H ) Targeted proteomics analysis of enrichment fractions . The mean of the total intensity area of 4–8 peptides per protein is indicated . ( I ) Same data as G , except area relative to S1 is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 00610 . 7554/eLife . 18312 . 007Figure 2—source data 1 . Analysis of the shotgun proteomics experiments characterizing the protein composition of the D10 stereocilia-membrane purification from the chick inner ear . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 00710 . 7554/eLife . 18312 . 008Figure 2—source data 2 . Output of the mclust analysis of the D10 purification fractions . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 00810 . 7554/eLife . 18312 . 009Table 2 . D10 purification . Protein assays and PTPRQ targeted proteomics assays were carried out on two preparations , each of ~1000 chick inner ears . Total PTPRQ PRM intensity was determined by multiplying the PTPRQ intensity from PRM targeted-proteomics experiments ( per µg protein ) by the total amount of protein ( in µg ) in each sample . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 009FractionDescriptionTotal protein ( µg ) , ± rangePTPRQ PRM intensity , totalPTPRQ PRM intensity , fraction of PNSS1Post nuclear supernatant109 , 000 ± 4 , 0002 . 63 ± 0 . 40 × 1010100 ± 15%S27000 rpm supernatant100 , 000 ± 4 , 0000 . 35 ± 0 . 0513 ± 3%P2Dense membranes36 , 400 ± 7002 . 56 ± 0 . 1797 ± 16%M3D10 bound2680 ± 900 . 34 ± 0 . 0413 ± 2%S6Cytoplasm29 , 200 ± 1000 . 32 ± 0 . 0712 ± 2%P6Membranes984 ± 10 . 28 ± 0 . 0111 ± 2%S7RIPA-soluble1037 ± 01 . 13 ± 0 . 1143 ± 8% Previous proteomics experiments indicated that PTPRQ and ATP2B2 are the most abundant membrane proteins of chick vestibular stereocilia ( Shin et al . , 2013 ) ; other notable membrane proteins enriched in stereocilia include SLC9A6 and SLC9A9 ( Hill et al . , 2006a ) ; ATP8B1 , NPTN , STARD10 , and EFR3A ( Zhao et al . , 2012 ) ; CDH23 , PTPRF , LGALS1 , C2CD2L , KIAA1211 , and KIAA1549 ( Shin et al . , 2013 ) , as well as SLC34A2 , a Na+-Pi transporter enriched in in cochlear stereocilia ( Avenarius et al . , 2014 ) . With the exception of KIAA1211 , which was not detected in the shotgun experiments , all of these stereocilia membrane proteins were enriched as the preparation proceeded or were found only in fraction S7 ( Figure 2D ) . To determine in an unbiased way whether stereocilia membranes increased in relative abundance , we calculated a slope for each protein of the preparation across the fractions where we expected enrichment ( S1 , P2 , M3 , P6 , and S7 ) . Of the 3036 proteins detected in at least four of the six fractions , PTPRQ had the largest 'enrichment slope' ( Figure 2A ) . The MYO3A-MYO3B group , with a somewhat steeper slope , was only detected in three fractions . SLC34A2 was highly enriched , as was XIRP2; while the function of the latter protein is poorly understood , it does localize to the membrane-cytoskeleton interface in stereocilia and could bind membranes itself ( Francis et al . , 2015 ) . Many other proteins enriched in the D10 preparation have annotated transmembrane domains , although some of these proteins likely derive from membranes other than those of stereocilia . Other proteins were known stereocilia proteins ( ESPN , ANKRD24 , MYO1H ) . Although ATP2B2 is part of a protein group also containing ATP2B1 and ATP2B4 , which are not located in stereocilia ( Dumont et al . , 2001 ) , we estimated ATP2B2’s contribution to the ATP2B1/2/4 group riBAQ value for each fraction; we divided the intensities from peptides only found in each isoform by the sum of all isoform-specific peptides , generating the fractional contribution for each isoform . After making this adjustment , ATP2B2 was also one of the most enriched proteins ( Figure 2A ) . Gene ontology analysis ( Ashburner et al . , 2000; Gene Ontology Consortium , 2015 ) using DAVID ( Huang et al . , 2009b , 2009a ) indicated that the ten most enriched GO terms in S7 relative to S1 referred to membranes , mitochondria , or plasma membrane ( Figure 2B ) ; the latter term was most highly enriched in the last enrichment fraction ( p=3 × 10–6 ) . Similarly , 'membrane' is the most highly enriched ( p=3 × 10–37 ) keyword from the Protein Information Resource ( Barker et al . , 2000 ) . In addition , the steeper the enrichment slope , the more likely a protein was annotated as containing a transmembrane helix ( Figure 2C ) . Taken together , our analysis shows that the D10 preparation substantially enriches membranes , especially stereocilia membranes , reducing protein complexity so that additional affinity purification steps can be used to identify protein complexes of stereocilia membranes . The most abundant membrane proteins in S7 were either from the plasma membrane ( e . g . , ATP1A1 , at riBAQ = 1 . 1 × 10–2 ) or mitochondrial membranes ( e . g . , ATP5A1 , at riBAQ = 1 . 1 × 10–2 ) . In comparison , ATP2B2 , marking stereocilia membranes , was present at riBAQ = 0 . 7 × 10–2 . In frog stereocilia , ATP2B2 and ATP1A1 each have membrane densities of ~3000 µm−2 ( Yamoah et al . , 1998; Burnham and Stirling , 1984 ) ; in rat liver mitochondria , ATP5A1 has a density of ~7500 µm−2 ( Schwerzmann et al . , 1986 ) . Taken together , these data suggest that stereocilia membranes account for ~5% of the solubilized membrane material in S7 . We used cluster analysis with mclust ( Fraley and Raftery , 2002 ) , a model-based hierarchical clustering algorithm that allows estimation of the best number of clusters for a dataset , to determine which proteins shared behavior across the stereocilia enrichment procedure . Using only proteins that were reproducibly detected across the preparation , mclust indicated that 25 clusters were appropriate for the normalized data ( Figure 2E; Figure 2—source data 2 ) . To determine which of these clusters were preferentially enriched in the D10 preparation , we applied a linear fit to the data from the sequential enrichment fractions ( S1 , P1 , M3 , P6 , and S7 ) . Cluster 8 had the greatest enrichment slope , and not surprisingly , it contained both PTPRQ and ATP2B2 , as well as XIRP2 ( Figure 2E ) . We also used targeted proteomics to confirm that proteins of interest were enriched in the D10 preparation ( Figure 2F–I ) . By directly examining the intensity associated with a known peptide , targeted proteomics usually offers substantially greater sensitivity and selectivity than shotgun proteomics with label-free quantitation ( Picotti and Aebersold , 2012 ) . We used PRM ( Gallien and Domon , 2015; Ronsein et al . , 2015 ) with an Orbitrap Fusion mass spectrometer to measure two to four peptides for each protein , monitoring at least two daughter ions per peptide . To confirm that the assays measured the peptide of interest , we spiked in heavy-labeled standard peptides; we also matched MS2 spectra collected during peaks to the protein of interest . Using this approach , we determined that PTPRQ was enriched between S1 and S7 by around 50-fold ( Figure 2 ) ; the enrichment of ATP2B2 was less , perhaps because substantial amounts were located in hair-cell somas . We sought to identify tight membrane complexes containing MYO7A . MYO7A is known to interact with protein complexes located at stereocilia membranes , including those of tip links and ankle links , and is associated with membranes in other cells ( Soni et al . , 2005 ) . While MYO7A is also present in hair cell somas , a rough estimate based on the amount in chick hair bundles ( Table 3 ) indicates that purification would require an enrichment of approximately 106-fold from dissected inner ears . The 102-fold purification afforded by the stereocilia-membrane enrichment indicates that additional affinity purification steps are required to bring MYO7A complex abundance to the level where it can be definitively detected in shotgun mass-spectrometry experiments . 10 . 7554/eLife . 18312 . 010Table 3 . Identification of MYO7A-binding proteins in twist-off-purified hair bundles . The presented data , as well as technical aspects of the chick and mouse hair-bundle purification and shotgun mass spectrometry protein detection and quantitation , have been described previously ( Krey et al . , 2015; Shin et al . , 2013; Wilmarth et al . , 2015 ) . BUN only , only found in hair bundle samples . UTR only , only found in whole utricle samples . ND , not detected . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 010ProteinChick molecules per stereociliumChick enrichmentMouse ( P23 ) molecules per stereociliumMouse ( P23 ) enrichmentMYO7A7961 . 6x22928xMYO693002 . 2x13052 . 6xGIPC34943 . 5x5512xMYO1C12617xNDUTR onlyMYO1H9129x128406xANKRD2456122x1BUN onlyPDZD75010xNDNDSORBS1162 . 0xNDUTR onlyLMO7103 . 4xNDND We therefore used immunoaffinity methods to further enrich MYO7A after purification of stereocilia membranes ( Figure 3A ) ; although the majority of MYO7A was in the insoluble fraction ( P7 ) of the RIPA extraction of D10 membranes , a substantial amount was solubilized ( S7 ) . For the subsequent step , we used the 138-1 monoclonal antibody , which recognizes chicken MYO7A ( Soni et al . , 2005 ) . We applied the S7 RIPA extract to control beads , constructed with purified mouse IgG , then applied the unbound material to 138-1 beads . Each aliquot of beads was washed thoroughly , then eluted with SDS; eluted proteins were then analyzed by shotgun and targeted mass spectrometry . Passing the extracts over control beads allowed us to identify proteins that nonspecifically bound to the immunoaffinity reagents , many of which were abundant; these proteins were then computationally subtracted from the proteins identified in the 138-1 eluates . We carried out three independent immunoprecipitation experiments , each with ~500 ear-equivalents; one to three technical replicates from each eluate were subjected to shotgun mass spectrometry , and the results from the six shotgun runs were analyzed together . All data are deposited at ProteomeXchange with the identifier PXD004221 , and the analysis is tabulated in Figure 3—source data 1 . 10 . 7554/eLife . 18312 . 011Figure 3 . MYO7A immunoaffinity purification from D10 enriched membranes . ( A ) Flow chart describing immunoaffinity purification strategy using the 138-1 monoclonal antibody , which recognizes chicken MYO7A . Mass spectrometry was carried out on the total ( S7; RIPA-soluble fraction ) , 138-1 eluate , and control eluate . F/T , unbound material; MS/MS , tandem mass spectrometry analysis . ( B–E ) Shotgun proteomics analysis of 138-1 immunoaffinity purification . ( B ) Relative protein levels of 1719 proteins detected in at least one run of total or 138-1 eluate . Distance from the diagonal unity line ( gray dashed ) indicates enrichment by 138-1; note that MYO7A ( red symbol ) and PDZD7 ( green ) are highly enriched . ( C ) Volcano plot illustrating enrichment and statistical significance . The x-axis displays the log10 of each protein’s enrichment by 138-1 immunoprecipitation relative to the S7 starting material , while the y-axis indicates the log10 value of the FDR ( false discovery rate ) adjusted p-value for that enrichment value . Dashed line indicates significance at p=0 . 05 . ( D ) Top 32 proteins in 138-1 eluates by enrichment . Only proteins not detected in control runs and detected in at least four of six 138-1 runs are displayed . ( E ) Top 27 proteins in 138-1 eluates by stoichiometry relative to MYO7A . Only proteins detected in at least four of six 138-1 runs are displayed; proteins detected in control eluates ( e . g . , actin ) were also included . Unadjusted stoichiometry: ( 138-1 riBAQ for protein of interest ) / ( 138-1 riBAQ for MYO7A ) . Adjusted stoichiometry: ( unadjusted stoichiometry ) × ( total riBAQ for MYO7A ) / ( total riBAQ for protein of interest ) . Figure 3—figure supplement 1 displays an immunoblot analysis of the 138-1 immunoprecipitation , examining MYO7A and its Usher syndrome partners USH1C and CDH23 . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 01110 . 7554/eLife . 18312 . 012Figure 3—source data 1 . Analysis of the shotgun proteomics experiments characterizing the protein composition of the 138-1 anti-MYO7A immunoaffinity purification from the chick inner ear . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 01210 . 7554/eLife . 18312 . 013Figure 3—figure supplement 1 . Immunoprecipitation of upper tip-link density components from D10-enriched stereocilia membranes . ( A ) MYO7A ( asterisk ) . IP input was 100% . Key: IP input , ears starting total extract ( S7 ) ; C , control immunoprecipitates; 7A , 138-1 immunoprecipitates; IPs , immunoprecipitates . ( B ) USH1C ( asterisk ) . IP input was 100% . ( C ) CDH23 . IP input was 4 ear-equivalents , which corresponds to 100% ( 4 ears ) and 20% ( 19 ears ) . With image enhancements to the blot , CDH23 bands were apparent in 138-1 eluates but not in control eluates . Key: Pre-D10 , dense membranes ( P2 ) ; ears , ear-equivalents loaded; Enhanced , blot adjusted by changes to dark level and gamma . Asterisks indicate positions of high ( * ) and lower ( ** ) mass bands immunoreactive for CDH23 . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 013 To characterize the immunoaffinity purification procedure , we first investigated whether previously-described MYO7A complexes were present in the 138-1 immunoprecipitates . Although a MYO7A-USH1C-USH1G-CDH23 complex has been inferred from cell-culture and complex reconstitution experiments ( Boëda et al . , 2002; Wu et al . , 2011 ) , no evidence has been presented to date for direct association of these proteins in the inner ear . When we immunoprecipitated MYO7A from stereocilia S7 detergent extracts ( Figure 3—figure supplement 1A ) and examined eluates using immunoblotting , we detected both a significant amount of USH1C ( Figure 3—figure supplement 1B ) and a small amount of CDH23 ( Figure 3—figure supplement 1C ) in the immunoprecipitates . Because MYO7A and its interacting proteins should be highly enriched by the 138-1 antibody from the starting material , we analyzed immunoaffinity purification experiments using shotgun mass-spectrometry , comparing the relative abundance of each protein in the 138-1 eluate and in the stereocilia RIPA extract . This ratio—the immunoaffinity enrichment—will be high for MYO7A itself and for proteins that specifically coprecipitated with MYO7A . Figure 3B shows all proteins detected in the experiment , with those detected only in the 138-1 immunoprecipitate or in S7 segregated from those detected in both . For proteins detected in both fractions , the presence of a protein to the left and above the unity line ( dashed ) indicates that it is enriched; as expected , MYO7A itself was highly enriched ( ~104-fold ) . The statistical significance , adjusted for the false-discovery rate ( FDR ) , is indicated in a volcano plot ( Figure 3C ) . MYO7A and PDZD7 were not only highly enriched , but the enrichment was highly significant statistically ( p=3 × 10–8 and p=2 × 10–5 ) . We examined the proteins that were detected at least 4/6 times in the 138-1 runs and that were not detected in the control IgG eluates ( Figure 3D ) . Remarkably , most of the most highly enriched proteins have been previously detected as key hair-bundle proteins ( Shin et al . , 2013 ) ; these include myosin motors MYO3B , MYO1H , and MYO6; actin-associated proteins ESPN , LMO7 , FSCN2 , and GRXCR1; and scaffolding proteins PDZD7 and GIPC3 ( Table 3 ) . Two other scaffolding proteins , SORBS1 and SORBS2 , were also detected . Myosins apparently are not associated by binding to co-purifying actin filaments; actin was present at a low stoichiometry relative to MYO7A ( <1 mol/mol ) , indicating that the concentration of filaments was very low . Moreover , treatment of immunoprecipitated material with 5 mM Mg•ATP did not release the other myosins from MYO7A complexes . We also noted the presence of multiple calmodulin-related proteins , including MYL4 , CALML4 , MYL1 , and CALM itself; these molecules are candidate light chains for MYO7A or co-precipitating myosins . In these shotgun experiments , we did not detect USH1G or CDH23 in 138-1 samples; we did , however , detect USH1C in 138-1 eluate , albeit only in 2/6 runs . PDZD7 was highly enriched along with MYO7A; both were present in the 138-1 immunoaffinity purification at relative concentrations >1000-fold higher than in S7 ( Figure 3B–D ) . While the calculated stoichiometry of PDZD7 relative to MYO7A was low ( ~1% ) , PDZD7 was present in the stereocilia membrane extract at a far lower level than MYO7A . Indeed , a large fraction of the PDZD7 ( ~20% ) was precipitated from the extract , indicating that many PDZD7 molcules were present in MYO7A complexes ( Figure 3E ) . PDZD7 has been localized to the ankle links ( Grati et al . , 2012 ) , where it is essential for localization of protein complexes containing ADGRV1 , USH2A , and DFNB31 ( Chen et al . , 2014; Zou et al . , 2014 ) . Depending on the splice form , PDZD7 has two or three PDZ domains , and is a paralog of both USH1C and DFNB31 ( Ebermann et al . , 2010 ) . An interaction of PDZD7 with MYO7A has not been reported previously , although the interactions of MYO7A with USH1C ( Boëda et al . , 2002 ) and DFNB31 ( Delprat et al . , 2005 ) suggest that the MYO7A-PDZD7 complex is plausible . Containing five ankyrin domains , ANKRD24 is enriched in chick and mouse hair bundles ( Shin et al . , 2013; Wilmarth et al . , 2015; Krey et al . , 2015 ) ; little else is known about ANKRD24 , although it is a hair-cell-specific protein in the chick ear ( Ku et al . , 2014 ) . By shotgun mass spectrometry analysis , most of the ANKRD24 in the S7 extract was present in MYO7A complexes ( Figure 3E ) . We used targeted proteomics to confirm that proteins identified in the shotgun analysis were indeed co-immunoprecipitated with MYO7A . We developed assays not only for several proteins detected in the shotgun experiments , but also for other proteins predicted to be in MYO7A complexes at tip links and ankle links . These experiments used heavy-labeled standard peptides to confirm retention times and daughter-ion patterns , and we also matched MS2 spectra to peaks to confirm the assays’ veracity . Three biological replicates of 500 chick ears were used , with two technical replicates from each . Each peptide was analyzed in two or three of the biological replicates , yielding 4–6 total replicates for each . The analysis is tabulated in Figure 4—source data 1 , and displayed in Figure 4 . 10 . 7554/eLife . 18312 . 014Figure 4 . Targeted mass spectrometry analysis of MYO7A immunoaffinity purification from chick stereocilia . ( A–C ) Diagrams illustrating targeted analysis . ( A ) Peptides eluting from the nano-scale liquid chromatography ( NanoLC ) system are introduced into a Tribrid Fusion mass spectrometer by electrospray ( Ion source ) . Monitored peptides ( blue ) are selected within the quadrupole mass analyzer , then are fragmented using higher-energy collisional dissociation ( HCD ) . Fragment mass spectra ( MS2 ) acquired from the Orbitrap analyzer are searched against a protein database to confirm the identity of the monitored peptide . ( B ) Robustly detected daughter ions ( here y4 , y6 , and y7 ) can be monitored over time . ( C ) Each daughter ion is monitored over time; signal from all monitored daughter ions is summed and the time-summed intensity plot is integrated to determine the signal for the peptide of interest . ( D–K ) Targeted mass-spectrometry analysis of indicated proteins . Panels are: left , time-intensity plot summed over all daughter ions of indicated peptide ( 138-1 and control eluates ) ; center , database-matched MS2 spectrum ( 138-1 eluate ) ; right , integrated intensities for indicated peptides ( mean ± SEM ) . ( D ) MYO7A . ( E ) MYO6 . ( F ) GIPC3 . ( G ) ANKRD24 . ( H ) PDZD7 . ( I ) ADGRV1 . ( J ) USH1C . ( K ) USH1G . Key: C , control eluates; 7A , 138-1 eluates . Arrowhead indicates elution position for heavy-labeled peptide and region where MS2 spectra match to the indicated protein . Statistical tests used two-tailed Student’s t-test , with significance indicated in the figure as follows: *p<0 . 05; **p<0 . 01; ***p<0 . 001 . Exact p-values and 95% confidence intervals are tabulated in Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 01410 . 7554/eLife . 18312 . 015Figure 4—source data 1 . Analysis of the targeted proteomics experiments characterizing the protein composition of the 138-1 anti-MYO7A immunoaffinity purification from the chick inner ear . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 015 For most proteins we examined , we detected strong peaks for peptides of interest in 138-1 immunoprecipitates but not in controls ( Figure 4 ) . For example , we detected daughter ions of the MYO7A peptide TTLTDSATTAK in 138-1 but not control samples ( Figure 4D , left ) ; this peak was identified definitively by matching an MS2 spectrum obtained during the peak to MYO7A in the database ( Figure 4D , middle ) . This peptide was also detected in the extract used as the starting material ( total; not shown ) . We examined four peptides , each of which displayed roughly equivalent levels of intensity in 138-1 immunoprecipitates as compared to totals ( Figure 4D , right ) . Because the amount of material loaded in total samples ( ~2 ear-equivalents ) was much less than loaded for the immunoprecipitation samples ( ~100 ear-equivalents ) , the y-axis reflects neither recovery nor enrichment but is useful for relative comparison of different proteins . As in the shotgun proteomics experiments ( Figure 3 ) , MYO6 was readily detected in the eluates of 138-1 immunoprecipitates but not in controls ( Figure 4E ) . Although a smaller fraction of MYO6 in the S7 extract was precipitated as compared to MYO7A ( right panels ) , MYO6 was >10-fold more abundant in the extract than MYO7A , suggesting that MYO6 and MYO7A could be present in a nearly equimolar complex . GIPC3 and ANKRD24 were also specifically precipitated using the 138-1 antibody ( Figure 4F , G ) . Finally , targeted proteomics allowed us to confirm that USH1G is in a complex with MYO7A ( Figure 4K ) , as is USH1C ( Figure 4J ) . Our targeted proteomics assay for CDH23 was unfortunately not sensitive enough to detect it in the 138-1 eluates . Our combined immunoblotting and targeted proteomics data thus indicates that a MYO7A complex in stereocilia with USH1C , USH1G , and—probably—CDH23 is plausible . Other proteins known to be located in ankle links also co-precipitated with MYO7A . While we did detect ADGRV1 and DFNB31 in 138-1 immunoprecipitates by shotgun proteomics , their inconsistent appearance ( identified in only two of six mass-spectrometry runs each ) prevented us from definitively concluding that they were in a complex with MYO7A . To provide stronger evidence that MYO7A forms a protein complex with PDZD7 and ADGRV1 , we used targeted proteomics to determine how much of each protein was precipitated by 138-1 anti-MYO7A or control immunoaffinity beads . PDZD7 was present at significantly greater levels in 138-1 precipitations than in control precipitations ( Figure 4H ) , confirming its specific precipitation with MYO7A . By contrast , only one peptide for ADGRV1 showed significant enrichment in 138-1 immunoprecipitates compared to controls , and the fraction precipitated was very small ( Figure 4I ) . We suggest that PDZD7:MYO7A complexes are more abundant than ones containing ADGRV1 , or that ADGRV1 binds less tightly than PDZD7 . To confirm that PDZD7 interacts directly with MYO7A and extend the results to mammals , we co-expressed mouse or human proteins in HEK293 cells and immunoprecipitated GFP-MYO7A with the 138-1 monoclonal antibody . When detected using an anti-PDZD7 antibody , PDZD7 fused to mCherry was specifically immunoprecipitated by 138-1 along with GFP-MYO7A ( Figure 5A ) . The MYO7A-PDZD7 stoichiometry was high , as the majority of the PDZD7 was immunoprecipitated . ANKRD24 also specifically co-immunoprecipitated with MYO7A ( Figure 5B ) , although a much lower fraction of the total ANKRD24 was recovered . By contrast , PDZD7 tagged with Strep-tag II did not co-immunoprecipitate with ANKRD24-GFP , supporting the specificity of the interactions of PDZD7 and ANKRD24 with MYO7A ( Figure 5C ) . 10 . 7554/eLife . 18312 . 016Figure 5 . Interaction of mouse MYO7A and partners in HEK293 cells . PDZD7 , ANKRD24 , and GFP-MYO7A proteins are mouse; HA-MYO7A is human . ( A–C ) Immunoprecipitation and protein immunoblotting . Each panel has the immunoblot detection ( green ) superimposed on the ink stain for total protein ( magenta ) . Note that MYO7A , immunoprecipitated in each case , is usually visible on the ink stain . Molecular mass markers ( in kD ) are indicated on the left . Left side , starting material for immunoprecipitation ( total ) , with the fraction loaded relative to the immunoprecipitate indicated . Right side , immunoprecipitates . ( A ) PDZD7 and MYO7A . Immunoprecipitation with 138-1 antibody; immunoblot with anti-PDZD7 . ( B ) ANKRD24 and MYO7A . Immunoprecipitation with 138-1 antibody; immunoblot with anti-GFP . ( C ) PDZD7 and ANKRD24 . Immunoprecipitation with anti-GFP antibody; detection with anti-Strep . ( D–F ) Immunoprecipitation and mass spectrometry . ( D ) Untransfected HEK293 cells . ( E ) HEK293 cells transfected with mCherry-Pdzd7 plasmid alone . No PDZD7 was immunoprecipitated with the 138-1 antibody . ( F ) HEK293 cells co-transfected with Gfp-Myo7a and mCherry-Pdzd7 plasmids . MYO7A and PDZD7 were highly enriched in the 138-1 immunoprecipitates . Note that no PDZD7 was detected in HEK293 cells transfected with Gfp-Myo7a alone . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 01610 . 7554/eLife . 18312 . 017Figure 5—source data 1 . Analysis of the shotgun proteomics experiments characterizing the protein composition of the 138-1 anti-MYO7A immunoaffinity purification from transfected HEK cells . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 017 To determine whether any other HEK proteins were utilized in making the MYO7A-PDZD7 complex , we analyzed 138-1 immunoprecipitates by shotgun mass spectrometry . All data are deposited at ProteomeXchange with the identifier PXD004266 . All analyses are presented in Figure 5—source data 1 . As expected , neither MYO7A nor PDZD7 were detected in untransfected HEK cells ( Figure 5D ) ; when PDZD7 alone was expressed , it was present at a relatively high total fraction in the HEK cell extract but was not immunoprecipitated by 138-1 . By contrast , when MYO7A was co-expressed , PDZD7 was present in the 138-1 immunoprecipitate at a level ~0 . 4x that of MYO7A , confirming a near-stoichiometric interaction . No other proteins were co-precipitated at similar abundance , with the exception of one abundant peptide for the protein FER1L1 , which was also artifactually precipitated by 138-1 in the untransfected controls . To visualize the PDZD7-MYO7A interaction in tissue-culture cells , we used the nanoscale pull-down method ( Bird et al . , 2016 ) , which relies on a fusion between a binding partner and a MYO10 'heavy meromyosin-like' construct , which contains the motor , light-chain-binding region , and single-alpha-helix domain . This fusion protein—and any co-associating proteins—specifically concentrate at the tips of COS7 filopodia , where they are easily visualized ( Berg and Cheney , 2002; Kerber and Cheney , 2011 ) . The human MYO7A tail ( residues 967–2175 , including the two tandem MyTH4-FERM domains and the SH3 domain ) was fused to a construct containing GFP and bovine MYO10 heavy meromyosin ( MYO10HMM; motor and light-chain binding domains ) . Co-expression of the MYO10HMM-MYO7Atail fusion with mCherry-USH1G led to robust targeting to filopodia tips ( Figure 6A , B ) . As seen previously , mCherry-USH1C formed intracellular actin cables; co-expressed MYO10HMM-MYO7Atail targeted there too ( Figure 6B ) . 10 . 7554/eLife . 18312 . 018Figure 6 . Co-localization of PDZD7 with MYO10-MYO7A fusion . COS7 cells were transfected with the indicated constructs and stained with phalloidin to detect actin . MYO10HMM and MYO10HMM-MYO7Atail constructs ( derived from bovine MYO10 and human MYO7A ) were fused to GFP , which was imaged directly , as was mCherry fused to the indicated molecules . ( A ) mCherry-USH1G robustly localizes to filopodia tips when expressed with MYO10HMM-MYO7Atail . ( B ) mCherry-USH1C generates actin cables , to which MYO10HMM-MYO7Atail targets . ( C ) mCherry-PDZD7 does not concentrate at filopodia tips when the control MYO10HMM construct is expressed . Occasional exceptions are seen ( yellow asterisk ) . ( D ) mCherry-PDZD7 does concentrate at filopodia tips when the MYO10HMM-MYO7Atail construct is expressed . ( E ) Very little Strep-ANKRD24 co-localizes with MYO10HMM-MYO7Atail at filopodia tips . Main panels are 40 × 40 µm; inset panels are 4 × 4 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 018 mCherry-PDZD7 targeted to filopodia tips , but only if MYO10HMM-MYO7Atail was expressed ( Figure 6D ) ; with a few exceptions , no mCherry-PDZD7 was clustered with the control construct MYO10HMM ( Figure 6C ) . By contrast , only minimal amounts of ANKRD24-Strep targeted to filopodia , confirming its interaction with the MYO7A tail was weak , nonexistent , or masked in tissue-culture cells ( Figure 6E ) . To determine whether PDZD7 depends on MYO7A for localization in hair cells , we compared wild-type and Myo7a8J mice , which are functionally null ( Zheng et al . , 2012 ) . High-resolution confocal imaging revealed additional features of hair bundles in Myo7a8J mouse utricles ( Figure 7A–G ) . As previously reported , vestibular stereocilia of 8J/8J homozygotes were not arranged in a coherent bundle but instead projected randomly; moreover , stereocilia heights in Myo7a8J mice were irregular , with many stereocilia being longer than the longest stereocilia in wild-type mice ( Figure 7B , F , M ) . Not only were stereocilia not arranged in a staircase manner , but unlike in wild-type bundles ( Figure 7E ) , they were often arranged radially around the cytoplasmic channel where the kinocilium inserts , the fonticulus ( Figure 7G , asterisk ) . The fonticulus was usually found in the center of cuticular plate , indicating a loss of subcellular planar polarity ( Deans , 2013 ) . 10 . 7554/eLife . 18312 . 019Figure 7 . MYO7A localization in P8 mouse hair cells . ( A ) MYO7A immunoreactivity in +/8J heterozygote utricle . ( B ) MYO7A immunoreactivity in a 8J/8J homozygote utricle . Antibody signal is gone . ( C ) MYO7A immunoreactivity in utricle stereocilia isolated on poly-lysine-coated glass; structured-illumination microscopy . Note punctate labeling outside of the actin core . ( D ) Phalloidin staining of +/8J heterozygote utricle . ( E ) Cross-section of +/8J heterozygote utricle immediately above the cuticular plate . The cytoplasmic channel where the kinocilium inserts , the fonticulus , is located asymmetrically ( * ) . ( F ) Phalloidin staining of 8J/8J homozygote utricle . Many stereocilia are abnormally long; bundle cohesion is absent . ( G ) Cross-section of 8J/8J homozygote utricle immediately above the cuticular plate . Note the fonticulus is centrally located ( * ) . ( H ) ADGRV1 immunoreactivity in +/8J heterozygote utricle; note band at ankle links region . ( I ) ADGRV1 immunoreactivity in a 8J/8J homozygote utricle . Antibody signal is gone . ( J ) ATP2B2 immunoreactivity in +/8J heterozygote utricle . ( K ) ATP2B2 immunoreactivity in a 8J/8J homozygote utricle . Antibody signal is unchanged . ( L ) MYO7A immunoreactivity in +/8J heterozygote cochlea; structured-illumination image . MYO7A labeling is punctate; it is sparse in the hair bundle , and strong in the cytoplasm . Inner hair cell stereocilia are bent backwards against the coverslip , causing them to overlap the outer hair cell area . ( M ) MYO7A immunoreactivity in a 8J/8J homozygote cochlea . Antibody signal is absent . ( N ) ATP2B2 immunoreactivity in +/8J heterozygote cochlea . Note weak staining in inner hair cells , but extremely strong staining in outer hair cells . ( O ) ATP2B2 immunoreactivity in a 8J/8J homozygote cochlea . Although the morphology of bundles from inner and outer hair cells has changed dramatically , ATP2B2 is still targeted to each in approximately the same density . Scales: panels A , B , D , and F are 20 × 20 µm; panels in C are 6 . 7 × 10 µm; panels E and G are 10 × 10 µm; panels H–K are 25 × 25 µm; panels L–O are 20 × 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 019 Depending on the report , MYO7A localization in mouse vestibular hair bundles has been reported to be at stereocilia bases , at tip links , and throughout the bundle ( Hasson et al . , 1997; Boëda et al . , 2002; Senften et al . , 2006; Grati and Kachar , 2011 ) . To localize MYO7A , we used an antibody that did not label 8J/8J homozygous bundles ( Figure 7B ) ; this antibody produced punctate labeling that was often stronger near stereocilia tips ( Figure 7A ) . To allow imaging of utricle stereocilia by superresolution microscopy with structured-illumination ( Gustafsson , 2000 ) , we isolated stereocilia by absorption to poly-L-lysine-coated coverslips ( Neugebauer and Thurm , 1984 , Hasson et al . , 1997 ) ; these experiments showed that MYO7A bound along the periphery of the actin core , juxtaposed with the membrane ( Figure 7C ) . As previously reported for cochlea ( Michalski et al . , 2007 ) , the localization of ADVGRV1 to the ankle link region in utricle hair bundles depended on functional MYO7A ( Figure 7H–I ) . By contrast , localization of ATP2B2 was unaffected in 8J/8J homozygous utricle bundles ( Figure 7J–K ) . To visualize MYO7A in cochlear hair cells , we used structured-illumination microscopy . MYO7A was punctate , with weak labeling in hair bundle , and stronger labeling in the cell soma ( Figure 7L ) . As previously reported ( Self et al . , 1998 ) , stereocilia of 8J/8J homozygotes were highly disorganized and had irregular lengths; no MYO7A immunoreactivity was detected ( Figure 7M ) . By contrast , ATP2B2 localization in inner and outer hair cells was relatively normal , despite the bundle disarray; in both heterozygotes and homozygotes , labeling was much stronger in outer hair cells than in inner hair cells ( Dumont et al . , 2001; Beurg et al . , 2010 ) . We sought to confirm the localization of PDZD7 at ankle links and to determine whether its location depended on MYO7A . To best visualize ankle links in a z-stack of x-y confocal images , we maximum-projected x-z slices to a depth of 5–10 µm , allowing superposition of multiple hair bundles in a profile perspective . As previously shown ( Grati et al . , 2012; Zou et al . , 2014 ) , PDZD7 localized to ankle links of heterozygous P8 Myo7a+/8J utricle hair cells ( Figure 8A ) . Some cell bodies of hair cells had large amounts of PDZD7 immunoreactivity ( Figure 8A , F ) , and PDZD7 was located in the pericuticular necklace region and the funiculus ( Figure 8B ) . Both regions have been implicated in vesicular trafficking . 10 . 7554/eLife . 18312 . 020Figure 8 . PDZD7 localization at ankle links depends on MYO7A in P8 mice . ( A ) PDZD7 immunoreactivity in wild-type utricle . Note PDZD7 in band near the stereocilia ankles ( arrows ) , consistent with previous localization at ankle links . ( B ) Deconvolution analysis of utricle hair bundle at the level of the cuticular plate . PDZD7 immunoreactivity appears as a ring around the cytoplasmic channel in the cuticular plate . PDZD7 is also present in the pericuticular necklace surrounding the cuticular plate ( arrows ) ; image gamma was adjusted to 1 . 5 to allow visualization of pericuticular labeling without saturating cytoplasmic-channel labeling . ( C ) PDZD7 immunoreactivity in utricle hair bundle isolated on poly-lysine-coated glass; structured-illumination microscopy ( SIM ) . Immunoreactivity near the stereocilia tapers is apparent ( arrows ) . ( D ) Co-labeling of MYO7A and PDZD7 in utricle stereocilia isolated on glass using SIM . PDZD7 labeling above tapers is indicated ( arrows ) . Insert , magnification of box in right-hand merged image . M , MYO7A; P , PDZD7; A , actin . Note that MYO7A and PDZD7 punctae occasionally overlap ( yellow ) . ( E ) Quantitation of PDZD7 and MYO7A along utricle stereocilia length . Left panel , quantified stereocilia are indicated . Note that the three stereocilia from panel D are included . Right panel , average fluorescence signals from each channel from stereocilia aligned at the taper . Note the decreased stereocilia diameter near 0 µm , which corresponds to the peak of PDZD7 labeling ( at ~0 . 5 µm ) . ( F ) MYO7A and PDZD7 punctae partially overlap in utricle stereocilia shafts . Individual channels and the merge of the PDZD and MYO7A channels are shown . Arrows indicate punctate that are present both in PDZD7 and MYO7A channels; arrowheads indicate PDZD7 at ankle-links region . ( G ) Single utricle hair cell after in utero electroporation with mCherry-PDZD7 at E12 and analysis at P2 . mCherry-PDZD7 is located in the ankle-link region ( arrows ) . ( H ) Single cochlear outer hair cell labeled after injectoporation with mCherry-PDZD7 at P3 and imaged after 2 days in culture . Labeling is associated with stereocilia base , although the top view obscures that localization . Airyscan processing . Additional transfected cells are displayed in Figure 8—figure supplement 1 . ( I ) Another mCherry-PDZD7 cell , viewed in profile using x-z reslice . Note signal is concentrated near stereocilia bases ( arrows ) . Airyscan processing . ( J ) PDZD7 immunoreactivity in utricle hair cells of Myo7a+/8J heterozygote mice . Labeling at ankle-link region is clear ( arrows ) . ( K ) PDZD7 at the ankle region of heterozygote utricle . ( L ) PDZD7 in utricle hair cells of Myo7a8J/8J null mice . While PDZD7 is still in the cytoplasm , it is no longer also located near ankle links . Labeling in the cytoplasmic channel and kinocilium base remains . ( M ) PDZD7 at ankle region in homozygous utricles . Little or no PDZD7 is detected at ankle links . ( N ) PDZD7 in cochlea hair cells of Myo7a+/8J heterozygote mice using SIM . Labeling at ankle-link region is clear ( arrows ) . ( O ) PDZD7 in cochlea hair cells of Myo7a8J/8J null mice using SIM . There is no PDZD7 immunoreactivity in bundles . Scales: panel A is 20 × 40 . 4 µm; panel B is 7 . 5 × 7 . 5 µm; panel C is 10 × 10 µm; panels in D are 6 . 4 × 13 µm; panels in F are 17 . 5 × 17 . 5 µm; panel G is 20 × 40 . 4 µm; panel H is 12 × 12 µm; panel I is 5 × 5 µm; panels J and L are 30 × 30 µm; panels K and M are 15 × 15 µm; panels N and O are 10 × 20 . 2 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 02010 . 7554/eLife . 18312 . 021Figure 8—figure supplement 1 . Localization of mCherry-PDZD7 at tips of stereocilia . This hair cell expressed PDZD7 at a relatively high level , and the hair bundle was severely disrupted . mCherry-PDZD7 was clearly seen at stereocilia tips , however ( asterisks ) . Green , mCherry-PDZD7; magenta , phalloidin stain for actin . Panel is 20 × 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18312 . 021 We also imaged PDZD7 using structured-illumination microscopy of isolated stereocilia; both intact bundles ( Figure 8C ) and dispersed stereocilia ( Figure 8D ) were obtained . When we labeled isolated stereocilia with antibodies against PDZD7 , some labeling was evident near basal tapers of stereocilia ( Figure 8C ) . When we quantified PDZD7 and MYO7A labeling ( Figure 8D–E ) , we noted consistent PDZD7 labeling at stereocilia tapers , despite significant background labeling; MYO7A labeling was more dense in distal regions of the stereocilia . There was some overlap between PDZD7 and MYO7A labeling at stereocilia tapers , however ( Figure 8D , right panel , inset ) . To determine whether PDZD7 and MYO7A co-localize in intact hair bundles , we used confocal imaging with deconvolution processing . While MYO7A and PDZD7 generally did not overlap , some punctae within the stereocilia were labeled clearly with both antibodies ( Figure 8F ) . To confirm the localization of PDZD7 at the stereocilia ankle link region , we transfected hair cells with mCherry-PDZD7 and investigated its location . We first used in utero electroporation of E12 mouse otocysts ( Brigande et al . , 2009 ) , then analyzed utricles at P2 . In two transfected utricles , we identified >20 transfected hair cells; the mCherry signal was low in all , but in cells with the highest expression of mCherry-PDZD7 , the signal was clearly discernable at the base of the hair bundle ( Figure 8G ) . We also used injectoporation ( Xiong et al . , 2012; Zhao et al . , 2014; Xiong et al . , 2014 ) to deliver the mCherry-PDZD7 plasmid to cochlear hair cells . Many cells were transfected , with a wide range of expression levels . Cells with the highest levels of mCherry signal had very disorganized hair bundles ( Figure 8—figure supplement 1 ) , a phenotype we rarely see in cells injectoporated with control plasmids ( Xiong et al . , 2012; Zhao et al . , 2014; Xiong et al . , 2014 ) . By contrast , cells with the lowest levels of signal had mCherry-PDZD7 located specifically in hair bundles , largely at stereocilia bases ( Figure 8H–I ) . We noted that many cells with intermediate levels of mCherry-PDZD7 had some signal located specifically at stereocilia tips ( Figure 8—figure supplement 1 ) . PDZD7 depended on MYO7A for localization to ankle links ( Figure 8J–O ) . While localization was normal in heterozygous Myo7a+/8J utricles ( Figure 8J–K ) , when we examined 8J/8J homozygous Myo7a8J/8J utricles at P8 , we noted that PDZD7 was absent from the ankle-link region , although it remained present in the apical region of hair cell somas ( Figure 8L–M ) . Strong PDZD7 appeared in the cytoplasmic channel where the kinocilium inserted; in some cases the labeling extended up into the hair bundle region , but was distinct from ankle-link labeling ( Figure 8L–M ) . In the cochlea , PDZD7 labeling was prominent in the ankle-link region of bundles of heterozygous cochlear inner hair cells visualized using structured-illumination microscopy ( Figure 8N ) . By contrast , PDZD7 labeling was absent from homozygous Myo7a8J/8J bundles ( Figure 8O ) .
Our original hair-bundle isolation method , which uses agarose to embed and excise the bundles , yields stereocilia that are >90% pure ( Gillespie and Hudspeth , 1991; Shin et al . , 2013; Krey et al . , 2015 ) . Considerable fine dissection is required , however , to prepare the inner-ear organs and clean up the agarose-embedded hair bundles; these steps make the daily throughput of bundles relatively low . In addition , the agarose matrix that extricates bundles complicates subsequent steps , such as detergent extraction of membrane proteins . While protein complexes can be detected by immunoprecipitation using the twist-off method ( Yamoah et al . , 1998; Hill et al . , 2006b ) , the D10 purification strategy is far better for obtaining large amounts of stereocilia-membrane material that is suitable for complex extraction and purification ( Table 1 ) . The D10 preparation has significant contamination , however; we estimate that stereocilia membranes account for only ~5% of the RIPA-solubilized material . Other membranes are present in part because contaminating intracellular organelles are disrupted by a freeze-thaw cycle , preventing their precise separation from stereocilia based on distinct , uniform sedimentation properties ( Cox and Emili , 2006 ) . Many of the most important stereocilia proteins are present at vanishingly small levels , however , so for stereocilia membrane enrichment , we chose to maximize total recovery of stereocilia membrane protein rather than purity . Immunoaffinity purification , especially with high-affinity monoclonal antibodies , allows tremendous enrichment , even if multiple affinity enrichment steps are ultimately needed for some stereocilia protein complexes . We stabilized protein complexes of the stereocilia using chemical crosslinking with membrane-impermeant crosslinker DTSSP , which should enhance our ability to detect low-abundance membrane protein interactions ( Vasilescu et al . , 2004; Gokhale et al . , 2012; Kim et al . , 2012; Corgiat et al . , 2014 ) . By using DTSSP , we were able to use stringent extraction and washing conditions , minimizing the dissociation of protein complexes that included extracellular domains . Because the disulfide bond in DTSSP can be split by treating samples with a strong reducing agent prior to immunoblotting or mass spectrometry , we were able to examine proteins associated with MYO7A . Were all proteins precipitated in our experiments with the 138-1 antibody in specific complexes with MYO7A ? Probably not . While DTSSP should not penetrate into cells and hence artifactually crosslink nearby proteins , it is possible that trace amounts of large cytoskeletal complexes are purified with the immunoaffinity procedure . Subsequent validation of interactions detected , for example showing co-immunoprecipitation , protein-protein binding , or mislocalization in mutant hair cells , is thus of high importance . Our data show that MYO7A interacts with PDZD7 , a key ankle-link component . Originally identified as a modifier gene for Usher syndrome type 2 genes , PDZD7 has more recently been identified as a nonsyndromic deafness gene itself ( Booth et al . , 2015; Vona et al . , 2016 ) . Genetics experiments showed that positioning of ADGRV1 at ankle links depends critically on PDZD7 ( Zou et al . , 2014 ) ; consistent with those results , PDZD7 binds directly to ADGRV1 ( Chen et al . , 2014 ) . Our experiments suggest how this positioning might occur; PDZD7 could bind to MYO7A , and then MYO7A could transport the PDZD7-ADGRV1 complex to its final location . While present at lower levels than PDZD7 , we did detect ADGRV1 in co-immunoprecipitates with MYO7A by both shotgun and targeted mass spectrometry; our experiments do not reveal , however , whether all three are simultaneously in the same protein complex . PDZD7 requires functional MYO7A for localization to ankle links . MYO7A is also required for localization of ADGRV1; while MYO7A might bind directly to ADGRV1 , the necessity of PDZD7 for the formation of the ankle-link complex suggests that MYO7A localizes both PDZD7 and ADGRV1 at ankle links . Association of MYO7A with ankle-link proteins explains the early observation that MYO7A concentrates at the ankle-link region of vestibular stereocilia of mature frogs ( Hasson et al . , 1997 ) and cochlea stereocilia of immature mice ( Senften et al . , 2006; Lëfevre et al . , 2008 ) . Positioning at this location by MYO7A is surprising , however , as the ankle links are very close to the pointed ends of the filaments located at the periphery of the stereocilia actin core . We suggest three possible explanations . First , because MYO7A moves from pointed to barbed ends of actin filaments ( Inoue and Ikebe , 2003 ) , MYO7A may move PDZD7 and ADGRV1 towards stereocilia tips but stall if the N-terminal , extracellular end of ADGRV1 was anchored on the adjacent stereocilium . In this case , MYO7A would put ankle links under tension , which could also explain the frequent oblique positioning of the links ( Goodyear and Richardson , 1992 , 1999 ) . Second , MYO7A could hand off PDZD7 to MYO6 anywhere in the stereocilium; because MYO6 moves towards pointed ends of actin filaments , it would properly localize the ankle-link proteins at stereocilia bases . Indeed , MYO6 is thought to move PTPRQ to the taper region ( Sakaguchi et al . , 2008 ) . Finally , MYO7A and MYO6 may be in a multimolecular , bidirectional motor complex; like with other bidirectional motor complexes ( Hancock , 2014 ) , reciprocal regulation of the two motors could steer the complex to the top or bottom of the stereocilia . Indeed , MYO6 was the most abundant protein co-precipitated with MYO7A , and we are presently investigating the specificity of this interaction . Regardless of the mechanism , the data here show that MYO7A binds PDZD7 and is necessary for at least one step in its localization at the base of the hair bundle . It is notable that the paralogs PDZD7 , USH1C , and DFNB31 are all deafness genes and interact with MYO7A . Moreover , PDZD7 and DFNB31 both homodimerize and form heterodimers ( Chen et al . , 2014 ) ; USH1C homodimerizes ( Siemens et al . , 2002 ) , but whether it can heterodimerize with PDZD7 or DFNB31 is unknown . Several other proteins are specifically co-precipitated with MYO7A; those with high enrichment and relatively high stoichiometry include MYO6 , MYL4 , COLEC12 , ANKRD24 , GIPC3 , and MYL1 . Although each of these will require in-depth characterization such as we have done here for PDZD7 , we can speculate to their possible roles in hair cells . MYL1 and MYL4 are myosin light chains , and may substitute for CALM in binding to one or more IQ domains in MYO7A ( or MYO6 ) . The presence of MYO6 is quite interesting , as it could mediate pointed-end-directed motility in a complex that also contains a barbed-end-direct motor . More specific investigation of how MYO6 and MYO7A interact will be required; our preliminary co-expression experiments indicated only a low-stoichiometry interaction when both were expressed together in HEK cells ( Figure 5—source data 1 ) . A relatively large fraction of ANKRD24 present in the stereocilia-membrane extract was co-precipitated , yet ANKRD24 did not appear to bind directly to MYO7A ( Figures 5 , 6 ) . Presumably ANKRD24 interacts with MYO7A via other proteins , indicating that a larger MYO7A complex may be present that warrants further study . Other proteins were highly enriched but less abundant , including MYO3B , MYO1H , ESPN , LMO7 , FSCN2 , and GRXCR1 . All of these proteins interact with the actin cytoskeleton , and it is likely that some of them co-precipitate because of their association with substoichiometric levels of actin filaments also bound to MYO7A . Future experiments will be needed to discern which of these proteins interacts specifically with MYO7A complexes and which co-precipitate on associated actin . A fraction of the MYO7A derived from enriched membranes was bound to USH1C and USH1G , and likely had bound CDH23 as well . While our experiments were not designed to accurately measure stoichiometry of MYO7A complexes , the data were nonetheless consistent with a roughly equimolar complex of the three proteins . This MYO7A-USH1C-USH1G complex was not necessarily the tip-link complex itself , as MYO7A could simply transport the scaffolding proteins to the correct location at stereocilia tips , hand off the complex to CDH23 , then dissociate . Notably , CDH23 does not depend on MYO7A for localization in hair bundles ( Senften et al . , 2006; Boëda et al . , 2002 ) , although USH1C does ( Boëda et al . , 2002; Lëfevre et al . , 2008 ) , suggesting that the cadherin and scaffolding proteins are transported separately . This interpretation is consistent with our determination that only a very small fraction of CDH23 was in a complex with MYO7A . Unconventional myosins like MYO7A participate in transport , anchoring , tension sensing , actin organization , and cell adhesion ( Hartman et al . , 2011 ) . Given the substantial number of scaffolding proteins found with MYO7A , it is reasonable to conclude that the motor transports various proteins towards stereocilia tips . Nevertheless , consistent localization at ankle links and tip links suggests that MYO7A may also anchor protein complexes and sense tension applied to them . Furthermore , examination of Myo7a8J/8J hair bundles indicates clearly that both their actin organization and stereocilia adhesion are disrupted . The division of MYO7A’s tasks into several categories is thus misleading , as those activities likely all overlap—MYO7A is a multifunctional protein , interacting with several different protein complexes in stereocilia and carrying out transport and functional roles . Our hypothesis is that knowing the composition of MYO7A complexes is an essential step in understanding the function of this motor in stereocilia . Our experiments are designed to provide a rough estimate of the abundance of proteins present in MYO7A complexes . To accurately measure the stoichiometry of the detected complexes , additional purification will be required . We can identify co-precipitating proteins using enrichment analysis , but despite a nearly 10 , 000-fold purification of MYO7A , the immunoaffinity eluates have many proteins that have bound nonspecifically . This result is entirely expected; the proteins that interact with MYO7A are rare within stereocilia , so MYO7A protein complexes are present at extremely small levels in the stereocilia detergent extract . In addition , because of their low levels , extensive washing of MYO7A immunoprecipitates is required , biasing our analysis towards high-affinity interactions . Future studies that use a second antibody purification step following precipitation with 138-1 , for example with an anti-PDZD7 antibody , will be needed to generate a more pure complex that can be interrogated for additional binding partners and stoichiometries of each component . Such experiments will likely require further scaling up of the starting material , given the likely losses during purification . Nevertheless , there is great value in large-scale purification of protein complexes from stereocilia , as these experiments will allow us to identify interacting components—like those in the transduction-channel complex—that may not be detected by any other strategies .
All proteins are referred to by their official gene symbols , all capitals ( http://www . uniprot . org; http://www . genecards . org; http://www . informatics . jax . org/mgihome/nomen/gene . shtml#ps ) . Actin is the principal exception , as stereocilia contain a mixture of actin isoforms that are not readily distinguished by mass spectrometry . When the mouse gene name differed from gene names in other species ( e . g . , Whrn vs . DFNB31 ) , we chose the more systematic name for the protein ( DFNB31 in this example ) . Hybridoma cells for the D10 monoclonal antibody were obtained from Guy Richardson ( University of Sussex , UK ) ; cells for the 138-1 monoclonal ( RRID:AB_2282417 ) were obtained from Developmental Studies Hybridoma Bank ( Iowa City , Iowa ) . Antibodies were purified in serum-free medium using a bioreactor ( VGTI Monoclonal Core , OHSU ) . The control antibody for immunoaffinity purification was ChromPure Mouse IgG , whole molecule ( Jackson ImmunoResearch , West Grove , PA; #015-000-003; RRID:AB_2337188 ) . Primary antibodies for immunoblotting or immunocytochemistry included mouse monoclonal anti-PDZD7 ( AbFrontier , Seoul , Korea; #YF-PA20973 , 1:500 ) , rabbit anti-MYO7A ( Proteus , Ramona , CA; #25-6790; RRID:AB_10015251 ) , rabbit anti-ADGRV1 ( from Dominick Cosgrove ) , and F2a rabbit anti-ATP2B2 ( Dumont et al . , 2001 ) . The position of the fluorescent protein in the symbol for each fusion protein corresponds to the location ( N- or C-terminus ) relative to the stereocilia protein . The mouse GFP-MYO7A plasmid , constructed in pEGFP-C1 , was obtained from David Corey . To generate the human HA-MYO7A construct , a MGC premier human MYO7A ORF clone ( BC172349 ) was purchased from TransOMIC ( Huntsville , AL; Catalog# TOH6003 ) , amplified , and directionally cloned into pCMV6-AN-HA vector ( OriGene , Rockville , MD; Catalog# PS100013 ) using AscI and NotI restriction enzymes . mCherry-PDZD7 , mCherry-USH1C , and mCherry-USH1G , where mouse Pdzd7 , mouse Ush1c ( isoform b1 ) , or human Ush1g were cloned into pmCherry-C1 , were described previously ( Grati et al . , 2012 ) . To make PDZD7-Strep , where 'Strep' refers to the tandem Strep tags called Strep-tag II ( Schmidt et al . , 2013 ) , mouse Pdzd7 was cloned into a modified pRP[Exp]-CMV vector ( VectorBuilder; Santa Clara , CA ) containing the tag at the 3’ end of the insertion site . Mouse Pdzd7 was also cloned into the pEGFP-N1 vector , which had been modified to remove the EGFP and add an HA tag , to generate HA-PDZD7 . Mouse ANKRD24-GFP was constructed by VectorBuilder in the pRP[Exp]-CMV vector . For large-scale protein purification experiments of Figures 2–4 , the large numbers of chicken ears made pre-experiment power analysis impractical . However , the stochastic sampling that plagues shotgun mass spectrometry experiments ( Figures 2–3 ) suggested that reliable detection of low abundance proteins would require multiple mass spectrometry runs , each a technical replicate . For the purification analysis ( Figure 2 ) , technical replicates are aliquots of the purification fractions; for the immunoaffinity purification experiments ( Figure 3 ) , technical replicates are divided 138-1 eluates . For practical reasons , for each experiment , we carried out two to three preparations ( biological replicates ) of ~500 chicken ears each . In the targeted proteomics experiments , not all proteins ( and their peptides ) were analyzed in each experiment ( biological replicate ) . We ensured that each peptide analyzed in Figure 4 was examined in at least two technical replicates from each of two biological replicates , for a total of four mass spectrometry runs each . For the immunoprecipitation and mass spectrometry experiments of Figure 5D–F , which reproduce the immunoprecipitation and immunoblotting experiment of Figure 5A ( which itself was repeated >5 times ) , we prepared a single biological replicate ( pooled from multiple plates ) , and divided the 138-1 eluate so that two identical technical replicates could each be analyzed by shotgun mass spectrometry . Although the small number of replicates precluded meaningful statistical analysis , the results qualitatively corroborated the immunoblotting experiments . To test whether any proteins were differentially expressed in shotgun proteomics experiments of Figure 3 , each protein’s riBAQ value was transformed into log2 scale and normalized by global median normalization ( Yang et al . , 2002 ) . Statistical significance between conditions was determined using a modified two-sided t-test by empirical Bayes ( Smyth , 2004 ) , with false-discovery-rate adjustment of p-values for a multiple test correction ( Benjamini and Hochberg , 1995 ) . Rather than filter proteins by the number of identifications per condition , we kept as many proteins as possible as the model could be fitted . The computation was done using the limma package ( Ritchie et al . , 2015 ) in R Statistical Computing Environment ( www . r-project . org ) . In experiments that used targeted proteomics to determine whether a protein was specifically immunoprecipitated by the 138-1 antibody as compare to a control antibody , we used the two-tailed Student’s t-test to compare a peptide’s signal in the 138-1 eluates with that in the control eluates . In each case , 4–6 replicates from 138-1 and control eluates were compared . For immunoaffinity isolation , we used MyOne Tosylactivated Dynabeads ( Life Technologies , Grand Island , NY; #65502 ) , which are based on 1 µm superparamagnetic beads . Antibodies were coupled at 40 µg antibody per mg of beads in 0 . 1 M sodium borate pH 9 . 5 , 1 M ammonium sulfate; coupling went overnight at 37°C with shaking . Unreacted groups were blocked overnight at 37°C with shaking in PBS containing 0 . 05% Tween 20 , and 0 . 5% BSA . Antibody-coupled beads were stored at 4°C in the same buffer with 0 . 02% NaN3 . The D10 bead stock concentration was 50 mg/ml , with the coupled antibody at 2 mg/ml . Fertilized chicken eggs were obtained from AA Lab Eggs ( Westminster , CA ) . Temporal bones were removed from E19-E21 chicks and were placed in ice-cold oxygenated chicken saline ( 155 mM NaCl , 6 mM KCl , 2 mM MgCl2 , 4 mM CaCl2 , 3 mM D-glucose , 10 mM HEPES , pH 7 . 25 ) for no more than 2 hr , with an exchange of saline after 1 hr . Sensory inner ear organs were removed using micro-dissection and were stored in ice-cold oxygenated saline for up to 4 hr during dissection . Organs were rinsed with 4–5 changes of chicken saline ( minimum 10-fold dilution per rinse ) to remove excess soluble protein . 3 , 3'-dithiobis ( sulfosuccinimidyl propionate ) ( DTSSP; Life Technologies #21578 ) was added at 1 mg/ml in chicken saline for 1 hr at 4°C; organs were subjected to rotation to permit maximal access of DTSSP to extracellular spaces . To quench the DTSSP reaction , 1 ml of 1 M Tris pH 8 was added and samples were incubated for 5 min . All excess buffer was removed , then the organs were frozen by plunging into liquid N2; organs were stored at -80°C prior to use . Organs were thawed with chicken saline containing 1:100 Protease Inhibitor Cocktail ( Sigma-Aldrich , St . Louis MO; #P8340 ) and 2% normal donkey serum ( NDS; Jackson ImmunoResearch ) at approximately 5 ml per 100 ears . The organs were homogenized with 20 strokes at 2400 rpm using a glass/Teflon homogenizer . The homogenate was centrifuged at 120 ×g for 5 min at 4°C . The supernatant was collected , and then the homogenization was carried out two more times . The pellet was subsequently washed 2–3 more times with chicken saline containing NDS and protease inhibitors . All supernatants ( typically 50–60 ml per 1000 ears ) were combined as the post-nuclear supernatant ( S1 ) ; the nuclear pellet ( P1 ) was discarded . The post-nuclear supernatant was layered on to 2 . 2 M sucrose cushions ( 11 ml supernatant , 1 ml cushion ) . The samples were spun at 8400 ×g for 30 min at 4°C . The supernatant was removed ( S2 ) ; to collect the dense-membrane pellet , the cushion was removed and the tubes were washed out with chicken saline with protease inhibitors and serum . Dense membranes ( P2 ) were homogenized using five strokes in a glass/Teflon homogenizer to remove lumps . The volume yield was usually ~20–25 ml for 500 ears . D10 beads ( 12 . 5 mg/ml beads , D10 at 500 µg/ml ) were washed with chicken saline with serum and were added to the dense membranes at 1 µl per ear: the mixture was rotated overnight at 4°C . The beads were then collected with a magnet , then washed 5x with chicken saline containing serum and 3x with chicken saline; they were then frozen while the next aliquot was prepared . The original flow-through solution was bound to fresh D10 beads using the same conditions; the next day , the beads were collected with a magnet , washed , and pooled with the original beads . To elute stereocilia membranes , the pooled and washed D10 beads were sonicated using a sonicator ( Sonics & Materials , Newtown , CT; model VCX 130 ) with a 2 mm probe; sonication was carried out in saline with protease inhibitors in 2–3 ml batches ( in ice water to keep cold ) . Sonication was for 5–10 s at 25–50% power , followed by cooling in ice water for 1–2 min . A magnet was used to concentrate the beads and the solution was removed . The sonication was repeated for a total of 20 ml of eluate; this solution was spun at 112 , 500 ×g ( rmax; 35 , 000 rpm in a Beckman 70Ti rotor ) ; the pellet was retained . Sonication was repeated on the D10 beads with 6 × 3 ml additional aliquots; these aliquots were pooled and centrifuged . The supernatants from the two centrifugation steps were pooled ( cytosolic fraction ) . Membrane pellets were resuspended using sonication with saline plus protease inhibitors and were combined; the pool was diluted to ~500 ear-equivalents per tube . The solution was spun at 125 , 000 ×g ( rmax; 45 , 000 rpm in Beckman TLA55 rotor ) for 30 min at 4°C . The supernatant ( S7 ) was removed and the pellet ( enriched stereocilia membranes ) was frozen at −80°C . Stereocilia membranes were sonicated with 500 µl RIPA buffer ( 50 mM Tris pH 8 . 0 , 150 mM NaCl , 0 . 1% SDS , 1% NP-40 , 0 . 5% deoxycholate , 1:100 protease inhibitors ) as above for each 500 ears; extracts were spun at 125 , 000 ×g ( rmax ) for 15 min at 4°C . The extraction was repeated twice on the pellet and the three supernatants were combined and diluted to 1 . 5 ml total volume ( 10 ears/30 µl ) . Aliquots used for immunoblots were precipitated with TCA and stored for later use . Immunoaffinity purification was carried out serially; the RIPA extract was first incubated with beads with control antibody , then the unbound material was then incubated with beads coupled with the specific antibody ( e . g . , 138-1 anti-MYO7A ) . The RIPA extract ( 1 . 5 ml; 500 ear-equivalents ) or flow-through material was added to 50 µl antibody-coupled beads; the beads and extract were rotated for 1 hr at room temperature . Beads were collected with a magnet , washed at least 5x with RIPA buffer , and eluted 5x with 20 µl 2% SDS . All eluates were combined for each set; typically , 20 µl was reserved for gel analysis , and 80 µl was used for mass spectrometry . Samples from the enrichment scheme were diluted to 0 . 05 or 0 . 2 ear-equivalents per 20 µl by calculating the ear number input and total volume of each fraction . NuPAGE 4X LDS sample buffer ( Life Technologies #NP0008 ) , and 500 mM DTT were added to achieve 1X LDS sample buffer and 50 mM DTT . Samples were boiled at 95°C for 2–5 min , and resolved using 4–12% SDS-PAGE gels with MES or MOPS buffer , or 3–8% SDS-PAGE gels with acetate buffer ( NuPAGE gels and buffers , Life Technologies ) . Proteins were transferred to PVDF ( Immobilon-P; EMD Millipore , Billerica MA ) , and were visualized with India Ink ( 1:5000 ) in PBS/0 . 05% Tween-20 . Membranes were blocked with Prime Blocking Agent ( GE Healthcare Life Sciences , Pittsburgh PA; #RPN418 ) , and probed with specific primary antibodies which were detected with species-specific HRP-coupled secondary antibodies ( Jackson ImmunoResearch #111-035-144 and #115-035-003 ) and ECL Prime ( GE Healthcare Life Sciences #RPN2232 ) . In-solution tryptic digests of the samples were prepared using the enhanced filter-aided sample preparation ( eFASP ) method ( Erde et al . , 2014 ) . Proteins were digested in an Amicon Ultra 0 . 5 ml 30K filter unit ( Millipore; UFC503056 ) in 100 µl digestion buffer with 200 ng sequencing-grade modified trypsin ( Promega , Madison , WI; #V5111 ) at 37°C for 12–16 hr . Peptides were isolated by centrifugation and were extracted with ethyl acetate to remove remaining deoxycholic acid ( Erde et al . , 2014 ) . Peptide samples were analyzed with an Orbitrap Fusion Tribrid mass spectrometer ( Thermo Fisher Scientific , Waltham , MA ) coupled to a Thermo/Dionex Ultimate 3000 Rapid Separation UPLC system and EasySpray nanosource . The samples were loaded onto an Acclaim PepMap C18 , 5 μm particle , 100 μm × 2 cm trap using a 5 μl/min flow rate and then separated on an EasySpray PepMap RSLC , C18 , 2 μm particle , 75 μm × 25 cm column at a 300 nl/min flow rate . Solvent A was water and solvent B was acetonitrile , each containing 0 . 1% ( v/v ) formic acid . After loading at 2% B for 5 min , peptides were separated using a 205-min gradient from 7 . 5–30% B , 5-min gradient from 30–90% B , 6-min at 90% B , followed by a 19 min re-equilibration at 2% B . MaxQuant ( Cox and Mann , 2008 ) and the search engine Andromeda ( Cox et al . , 2011 ) were used to identify peptides and assemble proteins from the mass spectrometer RAW files . MaxQuant was used to calculate iBAQ ( Schwanhäusser et al . , 2011 ) for each protein , and we used an Excel spreadsheet to calculate riBAQ ( Shin et al . , 2013; Krey et al . , 2014 ) and enrichment values . Mass spectrometry data , as well as spreadsheets with all derived values , are available from ProteomeXchange ( http://www . proteomexchange . org ) using the accession numbers PXD004222 ( "D10 Stereocilia Membrane Enrichment" ) , PXD004221 ( "Myosin VIIA 138-1 immunoaffinity purification" ) , and PXD004266 ( "HEK cell MYO7A-PDZD7 immunoprecipitation" ) ; information conforming to Minimal Information About a Proteomics Experiment ( MIAPE ) standards ( Taylor et al . , 2007 ) is included in the submissions . For clustering with mclust ( Fraley and Raftery , 2002 ) , shotgun riBAQ data from the D10 enrichment procedure were used if the protein was detected in at least one of the technical replicates of both biological replicates; the 2–4 measurements were then averaged . Only those proteins detected in all samples analyzed ( S1 , S2 , P2 , S3 , M3 , S6 , P6 , P7 , and S7 ) were subjected to clustering . For targeted MS/MS , we measured peptides from the same three immunoprecipitations , each of ~500 ear-equivalents of stereocilia , as were used for the shotgun analysis . In-solution tryptic digests of the samples were prepared using the eFASP method . Proteins were digested in the filter unit in 100 µl digestion buffer with 200 ng sequencing-grade modified trypsin at 37°C for 12–16 hr . After isolating peptides by centrifugation , we extracted them with ethyl acetate to remove remaining deoxycholic acid ( Erde et al . , 2014 ) . We obtained heavy peptide standards ( Spiketides_L ) from JPT Peptide Technologies ( Berlin , Germany ) for demonstration that the monitored transitions originated from the intended peptide . Peptide samples were analyzed with an Orbitrap Fusion Tribrid mass spectrometer configured as described above; samples were loaded onto the trap and separated as above . Solvent A was water and solvent B was acetonitrile , each containing 0 . 1% ( v/v ) formic acid . After loading at 2% B for 5 min , peptides were separated using a 55-min gradient from 7 . 5–30% B , 10-min gradient from 30–90% B , 6-min at 90% B , followed by a 19 min re-equilibration at 2% B . Peptides were analyzed using the targeted MS2 mode of the Xcalibur software in which the doubly or triply charged precursor ion corresponding to each peptide was isolated in the quadrupole , fragmented by HCD , and full m/z 350–1600 scans of fragment ions at 30 , 000 resolution collected in the Orbitrap . Targeted MS2 parameters included an isolation width of 2 m/z for each precursor of interest , collision energy of 30% , AGC target of 5 × 104 , maximum ion injection time of 100 ms , spray voltage of 2400 V , and ion transfer temperature of 275°C . No more than 75 precursors were targeted in each run and no scheduling was used . Three unique peptides for each protein of interest were chosen for isolation based on previous data-dependent discovery data or from online peptide databases ( www . peptideatlas . org , www . thegpm . org ) . Precursor isolation lists for all peptides of interest were exported from the software package Skyline ( http://proteome . gs . washington . edu/software/ skyline/ ) and imported into the Orbitrap control software . Skyline was used to analyze targeted MS/MS data . Chromatographic and spectral data from RAW files were loaded into Skyline and manually analyzed to determine fragment ion peaks corresponding to each peptide . RAW files were also processed using Proteome Discoverer ( Thermo Fisher Scientific ) software in order to match MS/MS spectra to an Ensembl spectral database using Sequest HT . Fragment ion peaks for each peptide were chosen according to the following criteria: 1 ) three or more co-eluting fragment ions contributed to the peak signal , 2 ) at least two or more data points were collected across the peak , and 3 ) one or more spectra within the peak were matched to correct peptide sequence within the spectral database . If spectra within a specific sample were not identified then a ) the retention time of the chosen peak must be within 2 min of the retention time of an identified peak for that peptide from another sample and b ) the type of daughter ions contributing to the peak must match the identified peptide peak from another sample . If no peak matching these criteria was found in a particular sample the peak area was counted as zero . Chromatographic peak areas from all detected fragment ions for each peptide were integrated and summed to give a final peptide peak area . Because heavy peptide standards were not used in all targeted experiments , the intensities for each sample were normalized by the relative MYO7A 138-1 eluate intensities for each experiment ( Figure 4—source data 1 ) . Briefly , for each MYO7A peptide , the intensity for a given experiment was divided by the average of all experiments for that peptide . These peptide normalization factors were then averaged across all MYO7A peptides , yielding an experiment normalization factor . The area for each peptide analyzed was then divided by the appropriate experiment normalization factor . The normalized peak areas were then averaged for the 4–6 replicates , and divided by the average of all S7 starting material ( total ) experiments . This calculation yielded the IP/total value , which was displayed for each peptide in Figure 4 . We then used the Student's t-test to determine whether the results from the 138-1 experiments were different from the control experiments . All calculations , including the statistical analysis , are tabulated in Figure 4—source data 1 . All targeted mass-spectrometry data are available at: https://panoramaweb . org/labkey/project/OHSU%20-%20Barr-Gillespie%20Lab/MYO7A%20PDZD7%20complex/begin . view . HEK293 cells ( RRID:CVCL_0045 ) were maintained in Dulbecco’s Modified Eagle Medium ( DMEM; Thermo Fisher Scientific ) with glucose , pyruvate , and glutamine ( Sigma-Aldrich D6429 ) , supplemented with 10% FBS ( Atlanta Biologicals , Norcross , GA; #S11150 ) . Cells were seeded 24 hr before transfection in 6 well plates at 2 × 105 cells per well . Cells were transfected with 0 . 4 µg DNA using Effectene ( Qiagen , Hilden , Germany ) , then were harvested 48 hr later by removing medium , and freezing at -80°C . Proteins were extracted by agitating the plates with two washes of 500 µl RIPA buffer plus protease inhibitors ( Sigma-Aldrich #P8340 ) . The extract was cleared by centrifugation at 125 , 000 ×g ( rmax; 45 , 000 rpm in TLA55 ) for 15 min at 4°C . Immunoprecipitations were carried out using 100 µl of extract and 20 µl of 138-1 beads ( in PBS , 0 . 05% Tween-20 , and 0 . 5% BSA at 25 mg/ml with coupled antibody at 1 mg/ml ) . The beads and extract were rotated for 1 hr at room temperature , collected with a magnet , washed at least 5x with RIPA , and eluted 3x with 40 µl 2% SDS . All eluates were combined; 80 µl of sample buffer with DTT was added to bring the volume to 200 µl . Total extracts ( 200 µl ) were prepared from 100 µl of extract and 100 µl of 2x sample buffer; totals were diluted where necessary to achieve comparable signals to immunoprecipitates . The nanoscale pull-down assay was conducted in COS7 cells ( RRID:CVCL_0224 ) as described ( Bird et al . , 2016 ) . Plasmid constructs encoding the GFP-tagged motor , IQ , and neck domains of bovine MYO10 ( construct HMM-M10 ) was a gift from Richard Cheney ( Berg and Cheney , 2002 ) . The full-length cDNA clone ( TOH6003 ) encoding human MYO7A isoform 2 ( NP_001120652 . 1 ) was purchased from TransOMIC technologies ( Huntsville , AL ) . Complementary DNA sequence encoding the human MYO7A tail , encompassing MyTH4-1 , FERM-1 , MyTH4-2 , and FERM-2 domains ( the 1208 C-terminal amino acids ) , was introduced into cDNA clone TOH6003 and fused to the HMM-M10 construct using EcoR I . Plasmid clones encoding the chimeric protein were selected and fully sequenced; clone M107 was used in this study . COS7 cells were transfected using JetPrime transfection reagent ( Mirus Bio LLC , Madison , WI ) at 30–50% confluence . Immunocytochemistry was performed as described ( Grati and Kachar , 2011 ) using a Zeiss LSM710 confocal microscope equipped with a 63X , 1 . 4 NA objective . Mouse strains were C57BL/6 ( RRID:IMSR_JAX:000664 ) , as well as heterozygote and homozygote Myo7a8J ( RRID:MGI:2182923 ) . Inner ears from postnatal day eight ( P8 ) mice were removed and fixed for 3 hr in 4% paraformaldehyde ( in PBS ) . Ears were rinsed with PBS , and utricles were incubated in 50 µg/ml protease XXVII ( Sigma ) for 10 min to remove otolithic membranes . Organ of Corti tissues were dissected from the inner ears and the lateral wall was removed . Organs were rinsed 2 times in PBS , permeabilized for 10 min in 0 . 5% Triton-X in PBS , and blocked for 1 hr in PBS containing 5% normal donkey serum and 1% bovine serum albumin . Utricles were incubated overnight at 4°C in blocking solution containing primary antibodies . After three 5–10 min washes in PBS , utricles were incubated for 3 hr at room temperature in blocking solution containing donkey anti-mouse IgG AlexaFluor 488 ( Invitrogen , 1:1000 ) or donkey anti-rabbit IgG AlexaFluor 488 ( Invitrogen , 1:1000 ) secondary antibodies and CF633 Phalloidin ( Biotium , Fremont , CA; 1:500 ) . Utricles were washed 3 times for 10 min each in PBS , post-fixed in 2% paraformaldehyde in PBS for 5 min , rinsed twice with PBS and then mounted with VECTASHIELD ( Vector Laboratories , Burlingame , CA ) on slides with 0 . 12 mm Secure-Seal spacers ( Thermo Fisher Scientific #S-24737 ) . Cochlea images were acquired using a 100X , 1 . 46 NA Plan-Apochromat objective on a Zeiss Elyra PS . 1 system that reconstructs super-resolution images from a series of images acquired under spatially structured illumination ( SIM ) ( Gustafsson , 2000 ) . Images were processed for SIM reconstruction in Zen 2012 ( Zeiss ) and three-dimensional projections of each z-stack were visualized and converted into TIFF files using the 'Surpass' mode and 'Snapshot' tool on Biplane Imaris 7 . Utricle images were acquired using either standard confocal microscopy or Airyscan processing on a Zeiss LSM780 or LSM880 with Airyscan using a Plan-Apochromat 63X , 1 . 4 NA objective . Airyscan processing was carried out using Zen 2012 ( Zeiss ) . All z-stacks were processed using Bitplane Imaris 7 . Three dimensional projections of each z-stack were visualized using the 'Surpass' mode and were cropped along the x or y dimension to include one row of hair cells using the 'Crop 3D' tool . The cropped 3D projection was then rotated in Surpass mode to visualize a y-z or x-z slab of the stack with the hair bundles oriented vertically and the 'Snapshot' tool was used to generate a tiff image of the projection . Utricles were dissected and otoconia were removed with an eyelash . Square #1 . 5 glass coverslips ( Corning ) were washed with water and 70% ethanol , then autoclaved; they were then coated with 100 µg/ml poly-L-lysine for 10–20 min . Poly-L-lysine was removed and coverslips dried for 30–60 min . Dissected utricles were dropped onto the coverslips in dissection media and gently pressed to the coverslip with the hair bundle side down . The utricle and dissection solution was then removed and 200 µl of 4% formaldehyde ( Electron Microscopy Sciences ) in PBS was added to each coverslip for 20 min . Coverslips were rinsed three times with PBS and then placed on top of a petri dish lid inside a humidity chamber . Samples were permeabilized for 15 min in 0 . 2% Triton X-100 and 2% normal donkey serum in PBS , and blocked for 1 hr in 5% normal donkey serum in PBS . Coverslips were incubated overnight at 4°C with primary antibodies diluted in blocking solution , then rinsed 3x for 10 min each . Coverslips were then incubated for 3–4 hr in blocking solution with donkey AlexaFluor 488 , 568 and/or 647 ( 1:1000 , Invitrogen ) secondary antibodies ( unless primary antibody was directly labeled ) and 0 . 4 U/ml Alexa Fluor 568 Phalloidin ( Molecular Probes , Invitrogen ) or CF633 Phalloidin ( Biotium , 1:500 ) followed by three 10-min rinses with PBS . Coverslips were rinsed briefly in water; excess water was blotted on a paper towel , then coverslips were mounted on slides using EverBrite mounting medium ( Biotium ) . Isolated stereocilia images were acquired using a 100X , 1 . 46 NA Plan-Apochromat objective on the Zeiss Elyra PS . 1 system . Images were processed for SIM reconstruction in Zen 2012 ( Zeiss ) and selected Z-planes were exported as tiff images in Biplane Imaris 7 . For the analysis in Figure 8E , the line tool was used in Fiji to draw a line from the taper of each indicated stereocilium to its tip; the line was approximately the width of the stereocilia shafts . 'Plot Profile' was used to determine the pixel intensity along each stereocilium . In Excel , the profiles were aligned at the tapers , then all 11 profiles were averaged . Because each stereocilium is of a different length , only the taper and shaft regions are in alignment . In utero electroporation was carried out as described previously ( Ebrahim et al . , 2016 ) . Briefly , C57BL/6 males were crossed to CD1 females to generate embryos for transuterine microinjection . Pregnant females with E11 . 5 embryos were anesthetized with Nembutal ( 60–65 µg/gram body weight ) in a 20 . 8 mg/ml MgSO4 , 10% ethanol , and 40% propylene glycol . The uterine horn was exposed using a ventral laparotomy and embryos were visualized and positioned using a low-intensity halogen light . A microinjection pipette , backfilled with concentrated DNA plasmid ( >3 µg/µl ) , was secured in place in a pipette holder coupled to a Picospritzer III microinjector . The micropipette was inserted into the otocyst , and compressed nitrogen was used to deliver the inoculum . A square wave electroporator ( Protech International CUY21SC ) was then used to deliver a pulse of 60–100 mA to the injected embryo . Once the transuterine and electroporation steps were complete , the abdominal wall was closed using an absorbable suture . Immunocytochemistry was carried out as described above . Experiments were performed essentially as described previously ( Xiong et al . , 2012; Zhao et al . , 2014 ) . The organ of Corti was isolated from P3 mice and placed in DMEM/F12 medium with 1 . 5 µg/ml ampicillin . For electroporation , glass electrodes ( 2 µm diameter ) were used to deliver plasmid ( 500 ng/µl in 1x HBSS ) to the sensory epithelium . A series of three pulses was applied; pulses were at 1 sec intervals with a magnitude of 60 V and duration of 15 msec ( ECM 830 square wave electroporator; BTX ) . The tissue was fixed after two days in culture and stained with phalloidin; the mCherry signal was directly imaged for PDZD7 . | Inside the inner ear , sensory cells called hair cells detect and respond to sounds and head movements . These cells have a mechanically sensitive structure called the hair bundle , which is made of many thin projections called stereocilia . The stereocilia are linked so that when they bend in response to a sound or head movement , the whole hair bundle moves as one . A protein called myosin VIIA ( MYO7A ) is thought to be involved in forming links at the base of stereocilia ( so-called ‘ankle links’ ) and relaying signals from the stereocilia to the rest of the hair cell . However , it is not known how MYO7A interacts with the proteins that make up the ankle links . To address this question . Morgan , Krey et al . developed a new method for isolating groups of proteins from the inner ear of chick embryos that are only found in low quantities . Using this method , it was possible to isolate MYO7A along with other proteins it associates with . One of these proteins – called PDZD7 – is known to be part of ankle links . The next step following on from this work is to use this new method to study other important groups of proteins that are even more scarce in hair bundles . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2016 | PDZD7-MYO7A complex identified in enriched stereocilia membranes |
How thermal , mechanical and chemical stimuli applied to the skin are transduced into signals transmitted by peripheral neurons to the CNS is an area of intense study . Several studies indicate that transduction mechanisms are intrinsic to cutaneous neurons and that epidermal keratinocytes only modulate this transduction . Using mice expressing channelrhodopsin ( ChR2 ) in keratinocytes we show that blue light activation of the epidermis alone can produce action potentials ( APs ) in multiple types of cutaneous sensory neurons including SA1 , A-HTMR , CM , CH , CMC , CMH and CMHC fiber types . In loss of function studies , yellow light stimulation of keratinocytes that express halorhodopsin reduced AP generation in response to naturalistic stimuli . These findings support the idea that intrinsic sensory transduction mechanisms in epidermal keratinocytes can directly elicit AP firing in nociceptive as well as tactile sensory afferents and suggest a significantly expanded role for the epidermis in sensory processing .
Cutaneous primary sensory afferents are the first in a chain of neurons that convert environmental stimuli into recognizable sensations of touch , heat , cold and pain . Sensory neurons are diverse in nature and exhibit unique chemical , morphological and electrophysiological properties that allow specific responses to applied stimuli . In response to stimuli , the skin produces neuroactive substances that are postulated to directly and indirectly modulate the activity of sensory fibers ( Groneberg et al . , 2005 ) . These substances include glutamate ( Nordlind et al . , 1993; Fischer et al . , 2009 ) , ATP ( Cook and McCleskey , 2002; Inoue et al . , 2005; Dussor et al . , 2009; Barr et al . , 2013 ) , acetylcholine ( ACh ) ( Grando et al . , 1993; Wessler et al . , 1998 ) , epinephrine ( Khasar et al . , 1999; Pullar et al . , 2006 ) , CGRP ( Hou et al . , 2011 ) , neurotrophic growth factors ( Truzzi et al . , 2011 ) and cytokines ( Shi et al . , 2013 ) . The skin also expresses ligand-gated ( glutamate , ATP , nicotinic , muscarinic , 5-hydroxytryptamine , glycine and gamma-aminobutyric ) and voltage-gated ( sodium , calcium , transient receptor potential [TRP] , potassium and cyclic nucleotide ) ion channels and growth factor and cytokine receptors ( Olah et al . , 2012 ) . The expression of neuroactivators and voltage and ion-gated channels indicates that complex autocrine and paracrine signaling between epithelial and neural tissues underlie sensory signaling ( Conti-Fine et al . , 2000; Peier et al . , 2002; Zhao et al . , 2008; Atoyan et al . , 2009; Dussor et al . , 2009 ) . It has been proposed that non-neuronal cells of the skin , specifically keratinocytes , contribute to the initial transduction process through regulated release of neuroactive substances ( Zhao et al . , 2008; Dussor et al . , 2009; Mandadi et al . , 2009; Hou et al . , 2011; Barr et al . , 2013 ) . Testing this in an intact system has been difficult because the complexity in skin-nerve interactions prohibits isolation of the skin and neuronal output ( a behavioral reflex or the pattern of axonal firing ) since any natural stimulation ( e . g . , mechanical or thermal ) simultaneously affects both keratinocytes and sensory neurons . To address this problem , mice with targeted expression of light-activated channelrhodopsin ( ChR2 ) can be used to determine the contribution of each cell type to cutaneous associated behavior ( withdrawal reflex ) and generation of afferent APs . For example , Ji and colleagues ( Ji et al . , 2012 ) showed that blue light stimulation of the skin of transgenic rats that expressed ChR2 in primary afferents under the Thy-1 . 2 promoter exhibited nocifensive type responses . Similarly , Daou et al . ( Daou et al . , 2013 ) showed light-induced behavioral sensitivity in mice in which the Nav1 . 8 promoter drove expression of ChR2 in a subset of primary afferents . In another optogenetic model , Maksimovic and colleagues directed ChR2 expression to the non-neuronal Merkel cells of the epidermis . Using an ex vivo electrophysiologic preparation they showed that blue light stimulation of the isolated skin elicited AP trains in slowly adapting type 1 ( SA1 ) afferents , thus confirming the essential transducer role of Merkel cells in transmission of mechanical stimuli by SA1 tactile afferents . To further examine how the epidermis and cutaneous afferents communicate we analyzed mice in which ChR2 was targeted to either sensory neurons or keratinocytes to determine the contribution of each cell type to cutaneous associated behavior ( withdrawal reflex ) and generation of afferent APs . Similar to Daou et al . ( Daou et al . , 2013 ) , we found that light stimulation of the skin and activation of ChR2 in sensory afferents elicits robust nocifensive behaviors in mice . Remarkably , for mice that only express ChR2 in skin keratinocytes , light stimulation was also sufficient to generate nocifensive behaviors and regulate firing properties and evoke APs in specific subsets of cutaneous afferents , several which are known to activate in response to painful stimuli . In addition , expression of the chloride pump NpHR3 . 0 in keratinocytes significantly reduced AP firing in cutaneous afferents . These data indicate that Merkel cells are not unique in their ability to directly generate action potentials in sensory neurons and that light-mediated activation of keratinocytes is sufficient to engage an endogenous mechanism that can directly regulate cutaneous afferent firing .
In these electrophysiological experiments we have recorded from 200 characterized cutaneous afferents ( 86 C-fibers , 37 Aδ , 77 Aβ ) from the three different mouse genotypes ( 49 Prph-ChR2 , 80 KRT-ChR2 , 71 KRT-NpHR ) . The response properties to natural stimuli ( pressure , heat , cold ) for the different fiber types can be summarized as follows: Aβ-LTMRs had mechanical thresholds from 5 to 10 mN ( mean 5 . 5 mN ) , while Aδ-LTMRs thresholds ranged from 1 to 5mN , with a mean of 2 . 3 mN . For A-HTMRs , Aβ-HTMRs had mechanical thresholds ranging from 10 to 25 mN , with a mean of 17 . 5 mN; Aδ-HTMRs thresholds were 5–100 mN , with a mean of 26 . 7 mN . Cutaneous C-fibers showed a range of response properties , with mechanical thresholds from 5 to 50 mN ( mean 23 mN ) , heat thresholds of 37–50°C ( mean 44°C ) , and cold thresholds of 1–18°C ( mean 11°C ) . No significant differences in these values were observed between genotypes . We first determined the extent to which ChR2 activation in sensory neurons mimicked natural stimulation . Mice harboring a cre-responsive ChR2-YFP fusion gene in the Rosa locus ( Ai32 mice ) were crossed with peripherin ( Prph ) -cre mice to target ChR2 to unmyelinated and myelinated primary sensory neurons . The YFP tag allowed visualization of ChR2-positive projections in the skin and cell bodies in the dorsal root ganglion ( DRG ) of Prph-ChR2 mice ( Figure 1A , B ) . Myelinated and unmyelinated fibers expressed ChR2 as indicated by ChR2-YFP-positive fibers in the skin ( Figure 1A ) and physiological recordings ( Figure 1D , E ) . Behaviorally , all Prph-ChR2 mice ( 5 out of 5 mice tested ) demonstrated robust light-induced tail-flick or hindpaw withdrawal in <30 ms in response to a 473 nm laser light flash , consistent with previous findings ( Grando et al . , 1993; Daou et al . , 2013 ) . Wildtype littermate mice ( n = 5 ) were unresponsive . 10 . 7554/eLife . 09674 . 003Figure 1 . Light stimulates various types of cutaneous afferents in Prph-ChR2 transgenic mice . ( A ) . ChR2-YFP expression in unmyelinated and myelinated ( lanceolate endings of hair shaft , panels on right ) fibers of Prph-ChR2 mouse skin . Arrows indicate nerve fibers in dermis and epidermis ( Epi ) ; DAPI ( blue ) labeling demarcates keratinocytes . ( B ) . ChR2 is expressed in DRG neurons of Prph-ChR2 but not KRT-ChR2 mice . CGRP labels peptidergic neurons . ( C ) . Ex vivo preparation used for functional characterization of cutaneous afferents in response to mechanical , heat and laser stimulation . ( D ) . Response of a Prph-ChR2 Aδ-HTMR to mechanical and blue laser stimulation . ( E ) . Recordings from a CMHC nociceptor from a Prph-ChR2 mouse in response to mechanical , thermal and light stimulation . Calibration bars in ( A ) = 250 µm , ( B ) = 100 µm , ( E ) = 60 mV/1 s , top trace; 40 mV/1 s , bottom trace . DOI: http://dx . doi . org/10 . 7554/eLife . 09674 . 003 We then used an ex vivo skin/nerve/DRG/spinal cord preparation ( Figure 1C ) ( McIlwrath et al . , 2007; Lawson et al . , 2008 ) to characterize cutaneous afferent response properties in Prph-ChR2 mice ( Figure 1D , E ) . ChR2 neurons responded to blue light pulses ranging from 39 . 7 mW ( 5–10 , 000 ms ) to 0 . 7 mW ( 1000 ms pulse ) . Recordings were made from 49 characterized sensory neurons from 7 mice with 26 responders that included 1 A-fiber and 25 C-fiber nociceptors ( identified based on their response to noxious mechanical or thermal stimuli ) ( Table 2 ) . Among laser-responsive C-fibers , 21 responded to mechanical stimuli and of these , 14 responded to heat and/or cold stimuli . Four were classified as responding only to heat stimulation and 7 responded only to mechanical stimuli . Activation of Prph-ChR2 afferents revealed complex intrinsic firing properties . A Prph-ChR2 Aδ-HTMR ( A-delta-high threshold mechanoreceptor ) exhibited a tonic response to mechanical stimulation whereas blue light evoked a phasic response ( Figure 1D ) . In a CMHC nociceptor ( C-fiber responding to mechanical , noxious heat and cold stimuli ) , suprathreshold light stimulation produced tonic firing whereas suprathreshold mechanical stimulation evoked a more phasic response ( Figure 1E ) . Latency to first response to mechanical and light stimulation was similar . Peak instantaneous frequencies ( IF ) were significantly higher for suprathreshold mechanical stimulation , averaging 33 . 9 Hz for mechanical vs 8 . 6 Hz for light stimulation ( @ 39 . 7 mW ) for all mechanically responsive C-fibers . Interestingly , the average peak IF seen with laser light was similar to that seen in polymodal nociceptors ( the majority of cutaneous afferents ) in response to noxious heat ( McIlwrath et al . , 2007; Lawson et al . , 2008 ) . This raised the possibility that afferent-expressed ChR2 activation can evoke a ‘baseline’ response of putative nociceptors that reflects the intrinsic properties of these cells and that more naturalistic responses require collaboration of surrounding cells , including keratinocytes . To determine if keratinocytes contribute to afferent activation , mice that express ChR2-YFP ( ChR2 ) specifically in keratinocytes were generated by crossing Ai32 mice with Krt14 keratin Cre mice ( KRT14-Cre ) . KRT-ChR2 mice exhibited robust expression of ChR2 in epidermal keratinocytes and hair follicles of hairy skin and basal and suprabasal keratinocytes of glabrous skin ( Figure 2A ) . ChR2 expression does not occur in other dermal structures ( vasculature , muscle ) or in the DRG ( Figure 1B ) . KRT-ChR2 mice also exhibited behavioral responses to blue light stimulation ( Figure 2B , Table 1 ) , but at lower frequencies and with greater latencies relative to Prph-ChR2 mice . The average withdrawal latency for KRT-ChR2 mice was 15 . 75 s ± 2 . 26 ( SEM ) ( see Video 1 ) , compared to the millisecond withdrawal responses exhibited by Prph-ChR2 mice . Testing was done in a blinded manner and all KRT-ChR2 mice responded at least one time out of 10 trials with laser stimulation restricted to a 30 s maximum . Measures on human skin using a thermistor showed a slight laser-induced increase in surface temperature ( from 27 . 5°C to 30 . 5°C ) over the 30 s stimulation period , indicating that KRT-ChR2 mouse responses were not due to laser heating of the skin . That light activation of ChR2-keratinocytes could evoke nocifensive-type behaviors suggested that robust communication occurs between keratinocytes and sensory afferents that transmit nociceptive stimuli . 10 . 7554/eLife . 09674 . 005Figure 2 . Blue light stimulates multiple subtypes of cutaneous afferents in KRT-ChR2 transgenic mice . ( A ) . ChR2-YFP expression in keratinocytes of glabrous skin of KRT-ChR2 mouse . PGP9 . 5-positive nerve fibers ( red ) are in dermis and epidermis ( arrows ) . ( B ) . Plot of behavioral responses to blue laser across time intervals for Prph-ChR2 and KRT-ChR2 mice . All Prph-Cre mice showed an immediate response ( within 5 s of stimulation ) . All KRT-ChR2 mice also responded at least once in 10 trials and with variable latencies ( see Table 1 ) . ( C ) . Example showing activation of a CMH fiber type in response to blue laser applied to KRT-ChR2 skin in the ex vivo preparation . Responses of this fiber to mechanical and heat stimuli are shown below laser response . ( D ) . Example of a train of action potentials elicited in a CH fiber type in response to laser activation of the KRT-ChR2 skin . Responses of this fiber to heat stimuli are shown below laser response . ( E ) . In this KRT-ChR2 Aβ HTMR afferent laser stimulation does not produce firing when presented alone , but does in combination with subthreshold ( 5 mN ) mechanical stimulation . ( F ) . Light directly activates this KRT-ChR2 CMHC fiber and summates with noxious heat stimulation . ( G ) . SA1 Aβ-low threshold mechanoreceptor responds to mechanical and laser stimulation . ( H ) . SA1s terminate on ChR2-YFP ( green ) positive Merkel cells co-labeled with anti-K20 ( orange ) . Anti-NFH ( red ) labels SA1 fiber . Calibration bars in ( A ) and ( H ) = 100 µm . ( I ) . Light-evoked responses from a SA-1 fiber at varying intensities ( 1–40 mW ) with instantaneous frequency depicted . Pulses were 5 s in duration with 30 s between pulses . ( J ) . Normalized mean firing rate vs light intensity plotted on a log-intensity scale . Data from 8 afferents are averaged from ascending and descending steps of light intensity , and were fit with a Boltzman sigmoidal function ( R2 = 0 . 98 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09674 . 00510 . 7554/eLife . 09674 . 006Table 1 . KRT-ChR2 mice respond to blue light stimulation of paw skinDOI: http://dx . doi . org/10 . 7554/eLife . 09674 . 006Mouse strainSexResponses/10KRT-ChR2 1Female4KRT-ChR2 2Female3KRT-ChR2 3Female1KRT-ChR2 4Male3KRT-ChR2 5Male1KRT-ChR2 6Male3Mean2 . 5KRT-CreMale0KRT-CreMale0WTFemale0KRT-CreFemale0WTFemale0Mean0 . 0All KRT-ChR2 mice respond to light applied to foot plantar skin whereas control littermates ( n = 5 ) showed no response . The number of nocifensive responses ( paw lifting , biting , licking ) out of 10 stimulations was recorded . In total , light evoked responses in KRT-ChR2 mice in 17 of 60 total trials ( 28% ) . Control KRT-Cre mice lack the ChR2 gene whereas WT controls lack both transgenes . 10 . 7554/eLife . 09674 . 007Video 1 . KRT-ChR2 mice exhibit nocifensive behaviors in response to blue light . Blue light stimulation of channelrhodopsin expressing keratinocytes in the skin of KRT-ChR2 mice induces behavioral withdrawal responses . This mouse exhibits foot lifting at ∼9 s after light exposure on the glabrous skin of the hind foot . DOI: http://dx . doi . org/10 . 7554/eLife . 09674 . 007 To further investigate keratinocyte-sensory neuron communication we used ex vivo preparations that employed both intracellular and fiber teasing recording techniques . Electrophysiological recordings were obtained from 80 cells isolated from 16 KRT-ChR2 mice ( Table 2 ) . Laser activation induced APs in 6 out of 24 unmyelinated nociceptive fiber neurons ( Figure 2C , D ) and in 4 out of 14 myelinated high-threshold mechanoreceptors ( HTMRs ) ( not shown ) . These responses in heat-sensitive neurons are not due to laser-generated heat , as measures using a thermistor show minimal rise ( ∼1 °C ) in temperature over the 5 s recording interval . In addition , 3 myelinated HTMR fibers exhibited apparent summation when the laser was presented with natural stimuli . An example of this summation is shown in Figure 2E . This myelinated HTMR fiber had a mechanical threshold of 10 mN and neither a 5 mN mechanical stimulus nor the maximal intensity of blue light evoked a response . However , simultaneous application of 5 mN mechanical stimulation and light stimulation was sufficient to elicit APs . Recordings from 18 C-fiber nociceptors were maintained long enough to make multiple presentations of natural , laser and combined laser and natural stimuli . In 12 of these fibers , combined laser and natural stimulation evoked significantly more APs than natural stimuli alone ( p < 0 . 01 paired T-test , n = 12 ) ( Figure 2F ) . The remaining 6 C-fiber nociceptors did not display any summation when pairing laser and natural stimuli ( not shown ) . Comparison of the functional properties of laser responsive and unresponsive nociceptive fibers revealed no significant differences . Laser activation also elicited AP firing in all 21 myelinated slowly adapting type 1 ( SA1 ) low-threshold mechanoreceptors ( LTMRs ) , which is most likely due to activation of Merkel cells ( Maricich et al . , 2009; Maksimovic et al . , 2014 ) , which , like epidermal keratinocytes , express the KRT14 keratin ( Figure 2G–J ) . However , laser stimulation failed to activate any APs in myelinated rapidly adapting LTMRs . 10 . 7554/eLife . 09674 . 004Table 2 . Number of primary afferents recorded from Prph-ChR2 , KRT-ChR2 and KRT-NpHR mice that showed responses to light stimulationDOI: http://dx . doi . org/10 . 7554/eLife . 09674 . 004Prph-ChR2KRT-ChR2KRT-NpHRCell typeResponsiveUnresponsiveResponsive ( direct ) UnresponsiveResponsiveUnresponsiveSA10321 ( 21 ) 0160RA ( Aβ ) LTMR0401509RA ( Aδ ) LTMR010602A-HTMR ( Aβ ) 113 ( 2 ) 125A-HTMR ( Aδ ) 024 ( 2 ) 657CM701 ( 0 ) 124CC020101CH434 ( 2 ) 301CMC011 ( 0 ) 111CMH1136 ( 2 ) 075CMHC336 ( 2 ) 021Fibers that were activated directly by light stimulation of KRT-ChR2 keratinocytes are in parentheses . Cell types recorded from are: SA1 , slowly adapting type 1; RA ( Aβ ) , rapidly adapting A beta low-threshold mechanoreceptor; RA ( Aδ ) , rapidly adapting A delta low-threshold mechanoreceptor , A-HTMR , high-threshold mechanoreceptor ( Aβ ) ; A-HTMR , high-threshold mechanoreceptor ( Aδ ) ; CM , C mechanoreceptor; CC , C cold receptor; CH , C heat receptor; CMC , C mechano-cold receptor; CMH , C mechano-heat receptor; CMHC , C mechano-heat and cold receptor . AP firing following laser stimulation of keratinocytes was generally less robust than AP firing in Prph-ChR2 afferents ( avg peak IF = 0 . 3Hz vs 8 . 6Hz , respectively ) . The exception was in recordings from SA1 fibers , which showed a robust , but atypical pattern of firing to light stimulation ( Figure 2G , I ) . In response to mechanical stimulation SA1 fibers exhibit a characteristic response consisting of an initial high frequency burst of action potentials followed by a sustained firing , but at a lower frequency . Although light stimulation of these fibers could evoke high frequency bursts of activity , these bursts did not occur at the initial onset of the light stimulus ( mechanical mean peak IF = 218 . 2Hz; optical mean peak IF = 109 . 2 Hz ) ( Figure 2G ) . The SA1 response to light was stable , could be elicited repeatedly and was intensity dependent ( Figure 2I , J ) . To confirm that KRT-ChR2 keratinocytes are indeed activated by blue light , we examined the electrophysiological properties of these cells using whole cell patch clamp analysis . Keratinocytes do not normally generate APs , but they do have resting membrane potentials generated by currents mediated by ion ( e . g . , K+ , Cl− ) channels intrinsic to the plasma membrane ( Wohlrab et al . , 2000 ) . Patch clamp recordings were made from keratinocytes isolated from adult tail skin of KRT-ChR2 mice ( Figure 3A ) . Recordings from 11 ChR2-YFP keratinocytes all showed inward current in response to a brief ( 1 s ) flash of blue light ( peak current: median 26 . 3 pA; steady current: 16 . 5 pA ) ( Figure 3B , C ) . No light-induced currents were recorded in keratinocytes cultured from wildtype mice ( n = 4 cells ) . 10 . 7554/eLife . 09674 . 008Figure 3 . Light elicits current activation in cultured keratinocytes . ( A ) . Fluorescent ChR2-YFP protein in plasma membrane of keratinocytes cultured from skin of KRT-ChR2 mice . ( B ) . IR-DIC images of patch pipette on single keratinocyte that was recorded from and then filled with Alexa 555 dye . ( C ) . Representative trace illustrates typical current evoked by blue light stimulation of KRT-ChR2 . Yellow light stimulation of KRT-NpHR keratinocytes also produced a change in voltage properties of the cell . Control KRT-Cre keratinocytes that were isolated in parallel showed no response to light ( not shown ) . Bar in A is 40 µM . DOI: http://dx . doi . org/10 . 7554/eLife . 09674 . 008 A loss of function approach using transgenic mice that express halorhodopsin ( eNpHR3 . 0 , ‘NpHR’ ) in keratinocytes was also used to further demonstrate the role of epidermal cells in afferent activation . Halorhodopsin is a yellow-to-red light-activated chloride pump that when expressed in neurons generates hyperpolarization , inhibits AP firing and neural activity ( Raimondo et al . , 2012 ) . Using keratinocyte cultures from KRT-NpHR mice we recorded from 5 cells that all exhibited a hyperpolarizing response to orange light illumination . The median hyperpolarization was −1 . 1 mV . Using ex vivo preparations employing intracellular and fiber teasing techniques , 46 myelinated and 25 unmyelinated cells were recorded from 5 KRT-NpHR mice ( Table 2 ) . Application of yellow laser ( 589 nm ) to the skin reduced AP firing in response to mechanical or heat stimulation in 12 of 25 C-fiber nociceptors and 7 of 19 myelinated nociceptors ( Figure 4 ) . This reduction was fiber type dependent with the most pronounced effects in mechanically sensitive C-fiber nociceptors ( p = 0 . 02 Paired T-test n = 7 ) and slowly adapting type I LTMRs ( p < 0 . 01 Paired T-Test n = 10 ) ( Table 2 ) . There were no effects observed on myelinated rapidly adapting LTMRs . It should also be noted that while in some presentations this yellow light-induced reduction in firing was 100% ( Figure 4A , B ) , the average reduction in affected fibers was lower , that is , 44% in C fibers ( n = 12 ) , 48% in A-HTMRs ( n = 7 ) and 44% in 16 SA1 fibers . In addition , in some cases where 100% reduction was observed , on subsequent light exposures the reduction in firing was less pronounced ( Figure 4B ) . 10 . 7554/eLife . 09674 . 009Figure 4 . Yellow light inhibits AP firing in multiple subtypes of cutaneous afferents in KRT-NpHR mice . ( A ) . Yellow light decreases AP firing in response to mechanical stimulation in this Aδ-HTMR afferent . ( B ) . In this CMH-fiber the response to mechanical stimulation is decreased with the initial yellow laser stimulation; a smaller decrease in AP firing occurred with a second laser presentation . ( C ) . This CMH-fiber showed decreased firing in response to heat in the presence of yellow laser stimulation . ( D ) . Responses of a SA1 fiber to mechanical stimulation are significantly reduced by activation of NpHR in epidermal keratinocytes ( which are likely Merkel cells ) . Laser stimuli ( orange bars ) occurred 1 s prior to mechanical ( black bar ) or heat ( red bar ) stimuli . Duration of each stimulus was either 5 s ( mechanical and heat ) or 6 s ( laser ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09674 . 009
These studies show in an intact skin preparation that ChR2-induced stimulation of skin keratinocytes , in isolation from other cells , is sufficient to induce AP firing in several types of sensory neurons . For some neuron subtypes , light activation of keratinocytes induces action potential firing similar to that evoked in response to natural stimuli . For other afferents , keratinocyte activation produced sub-threshold effects that potentiated the response to natural stimulation . For example , we recorded from afferents where light activation of keratinocytes alone did not elicit action potentials , but when combined with sub-threshold mechanical stimuli , produced multiple action potentials . These results suggest that keratinocytes are not only intimately involved in the generation of sensory neuron activity , but that the nature of this interaction is heterogeneous , differing for the many subtypes of sensory neurons that innervate the skin . Contributing to this heterogeneity may be the type or relative level of neuroactivator compound released by keratinocytes in response to mechanical , thermal or noxious stimulation or interactions with other cell types or structures in the skin , for example , immune cells or vascular structures . Our electrophysiologic findings indicate that activation of Aδ and C fiber nociceptors likely underlies the behavioral sensitivity evoked by light in KRT-ChR2 mice . In addition , light stimulation of ChR2 expressed by Merkel cells likely transduces a signal that directly activates SA1 low threshold mechanoreceptors , as shown by Maksimovic ( Maksimovic et al . , 2014 ) . That ChR2 in epidermal cells other than Merkel cells can activate numerous neuronal subtypes that are known to transmit thermal , mechanical and painful stimuli significantly expands the role of the epidermis in sensory processing . The ability of keratinocytes to signal to sensory afferents and transmit pain is also supported by recent findings of Pang and colleagues ( Pang et al . , 2015 ) . In these studies TRPV1 global knockout mice were genetically engineered to ectopically express TRPV1 selectively in keratinocytes . In these mice capsaicin could evoke nocifensive behaviors and c-fos expression in spinal cord dorsal horn neurons . As capsaicin application should only have activated keratinocyte-expressed TRPV1 , it was concluded that these responses , which require activation of nociceptors , were initiated by keratinocytes , which in turn induced firing in primary afferents . Afferents that fire APs in response to light stimulation of keratinocytes were either polymodal , responding to mechanical and thermal stimuli , or unimodal , responding only to mechanical or thermal stimuli . For example , over half of the C-heat ( CH ) fibers , which only signal noxious heat and express TRPV1 ( Jankowski et al . , 2012 ) , responded to keratinocyte activation . This suggests that keratinocytes have the ability to communicate directly with neurons that express TRPV1 , an ion channel that transmits noxious heat and is required for inflammatory pain signaling ( Woodbury et al . , 2004; Baumbauer et al . , 2014 ) . Interestingly , LTMRs afferents , which form lanceolate endings around hair follicles ( Figure 1A ) , were not activated by illumination of the skin in either Prph-ChR2 mice or KRT-ChR2 mice . A possible reason for this may be that these nerve fibers and/or the associated keratinocytes were not effectively illuminated due to the depth of the skin . However , in ongoing studies using Advillin-cre- and trkB-CreER-ChR2 mice , action potentials can be evoked in Aβ and Aδ LTMRs using the same light stimulus ( not shown ) . Thus , it is possible that in Prph-ChR2 mice , an insufficient level of ChR2 for activation of LTMRs may exist . Another possibility is that the peripherin promoter only targets C-LTMR afferents . Unfortunately , the only cells we recorded from with lanceolate endings in these prreparations were myelinated RA-LTMRs . In addition to the afferent stimulation , afferent activity could also be repressed by optogenetic stimulation of epidermal cells expressing NpHR . Light stimulation of NpHR and the predicted intracellular directed Cl flux led to significant reduction in many C-fiber , Aδ and SA1 afferent responses to mechanical and/or heat stimulation of the skin . Although the physiological and cellular mechanisms underlying this chloride-mediated change in keratinocyte signaling are yet to be resolved , the reduction in AP activity suggests a possible role for Cl− in mediating neural-keratinocyte communication . Keratinocytes are known to exhibit chloride conductance ( Rugolo et al . , 1992 ) , and Cl− has been shown to contribute to changes in resting potential ( Wohlrab et al . , 2000 ) and keratinocyte hyperpolarization in response to mechanical stimuli evoked by hypotonic stress ( Gonczi et al . , 2007 ) . Future studies , to determine if KRT-NpHR mice exhibit reduced behavioral responses in response to noxious stimuli , will require a system in which dual presentation of the stimulus , for example , heat and yellow light , are delivered . The afferent responses evoked by keratinocyte stimulation were not at the same level evoked by natural stimuli , but this was not expected . It is most likely that keratinocyte activation is one contributor to natural stimuli-evoked sensory signaling ( at least for some cells ) and , in addition to neuronal activation , is a critical component of sensory transmission . Evidence for this is the clear activation of primary afferents by blue light stimulation of keratinocytes and the observed summation of AP firing in afferents exposed to light and mechanical or thermal stimuli . Importantly , physiological relevance is also indicated by the in vivo nocifensive behavior and clear withdrawal response elicited by light stimulation of KRT-ChR2 mice . These responses were much slower compared to behavioral response times measured in Prph-ChR2 mice , which express the ChR2 ion channel in the primary afferent . This difference may reflect the time needed for release by keratinocytes of neuroactivator compound ( s ) to a level sufficient to evoke an AP as well as the heterogeneity of fiber types innervating the epidermis . Further study of the types of neuroactivator compounds released by light stimulated ChR2 keratinocytes and the effect of these activators on specific types of primary afferents will address these issues . Disturbances in epidermal-neuronal signaling in inflamed or damaged skin result in abnormal sensory transmission that underlies associated pain , itch and paresthesia ( Urashima and Mihara , 1998; Kinkelin et al . , 2000 ) . The present findings support the idea that keratinocytes , as activators of cutaneous neurons , have a central role in the onset and maintenance of such abnormal transmission . These findings also suggest that altered release of keratinocyte expressed neuromodulators ( e . g . , ATP , CGRP ) , neurotransmitters ( e . g . , ACh ) or activity of neurotransmitter receptors and ion channels could drive changes in transmission and importantly , may do so in a neuron subtype specific manner .
Male and female mice ages 6–10 wks were used . Mice expressing ChR2 in sensory neurons were generated by crossing Ai32 mice with peripherin-Cre mice ( Zhou et al . , 2002 ) , which were generously provided by Dr . Rebecca Seal ( Department of Neurobiology , University of Pittsburgh ) . Transgenic mice that express ChR2 in keratinocytes were generated by crossing Ai32 mice ( B6;129S-Gt ( ROSA ) 26Sortm32 . 1 ( CAG-COP4*H134R/EYFP ) Hze/J ) with KRT14-Cre mice ( Tg ( KRT14-cre ) 1Amc/J ) , both obtained from Jackson Laboratories ( Bar Harbor , ME ) . Mice expressing halorhodopsin ( eNpHR3 . 0-EYFP ) in keratinocytes were generated by crossing Ai39 mice ( B6;129S-Gt ( ROSA ) 26Sortm39 ( CAG-HOP/EYFP ) Hze/J ) with KRT14-Cre mice . All experiments were approved by the Institutional Animal Care and Use Committee at the University of Pittsburgh ( protocol # 14074296 ) . Skin and dorsal root ganglia were post-fixed in 4% paraformaldehyde , cryoprotected in 25% sucrose , embedded in gelatin , sectioned on a sliding microtome and labeled using target-specific antibodies followed by a fluorescently tagged secondary . Sections were stained with antibodies to keratin K20 ( 1:20 , mouse; Signet Covance , MA ) , NF145 ( 1:200 , rabbit; Millipore , MA ) or PGP9 . 5 ( 1:1000 , rabbit; Ultraclone , UK ) followed by appropriate secondary antibodies ( Jackson ImmunoResearch ) used at 1:500 dilution . Fluorescent images were captured using a digital camera attached to a Leica DM4000B fluorescence microscope ( Leica , Wetzlar , Germany ) and processed for brightness and contrast using Adobe Photoshop . Laser-induced paw withdrawal latency was measured using an 80 mW , 473 nm wavelength laser from a distance of 8–10 mm while animals were confined in a glass container . For KRT-ChR2 and control mice the number of nocifensive responses ( paw lifting , biting , licking ) out of 10 stimulations was recorded . Comprehensive phenotyping of individual afferents was done using an ex vivo skin/nerve/DRG preparation as previously described ( McIlwrath et al . , 2007 ) . Mice were anesthetized with ketamine/xylazine mixture ( 90/10 mg/kg , respectively ) and perfused with oxygenated artificial cerebrospinal fluid ( aCSF ) . The hairy skin of one hindpaw , saphenous nerve , DRGs , and spinal cord were dissected in continuity and placed in a bath of warm ( 31°C ) circulating oxygenated aCSF . The skin was placed on an elevated metal platform exposing the epidermis to air for mechanical , thermal and laser stimulation . Electrophysiological recordings were performed by impaling individual neuronal somata using sharp quartz microelectrodes . Electrical stimuli were delivered through a suction electrode on the nerve to locate sensory neurons that innervate the skin . Receptive fields were localized and characterized based on responses to mechanical and/or thermal stimulation . Responsiveness to laser stimulation was determined using an 80 mW , 473 nm wavelength laser ( to activate ChR2 ) or a 34 mW , 589 nm wavelength laser ( to activate halorhodopsin ) ( Laserglow Technologies , Toronto , Canada ) affixed to a micromanipulator . The distance from the skin was adjusted to produce a 1–2 mm diameter illuminated area . In the KRT-ChR2 experiments blue light and mechanical or thermal stimuli were applied simultaneously . The tip of the mechanical stimulator is 1 mm in diameter and typically did not block the entire receptive field available for laser stimulation . In addition , the light was delivered at a 45O angle , allowing penetration of the skin beneath the probe . In the KRT/HpHR experiments the yellow light preceded the natural stimulus by 1 s . Neurons with conduction velocities < 1 . 2 m/s were classified as C-fibers , while all others were classified as A-fibers . Fiber teasing experiments were performed using previously established protocols ( Zimmermann et al . , 2009 ) to further examine afferents in KRT-ChR2 and KRT-NpHR mice . The preparation was prepared in the same manner as the skin/nerve/DRG preparation , except the saphenous nerve was cut slightly proximal to the junction with the femoral nerve . Recordings were performed using a bipolar platinum electrode , and stimuli were administered to the epidermis . Adult mouse keratinocytes were cultured following the procedure of ( Redvers and Kaur , 2005 ) . Tail skin was digested in dispase II ( 8 mg/ml dissolved in HBBS containing 1% pen/strep ) overnight at 4°C . The epidermal sheet was removed , digested in trypsin-ethylenediamine acid solution ( Life Technologies , Waltham , MA ) and the dissociated cells plated onto 12 mm glass coverslips coated with type 4 collagen at 104 cells/coverslip . Cells were cultured in Keratinocyte Serum Free Medium ( K-SFM , Life Technologies ) supplemented with 0 . 1% pen/strep , 10 ng/ml epidermal growth factor and 0 . 1 nM cholera toxin . Patch clamp recordings were performed at 7–14 d post plating . Whole cell patch clamp recordings were made on keratinocytes grown on coverslips exposed to a one second blue light pulse . Keratinocytes on coverslips were transferred to a recording chamber that was continuously perfused with extracellular bath solution containing ( in mM ) : NaCl 140 , KCl 5 . 4 , CaCl2 1 . 8 , MgCl2 1 . 0 , HEPES ( N-2-hydroxyethylpiperazine-N'-2-ethanesulfonic acid ) 10 . 0 and D-glucose 11 . 1 ( Inoue et al . , 2005 ) . The pH was adjusted to 7 . 4 with NaOH . Cells were visualized using a microscope with infrared differential interference contrast ( IR-DIC ) optics ( Olympus , Pittsburgh , PA , BX-51WI ) . Patch pipettes made from borosilicate thin walled glass capillaries ( Warner Instruments , G150F-6 ) using a P-97 micropipette puller ( Sutter Instrument Company , Novato , CA ) had a tip resistance of 10–15 MΩ . The composition of pipette solution was ( in mM ) ; 135 potassium gluconate , 5 KCl , 0 . 5 CaCl2 , 5 EGTA , 5 Hepes , 5 ATP-Mg , 0 . 025 Alexa 555 , pH 7 . 2 . All experiments were conducted at room temperature ( 19°C ) . Whole-cell patch clamp recordings were made using an Axopatch 200B amplifier ( Molecular Devices , Sunnyvale , CA ) . The currents were clamped at −50 mV and a one second blue light pulse was delivered from a xenon light source ( Lambda DG-4 , Sutter Instrument Company ) using a 40x water immersion objective and GFP filter set . Data were digitized using a Digidata 1322A ( Molecular Devices ) and stored and analyzed using pClamp 10 software ( Molecular Devices ) . | When a person touches a hot saucepan , nerve cells in the skin send a message to the brain that causes the person to pull away quickly . Similar messages alert the brain when the skin comes in contact with an object that is cold or causes pain . These nerve cells also help to transmit information about other sensations like holding a ball . Scientists believe that skin cells may release messages that influence how the nerves in the skin respond to sensations . But it is difficult to distinguish the respective roles of skin cells and nerve cells in experiments because these cells often appear to react at the same time . Researchers have discovered that a technique called optogenetics , which originally developed to study the brain , can help . Optogenetics uses genetic engineering to create skin cells that respond to light instead of touch . Baumbauer , DeBerry , Adelman et al . genetically engineered mice to express a light-sensitive protein in their skin cells . When these skin cells were exposed to light , the mice pulled away just like they would if they were responding to painful contact . This behavior coincided with electrical signals in the nerve cells even though the nerve cells themselves were not light sensitive . In further experiments , mice were genetically engineered to express another protein in their skin cells that prevents the neurons from being able to generate electrical signals . When these skin cells were exposed to light , the surrounding nerve cells produced fewer electrical signals . Together , the experiments show that skin cells are able to directly trigger electrical signals in nerve cells . Baumbauer , DeBerry , Adelman et al . 's findings may help researchers to understand why some patients with particular inflammatory conditions are in pain due to overactive nerve cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Keratinocytes can modulate and directly initiate nociceptive responses |
The human gut microbiota impacts host metabolism and has been implicated in the pathophysiology of obesity and metabolic syndromes . However , defining the roles of specific microbial activities and metabolites on host phenotypes has proven challenging due to the complexity of the microbiome-host ecosystem . Here , we identify strains from the abundant gut bacterial phylum Bacteroidetes that display selective bile salt hydrolase ( BSH ) activity . Using isogenic strains of wild-type and BSH-deleted Bacteroides thetaiotaomicron , we selectively modulated the levels of the bile acid tauro-β-muricholic acid in monocolonized gnotobiotic mice . B . thetaiotaomicron BSH mutant-colonized mice displayed altered metabolism , including reduced weight gain and respiratory exchange ratios , as well as transcriptional changes in metabolic , circadian rhythm , and immune pathways in the gut and liver . Our results demonstrate that metabolites generated by a single microbial gene and enzymatic activity can profoundly alter host metabolism and gene expression at local and organism-level scales .
The human gut microbiome is known to play a crucial role in human energy harvest and homeostasis ( BäckhedBackhed et al . , 2004; Turnbaugh et al . , 2006 ) . Lean and obese people harbor different gut bacterial communities , suggesting that developing gut bacterial imbalances may contribute to obesity ( Ley et al . , 2006; Turnbaugh et al . , 2006; Turnbaugh et al . , 2008 ) . Importantly , transplantation of the fecal microbiota from obese humans to germ-free ( GF ) mice has been shown to result in the development of obesity-associated metabolic phenotypes in recipient mice ( Ridaura et al . , 2013 ) . These studies establish a causal relationship between gut bacteria and host metabolic status . The molecular mechanisms by which gut microbes regulate host metabolism , however , remain largely unknown . This lack of mechanistic understanding regarding the functions of microbial species and their metabolic capabilities has limited the effectiveness of both dietary and therapeutic approaches to improving host physiology ( Jia et al . , 2008; Wallace et al . , 2010 ) . The investigation of microbial metabolite production represents both an important opportunity and a challenge in the search to uncover the causal underpinnings of the effects of gut bacteria on host metabolism . One of the most concrete effects that human-associated bacteria have on the host is the production of small molecule metabolites , some of which accumulate to levels in the body higher than that of a typical drug ( Donia and Fischbach , 2015 ) . Recent research suggests that bacterial metabolites play important roles in host metabolism by regulating host glucose and energy homeostasis ( De Vadder et al . , 2014; Gao et al . , 2009; Todesco et al . , 1991 ) . The complexity of gut microbial ecosystems and associated microbial and host-derived microbial metabolites , however , presents significant obstacles on the path to defining how individual compounds elicit specific in vivo effects . Means to control specific metabolites is critical to understanding how these molecules affect host physiology . In this work , we selectively modulate the in vivo levels of bile acids and demonstrate that this controlled alteration of the metabolite pool exerts distinct effects on host physiology . Bile acids are steroidal natural products that are synthesized from cholesterol in the liver and constitute an important part of the molecular environment of a healthy human gut ( Ridlon et al . , 2006 ) . Upon ingestion of a meal , bile acids are secreted from the liver and gallbladder into the duodenum where , with the activities of pancreatic enzymes , they form micelles that solubilize lipids and fat-soluble vitamins that are otherwise poorly absorbed . Remaining free bile acids are efficiently reabsorbed from the ileum via the action of bile acid transporters and recirculated back to the liver . Approximately 3–5% of bile acids escape enterohepatic recirculation and enter the colon at a rate of 400–800 mg/day , forming a concentrated pool of metabolites ( 200 to 1000 μM ) ( Hamilton et al . , 2007 ) . In the colon , these molecules are modified by the resident bacteria in near-quantitative fashion , forming a class of on the order of 50 different metabolites called secondary bile acids ( Figure 1A ) . In addition to their role in digestion , many primary and secondary bile acids act as ligands for host nuclear receptors , including the farnesoid X receptor ( FXR ) , the pregnane X receptor ( PXR ) , the vitamin D receptor ( VDR ) , the liver X receptor ( LXR ) and the G-protein-coupled receptor TGR5 ( Fiorucci and Distrutti , 2015; Katsuma et al . , 2005; Makishima et al . , 2002; Song et al . , 2000; Staudinger et al . , 2001 ) . By acting as agonists or antagonists for these receptors , bile acids further impact the regulation of glucose tolerance and homeostasis , insulin sensitivity , lipid metabolism , triglyceride and cholesterol levels , and energy expenditure by the host ( Fiorucci and Distrutti , 2015; Modica et al . , 2010 ) . Additionally , bile acids regulate their own biosynthesis via an FXR-mediated negative feedback mechanism , which affects downstream nutrient availability for the host ( Modica et al . , 2010 ) . As a result of these interactions , bile acid imbalance has been implicated as having a causal effect in the development of diet-induced obesity ( Fiorucci and Distrutti , 2015 ) . Conversely , modification of the bile acid pool by commensal bacteria has been suggested to induce beneficial changes in host metabolism ( Joyce et al . , 2014 ) . The mechanisms underlying these effects , however , remain largely undefined . Due to the large number of compounds and receptors involved as well as the additional role of bile acids as biological detergents , the in vivo roles of specific bile acids have been difficult to untangle . Our novel approach to deconvoluting the physiological role of structurally distinct bile acids is to control the in vivo activity of selective bacterial bile salt hydrolases ( BSH ) . BSH hydrolyze conjugated bile acids that have been linked to either taurine or glycine by host liver enzymes , revealing unconjugated bile acids ( Figure 1A ) . This deconjugation step occurs prior to subsequent bacterial conversion of primary bile acids ( e . g . cholic acid and chenodeoxycholic acid ) to secondary bile acids ( e . g . deoxycholic acid and lithocholic acid ) ( Ridlon et al . , 2006 ) . Prior work suggests that BSH play a critical role in regulating host metabolism . However , these studies have not yet uncovered how specific bile acid metabolites exert their in vivo effects on host metabolism , and conflicting results have been reported regarding whether BSH activity should be increased or decreased to achieve host metabolic benefits ( Joyce et al . , 2014; Li et al . , 2013 ) . Research efforts to date have either examined correlative relationships between BSH activities , bile acid levels , and metabolic indications ( Li et al . , 2013 ) or investigated the metabolic effects of ‘unconjugated’ versus ‘conjugated’ groups of bile acids ( Joyce et al . , 2014 ) . It is imperative to be able to differentiate bile acids in vivo based on their structure in order to understand their effects on host metabolism . As an important example , taurocholic acid ( TCA ) and tauro-β-muricholic acid ( TβMCA ) are both conjugated bile acids but exert different physiological effects: TCA is an FXR agonist , while TβMCA is an FXR antagonist ( Figure 1B ) ( Li et al . , 2013; Sayin et al . , 2013 ) . Herein , we uncover a group of bacteria within the abundant human gut commensal genus Bacteroides that possess selective BSH activity . We then identify the gene responsible for this activity in Bacteroides thetaiotaomicron and construct a knockout strain . By monocolonizing germ-free ( GF ) mice with the wild-type or BSH-deleted strain , we demonstrate that we can predictably alter the in vivo bile acid pool using this selective enzyme and that this change has significant effects on host metabolic status . Our results demonstrate that the deletion of a single bacterial gene can exert significant effects on host metabolism in a gnotobiotic environment and highlight the importance of modulating specific compounds when seeking to understand the effects of bacterial metabolites on host physiology .
BSH ( EC 3 . 5 . 1 . 24 ) are found across a wide range of bacterial genera from the two dominant gut phyla , Bacteroidetes and Firmicutes ( Jones et al . , 2008 ) . However , the structural and activity characterization of these enzymes has been largely limited to Gram-positive species ( i . e . Clostridia , Lactobacillus , Bifidobacterium , Listeria ) ( Begley et al . , 2006; Rossocha et al . , 2005 ) . These enzymes largely demonstrate non-selective activities , cleaving all conjugated bile acids independent of either the bile acid core or amino acid conjugate ( taurine or glycine ) ( Ridlon et al . , 2006 ) . While differential reactivity toward conjugated substrates has been observed in some Gram-positive strains , in these cases , the selectivity has been based on a preference for one amino acid over the other , not on the structure of the steroidal core ( De Boever P and Verstraete , 1999; Grill et al . , 1995; Kim et al . , 2004; Ridlon et al . , 2006 ) . In contrast , the activity of Gram-negative bacteria has been largely underexplored . While Bacteroides fragilis ATCC 25285 was reported to exhibit non-selective BSH activity ( Stellwag and Hylemon , 1976 ) , some Bacteroides vulgatus strains were observed to cleave taurochenodeoxycholic acid ( TCDCA ) and TβMCA but minimally cleaved TCA ( Chikai et al . , 1987; Kawamoto et al . , 1989 ) , thus exhibiting a degree of selectivity based on the hydroxylation pattern of the steroid . These results suggested to us that perhaps other strains within the phylum Bacteroidetes might display steroidal core-based selectivity . To investigate this question , we performed a screen of the BSH activity of twenty Bacteroidetes strains found in the human gut ( Figure 2A and Figure 2—figure supplement 1 and 2 ) ( Kraal et al . , 2014 ) . We also tested Clostridium perfringens and Lactobacillus plantarum , two Gram-positive species with known non-selective BSH activities , for comparison . We incubated pre-log phase cultures of individual strains with a group of either the most abundant tauro- or the most abundant glyco-conjugated bile acids found in the human and murine GI tracts . We monitored deconjugation over time by UPLC-MS and determined that all hydrolysis reactions had reached steady state by 48 hr ( Figure 2B , Figure 2—figure supplement 3 ) . We then quenched the cultures and profiled bacterial bile acid metabolism . As expected , C . perfringens ATCC 13124 , L . plantarum WCFS1 , and B . fragilis ATCC 25285 deconjugated all conjugated bile acid substrates tested . Strikingly , the majority of Bacteroidetes strains tested displayed some degree of selectivity for conjugated bile acid substrates , with a preference for deconjugating tauro- over glyco-conjugated substrates . A subset of these strains ( B . thetaiotaomicron VPI-5482 , B . caccae ATCC 43185 , B . fragilis 638R , Bacteroides sp . D2 , and Bacteroides sp . 2_1_16; Group I – red , Figure 2A ) exhibited selectivity exclusively based on the steroidal core structure , deconjugating C12 = H primary bile acids ( i . e . TCDCA , GCDCA , and TβMCA ) but not C12 = OH primary bile acids ( i . e . TCA and GCA ) . To our knowledge , this study represents the first systematic evaluation of BSH activity in the common gut-bacterial phylum Bacteroidetes . Given that specific conjugated and unconjugated bile acids bind to different host receptors and have the potential to exert different downstream effects , the selectivity uncovered here may have important physiological consequences depending on which Bacteroides species colonize the host . To further explore this possibility and define the effects of selective BSH on host physiology , we monocolonized GF mice with isogenic strains of wild-type and BSH-deleted B . thetaiotaomicron as described below . We recognized that deletion of the BSH enzyme from one of the Group I Bacteroides species would provide us with a paired set of isogenic strains ( wild-type and knockout ) that would allow us to rationally manipulate the in vivo bile acid pool in a highly specific manner . In mice , the two most abundant primary bile acids are TCA and TβMCA ( Sayin et al . , 2013 ) . Based on the observed selectivity for deconjugating C12 = H but not C12 = OH core primary bile acids , we predicted that colonization with a BSH wild-type strain would result in lower levels of TβMCA ( C12 = H ) relative to knockout colonized mice , while the levels of TCA ( C12 = OH ) in both groups would remain constant . All the five Group I strains displayed weak-to-moderate deconjugation of TβMCA in vitro ( Figure 2A ) . Importantly , we did not detect any products of TCA deconjugation from any of these strains . This result suggested that the levels of deconjugated CA in mice colonized with these bacteria would remain low to undetectable , while the levels of deconjugated βMCA could build up due to enterohepatic recirculation . We decided to focus our efforts on generating paired isogenic strains in one of these species , B . thetaiotaomicron ( Bt ) . Although this strain displayed relatively weak TβMCA-deconjugating activity , Bt had been previously shown to be amenable to genetic manipulation , allowing knockout of putative BSH genes ( Cullen et al . , 2015; Koropatkin et al . , 2008 ) . We performed a BLASTP search of the characterized BSH from C . perfringens ( Ridlon et al . , 2006 ) against the Bt genome and identified two genes , BT2086 and BT1259 , as putative BSH . We constructed unmarked deletions of these genes using allelic exchange and then tested the resultant mutants for their ability to deconjugate bile acids in whole cell culture using UPLC-MS . The BtΔ2086 mutant ( henceforth referred to as Bt KO ) had lost the ability to cleave conjugated bile acid substrates . In contrast , the BtΔ1259 mutant displayed no loss-of-function phenotype ( Figure 2C ) . Complementation of the Bt KO strain with BT2086 restored BSH activity ( Figure 2C ) , confirming that BT2086 is necessary for bile acid deconjugation in Bt . Since bile salt hydrolases and penicillin V amidases ( PVA ) both belong to the cholylglycine hydrolase ( CGH ) family and share a high degree of sequence homology , it is possible that BT1259 is a PVA , although additional experiments would be needed to definitively establish this activity ( Jones et al . , 2008; Panigrahi et al . , 2014 ) . Finally , we verified that when incubated with both TβMCA and TCA , Bt wild-type ( Bt WT ) deconjugated TβMCA but not TCA , whereas the Bt KO strain did not deconjugate either bile acid ( Figure 2D ) . A phylogenetic grouping of the 20 Bacteroidetes strains assayed revealed that while the species that deconjugate bile acids based on the amino acid conjugate ( Group II – gray , Figure 2A ) form a partial clade ( Figure 3A ) , the strains that exhibit selectivity based on the steroid core ( Group I – red ) and those that display no selectivity ( Group III – blue ) are not separated into distinct clades . A BLAST-P search using BT2086 as a query gene identified candidate BSH genes in 19 of the 20 Bacteroidetes strains tested . Bacteroides finegoldii DSM 17565 did not display BSH activity and also lacked a putative BSH . A phylogenetic tree resulting from the multiple sequence alignment of these 19 candidate BSH genes revealed a lack of homology among enzymes within a given activity group ( Figure 3B ) . Group II enzymes , which had formed a clade at the strain level , are now separated into two groups , and steroid core-selective strains ( Group I ) do not cluster significantly . Taken together , these findings suggest that preference for C12 = H over C12 = OH primary bile acid cores is an activity that may have evolved multiple times independently from related members of the BSH superfamily . To test our hypothesis that deleting a single bacterial gene , the bile salt hydrolase BT2086 , would result in a predictable and selective alteration of the in vivo bile acid pools , GF mice were monocolonized with Bt WT or Bt KO ( monocolonization experiment , Figure 4A ) . To further assess effects of this single microbial gene on overall host metabolism and energy utilization , we also performed an experiment in CLAMS ( Comprehensive Lab Animal Monitoring System ) cages using three groups of animals: ( 1 ) mice monoassociated with Bt WT , ( 2 ) mice monoassociated with Bt KO or ( 3 ) GF control mice which remained sterile ( CLAMS experiment , Figure 4A ) . For both studies , over a 4-week period , mice were fed a high-fat , high-sugar diet designed to mimic a Western-style human diet ( 60% kcal% fat ) . For the last week of the CLAMS experiment , mice were transferred from gnotobiotic isolators to pre-sterilized metabolic cages with continuous monitoring in the CLAMS system in order to carefully monitor metabolic status . We first confirmed that BT2086 was expressed in vivo by performing qRT-PCR on cecal contents from Bt WT-colonized mice ( Figure 4—figure supplement 1 ) . As expected , no BT2086 transcripts were detected in the cecal contents of BT KO-colonized mice . We then performed bile acid analyses on tissues and blood from mice in both experiments . As we predicted , Bt KO-colonized mice displayed higher levels of TβMCA in cecal contents than Bt WT-colonized mice in the monocolonization experiment ( Figure 4B ) . Bt KO-colonized mice also exhibited significantly lower levels of βMCA ( p<0 . 0001 ) , the product of TβMCA hydrolysis , than Bt WT-colonized mice . Importantly , the levels of TCA remained unchanged between the two groups , and no CA was detected in either group . These results are consistent with our in vitro data showing that the Bt BSH can deconjugate C12 = H but not C12 = OH primary bile acids . We observed the same significant difference in βMCA levels in feces ( Figure 4C , red highlight boxes ) . In the CLAMS experiment , in agreement with previous reports ( Sayin et al . , 2013 ) , GF mice had significantly higher overall bile acid levels than colonized mice ( p=0 . 0012 Bt WT vs GF , p=0 . 0071 BT KO vs GF ) . Consistent with the monocolonization experiment , cecal contents of Bt KO-colonized mice displayed significantly lower levels of βMCA ( p<0 . 0001 , Bt WT vs GF and BT KO vs GF ) than cecal contents of Bt WT-colonized mice , while CA remained undetectable in both groups ( Figure 4B ) . We also profiled the bile acid composition in the distal ileum , the site of active bile acid reuptake from the small intestine , in the CLAMS experiment . As expected , the bile acid concentrations were approximately fivefold higher in this compartment than in cecal and fecal contents ( Figure 4D ) ( Sayin et al . , 2013 ) . We observed the same trend as we had noted in cecal contents , with higher TβMCA levels in Bt KO-colonized mice , although the differences were not statistically significant ( p=0 . 9343 ) . During sacrifice , we noted that the distribution of this food debris was not uniform along the length of the small intestine . This heterogeneity of contents in the distal ileum may help explain the large range of bile acid measurements observed in this compartment . In contrast to the cecum , feces , and distal ileum , the liver and circulating plasma ( Figure 4—figure supplement 2 ) of Bt WT- and Bt KO-colonized mice contained similar bile acid compositions , with no significant differences noted . These data are consistent with previous observations that the greatest differences between GF and conventionally raised mice were in the cecum and colon , not in the liver or the blood ( Sayin et al . , 2013 ) . We also observed a significant upregulation of bile acid synthesis genes in the liver ( vide infra ) , suggesting that de novo bile acid synthesis may lessen the observed differences between the two groups . Taken together , our data show that we can rationally manipulate the in vivo bile acid pool in the cecum and to a lesser extent in the small intestine and distal colon ( i . e . feces ) using a Bacteroides BSH enzyme that selectively cleaves C12 = H but not C12 = OH conjugated primary bile acids . Importantly , this selective hydrolysis allows us to modulate the levels of TβMCA , a known FXR antagonist , while leaving the levels of TCA , an FXR agonist unchanged . Having shown that Bt BSH status selectively determines composition of the bile acid pool in monocolonized GF mice , we next sought to explore how these specific changes in bile acid levels affected host metabolism . Strikingly , Bt KO-colonized mice gained less weight on the high-fat diet than Bt WT-colonized mice in the monocolonization experiment ( Figure 5A ) . This result is notable because it has been shown that GF mice are more resistant to weight gain when fed a high-fat diet ( Bäckhed et al . , 2007 ) . In addition , we performed a relatively short diet intervention compared to other studies that have used HFD to study metabolic changes ( Jiang et al . , 2015; Joyce et al . , 2014; Rao et al . , 2016; Serino et al . , 2012 ) , and we did not expect to observe significant changes in body weight over the course of a shorter experiment . Importantly , the host effects observed are not due to differences in colonization efficiency . In both experiments , Bt WT and Bt KO efficiently colonized the GI tract and remained the only bacterial species in the mono-associated animals ( Figure 5B and Figure 4—figure supplement 1 ) . These data suggest that the observed metabolic changes are rather due to alterations in the bile acid pool driven by the presence or absence of the Bt BSH . Consistent with the reduced weight gain phenotype , we observed lower levels of triglycerides , cholesterol , and free fatty acids in plasma ( Figure 5C ) as well as lower triglyceride levels in liver ( Figure 5D ) of Bt KO-colonized compared to Bt WT-colonized mice in the monocolonization experiment . Bt KO-colonized mice also exhibited less liver steatosis than Bt WT-colonized mice , consistent with the lower liver triglyceride levels in the former group ( Figure 5E ) . In order to further investigate the effects of Bt BSH status on host metabolism , we transferred Bt KO- or Bt WT-colonized or GF mice to metabolic cages ( CLAMS experiment ) . After a 24 hr acclimation period , we monitored metabolic inputs and outputs for 6 days . We observed significant metabolic differences between the three groups of mice . Both Bt KO-colonized mice and GF mice displayed a lower respiratory exchange ratio ( RER ) than Bt WT-colonized mice ( Figure 6A ) . RER is calculated as the ratio of carbon dioxide produced to oxygen consumed and is used as a measurement of the relative utilization of carbohydrates versus lipids as an energy source ( carbohydrate utilization RER = 1 , lipid RER = 0 . 7 ) . Thus , our data indicate that both the Bt KO-colonized and GF mice are utilizing more lipids for energy than carbohydrates relative to the Bt WT-colonized mice . While Bt KO-colonized mice consumed more oxygen than Bt WT-colonized mice ( Figure 6B ) , there were no significant differences in carbon dioxide production between groups ( full day , Bt WT vs Bt KO p=0 . 4041; Bt WT vs GF p=0 . 3239; Bt KO vs GF p=0 . 0606 ) ( Figure 6C ) . These data are consistent with the lower RER observed in Bt KO-colonized mice . No statistically significant differences in locomotor activity were noted between the three groups ( Figure 6—figure supplement 1 ) . We then used linear regression to investigate the relationship between metabolic rate and body weight in the three groups of mice . Conventionally raised mice as well as humans display a positive linear correlation between energy expenditure and body mass ( Fricker et al . , 1989; Moruppa , 1990 ) . While Bt WT-colonized mice displayed this linear relationship ( p=0 . 0168 , R2 = 0 . 7134 ) , strikingly , both Bt KO-colonized ( p=0 . 6806 , R2 = 0 . 03017 ) and GF ( p=0 . 6930 , R2 = 0 . 02782 ) mice did not ( Figure 6D ) . These data suggest that the deletion of a single bacterial gene , a selective bile salt hydrolase , results in loss of the relationship between metabolic rate and body weight in the host . Taken together , our data from both the monocolonized experiment and the CLAMS experiment suggest that the Bt KO-colonized mice exhibit a metabolic phenotype distinct from Bt WT-colonized mice . The reduced respiratory exchange ratio and weight gain of Bt KO-colonized mice suggest a reduced energy availability profile that is consistent with either reduced food consumption or less efficient caloric extraction from food . In the monocolonization experiment , Bt KO-colonized mice consumed less food during HFD feeding than Bt WT-colonized mice ( −2 . 28 g ± 1 . 36 g vs . +3 . 39 g±1 . 61 g per cage per week , respectively , compared to weekly average for all cages , p=0 . 0165 ) . This result indicates that decreased caloric intake may be a contributing factor in the former group’s decreased weight gain . Since bile acids act as biological detergents that aid in digestion , it is conceivable that the differences in bile acid pools between the groups could alter caloric extraction efficiency . To test this hypothesis , we performed a detergent assay in which we determined the ability of the bile acid pools to solubilize a mixture of fats representative of lipolysis products in the small intestine ( Hofmann , 1963 ) . Bile acid pools for Bt KO- and Bt WT-colonized mice were reconstituted using the mean values for individual compounds measured in the distal ileum and incubated with a 1:1:1 mixture of oleic acid , sodium oleate and 1-oleoyl-rac-glycerol under conditions representative of those in the small intestine ( 150 mM NaCl , pH 6 . 3 , 37˚C ) ( Hofmann , 1963 ) . Sodium dodecyl sulfate ( SDS ) was used as a positive control at its critical micelle concentration ( 8 . 2 mM ) . At both 5 hr and 24 hr time points , we did not detect any differences in solubilization at four different fat concentrations as measured by the turbidity of the resulting mixtures ( Figure 7A ) . Taken together , these data suggest that the metabolic differences observed between the Bt WT- and Bt KO-colonized mice are not due to different detergent abilities of the bile acid pools . In further support of this conclusion , fecal bomb calorimetry did not reveal any differences in energy remaining in fecal pellets from Bt WT-colonized , Bt KO-colonized , or GF mice , indicating that there were no notable differences in caloric energy extraction from food between these groups ( Figure 7B ) . These data suggest that the observed metabolic differences between Bt WT- and Bt KO-colonized mice may be due to differences in bile acids acting as signaling molecules in the host . In order to investigate the gene regulatory mechanisms underlying the metabolic changes observed in Bt KO- compared to Bt WT-colonized mice , we performed RNA-sequencing ( RNA-Seq ) on distal ileum from the monocolonization experiment ( Figure 8—figure supplement 1 ) . We decided to focus our analysis on the distal ileum for three reasons . First , while known bile acid receptors are highly expressed in both liver and intestinal tissue , we observed larger differences in bile acid pool composition in the GI tract ( i . e . small intestine , cecum , feces ) than the liver and blood , suggesting that differences in bile-acid-mediated signaling effects will likely be greater in the small intestine than in the liver . Second , bile acid concentrations are significantly higher in the small intestine than the cecum and colon , the other sites at which we observed differences in bile acid pool composition ( approximately 5-fold and 100-fold higher , respectively ) ( Sayin et al . , 2013 ) . Third , following passage through the small intestine , bile acids are absorbed and recirculated back to the liver primarily in the distal ileum ( Dawson et al . , 2009 ) , making this site the nexus for bile acid sensing and transport in the GI tract . Global transcriptional analysis of the distal ileum identified 12 , 432 genes , of which 428 genes were differentially expressed ( adjusted FDR ≤ 0 . 05 , fold-change ≥±1 . 5 ) between the Bt KO- and Bt WT-colonized mice . Of those genes , the majority ( 314 genes ) were increased in the Bt KO-colonized mice ( Figure 8A ) . Multidimensional scaling analysis ( MDS ) revealed that the two monocolonized groups segregate based on their transcriptional profiles ( Figure 8B ) . Gene Ontology ( GO ) and KEGG pathway analyses of RNA-Seq expression data revealed coordinated changes in gene expression related to metabolism , circadian rhythm , immune response , and histone modifications ( Figure 8C ) . The largest group of differentially expressed genes were those related to host metabolism . We observed significant changes in genes related to carbohydrate and lipid metabolism , amino acid degradation and nitrogen metabolism , and xenobiotic metabolism . In particular , genes involved in the transport ( Slc2a1 ) and breakdown ( Hk1/2 , Pfkl/m ) of glucose were upregulated , whereas G6pc ( glucose-6-phosphatase ) , the final enzyme in the gluconeogenesis pathway , was significantly downregulated ( 8 . 8-fold ) , indicating a shift away from gluconeogenesis and toward glycolysis in the distal ileum of Bt KO-colonized mice . We confirmed the transcriptional change of G6pc in distal ileum using qPCR ( Figure 8D ) . Consistent with these findings , we observed significantly higher blood glucose levels in Bt KO-colonized mice compared to Bt WT-colonized mice in the CLAMS experiment ( p=0 . 0228 ) , indicating an increase in glucose available for glycolysis in the distal ileum ( Figure 9A ) . The expression pattern for genes related to lipid metabolism was more complex , with pathways related to both lipogenesis and lipid breakdown upregulated in Bt KO-colonized animals . Two key genes in the ketogenesis pathway , Bdh1 ( 3-hydroxybutyrate dehydrogenase 1 ) and Hmgcs2 ( 3-hydroxy-3-methylglutaryl-CoA synthase 2 ) , were significantly upregulated ( Figure 8C ) , indicating an increase in the use of lipid and ketogenic amino acid degradation for energy production in the host . Additional genes related to amino acid degradation ( Hao2 , Nos1 , Pcca , Tat ) were also significantly upregulated . Expression of genes involved in the biosynthesis of both glycerophospholipids , in particular phosphatidic acid ( Dgkg , Dgkh , Gpam , Mboat1 , Mboat2 ) , and sphingolipids , in particular cerebrosides and gangliosides ( Glb1 , St3gal5 , St6galnac6 , Ugt8a ) , was also higher in KO-colonized mice . Complex fats synthesized via de novo lipogenesis serve as ligands for PPAR type II nuclear receptors ( Lodhi et al . , 2011 ) . RNA-Seq data revealed that Pparg expression was significantly up-regulated in Bt KO-colonized mice ( Figure 8C ) . Activation of Pparg has been shown to both enhance glucose metabolism and increase lipid uptake ( Martin et al . , 1998 ) , consistent with our broader transcriptional analysis . We confirmed that Pparg expression was significantly upregulated in KO-colonized animals by qPCR ( p=0 . 0207 ) ( Figure 8D ) . Collectively , these data suggest that ileal cells in KO-colonized mice have shifted toward a regime of enhanced glycolysis and increased lipid uptake for the purposes of both the synthesis of complex fats and the breakdown of lipids for energy . Transcriptional analysis also revealed changes in genes regulating circadian rhythm . The observed inverse relationship between expression of the transcriptional activators ( Npas2 and Arntl , decreased in Bt KO-colonized mice ) and circadian repressors ( Per1 , Per2 , Per3 , Cry2 , increased in Bt KO-colonized mice ) is consistent with the transcription-translation negative feedback loop that establishes diurnal rhythms ( King and Takahashi , 2000 ) . The relative changes in circadian rhythm regulation genes were validated using qPCR ( Figure 8E ) . These data indicate that tissues in the distal ileum of Bt KO-colonized mice exist in an altered circadian synchronization state compared to those of Bt WT-colonized mice . Genes involved in immune homeostasis and histone modifications were also differentially expressed . Of particular note , Toll-like receptors ( Tlr1 , Tlr2 , Tlr4 ) , innate immune receptors that play key roles in recognizing microbially produced molecules ( Akira et al . , 2001 ) , were significantly upregulated in our Bt KO-colonized mice . Taken together , these data suggest that bile acid pool alteration elicited a broader scope of changes in the host beyond those directly related to energy production and lipid synthesis . We next sought to investigate the hypothesis that the two bile acid pools would differentially and predictably affect FXR signaling in the small intestine and the liver . Prior work has shown that the gut microbiome mainly affects FXR targets in the ileum but not the liver ( Sayin et al . , 2013 ) . Specifically , activation of ileal FXR leads to production of fibroblast growth factor 15/19 ( FGF15 in mice and FGF19 in humans ) . FGF15 then translocates to the liver where it binds to the FGFR4/β-Klotho complex and represses the expression of Cyp7a1 , which encodes an enzyme catalyzing the rate-limiting step in bile acid synthesis from cholesterol ( Ding et al . , 2015 ) . In this way , activation of FXR in the ileum downregulates bile acid synthesis in the liver . In our system , the levels of the FXR antagonist TβMCA were higher in the cecal contents of Bt KO- versus Bt WT-colonized mice , while the levels of the FXR agonist TCA remain constant between these two groups . Based on these results , we predicted that we would observe inhibition of FXR-dependent pathways in the distal ileum and perhaps the liver in Bt KO-colonized mice . We measured expression of FXR-dependent genes in these tissues using qPCR . Contrary to our expectations , we did not observe a significant difference in genes downstream of FXR , including Nr0b2/Shp ( p=0 . 2018 ) , Fgf15 ( p=0 . 6213 ) , and Fabp6/Ibabp ( p=0 . 6425 ) , in the distal ileum ( Figure 9B ) . We did observe a downregulation of Nr0b2/Shp and upregulation of Cyp7a1 in the liver of Bt KO-colonized mice , results that are consistent with increased TβMCA-mediated FXR antagonism in Bt KO-colonized mice ( Figure 9C ) . The total bile acid pool concentration in cecal contents was higher in Bt KO-colonized mice ( Figure 4A ) , consistent with an increase in Cyp7a1 transcription resulting in an increase in bile acid synthesis . We also observed decreases in the expression of other genes in the liver that are regulated by FXR , including Apoc2 , which encodes a protein that is secreted into plasma and activates lipoprotein lipase , as well as increases in genes that are negatively regulated by the FXR target gene Nr0b2/Shp , including sterol regulatory element-binding protein 2 ( Srebf2/Srebp2 ) and glucose-6-phosphatase ( G6pc ) ( Figure 9C ) . While the former gene regulates cholesterol biosynthesis in the liver , the later gene catalyzes the final step in gluconeogenesis . The increase in G6pc in the liver of Bt KO-colonized mice is notable because this gene is significantly downregulated in the distal ilea of these mice ( Figure 8D ) . Taken together , our data are consistent with a scenario in which bile-acid-mediated FXR antagonism is affecting pathways in the liver but not the ileum of Bt KO-colonized mice . While some patterns of gene expression in the liver may be explained by FXR signaling , changes in the expression levels of certain notable pathways are not consistent with FXR-controlled regulation . We would expect to see an increase in the expression of the gene encoding sterol regulatory element binding protein 1 c ( Srebf1/Srebp1c ) as well as the downstream genes Fas and Acc , which are involved in de novo fatty acid synthesis , in the liver of Bt KO-colonized animals . No significant differences in expression of these genes , however , were observed between Bt KO- and WT-colonized mice ( Srebf1/Srebp1c , p=0 . 5018; Fas , p=0 . 3292; Acaca , p=0 . 1302 ) ( Figure 9C ) . In addition , we observed significant decreases in genes not known to be under the control of FXR , including Cd36 ( p=0 . 0015 ) , a gene encoding a fatty acid transporter , the immune-related genes tumor necrosis factor alpha ( Tnf/Tnfα , p=0 . 0225 ) and EGF-like module-containing mucin-like hormone receptor-like 1 ( Adgre1/Emr1 , p=0 . 0011 ) , and the G-protein-coupled receptor S1pr2 target gene sphingosine kinase 2 ( Sphk2 , p=0 . 0274 ) ( Nagahashi et al . , 2015 ) , in the liver of Bt KO-colonized mice ( Figure 9C ) . These results indicate that other host receptors may be involved in the transcriptional changes and metabolic differences observed . Taken together , our data suggest that changing the in vivo bile acid pool using selective expression of a bacterial bile salt hydrolase results in significant alterations in host gene expression , and that these changes are due not to the detergent properties of bile acids but rather to their activities as signaling molecules .
In this work , we identified a group of gut strains from the bacterial phylum Bacteroidetes that exhibit selective bile salt hydrolase activity . These bacteria selectively hydrolyze conjugated bile acid substrates based on the hydroxylation pattern of the steroidal core as opposed to the amino acid conjugate . Since the majority of BSH characterized to date from Bacteroidetes and Firmicutes are promiscuous and do not display selective deconjugation activity based on the bile acid substrate , it is possible that selective BSH activity may be an evolved trait . The lack of distinct clustering of Group I ( i . e . steroidal core-selective ) BSH at both the strain and protein levels suggests this activity that may have arisen multiple times in evolutionary history from different bacterial hydrolase precursors . Structural comparisons of closely related BSH with different selectivity profiles may reveal individual amino acids that could be responsible for the activities observed . It is also possible that differential trafficking of either the bile acid substrate or product or of the BSH protein itself ( Begley et al . , 2006 ) in these Bacteroidetes strains may be responsible for some of the differences in reactivity . Additional microbiological , biochemical , and structural studies will be needed to answer these questions . After identifying the gene responsible for BSH activity in B . thetaiotaomicron and generating a mutant ( Bt KO ) , we leveraged these isogenic strains in order to manipulate the in vivo bile acid pool in a highly specific manner in monocolonized GF mice . Bt KO-colonized mice , which contained significantly higher cecal TβMCA levels than Bt wild type ( WT ) -colonized mice , gained less weight on a HFD , had lower liver and plasma lipid levels , and displayed a respiratory exchange ratio that was shifted toward lipid utilization . These changes in host metabolism are particularly striking in light of the fact that the only difference between these two groups of mice was the presence or absence of a single bacterial gene . Remarkably , the presence of this BSH gene in BT WT-colonized mice was able to recover the positive linear correlation between energy expenditure and lean body mass normally observed in both conventional mice and humans ( Fricker et al . , 1989; Moruppa , 1990 ) . This result suggests that specific genes in the gut microbiome may contribute to the establishment of host phenotypes not previously considered to be affected by the resident microbiota . At a transcriptional level , genes related to metabolic pathways , circadian rhythm , immune modulation , and histone modifications were significantly altered in Bt KO- compared to Bt WT-colonized mice . Since TβMCA is a known FXR antagonist , we expected to observe changes in host gene expression that were consistent with downregulation of FXR-mediated pathways in Bt KO-colonized mice . The decreased expression of FXR target genes Nr0b2/Shp and Apoc2 as well as the increased expression of Cyp7a1 , the rate-limiting enzyme in bile acid biosynthesis , are consistent with a regime of FXR antagonism in the livers of Bt KO- compared to Bt WT-colonized mice . These transcriptional changes suggest that the observed increase in total bile acids in the cecal contents of Bt KO-colonized mice is due to FXR-dependent bile acid biosynthesis in the liver . Other phenotypic and transcriptional differences observed between Bt KO- and Bt WT-colonized mice are not readily explained by FXR antagonism , however . Conventionally colonized FXR knockout ( Nr1h4-/- ) mice display less weight gain on a high-fat diet than wild-type mice ( Prawitt et al . , 2011 ) and also have decreased liver expression of Nr0b2/Shp and increased expression of Cyp7a1 ( Sayin et al . , 2013 ) , consistent with our results in Bt KO-colonized mice . Nr1h4-/- mice , however , exhibit increased triglyceride and cholesterol levels in plasma ( Cariou et al . , 2006; Lambert et al . , 2003; Sinal et al . , 2000 ) and low blood glucose and delayed intestinal glucose absorption when fasted ( Cariou et al . , 2006; van Dijk et al . , 2009 ) . Bt KO-colonized mice displayed the opposite phenotypes , including increased glucose and decreased triglyceride and cholesterol levels in plasma and a shift toward increased expression of glucose uptake and utilization genes in the distal ileum when fasted . While the expression of the FXR target genes Nr0b2/Shp , Fgf15 , and Fabp6/Ibabp in the ileum are decreased in Nr1h4-/- mice ( Sayin et al . , 2013 ) , we observed no transcriptional differences in these genes in Bt KO- and Bt WT-colonized mouse ilea . In addition , while the expression of hepatic gluconeogenesis genes is decreased in FXR-deficient mice ( Cariou et al . , 2005; Duran-Sandoval et al . , 2005; Ma et al . , 2006 ) , we observed an increase in the expression of glucose-6-phosphatase ( G6pc ) in the liver of Bt KO-colonized mice . Taken together , these comparisons may indicate that many of the phenotypic and transcriptional differences noted in BT KO-colonized mice are either FXR-independent or not directly dependent on FXR-mediated signaling . These results raise the possibility , then , that bile acid signaling through other host receptors may be in part responsible for the observed differences in host metabolism . Returning to the RNA-Seq data , we noted that there were significant differences in the expression of ileal genes involved in xenobiotic metabolism in Bt KO-colonized compared to BT WT-colonized mice . The pregnane X receptor ( PXR ) has been shown to play a central role in the response to xenobiotics , and in particular , in the transcriptional regulation of cytochrome P450 3A ( Cyp3A ) genes ( Bertilsson et al . , 1998 ) . The expression of Cyp3a11 , a mouse gene known to be regulated by PXR ( Kliewer et al . , 1998 ) , was significantly decreased ( 3 . 7-fold ) in Bt KO-colonized animals . This result indicates that PXR-dependent pathways may be suppressed in these mice compared to Bt WT-colonized mice . PXR also plays an important role in glucose and lipid homeostasis and energy metabolism ( Gao and Xie , 2010; Kodama et al . , 2004; Kodama et al . , 2007; Nakamura et al . , 2007 ) . PXR knockout ( Nr1i2-/- ) mice gain less weight on a high-fat diet than wild-type mice and also display decreased liver steatosis and hepatic triglyeride levels ( Spruiell et al . , 2014 ) . Importantly , in contrast to FXR knockout ( Nr1h4-/- ) mice , Nr1i2-/- mice fed a high-fat diet exhibit increased fasting blood glucose levels and unchanged fasting insulin levels compared to wild-type mice ( He et al . , 2013; Spruiell et al . , 2014 ) . These metabolic phenotypes are consistent with those observed in Bt KO-colonized mice . Finally , PXR has been shown to be necessary and sufficient for the activation of the fatty acid transport gene Cd36 in the liver ( Zhou et al . , 2006 ) , and we observed a decrease in hepatic Cd36 expression in BT KO-colonized mice . Taken together , our data are consistent with a regime of reduced PXR activation in Bt KO-colonized mice and perhaps suggest that PXR signaling may be involved in some of the metabolic phenotypes observed . We cannot rule out the possibility that host receptors beyond FXR and PXR may be involved in the differences noted between Bt KO- and Bt WT-colonized mice . Exploration of bile acids as modulators of host metabolic , circadian rhythm , and immune response via binding to nuclear receptors and GPCRs is an experimental trajectory that warrants further investigation . Moreover , although our results support the conclusion that the observed changes in host metabolism are mediated by signaling properties of bile acids and not by their detergent activities , it is not yet clear which gene-level changes are driving the organism-level metabolic effects . Although our biochemical , transcriptional , and CLAMS data are consistent with a regime of decreased food intake in Bt KO- compared to Bt WT-colonized mice , we did not observe significant differences in plasma levels of leptin ( p=0 . 4648 , Bt WT vs Bt KO ) and ghrelin ( p=0 . 7783 , Bt WT vs Bt KO ) , hormones regulating satiety and hunger ( Figure 8C ) . Additional studies are needed to explore the contributions of bile acid signaling to host energy expenditure and feeding behavior . Finally , because mice and humans possess different primary bile acids , there is the question of whether the observed changes in both bile acid composition and host metabolism are relevant for humans . While the major primary bile acids in mice are TβMCA and TCA ( Sayin et al . , 2013 ) , humans produce glyco- and tauro-conjugated CDCA and CA ( Russell , 2003 ) . Our in vitro results show that selective Bacteroides strains cleave conjugated C12 = H ( e . g . βMCA , CDCA ) but not C12 = OH ( e . g . CA ) primary bile acids . Based on our in vivo results , one would predict that these strains would cleave conjugated CDCA while leaving conjugated CA untouched in the human gut . Furthermore , analysis of data from the first and second phases of the Human Microbiome Project has revealed that the composition of the human gut community , specifically species of Bacteroidetes , is highly personalized . While Firmicutes were more temporally variable within individuals , Bacteroidetes species , and in particular the Bacteroides genus , displayed mainly inter-individual variation ( Kraal et al . , 2014; Lloyd-Price et al . , 2017 ) . Our results suggest that Bacteroides species status in individuals may in part determine downstream bile acid pool composition in these people . Finally , the FXR pathway as well as other host receptor pathways that may act as bile acid targets are highly conserved in mammals ( Reschly et al . , 2008 ) , suggesting that discoveries about fundamental host signaling in mice are also likely to be operable in humans . Future studies in mice with humanized bile acid pools may reveal how selective Bacteroides BSH activity is likely to affect metabolism in the human host .
Conjugated and unconjugated bile acids were purchased from Steraloids Inc . ( Newport , RI ) . Oleic acid , sodium oleate and 1-oleoyl-rac-glycerol were purchased from Sigma Aldrich . All Bacteroidetes strains were cultured at 37°C in brain heart infusion agar ( Bacto ) supplemented with with 5 mg/L hemin , 2 . 5 uL/L Vitamin K , and 500 mg/L cysteine HCl ( BHI+ ) . All strains were cultured under anaerobic conditions using an anaerobic chamber ( Coy Laboratory Products ) with a gas mix of 5% hydrogen and 20% carbon dioxide ( balance nitrogen ) . Escherichia coli strains were grown aerobically at 37˚C in LB medium supplemented with ampicillin to select for the pExchange-tdk plasmid . A BLASTP search was performed on Integrated Microbial Genomes , the US Department of Energy’s Joint Genome Institute ( IMG JGI ) using the bile salt hydrolase from C . perfringens ( NCBI Protein accession code WP_003461725 ) as the query sequence , with a cutoff expectation value of 1 × 10−5 . Plasmids and primers are listed in the Key Resources Table . All mutants were created in the B . thetaiotaomicron VPI-5482 Δtdk background . The BtΔ2086 and BtΔ1259 mutants were constructed using a previously described counterselectable allelic exchange method ( Koropatkin et al . , 2008 ) . Briefly , ~1 kb fragments upstream and downstream of the BT2086 and BT1259 genes were cloned and fused using primer pairs ( BT2086KO UF/UR and DF/DR; BT1259KO UF/UR and DF/DR ) and ligated into the suicide vector pExchange-tdk . The resulting vectors were electroporated into Escherichia coli S17-1 λ pir and then conjugated into B . thetaiotaomicron . Single-crossover integrants were selected on BHI-blood agar plates containing 200 μg/ml gentamicin and 25 μg/mL erythromycin , cultured in TYG medium overnight , and then plated onto BHI-blood agar plates containing 200 μg/ml 5-fluoro-2-deoxyuridine ( FUdR ) . Candidate BT2086 and BT1259 deletions were screened by PCR using the diagnostic primers listed in the Key Resources Table and confirmed by DNA sequencing to identify isolates that had lost the gene . The B . thetaiotaomicron complementation strains were constructed using a previously described method with slight modifications . Assembled construct designs were based on the mobilizable Bacteroides element NBU2 ( Wang et al . , 2000 ) . Briefly , BT2086 was PCR-amplified , cloned as an NdeI/XbaI fragment into the constitutive expression vector pNBU2_erm_us1311 , which contains the 300 bp region upstream of BT1311 ( σ70 ) , and transformed into E . coli S17-1 λ pir chemically competent cells ( Cullen et al . , 2015; Degnan et al . , 2014 ) . E . coli S17 lambda pir containing the desired plasmid or the pNBU2_erm_us1311 control vector were cultured aerobically in 5 mL of LB media at 37°C , and the Bacteroides recipient strain ( BtΔ2086 ) was cultured anaerobically in 5 mL BHI+ media at 37°C . The E . coli S17 donor strains and B . theta recipient strain were then subcultured in 5 mL of fresh media . At mid to late log growth , the transformed S17-1 cells were spun down , resuspended with Bacteroides strain ( BtΔ2086 ) culture in 1 mL BHI+ media , spreaded on to a BHI+ 10% horse blood agar plate , and incubated aerobically at 37°C agar side down . After 16–24 hr , bacterial biomass from the conjugation plates was scraped and resuspended in 5 mL of BHI+ media and spread on to a BHI-blood agar plate containing 200 μg/mL gentamycin and 25 μg/mL erythromycin . Colonies were confirmed via PCR and sequencing using the diagnostic primers listed in the Key Resources Table . Recovery of function of the complementation strain was confirmed via UPLC-MS with 100 μM TUDCA as substrate . BLASTP searches were performed on Integrated Microbial Genomes ( IMG JGI ) using BT2086 as the query sequence , with a cutoff expectation value of 1 × 10−5 . Putative Bacteroidetes BSHs were identified from 19 of the 20 Bacteroidetes strains tested . A multiple sequence alignment was calculated using MUSCLE ( Edgar , 2004 ) . A phylogenetic tree was then computed from this alignment using PhyML ( Guindon et al . , 2010 ) , choosing the LG substitution model , the SPR and NII ( best ) tree improvement method , 10 random starting trees , and bootstrap with 1000 replicas . The phylogenetic tree was visualized using iTOL ( Letunic and Bork , 2011 ) . A phylogenetic analysis of 20 Bacteroidetes strains was performed using PhyloPhlAn ( Segata et al . , 2013 ) . All 20 fully or partial sequenced microbial genomes were retrieved from IMG and the National Center for Biotechnology Information ( NCBI ) . All strains were cultured in 4 mL of BHI+ medium overnight . The following day they were diluted to pre-log phase ( OD600 = 0 . 1 ) in fresh BHI+ to a final volume of 4 mL . Stock solution of taurine conjugated bile acids ( TCA , TDCA , TCDCA , TUDCA , TLCA and TβMCA ) or glycine conjugated bile acids ( GCA , GDCA , GCDCA , GUDCA and GLCA ) were added to each culture to obtain a final concentration of 50 µM of each bile acid . Cultures were then incubated in the anaerobic chamber at 37°C for 48 hr . At the 24 hr and 48 hr time points , 2 mL of each culture was extracted via the method described under ‘Bile Acid Analysis - Sample Preparation for Bacterial Culture’ . Germ-free C57BL/6 mice were maintained in gnotobiotic isolators at the Massachusetts Host-Microbiome Center under a strict 12 hr light cycle and a constant temperature ( 21 ± 1°C ) and humidity ( 55–65% ) . All experiments were conducted on 8–9 week old male mice . GF status prior to the gavage was confirmed on a bi-weekly basis microbiologically through culturing mouse stool on broad-spectrum plates in both aerobic and anaerobic conditions , as well as Gram staining homogenized mouse stool in 1xPBS . TSA Blood Agar plates were used for aerobic conditions , while Brucella Blood Agar plates were used for anaerobic conditions . Monocolonization or GF status following gavage was confirmed by plating of fecal pellets ( described below ) and by 16S rRNA gene PCR and 16S sequencing . All experiments involving mice were performed using IACUC approved protocols under Brigham and Women’s Hospital Center for Comparative Medicine . For the monocolonization experiment , 12 weight-matched mice per group were colonized with either B . thetaiotaomicron wild type ( Bt WT ) or the Δ2086 mutant ( Bt KO ) by oral gavage of overnight cultures as previously described ( Marcobal et al . , 2011 ) . Frozen feces ( day 2 post-colonization ) were plated to determine CFU/g . The mice were co-housed in their respective groups in gnotobiotic isolators for the entire duration of the experiment . The mice were fed a standard diet containing 24% of calories from fat , 23% from protein , and 53% from carbohydrates ( Autoclavable Mouse Breeder Diet 5021; LabDiet ) for the first 4 days after gavage . After a 4-day acclimation period post-gavage , the mice were switched to a high-fat diet ( Research Diets D12492 ) with 60 kcal% of fat sterilized by 10–20 kGy of gamma-irradiation . Mouse feces were collected 2 days after colonization to check bacterial colonization efficiency . This was achieved by homogenizing 1–2 fecal pellets in 1 mL PBS and then plating out 1:10 serial dilutions of the homogenate on BHI+ agar plates in the anaerobic chamber . BSH enzyme activities in different experimental groups were also checked via UPLC-MS for validation purposes . Fecal samples were collected on days 2 , 4 , 11 , 18 , 25 and 32 post-colonization and frozen at −80˚C prior to analysis . Mice were fasted for 4 hr prior to sacrifice , at which point tissues and blood were collected . For the CLAMS experiment , GF mice were colonized as above with a third group of age- and weight-matched GF mice used as an additional control group ( eight mice per group ) . Fresh feces ( day 2 post-colonization ) were plated to determine CFU/g . On day 24 and day 25 post-colonization , mice were transported in pre-sterilized CLAMS cages to Brigham and Women’s Hospital ( BWH ) Metabolic Core facility to conduct metabolic studies . Animals were housed individually in metabolic chambers maintained at 22˚C under a 12 hr light/dark cycle with a constant access to food and water and maintained on a high-fat diet ( Research Diets D12492 ) . One mouse from the Bt WT-colonized group was excluded from the study because this animal refused food , lost 35% of its initial body weight in the CLAMS , and displayed GI tract abnormalities during sacrifice . Whole body metabolic rate was measured using the Oxymax open-circuit indirect calorimeter , Comprehensive Lab Animal Monitoring System ( CLAMS , Columbus Instruments ) . Body composition was examined with Echo MRI ( Echo Medical Systems , Houston , Texas ) using the 3-in-1 Echo MRI Composition Analyzer ( Kazak et al . , 2017; Mina et al . , 2017 ) , and respiratory exchange ratio ( RER ) , calorific value ( CV ) , and energy expenditure ( EE ) are calculated by the equations below:RER=VCO2VO2CV=3 . 815+1 . 232∗REREE=CV∗VO2 During sacrifice , whole blood was collected into commercially available EDTA-coated tubes ( Milian Dutscher group ) . Cells were removed from plasma by centrifugation for 15 min at 2 , 000 g at 4°C . Plasma was transferred to a new eppendorf tube from the supernatant and stored in −80°C for further investigation . Cecal contents were collected at sacrifice ( day 32 ) using sterile tools on a sterile field and plated to confirm maintenance of monocolonized or GF status throughout the experiment . No CFU were detected in the GF group . RNA extracted from these cecal contents was used for qRT-PCR using 16 s rRNA and Bt BSH primers ( Key Resources Table , Figure 4—figure supplement 1 ) . Plasma insulin , glucose and ghrelin levels were analyzed by the Vanderbilt Mouse Metabolic Phenotyping Center . Plasma glucose was measured by a glucose oxidase method using an Analox Instruments GM9 glucose analyzer ( Stourbridge , UK ) . Plasma insulin was measured by radioimmunoassay ( Millipore ) . Total ghrelin was measured by radioimmunoassay ( Millipore ) . Plasma leptin , glucagon levels were analyzed by ELISA kits ( Crystal Chem ) . Total plasma cholesterol , triglyceride and free fatty acids ( FFA ) were measured by standard enzymatic assays , and liver tissues were extracted ( Folch et al . , 1957 ) and analyzed by the Vanderbilt University Metabolic Phenotyping Center ( VUMC ) using GC . Histology of the liver samples for steatosis was performed using a reported method ( Rao et al . , 2016 ) in the Harvard Rodent Histopathology Core . Briefly , a portion of liver sample was cut and formalin-fixed , trimmed , cassetted and embedded in paraffin and stained with hematoxylin and eosin . Liver histology was assessed for steatosis on blinded sections . To determine the remaining caloric content in the mouse feces , bomb calorimetry was carried out using the Parr Oxygen Bomb equipped with a Parr 6725 Semimicro Calorimeter module and aParr 6772 Calorimetric Thermometer module at the Brigham and Women’s Hospital ( BWH ) Metabolic Core facility . Briefly , 30–100 mg of pooled fecal samples from the sample mice were dehydrated at 60˚C for 48 hr in a micro centrifuge tube . Calculated heats ( cal/g ) take into account diurnal variations in fecal output as well as any contaminants that had entered into the sample . Unless otherwise indicated in the figure legends , differences between experimental groups or conditions were evaluated using unpaired two-tailed Welch’s t test for pairwise comparison , one-way ANOVA for multiple comparisons . Significance was determined as p value < 0 . 05 . Statistical analysis and plotting for metabolic studies was performed in the R programming language with CalR , a custom package for analysis of indirect calorimetry using analysis of covariance with a graphical user interface ( Mina et al . , 2017 ) . | The microbiome , the collection of bacteria that live in and on human bodies , has a strong influence on how well the body works . However , the diversity of the microbiome makes it difficult to untangle exactly how it has these effects . For example , it is poorly understood how the hundreds of species of bacteria that live in the gut affect metabolism – the chemical processes that make life possible . But they are known to influence how metabolic diseases like diabetes and obesity develop . When we eat a meal , the body releases compounds called bile acids to help to digest the food . Once the bile acids reach the colon , the bacteria residing there use enzymes to chemically modify the compounds . Imbalances in the resulting pool of over 50 different bile acids may accelerate how quickly people develop metabolic disorders . It is not clear , however , which bile acids have helpful or harmful effects on metabolism . Yao et al . first identified a selective version of a prevalent gut bacterial enzyme called a bile salt hydrolase . This enzyme was then deleted from a common gut bacterium using genetic tools . Finally , Yao et al . colonized mice lacking any bacteria ( i . e . , germ-free mice ) with either the original bacterium or the hydrolase-deleted bacterium . Mice colonized with the hydrolase-deleted bacteria gained less weight on a high fat diet and had lower levels of fat in their blood and liver . These mice also shifted to burning fats instead of carbohydrates for energy . The changes in the bile acid pool produced in mice colonized with hydrolase-deleted bacteria did not only affect metabolism . Yao et al . found differences in the activity of genes important for other biological processes as well , such as those that control circadian rhythms and immune responses . Further research is needed to investigate whether limiting the activity of the bile salt hydrolase enzyme has similar effects in humans . If so , developing drugs or probiotics that target the enzyme could lead to new treatments for people with metabolic diseases like obesity and fatty liver disease . Investigating the biological effects of other bacterially modified bile acids may identify other possible treatments as well . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2018 | A selective gut bacterial bile salt hydrolase alters host metabolism |
Recently , a prototype dengue early warning system was developed to produce probabilistic forecasts of dengue risk three months ahead of the 2014 World Cup in Brazil . Here , we evaluate the categorical dengue forecasts across all microregions in Brazil , using dengue cases reported in June 2014 to validate the model . We also compare the forecast model framework to a null model , based on seasonal averages of previously observed dengue incidence . When considering the ability of the two models to predict high dengue risk across Brazil , the forecast model produced more hits and fewer missed events than the null model , with a hit rate of 57% for the forecast model compared to 33% for the null model . This early warning model framework may be useful to public health services , not only ahead of mass gatherings , but also before the peak dengue season each year , to control potentially explosive dengue epidemics .
Dengue is an arboviral infection of major international public health concern ( Guzman and Harris , 2015 ) . Dengue is endemic in more than 100 countries in the tropics and sub-tropics , with Brazil reporting more cases than any other country in the world ( Bhatt et al . , 2013; Teixeira et al . , 2009 ) . Dengue is caused by four distinct dengue virus serotypes ( DENV 1–4 ) , which are transmitted to humans by Aedes mosquitoes ( Guzman and Harris , 2015 ) . The distribution of both Ae . aegypti and Ae . albopictus is widespread across Brazil , with Ae . aegypti found predominantly in urban settings , breeding in artificial containers , and Ae . albopictus more commonly found in rural and peri-urban settings ( Kraemer et al . , 2015 ) . However , dengue incidence is unequally distributed in Brazil , with higher and sustainable incidence along the Atlantic coast and in the central region . Temperature and rainfall regimes seem to control the magnitude and seasonality of dengue transmission ( Campbell et al . , 2015 ) . Large outbreaks are typically observed after rainy and warm periods , at the end of summer , particularly in densely populated urban areas ( Teixeira et al . , 2009 ) . The presence and abundance of dengue mosquitoes is a necessary but not sufficient condition for dengue transmission and the occurrence of large outbreaks . Besides vector infestation , an important factor regulating transmission is the introduction of new serotypes of virus in areas with a high susceptible population . This may be facilitated by the increasing international and internal mobility across the country . Thus , large and touristic cities are prone to the introduction and maintenance of virus circulation . International mass gathering events have become an important health issue in the recent years ( Abubakar et al . , 2012 ) as they create the opportunity for the introduction of new pathogens in a susceptible population , as well as exposing visitors to new and unknown local risks ( Matos and Barcellos , 2010 ) . Early warning systems , which take into account multiple dengue risk factors , can assist public health authorities to implement timely control measures ahead of imminent dengue outbreaks . Seasonal climate forecasts combined with early dengue surveillance system data provide an opportunity to anticipate dengue outbreaks several months in advance ( Lowe et al . , 2014 ) . To date , several studies have assessed the use of climate information in early warning systems for diseases , such as malaria and Rift Valley fever ( Anyamba et al . , 2009; Thomson et al . , 2006 ) . The incorporation of climate information for dengue early warning systems has also been explored ( Degallier et al . , 2010; Lowe et al . , 2011; Stewart-Ibarra and Lowe , 2013 ) . However , to our knowledge , real-time climate forecasts have not been previously applied to predict dengue epidemics in a practical real-life framework . From 12 June to 13 July 2014 , Brazil hosted the 2014 Fédération Internationale de Football Association ( FIFA ) World Cup , a mass gathering of more than 3 million Brazilian and international spectators , travelling between 12 different host cities . Before the event , the potential risk of transmission of several communicable diseases , including dengue fever , was highlighted ( Gallego et al . , 2014; Wilson and Chen , 2014 ) . Several research groups published dengue outlooks ahead of the World Cup . Approaches included analysing historical time series distributions of city or state level data ( Hay , 2013 ) and mapping of historical averages , while accounting for seasonality and areas of permanent transmission ( Barcellos and Lowe , 2014a ) . Some groups formulated deterministic ( Massad et al . , 2014 ) and statistical ( van Panhuis et al . , 2014 ) models to estimate the number of tourists expected to contract dengue fever . Another study ( Lowe et al . , 2014 ) assessed the potential for dengue epidemics during the tournament by providing probabilistic forecasts of dengue risk for the 553 microregions of Brazil with risk-level warnings issued for the 12 cities where the matches were played . The dengue early warning system , formulated using a Bayesian spatio-temporal model framework ( Lowe et al . , 2011 , 2013 ) , was driven by real-time seasonal climate forecasts for the period March-April-May and the dengue cases reported to the Brazilian Ministry of Health in February 2014 . This information was combined to produce a dengue forecast at the start of March 2014 . Predicted probability distributions of dengue incidence rates ( DIR ) were summarised and translated into risk warnings , using a two-tier threshold approach . First , the probability of DIR falling into categories of low , medium and high risk was determined using dengue risk thresholds of 100 and 300 cases per 100 , 000 inhabitants , defined by the National Dengue Control Programme of the Brazilian Ministry of Health ( Ministério da Saúde , 2008 ) . Second , probability trigger thresholds were calculated by selecting optimal cut-off values that maximised sensitivity and specificity , for each dengue risk threshold ( medium and high ) . Using criteria related to the probability trigger thresholds , forecast warnings of low , medium or high dengue risk were determined for the 12 microregions hosting World Cup matches ( see Materials and methods for further details ) . The forecasts were produced and made available three months ahead of the event ( see Lowe et al . , 2014 ) . In this article , we provide an evaluation of the forecast model predictions by using the observed dengue incidence rates for June 2014 to assess the ability of the model framework to successfully assign dengue risk warning categories for the host microregions and all microregions in Brazil . We also compare the forecast model framework to a null model , based on seasonal averages of previously observed dengue incidence . We then discuss the challenges and limitations of producing disease risk forecasts in a real-time setting , such as the use of incomplete surveillance data to drive the model , the coarse spatial resolution of the forecasts , the definition of risk and alarm trigger thresholds , the lack of information regarding the ( re ) introduction of different serotypes or vector control activities , and the difficulties in communicating probabilistic forecasts . Finally , we suggest future model developments and advocate a multi-model approach to dengue prediction in the future .
Table 1 shows the dengue forecasts for June 2014 for the microregions where stadiums were located , issued three months before the World Cup , and published before the event ( Lowe et al . , 2014 ) . For comparison , the observed DIR values are included in the table , along with the observed risk categories , determined using the medium and high dengue risk thresholds . Note that the dengue risk thresholds used by the National Dengue Control Programme are based on yearly dengue incidence rates ( Ministério da Saúde , 2008 ) . Therefore , we converted the monthly incidence to yearly equivalent incidence to make use of the medium and high dengue risk thresholds at a monthly time scale ( see Materials and methods ) . Figure 1 shows the corresponding predictive distributions , the posterior predicted mean DIR and upper 95% prediction interval ( dashed and dotted lines ) and observed DIR ( marked with an arrow ) . The comparison of the second and last columns of Table 1 reveals that the model correctly predicted dengue risk categories ( highlighted in blue ) for Fortaleza and Natal ( high ) , Belo Horizonte , Manaus and Salvador ( medium ) and Curitiba and Porto Alegre ( low ) . In Recife , the predicted category was high , but the observed category was medium . However , for Recife , the mean predicted DIR was almost equal to the observed DIR and the point estimate fell within the medium category ( see Table 1 , Figure 1 ) . The definition of the alarm trigger threshold placed this microregion in the high category . This result highlights the difficulties of translating probabilistic information into simpler warnings , based on predefined probability trigger thresholds . The model 'missed' the unprecedented high incidence that was observed in Brasília and São Paulo in June 2014 . However , the model predicted a 7% forecast probability of observing high risk in Brasília . 10 . 7554/eLife . 11285 . 003Table 1 . Dengue risk forecast warnings and corresponding observations for June , 2014 for host microregions . Dengue risk forecast warnings and observed category for June 2014 , for the microregions hosting the World Cup tournament . Low risk was defined as fewer than 100 cases per 100 , 000 inhabitants , medium risk as between 100 and 300 cases per 100 , 000 inhabitants , and high risk as greater than 300 cases per 100 , 000 inhabitants . If the probability of low risk was less than 68% , a medium risk forecast warning was issued . If the probability of high risk was concurrently greater than 18% , the forecast warning was upgraded to high risk . The observed DIR value is included . Microregions where the observed DIR fell into the same category as forecast ( e . g . the forecast warning category was high and the observed DIR category was high ) , are shaded . DOI: http://dx . doi . org/10 . 7554/eLife . 11285 . 003MicroregionForecast warningProbability ( pL , pM , pH ) Observed DIRObserved categoryBelo HorizonteMediump ( 65% , 24% , 11% ) 126MediumBrasíliaLowp ( 73% , 20% , 7% ) 725HighCuiabáLowp ( 71% , 22% , 7% ) 168MediumCuritibaLowp ( 100% , 0% , 0% ) 4LowFortalezaHighp ( 34% , 20% , 46% ) 507HighManausMediump ( 63% , 25% , 12% ) 110MediumNatalHighp ( 32% , 20% , 48% ) 780HighPorto AlegreLowp ( 100% , 0% , 0% ) 1LowRecifeHighp ( 57% , 24% , 19% ) 161MediumSalvadorMediump ( 56% , 27% , 17% ) 149MediumSão PauloLowp ( 99% , 1% , 0% ) 161MediumRio de JaneiroMediump ( 62% , 25% , 13% ) 32Low10 . 7554/eLife . 11285 . 004Figure 1 . Predictive distributions and observed DIR for June 2014 for host microregions . Posterior predictive distributions of dengue incidence rates ( DIR ) ( base-10 logarithmic scale ) for June 2014 showing the probability of low risk ( blue ) , medium risk ( orange ) and high risk ( pink ) for June 2014 , in the microregions hosting the World Cup tournament: ( a ) Belo Horizonte , ( b ) Brasília , ( c ) Cuiabá , ( d ) Curitiba , ( e ) Fortaleza , ( f ) Manaus , ( g ) Natal , ( h ) Porto Alegre , ( i ) Recife , ( j ) Salvador , ( k ) São Paulo and ( l ) Rio de Janeiro . Observed DIR indicated by black arrow . Posterior predictive mean and upper 95% prediction ( credible ) interval of the distribution indicated by a dashed and dotted line , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 11285 . 004 Probabilistic forecasts were generated not only for the twelve host microregions , but for all 553 microregions of Brazil . This gives an idea of how the model framework might contribute towards a nationwide dengue early warning system in the future . Figure 2 shows a ternary probabilistic forecast map ( Jupp et al . , 2012; Lowe et al . , 2014 ) and the corresponding observed dengue incidence rate categories ( low , medium and high ) . The model correctly predicted , with high certainty ( the greater the colour saturation , the greater the certainty ) , low dengue risk in South Brazil and large areas of the Amazon . Areas with a higher chance of observing high risk were correctly detected for areas in North East Brazil . Actual dengue incidence rates were higher than expected in Brasília , although the likelihood of observing higher dengue incidence for the surrounding region was relatively greater than observing lower incidence . For some microregions in the state of São Paulo , the model was uncertain of the most likely category ( indicated by pale colours ) . Some of these areas experienced high dengue incidence rates in June 2014 . 10 . 7554/eLife . 11285 . 005Figure 2 . Probabilistic dengue forecast and observed dengue incidence rate categories for Brazil , June 2014 . ( a ) Probabilistic dengue forecast for June 2014 . The continuous colour palette ( ternary phase diagram ) conveys the probabilities assigned to low-risk , medium-risk , and high-risk dengue categories . Category boundaries defined as 100 cases per 100 , 000 inhabitants and 300 cases per 100 , 000 inhabitants . The greater the colour saturation , the more certain is the forecast of a particular outcome . Strong red shows a higher probability of high dengue risk . Strong blue indicates a higher probability of low dengue risk . Colours close to white indicate a forecast similar to the benchmark ( long-term average distribution of dengue incidence in Brazil , June , 2000–2013: pL=68% , pM=16% , pH=16% ) , marked by a cross . ( b ) Observed dengue incidence rate ( DIR ) categories for June , 2014 . DOI: http://dx . doi . org/10 . 7554/eLife . 11285 . 005 Figure 3 shows the probability of DIR falling in the category that was actually observed . The deeper the colour shading , the greater the probability of observing the correct category . This gives an indication of the certainty of the model in predicting correct outcomes . In general , a high degree of certainty in the forecast is found in the south region , parts of the Amazon and many densely populated cities along the eastern coastline . However , as the historical distribution of DIR is not symmetrical , with a greater proportion of the distribution in the low category , compared to the high category ( as epidemics can be considered as 'extreme events' ) , it is interesting to consider each category individually . Figure 4a–c show conditional maps of the forecast probability given that low , medium and high DIR was observed , respectively . The grey areas indicate areas where the observed DIR fell in the other two categories and are therefore not considered in each individual map . The probability trigger thresholds defined in Lowe et al . , ( 2014 ) are taken into account to weight the graduated colour bars , ranging from 0% to 100% chance of the observed category . Using the forecasting model , if the probability of low risk were greater than 68% , a low risk warning would have been assigned . If the probability of low risk were less than or equal to 68% , a medium risk warning would have been assigned ( giving a medium trigger threshold of 32% ) . If simultaneously , high risk were greater than 18% , a high risk warning would have been assigned . Therefore , lower probabilities are assigned more weight ( represented by colour darkness ) in the high category plot than the low category plot . Given that low risk was observed , the model framework would have correctly assigned a low risk warning for 67% of the microregions . Given that high risk was observed , the model framework would have correctly assigned a high risk warning for 57% of the microregions . High risk was correctly forecast with considerable certainty in microregions in the north east of Brazil near Fortaleza ( see Figure 4c ) . Although the model 'missed' the high risk observed in Brasília , it was able to correctly detect , with a relatively high degree of certainty high risk in surrounding microregions . 10 . 7554/eLife . 11285 . 006Figure 3 . Forecast probability of observed DIR categories for June 2014 . Probability of observing the correct DIR category ( low , medium and high ) . The graduated colour bar represents the probability of observing any given category ( ranging from 0% , pale colours , to 100% , deep colours ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11285 . 00610 . 7554/eLife . 11285 . 007Figure 4 . Forecast probability of observed DIR in the low , medium and high category for June 2014 . Forecast probability given that ( a ) low , ( b ) medium and ( c ) high DIR was observed . Grey areas indicate that other DIR categories were observed and are therefore not considered . The graduated colour bar represents the probability of observing the given category ( ranging from 0% , pale colours , to 100% , deep colours ) . Note , the alarm trigger thresholds are marked with a star ( * ) . For ( a ) low risk warnings , pL > 68% , for ( b ) medium risk warnings , pM > 32% and for ( c ) high risk warnings , pH > 18% . Colour bars are weighted , with increased saturation beyond the alarm trigger threshold to reflect the correct assignation of warnings . DOI: http://dx . doi . org/10 . 7554/eLife . 11285 . 007 Useful predictions from a forecasting system are likely to be those that recommend changes from the activities that would otherwise have taken place anyway , which are typically based on the 'normal' dengue season . Beyond that , predictions that forecast higher than expected incidence are critical , as they could advocate increased interventions . To assess the performance of the forecast model framework beyond a simple seasonal profile , we defined a null model as the average DIR in each microregion for June 2000–2013 . We consider the ability of both the forecast model and the null model to predict 'high risk' dengue across Brazil . Table 2 shows a summary of contingency table results for observed DIR exceeding the high risk epidemic threshold ( 300 cases per 100 , 000 inhabitants ) using the probabilistic category forecast model and the null model for June 2014 . Results show that for the June 2014 event , the forecast model predicted a greater number of true positives ( hits ) and fewer false negatives ( misses ) than the null model ( see Table 2 , Materials and methods ) . This gave a hit rate of 57% ( miss rate of 43% ) when using the forecast model and a hit rate of 33% ( miss rate of 67% ) when using the null model . However , the forecast model also tended to produce more false positives ( or false alarms ) than the null model ( see Table 2 ) . The two types of error ( false alarms and missed events ) have very different consequences for public health . For example , failing to predict an epidemic that then occurs ( type II error – a miss ) is much more damaging than predicting an epidemic that does not materialise ( type I error – a false alarm ) ( Stephenson , 2000 ) . Figure 5 shows hit rates and false alarm rates for both the forecast and null model , calculated in 'leave one year out' cross-validation mode from 2000–2013 , i . e . by excluding the year for which the prediction is valid when estimating model parameters ( see Materials and methods ) . Results for the 2014 event are also included . The hit rate for the forecast model exceeds that of the null model for all years , expect 2004 , when dengue incidence was at its lowest across the whole of Brazil . 10 . 7554/eLife . 11285 . 008Table 2 . Summary of contingency table results for observed DIR exceeding the epidemic risk threshold . Summary of contingency table results for observed DIR exceeding the high risk epidemic threshold ( 300 cases per 100 , 000 inhabitants ) using the probabilistic category forecast model and the null model ( mean DIR , June 2000–2013 ) for June 2014 . DOI: http://dx . doi . org/10 . 7554/eLife . 11285 . 008Performance measuresForecast model probabilisticNull model seasonal meanHit8146False alarm ( type I error ) 9455Miss ( type II error ) 6095Correct rejection318357Hit rate57%33%False alarm rate23%13%Miss rate43%67%10 . 7554/eLife . 11285 . 009Figure 5 . Hit rate and false alarm rate for predicting dengue in the high risk category for June 2000–2014 using the forecast model and null model . Comparison of ( a ) hit rates and ( b ) false alarm rates for the event of observed DIR exceeding the high risk epidemic threshold ( 300 cases per 100 , 000 inhabitants ) using the probabilistic category forecast model ( blue circles ) and the null model ( orange triangles ) for June 2000–2014 . The vertical bars around each point represent the 95% confidence intervals . The vertical dotted line separates the leave-one-out cross validation results ( 2000–2013 ) from the true predicted results for 2014 . DOI: http://dx . doi . org/10 . 7554/eLife . 11285 . 009 To assess the additional value of the forecasting system beyond that of the seasonal profile , it is useful to consider the full posterior predictive distributions from the model , compared to the null model and associated prediction intervals . Figure 6 shows time series of observed and predicted dengue incidence rates for June 2000–2014 for the 12 host microregions . The posterior predictive mean and upper 95% prediction ( credible ) interval from the forecast model , and the sample mean and upper 95% prediction interval from the null model are also included . The forecast and null model predictions are calculated in 'leave one year out' cross-validation mode from 2000–2013 , i . e . , by excluding the year for which the prediction is valid when estimating model parameters ( see Materials and methods ) . Note , the predictions for 2014 are the only 'true' forecasts ( i . e . , no information is included beyond the forecast issue date ) . When considering the posterior predictive mean of the forecast model , some of the inter-annual variations in the observations are captured by the model , for example in Belo Horizonte , Manaus and Salvador . However , in some other places , the mean prediction from the forecast and the null model are either very similar or , in some years , the null model mean is closer to the observed value , for example , in Cuiabá in 2009 . During years with relatively low DIR , the predictions from the forecast model tend to be more precise than the null model , with narrower prediction ( credible ) intervals . Further , when DIR is exceptionally high , the forecast model is able to account for this increased possibility of an outbreak in most cases , compared to the null model . This is evident for the dengue epidemics that occurred in Belo Horizonte in 2010 and 2013 ( Figure 6a ) , Salvador in 2010 ( Figure 6j ) and Manaus in 2011 ( Figure 6f ) . Although the forecast model is far from perfect , in general , it is better able to detect extreme dengue incidence rates than the null model . 10 . 7554/eLife . 11285 . 010Figure 6 . Time series of observed and predicted DIR for June 2000–2014 for host microregions . Observed DIR ( pink squares ) , posterior mean DIR ( blue circles ) and upper 95% prediction ( credible ) interval from forecast model ( blue dashed line ) and mean DIR ( orange triangles ) and upper 95% prediction interval ( orange dashed line ) from null model , June 2000–2014 in the host microregions ( a ) Belo Horizonte , ( b ) Brasília , ( c ) Cuiabá , ( d ) Curitiba , ( e ) Fortaleza , ( f ) Manaus , ( g ) Natal , ( h ) Porto Alegre , ( i ) Recife , ( j ) Salvador , ( k ) São Paulo and ( l ) Rio de Janeiro . The vertical dotted line separates the leave-one-out cross validation results ( 2000–2013 ) from the true predicted results for 2014 . DOI: http://dx . doi . org/10 . 7554/eLife . 11285 . 010
Several studies have developed models for dengue fever , using climate and other risk factors , and tested predictive performance in retrospective mode . However , none of these studies have incorporated real-time seasonal climate forecasts and epidemiological data to predict future dengue risk . Therefore , to our knowledge , this work constitutes the first evaluation of a nationwide dengue early warning , issued before a global mass gathering . The dengue early warnings were disseminated to the Ministry of Health , the general public and visitors travelling to Brazil , prior to the World Cup . The predictions were incorporated into the European Centre for Disease Control ( ECDC ) health risk assessment ( ECDC , 2014 ) and reported by more than 18 international press outlets . As a result , the forecast by Lowe et al . ( 2014 ) , along with others ( Hay , 2013; Massad et al . , 2014 ) , further contributed by raising general awareness about dengue fever and the risk of contracting the disease when travelling to endemic regions . This dengue early warning framework may be useful , not only ahead of mass gatherings , but also before the peak dengue season each year , to control or contain potentially explosive dengue epidemics . The use of real-time seasonal climate forecasts and early epidemiological reports in routine dengue early warnings is now a priority for the Brazilian Climate and Health Observatory ( www . climasaude . icict . fiocruz . br ) , in collaboration with the Brazilian Institute for Space Research . We hope this prototype will serve as a demonstration for scientists , health surveillance teams and decision makers of the data and tools required to produce , communicate and evaluate timely predictions of climate-sensitive disease risk .
We obtained dengue data for June 2014 from the Notifiable Diseases Information System ( SINAN ) , organised by the Brazilian Ministry of Health . We then aggregated the cases for the 5570 municipalities , 42% of which have less then 10 , 000 inhabitants , to the microregion level . A microregion is defined as an aggregate of neighbouring municipalities , with common economic interests and frequent population exchanges . This helps to alleviate problems of low population numbers and misreporting , due to variations in availability of health services/epidemiological facilities at the municipality level . This data includes confirmed cases of dengue fever , including mild infections , dengue haemorrhagic fever , and shock syndrome . Dengue cases can be confirmed by laboratory exams or clinical and epidemiological evidence . In the second case , a patient must present at least two of the following symptoms: high fever , severe headache , severe eye pain , joint and muscle pain , mild bleeding manifestation or low white cell count . In addition to these symptoms , the patient must have been in areas where dengue is being transmitted or where there has been an infestation of Ae . aegypti in the past 15 days ( Ministério da Saúde , 2005 ) . Dengue notification records are considered a priority in the epidemiological surveillance system in Brazil . Data flow is accelerated in relation to other diseases . About 50% of cases are reported within 3 days after the first symptoms , and 90% of cases are digitised within 7 days of notification ( Barbosa et al . , 2015 ) . During outbreaks , this flow tends to be speeded up , adopting optimisation measures , for example , using 'clinical and epidemiological evidence' to confirm cases that could not be submitted for laboratory confirmation . These criteria allow accelerated case reporting , prioritising sensitivity and opportunity rather than specificity of information ( Duarte and França , 2006 ) . In fact , during high incidence periods , the proportion of cases confirmed by laboratory criteria is lower than during low incidence periods and 'clinical and epidemiological evidence' is a common procedure for case confirmation . On the other hand , about 50% of suspect cases are subsequently confirmed after laboratory tests in epidemic periods , while in periods of low transmission intensity , approximately 30% of cases are confirmed , revealing good predictive value of suspected cases ( Barbosa et al . , 2015 ) . Therefore , after post-processing , some cases are discarded because they have negative serology . However , beyond these 30% of cases , identification of serotype is not carried out by laboratory tests , which hinders the understanding of transmission dynamics and population susceptibility level . Since recent data are still subject to confirmation of cases and elimination of duplicate registers , the initial figures of dengue cases may be modified in the following months and will most likely be official by the end of 2015 . These data will then be made publicly available via the Health Information Department ( DATASUS , http://dtr2004 . saude . gov . br/sinanweb/ ) . We used 2014 population estimates obtained from the Brazilian Institute for Geography and Statistics ( IBGE , 2014 ) to convert the case data into dengue incidence rates ( DIR ) , per 100 , 000 inhabitants . Other estimates of population for inter census years are produced by national institutions and may present discrepancies , mainly for small populations and newly created municipalities . According to the methodology used by the Ministry of Health , the DIR is calculated for a geographical space in a given year ( PAHO , 2008 ) . As the dengue risk thresholds used by the National Dengue Control Programme are based on yearly dengue incidence rates ( Ministério da Saúde , 2008 ) , it is necessary to use a proportion ( 1/12 ) of yearly population estimate as the denominator in the dengue incidence rate calculation , to make use of this at a monthly time scale . Therefore , we converted the monthly incidence to yearly equivalent incidence to make use of the risk thresholds of 100 and 300 cases per 100 , 000 per year . This is consistent with the metrics published in the Epidemiological Bulletins of the Ministry of Health . Our goal was to provide measures that can be easily interpreted by the Dengue Control Programme and translated into well understood risk levels ( low , medium , high ) . A spatio-temporal Bayesian hierarchical model ( Lowe et al . , 2011 , 2013 , 2014 ) was formulated , using monthly dengue cases , from 2000 to 2013 , for 553 Brazilian microregions as the response variable . Based on findings from previous studies ( Lowe et al . , 2013 ) , the climate variables used to formulate the model were three-month average temperature ( Fan and Van den Dool , 2008 ) and precipitation ( Adler et al . , 2003 ) anomalies ( departures from the long-term average ) , over the three months preceding the dengue month of interest . This is equivalent to a two month lag when considering the mid-point of the three month average . Lags of 1–3 months are typically used when modelling dengue ( Lowe et al . , 2015b ) , to try and capture the impact of rainfall on mosquito breeding sites and the effect of temperature on the mosquito life cycle , although these relationships are still not well understood . Other explanatory variables included population density , altitude , and dengue relative risk ( ratio of observed to expected cases ) lagged by four months . Zone-specific seasonality was accounted for using autocorrelated annual cycles ( i . e . by allowing each calendar month to depend on the previous month ) for different Brazilian ecological zones ( e . g . Amazon , Caatinga , Cerrado , Atlantic Pampa , Pantanal ) . Unknown confounding factors ( e . g . health care and vector control disparities between microregions ) and dependency structures ( i . e . , human mobility between neighbouring areas ) were allowed for using area-specific unstructured and structured random effects ( see Lowe et al . ( 2014 ) for further details ) . To produce the forecast for June 2014 , the model was driven by ( 1 ) real-time seasonal precipitation and temperature anomaly forecasts ( Coelho et al . , 2006 ) , produced in mid-February by the Center for Weather Forecasting and Climate Research ( CPTEC ) ( valid for the March-May [MAM] season ) and ( 2 ) the observed epidemiological situation ( ratio of observed to expected cases ) for February , 2014 , collated in March , 2014 by the Ministry of Health [see Lowe et al . ( 2014 ) for details] . Note , the precipitation seasonal forecasts used in this study were produced by CPTEC as part of EUROBRISA: A Euro-Brazilian Initiative for improving South American seasonal forecasts ( http:eurobrisa . cptec . inpe . br ) . Posterior predictive distributions were simulated for every microregion to determine the probability of dengue incidence rates exceeding predefined risk thresholds ( see Figure 1 ) . Probability forecasts ( pL , pM , pH ) were issued for low ( fewer than 100 dengue cases per 100 , 000 inhabitants ) , medium ( between 100 and 300 dengue cases per 100 , 000 inhabitants ) and high ( more than 300 dengue cases per 100 , 000 inhabitants ) risk . These results were presented using a visualisation technique ( Jupp et al . , 2012 ) , where the forecast for each microregion was expressed as a colour determined by a combination of three probabilities , with colour saturation used to indicate certainty for a particular category ( see Figure 2a ) . We then used a receiver operating characteristic ( ROC ) analysis of past forecasts and observations from 2000–2013 to define optimal probability thresholds for warnings . If the probability of low risk was less than 68% , a medium risk forecast warning was issued . If the probability of high risk was concurrently greater than 18% , the forecast warning was upgraded to high risk ( see Table 1 ) . After the event , we compared the published probabilistic predictions with observed DIR data for June 2014 . We defined a null model as the seasonal average of past dengue incidence ( i . e . , mean DIR for June 2000–2013 ) . We assessed the ability of the forecast model and the null model to determine the binary event of DIR exceeding 300 cases per 100 , 000 inhabitants ( i . e . the high risk threshold ) for n=553 microregions in Brazil ( Table 2 ) . Table 3 shows the two ways for the forecast to be correct ( either a hit or a correct rejection ) and two ways for the forecast to be incorrect ( either a false alarm or a miss ) . Cell count a is the number of events correctly forecast to occur , i . e . the number of hits; cell count b is the number of events incorrectly forecast to occur , i . e . , the number of false alarms; cell count c is the number events incorrectly forecast not to occur , i . e . , the number of misses; and cell count d is the number of event correctly forecast not to occur , i . e . , the number of correct rejections ( Jolliffe and Stephenson , 2012 ) . We calculated performance measures , such as the hit rate; the proportion of events ( i . e . , epidemics ) that were correctly predicted ( a/ ( a+c ) , also know as true positive rate or sensitivity ) and the false alarm rate; the proportion of events that were predicted but did not occur ( b/ ( b+d ) , also know as false positive rate or 1-specificity ) . The false alarm rate can be interpreted as the rate of making a 'type I error' , whereas the 'miss rate , ' equal to one minus the hit rate , measures the rate of making a 'type II error' ( Stephenson , 2000 ) . 10 . 7554/eLife . 11285 . 011Table 3 . The four possible outcomes for categorical forecasts of a binary event . DOI: http://dx . doi . org/10 . 7554/eLife . 11285 . 011Event observed YesNoTotalForecast warning issued YesHit ( a ) False alarm ( b ) a+bNoMiss ( c ) Correct rejection ( d ) c+dTotala+cb+da+b+c+d=n We compared time series of observed and predicted DIR from the forecast and null model for June 2000–2014 in the twelve host microregions . The mean of the posterior predictive distribution from the forecast model and the 95% prediction ( credible ) interval , obtained from the 2 . 5% and 97 . 5% percentiles of the distribution were calculated for June 2014 , using the forecast model fitted to data from 2000–2013 ( note , the lower 95% prediction interval from the forecast model was nearly always equal to zero and is therefore not shown in the figures ) . For the years 2000–2013 , the model was fitted 14 times , excluding one year at a time when estimating model parameters , to produce 'cross validated' predictions to test against 'out-of-sample' data ( i . e . the year for which the predictions are valid ) . For the null model , the 95% prediction intervals for the sample mean was calculated as y¯±tn−1 , α/2s1+1n , where y¯ is the sample mean , tn−1 , α/2 is the 100 ( 1-α/2 ) th percentile of T Distribution , with n−1 degrees of freedom , and s is the standard error . For the 2014 null model prediction , the mean , standard error and 95% prediction intervals were calculated using past data for June 2000–2013 ( n=14 ) . For 2000–2013 , the null model mean , standard error and 95% prediction intervals were calculated in cross-validated mode , by excluding one year at a time ( n=13 ) ( note , the lower 95% prediction interval for the null model was nearly always less than zero and is therefore not shown in the figures ) . | Dengue is a viral infection spread by mosquitoes and is widespread in tropical and sub-tropical regions . Dengue epidemics in Brazil often occur without warning , and can overwhelm the public health services . Forecasts of seasonal climates combined with early data from a dengue surveillance system could help public health services anticipate dengue outbreaks several months in advance . However , this information has not been previously exploited to predict dengue epidemics in a practical real-life framework . Recently , a group of researchers developed a prototype of a dengue early warning system based on 13 years worth of data , and used it to predict the risk of dengue three months ahead of the 2014 FIFA World Cup in Brazil . Now Lowe et al . – including most of the researchers involved in the earlier work – have evaluated the prototype against the actual reported cases of dengue during the event . Brazil is divided into over 550 'microregions' , and the forecasts correctly predicted high risk of dengue for 57% of the microregions reporting high levels of dengue during the games . Forecasts based on seasonal dengue averages would have only detected high risk in 33% of these microregions . The forecasts also correctly predicted the dengue risk level in seven out of the twelve cities where the World Cup games were hosted . However , the prototype failed to predict the high risk in both São Paulo and Brasília . Lowe et al . speculate that this may have been due to changes in how water was stored in these cities ( standing water is a breeding site for mosquitoes ) and the circulation of a new strain of the dengue virus . The implementation of seasonal climate forecasts and early reports of dengue cases into an early warning system is now a priority for public health authorities . This action is likely to help them to prepare for and minimize epidemics of dengue and other diseases that are spread by mosquitoes , such as chikungunya and Zika virus . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"epidemiology",
"and",
"global",
"health",
"microbiology",
"and",
"infectious",
"disease"
] | 2016 | Evaluating probabilistic dengue risk forecasts from a prototype early warning system for Brazil |
Integrins are heterodimeric cell surface adhesion and signaling receptors that are essential for metazoan existence . Some integrins contain an I-domain that is a major ligand binding site . The ligands preferentially engage the active forms of the integrins and trigger signaling cascades that alter numerous cell functions . Here we found that the adenylate cyclase toxin ( CyaA ) , a key virulence factor of the whooping cough agent Bordetella pertussis , preferentially binds an inactive form of the integrin complement receptor 3 ( CR3 ) , using a site outside of its I-domain . CyaA binding did not trigger downstream signaling of CR3 in human monocytes and CyaA-catalyzed elevation of cAMP effectively blocked CR3 signaling initiated by a natural ligand . This unprecedented type of integrin-ligand interaction distinguishes CyaA from all other known ligands of the I-domain-containing integrins and provides a mechanistic insight into the previously observed central role of CyaA in the pathogenesis of B . pertussis .
Integrins are dimeric transmembrane proteins complexes composed of an alpha and a beta subunit . There are 18 different alpha subunits and 8 beta subunits that combine in a limited number of combinations of which 24 are currently known in mammals ( Tan , 2012 ) . Integrins are essential for regulation of numerous cellular functions including cell signaling and adhesion . Nine of the eighteen integrin alpha subunits harbor a conserved I ( inserted ) -domain that is crucial for binding of endogenous ligands ( Johnson and Chouhan , 2014 ) . Four of these integrin alpha subunits ( αD , αL , αM and αX ) form heterodimers exclusively with the β2 subunit , thus forming the αDβ2 ( CD11d/CD18 ) , αLβ2 ( CD11a/CD18 , LFA-1 ) , αMβ2 ( CD11b/CD18 , complement receptor 3 ( CR3 ) , Mac1 ) and αXβ2 ( CD11c/CD18 , CR4 , p150/195 ) integrins , respectively ( Arnaout , 1990; Mazzone and Ricevuti , 1995; Sanchez-Madrid , 1983; Tan , 2012; Van der Vieren et al . , 1995 ) . The β2 integrins have specialized roles in immune and inflammatory responses and , like other integrins , employ a two-step mechanism of bi-directional signal transmission between the interior of cells and the extracellular milieu ( Anthis and Campbell , 2011; Tan , 2012 ) . Upon activation by various intracellular signals , the inside-out signaling is initiated through rearrangement of the integrin molecule from an inactive ( bent , closed , resting , low-affinity ) conformation to an active ( extended , open , high-affinity ) conformation . Subsequent ligand binding triggers outside-in signaling of the extended integrins through activation of Src family tyrosine kinases ( Jakus et al . , 2007; Mócsai et al . , 2002; 2010; Schymeinsky et al . , 2007 ) . Src kinases phosphorylate tyrosine residues within the so-called immunoreceptor tyrosine-based activation motif ( ITAM ) , on the cytoplasmic face of ITAM-containing transmembrane adaptor proteins , such as DAP12 or the FcR γ-chain ( FcRγ ) ( Jakus et al . , 2007; Mócsai et al . , 2006; 2010; Schymeinsky et al . , 2007 ) . These serve as docking sites for the tandem phosphotyrosine-binding Src homology 2 ( SH2 ) domains of the non-receptor spleen tyrosine kinase ( Syk ) ( Jakus et al . , 2007; Mócsai et al . , 2006; 2010; Schymeinsky et al . , 2007 ) . Recruitment and activation of Syk then leads to assembly of a multi-protein signaling complex that contains cytosolic Syk-binding molecules and initiates further downstream signaling , ultimately triggering various cellular responses that play a central role in the innate immune defense to infection ( Mócsai et al . , 2010 ) . The β2 integrin complement receptor 3 ( CR3 ) is used as receptor by the 1706 residue-long RTX ( Repeats in ToXin ) adenylate cyclase toxin-hemolysin ( CyaA , ACT , or AC-Hly ) of Bordetella pertussis , which plays a crucial role in virulence and immune evasion of the whooping cough agent ( Goodwin and Weiss , 1990; Guermonprez et al . , 2001; Khelef et al . , 1992; Vojtova et al . , 2006; Weiss et al . , 1984 ) . Upon binding to CR3 , CyaA penetrates myeloid phagocytes and paralyzes their bactericidal functions by catalyzing uncontrolled conversion of cytosolic ATP to the key signaling molecule cAMP ( Confer and Eaton , 1982; Guermonprez et al . , 2001 ) . The toxin also hijacks maturation and proinflammatory signaling of CR3-expressing dendritic cells and it likely subverts antigen presentation and induction of adaptive T cell immune responses to bacterial infection by intraepithelial dendritic cells of host respiratory mucosa ( Bagley et al . , 2002; Boschwitz et al . , 1997; Boyd et al . , 2005; Fedele et al . , 2010; Hickey et al . , 2008; Njamkepo et al . , 2000; Ross et al . , 2004; Spensieri et al . , 2006 ) . Here we reveal that CyaA acts as a unique type of ligand of the I-domain-containing integrin CR3 , preferentially recognizing an inactive state of the integrin through interaction at a novel binding site located outside of the I-domain . CyaA thereby avoids activation of downstream signaling of the engaged CR3 via the key signaling kinase Syk in human monocytes . Furthermore , we show that CyaA-catalyzed elevation of cAMP effectively blocked the iC3b opsonin-elicited activation of the CR3-Syk signaling pathway in human monocytes .
It was previously shown that Chinese hamster ovary ( CHO ) cells expressing human CR3 can be used as a suitable model for studying the interaction of CyaA with CR3 ( Guermonprez et al . , 2001 ) . Indeed , CR3 ( CD11b/CD18 ) expressed by CHO cells allowed the binding and cAMP-elevating ( cytotoxic ) activities of CyaA , while the highly homologous CR4 ( CD11c/CD18 ) was unable to bind CyaA despite sharing the same β2 ( CD18 ) subunit with CR3 ( Guermonprez et al . , 2001 ) . Therefore , to delineate the CyaA binding site ( s ) on CR3 , we performed swapping of the homologous alpha chain segments ( CD11b and CD11c ) of CR3 and CR4 ( Figure 1A ) . First , a CHO cell line stably expressing the CD18 subunit ( CHO-CD18 ) was established ( Figure 1—figure supplement 1 ) . This was next used for generating stable cell lines that expressed similar quantities of intact CD11b ( CHO-CD11b/CD18 ) , CD11c ( CHO-CD11c/CD18 ) , or of the chimeric CD11b-CD11c heterodimeric complexes on the cell surface ( Figure 1—figure supplement 2 ) . The capacity of such cells to bind CyaA was then assessed by flow cytometry and the susceptibility to CyaA penetration and enzymatic intoxication by the delivered AC domain of the toxin was measured as intracellular accumulation of cAMP . As shown in Figure 1B , replacement of the segment containing residues 614 to 682 ( segment 614-682 ) of CD11b by the corresponding portion of CD11c caused a sharp , about ten-fold decrease of CyaA binding to the CHO-CD11b614-682c/CD18 transfectants , as compared to cells expressing intact CR3 ( Figure 1B ) . This resulted in accordingly reduced toxin penetration and cAMP accumulation in cells ( Figure 1C ) . CyaA binding was also partially reduced on cells expressing CR3 chimeras having the residues 33 to 115 ( 72 ± 6% ) , 342 to 424 ( 61 ± 10% ) , or 686 to 1152 ( 89 ± 11% ) of CD11b replaced by the corresponding segments of CD11c , respectively ( Figure 1B ) . No reduction of CyaA binding was observed on cells expressing the remaining CD11b-CD11c/CD18 chimeras ( Figure 1B ) , which all accumulated comparable levels of cAMP as the cells expressing intact CR3 ( Figure 1C ) . Surprisingly , CyaA binding and subsequent intoxication of cells by cAMP were neither reduced by swapping of the homologous I-domain regions between CD11b and CD11c ( CD11b120-272c and CD11b275-339c ) , nor upon removal of the entire I-domain of CD11b ( CD11bΔ164-339 ) ( Figure 1B , C ) . This demonstrates that CyaA does not bind CR3 through the I-domain and rather recognizes the segment containing residues 614 to 682 that encompasses the C-terminal end of the last repeat of the β-propeller domain and the N-terminal portion of the thigh domain of CD11b . This finding is consistent with the sequence alignment by the Jotun Hein method ( Lasergene software , DNASTAR ) showing that the 614-682 segments of CD11b and CD11c are only 41% identical at the amino acid level , compared to the entire CD11b and CD11c subunits that exhibit an overall 61% sequence identity . Since , however , some other segments of CD11b appeared to be important for full toxin binding to CR3 as well , these results suggest that CyaA likely binds the integrin through a multivalent interaction , making contacts with several integrin segments . Moreover , CyaA bound CD11b only when this was associated with CD18 and not when the CD11b subunit was expressed alone on the surface of CHO-CD11b cells ( Figure 1—figure supplement 3 ) . 10 . 7554/eLife . 10766 . 003Figure 1 . Residues 614-682 of CD11b are crucial for CyaA binding and penetration into cells . ( A ) The CD11 subunits of β2 integrins consist of a long N-terminal extracellular domain , a single-pass transmembrane segment ( TS ) and a short C-terminal cytoplasmic tail , respectively . The N-terminal part of the extracellular domain harbors seven β-sheet repeats ( numbers in boxes ) , forming a β-propeller domain , which is followed by the thigh , calf-1 and calf-2 domains . The I-domain segment , inserted between repeats 2 and 3 of the β-propeller domain , plays a critical role in interaction of the I-domain-containing integrins with their endogenous ligands . To map the CyaA binding site on the CD11b subunit , segments of CD11b ( green ) were systematically replaced with their CD11c counterparts ( blue ) . In the CD11bΔ164-339 molecule , the entire I domain of CD11b was deleted . ( B ) 2x105 CHO cells expressing integrin molecules were incubated with 2 µg/ml of CyaA-biotin , the surface-bound toxin was labeled with streptavidin-PE and the cells were analyzed by flow cytometry . CyaA binding was expressed as percentage of toxin binding to CHO cells expressing the native form of CD11b/CD18 . Each bar represents the mean value with SD of at least five independent experiments performed in duplicate or triplicate . Significantly reduced binding of CyaA to mutant integrins in comparison with intact CD11b/CD18 is indicated ( **** , p<0 . 0001; ANOVA ) . ( C ) 1x105 CHO cells expressing integrin molecules were incubated with various concentrations of CyaA and the amounts of accumulated cAMP were determined in cell lysates by ELISA . Each point represents the mean value ± SD of at least seven determinations from at least three independent experiments . Significant differences between mean values of cAMP intoxication of cells expressing intact CD11b/CD18 and mutant integrins are shown ( **** , p<0 . 0001; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 00310 . 7554/eLife . 10766 . 004Figure 1—figure supplement 1 . Expression of the CD18 subunit on the surface of CHO cells . A CHO cell line was transfected with a plasmid construct encoding CD18 and the positively transfected cells were selected using a cell sorter . The expression of the CD18 subunit ( violet ) on the cell surface was examined by flow cytometry upon staining of 2x105 cells with the anti-CD18 mAb MEM-48 . CHO cells transfected with an empty vector were processed in parallel and used as negative control ( grey ) . A typical flow cytometry histogram from one representative binding experiment out of five performed is given . RFI , relative fluorescence intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 00410 . 7554/eLife . 10766 . 005Figure 1—figure supplement 2 . Binding of mAbs and CyaA to CHO cells expressing CD11b/CD18 , CD11c/CD18 and the CD11b-CD11c/CD18 chimeras . A CHO cell line stably expressing intact CD18 subunit was transfected with plasmid constructs encoding CD11b , CD11c , or their respective chimeric variants . Cells stably expressing the integrin molecules were selected using a cell sorter . The expression levels of CD11b/CD18 ( green ) , CD11c/CD18 ( blue ) and CD11b-CD11c/CD18 chimeras ( red ) on the cell surface were examined by flow cytometry upon staining of 2x105 cells with the ICRF 44 mAb ( recognizing the I-domain of CD11b ( left panels ) ) , the OKM1 mAb ( recognizing segment 614-682 of CD11b ( middle left panels ) ) , or the 3 . 9 mAb ( recognizing the I-domain of CD11c ( middle right panels ) ) . The cells were also analyzed for capacity to bind CyaA: 2x105 transfected cells were incubated with 2 μg/ml of CyaA-biotin , the surface-bound toxin was labeled with streptavidin-PE and the cells were analyzed by flow cytometry ( right panels ) . A typical histogram from one representative binding experiment out of five performed is shown for binding of the respective mAb ( ICRF 44 , OKM1 or 3 . 9 ) , or CyaA to each integrin variant . CHO cells expressing no β2 integrin ( transfected with empty vectors ) were processed in parallel and used as negative control ( grey ) . RFI , relative fluorescence intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 00510 . 7554/eLife . 10766 . 006Figure 1—figure supplement 3 . CyaA recognizes CD11b only in the heterodimeric complex with CD18 . ( A and B ) 1x105 CHO cells stably expressing intact dimeric CD11b/CD18 or its individual monomeric subunits CD11b or CD18 were stained with the anti-CD11b mAb OKM1 ( A ) or the anti-CD18 mAb MEM-48 ( B ) and analyzed by flow cytometry . A typical overlay flow cytometry histogram from one representative binding experiment out of three performed is given . ( C ) 1x105 integrin-transfected CHO cells were incubated with different concentrations of Dy647-labeled CyaA and the cells were analyzed by flow cytometry . CyaA binding data were deduced from the mean fluorescence intensities of three independent experiments and each point represents the mean value ± SD . Binding of CyaA to cells expressing CD11b , CD18 or no β2 integrin was at all measured toxin concentrations significantly lower than toxin binding to cells expressing intact CD11b/CD18 ( **** , p<0 . 0001; ANOVA ) . CHO cells transfected with an empty vector and expressing no β2 integrin were processed in parallel and used as negative control ( all panels ) . RFI , relative fluorescence intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 006 To exclude that the reduction of CyaA binding may have resulted from a structural alteration of the integrin chimeras , we next mapped the CyaA binding site on intact CR3 by performing competition experiments with a set of CD11b-specific monoclonal antibodies ( mAbs; Figure 2A and Figure 2—figure supplement 1 ) that do not activate CR3 ( Figure 2—figure supplement 2 ) . The I-domain-recognizing mAbs ICRF 44 , 44 and 2LPM19c ( Figure 2A and Figure 2—figure supplement 1 ) , the CD11c-specific mAb 3 . 9 , the CD18-specific mAb MEM-48 and the isotype control mAbs , all failed to substantially reduce CyaA binding to CR3 ( Figure 2B ) . In contrast , CyaA was strongly outcompeted from binding to intact CR3 by the excess of OKM1 , VIM12 , or M1/70 mAbs ( Figure 2B ) that recognize epitopes located between the residues 614 to 682 of CD11b ( Figure 2A and Figure 2—figure supplement 1 ) . This strongly supports the conclusion reached by the CD11b-CD11c domain swapping experiments that the segment encompassing residues 614 to 682 of CD11b is specifically involved in binding of CyaA . In addition , CyaA binding to CR3 was strongly inhibited by the MEM-174 mAb ( Figure 2B ) that recognizes the CD11b segment comprised between residues 342 to 424 ( Figure 2A and Figure 2—figure supplement 1 ) . This also concurs with the results of the segment swapping experiments , showing that the segment 342-424 along with some other segments of CD11b may also be contributing to CyaA binding to CR3 . 10 . 7554/eLife . 10766 . 007Figure 2 . Antibodies recognizing the same segments of CD11b as CyaA block its binding to CR3 . ( A ) Schematic representation showing the binding segments of a set of mAbs in the CD11b subunit of CR3 , which were mapped by flow cytometry . The ICRF 44 , 44 and 2LPM19c mAbs recognize the I-domain of CD11b , the major ligand binding site of CR3 . The MEM-174 , or OKM1 , VIM12 and M1/70 mAbs target amino acid segments 342-424 or 614-682 of CD11b , respectively , which are important for CyaA binding . ( B ) 2x105 CHO-CD11b/CD18 cells were preincubated without or with saturating concentrations of different mAbs and then incubated with 2 µg/ml of CyaA-biotin . Surface-bound CyaA-biotin was labeled with streptavidin-PE and the cells were analyzed by flow cytometry . CyaA binding was expressed as percentage of toxin binding to CHO-CD11b/CD18 cells treated without mAb . Each bar represents the mean value with SD of at least eight determinations from at least three independent experiments . Significant differences between mean values of CyaA binding to mAb-untreated cells and cells treated with different mAbs are indicated ( *** , p<0 . 001; **** , p<0 . 0001; ANOVA ) . 3 . 9 , CD11c-specific mAb; MEM-48 , CD18-specific mAb; IgG1 , IgG2a and IgG2b , isotype control mAbs . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 00710 . 7554/eLife . 10766 . 008Figure 2—figure supplement 1 . Flow cytometry profiles of anti-CD11b mAbs on CHO cells expressing the integrin CD11b/CD18 and its mutant variants . 2x105 CHO cells stably expressing intact CD11b/CD18 , or the mutant variant lacking the I-domain ( CD11bΔ164-339/CD18 ) , or two CR3-CR4 chimeras ( CD11b342-424c/CD18 and CD11b614-682c/CD18 ) with a strongly reduced capacity to bind CyaA were left unstained ( grey ) or were stained with a panel of anti-CD11b mAbs ( red ) . The cells were analyzed for their capacity to bind mAbs by flow cytometry and typical overlay histograms from one representative binding experiment out of five performed are shown . RFI , relative fluorescence intensity . The data confirmed that the ICRF 44 , 44 and 2LPM19c mAbs recognize the I-domain of CR3 and revealed that the MEM-174 mAb has an epitope in the segment 342-424 of CD11b and the OKM1 , VIM12 and M1/70 mAbs target the segment 614-682 of CD11b . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 00810 . 7554/eLife . 10766 . 009Figure 2—figure supplement 2 . Anti-CD11b antibodies do not activate CR3 . 2x105 CHO-CD11b/CD18 cells were preincubated without or with saturating concentrations of the indicated anti-CD11b mAbs and the CR3 was stained with the integrin activation-reporting anti-CD18 mAb MEM-148 . MEM-148 binding was determined by flow cytometry and expressed as percentage of MEM-148 binding to CHO-CD11b/CD18 cells treated without mAb . Each bar represents the mean value with SD of two independent experiments performed in duplicate . Significant differences between mean values of MEM-148 binding to mAb-untreated cells and cells treated with different anti-CD11b mAbs are indicated ( *** , p<0 . 001; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 009 To confirm that the 614-682 segment of CD11b is the principal binding site of CyaA , we transferred it into the context of the CD11c subunit of CR4 ( Figure 3A ) . Since mAbs recognizing CD11b and CD11c with equal efficacy are not available , a fluorescent YFP tag was added to the C-termini of CD11b , CD11c and CD11c614-682b molecules ( Figure 3A ) . This allowed cytometric selection of stably transfected CHO clones expressing similar quantities of the integrins on the cell surface ( Figure 3—figure supplement 1 ) . As shown in Figure 3B , the CD11c614-682b-YFP/CD18 chimera bound different concentrations of CyaA with substantially higher efficacy ( on average 24-fold higher ) than the intact CD11c-YFP/CD18 complex and the chimera exhibited only a slightly lower efficacy of toxin binding ( on average 1 . 7-fold lower ) than the CD11b-YFP/CD18 integrin itself ( Figure 3B ) . Consistently , the CHO-CD11c614-682b-YFP/CD18 cells were efficiently penetrated by CyaA and accumulated high levels of cAMP ( Figure 3C ) . The segment encompassing residues 614 to 682 of CD11b , hence , constitutes an autonomous CyaA-binding structure that enables toxin penetration into cells even when located in the context of the CD11c/CD18 integrin molecule . 10 . 7554/eLife . 10766 . 010Figure 3 . Residues 614 to 682 of CD11b confer CyaA binding to CD11c . ( A ) Schematic representation of the CD11c subunit ( blue ) with residues 614–682 replaced by the corresponding residues of CD11b ( green ) . All CD11 molecules were C-terminally fused to yellow fluorescent protein ( YFP ) to allow quantification of expression . ( B ) 2x105 CHO cells expressing integrin molecules were incubated with different concentrations of CyaA-biotin and the surface-bound toxin was labeled with streptavidin-PE . The cells were analyzed by flow cytometry and mean fluorescence intensities of CyaA binding were plotted against the concentrations of CyaA . Each point represents the mean value ± SD of two independent experiments performed in triplicate . CyaA binding to cells expressing CD11c614-682b-YFP/CD18 , or CD11b-YFP/CD18 was at all measured CyaA concentrations significantly higher in comparison with cells expressing intact CD11c-YFP/CD18 ( p<0 . 0001; ANOVA ) . RFI , relative fluorescence intensity . ( C ) 1x105 CHO cells expressing integrin molecules were incubated with various concentrations of CyaA and the amounts of accumulated cAMP were determined in cell lysates by ELISA . Each point represents the mean value ± SD of two independent experiments performed in duplicate . Significant differences between mean values of cAMP intoxication of cells expressing CD11c614-682b-YFP/CD18 , or CD11b-YFP/CD18 and intact CD11c-YFP/CD18 are shown ( * , p<0 . 05; ** , p<0 . 01; **** , p<0 . 0001; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 01010 . 7554/eLife . 10766 . 011Figure 3—figure supplement 1 . Expression of integrin variants fused with a fluorescent YFP protein on the surface of CHO cells and binding of CyaA to transfected cells . A CHO cell line stably expressing the intact CD18 subunit was transfected with plasmid constructs encoding CD11b-YFP , CD11c-YFP or CD11c614-682b-YFP and cells stably expressing the integrin molecules were selected using a cell sorter . The expression levels of CD11b-YFP/CD18 ( green ) , CD11c-YFP/CD18 ( blue ) and CD11c614-682b-YFP/CD18 chimera ( red ) on the surface of 2x105 cells were examined by flow cytometry for YFP ( left panels ) , or upon staining of cells with mAb recognizing CD11b ( MEM-174 , middle left panels ) , or CD11c ( 3 . 9 , middle right panels ) . The cells were also analyzed for capacity to bind CyaA: 2x105 transfected cells were incubated with 2 μg/ml of CyaA-biotin , the surface-bound toxin was labeled with streptavidin-PE and the cells were analyzed by flow cytometry ( right panels ) . CHO cells expressing no β2 integrin ( transfected with empty vectors ) were processed in parallel and used as negative control ( grey ) . Typical flow cytometry histograms from one representative binding experiment out of four performed are shown . RFI , relative fluorescence intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 011 To analyze the binding interaction of CyaA with CR3 in molecular detail , we employed a set of in silico approaches to predict the key residues involved . First , the known 3D structure of the homologous CR4 integrin ( Xie et al . , 2010 ) was used as a template for construction of a homology model of the 3D structure of CR3 . In parallel , the structure of the CR3 binding site located within residues 1166 to 1287 ( segment 1166-1287 ) of CyaA ( El-Azami-El-Idrissi et al . , 2003 ) was predicted ab initio , using I-TASSER ( Roy et al . , 2010 ) . Flexible side chain docking of CR3 and of the segment 1166-1287 of CyaA by the ClusPro server ( Comeau et al . , 2007 ) was next used to predict the most probable interacting residues of the two proteins ( Figure 4A , B ) . Second , the α subunits of all four β2 integrins were aligned by the Jotun Hein method ( Lasergene software , DNASTAR ) and residues that are present at a given position only in the 614-682 segment of CD11b were considered as potentially involved in CyaA binding . Finally , the theoretical isoelectric points of the interacting segments ( 3 . 76 for 1166-1287 segment of CyaA and 8 . 27 for 614-682 segment of CD11b ) were taken into account . The sum of these considerations suggested that the negatively charged residues of the 1166-1287 segment of CyaA might be involved in the interaction with positively charged residues within the 614-682 segment of CD11b . 10 . 7554/eLife . 10766 . 012Figure 4 . Electrostatic interaction of oppositely charged residues underlies CyaA binding to CR3 . ( A ) 3D structure of CR3 was modeled by homology onto the known 3D structure of CR4 . The structure of the CD11b binding site within the segment 1166-1287 of CyaA was predicted using I-TASSER . For clarity , only the CD11b subunit is shown . ( B ) To identify interacting residues , a flexible side chain docking of the segment 1166-1287 of CyaA to CR3 was performed using the ClusPro server . ( C ) Different concentrations of Dy647-labeled intact CyaA or its variants with point mutations in the CD11b binding site of the toxin were incubated with 2x105 CHO cells expressing intact CR3 and the cells were analyzed by flow cytometry . Mean fluorescence intensities of binding of intact CyaA or its variants were plotted against the toxin concentrations . Each point represents the mean value ± SD of four independent experiments . Binding of CyaA mutant variants to cells was at all measured concentrations significantly lower than binding of intact CyaA ( p<0 . 0001; ANOVA ) . RFI , relative fluorescence intensity . ( D ) Different concentrations of intact CyaA and its mutant variants were incubated with 1x105 CHO cells expressing intact CR3 and intracellular levels of cAMP were determined by ELISA . Each point represents the mean value ± SD of two independent experiments performed in triplicate . Intoxication of cells by CyaA mutant variants was at all measured concentrations significantly lower than intoxication of cells by intact CyaA ( p<0 . 0001; for CyaAE1232+D1234A at 20 ng/ml p<0 . 001; ANOVA ) . ( E ) 2x105 CHO cells expressing integrin molecules were incubated with 2 µg/ml of CyaA-biotin , the surface-bound toxin was labeled with streptavidin-PE and the cells were analyzed by flow cytometry . CyaA binding was expressed as percentage of toxin binding to CHO cells expressing the native form of CD11b/CD18 . Each bar represents the mean value with SD of two independent experiments performed in duplicate . Significantly reduced binding of CyaA to mutant integrins in comparison with intact CD11b/CD18 is indicated ( **** , p<0 . 0001; ANOVA ) . ( F ) 1x105 CHO cells expressing different integrin molecules were incubated with various concentrations of CyaA and the amounts of accumulated cAMP were determined in cell lysates by ELISA . Each point represents the mean value ± SD of six independent experiments . Intoxication of cells expressing CD11bR662A+R664A+R666A/CD18 or no β2 integrin was in concentrations ranging from 5 to 20 ng/ml of CyaA significantly lower than intoxication of cells expressing intact CD11b/CD18 ( p<0 . 0001; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 01210 . 7554/eLife . 10766 . 013Figure 4—figure supplement 1 . Expression of CD11b/CD18 and of its mutant variants on the surface of CHO cells and binding of CyaA to transfected cells . A CHO cell line stably expressing the intact CD18 subunit was transfected with plasmid constructs encoding intact CD11b or its mutant variants with the charged and hydrophilic residues within the segment 614-682 replaced with alanine residues . Cells stably expressing integrin molecules were selected using a cell sorter . The expression levels of CD11b/CD18 ( green ) and its mutant variants ( red ) on the cell surface were examined by flow cytometry after staining of 2x105 cells with the 2LPM19c mAb recognizing the I-domain of CD11b ( left panels ) , or the OKM1 mAb recognizing the mutagenized segment 614-682 of CD11b ( middle panels ) . The cells were also analyzed for capacity to bind CyaA: 2x105 transfected cells were incubated with 2 μg/ml of CyaA-biotin , the surface-bound toxin was labeled with streptavidin-PE and the cells were analyzed by flow cytometry ( right panels ) . CHO cells expressing no β2 integrin ( transfected with empty vectors ) were processed in parallel and used as negative control ( grey ) . Typical flow cytometry histograms from one representative binding experiment out of four performed are shown . RFI , relative fluorescence intensity . In contrast to other substitutions in the segment 614-682 of CD11b , the triple substitution E625A+N627A+R629A abolished binding of OKM1 , suggesting that the epitope recognized by OKM1 , or part of it , comprises these residues . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 01310 . 7554/eLife . 10766 . 014Figure 4—figure supplement 2 . Staining of CHO cells expressing intact CD11b/CD18 and of its two mutant variants with the four mAbs that block the binding of CyaA . 2x105 CHO cells stably expressing intact CD11b/CD18 , or its mutant variants exhibiting significantly reduced capacity to bind CyaA , were stained with the anti-CD11b mAbs that block CyaA binding . The cells were analyzed by flow cytometry and antibody binding was expressed as percentage of mAb binding to CHO cells expressing the native form of CD11b/CD18 . Each bar represents the mean value with SD of four independent experiments . Significant differences between mean values of mAb binding to cells expressing mutant integrin and cells expressing intact CD11b/CD18 are indicated ( **** , p< 0 . 0001; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 014 Therefore , the predicted negatively charged glutamate and aspartate residues of the 1166-1287 segment of CyaA were substituted by alanine . As shown in Figure 4C , D , the CyaAD1193A+D1194A+E1195A and CyaAE1232+D1234A constructs exhibited a strongly reduced capacity to bind and penetrate CR3-expessing CHO cells . Further , combinations of alanine substitutions were also introduced in place of charged and hydrophilic residues within the 614-682 segment of CD11b and six mutant CD11b molecules were stably expressed on the surface of CHO-CD18 cells ( Figure 4—figure supplement 1 ) . No statistically significant drop of CyaA binding ( Figure 4E ) and cAMP elevation ( Figure 4F ) was observed in the cells expressing CR3 variants with the combinations of substitutions R619A+K621A , E625A+N627A+R629A , K644A+K646A , or E650A+R652A within the 614-682 segment of CD11b , respectively . In contrast , alanine substitutions of the three positively charged arginine residues at positions 662 , 664 and 666 of CD11b almost completely abolished binding of CyaA to the CR3 mutant variant and a significant reduction of CyaA binding was also observed with cells expressing the CD11bK659+T661A/CD18 construct ( Figure 4E ) . As shown in Figure 4—figure supplement 2 , this loss or reduction of CyaA binding was not due to an alteration of the structure of the CyaA binding site within the mutant integrins , as these bound the CyaA-blocking mAbs MEM-174 , OKM1 and VIM12 to the same extent as the intact CR3 . The only exception was the CyaA-blocking mAb M1/70 , which exhibited reduced binding to both mutant integrins , similarly as CyaA . This indicates that the M1/70 mAb and CyaA may recognize the same epitope within the 614-682 segment of CD11b . In comparison to CHO cells expressing intact CR3 , CyaA produced importantly lower intracellular cAMP levels in cells expressing CD11bR662A+R664A+R666A/CD18 ( Figure 4F ) . Taken together , our mutagenesis experiments confirmed the computational predictions and identified the residues involved in CD11b recognition by CyaA . Since endogenous ligands preferentially bind the I-domain of activated CR3 , the unique location of the CyaA binding site prompted us to test whether CyaA differentiates between the inactive ( bent ) and active ( extended ) conformation of CR3 . For this experiment , a leukocyte-enriched fraction was isolated from fresh human blood and treated ( or not ) with phorbol 12-myristate 13-acetate ( PMA ) that is a known agonist activating CR3 ( Diamond and Springer , 1993 ) . The cells were subsequently stained with mAbs and exposed to different concentrations of CyaA , before being analyzed by flow cytometry . Monocytes were gated based on light-scatter characteristics and expression of CD14 and used for assessment of CyaA binding . As shown in Figure 5A , B , PMA-stimulated monocytes expressed the same amounts of CR3 as untreated monocytes ( detected with OKM1 mAb ) , but bound the toxin with a significantly lower efficiency . This was clearly due to conformational rearrangement of CR3 on PMA-activated monocytes , as the activated cells bound approximately two-fold more of the activation-reporting mAb MEM-148 ( Drbal et al . , 2001 ) than the non-activated monocytes ( Figure 5A ) . The monocytes pretreated with PMA then accumulated substantially lower intracellular levels of cAMP upon exposure to CyaA than the non-activated cells ( Figure 5C ) . Similarly , activation of CR3 by pretreatment of cells with Mn2+ ions ( Altieri , 1991; Li et al . , 2013 ) reduced toxin binding and cAMP accumulation in monocytes ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 10766 . 015Figure 5 . CyaA preferentially recognizes the inactive ( bent ) conformation of CR3 . ( A , B ) A leukocyte-enriched fraction prepared from fresh whole blood was treated without or with 100 nM PMA to activate CR3 . The cells were promptly stained with the OKM1 mAb recognizing both CR3 conformations , or with the MEM-148 mAb recognizing the extended integrin conformation ( A ) , or with different concentrations of Dy647-labeled CyaA ( B ) in a combination with anti-CD14 mAb . After 2 min , cells were analyzed by flow cytometry , monocytes were gated based on light-scatter characteristics and expression of CD14 and used for calculation of mAbs and CyaA binding . Binding of mAbs , or CyaA ( at the highest tested concentration ) to PMA-untreated cells was taken as 100% . Each value represents the mean with SD of three independent experiments performed in duplicate using three different donors . Significant differences between mean values of mAbs or CyaA binding to cells treated with buffer alone and cells treated with PMA are shown ( * , p<0 . 05; ** , p< 0 . 01; *** , p<0 . 001; Student’s t-test ) . ( C ) 1x105 human primary monocytes were pretreated without or with 100 nM PMA and incubated with indicated concentrations of CyaA . The amounts of accumulated cAMP were determined in cell lysates by ELISA . Each point represents the mean value ± SD of seven independent experiments performed in duplicate using cells of seven different donors . Significant differences between mean values of cAMP intoxication of monocytes incubated in the absence and in the presence of PMA are shown ( * , p<0 . 05; *** , p<0 . 001; **** , p<0 . 0001; Student’s t-test ) . ( D-G ) The bent and extended conformers of sCR3 were immobilized to a Bio-Rad ProteOn XPR36 GLC sensor chip and the MEM-48 mAb recognizing both conformations of sCR3 ( D ) , or the MEM-148 mAb recognizing the extended integrin conformation ( E ) were used as controls . To analyze the interaction between the bent ( F ) , or extended ( G ) conformation of sCR3 and the toxin , CyaA∆H was passed over the chip surface at concentrations of 20 , 40 , 80 , 160 and 320 µg/ml . The data were analyzed by global fitting of the response curves using a bivalent analyte model and calculated kinetic parameters are given in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 01510 . 7554/eLife . 10766 . 016Figure 5—figure supplement 1 . CyaA preferentially recognizes the inactive ( bent ) conformation of CR3 . ( A , B ) A leukocyte-enriched fraction prepared from fresh whole blood was treated without or with 1 mM Mn2+ ions to activate CR3 . The cells were promptly stained with the OKM1 mAb recognizing both CR3 conformations , or with the MEM-148 mAb recognizing the extended integrin conformation ( A ) , or with different concentrations of Dy647-labeled CyaA ( B ) in a combination with anti-CD14 mAb . After 2 min , cells were analyzed by flow cytometry , monocytes were gated based on light-scatter characteristics and expression of CD14 and used for calculation of mAbs and CyaA binding . Binding of mAbs , or CyaA ( at the highest tested concentration ) to Mn2+-untreated cells was taken as 100% . Each value represents the mean with SD of three independent experiments performed in duplicate using three different donors . Significant differences between mean values of mAbs or CyaA binding to cells treated with buffer alone and cells treated with Mn2+ ions are shown ( * , p<0 . 05; ** , p<0 . 01; Student´s t-test ) . ( C ) 1x105 primary human monocytes were pretreated without or with 1 mM Mn2+ to activate CR3 and then incubated with indicated concentrations of CyaA . The amounts of accumulated cAMP were determined in cell lysates by ELISA . Each point represents the mean value ± SD of four independent experiments ( three performed in duplicate ) using cells of four different donors . Significant differences between mean values of cAMP intoxication of monocytes incubated in the absence and in the presence of Mn2+ are shown ( * , p<0 . 05; ** , p<0 . 01; Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 01610 . 7554/eLife . 10766 . 017Figure 5—figure supplement 2 . Isolation and characterization of the bent and extended conformers of sCR3 . ( A ) Soluble ectodomain complex of CR3 secreted from CHO transfected cells was purified from cell media by affinity chromatography and the bent and extended conformers were separated by size exclusion chromatography . An overlay of chromatograms of re-analyzed samples containing either the extended ( red ) or the bent ( green ) conformer of sCR3 is shown . ( B , C ) The efficacy of separation was confirmed by transmission electron microscopy upon application of the extended ( B ) and bent ( C ) integrin conformers on glow-discharged carbon copper grids , followed by staining with uranyl formate . Inlet figures in red frames show zoomed views of the sCR3 conformers . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 017 To further analyze the impact of CR3 conformation on CyaA binding , a soluble secreted form of the CR3 ectodomain complex ( sCR3 ) was purified from supernatants of CHO cultures by immunoaffinity chromatography on MEM-174 mAb-Sepharose beads . The bent and extended conformers of sCR3 were resolved by size exclusion chromatography ( Figure 5—figure supplement 2A ) , their conformation was confirmed by electron microscopy ( Figure 5—figure supplement 2B , C ) and the two sCR3 conformers were immobilized onto a surface plasmon resonance ( SPR ) sensor chip . Coupling densities and random orientation of sCR3 on the chip was verified by passage of the control mAb MEM-48 that recognizes both integrin conformations ( Figure 5D ) . As shown in Figure 5E , the immobilization procedure preserved the conformation of the integrin molecules , since the integrin activation-reporting mAb MEM-148 bound only the immobilized extended sCR3 and not its bent form . Since aggregation and unspecific binding to the SPR chip surface of intact CyaA was observed at the required toxin concentrations , we used a CyaA∆H construct , which lacks the hydrophobic pore-forming domain and is soluble even at high concentrations ( Šebo and Ladant , 1993 ) . Real-time SPR measurements of CyaA∆H interaction with the immobilized sCR3 conformers exhibited typical association and dissociation curves ( Figure 5F , G ) . Analysis of the binding curves by a bivalent analyte model then revealed that the association rate was higher and the dissociation rate was lower for the complex of the bent sCR3 conformer with CyaA∆H ( Kd = 2 . 1x10-7 M ) than for the complex of the extended sCR3 conformer with CyaA∆H ( Kd = 6 . 4x10-7 M ) ( Table 1 ) . This demonstrates that the bent CR3 conformer bound the toxin with a higher affinity than the extended conformer , confirming the data obtained by binding experiments on PMA- or Mn2+-activated primary monocytes . 10 . 7554/eLife . 10766 . 018Table 1 . Kinetic parameters of CyaA∆H binding to sCR3 calculated by a bivalent analyte model . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 018LigandaAnalyteka1 [×103 M-1s-1]kd1 [×10-3 s-1]ka2 [×10-4 RU-1s-1]kd2 [×10-4 s-1]Bent sCR3CyaA∆H5 . 8 ± 0 . 71 . 2 ± 0 . 21 . 3 ± 0 . 23 . 8 ± 0 . 5Extended sCR32 . 8 ± 0 . 21 . 8 ± 0 . 42 . 7 ± 0 . 35 . 1 ± 0 . 6aTwo independent SPR binding experiments were performed in duplicate and the differences between mean values of CyaA∆H binding to the bent and extended conformation of sCR3 were statistically significant ( p<0 . 01; Student’s t-test ) . Binding of endogenous ligands to the I-domain of CR3 is known to trigger activation of Syk . This initiates downstream signaling and provokes multiple cellular activation responses , including complement-mediated opsonophagocytosis of bacteria ( Shi et al . , 2006 ) . We hence examined if the I-domain-independent interaction of CyaA with CR3 activates Syk in human monocytic cells . THP-1 monocytes , which bind CyaA in a CR3-dependent and saturable manner ( Figure 6—figure supplement 1 ) , were exposed to different concentrations of CyaA over a range of incubation times . Phosphorylated Syk was next immunoprecipitated from cell lysates with anti-phosphotyrosine mAb and was detected by anti-Syk mAb in immunoblots . As a positive control , the CR3-Syk signaling pathway was activated by iC3b-opsonized zymosan particles ( Shi et al . , 2006 ) . As shown in Figure 6A , Syk phosphorylation on tyrosine residues was observed after 30 min of incubation of THP-1 cells with iC3b-opsonized zymosan . No phosphorylation of Syk was , however , detected following exposure of THP-1 cells for up to 60 min to 30 ng/ml CyaA , a concentration close to physiological toxin levels ( Eby et al . , 2013 ) . Similarly , Syk activation was not observed when THP-1 cells were incubated with the enzymatically inactive CyaA-AC- toxoid ( Figure 6B ) . Neither was Syk phosphorylation triggered by THP-1 cell exposure to concentrations of up to 3 µg/ml of active CyaA toxin ( Figure 6C ) , or of its CyaA-AC- toxoid ( Figure 6D ) for 15 or 30 min , respectively . Hence , in contrast to iC3b , which binds through the I-domain of CR3 and triggers activation of Syk ( Shi et al . , 2006 ) , the I-domain-independent mode of CyaA binding did not trigger any CR3 signaling through Syk pathway activation . 10 . 7554/eLife . 10766 . 019Figure 6 . CyaA binding to CR3 does not trigger Syk activation . ( A , B ) 3x106 THP-1 cells were incubated with 30 ng/ml of CyaA ( A ) , or CyaA-AC- ( B ) for indicated times . ( C , D ) 3x106 THP-1 cells were incubated with different indicated concentrations of CyaA for 15 min ( C ) , or CyaA-AC- for 30 min ( D ) . ( A-D ) Treated cells were lysed and cell lysates were immunoprecipitated ( IP ) with anti-phosphotyrosine ( anti-PY ) mAb . Syk-P immunoprecipitated from whole cell lysates was detected by immunoblotting ( IB ) with anti-Syk mAb and normalized to total Syk detected in whole cell lysates . Cells treated with iC3b-opsonized zymosan were taken as a positive control and the cells treated with buffer , or unopsonized zymosan were taken as negative controls . Each bar represents the mean value with SD of three independent experiments . In comparison to buffer-treated cells , a significant increase of Syk activation was observed only in cells treated with iC3b-opsonized zymosan ( **** , p<0 . 0001; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 01910 . 7554/eLife . 10766 . 020Figure 6—figure supplement 1 . CyaA binds efficiently and specifically the THP-1 cells . ( A ) 2x105 THP-1 cells were treated with the anti-CD11b OKM1 mAb , or with an isotype-matched mouse IgG2b control antibody and the cells were analyzed by flow cytometry . Each bar represents the mean value ± SD of two independent experiments performed in duplicate . Significant difference between mean values of OKM1 and the control mAb binding to THP-1 cells is indicated ( **** , P < 0 . 0001; Student’s t-test ) . ( B ) 2x105 THP-1 cells were preincubated without or with 25 μg/ml of the anti-CD11b OKM1 mAb for 30 min and subsequently treated with different concentrations of CyaA-biotin . The surface-bound toxin was labeled with streptavidin-PE and the cells were analyzed by flow cytometry . Mean fluorescence intensities of CyaA binding were plotted against toxin concentrations . Each point represents the mean value ± SD of two independent experiments performed in duplicate . Binding of CyaA to cells preincubated with the OKM1 mAb was at all measured concentrations significantly lower than binding of CyaA to cells pre-treated with buffer alone ( P < 0 . 01; Student’s t-test ) . RFI , relative fluorescence intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 020 We further investigated whether intracellular cAMP signaling produced by the cell-invading CyaA toxin would interfere with the outside-in signaling of CR3 . As shown in Figure 7A , Syk was efficiently activated by iC3b-opsonized zymosan in THP-1 cells that were pre-treated with buffer , or with 300 ng/ml of the enzymatically inactive CyaA-AC- toxoid . However , iC3b-coated particles failed to elicit any Syk phosphorylation in THP-1 cells that were pre-incubated with 300 ng/ml of CyaA ( Figure 7A ) . Control experiments demonstrated that CyaA and CyaA-AC- bound THP-1 cells with the same efficacy ( Figure 7—figure supplement 1 ) and that toxin binding did not alter the level of CR3 expression on the cell surface ( Figure 7—figure supplement 2 ) . Nevertheless , a partial reduction of binding of the iC3b-coated particles to THP-1 cells preincubated with CyaA was observed , as compared to cells preincubated with buffer alone or with the CyaA-AC- toxoid ( Figure 7—figure supplement 3 ) . Hence , the cAMP signaling action of CyaA concurrently prevented Syk activation by CR3-bound iC3b as well as it provoked a decrease of the iC3b opsonin binding . 10 . 7554/eLife . 10766 . 021Figure 7 . CyaA-produced cAMP blocks opsonin-induced Syk activation . ( A ) 3x106 THP-1 cells were incubated with 300 ng/ml of CyaA , CyaA-AC- or buffer alone for 15 min and subsequently incubated with iC3b-opsonized zymosan for 30 min to activate Syk . Cells treated with buffer followed by unopsonized zymosan were used as a control . ( B ) 3x106 THP-1 cells were pre-incubated with iC3b-opsonized zymosan for 15 min to activate Syk and subsequently incubated with 300 ng/ml of CyaA , CyaA-AC- or buffer alone for 30 min . Cells treated with unopsonized zymosan followed by buffer were taken as negative control . Processing of cells and detection of Syk were performed as in the legend to Figure 6 . Each bar represents the mean value with SD of three independent experiments . Significant differences between mean values of Syk activation in cells treated with iC3b-opsonized zymosan in the absence or presence of CyaA or CyaA-AC- are shown ( ** , p<0 . 01; *** , p<0 . 001; **** , p<0 . 0001; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 02110 . 7554/eLife . 10766 . 022Figure 7—figure supplement 1 . CyaA and its enzymatically inactive variant CyaA-AC- bind THP-1 cells with the same efficacy . ( A ) 2x105 THP-1 cells were treated with buffer alone , or with 300 ng/ml of CyaA or CyaA-AC- and then stained with CyaA-specific mAb 3D1 , followed by Cy5-labeled secondary antibody ( GAM-Cy5 ) . Cells were analyzed by flow cytometry and a typical overlay histogram from one representative binding experiment out of two independent experiments performed in duplicate is shown . ( B ) The CyaA and CyaA-AC- binding data were deduced from the mean fluorescence intensities of the flow cytometry experiments and the mean values with SD were expressed as fold change with respect to control cells treated with buffer alone . The mean values of CyaA and CyaA-AC- binding to THP-1 cells did not significantly differ ( ns; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 02210 . 7554/eLife . 10766 . 023Figure 7—figure supplement 2 . CyaA and CyaA-AC- do not alter the level of CR3 expression on the surface of THP-1 cells . ( A , B ) 2x105 THP-1 cells were treated with 300 ng/ml of CyaA ( green ) , CyaA-AC- ( red ) or buffer ( blue ) for 15 min at 37°C , followed by staining with anti-CD11b mAbs ICRF 44 ( A ) or OKM1 ( B ) . The cells were analyzed by flow cytometry and a typical overlay histogram from one representative binding experiment out of three independent experiments performed is shown for each mAb . THP-1 cells treated with buffer alone are shown as negative control ( grey ) . RFI , relative fluorescence intensity . ( C , D ) The mAbs binding data were deduced from the mean fluorescence intensities of the flow cytometry experiments and the mean values with SD were expressed as fold change with respect to control unstained cells . The mean values of ICRF 44 ( C ) or OKM1 ( D ) mAb binding to THP-1 cells treated with CyaA or CyaA-AC- did not significantly differ ( ns; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 02310 . 7554/eLife . 10766 . 024Figure 7—figure supplement 3 . CyaA reduces binding of iC3b-zymosan to THP-1 monocytes . ( A ) 2x105 THP-1 cells were treated with 300 ng/ml of CyaA ( green ) , CyaA-AC- ( red ) or buffer ( blue ) for 15 min at 37°C , followed by iC3b-zymosan for 30 min at 37°C . Cells were analyzed by flow cytometry and a typical overlay histogram from one representative binding experiment out of two independent experiments performed in duplicate is shown . Cells treated with buffer alone are shown in grey . ( B ) The iC3b-zymosan binding data were deduced from the mean fluorescence intensities of the flow cytometry experiments and the mean values with SD were expressed as percentage of iC3b-zymosan binding to cells in the absence of CyaA or CyaA-AC- . Significant differences between mean values of iC3b-zymosan binding to buffer-treated cells and cells treated with CyaA or CyaA-AC- are indicated ( **** , p<0 . 0001; ns , not significant; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 02410 . 7554/eLife . 10766 . 025Figure 7—figure supplement 4 . Preincubation of THP-1 monocytes with iC3b-opsonized zymosan does not reduce binding of CyaA or of CyaA-AC- to cells . ( A , B ) 2x105 THP-1 cells were preincubated with buffer alone or iC3b-opsonized zymosan for 30 min at 37°C and subsequently treated with biotinylated CyaA ( A ) or CyaA-AC- ( B ) for 10 min at 37°C . The surface-bound toxin was labeled with streptavidin-PE and the cells were analyzed by flow cytometry . A typical overlay histogram from one representative binding experiment out of three independent experiments performed is shown . RFI , relative fluorescence intensity . ( C-D ) The CyaA and CyaA-AC- binding data were deduced from the mean fluorescence intensities of the flow cytometry experiments and the mean values with SD were expressed as fold change with respect to cells treated with buffer alone . Preincubation of THP-1 monocytes with iC3b-opsonized zymosan did not significantly change binding of CyaA ( C ) or CyaA-AC- ( D ) to cells ( ns; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 02510 . 7554/eLife . 10766 . 026Figure 7—figure supplement 5 . CyaA and its enzymatically inactive variant CyaA-AC- do not alter viability of THP-1 cells over the duration of the signaling experiments . ( A ) THP-1 cells were treated with buffer alone or with 300 ng/ml of CyaA or CyaA-AC- for 45 min at 37°C ( i . e . at the same conditions as during the signaling experiments ) . After washing , cells were stained with FITC-labeled Annexin V for 20 min at 20°C and analyzed by flow cytometry in the presence of 1 µg/ml of propidium iodide ( PI ) . A typical flow cytometry dot plot from one representative binding experiment out of two independent experiments performed in triplicate is given for each sample . Cell death/apoptosis was evaluated by quadrant gating: Q1 , necrotic cells; Q2 , late apoptotic cells; Q3 , early apoptotic cells; Q4 , live cells . ( B ) The percentages of cells were calculated for each quadrant from the flow cytometry experiments and the mean values with SD were plotted as a bar chart . The mean percentages of live cells treated with CyaA or CyaA-AC- did not significantly differ from buffer-treated cells ( ns; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 02610 . 7554/eLife . 10766 . 027Figure 7—figure supplement 6 . CyaA-produced cAMP blocks opsonin-induced Syk activation . ( A ) 3x106 primary human monocytes were incubated with 300 ng/ml of CyaA , CyaA-AC- or buffer alone for 15 min and were subsequently incubated with iC3b-opsonized zymosan for 30 min to activate Syk . Cells treated with buffer followed by unopsonized zymosan were used as a control . ( B ) 3x106 primary human monocytes were pre-incubated with iC3b-opsonized zymosan for 15 min to activate Syk and subsequently incubated with 300 ng/ml of CyaA , CyaA-AC- or buffer alone for 30 min . Cells treated with unopsonized zymosan followed by buffer were taken as negative control . Processing of cells and detection of Syk were performed as in the legend to Figure 6 . Each bar represents the mean value with SD of three independent experiments . Significant differences between mean values of Syk activation in cells treated with iC3b-opsonized zymosan in the absence or presence of CyaA or CyaA-AC- are shown ( ** , p<0 . 01; *** , p<0 . 001; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 027 Importantly , when Syk was pre-activated by incubation of THP-1 cells with iC3b-opsonized zymosan , the subsequent addition of CyaA , but not of CyaA-AC- , provoked a significant drop of tyrosine phosphorylation of Syk ( Figure 7B ) . Control experiments showed that preincubation of THP-1 cells with iC3b-opsonized zymosan did not reduce binding of CyaA or CyaA-AC- , respectively ( Figure 7—figure supplement 4 ) . Moreover , the impairment of Syk phosphorylation upon CyaA treatment was not due to decreased cell viability ( Figure 7—figure supplement 5 ) . It can , hence , be concluded that Syk inactivation was due to signaling of the cAMP produced in cells by the penetrating CyaA toxin . Similar results were obtained when the same experiments were performed with primary human monocytes isolated from leukopacks of healthy donors ( Figure 7—figure supplement 6 ) . Taken together , our data show that the cAMP signaling of CyaA decreases the capacity of CR3 to bind the iC3b opsonin , prevents activation of Syk by iC3b-engaged CR3 and inactivates also Syk that has already been activated by iC3b/CR3 .
We show here that CyaA hijacks the complement receptor 3 of phagocytes by subversively binding to its bent , inactive conformation by binding to a unique site located outside of the I-domain . This novel type of CR3 interaction then enables rapid toxin penetration into phagocytic cells and elevation of cytosolic cAMP concentration , thus ablating the pro-phagocytic signaling of CR3 via Syk ( Figure 8 ) . This unprecedented mode of binding and action differentiates CyaA from all other known ligands of the I-domain-containing integrins . 10 . 7554/eLife . 10766 . 028Figure 8 . CyaA acts as a unique ligand of the I-domain-containing integrin CR3 . CyaA secreted by B . pertussis binds CR3 outside of its I-domain , using a unique site that encompasses the C-terminal end of the last repeat of the β-propeller domain and the N-terminal portion of the thigh domain of CD11b . CyaA preferentially binds the integrin in a non-activated ( bent , low-affinity ) conformation and engagement of CR3 by CyaA does not trigger Syk activation in monocytes . Moreover , CyaA-catalyzed elevation of cAMP effectively blocks the iC3b opsonin-elicited activation of CR3-Syk signaling in monocytes . It remains to be elucidated how CyaA-produced cAMP signaling suppresses Syk activity . Binding outside of the I-domain in an activation-independent mode thus enables the toxin to hijack CR3 and block its signaling , thereby enabling B . pertussis to evade CR3-mediated phagocytosis . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 028 While the I-domain is essential for binding of ligands to I-domain-containing integrins , the CyaA toxin binds CR3 independently of the presence of the I-domain . This possibility has already been suggested by Guermonprez and co-workers , who observed that in contrast to I-domain-dependent CD11b ligands , CyaA binding does not require the presence of Mg2+ ions ( Guermonprez et al . , 2001 ) . Indeed , we show that CyaA primarily recognizes the CD11b segment consisting of residues 614 to 682 . This encompasses the C-terminal end of the last β-propeller repeat and the N-terminal portion of the thigh domain . Besides interacting with the I-domain , certain ligands appear to interact to some extent also with the β-propeller domain of alpha subunits of β2 integrins ( Li and Zhang , 2003; Yalamanchili et al . , 2000 ) . The β-propeller domain consist of seven β-sheet repeats ( Figure 1A ) that are arranged in a torus , or propeller , with each β-sheet representing one blade of the propeller ( Oxvig and Springer , 1998; Springer , 1997; Xie et al . , 2010 ) . Its fold is most closely related to the β subunit of the trimeric G-protein ( Springer , 1997 ) and proper folding of the β-propeller requires association of the α subunit with the β2 subunit ( Huang and Springer , 1997; Lu et al . , 1998 ) . The β-propeller is followed by the thigh domain , exhibiting a C2-set immunoglobulin fold ( Chothia and Jones , 1997 ) , and by a genu region ( Xie et al . , 2004; 2010 ) . The structural rearrangement at the thigh/genu interface is involved in the activation-dependent extension of the integrin α subunit ( Xie et al . , 2004 ) . It was previously suggested that the β-propeller/thigh region segment of CD11b ( residues 599 to 679 ) , comprising the principal CyaA binding site , can fold also in the absence of the CD18 subunit , since it was recognized by the OKM1 mAb when the CD11b subunit was expressed alone on COS cells ( Lu et al . , 1998 ) . Indeed , we observed here that OKM1 recognized CD11b that was expressed alone on CHO-CD11b cells ( Figure 1—figure supplement 3A ) . However , CyaA failed to bind any significantly to cells expressing the isolated CD11b subunit on their surface ( Figure 1—figure supplement 3C ) . In contrast to the OKM1 epitope , hence , the CyaA-binding site within the segment 614-682 of CD11b may acquire the correct structure ( fold ) only when the CD11b subunit is in complex with CD18 . Alternatively , an interaction of CyaA with the CD18 subunit may also be involved in CyaA binding to CR3 , even though CD18 alone was unable to bind CyaA on its own ( Figure 1—figure supplement 3C ) . In addition , recognition of additional segments of CD11b appears to contribute to the full capacity of CR3 to bind CyaA . Engagement of these segments likely facilitates or stabilizes the higher-affinity interaction of CyaA with the principal binding site ( residues 614 to 682 ) . Toxin interactions at several sites on CD11b would then synergize in bringing about the highly selective interaction of CyaA with the CD11b subunit of CR3 . Indeed , we have previously described that an initial low-affinity interaction of CyaA with N-linked glycan chains of the C-terminal part of the CD11b subunit is a critical prerequisite for toxin binding to CR3 ( Hasan et al . , 2015; Morova et al . , 2008 ) . A low-affinity interaction with abundant N-linked glycans on the integrin molecule most likely serves as the initial step of toxin pre-concentration from solution into the two-dimensional space of the cell surface . This increase of local CyaA concentration would then increase the probability of proper positioning of the toxin molecule for its subsequent high affinity interaction with specific amino acid residues of the CR3 glycoprotein . This two-step mode of CyaA-CR3 interaction is supported by the SPR binding curves which fit well using the bivalent analyte model , describing a two-step association process . Compared to a mechanism of direct single step binding from solution , a sequential binding mechanism would , indeed , lower the affinity constraints on interaction of CyaA with the ectodomain of CR3 . The stability of the CyaA-CR3 complex may then be further increased by an irreversible insertion of the hydrophobic domain and of the acyl chains of CyaA into the cellular membrane , which most likely accounts for the observed high apparent affinity of toxin binding to CR3-expressing cells ( Guermonprez et al . , 2001 ) . It was recently proposed that CyaA can specifically bind yet another β2 integrin on leukocytes , the LFA-1 ( CD11a/CD18 ) complex ( Paccani et al . , 2011 ) . However , this is at odds with the initial observation of Guermonprez and co-workers that CyaA does selectively and with high affinity bind the dendritic cells and macrophages expressing CD11b/CD18 , whereas B and T cell lines expressing exclusively the LFA-1 complex were recognized with substantially lower efficacy , if at all ( Guermonprez et al . , 2001 ) . To rule out that this difference in CyaA binding was due to unequal expression levels of the integrins on various cell types , we added a fluorescent YFP tag to the C-terminus of the CD11a subunit and generated a cell line ( CHO-CD11a-YFP/CD18 ) expressing the same quantities of the integrin molecules on the cell surface as found on the CR3-expressing cell line ( CHO-CD11b-YFP/CD18 ) ( Figure 9—figure supplement 1 ) . As shown in Figure 9A , over a range of CyaA concentrations , the CHO-CD11a-YFP/CD18 cells bound about two orders of magnitude less CyaA toxin than the CD11b-YFP/CD18-expressing cells . Moreover , the difference in binding of CyaA to CD11a-YFP/CD18-expressing cells , CD11c-YFP/CD18-expressing cells , or the mock-transfected cells expressing no β2 integrin at all , was not statistically significant ( Figure 9A ) . To exclude the concern that the residual binding of CyaA to CD11a-YFP/CD18 could be due to the presence of the YFP tag , we generated a cell line expressing comparable amounts of intact CD11a/CD18 ( Figure 9—figure supplement 2A ) and demonstrated that CyaA binds these cells as poorly as cells expressing CD11a-YFP/CD18 with the YFP tag ( Figure 9—figure supplement 2B ) . In addition , we purified the intact LFA-1 integrin from human peripheral blood mononuclear cells by immunoaffinity chromatography on MEM-25 mAb-Sepharose beads and immobilized it onto an SPR sensor chip . Real-time SPR measurements revealed no CyaA∆H binding to the intact LFA-1 ( Figure 9—figure supplement 3; cf . CyaA∆H binding to CR3 in Figure 5F , G ) . Finally , cells expressing CD11a-YFP/CD18 were intoxicated by CyaA to equally low cAMP levels as the control cells expressing CD11c-YFP/CD18 , or lacking any β2 integrin , while cAMP intoxication of cells expressing CD11b-YFP/CD18 was over two orders of magnitude higher than that of the control cells ( Figure 9B ) . All these data clearly demonstrate that CyaA binds and intoxicates cells expressing LFA-1 with equally low efficacy as cells expressing CR4 , or lacking any β2 integrin at all . 10 . 7554/eLife . 10766 . 029Figure 9 . CyaA binds and intoxicates cells expressing LFA-1 with equally low efficacy as cells expressing CR4 , or lacking any β2 integrin at all . ( A ) 2x105 stably transfected CHO cells expressing CD11a-YFP/CD18 , CD11b-YFP/CD18 , CD11c-YFP/CD18 , or no β2 integrin were incubated with different concentrations of Dy647-labeled intact CyaA and analyzed by flow cytometry . Mean fluorescence intensities of CyaA binding were plotted against the concentrations of CyaA . RFI , relative fluorescence intensity . ( B ) 1x105 CHO cells expressing integrin molecules were incubated at indicated concentrations of CyaA and the amounts of accumulated cAMP were determined in cell lysates by ELISA . ( A and B ) Each point represents the mean value ± SD of three independent experiments performed in duplicate . CyaA binding to or cAMP intoxication of cells expressing CD11a-YFP/CD18 , CD11c-YFP/CD18 , or no β2 integrin was at all toxin concentrations significantly lower than toxin binding to or cAMP intoxication of cells expressing intact CD11b-YFP/CD18 ( p<0 . 0001; ANOVA ) . However , CyaA binding to or cAMP intoxication of cells expressing CD11a-YFP/CD18 was at all measured toxin concentrations found to be statistically the same as toxin binding to or cAMP intoxication of cells expressing CD11c-YFP/CD18 or no β2 integrin at all ( P > 0 . 1; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 02910 . 7554/eLife . 10766 . 030Figure 9—figure supplement 1 . Expression of integrin variants fused with a fluorescent YFP protein on the surface of CHO cells . A CHO cell line stably expressing the intact CD18 subunit was transfected with plasmid constructs encoding CD11b-YFP or CD11a-YFP and cells stably expressing the integrin molecules were selected using a cell sorter . The expression levels of CD11b-YFP/CD18 ( green ) and CD11a-YFP/CD18 ( violet ) on the surface of 2x105 cells were examined by flow cytometry for YFP ( left panels ) , or upon staining of cells with a mAb recognizing CD11b ( M1/70 , middle panels ) , or a mAb recognizing CD11a ( MEM-25 , right panels ) . CHO cells expressing no β2 integrin ( transfected with empty vectors ) were used as negative control ( grey ) . Typical flow cytometry histograms from one representative binding experiment out of four performed are shown . RFI , relative fluorescence intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 03010 . 7554/eLife . 10766 . 031Figure 9—figure supplement 2 . Residual binding of CyaA to CD11a-YFP/CD18 is not due to the presence of the YFP tag . ( A ) A CHO cell line stably expressing the intact CD18 subunit was transfected with a plasmid construct encoding CD11a and cells stably expressing CD11a/CD18 were selected using a cell sorter . The expression levels of CD11a/CD18 ( red ) on the surface of 2x105 cells were examined by flow cytometry upon staining of cells with the anti-CD11a mAb MEM-25 ( left panels ) and compared with expression levels of CD11a-YFP/CD18 ( violet ) . Both CD11a/CD18- and CD11a-YFP/CD18-epressing cells were also examined by flow cytometry for YFP ( right panels ) . CHO cells expressing no β2 integrin ( transfected with empty vectors ) were used as negative control ( grey ) . Typical flow cytometry histograms from one representative binding experiment out of four performed are shown . RFI , relative fluorescence intensity . ( B ) 2x105 stably transfected CHO cells expressing CD11a/CD18 , CD11a-YFP/CD18 , CD11b/CD18 , or no β2 integrin were incubated with different concentrations of Dy647-labeled intact CyaA and analyzed by flow cytometry . Mean fluorescence intensities of CyaA binding were plotted against the concentrations of CyaA . RFI , relative fluorescence intensity . Each point represents the mean value ± SD of two independent experiments performed in duplicate . CyaA binding to cells expressing CD11a/CD18 , CD11a-YFP/CD18 or no β2 integrin was at all toxin concentrations significantly lower than toxin binding to cells expressing intact CD11b/CD18 ( p<0 . 0001; ANOVA ) . In contrast , CyaA binding to cells expressing CD11a/CD18 at all measured toxin concentrations was found to be statistically the same as toxin binding to cells expressing CD11a-YFP/CD18 or no β2 integrin at all ( P > 0 . 1; ANOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 03110 . 7554/eLife . 10766 . 032Figure 9—figure supplement 3 . CyaA does not interact with the intact native LFA-1 integrin . ( A ) LFA-1 was purified from human peripheral blood mononuclear cells by immunoaffinity chromatography and immobilized onto an SPR sensor chip . To analyze the interaction between LFA-1 and the toxin , CyaA∆H was passed over the chip surface at concentrations of 20 , 40 , 80 , 160 and 320 µg/ml . ( B ) The MEM-48 mAb recognizing the CD18 subunit of LFA-1 was used as positive control . DOI: http://dx . doi . org/10 . 7554/eLife . 10766 . 032 Our findings have numerous implications for understanding of the potent immunosubversive action of CyaA on phagocytes and of its central role in B . pertussis virulence and escape from innate immunity control . Despite the fact that CyaA was repeatedly reported to subvert FcR- and CR3-mediated opsonophagocytosis ( Confer and Eaton , 1982; Kamanova et al . , 2008; Lenz et al . , 2000 ) , little attention has been paid to the effector molecules and pathways targeted by CyaA-generated cAMP signaling . The Syk tyrosine kinase pathway , pinpointed here as the target of CyaA action , plays a central role in both FcR- and CR3-mediated phagocytosis , as it is activated immediately upon binding of immunoglobulin or iC3b opsonins to the respective FcR and CR3 receptors ( Crowley et al . , 1997; Kiefer et al . , 1998; Shi et al . , 2006 ) . We show that CyaA bypasses the Syk-activating interaction with CR3 through a subversive preferential recognition of its bent , resting conformation and by engaging the CR3 structures outside of the I-domain of CD11b . In contrast to iC3b opsonin , binding of CyaA , or of its non-enzymatic CyaA-AC- toxoid , did not trigger any CR3-mediated activation of Syk . Strikingly , CyaA-catalyzed elevation of cAMP in phagocytes effectively blocked subsequent activation of Syk through iC3b binding to CR3 . Most importantly , when Syk was pre-activated by incubation of phagocytes with iC3b , the subsequent CR3-dependent invasion of CyaA into cells enabled cAMP signaling-mediated inactivation of Syk ( Figure 8 ) . It remains to be elucidated how CyaA-produced cAMP signaling suppresses Syk activity . Two plausible working hypotheses are , nevertheless , worth mentioning in this context . The first would be based on the observation that increased cAMP levels activate the inhibitory C-terminal Src kinase ( Csk ) via the cAMP-dependent protein kinase PKA ( Vang et al . , 2001 ) . Csk may then directly inactivate the Src kinases by phosphorylation ( Vang et al . , 2001 ) . Inactive Src kinases would then be unable to accomplish the double phosphorylation of the ITAM-containing adaptor , which is a prerequisite for recruitment and activation of Syk ( Jakus et al . , 2007; Mócsai et al . , 2006; 2006; 2010; Schymeinsky et al . , 2007 ) . A second plausible mechanism would be based on our recent observation that CyaA/cAMP-mediated signaling triggers activation of the ubiquitously expressed SH2-containing non-receptor protein tyrosine phosphatase SHP-1 ( PTPN6 ) ( Cerny et al . , 2015 ) . This phosphatase has been previously implicated in regulation of tyrosine phosphorylation of Syk and in blocking of FcγR–mediated phagocytosis ( Huang , 2003; Kant et al . , 2002 ) . Experiments aimed at deciphering the mechanism by which cAMP signaling inactivates Syk are ongoing . CR3-mediated phagocytosis involves active RhoA GTP-ase , which acts downstream of Syk ( Caron and Hall , 1998 ) . We have previously shown that cAMP signaling of CyaA rapidly decreases RhoA activity and induces dephosphorylation of the actin filament-severing protein cofilin ( Kamanova et al . , 2008 ) . CyaA thereby provokes massive actin cytoskeleton rearrangements and unproductive membrane ruffling of macrophage cells ( Kamanova et al . , 2008 ) . We therefore speculate that this inhibitory intervention of the toxin at the two different signaling levels of Syk and RhoA , respectively , synergizes in bringing about a complete and rapid inhibition of CR3-mediated phagocytosis and complement-mediated opsonophagocytic killing of bacteria . CyaA would thereby subvert the central mechanism of innate defense against B . pertussis in naive hosts devoid of pathogen-specific antibodies , thus promoting bacterial colonization and virulence . Besides FcR and CR3-mediated phagocytosis , Syk activity regulates a number of additional processes involved in the control of bacterial infections by the immune system . These include innate immunity mechanisms such as phagocyte chemotaxis , generation of reactive oxygen intermediates , neutrophil extracellular trap formation and degranulation of phagocytes ( Forsberg et al . , 2001; Gevrey et al . , 2005; Mócsai et al . , 2002; Van Ziffle and Lowell , 2009; Willeke , 2003 ) . Moreover , Syk and its homologue , the ζ-chain-associated protein kinase of 70 kDa ( ZAP-70 ) , also play a crucial role in B- and T-cell receptor signaling during induction of adaptive immune responses ( Turner et al . , 2000 ) . Indeed , chemotactic activity and oxidative burst of phagocytes , as well as the Th1/Th2 adaptive immune response balance , were both previously shown to be sensitive to the cAMP-elevating activity of CyaA ( Confer and Eaton , 1982; Friedman et al . , 1987; Pearson et al . , 1987; Rossi Paccani et al . , 2009; Weingart et al . , 2000 ) . This suggests that disruption of Syk/ZAP70 signaling by CyaA impairs a broad range of processes that play a crucial role in host defense against infection . Our demonstration of targeting and efficient blocking of Syk activation by CyaA-produced cAMP thus provides a novel mechanistic insight into the prominent role played by CyaA in mediating immune evasion of Bordetella species pathogenic to mammals . This underpins the crucial role played by the CyaA toxin in the early stages of host colonization by B . pertussis ( Goodwin and Weiss , 1990; Khelef et al . , 1992 ) . It further indicates that vaccine-induced neutralizing antibody response against CyaA is likely to be importantly contributing to protection against infection and colonization by B . pertussis . Absence of the CyaA antigen from current acellular pertussis ( aP ) vaccines is thus of concern ( Sebo et al . , 2014 ) . It may represent one of the factors contributing to recent resurgence of pertussis in the most developed countries that use the aP vaccine ( Cherry , 2012; Misegades et al . , 2012; Octavia et al . , 2012; Tartof et al . , 2013; Witt et al . , 2012; 2013 ) .
Monoclonal antibodies ( mAbs ) 2LPM19c ( mouse IgG1 ) and 44 ( mouse IgG1 ) , specific for human CD11b , were obtained from Santa Cruz Biotechnology , Santa Cruz , CA . MAb M1/70 ( rat IgG2b ) specific for murine and human CD11b was purchased from BD Pharmingen . MAb ICRF44 ( mouse IgG1 ) specific for human CD11b was obtained from BioLegend , San Diego , CA . MAb VIM12 ( mouse IgG1 ) specific for human CD11b was obtained from Caltag Laboratories , Burlingame , CA . MAb 3 . 9 ( mouse IgG1 ) specific for human CD11c was obtained from Ancell Corporation , Bayport , MN . MAbs specific for human CD11b ( MEM-174 , mouse IgG2a ) and CD18 ( MEM-48 , mouse IgG1 ) were a kind gift of V . Horejsi ( Czech Academy of Sciences , Prague , Czech Republic ) . MAb OKM1 ( mouse IgG2b ) specific for human CD11b was purified from the OKM1 hybridoma obtained from the European Collection of Cell Cultures , Porton Down , UK . Anti-human CD11a ( MEM-25 , mouse IgG1 ) , anti-human CD18 ( MEM-148 , mouse IgG1 ) , anti-phosphotyrosine ( P-Tyr-01 , mouse IgG1 ) and anti-Syk ( SYK-01 , mouse IgG1 ) mAbs were purchased from Exbio , Vestec , Czech Republic . Anti-human CD14 ( TÜK4 , mouse IgG2a ) mAb was obtained from Dako , Glostrup , Denmark . Mouse mAb 3D1 recognizing CyaA was a kind gift of E . Hewlett ( University of Virginia , Charlottesville , VA ) . Monoclonal antibodies were unlabeled and/or conjugated with Alexa Fluor 488 ( AF488 ) , fluorescein isothiocyanate ( FITC ) , R-phycoerythrin ( PE ) , allophycocyanin ( APC ) or biotin , respectively . Cy5-conjugated goat anti-mouse IgG F ( ab′ ) 2 fragment ( GAM-Cy5 ) was obtained from Jackson ImmunoResearch Laboratories , West Grove , PA . Horseradish peroxidase labeled anti-mouse IgG antibody was purchased from GE Healthcare , Piscataway , NJ . CHO-K1 Chinese hamster ovary cells ( ATCC CCL-61 ) and THP-1 human monocytic cells ( ATCC TIB-202 ) were obtained from the American Type Culture Collection ( ATCC , Manassas , VA ) and tested for the absence of mycoplasma contamination by Hoechst stain . CHO cells were grown in F12K medium ( GIBCO Invitrogen , Grand Island , NY ) supplemented with 10% fetal calf serum ( FCS ) ( GIBCO Invitrogen , Grand Island , NY ) and antibiotic antimycotic solution ( ATB , 0 . 1 mg/ml streptomycin , 1000 U/ml penicillin and 0 . 25 µg/ml amphotericin; Sigma-Aldrich , St . Louis , MO ) . THP-1 cells were cultured in RPMI 1640 ( Sigma-Aldrich , St . Louis , MO ) supplemented with 10% fetal calf serum and antibiotic antimycotic solution . Commercial anonymous human blood and leukopacks were purchased from the Blood Bank of Thomayer Hospital , Prague , Czech Republic , and informed consent was therefore not applicable . Handling of cells of human origin was performed in compliance with the quality and safety requirements of the Act No . 296/2008 Coll . and the decree No . 422/2008 Coll . , and the used protocols were in accordance with internal guidelines of the Institute of Microbiology of the CAS , v . v . i . pT7CACT1 was used for co-expression of cyaC and cyaA genes in production of recombinant CyaC-activated CyaA in Escherichia coli under the control of the isopropyl-β-D-thiogalactopyranoside-inducible lacZ promoter ( Osicka et al . , 2000 ) . Oligonucleotide-directed PCR mutagenesis was used to construct pT7CACT1-derived plasmids for expression of CyaA mutant variants ( CyaA-D1193A+D1194A+E1195A and CyaA-E1232+D1234A , respectively ) harboring negatively charged aspartate and glutamate residues of the CD11b binding site of the toxin substituted with alanine residues . CyaA-AC- toxoid , unable to convert ATP to cAMP , was expressed from pT7CACT1-derived plasmid generated by placing a synthetic BamHI linker ( 5’-GGATCC-3’ ) , encoding a dipeptide GlyPhe , into the EcoRV site between codons 188 and 189 of cyaA , as described previously ( Osicka et al . , 2000 ) . pACT∆385-828 was used to produce a deletion mutant of CyaA , CyaA∆H , lacking the hydrophobic segment between residues 385-828 ( Šebo and Ladant , 1993 ) . Human cDNAs encoding CD11b and CD18 were a kind gift of D . Golenbock , Boston University School of Medicine , Boston , MA ( Ingalls et al . , 1998 ) . Human cDNAs encoding CD11a and CD11c were purchased from Geneservice , Cambridge , UK . The cDNAs for CD11a , CD11b and CD11c were cloned into the pcDNA3 expression vector ( Invitrogen , Carlsbad , CA ) and the cDNA for CD18 into the pcDNA3 . 1/Zeo ( + ) expression vector ( Invitrogen , Carlsbad , CA ) under the control of the human cytomegalovirus immediate-early promoter . For production of chimeric CD11b-CD11c molecules , silent mutations not altering the sequence of encoded proteins were introduced by oligonucleotide-directed PCR mutagenesis to generate restriction sites in homologous regions of the cDNAs for CD11b and CD11c . These restriction sites were then used to construct the chimeric CD11b molecules having different segments replaced by the corresponding portions of CD11c ( Figure 1A ) , or to obtain the CD11b molecule with deleted I-domain ( Figure 1A ) , or to construct a CD11c molecule harboring the 614-682 segment replaced by the CD11b counterpart segment ( Figure 3A ) . All cDNAs for expression of CD11-derived proteins were constructed in the pcDNA3 vector . To express CD11-derived proteins C-terminally fused to yellow fluorescent protein ( YFP ) , a DNA fragment encoding YFP , originating from the plasmid pEYFP-C3 ( Clontech , Ozyme , Paris , France ) , was fused in frame to the 3’-end of cDNAs ( cloned in pcDNA3 ) encoding CD11a , CD11b , CD11c and CD11c614-682b , respectively , using standard PCR and molecular cloning techniques . Oligonucleotide-directed PCR mutagenesis was used to construct pcDNA3-CD11b-derived plasmids for expression of the CD11b mutant variants CD11bR619A+K621A , CD11bE625A+N627A+R629A , CD11bK644A+K646A , CD11bE650A+R652A , CD11bK659+T661A and CD11bR662A+R664A+R666A , respectively . To express the secreted ectodomain of the integrin CR3 ( sCR3 ) , the segments encoding transmembrane helices and the C-terminal cytoplasmic tails of CD11b ( codons 1106 to 1152 ) and of CD18 ( codons 700 to 769 ) were replaced by short DNA segments encoding a C-terminal 6 x His tag . CyaA and its mutant variants were produced in the presence of the activating protein CyaC , using the E . coli strain XL1-Blue ( Stratagene , La Jolla , CA ) and the proteins were purified by a combination of ion exchange chromatography on DEAE-Sepharose and hydrophobic chromatography on Phenyl-Sepharose ( Osicka et al . , 2000 ) . Endotoxin was removed by repeated washes of the toxin-bound resin with 60% isopropanol ( Franken et al . , 2000 ) . The preparations used in signaling experiments thus contained less than 0 . 1 EU/1 µg of CyaA as determined by Limulus Amebocyte Lysate assay ( QCL-1000 , Lonza , Walkersville , MD ) . On-column labeling of CyaA and its mutant variants was performed after the DEAE-Sepharose purification step . Protein samples were diluted 4-times in ice-cold 50 mM Tris-HCl , pH 8 . 0 containing 1 M NaCl and loaded on Phenyl-Sepharose beads . To label CyaA with biotin , the column was extensively washed with PBS ( 12 mM Na2HPO4 , 2 mM KH2PO4 , pH 7 . 4 , 3 mM KCl and 132 mM NaCl ) and the beads were resuspended in PBS containing NHS-Sulfo-LC-Biotin ( Pierce , Rockford , IL ) in a concentration to reach a biotin:CyaA molar ratio of ~20:1 . Biotin coupling was performed at 25°C and was stopped after 40 min by washing of the resin with 50 mM Tris-HCl , pH 8 . 0 solution , and then extensively with PBS . Purified biotinylated CyaA ( CyaA-biotin ) was then eluted with 50 mM HEPES , 8 M urea and 2 mM EDTA . To label CyaA and its mutant variants with Dyomics 647 ( Dy647 ) , the Phenyl-Sepharose column was washed with 50 mM sodium bicarbonate ( pH 8 . 3 ) and the beads were subsequently resuspended in the same buffer containing Dy647-NHS ester ( Dyomics , Jena , Germany ) in a concentration to reach a Dy647:protein molar ratio of ~6:1 . Labeling was performed at 25°C for 2 hr , the column was washed with 50 mM Tris-HCl ( pH 8 . 0 ) and labeled proteins were eluted in a buffer containing 50 mM Tris-HCl ( pH 8 ) and 8 M urea . The binding , cell-invasive and hemolytic activities of labeled CyaA-derived proteins were determined as described elsewhere ( Osicka et al . , 2000 ) and were comparable ( >90% ) to activities of untreated CyaA . A CHO cell line stably expressing the CD18 subunit ( CHO-CD18 ) was established and used for cell surface expression of the CD11 subunits or their respective chimeric/mutant variants . Briefly , 0 . 8 μg of a highly purified plasmid DNA containing the appropriate integrin cDNA insert was mixed with 2 μl of Lipofectamine 2000 ( Invitrogen , Grand Island , NY ) in 100 μl of Opti-MEM I medium ( Invitrogen , Grand Island , NY ) and the mixture was incubated at 25°C for 20 min . The DNA:lipofectamine mix was added to CHO cells ( > 90% confluency ) in F12K medium with 10% FCS . After 6 hr at 37°C , the medium was changed , the transfected cells were cultured for 24 hr and then transferred to F12K medium supplemented with 10% FCS , ATB and 1000 μg/ml of G418 ( InvivoGen , San Diego , CA ) and/or 600 μg/ml of zeocin ( InvivoGen , San Diego , CA ) to select stable transfectants for 10 days . Stably transfected CHO cells were stained with anti-CD11a , anti-CD11b , anti-CD11c and/or anti-CD18 mAbs labeled with AF488 , FITC , or APC and cells expressing high levels of integrin molecules were selected , using a FACS Vantage cell sorter ( Becton Dickinson , Franklin Lakes , NJ ) , by sorting single cells into individual wells of 96-well plates containing F12K medium supplemented with 10% FCS , ATB and 600 μg/ml of G418 and/or 300 μg/ml of zeocin . After 3 weeks , the cells from positive wells were expanded , stained with mAbs recognizing the integrin subunits and CHO clones expressing similar amounts of integrin molecules on the cell surface ( the maximum difference between surface expression of mutant integrins and intact CR3 was ± 30% ) were identified by flow cytometry on a FACS LSR II instrument ( BD Biosciences , San Jose , CA ) . To monitor that the levels of integrin expression on the cell surface of CHO transfectants remained constant , the amounts of surface-expressed integrin molecules were systematically quantified in each experiment by staining with mAbs and subsequent flow cytometry analysis . The obtained mean fluorescence intensity ( MFI ) values from integrin detection were used to normalize the relative values of CyaA binding to cells . To produce the secreted form of the integrin ectodomain ( sCR3 ) , CHO cells were co-transfected with plasmids harboring cDNAs encoding sCD11b and sCD18 . A stably transfected cell line was established and individual cells were cloned from the bulk cell population using a FACS Vantage cell sorter . Stably transfected CHO-sCR3 cells were screened for secretion of the integrin ectodomain by a sandwich ELISA , using the MEM-48 mAb recognizing the CD18 subunit as a capture antibody and the biotinylated MEM-174 mAb directed against the CD11b subunit as detection antibody , respectively . The CHO-sCR3 clone with the highest secretion level of sCR3 was selected for further work . CHO-sCR3 cells were cultured in F12 medium supplemented with 10% FCS at 37°C and culture supernatants were harvested every week . sCR3 was purified from the filtered culture supernatant by using an immunoaffinity chromatography matrix prepared by a covalent linkage of MEM-174 mAb to Cyanogen bromide ( CNBr ) -activated Sepharose 4B beads ( GE Healthcare , Piscataway , NJ ) . The bent and extended conformers of the sCR3 ectodomain were separated by gel filtration chromatography on Superdex 200 HR ( GE Healthcare , Piscataway , NJ ) in HBSS buffer complemented with 2 mM CaCl2 and 2 mM MgCl2 . The freshly separated conformers of sCR3 were directly applied on glow-discharged carbon copper grids ( Benada and Pokorny , 1990 ) , stained with 0 . 75% uranyl formate , and visualized in a Philips CM 100 transmission electron microscope ( FEI , Eidhoven , Netherlands ) equipped with a MegaViewII slow scan camera controlled by AnalySis 3 . 2 software ( Olympus Soft Imaging Solutions , Münster , Germany ) . All assays were performed in HEPES-buffered salt solution ( HBSS buffer; 10 mM HEPES , pH 7 . 4 , 140 mM NaCl , 5 mM KCl ) complemented with 2 mM CaCl2 , 2 mM MgCl2 , 1% ( w/v ) glucose and 1% ( v/v ) FCS ( cHBSS ) in 96-well culture plates ( Nunc , Roskilde , Denmark ) . For staining of integrin molecules on cell surface by mAbs , 2x105 transfected CHO cells were incubated for 30 min at 4°C in 50 µl of cHBSS buffer containing mAbs diluted according to manufacturer´s instructions . For CyaA binding assay , 2x105 transfected CHO cells were incubated in 100 µl of cHBSS buffer containing 2 µg/ml of CyaA-biotin ( the concentration giving approximately a half-maximal binding of the toxin to CHO cells expressing intact CR3 ) for 30 min at 4°C . The surface-bound CyaA-biotin was stained with streptavidin-PE ( diluted 1:400; eBioscience , San Diego , CA ) for 30 min at 4°C . After washing , cells were resuspended in HBSS and analyzed by flow cytometry in the presence of 1 µg/ml of propidium iodide , or Hoechst 33258 . Data were analyzed using the FlowJo software ( Tree Star , Ashland , OR ) and appropriate gatings were used to exclude cell aggregates and dead cells . CyaA binding was expressed as percentage of toxin binding to CHO cells expressing the native form of CR3: ( ( MFI of cells expressing the mutant integrin - MFI of mock transfected CHO cells ) / ( MFI of cells expressing intact CR3 - MFI of mock transfected CHO cells ) ) x 100 . For blocking of CyaA binding to CR3 by mAbs , 2x105 CHO-CD11b/CD18 cells were preincubated for 15 min at 4°C in the presence of saturating concentrations of different mAbs in 45 µl of cHBSS buffer . CHO-CD11b/CD18 cells incubated with isotype control mAbs or with cHBSS buffer alone were used as controls . 5 µl of CyaA-biotin was then added to the cells in the continuous presence of the mAbs to a final concentration of 2 µg/ml and incubation was continued for 30 min at 4°C . CyaA binding was determined by flow cytometry as described above and expressed as percentage of toxin binding to CHO-CD11b/CD18 cells treated without mAb: ( ( MFI of mAb-treated cells incubated with CyaA - MFI of mAb-untreated cells incubated without CyaA ) / ( MFI of mAb-untreated cells incubated with CyaA - MFI of mAb-untreated cells incubated without CyaA ) ) x 100 . 1x105 CHO transfected cells were incubated with various concentrations of CyaA ranging from 0 to 20 ng/ml for 30 min at 37°C in D-MEM and the reaction was stopped by addition of 0 . 2% Tween-20 in 50 mM HCl . Samples were boiled for 15 min at 100°C , neutralized by addition of 150 mM unbuffered imidazole and cAMP levels were determined as described previously ( Karimova et al . , 1998 ) . Due to the extremely high specific catalytic activity of the AC domain , the concentration of CyaA in the cAMP assay was two to three orders of magnitude lower than the toxin amount used in the CyaA binding assay . A leukocyte-enriched fraction was prepared from fresh human blood of healthy donors by hypotonic lysis of red blood cells . Briefly , 1 ml of whole blood was diluted to 25 ml of ice-cold 0 . 2% NaCl solution , incubated for 30 s , and isotonicity of the solution was restored by addition of 25 ml of ice-cold 1 . 6% NaCl . The suspension was centrifuged for 6 min at 250 g and the pelleted leukocytes were resuspended in 0 . 5 ml of the pre-warmed DMEM medium . The cells were treated without or with 100 nM phorbol 12-myristate 13-acetate ( PMA ) or 1 mM Mn2+ ions for 15 min at 37°C to activate CR3 . 100 µl aliquots containing 2x105 cells were immediately stained with OKM-1 mAb recognizing both CR3 conformations , or with MEM-148 mAb recognizing only the extended integrin conformation , or with different concentrations of Dy647-labeled CyaA , in a combination with anti-CD14 mAb . After 2 min , cells were analyzed by flow cytometry in the presence of 1 µg/ml of Hoechst 33258 . Data were analyzed using the FlowJo software ( Tree Star , Ashland , OR ) and live monocytes , gated based on light-scatter characteristics and CD14-positivity ( monocytes did not change the level of surface expression of CR3 after incubation with PMA/Mn2+ and/or CyaA ) , were used for assessment of CyaA binding . Human peripheral blood mononuclear cells were isolated from leukopacks of healthy donors using the Ficoll-Paque Plus density gradient centrifugation and were lysed with 50 mM Tris-HCl ( pH 7 . 4 ) , 150 mM NaCl , 1% Triton X 100 , containing a protease inhibitor cocktail ( Roche Diagnostics GmbH , Mannheim , Germany ) . LFA-1 was purified from cell lysates using the anti-CD11a mAb MEM-25 coupled to CNBr-activated Sepharose 4B beads ( GE Healthcare , Piscataway , NJ ) . SPR measurements were performed at 25°C using a ProteOn XPR36 protein interaction array system ( Bio-Rad Laboratories , Hercules , CA ) . The freshly separated bent and extended conformers of the sCR3 ectodomain and the freshly purified LFA-1 integrin were diluted to concentrations of 1–20 µg/ml in 10 mM acetate buffer ( pH 5 . 0 ) and immobilized on a GLC chip using a ProteOn amine coupling kit ( Bio-Rad Laboratories ) at a flow rate of 30 µl/min . Non-reacted activated groups were blocked by the injection of 1 M ethanolamine ( pH 8 . 5 ) . The subsequent SPR measurements were carried out in 10 mM HEPES , pH 7 . 4 , 150 mM NaCl , 2 mM CaCl2 and 0 . 005% Tween 20 at a flow rate of 100 µl/min for both the association and the dissociation phase of the sensograms . To minimize non-specific binding and mass transfer effects , three coupling concentrations of each sCR3 conformer ( leading to refractive index changes of 400 , 800 and 1200 RU ) were tested . Initial experiments at each of the sCR3 coating concentrations were conducted with CyaA∆H at concentrations ranging from 20 to 320 µg/ml . The initial estimates of koff values then showed a concentration independence at all coupling levels , indicating that artifacts from analyte rebinding and mass transfer effects could be ruled out . Consequently , a coupling level of 1200 RU and flow rate of 100 µl/ml was chosen for all remaining experiments . The CyaA∆H protein was serially diluted in running buffer to concentrations of 20 , 40 , 80 , 160 and 320 µg/ml and injected in parallel ( ‘one-shot kinetics’ ) over the immobilized sCR3 and LFA-1 surface . Response curves were evaluated using the ProteOn Manager software ( Bio-Rad Laboratories ) . The sensograms were corrected for sensor background by interspot referencing ( the sites within the 6×6 array , which are not exposed to ligand immobilization but are exposed to analyte flow ) , and double referenced by subtraction of analyte ( channel 1–5 ) using a ´‘blank´’ injection ( channel 6 ) . The data were analyzed by global fitting of the response curves using a 1:1 Langmuir binding kinetics and a bivalent analyte model to determine the kinetic association and dissociation rate constants . The Langmuir-type model assumes the interaction between ligand ( L ) and analyte ( A ) resulting in a direct formation of the final complex ( LA ) : L+A↔ka , kdLA , where ka and kd are the association and the dissociation rate constants , respectively . The bivalent analyte model assumes two-step association process: L+A↔ka1 , kd1LA+L↔ka2 , kd2LLA , where the first binding event is described by ka1 and kd1 , while ka2 and kd2 describe the association and dissociation of the second binding event , respectively . Fitting of the binding curves revealed that the bivalent analyte model described the interaction of CyaA∆H with immobilized sCR3 significantly better in terms of reduced χ2 value and residual statistics than the simple 1:1 Langmuir binding model . d[L]/dt = − ( 2 × ka1 × [L] × [A] – kd1 × [LA] ) – ( ka2 × [LA] × [L] – 2 × kd2 × [LLA] ) d[LA]/dt = ( 2 × ka1 × [L] × [A] – kd1 × [LA] ) – ( ka2 × [LA] × [L] – 2 × kd2 × [LLA] ) d[LLA]/dt = ( ka2 × [LA] × [L] – 2 × kd2 × [LLA] ) Experiments were performed on the human monocytic cell line , THP-1 , and primary human monocytes isolated from leukopacks using the Ficoll-Paque Plus density gradient centrifugation followed by a magnetic cell-sorting using anti-CD14 magnetic beads to collect the CD14 positive fraction with an AutoMACS Pro Separator ( Miltenyi Biotec , Bergisch Gladbach , Germany ) . To opsonize zymosan with iC3b , the complement activation cascade in serum was used . Zymosan A ( Sigma-Aldrich , St . Louis , MO ) was incubated in 50% human serum at 37°C for 30 min . Zymosan particles were washed with PBS to remove unattached components of serum and dissolved in DMEM without FCS before use for treatment of cells . Unopsonized zymosan was prepared by incubating zymosan particles in PBS at 37°C for 30 min . CyaA or genetically detoxified CyaA ( CyaA-AC- ) was diluted to final concentrations from concentrated stocks in DMEM without FCS . For the time course experiments , 3x106 THP-1 cells were treated with 30 ng/ml of toxin for different time points at 37°C . For concentration dependence experiments , 3x106 THP-1 cells were incubated at 37°C with different amounts of CyaA or CyaA-AC- for 15 or 30 min , respectively . 3x106 THP-1 cells treated with 300 µg of opsonized or unopsonized zymosan were used as positive and negative controls , respectively . THP-1 cells treated with TUC buffer ( 50 mM Tris-HCl ( pH 8 . 0 ) , 8 M urea , 2 mM CaCl2 ) diluted to the same level as the highest toxin concentration were used as an additional control . To investigate the effect of elevated cAMP levels on Syk activation , 3x106 THP-1 cells or primary monocytes were pretreated with 300 ng/ml of CyaA , CyaA-AC- , or TUC buffer for 15 min , followed by treatment with iC3b-opsonized zymosan for 30 min at 37°C . Cells treated with TUC buffer followed by unopsonized zymosan were used as a control . To examine the effect of cAMP on already activated Syk , 3x106 THP-1 cells or primary monocytes were incubated for 15 min with iC3b-opsonized zymosan followed by addition of 300 ng/ml of CyaA , CyaA-AC- , or TUC buffer for 30 min at 37°C . Cells incubated with unopsonized zymosan followed by TUC buffer were used as a control . The reactions were carried out in DMEM with 10% FCS and no antibiotics . Triton X lysis buffer ( 50 mM Tris-HCl ( pH 7 . 4 ) , 150 mM NaCl , 1 mM Na3VO4 , 10 mM NaF , 0 . 13% SDS , 1% Triton X 100 and EDTA-free protease inhibitor cocktail ( Roche Diagnostics GmbH , Mannheim , Germany ) ) was used to stop the reaction and to lyse the cells at the respective time points . Cell lysates were collected and debris was removed by centrifugation . Aliquots of lysates were taken to detect the total amounts of Syk in individual samples . Tyrosine phosphorylated proteins were immunoprecipitated using anti-phosphotyrosine mAb ( P-Tyr-01 ) for 3 hr at 4°C and bound to protein A-coupled sepharose beads ( GE Healthcare , Uppsala , Sweden ) overnight at 4°C . The beads were washed in Triton X lysis buffer and boiled with Laemmli loading buffer . The proteins were separated by SDS-PAGE and transferred onto a nitrocellulose membrane ( Pall Corporation , Pensacola , FL ) . The membrane was blocked with 5% BSA in TBS-T ( 25 mM Tris-HCl ( pH 7 . 4 ) , 300 mM NaCl , 2 . 6 mM KCl , 0 . 3% Tween 20 ) for 60 min at 25°C and incubated with anti-Syk mAb ( SYK-01 ) for 1 h at 25°C . Upon washing with TBS-T , the membrane was incubated with horseradish peroxidase-conjugated anti-mouse IgG antibody for 1 hr and washed with TBS-T . The blots were developed using SuperSignal West Femto maximum sensitivity substrate ( Thermo Scientific , Rockford , IL ) and the chemiluminescent signal was recorded using an ImageQuant LAS 4000 Imager ( GE Healthcare , Uppsala , Sweden ) . The 3D structure model of the CR3 complex was generated with the Modeler suite of programs ( Eswar et al . , 2003 ) using the known 3D structure of the highly homologous CR4 complex ( PDB ID 3K72 , chains A B; omitting the highly flexible I-domain and the C-terminal domains that are not deposited in the PDB file ) ( Xie et al . , 2010 ) . The structure of CR3 was then equilibrated to allow partial spatial rearrangement of the protein . Short ( 2 ns ) MD simulation was performed using OpenMM ( Eastman and Pande , 2010 ) Zephyr ( Friedrichs et al . , 2009 ) ( code freely available on https://simtk . org/home/zephyr ) implementing GPU accelerated version of GROMACS suite of programs ( Van Der Spoel et al . , 2005 ) . Implicit solvation ( GBSA , ε=78 . 3 , “‘accurate water”’ with collision interval of 10 . 99 fs ) in combination with parm96 force-field was used ( Kollman , 1996 ) . The initial structure was next optimized and refined by simulation at 300 K with time step of 2 fs . The structure of the amino acid segment 1166–1287 of CyaA , which was found to be responsible for CD11b binding ( El-Azami-El-Idrissi et al . , 2003 ) , was predicted using I-TASSER ( Roy et al . , 2010 ) . To analyze the binding mode of CyaA to CR3 , a flexible side chain docking of the modeled region 1166–1287 of CyaA to the CR3 complex was performed using the ClusPro server ( Comeau et al . , 2007 ) . The 3D structure of the full-length CD11b subunit was subsequently modeled using the structure prediction server I-TASSER ( Roy et al . , 2010 ) . The resulting structure and orientation of the I-domain and of the C-terminal segment agreed with the crystal structure of CR4 ( Xie et al . , 2010 ) . Results were expressed as the arithmetic mean ± standard deviation ( SD ) of the mean . Student’s t-test was used to calculate statistical significance when two groups were compared . To test more than two groups , statistical analysis was performed by one-way ANOVA followed by Dunnett's post-test , comparing all the samples with the control . GraphPad Prism 6 . 0 ( GraphPad Software ) was used to perform statistical analysis . Significant differences are indicated by asterisks ( * , p<0 . 05; ** , p<0 . 01; *** , p<0 . 001; **** , p<0 . 0001 ) . | The outer surfaces of animal cells are coated with proteins , including many that are able to sense signals from the environment . The integrins are one such group of proteins . Particular ions or small molecules – collectively known as ligands – can bind to these proteins and activate cascades of signaling events inside the cell . An integrin called complement receptor 3 ( CR3 ) resides on the surface of many immune cells . CR3 binds to molecules found on the surface of bacteria , and prompts the immune cell to engulf and destroy the bacteria . The ligands bind to a region of CR3 called the I-domain , and it is thought that this domain is only able to accept ligands once the integrin protein has adopted an active form . Bordetella pertussis – the bacterium that causes a disease called whooping cough – subverts the immune defenses of the host . B . pertussis produces a toxin known as adenylate cyclase toxin ( CyaA ) that binds to CR3 in order to penetrate the immune cell and stop immune responses from being activated . However , it is not clear how CyaA is able to bind to CR3 without activating the signaling cascades . Here , Osicka et al . used biochemical techniques to address this question . The experiments reveal that CyaA mostly binds to an inactive form of CR3 through a unique site outside of the I-domain . It enables the toxin to use the integrin without triggering an immune response . Furthermore , the experiments show how CyaA prevents ligand signaling via CR3 proteins to allow B . pertussis to shut down the host’s first line of defense against infection . Osicka et al . ’s findings show how CyaA evades the host’s immune system and highlight the central role played by this toxin in B . pertussis infections . In the future , these findings could inform efforts to produce more effective vaccines against whooping cough . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2015 | Bordetella adenylate cyclase toxin is a unique ligand of the integrin complement receptor 3 |
This study sought to evaluate the performance of metabolic gestational age estimation models developed in Ontario , Canada in infants born in Bangladesh . Cord and heel prick blood spots were collected in Bangladesh and analyzed at a newborn screening facility in Ottawa , Canada . Algorithm-derived estimates of gestational age and preterm birth were compared to ultrasound-validated estimates . 1036 cord blood and 487 heel prick samples were collected from 1069 unique newborns . The majority of samples ( 93 . 2% of heel prick and 89 . 9% of cord blood ) were collected from term infants . When applied to heel prick data , algorithms correctly estimated gestational age to within an average deviation of 1 week overall ( root mean square error = 1 . 07 weeks ) . Metabolic gestational age estimation provides accurate population-level estimates of gestational age in this data set . Models were effective on data obtained from both heel prick and cord blood , the latter being a more feasible option in low-resource settings .
One cord blood sample was excluded because 100% of analyte values were missing . Imputation was conducted for the remaining samples missing analyte values ( n = 28 heel samples and 21 cord samples; no individual sample had more than 5/47 ( 11% ) analyte values missing ) . The final cohort consisted of 1523 samples from 1069 unique individual newborns . 1036 samples were collected immediately after birth ( range: 0 min - 2 hr 1 min ) from the umbilical cord , and 487 heel prick samples were collected an average of 14 hr 58 min after birth ( range: 25 min - 40 hr 30 min ) . The majority of samples received ( 93 . 2% of heel prick samples; 89 . 9% of cord blood samples ) were from term infants ( gestational age ≥37 weeks ) . 18 . 1% of heel prick samples and 15 . 9% of cord blood samples were derived from infants with a birthweight <2500 g . Of the 1069 infants included in the study , 454 contributed both heel and cord blood samples . A summary of participant demographics is provided in Table 1 . We determined the performance of previously published metabolic gestational dating algorithms in heel prick-derived data from the Bangladeshi infant cohort . Results of linear regression analyses for heel prick metabolic profiles demonstrated optimal performance among term infants between 38 and 39 completed gestational weeks ( Figure 1 ) . Residual plots for each of the three models in both heel and cord samples are provided in Figure 2 . In general , all models predicted gestational ages close to full term with the highest accuracy , while tending to overestimate gestational age in preterm infants and underestimate gestational age in post-term infants , in the Bangladesh cohort . A baseline model including only clinical covariates ( infant sex , birthweight and multiple birth status , Model 1 ) provided the least accurate estimation of gestational age relative to ultrasound-validated gestational age estimates , RMSE 1 . 46 weeks . By comparison , a model including analyte covariates ( Model 2 ) had an RMSE of 1 . 35 weeks . A full model containing all clinical and analyte data ( Model 3 ) demonstrated the lowest RMSE ( best performance ) of 1 . 07 weeks and correctly estimated gestational age to within 1 week for 63 . 9% , and within 2 weeks for 94 . 3% of all heel prick samples . Among small for gestational age infants , the full heel prick model had an RMSE of 1 . 12 weeks when growth restriction was defined as birthweight below the 10th percentile for gestational age and an RMSE of 1 . 30 weeks when defined as birthweight below the 3rd percentile for gestational age . By these definitions , our model accurately estimated gestational age to within 1 week for 62 . 8% and 53 . 4% of growth-restricted infants , respectively . As with heel prick data , algorithmic estimates of gestational age most accurate among term infants ( Table 2 ) . When applied to cord blood-derived data , the baseline model ( Model 1 ) and model including analytes ( Model 2 ) performed comparably ( RMSE of 1 . 51 weeks and 1 . 45 weeks , respectively ) . As with heel prick data , the full model ( Model 3 ) provided the best estimates of gestational age ( RMSE of 1 . 23 ) . Here , gestational age was correctly estimated to within 2 weeks for 90 . 4% of infants overall ( 90 . 7% and 85% for growth-restricted infants with birthweight below the 10th and 3rd percentiles , respectively; 84 . 2% for infants < 2500 g ) . A comparison of the two sample types indicated that metabolic dating models using data derived from heel prick samples provided more accurate gestational age estimates than models using cord blood samples . We evaluated the discrimination of gestational age across a dichotomous preterm birth threshold ( ≥37 weeks vs <37 weeks gestational age ) ( Figure 3 ) . Gestational age estimation models performed best when applied to metabolic profiles derived from heel prick samples . For both types of samples , the best performance was achieved by the full model containing all clinical and analyte data ( Model 3 ) ( area under the curve [AUC] 0 . 945 ( 95% CI 0 . 890 , 0 . 999 ) for heel prick profiles and AUC 0 . 894 ( 95% CI 0 . 853 , 0 . 935 ) for cord blood profiles ) . In this paper , we demonstrate that algorithms developed using newborn screening data from Ontario , Canada are effective in deriving estimates of gestational age in infants born in Matlab , Bangladesh that are accurate to within approximately 1 to 2 weeks of ultrasound-validated gestational age . Data derived from newborn heel prick samples consistently yielded more accurate estimates of gestational age than cord blood-derived data , likely reflecting the fact that our models were originally developed from data obtained from this sample type . Indeed , we have shown that the correlation between cord blood and heel prick-derived data varies significantly across analyte subtypes ( Appendix 1 ) . Accurate assessment of gestational age , preterm birth and small for gestational age is a recognized priority area where there is a need to improve program tracking and accountability ( March of Dimes , 2012; WHO , 2014 ) . Although birthweight data are collected in most settings , it is an unreliable surrogate for gestational age that is prone to overestimation of preterm birth rates in low- and middle-income settings where a high proportion of infants are born small for gestational age . Commonly-used gestational age assessments applied after birth are hampered by their reliance on complex scoring systems . A recent systematic review and meta-analysis of 18 newborn assessments based on a variety of neuromuscular , physical and other criteria determined that the most popular scoring systems ( the Ballard and Dubowitz scores ) systematically overestimated gestational age with wide margins of error ( Lee et al . , 2016 ) . Whereas gold standard first trimester ultrasound scans are accurate to within one week , the accuracy of measurements based on newborn examination varies from 2 to 4 weeks . Furthermore , newborn clinical assessments of gestational age such as Dubowitz and Ballard scoring , and neonatal anthropometrics have been demonstrated to be inaccurate surrogate markers of gestational age , specifically in rural communities of Bangladesh ( Lee et al . , 2016 ) . Metabolic gestational dating approaches emerged in response to the urgent need to improve the epidemiology and surveillance of preterm birth . Circulating newborn metabolites are known to be affected by gestational age and gestational age is routinely considered in the interpretation of newborn screening analysis ( Slaughter et al . , 2010; Oladipo et al . , 2011; Newborn Screening Ontario , 2017 ) . To date , three groups in North America have developed metabolic dating algorithms based on newborn health administrative datasets ( Jelliffe-Pawlowski et al . , 2016; Ryckman et al . , 2016; Wilson et al . , 2016 ) . Research has since sought to refine existing models through the addition of analytes known to correlate with gestational age and develop tiered models of varying complexity . Our own group has demonstrated that proportions of fetal and adult hemoglobins are some of the strongest individual predictors of gestational age , ( Wilson et al . , 2017 ) and we have also validated our algorithms across ethnic subgroups in Ontario ( Hawken et al . , 2017 ) . Efforts are currently underway to begin implementing metabolic gestational age dating in low-resource settings to determine the burden of preterm birth and intrauterine growth restriction . The results from our study offer a reason to be optimistic about these efforts . While the intent of metabolic gestational age dating at present is to provide population-based estimates of the burden of preterm birth , it is conceivable that this approach could also be used to guide care for individual newborns who are identified as preterm . Our study had a number of important strengths and limitations . Strengths of our approach include the use of internationally-derived samples to externally validate our models and using samples from a well-described cohort of infants with gestational age confirmed by first trimester ultrasound . The study design of the PreSSMat cohort in which our study was nested ensured that enrollment was open to a representative selection of women and newborns delivering in the Matlab icddr , b service area . Other strengths include the high quality of samples received for analysis , and the use of paired cord blood and heel prick samples to compare model performance metrics . The primary limitation of this study is the participation bias against very preterm and extremely preterm infants , whose parents expressed reluctance to subjecting their newborn to these collection procedures . As a result , we had a relatively small number of samples collected from very preterm and extremely preterm infants , limiting our ability to comment on model performance in these sub-groups . In this Bangladesh cohort , the gestational ages estimated from our models were most accurate in infants who were confirmed to be close to full-term by first trimester ultrasound . Algorithm-derived gestational ages tended to be overestimated in preterm infants and underestimated in post-term infants . This suggests that calibration in the large ( i . e . introducing a calibration slope adjustment ( Steyerberg , 2010 ) to model predictions could improve overall model performance in this external cohort , although this was not conducted in the current study . Our findings are encouraging for several reasons . First , this work provides early evidence that gestational dating models developed using metabolic data derived from a North American cohort perform well in low-resource populations . The model originally published by our group was developed using data from a Canadian-born cohort of 250 , 000 infants . In Ontario , the model was able to estimate gestational age to within one week ( RMSE 1 . 06 vs 1 . 07 for the Bangladeshi cohort ) overall and correctly ascertain gestational age to within 2 weeks for 94 . 9% of infants ( vs . 94 . 3% for the Bangladeshi cohort ) ( Wilson et al . , 2016 ) ; estimates that compare favorably against other currently-used postnatal gestational age estimation methods that produce estimates varying in accuracy from 2 to 4 weeks gestational age ( Taylor et al . , 2010; Spinnato et al . , 1984; Robillard et al . , 1992; Lee et al . , 2016; Alexander et al . , 1992 ) . Second , our metabolic models provided significantly improved estimates of gestational age among infants with birthweights < 2500 g , cases where current estimates based on symphysis fundal height and neuromuscular assessments perform poorly ( Spinnato et al . , 1984; Goto , 2013 ) . Lastly , we are encouraged by the potential utility of cord blood profiles for deriving gestational age estimates . Differences in cord blood and heel prick profiles described in our analysis likely stem from a number of factors related to timing of collection , including early postnatal fluctuations in neonatal TSH levels , ( Ryckman et al . , 2012; Büyükgebiz , 2013 ) and infant feeding status prior to collection . Although the performance of the models when applied to cord-blood-derived data was somewhat attenuated relative to heel prick data , development of cord-blood-specific models restricted to analytes less susceptible to fluctuations in the postnatal environment may further improve gestational age estimation . Ultimately , acceptable levels of error in gestational age measurements will need to be determined by public health and maternal child health officials . Given the acknowledged limitations of existing alternatives to ultrasound estimation , metabolic gestational dating approaches appear to offer reliable estimates that are unencumbered by user variability . As we prepare for the scale-up and implementation of metabolic gestational dating approaches for robust population-level estimates of preterm birth , our findings highlight a number of opportunities and challenges . First , heel prick samples taken for newborn screening are typically collected at least 24 hr after birth to accommodate postpartum fluctuations in analyte levels . In many settings around the world , mother-infant pairs are discharged from healthcare settings within the first 24 hr after delivery ( Campbell et al . , 2016 ) . As a result , the accuracy of existing metabolic dating algorithms would be compromised by the change in timing of sample collection . Second , newborn screening is not a standard service of practice in low- and middle-income countries , including Bangladesh . It was therefore unsurprising that anecdotal feedback from field nurses assisting with this study indicated that parents were hesitant to consent to heel prick procedures for their infants . Although on-site research staff received extensive training through videos , visual guides and in-person training , a preference for collection of cord blood samples over heel prick amongst research staff may also have affected the number and quality of samples collected . A quality assurance trial was required to improve sample collection and handling techniques . While our current models were originally optimized for application to heel prick data , we highlight an opportunity to optimize these algorithms for use on cord blood data . Transitioning to cord blood-based models would additionally bypass the need to impose discomfort on the child , stress on parents and staff , and also avoid the requirement for extensive training and screening of sample collection techniques . Finally , population-level metabolic screening provides the additional opportunity to provide insight into the prevalence of congenital conditions in participating jurisdictions . In summary , metabolic gestational age dating approaches offer a novel means for providing accurate population-level gestational age estimates . As we work toward implementing preterm birth surveillance initiatives in a variety of low-income settings ( Mundel , 2017 ) , the level of acceptable accuracy of metabolic algorithms should be considered . Application of models to cord blood metabolic profiles is the most feasible option at present , although derivation and optimization of such models are warranted . Utility of other maternal , pregnancy and infant factors that were not available to us in the current analysis for improving existing metabolic dating models may also be of benefit . Where population-level surveillance of preterm birth might be supported through the analysis of a few drops of blood taken shortly after birth , future work should aim to derive models that determine other priority birth outcomes .
Our objective was to validate the performance of previously published gestational age estimation models developed in Ontario , Canada ( Wilson et al . , 2016; Wilson et al . , 2017 ) in a cohort of infants born in Bangladesh . Specifically , we sought to compare estimates of gestational age derived from our algorithms , through the analysis of newborn blood spots , against estimates of gestational age determined by first-trimester ultrasound . A version of the protocol for this study has been published ( Murphy et al . , 2017 ) . Due to logistical challenges in initiating the study , fewer samples were collected than initially anticipated in our protocol and low numbers of infants with gestational age below 34 weeks . Our methods of sample collection and analysis remained the same . Newborn screening is a public health initiative that screens for rare , treatable conditions that typically produce no symptoms in the neonatal period . Programs vary in scope by jurisdiction , screening for one to over 50 conditions ( Therrell et al . , 2015 ) . In Ontario , as in many regions , drops of blood are taken by infant heel prick , typically within the first few days after birth , and dried onto filter paper . Dried blood spot samples are then analyzed by a series of assays including tandem mass spectrometry , colorimetric and immunoassays as well as high-performance liquid chromatography for metabolic , genetic and other analyte markers . Sample collection was conducted in the Matlab sub-district of Chandpur , Bangladesh where the International Centre for Diarrhoeal Disease Research , Bangladesh ( icddr , b ) has been running a Health and Demographic Surveillance System ( HDSS ) in Matlab since 1966 . Based on service provision , the HDSS area is divided into two jurisdictions: 1 ) the icddr , b service area where women of reproductive age and their children under 5 years of age receive care though icddr , b facilities; and 2 ) the government service area where individuals receive care from government facilities as in other areas of the country . The present study was conducted in the icddr , b service area , and nested within a cohort study entitled ‘Preterm and Stillbirth Study , Matlab’ ( PreSSMat ) that was designed to capture data on the biological determinants of adverse pregnancy outcomes , including preterm births . In the PreSSMat cohort , pregnant women were followed prospectively along the pregnancy continuum , with scheduled visits at 11–14 ( enrollment and ultrasound ) , 22–24 , and 32 weeks’ gestation , at delivery , and at 6 weeks post-partum to collect socio-demographic and clinical data as well as biological specimens . Preterm births were defined as all births that occurred at <37 weeks’ gestation . ‘Very preterm births’ were those that occurred at <32 weeks , and ‘extremely preterm births’ were those that occurred at <28 weeks . Small for gestational age ( SGA10 ) was defined as cases where birthweight was below the 10th percentile within categories of week of gestational age at delivery and infant sex . The percentiles were calculated and applied based on a North American distribution of birthweight within sex and gestational age categories . We also calculated SGA3 , which identifies infants below the 3rd percentile within gestational age and sex categories and is much more likely to reflect infants who suffered intrauterine growth restriction , especially in low and middle-income countries such as Bangladesh where birthweights are lower . Pregnant women were identified by community health workers through monthly home visits . All enrolled women underwent a gestational dating ultrasound at enrollment; otherwise no explicit inclusion or exclusion criteria were applied . All women enrolled in the PreSSMat cohort were eligible for participation in the current study . To examine the effect of timing of sample collection on newborn metabolic profiles , cord blood was collected immediately after birth and spotted on Whatman 903 filter paper . A second dried blood spot sample was also collected via heel prick within 72 hr of delivery or immediately prior to discharge , whichever happened first . The latter reflects the timing of collection for samples used to develop our previously published gestational age estimation models ( recommended timing of sample collection for healthy newborns in Ontario , Canada is 24–48 hr after birth ) . Samples were collected onto filter paper , air-dried and shipped weekly to Newborn Screening Ontario ( NSO ) , the provincial newborn screening facility in Ottawa , Canada . Samples were stored in a temperature and humidity-controlled environment prior to shipment . Eight 3 . 2 mm diameter samples were punched from each sample for testing of the following analytes: hemoglobin profiles; 17α hydroxyprogesterone ( 17-OHP ) ; thyroid stimulating hormone ( TSH ) ; immunoreactive trypsinogen ( IRT ) ; a panel of 12 amino acids and 31 acylcarnitines; t-cell receptor excision circles ( TREC ) ; biotinidase activity; and galactose-1-phosphate uridylyltransferase activity . Hemoglobin profiles were determined by high-performance liquid chromatography on a Bio Rad Variant nbs system; neonatal 17-OHP , TSH and IRT were measured using PerkinElmer AutoDELFIA Immunoassays; amino acid and acylcarnitine analysis was performed by electrospray ionization tandem mass spectrometry ( Waters TQD ) ; total TREC copy number was measured by quantitative polymerase chain reaction using a ThermoFisher Scientific Viia 7; biotinidase and galactose-1-phosphate uridyltransferase levels were measured using the Astoria-Pacific SPOTCHECK Pro system . Clinical covariates were retrieved from the PreSSMat database to facilitate clinical interpretation of newborn screening data , and also for inclusion as model parameters in this study . Figure 4 summarizes the study design . Newborn screening blood spots are subject to degradation if collected or handled inappropriately . Samples with insufficient good-quality dried blood to complete the full panel of assays were excluded from analysis . Samples with missing analyte values had the missing levels imputed ( see Appendix 1 for details ) . In the process of applying newborn screening procedures for the analysis of samples , results of ‘screen negative’ and ‘screen positive’ were generated for conditions screened for by the NSO program . Management of incidental clinical findings ( screen-positive cases ) has been reported elsewhere ( Murphy et al . , 2017 ) . Mothers provided informed consent for their infants to be included in the PreSSMat birth cohort and to have clinical data , cord blood and newborn heel prick samples collected and analysed . The present study was approved by the Research Review and Ethical Review Committees of the International Centre for Diarrhoeal Disease Research , Bangladesh ( PR-16039 ) on July 10 , 2016 . Approvals were also obtained from the Research Ethics Boards of the Ottawa Health Science Network ( 20160219–01H ) on June 10 , 2016 , and the Children’s Hospital of Eastern Ontario ( 16/20E ) on June 8 , 2016 . | Complications from preterm birth are the leading cause of death among children under five . Ultrasounds are routinely used in wealthy countries to track babies' development . In countries with limited resources , however , ultrasounds are rare , making it harder to estimate how many children are born prematurely . Blood tests may offer a way to determine whether a newborn was born too early when ultrasounds are not available . Many countries already require clinicians to collect a drop of blood from newborns via a heel-prick or from their umbilical cord . Testing these blood spots identifies babies at risk of rare conditions so they can receive prompt treatment . Chemicals in the blood vary depending on how long the newborn spent growing in its mother’s womb . Scientists have developed a mathematical formula that can estimate a baby’s gestational age based on these chemicals . Using blood spots to estimate gestational age worked well when this strategy was tested in Canada , a high-income country . More tests are needed to determine if it works in low-income countries . Now , Murphy et al . show their blood spot-testing strategy also reliably predicts the gestational age of babies in Matlab , Bangladesh . In the experiments , blood spots were collected from 1 , 069 newborns . This included 1 , 036 cord blood samples and 487 heel prick samples . Nearly all the samples came from full-term infants . A mathematical model estimated the infants' gestational age to within an average of one week of their true age when applied to heel-prick blood samples and to within two weeks of the baby’s true gestational age 94% of the time . The model also provided reliable estimates of babies’ gestational ages when cord blood samples were tested , which is useful as the Bangladeshi parents were more comfortable with this method of blood collection . Using this strategy to estimate how many babies are born too early in low-income countries may help the countries develop strategies to reduce preterm births . The estimates might also help identify preterm babies who need special care . | [
"Abstract",
"Introduction",
"Materials",
"and",
"methods"
] | [
"epidemiology",
"and",
"global",
"health"
] | 2019 | External validation of postnatal gestational age estimation using newborn metabolic profiles in Matlab, Bangladesh |
Brains regulate behavioral responses with distinct timings . Here we investigate the cellular and molecular mechanisms underlying the timing of decision-making during olfactory navigation in Caenorhabditis elegans . We find that , based on subtle changes in odor concentrations , the animals appear to choose the appropriate migratory direction from multiple trials as a form of behavioral decision-making . Through optophysiological , mathematical and genetic analyses of neural activity under virtual odor gradients , we further find that odor concentration information is temporally integrated for a decision by a gradual increase in intracellular calcium concentration ( [Ca2+]i ) , which occurs via L-type voltage-gated calcium channels in a pair of olfactory neurons . In contrast , for a reflex-like behavioral response , [Ca2+]i rapidly increases via multiple types of calcium channels in a pair of nociceptive neurons . Thus , the timing of neuronal responses is determined by cell type-dependent involvement of calcium channels , which may serve as a cellular basis for decision-making .
Brains process sensory information to generate various kinds of physiological responses with different timings ( i . e . , with different latencies to respond ) : For example , motor control , foraging , decision-making and the sleep-wake cycle range on timescales from milliseconds to days ( Buhusi and Meck , 2005; Richelle and Lejeune , 1980 ) . In decision-making , animals choose one from multiple behavioral options based on environmental sensory information , where a temporal delay is associated with the certainty of sensory information . In primates and rodents , increases in neural activity during the delay period according to the sensory information ( ‘evidence accumulation’ ) has been described as a key physiological basis for the timing of decision-making ( Carandini and Churchland , 2013; Gold and Shadlen , 2007; Schall , 2001; Shadlen and Newsome , 2001 ) . For example , clear sensory information causes a faster rise in neuronal firing rate to a threshold and faster behavioral choice , whereas uncertain information causes a slower rise in the firing rate and slower behavioral choice . Despite their essential roles in the timing of decision-making , the neural mechanisms that generate evidence accumulation still need to be clarified . Theoretical studies suggest that evidence accumulation is mediated by recurrent neural circuits ( Gold and Shadlen , 2007; Wang , 2008 ) while intracellular mechanisms , such as calcium signaling via N-methyl-D-aspartate ( NMDA ) receptors , calcium-activated nonspecific cation ( CAN ) channels , and/or voltage-gated calcium channels ( VGCCs ) , have also been proposed ( Curtis and Lee , 2010; Major and Tank , 2004 ) . The cellular and molecular bases of simple decision-making have been studied in invertebrate animals because of the simplicity and accessibility of their nervous systems ( Kristan , 2008 ) . For example , a neuron that biases decisions was identified based on optical monitoring of the neuronal activities in the medical leech ( Briggman et al . , 2005 ) , and genetic analyses identified a neuropeptide and a catecholamine receptor that underpin simple decision-making tasks in Drosophila melanogaster and in Caenorhabditis elegans , respectively ( Bendesky et al . , 2011; Yang et al . , 2008 ) . In addition , a recent behavioral study showed that Drosophila temporally accumulates sensory evidence for decision-making ( DasGupta et al . , 2014 ) . However , the physiological mechanisms of these decision-making tasks have not been elucidated , and thus not discussed in terms of their possible commonalities with the mechanisms in mammals . In this study , we reveal the cellular and molecular mechanisms of the timing of decision-making in C . elegans . We first show that the animals migrate in an appropriate direction with unexpectedly high efficiency , likely based on the gradient of a repulsive odorant . From simultaneous monitoring of behavior and neural activity in virtual odor gradients , we find that two pairs of sensory neurons regulate this behavioral response in an opposing manner with different temporal dynamics . A pair of ASH nociceptive neurons exhibits a time-differential-like response to an increase in the odor concentration , which leads to a bout of turns in random directions similar to a ‘reflex’ . In contrast , a pair of AWB olfactory neurons exhibits a time-integral response to a decrease in the odor concentration , which leads to turn suppression with a temporal delay resembling ‘deliberation’ . The AWB response is mediated by a gradual calcium influx mainly via L-type VGCCs whereas the ASH response is mediated by a rapid calcium influx via multiple types of calcium channels . Thus , our results indicate that the timing of a sensory response , such as deliberate decision-making or rapid reflex , is determined by cell type-dependent involvement of calcium channels .
C . elegans avoids the odorant 2-nonanone , and this odor avoidance behavior is regulated by stimulus-dependent transitions between the two behavioral states , ‘pirouette’ ( a period of short migrations divided by turns ) and ‘run’ ( a period of long straight migration ) ( Figure 1A and Figure 1—figure supplement 1A ) ( Bargmann et al . , 1993; Kimura et al . , 2010 ) . The pirouette strategy ( a form of ‘biased random walk’ ) is the major behavioral strategy used by these animals for chemotaxis and thermotaxis , and the choice of migratory direction at run initiation is considered to be random in this strategy ( Lockery , 2011; Pierce-Shimomura et al . , 1999 ) . We found , however , that C . elegans appropriately chose the migratory direction in 2-nonanone avoidance: 78 . 4% of the migratory directions at run initiation and 83 . 5% during runs were away from the odor source ( Figure 1B and Figure 1—figure supplement 1B ) . Thus , during odor avoidance , animals chose the appropriate direction about four times more frequently than the inappropriate direction ( ~80% versus~20% ) . This probability of run initiation in the appropriate direction for odor avoidance was far higher than that in salt-taxis , the best-studied chemosensory behavior of the animal ( 59 . 3% in Figure 1—figure supplement 1C and D , and ~56% in Pierce-Shimomura et al . , 1999 ) , and the probability exceeded , or was at least comparable to , that of odor-taxis in Drosophila larvae ( 73 . 9% ) ( Gomez-Marin et al . , 2011; Louis et al . , 2008 ) . Appropriate directions were chosen from multiple exploratory short migrations in random directions during pirouettes ( Figure 1C and D ) . These results indicate that C . elegans can efficiently choose the appropriate migratory direction on a repulsive odor gradient ( Figure 1E ) . 10 . 7554/eLife . 21629 . 003Figure 1 . C . elegans selectively initiates runs away from the odor source . ( A ) Examples of the tracks of 2 animals during 12 min of 2-nonanone avoidance assay , overlaid on a schematic drawing of a 9 cm plate . One of the tracks is magnified below . In the magnified view , pirouettes are red and runs are blue . Arrow heads and arrows indicate the directions of run initiation and those during runs , respectively . ( B ) Histogram indicates the bearings at run initiation during 2-nonanone avoidance ( i . e . , the bearing of the arrow heads in panel A , and the initial bearing of the blue arrows in panel E ) . The bearing was determined as B = 0° when animals migrated directly away from the odor source ( = down the gradient ) and ±180° when they migrated directly toward the source ( = up the gradient ) . Migration away from the odor source ( i . e . , within ±90° bearings; red arrow ) comprised 78 . 4% of all data . ( C ) Histogram indicates all the bearings after a turn during pirouettes , including those that later switched to runs ( i . e . , the initial bearings of the red plus blue arrows in panel E ) . ( D ) Percentages of turn numbers per pirouette during 2-nonanone avoidance ( solid line and filled circles ) and in salt-taxis ( dashed lines and open circles ) . The average number of turns in a pirouette was significantly larger during 2-nonanone avoidance than in salt-taxis ( 5 . 2 vs . 2 . 2 indicated by arrows , respectively; p<0 . 001 by Mann-Whitney test ) . ( E ) Schematic drawings of salt-taxis and 2-nonanone avoidance . Typical chemotaxis such as salt-taxis in C . elegans is regulated by the pirouette ( i . e . , biased random walk ) strategy , where the animal initiates runs mostly in a random direction after a pirouette ( Pierce-Shimomura et al . , 1999 ) . In contrast , in 2-nonanone avoidance , an appropriate migratory direction is chosen from multiple trials in a pirouette . We refer to this as the ‘pivot-and-go’ strategy . All the statistical details are shown in Supplementary file 1 . For panels B , C and D ( 2-nonanone avoidance ) , the data are from 100 wild-type animals . Panel D ( salt-taxis ) is from 64 wild-type animals . The following figure supplement is available for Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 00310 . 7554/eLife . 21629 . 004Figure 1—figure supplement 1 . Differences between 2-nonanone avoidance behavior and salt-taxis of C . elegans . ( A ) Pirouettes and runs were classified by the length of turn interval ( i . e . , migratory durations ) . A distribution of turn intervals during the odor avoidance was fitted by the sum of two exponentials , suggesting that turn interval is regulated by two probabilistic mechanisms ( Pierce-Shimomura et al . , 1999 ) . Each exponential is indicated by a dashed line , and the sum is indicated by a solid line . The fitted lines were calculated with the least squares method . The vertical axis is a logarithmic scale . A period at which the numbers of the short and long intervals were equal was determined as a threshold value tcrit ( vertical red dotted line ) . Turn intervals longer than tcrit were classified as runs , and shorter intervals and turns were classified in pirouettes . ( B ) The histogram indicates the bearing in each step ( i . e . , per second ) of the runs during 2-nonanone avoidance . Migration away from the odor source ( i . e . , within ±90° bearings; red arrow ) comprised 83 . 5% of all data . ( C ) A typical track of C . elegans during salt-taxis for 20 min ( 200 mM NaCl spotted onto grid points ) ( Iino and Yoshida , 2009 ) . As in Figure 1A , pirouettes and runs are indicated with red and blue , respectively . ( D ) Histogram indicates bearings at run initiation of 64 animals during salt-taxis . In this experiment , the bearing was determined as B = 0° when animals migrated directly toward the source ( up the gradient ) and ±180° when they migrated directly away from the source ( down the gradient ) . Migration toward the source ( i . e . , the data within ±90° , indicated by a red arrow ) was 59 . 3% of all data . ( E ) A distribution of turn interval during the 180 s down phase of the odor gradient ( data from Figure 4B , middle right panel ) was fitted by the sum of two exponentials similarly to panel A . The fitted lines were calculated with the least squares method . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 004 Next , to understand the correlations between sensory information and appropriate directional choice , we developed a method for measuring and determining the dynamic spatio-temporal pattern of the naturally evaporating and diffusing odor gradient ( Figure 2A and Figure 2—figure supplement 1A–D , and Video 1 ) . The measured odor gradient was then used to calculate the odor concentrations that each animal experienced at each position at every second ( Cworm in the middle panels of Figure 2B and Figure 2—figure supplement 1E ) . We found that the temporal changes in Cworm ( dCworm/dt ) were strongly correlated with the two behavioral states of the animals ( the bottom panels of Figure 2B and Figure 2—figure supplement 1E , and Figure 2C ) . During pirouettes , the dCworm/dt values were mostly positive because the animals did not migrate much while the odor concentration on the plate was increasing due to sustained evaporation of the odor from the source ( Figure 2—figure supplement 1D and Video 1 ) . During runs , in contrast , dCworm/dt values were mostly negative because the animals migrated down the gradient . These correlations may suggest that pirouettes and runs are caused by positive and negative dCworm/dt , respectively . In addition , we further found that the relationships between animals' responsiveness and instantaneous dCworm/dt largely differed between pirouettes and runs ( Figure 2D ) , suggesting that pirouettes and runs are physiologically distinct behavioral states in terms of sensory response . Taken together , these results suggest that the efficient transitions between discrete behavioral states based on odor concentration information may lead to the appropriate choice of migratory direction as a simple form of decision-making . 10 . 7554/eLife . 21629 . 005Figure 2 . Pirouettes and runs are distinct behavioral states , which are associated with positive and negative dCworm/dt , respectively . ( A ) Fitted odor gradient over the assay plate at 12 min , based on the actual measurements shown in Figure 2—figure supplement 1D . ( B ) ( Top ) Same with the magnified view of an animal's trajectory in Figure 1A . ( Middle , bottom ) Graphs showing the 2-nonanone concentration ( Cworm: middle ) or temporal changes in it ( dCworm/dt: bottom ) at this animal's position at each second during the odor avoidance behavior . As in Figure 1A , pirouettes and runs are red and blue , respectively . Most of the animals did not migrate much during the first 2 min and were excluded from the analysis ( green ) because of the rapid increases in the odor concentration during this period . See also Figure 2—figure supplement 1E for another example . ( C ) Correlation between dCworm/dt and pirouettes or runs . dCworm/dt of 2-nonanone was plotted against the animal's x position for each second during pirouettes ( top ) or runs ( bottom ) . The bars represent the median ± quartiles for each 5 mm fraction . ( D ) The responsiveness to the instantaneous dCworm/dt differed between pirouettes and runs . The turning rate was determined as the relationship between dCworm/dt during one second of migration and the probability of turning in the next second . Average turning rates ± SEM for every 0 . 005 μM/s bin during pirouettes ( red line ) or runs ( blue line ) are shown . The data in panels C and D are from the same 100 wild-type animals as in Figure 1 . The following figure supplement is available for Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 00510 . 7554/eLife . 21629 . 006Figure 2—figure supplement 1 . Measurement of the gaseous 2-nonanone gradient in the plate assay paradigm . ( A ) A schematic cross-section ( upper panel ) and top view ( lower panel ) of gas sampling . The plate is placed upside-down . In the lower panel , crosses indicate positions of the odor source and dots indicate sampling positions , where numbers represent x-y coordinates . In all positions , the tip of the needle was placed ~2 mm below the surface of the agar; an example of the sampling from ( x , y ) = ( 11 , 0 ) is shown in the upper panel . The gas sample was immediately subjected to gas chromatography ( GC ) analysis . Note that one plate was used for each sampling so that sampling did not disturb the odor gradient . ( B ) A sample record from the GC analysis of 4 μM 2-nonanone . A single large peak was detected at ~260 s , and the peak height was used for calibration . ( C ) Calibration curve for 2-nonanone . Each dot represents the average of 3–4 experiments , and data on the log-log plot were fit by two simple regression lines for lower ( squares ) and higher ( triangles ) concentrations because of the detector's characteristics ( see Materials and Methods ) . ( D ) 2-nonanone gradient measured along the x axis ( left panel ) or at ( x , y ) = ( 22 , 0 ) and ( 22 , 15 ) ( right panel ) at different periods of the assay . The sampling points are indicated in panel A . The gas ( 0 . 2 mL ) was sampled at 1 , 3 , 6 , 9 , and 12 min . Each data point represents the median ± quartile of 7–9 independent experiments . Notably , the error bar of each measured value was small in the right half of the plate ( x ≥ 0 ) , where most of the animals were located during the avoidance behavior . This was likely due to the fact that odor diffusion smooths out positional differences in odor concentration . No significant differences were observed along the y axis at the same time point ( right panel; Mann-Whitney test ) . ( E ) Another example ( the upper animal in the top panel of Figure 1A ) of correlations between an animal's behavioral pattern and changes in 2-nonanone concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 00610 . 7554/eLife . 21629 . 007Video 1 . Time-course changes in the fitted 2-nonanone concentration . Although the odor sources were two circles of ~5 mm diameter in the real experiment , they were treated as points in the simulation . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 007 To determine whether sensory stimuli are the causal reason for the behavioral response during odor avoidance , and the neural mechanisms linking stimulus and behavior , we developed a novel integrated microscope system for the quantitative optophysiological analyses of freely moving C . elegans on virtual odor gradients . We integrated the OSB system , an auto-tracking microscope system for calcium imaging and optogenetic analyses of neuronal activity in C . elegans ( Video 2 ) , with an odor-delivery sub-system ( Busch et al . , 2012; Tanimoto et al . , 2016 ) . Using this new OSB2 system , a moving C . elegans was continuously exposed to an odor flowing from syringe pumps ( Figure 3A and Figure 3—figure supplement 1A , and Video 3 ) . The odor concentration in the flow changed according to a predefined program , based on estimated values experienced in the plate assay paradigm ( Figure 2 ) . Using the OSB2 system , we first tested whether the dC/dt of the odor itself could regulate an animal's behavior . When animals experienced temporal increases in the odor concentration , they exhibited more frequent turns like in pirouettes ( Figure 3B ) . Conversely , when they experienced temporal decreases , they suppressed turns and exhibited long migrations similar to runs ( Figure 3B ) . During the odor-down phase , no statistical bias was detected in migratory direction ( Figure 3—figure supplement 1B ) . These results indicate that C . elegans responds to temporal changes in the odor concentration and regulates turning rates . 10 . 7554/eLife . 21629 . 008Figure 3 . AWB and ASH sensory neuron pairs regulate turning rate in response to dC/dt of 2-nonanone . ( A ) Schematic drawing of the OSB2 system . ( B ) Behavioral response to temporal changes in the 2-nonanone concentration . ( Top ) Track of a wild-type animal . The first 60 s ( gray ) is a period of no odor ( ‘odor-zero phase’ in the bottom left panel ) , 60–90 s ( black ) is a period with a constant increase in odor concentration from 0 to 1 μM ( ‘odor-up phase’ ) , and 90–180 s ( gray ) is a period with a constant decrease in odor concentration from 1 to 0 μM ( ‘odor-down phase’ ) . Red dots indicate turns . ( Bottom left ) Rastergram of turns . The upper portion of the panel shows the measured 2-nonanone concentration in the flow . In the lower portion of the panel , each turn is denoted by a dot , and each row represents the behavioral record of a single animal during the 180 s of analysis . The results of 20 animals are shown . ( Bottom right ) Ensemble averages ± SEM for the turning rate ( turns per second ) during each phase in the left panel ( n = 20 ) . The turning rates in the three conditions differed significantly from each other ( ***p<0 . 001 , Kruskal-Wallis test with post hoc Steel-Dwass test ) . ( C ) Response of AWB neurons . The averages ± SEM of ∆F/F0 for GCaMP3 ( middle ) and rastergram of turns of the animals ( bottom ) are shown ( n = 9 ) . ( D ) Effect of optogenetic activation of AWB neurons on the turning rate in the absence of odor . Transgenic animals expressing ChR2 ( C128S ) , a bi-stable variant of ChR2 , in AWB were cultivated in the absence or presence of all-trans-retinal ( ATR ) ( dashed or solid bars , respectively ) ; exogenous ATR is required for functional ChR in C . elegans ( Nagel et al . , 2005 ) . Average turning rates ± SEM of the 30 s periods before ( gray bars ) or after blue light illumination for 3 s ( blue bars ) are shown ( n = 20 each ) . **p<0 . 01 ( Mann-Whitney test ) . ( E ) Calcium imaging of ASH neurons using GCaMP3 ( n = 7 ) . ( F ) Effect of optogenetic silencing of ASH neurons on the turning rate . The transgenic animals expressing Arch in ASH neurons were cultivated in the absence or presence of ATR ( dashed or solid bars , respectively ) and illuminated with green light during the up-phase ( n = 16 each ) . The odor pattern was the same as that in panels B , C , and E . *p<0 . 05 ( Mann-Whitney test ) . ( G ) Model of the regulation of odor avoidance by ASH and AWB neurons . All the statistical details are shown in Supplementary file 1 . The following figure supplement is available for Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 00810 . 7554/eLife . 21629 . 009Figure 3—figure supplement 1 . Spatial arrangement of the odor stimulation and behavioral response in the OSB2 system . ( A ) Arrangement of the odor flow on the OSB2 system . The end of the tube was positioned ~1 mm from the animal , and odor flow covered the entire body of the animal . Visualization was obtained from the ocular lens of the microscope , and the airflow and direction of an animal's movement and of the stage were overlaid . See also Video 3 . ( B ) Distribution of migratory bearing during negative dC/dt . Bearings of each animal's migratory vector throughout decreases in odor concentration , which were formed by connecting the start and end points of each animals’ trajectory during the odor-down phase , were plotted for the 20 animals that were analyzed in Figure 3B . No significant bias was observed ( p>0 . 1 , Watson's U2 test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 00910 . 7554/eLife . 21629 . 010Video 2 . A demonstration video for calcium imaging with the OSB2 system . ( Left ) The bright field images for the tracking and the fluorescence images for calcium imaging were acquired simultaneously but separately in the tracking and calcium imaging subsystems , respectively . The video was compiled separately because of the different frame rates and ran on a Power Point file . Because of a technical limitation on the video playing , the videos were not completely synchronized . ( Right ) From the fluorescence images , the cell body of AWB neuron was tracked and centered off-line . The video speed is 2 . 8x . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 01010 . 7554/eLife . 21629 . 011Video 3 . Visualization of the odor flow on the OSB2 system . The view was from the ocular lens of the microscope . The tube end was on the left and the flow was from the left to the right , which was visualized by fog produced by Wizard Stick ( Zero Toys , USA ) . Contrast of the video image was enhanced for visualization of the flow . The worm was immobilized for reference . The video speed is 1x . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 011 We then performed calcium imaging of neural activity and revealed that two pairs of sensory neurons regulate the avoidance behavior by responding to either increases or decreases in the odor concentration . In C . elegans' sensory neurons , dynamic changes in [Ca2+]i in the cell bodies are generally similar to those in axons , which cause neurotransmitter release from the neurons ( Kato et al . , 2014; Zahratka et al . , 2015 ) . First , we investigated a pair of AWB olfactory neurons . AWB neurons are known to be primarily responsible for 2-nonanone avoidance and to exhibit an odor-OFF response when animals were stimulated with 2-nonanone-saturated buffer in a stepwise manner ( Ha et al . , 2010; Troemel et al . , 1997 ) . However , it was not clear whether and how AWB neurons respond to gradual and/or subtle changes in odor concentration in the air phase . We found that AWB neurons were gradually and continuously activated during the odor-down phase for 90 s ( Figures 3C , 90–180 s ) . Optogenetic activation of AWB neurons by the bi-stable variant of the light-gated cation channel ChR2 ( C128S ) ( Berndt et al . , 2009 ) significantly suppressed turns in the absence of the odor ( Figure 3D and Video 4 ) . In addition , we also found that ASH neurons , a pair of nociceptive neurons ( Bargmann , 2006; Kaplan , 1996 ) , were activated by an odor increase ( Figures 3D , 60–90 s ) . Consistently , optogenetic inactivation of ASH neurons during the odor-up phase by the light-driven H+ pump Arch ( Chow et al . , 2010 ) significantly suppressed the increase in the turning rate ( Figure 3F ) . Taken together , these results indicate that two distinct sensory pathways respond to the physiological range of odor concentration changes and opposingly regulate the avoidance behavior ( Figure 3G ) . 10 . 7554/eLife . 21629 . 012Video 4 . Optogenetic activation of AWB neurons . After 60 s without any stimulus , a transgenic animal expressing the bistable variant of channelrhodopsin , ChR2 ( C128S ) , was illuminated with blue light for 3 s to cause sustained AWB activation , and the behavior was monitored for the following 30 s . The video speed is 8x . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 012 Through mathematical analyses of ASH and AWB responses , we found , unexpectedly , that the two neuron pairs decode the temporal dynamics of odor concentration information using different computations . When C . elegans was stimulated with different rates of positive dC/dt , the ASH neurons were always activated rapidly ( Figure 4A , black lines in middle panels ) . These responses peaked soon after the onset of the odor-up phase and were mostly maintained during this phase , which can be approximated by time-differentials of the odor concentrations ( i . e . , dC/dt: Figure 4A , red lines in middle panels; see also Table 1 ) . The time-differential response of sensory neurons is consistent with recent studies ( Lockery , 2011; Schulze et al . , 2015 ) and the general idea that sensory neurons respond in a phasic manner by detecting stimulus changes ( Delcomyn , 1998 ) . A portion of ASH response ( e . g . , decay kinetics during the odor-plateau phase ) was not well fitted by the time-differential equation , which reflected the fact that ASH response is regulated by multiple mechanisms ( see below ) . 10 . 7554/eLife . 21629 . 013Figure 4 . ASH neurons are activated according to dC/dt for initiating turns , and AWB neurons are activated according to the leaky integration of the negative dC/dt for suppressing turns with a dC/dt-dependent delay . ( A ) ASH responses ( middle panels ) and turns ( lower panels ) in response to odor concentration increases from 0 to 1 μM in 45 s ( left most; n = 32 ) , 90 s ( middle left; n = 39 ) , 180 s ( middle right; n = 35 ) or no odor increase ( ‘no odor control’; right most; n = 41 ) are shown . ( Middle panels ) In the three conditions with odor increases , the response patterns of ASH neurons ( the average ± SEM: black lines and gray shadows , respectively ) were fitted by time-differentials of the average of measured odor concentration ( red lines ) , calculated by the rightmost equations . ( Lower panels ) Ensemble averages of the turning rates ± SEM in each 10 s bin were calculated . The original data is shown in the raster plot . Black horizontal dashed lines in lower panels indicate the upper limit of 99% prediction interval of all the turning rates during the odor-zero phase ( t = −60 ~ 0; gray area ) . Red vertical dotted lines indicate the time when each turning rate first exceeded the upper limit of prediction interval . In the first bin of odor-up phase ( indicated by a black horizontal bar in the lower panels ) , the turning rate in the 45 s and 90 s conditions increased significantly compared to the no-odor control ( ***p<0 . 001 , Kruskal-Wallis test with post hoc Steel-Dwass test ) . ( B ) AWB responses and turns in response to odor concentration decreases from 1 μM ( odor-plateau phase; gray area ) to 0 μM ( odor-zero phase ) in 45 s ( left-most; n = 42 ) , in 90 s ( middle left; n = 43 ) , in 180 s ( middle right; n = 48 ) and no odor decrease ( ‘odor-plateau control’; right-most; n = 38 ) . AWB responses to different dC/dt rates were fitted by a leaky integrator equation of negative dC/dt ( red lines ) . Black horizontal dashed lines in the lower panels indicate the lower limit of the 99% prediction interval of the turning rates during the odor-plateau phase . The times when the turning rates became lower than the limit ( red vertical dotted lines ) were delayed when the dC/dt rate was smaller ( *p<0 . 005 and ***p<0 . 001 , Kruskal-Wallis test with post hoc Steel-Dwass test ) . The statistical test was performed in the first 3 bins of the odor-down phase ( a black horizontal bar in lower panels ) . ( C ) , Effect of optogenetic silencing of the AWB response on negative dC/dt . Transgenic animals expressing Arch in AWB neurons were cultivated in the absence ( n = 18 ) or presence ( n = 19 ) of ATR and illuminated with green light during the odor-down phase . *p<0 . 05 , **p<0 . 01 , and ***p<0 . 001 ( Mann-Whitney test ) . In the panels , the gray shading means the period with no odor change , in which the prediction intervals were calculated . All the statistical details are shown in Supplementary file 1 . The following figure supplements are available for Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 01310 . 7554/eLife . 21629 . 014Figure 4—figure supplement 1 . AWB responses were not fitted sufficiently by time-differential equations . The AWB responses are the same as those in Figure 4B . k are described in Table 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 01410 . 7554/eLife . 21629 . 015Figure 4—figure supplement 2 . Estimated intracellular calcium concentrations in AWB neurons calculated from measured ΔF/F0 in Figure 4B . Estimated calcium concentration changes ( black lines ) in response to odor decreases from 1 to 0 μM in 45 s ( left ) , 90 s ( center ) , and 180 s ( right ) were also well-fitted by a leaky integrator equation of negative dC/dt ( red lines ) . The calcium affinity ( Kd = 405 nM ) used for the estimation was according to Akerboom et al . ( 2012 ) . Other parameters are described in Table 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 01510 . 7554/eLife . 21629 . 016Figure 4—figure supplement 3 . ASH responses were partially fitted by the time-integral equations . The ASH responses are the same as those in Figure 4A . Red arrows indicate the same timing with the red vertical dotted lines in Figure 4A . The parameters and goodness of fit are described in Table 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 01610 . 7554/eLife . 21629 . 017Table 1 . Models and parameters used in the fitting of ASH responses with time-differential equations . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 017ConditionsASH , wild-typeASH , odr-3 ( n2150 ) ASH , odr-3 ( n1605 ) ASH , wild-type slow componentDurations of up/down phaseUp 45 sUp 90 sUp 180 sUp 90 sUp 90 sUp 90 sModelX ( t ) =kIX ( t ) =kIX ( t ) =kIX ( t ) =kIX ( t ) =kIdX ( t ) dt=kI−1τX ( t ) I=dC ( t ) dtI=dC ( t ) dtI=dC ( t ) dtI=dC ( t ) dtI=dC ( t ) dtI=dC ( t ) dtParameters used for fitting to ΔF/F0k56 . 1 [μM−1·s]80 . 9 [μM−1·s]121 . 9 [μM−1·s]49 . 1 [μM−1·s]46 . 7 [μM−1·s]2 . 96 [μM−1]τ-----10 . 4 [s]Parameters used for fitting to estimated calcium concentrationk3 . 9 [s]6 . 2 [s]7 . 9 [s]2 . 8 [s]2 . 9 [s]-τ------fmax9 . 49 . 59 . 29 . 18 . 7-fmin0 . 790 . 790 . 770 . 760 . 73-Xbase103 . 8 [nM]92 . 2 [nM]107 . 0 [nM]108 . 2 [nM]109 . 7 [nM]- The behavioral response also changed rapidly: At the onset of the odor-up phase , the turning rates exceeded the upper limit of the 99% prediction interval for the rate during the odor-zero phase ( Figure 4A , red vertical dotted lines in lower panels ) : The 99% prediction interval is a value in which future data ( odor-up phase in this case ) will fall with 99% probability based on the observed data ( odor-zero phase ) ( Montgomery and Runger , 2002 ) . In the 45 s and 90 s odor-up conditions ( left-most and middle left panels ) , the turning rates were significantly different from the no odor control ( see Supplementary file 1 for statistical details ) . It should be noted that , even with the slowest odor concentration increase ( middle right panel ) , at which the overall turning rate was around the threshold , the turning rate increased and surpassed the threshold at the onset of the odor-up phase although no statistical difference was detected . These results suggest that the nociceptive ASH neurons compute the time-differential of odor concentration to rapidly cause an aversive response based on a small change in odor concentration . In contrast , the AWB responses to decreases in odor concentration were time-integral . The responses gradually increased and peaked with a considerable delay , which depended on the dC/dt rates after the onset of the odor-down phase ( Figure 4B , black lines in middle panels ) . These responses could not be fitted by the dC/dt itself but by a leaky integrator equation , in which -dC/dt acted as the input ( Figure 4B , red lines in middle panels and Figure 4—figure supplement 1; see also Table 2 for parameters and Table 3 for goodness of fit ) . In the leaky integrator model , the neuronal response is given by the sum of the past inputs , which decreases exponentially due to leakage . Furthermore , we estimated [Ca2+]i by including a sigmoidal relationship between calcium concentration and fluorescence intensity into the model ( Akerboom et al . , 2012; Kato et al . , 2014 ) . The estimated calcium concentration exhibited a similar pattern to the measured fluorescence change with an estimated basal concentration of about 60–90 nM ( Figure 4—figure supplement 2 and Table 2 ) , similar to that generally reported for resting neurons in general ( ~100 nM ) ( Clapham , 2007 ) . Taken together , these results suggest that AWB neurons compute the leaky integration of the negative dC/dt to temporally integrate the sensory information as the accumulation of [Ca2+]i . 10 . 7554/eLife . 21629 . 018Table 2 . Models and parameters used in the fitting of AWB responses with leaky integrator equations . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 018ConditionsAWB , wild-typeAWB , unc-13 ( e51 ) AWB , unc-31 ( e928 ) AWB , odr-3 ( n2150 ) AWB , odr-3 ( n1605 ) Durations of up/down phaseDown 45 sDown 90 sDown 180 sDown 90 sDown 90 sDown 90 sDown 90 sModeldX ( t ) dt=kI−1τX ( t ) dX ( t ) dt=kI−1τX ( t ) dX ( t ) dt=kI−1τX ( t ) dX ( t ) dt=kI−1τX ( t ) dX ( t ) dt=kI−1τX ( t ) dX ( t ) dt=kI−1τX ( t ) dX ( t ) dt=kI−1τX ( t ) I=−dC ( t ) dtI=−dC ( t ) dtI=−dC ( t ) dtI=−dC ( t ) dtI=−dC ( t ) dtI=−C ( t ) −C ( t−Δt ) ΔtI=−C ( t ) −C ( t−Δt ) Δt ( ∆t = 67 s ) ( ∆t = 66 s ) Parameters used for fitting to ΔF/F0k4 . 6 [μM−1]3 . 6 [μM−1]3 . 0 [μM−1]2 . 6 [μM−1]4 . 9 [μM−1]8 . 7 [μM−1]7 . 7 [μM−1]τ19 . 7 [s]28 . 1 [s]25 . 0 [s]34 . 5 [s]28 . 1 [s]25 . 2 [s]23 . 5 [s]Parameters used for fitting to estimated calcium concentrationk0 . 400 . 370 . 330 . 260 . 441 . 000 . 87τ21 . 1 [s]28 . 5 [s]25 . 0 [s]37 . 3 [s]30 . 7 [s]17 . 7 [s]17 . 7 [s]fmax9 . 710 . 18 . 59 . 68 . 910 . 910 . 4fmin0 . 810 . 840 . 710 . 800 . 740 . 910 . 87Xbase66 . 1 [nM]56 . 7 [nM]89 . 6 [nM]66 . 7 [nM]81 . 2 [nM]41 . 8 [nM]51 . 5 [nM]10 . 7554/eLife . 21629 . 019Table 3 . Parameters and goodness of fit results for mathematical models of ASH and AWB responses . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 019ConditionsASH , wild-typeAWB , wild-typeDurations of up/down phaseUp 45 sUp 90 sUp 180 sDown 45 sDown 90 sDown 180 sNumber of samples ( frames ) used for calculation of BICN = 135 ( t = −60 ~ 75 s ) N = 180 ( t = −60 ~ 120 s ) N = 270 ( t = −60 ~ 210 s ) N = 135 ( t = −60 ~ 75 s ) N = 180 ( t = −60 ~ 120 s ) N = 270 ( t = −60 ~ 210 s ) X ( t ) =kII=dC ( t ) dtk = 56 . 1 [μM−1·s] BIC = −222 . 2k = 80 . 9 [μM−1·s] BIC = −446 . 5k = 121 . 9 [μM−1·s] BIC = −637 . 8k = −58 . 4 [μM−1·s] BIC = −170 . 6k = −73 . 4 [μM−1·s] BIC = −372 . 5k = −67 . 3 [μM−1·s] BIC = −1157dX ( t ) dt=kI−1τX ( t ) I=dC ( t ) dtk = 5 . 8 [μM−1] τ = 12 . 0 [s] BIC = −331 . 1k = 18 . 9 [μM−1] τ = 4 . 4 [s] BIC = −472 . 8k = 11 . 5 [μM−1] τ = 11 . 7 [s] BIC = −804 . 9k = −4 . 56 [μM−1] τ = 19 . 7 [s] BIC = −591 . 9k = −3 . 58 [μM−1] τ = 28 . 1 [s] BIC = −806 . 2k = −3 . 01 [μM−1] τ = 25 . 0 [s] BIC = −1458 The dC/dt rate-dependent delay in AWB activation was correlated with the behavioral responses . The time when the turning rate became lower than the 99% prediction interval of the odor-plateau phase was delayed according to the dC/dt rate , which was associated with statistical differences from the odor-plateau control ( Figure 4B , red vertical dotted lines in lower panels ) . This result suggests that the turning rate is suppressed when the AWB activity exceeds a certain value ( Figure 4B , red horizontal arrows in middle panels ) . Interestingly , even when animals were stimulated with the most shallow odor gradient in which the average turning rate appeared gradually decreased ( middle right panels in Figure 4B ) , turn intervals of each animal could be classified into two groups like pirouettes and runs ( compare Figure 1—figure supplement 1E with 1A ) . This result suggests that , even with subtle changes in odor concentration , the behavioral response of C . elegans does not change gradually , but instead transits between high- and low-turning states . Optogenetic suppression of the AWB neurons with Arch during the odor-down phase significantly affected the transition ( Figure 4C ) . Taken together , these results indicate that AWB neurons regulate the transition from a pirouette to a run based on temporal integration of the negative dC/dt . To further understand whether the neural computations of odor concentration contribute to the choice of migratory direction during decision-making , we performed computer simulations of the odor avoidance behavior ( Figure 5A ) . During pirouettes , the model animal frequently repeated turns and short migrations in random directions , and a pirouette transited to a run when the time-integral of -dCworm/dt over time reached a threshold value . The model produced similar migration patterns to the odor avoidance behavior seen in real animals , in which most of the runs were down the odor gradient ( Figure 5B ) . Interestingly , when we changed the model animal's computation for transition of a pirouette to a run from the temporal integration of dC/dt ( the integration model ) to the simple dC/dt itself ( the differentiation model ) , the directional choice at run initiation significantly worsened ( Figure 5C; p<0 . 001 , Mardia-Watson-Wheeler test ) . This is because , in the differentiation model , the animals initiated runs even when they transiently sensed dC/dt <0 , due to the periodical head swing for example ( Iino and Yoshida , 2009; Yamazoe-Umemoto et al . , 2015 ) . In contrast , animals in the leaky integration model initiated runs only when they sensed dC/dt <0 for a certain period of time , resulting in an appropriate directional choice . In reality , the movement of the animal's anterior end ( where the sensory endings of the ASH and AWB neurons are exposed to the environment ) is more random than that in our model ( Videos 2 and 4 ) , suggesting that the temporally integrating property of AWB neurons may play a significant role in robust directional choice based on a noisy sensory input . 10 . 7554/eLife . 21629 . 020Figure 5 . A computer model reproduced the directional choice in the odor avoidance task in a temporal integration-dependent manner . ( A ) Model of the behavioral transition in 2-nonanone avoidance . During a pirouette , a model animal frequently repeated turns and short migrations . When a model animal initiated a migration away from the odor source and sensed dC/dt <0 , the high-turning state transited to a low-turning state ( i . e . , a run ) when the leaky integrator equation exceeded a threshold . During a run , the animals turned with much lower frequency than in pirouette and at a probability related to dC/dt >0 due to straying away from the original direction . ( B ) Three typical tracks of real ( left ) and model ( right ) animals . As shown in Figure 1A , the odor source is on the left side . ( C ) Histograms of the initial bearings of runs in the model animals . The high-turning-to-low-turning transition was dependent on temporal integration ( left ) or on differentiation ( right ) . p<0 . 001 ( Mardia-Watson-Wheeler test ) . All the statistical details are shown in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 020 Temporally integrating neural activity for decision-making and working memory is considered to be based on recurrent synaptic circuits in vertebrates ( Aksay et al . , 2007; Wang , 2008 ) . To determine whether a synaptic circuit input is required for the temporally integrating property of AWB neurons , we abolished the transmission of synaptic and/or dense core vesicles in most , if not all , neurons via a mutation in unc-13 or unc-31 , the orthologs of mammalian Munc-13 and CAPS ( calcium activated protein for secretion ) , respectively ( Richmond , 2005 ) . In the unc-13 and unc-31 mutants , however , the AWB responses were essentially similar to that in wild-type animals and fitted with the same leaky integrator equations ( Figure 6A and Table 2 ) , indicating that the temporally integrating neural activity occurs within the AWB neuron itself and does not require a recurrent circuit . 10 . 7554/eLife . 21629 . 021Figure 6 . Cell-autonomous computations in AWB neurons . ( A ) The AWB responses of unc-13 ( left ) and unc-31 ( right ) mutants to the odor decreases , which are the same as those shown in the middle left panel of Figure 4B , were essentially similar to those of wild-type animals and fitted by the leaky integrator equations ( n = 31 and 17 , respectively ) . ( B ) The AWB responses of odr-3 ( n2150 ) mutants ( n = 24 ) were fitted by the leaky integrator equation ( red solid line ) with the time interval ( ∆t ) being much longer than that of wild-type AWB ( red dotted line ) . The parameters are described in Table 2 . In the experiments for panels A and B , the behavioral responses were not analyzed because the unc-13 , unc-31 , and odr-3 mutations affect the activities of multiple neurons , including ASH and AWB . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 02110 . 7554/eLife . 21629 . 022Figure 6—figure supplement 1 . Responses of AWB and ASH neurons in odr-3 mutants . ( A ) AWB responses in odr-3 ( n2150 ) ( left panels; the same data as Figure 6B ) or odr-3 ( n1605 ) ( right panels; n = 28 ) mutants were fitted by the right-most equations . When fitting with the leaky integrator equation also used in Figure 4B , τ diverged to infinity ( middle panels; see Table 4 ) . ( B ) ASH activity in odr-3 ( n2150 ) mutants ( left; n = 15 ) and odr-3 ( n1605 ) mutants ( right; n = 12 ) in response to a concentration increase from 0 to 1 μM in 90 s are shown . The ASH activities of odr-3 mutants were also fitted by dC/dt ( red solid lines; red dashed lines are the fit to the wild-type response in Figure 4A ) although the coefficients k were different from those of wild-type animals ( Table 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 022 To reveal the molecular mechanisms of the temporal integration of sensory information in AWB neurons , we performed genetic manipulations . First , we identified the gene required for the calculation of dC/dt . Previous genetic studies suggest that the signaling pathway mediating odor response in AWB neurons is similar to that in mammalian olfactory/photoreceptor neurons . Upon binding to an odorant , a G-protein coupled receptor activates Gα protein , which eventually leads to the opening of cyclic nucleotide-gated cation channels for depolarization ( Bargmann , 2006 ) . ODR-3 , one of the 20 Gα homologs in C . elegans , is expressed in AWB , ASH and a few other pairs of sensory neurons . It is localized at the sensory ending , and its dysfunction causes severe chemotaxis defects , suggesting that ODR-3 transduces the sensory signal from odorant receptor ( Roayaie et al . , 1998 ) . Two alleles of loss-of-function mutations of the odr-3 gene , however , caused gradual activations of AWB neurons , whose peaks were even larger than those of wild-type animals ( Figure 6B and Figure 6—figure supplement 1A , black lines ) , suggesting that ODR-3 may play an inhibitory role in sensory signaling . Interestingly , while the AWB response patterns in wild-type animals were fitted by a leaky integrator equation with a time interval for the input ( ∆t ) =1 s ( solid red lines in Figure 4B and dotted red line in 6B ) , AWB response patterns in odr-3 mutants were better fitted by an equation with ∆t longer than 1 min than by ∆t = 1 s ( solid red line in Figure 6B , Figure 6—figure supplement 1A; see also Table 2 for parameters and Table 4 for goodness of fit ) . Thus , the time-differential property for the sensory input was greatly degraded with the odr-3 mutations while the time-integral property was not affected . This result suggests that the time-differential computation of odor concentration depends mostly on the ODR-3 Gα protein ( see Discussion for details ) . The result also supports the idea that AWB neurons possess a temporally integrating property for sensory inputs . The overall response of ASH neurons , which also express ODR-3 , was partially affected by the odr-3 mutations ( Figure 6—figure supplement 1B ) , suggesting that ODR-3 may play a primary role in sensory signaling in ASH neurons . 10 . 7554/eLife . 21629 . 023Table 4 . Parameters and goodness of fit for mathematical models of AWB responses in odr-3 mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 023ConditionsAWB , odr-3 ( n2150 ) AWB , odr-3 ( n1605 ) Durations of up/down phasedown 90 sdown 90 sNumber of samples ( frames ) used for calculation of BICN = 180 ( t = −60 ~ 120 s ) N = 180 ( t = −60 ~ 120 s ) X ( t ) =kII=−dC ( t ) dtk = 89 . 1 [μM−1·s] BIC = 25 . 1k = 77 . 3 [μM−1·s] BIC = −39 . 2dX ( t ) dt=kI−1τX ( t ) I=−dC ( t ) dtk = 2 . 08 [μM−1] τ → ∞ BIC = −477 . 7k = 1 . 77 [μM−1] τ → ∞ BIC = −578 . 5dX ( t ) dt=kI−1τX ( t ) I=−C ( t ) −C ( t−Δt ) Δtk = 8 . 71 [μM−1] τ = 25 . 2 [s] ∆t = 67 [s] BIC = −645 . 6k = 7 . 73 [μM−1] τ = 23 . 5 [s] ∆t = 66 [s] BIC = −779 . 9 Furthermore , we genetically identified the genes responsible for the time-integral calcium accumulation . In general , [Ca2+]i increases in neurons depend on its influx through the plasma membrane via VGCCs , which can then trigger rapid calcium release from the endoplasmic reticulum ( ER ) via IP3 receptors ( IP3Rs ) and/or ryanodine receptors ( RyRs ) as calcium-induced calcium release ( CICR ) ( Catterall , 2011; Clapham , 2007 ) . The pore-forming α1 subunits of VGCCs are classified into L-type , N/P/Q-type , and T-type subgroups . Previous studies have demonstrated the requirement for EGL-19 and/or UNC-2 , the sole orthologs of L- and N-type channels , respectively , in the responses of sensory neurons of C . elegans ( Busch et al . , 2012; Frøkjaer-Jensen et al . , 2006; Hilliard et al . , 2005; Kato et al . , 2014; Larsch et al . , 2015; Suzuki et al . , 2003; Zahratka et al . , 2015 ) although their precise roles in intracellular calcium dynamics have remained unclear . We found that an egl-19 reduction-of-function mutation ( rf ) ( Lee et al . , 1997 ) significantly affected the AWB response ( Figure 7A ) . Consistently , treatment with the EGL-19 antagonist Nemadipine-A ( NemA ) ( Kwok et al . , 2006 ) exhibited a similar , or possibly even stronger , effect . In contrast , mutations in unc-2 , itr-1 ( the IP3R ortholog ) or unc-68 ( the RyR ortholog ) ( Dal Santo et al . , 1999; Sakube et al . , 1997; Schafer and Kenyon , 1995 ) did not significantly affect the response ( Figure 7A and C left panel ) . These results suggest that the calcium accumulation in AWB neurons mostly depends on influx through the EGL-19 L-type VGCCs , but not other calcium channels . 10 . 7554/eLife . 21629 . 024Figure 7 . Calcium channels are involved in the dynamic regulation of [Ca2+]i in a cell type-dependent manner . ( A and B ) Responses of AWB ( panel A ) or ASH ( panel B ) neurons in strains with genetic and/or pharmacological suppression of N/P/Q-type VGCC UNC-2 , T-type VGCC CCA-1 , L-type VGCC EGL-19 , IP3R ITR-1 , and RyR UNC-68 . lf is a loss-of-function mutation , and rf is a reduction-of-function mutation . Note that loss-of-function mutations of egl-19 and itr-1 are not available due to possible lethality . ( C ) Suppression of the AWB response during the odor-down phase ( left ) and of the initial ASH response ( 10–15 s of the odor-up phase ) ( right ) with the genetic mutation ( M ) and/or the drug treatment ( D ) . ***p<0 . 001 ( Kruskal-Wallis test with post hoc Steel-Dwass test ) . ( D ) egl-19 is also responsible for the slow time-integral component in ASH neurons . Addition of a putative time-integral response using the leaky integrator equation to the transient response in the ‘egl-19 ( drug ) ’ reproduced the ASH response of the naive wild-type animals . Black line: wild-type response shown in panel B; black dashed line: egl-19 ( drug ) response in panel B; red dashed line: the time-integral model of positive dC/dt; red line: sum of the black and red dashed lines . The parameters are described in Table 1 . All the statistical details are shown in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 02410 . 7554/eLife . 21629 . 025Figure 7—figure supplement 1 . ASH response does not depend on synaptic transmission . ASH responses in wild-type ( left; n = 35 ) and unc-13 ( e51 ) ( right; n = 44 ) animals , analysed in parallel , are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 025 In ASH neurons , whose response pattern was also not affected by synaptic connections ( Figure 7—figure supplement 1 ) , the suppression of EGL-19 with NemA mainly affected the later phase of the response , but did not significantly affect the initial response ( Figure 7B and C right panel ) . The mathematical addition of the AWB-like time-integral response to the NemA-treated ASH response resembled the wild-type ASH response ( Figure 7D ) , suggesting that the ASH response consists of both fast and slow components . Because increases in the turning rate occurred at the onset of the odor-up phase ( Figure 4A ) , the fast component may have a major influence on the behavioral response . This fast component was not affected by a loss-of-function mutation in unc-2 or in a T-type VGCC homolog , cca-1 ( Steger et al . , 2005 ) , suggesting that this calcium response is not mediated by the typical VGCCs . Suppressing CICR by mutations of itr-1 or unc-68 , or using dantorolene , a RyR antagonist ( Krause et al . , 2004 ) , partially affected the magnitude , but not the temporal pattern , of the ASH response regardless of NemA treatment ( Figure 7B and C right panel ) . Taken together , these results suggest that the ASH response may be mediated by rapid calcium influx via an unidentified type of calcium channels and by slow influx via EGL-19 L-type VGCCs , both of which are amplified by CICR . In previous studies , ASH response was significantly suppressed by NemA-treatment as well as by the egl-19 ( n582 ) mutation upon stimulation with another repulsive odorant ( 1-octanol ) ( Zahratka et al . , 2015 ) although ASH response was essentially unaffected by the same egl-19 mutation upon stimulation with glycerol ( Pokala et al . , 2014 ) , suggesting that different sensory stimuli may be processed differently in the polymodal ASH neurons .
The behavioral transitions in 2-nonanone avoidance are regulated by ASH neurons for odor increase and AWB neurons for odor decrease ( Figure 8A ) . ASH neurons respond to unfavorable changes in the odor concentration in a time-differential manner and initiate turns rapidly . Such a ‘reflex’-like response is consistent with the nociceptive features of ASH neurons and their synaptic connectivity , whereby the animals respond to various noxious stimuli by inducing turns and reversals ( Bargmann , 2006; Kaplan , 1996 ) . Time-differential properties have been suggested for the transient responses of ASH and other sensory neurons with all-or-none stimuli in C . elegans ( Lockery , 2011 ) . This property has also been reported in a recent study of Drosophila olfactory neurons ( Schulze et al . , 2015 ) and is in agreement with the general idea that dynamic signal changes are mediated by phasic sensory receptors ( Delcomyn , 1998 ) . Mathematically , the ASH activity can also be approximated by the same leaky integrator equation as AWB , whose fitness is higher than the time-differential equation according to the Bayesian Information Criterion ( BIC ) ( Figure 4—figure supplement 3 and Table 3 ) . This is consistent with the recent report that ASH neurons temporally integrate sensory signals over several seconds ( Kato et al . , 2014 ) . However , the onset of responses was not sufficiently reproduced with the model ( Figure 4—figure supplement 3 , red arrows ) . Our result in Figure 7D suggests that the ASH response consists of the time-integral component mediated by EGL-19 L-type VGCCs and the fast and transient component , possibly mediated by as-yet-unidentified channels . Nociceptive ASH neurons may have developed specialized mechanisms to quickly cause aversive responses based on a slight change in the undesirable signal . 10 . 7554/eLife . 21629 . 026Figure 8 . Physiological and molecular models of decision-making by C . elegans during odor avoidance . ( A ) Computations of ASH and AWB neurons during odor avoidance behavior . ( B ) Model of the molecular mechanisms for temporal computation of odor information in AWB and ASH neurons . ( Left ) In AWB neurons , odor decreases likely cause the activation of Gα proteins as an odor-OFF response ( Bargmann , 2006; Usuyama et al . , 2012 ) , where an unidentified Gα positively transduces the signal and ODR-3 inhibits the signal for the time-differential computation . The Gα signaling is transmitted to the cGMP-gated cation channel TAX-2/TAX-4 ( Bargmann , 2006 ) to cause depolarization . Depolarization then triggers calcium influx via EGL-19 at the cell body , which causes the gradual accumulation of [Ca2+]i . ( Right ) In ASH neurons , the depolarization at the sensory ending triggers an unidentified rapid and transient calcium channels , as well as EGL-19 . The calcium influx through these channels is amplified by CICR via RyR ( UNC-68 ) and IP3R ( ITR-1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21629 . 026 In contrast , AWB neurons extract long-term information about favorable changes in odor concentration using the time-integral property , which leads to ‘deliberate’ transitions from pirouettes to runs . This time-integral property has not been characterized previously in the sensory response of C . elegans . The sensory responses of the animals to gradual signal changes have been reported in a few cases , although they were time-differential , stochastic or tonic , rather than time-integral ( Biron et al . , 2008; Kimura et al . , 2004; Luo et al . , 2014 ) . In reality , sensory signals likely change with noise rather than at a constant rate . Moreover , an animal's movement itself causes fluctuations in the sensory input . Thus , it is reasonable to conclude that animals extract sensory information over a longer time window by using a time-integral property , such as that described here for AWB neurons . To trigger behavioral transitions according to neuronal activity , a threshold for temporal integration is required . AIZ interneurons , the major postsynaptic targets of AWB neurons ( White et al . , 1986 ) , have been shown to elicit digital-like excitability and trigger turns upon activation ( Li et al . , 2014 ) , suggesting that AWB neurons inactivate AIZ neurons in an all-or-none manner at a certain threshold . Through genetic analyses , we investigated the molecular mechanisms underlying the computations in AWB and ASH neurons . First , we found that the temporal integration of sensory information occurs within the AWB neuron itself and is not dependent on synaptic connections . To the best of our knowledge , this result is the first experimental demonstration of evidence accumulation for decision-making within a neuron ( see below ) . We also found that the ODR-3 Gα protein possibly plays a time-differential role in the sensory signaling of AWB neurons . The result was surprising because ODR-3 has been considered to play a major stimulatory role in sensory signal transduction , not in its inhibitory modification ( Bargmann , 2006; Roayaie et al . , 1998 ) . Because ODR-3 is homologous to the inhibitory Gαi proteins ( Bargmann , 2006 ) , it is possible that the sensory signal via another unidentified Gα might be downregulated by the activity of ODR-3—such an inhibitory feed-forward loop could compute the time-differential of the input ( Figure 8B , left ) ( Alon , 2007 ) . In ASH neurons , ODR-3 appears to transduce the sensory signal itself ( likely with other Gα ) ; the different roles of ODR-3 in AWB and ASH neurons may be related to the fact that AWB and ASH neurons use different molecular pathways for signal sensation ( Bargmann , 2006 ) . Together with the finding that the time-integral calcium increase in AWB neurons is mainly mediated by EGL-19 L-type VGCCs , these results suggest the following model ( Figure 8B , left ) . A constant change in odor concentration leads to a constant and persistent depolarization with a time-differential activity of ODR-3 Gα at the sensory ending . This depolarization is conducted to the cell body and causes the constant activation of EGL-19 L-type VGCCs and the constant influx of extracellular calcium , which leads to the gradual accumulation of [Ca2+]i . In agreement with this model , L-type , but not N/P/Q-type or T-type , VGCCs are known to be activated continuously during depolarization ( Hille , 2001 ) . This gradual [Ca2+]i accumulation causes a temporal delay and deliberate decision , acting as a low-pass filter to smooth out noisy sensory inputs . By contrast , in ASH neurons , the depolarization causes the activation of EGL-19 L-type VGCCs and other type ( s ) of rapid and transient calcium channels , where a small influx of calcium leads to its amplification with CICR and to a rapid behavioral response ( Figure 8B , right ) . In principle , temporal integration of sensory information for behavioral choice in C . elegans during odor avoidance is analogous to evidence accumulation in perceptual decision-making in mammals . In both cases , animals extract a long-term trend in sensory information by temporally integrating the information for decision-making ( Gold and Shadlen , 2007; Schall , 2001 ) . While AWBs are sensory neurons , their integration property may reflect a central nervous system-like function; C . elegans' sensory neurons are known to possess higher-order functions , such as learning and memory , because of the quite small number of neurons in their nervous system ( de Bono and Maricq , 2005; Sasakura and Mori , 2013 ) . In higher animals , temporally integrating neural activity is required for decision-making as well as for other brain functions , such as working memory , and is generally considered to be mediated by recurrent neural circuits ( Gold and Shadlen , 2007; Schall , 2001; Wang , 2008 ) . This temporally integrating activity has also been suggested to be mediated by intracellular calcium signaling within a cell ( Curtis and Lee , 2010; Loewenstein and Sompolinsky , 2003; Major and Tank , 2004 ) ; however , this had not been demonstrated experimentally . We speculate that single-cell temporal integrators with L-type VGCCs , such as the AWB neurons , may also be involved in decision-making and working memory in higher animals . Interestingly , the odor avoidance behavior of C . elegans is regulated by dopamine ( Kimura et al . , 2010 ) , a neuromodulator involved in mammalian decision-making ( Schultz , 2007 ) . Thus , further genetic analysis of the behavioral choice during odor avoidance behavior in C . elegans may identify previously undescribed evolutionarily conserved molecular mechanisms that are responsible for decision-making .
The techniques used for culturing and handling C . elegans were essentially as described previously ( Brenner , 1974 ) . The C . elegans wild-type Bristol strain N2 , RRID:WB-STRAIN:JD21 cca-1 ( ad1650 ) , RRID:WB-STRAIN:MT1212 egl-19 ( n582 ) , RRID:WB-STRAIN:JT73 itr-1 ( sa73 ) , RRID:WB-STRAIN:CX3222 odr-3 ( n1605 ) , RRID:WB-STRAIN:CX2205 odr-3 ( n2150 ) , RRID:WB-STRAIN:CB55 unc-2 ( e55 ) , RRID:WB-STRAIN:MT7929 unc-13 ( e51 ) , RRID:WB-STRAIN:DA509 unc-31 ( e928 ) , and RRID:WB-STRAIN:CB540 unc-68 ( e540 ) were obtained from the Caenorhabditis Genetics Center ( University of Minnesota , USA ) . In all the behavioral and physiological experiments , young adult hermaphrodites were used . Quantitative analysis of the 2-nonanone avoidance of wild-type animals in the 9 cm agar plate was carried out as previously described ( Kimura et al . , 2010; Yamazoe-Umemoto et al . , 2015 ) . In brief , several adult animals per assay were transferred to the center of a 9 cm NGM agar plate either directly from a standard 6 cm nematode growth medium ( NGM ) plate with the food bacteria OP-50 ( ‘fed’ ) or after a 1 hr starvation on the NGM plate without OP-50 ( ‘starved’ ) . In the following analysis , we used a data set that is a mixture of 50 fed and 50 starved animals . Although we did not find a significant difference between fed and starved animals in 2-nonanone avoidance ( Kimura et al . , 2010 ) , the feeding state ( fed or starved ) could affect some aspects of C . elegans' behavior ( Bargmann , 2006 ) and we wanted to focus on feeding state-independent behavioral aspects of the animals . Two μL of 30% 2-nonanone ( diluted in EtOH ) were put in two spots on the surface of the agar plate ( Figures 1A and 2A ) , and images of the animals during the avoidance behavior were captured at 1 Hz for 12 min by our multi-worm tracking system with a high-resolution camera in a fixed position ( Kawazoe et al . , 2013; Yamazoe-Umemoto et al . , 2015 ) . In this study , we used a CMOS camera CSB4000F-10 ( Toshiba Teli Corp . , Japan ) equipped with a C mount adaptor and a Nikkor 50 mm f/1 . 2 lens ( Nikon Corp . , Japan ) . Because the camera captures the entire area of the 9 cm plate with a resolution of 2008 × 2044 pixels , an animal of length ~1 mm and width ~0 . 05 mm is depicted in ~25 pixels . x−y coordinates of the centroids of the animals in each image were measured by Move-tr/2D software ( Library Inc . , Japan ) , and were further analysed by Excel2010 ( Microsoft ) or R ( The R Project ) . Because the animals did not initiate avoidance during the first 2 min on average ( Kimura et al . , 2010 ) ( Figures 1A and 2B and Figure 2—figure supplement 1E ) , data between 121–720 s were used for the analysis ( Tanimoto et al . , 2017 ) . A pirouette is a period of frequent turns and migrations whose duration is shorter than a threshold value ( Pierce-Shimomura et al . , 1999 ) . The animal's behavioral state in one second was classified as a turn if the absolute value of angle change in migratory vector of the animal's centroid from the previous second ( i . e . , during 1 s ) was larger than 90° or if the migratory velocity was smaller than 0 . 1 mm/s in the following frames after the large angle change . According to this definition , the reverse and the omega turn ( Gray et al . , 2005 ) were recognized as turns . A distribution of turn intervals ( i . e . , migratory durations ) during 2-nonanone avoidance was well-fitted by a sum of two exponentials for shorter and longer intervals ( Figure 1—figure supplement 1A ) . A period at which the numbers of the short and long intervals were equal was 13 . 1 s and determined as tcrit according to the original definition ( Pierce-Shimomura et al . , 1999 ) . Migrations whose turn interval was longer than tcrit were classified as runs , and migrations shorter than tcrit as well as turns were classified in pirouettes . The directions of animal migrations for 1 s were defined in terms of the bearing , B , with respect to the 2-nonanone gradient , where B = 0° indicates migration directly away from the odor source ( i . e . , down the gradient ) and B = ±180° indicates migration directly toward the odor source ( i . e . , up the gradient ) . Bearing at run initiation in salt-taxis by the previous study was calculated from the results of taxis toward NH4Cl ( 56 . 0% ) and biotin ( 55 . 2% ) in Figure 9 of the report ( Pierce-Shimomura et al . , 1999 ) . Calibration curve for 2-nonanone measurement is described in more detail at Bio-protocol ( Yamazoe-Umemoto et al . , 2018 ) . To measure local concentrations of gaseous 2-nonanone in the assay plate , we used a gas chromatograph ( GC ) with a sensitive semiconductor detector , SGVA-N2 , which was optimized for 2-nonanone detection ( FIS Inc . , Japan ) . To make a calibration curve for the measurement , 0 . 36 , 1 . 07 , 3 . 56 , 35 . 6 , 59 . 4 , 97 . 2 , and 200 μL of liquid 2-nonanone ( Wako Pure Chemical , Japan ) were vaporized in a 50 L tank DT-T1 ( FIS Inc . ) , each corresponding to 0 . 04 , 0 . 12 , 0 . 4 , 4 . 0 , 6 . 8 , 11 . 1 , and 22 . 9 μM in the gas phase , respectively . After the volatilization period , 0 . 2 mL of the gas was sampled with a 2 mL plastic syringe with a needle from an outlet of the tank and was immediately injected into the GC . The volatilization periods were determined for each amount of the liquid to maximize the 2-nonanone signal . Synthetic air Alphagaz 1 ( Air Liquide , Japan ) was used as a carrier gas . With 260 s retention time , a single large peak of signal intensity ( mV ) was detected as 2-nonanone signal ( Figure 2—figure supplement 1B ) . The experiments were repeated 3–4 times for each concentration . The correlations between the peak height of the signal and the gaseous 2-nonanone concentration in a log-log plot were well-fitted by two simple regression lines for lower and higher concentrations ( Figure 2—figure supplement 1C; R2 >0 . 999 for both ) . In general , for semiconductor detectors , the correlation between the peak height of the signal and signal concentration in a log-log plot are well-fitted by two simple regression lines for lower and higher concentrations . Measuring odor gradient by gas chromatograph is described in more detail at Bio-protocol ( Yamazoe-Umemoto et al . , 2018 ) . For the odor sampling , a hole of 1 mm in diameter was made through the bottom of the plastic plate and the agar . Because the molecular weight of 2-nonanone ( FW 142 . 2 ) is larger as a volatile compound , it did not leak easily from such a small hole . 1 , 3 , 6 , 9 and 12 min after placing the odor at the two spots , a 2 mL plastic syringe , which is the same type as the one used in the calibration , was inserted into the plate through the hole from the bottom , and 0 . 2 mL of the gas phase was sampled ( Figure 2—figure supplement 1A ) . Each plate was used only once to avoid disturbance of the gradient by the sampling . The sampled gas was immediately injected into the GC for measurement . The concentration of 2-nonanone was calculated from the height of the signal peak according to the regression line for the calibration . For each data point , the measurements were repeated 7–9 times and median and quartile was calculated for the fitting . Fitting the odor gradient and calculation of Cworm are described in more detail at Bio-protocol ( Yamazoe-Umemoto et al . , 2018 ) . The least squares method was used to fit the measured concentration . In the closed plate , the odor concentration asymptotically approaches a constant value . Therefore the measured concentrations were fitted to a phenomenological curve with two exponential saturation functions: C ( x , y , t ) = a ( r1 ) ( 1-exp ( -b ( r1 ) t ) ) + a ( r2 ) ( 1-exp ( -b ( r2 ) t ) ) . r1 and r2 are the distances from the position ( x , y ) on the agar to the two odor sources . The asymptotic concentration a ( r ) and the increasing rate b ( r ) are functions of the distance r such as a ( r ) =a0 exp ( -a1r - a2r2 ) and b ( r ) =b0 exp ( -b1r - b2r2 ) . The assumption that C ( x , y , t ) is given by the sum of the two independent functions is valid for the low concentration regions x > 0 . The fitting parameters a0 = 20 . 68 μM , a1 = 0 . 7355 cm−1 , a2 = −0 . 05408 cm−2 , b0 = 0 . 8384 min−1 , b1 = 0 . 7835 cm−1 and b2 = −0 . 05761 cm−2 were determined by the Levenberg-Marquardt method ( Press et al . , 1992 ) . We consider the measured and the fitted odor gradient as reliable because it is consistent with the fact that the amount of 2-nonanone at the source was apparently reduced by 20–30% after 12 min and with a theoretically calculated simulation ( Yamazoe-Umemoto et al . , 2015 ) . The 2-nonanone concentration at a given temporal and spatial point of an animal’s centroid was calculated from the fitting curve and was designated as Cworm . Turning rate shown in Figure 2D was determined as the relationship between the dCworm/dt during one second of migration and the probability of turning in the next second . For the cell-specific expression of mCherry ( Shaner et al . , 2004 ) , GCaMP3 ( Tian et al . , 2009 ) , ChR2 ( C128S ) ( Berndt et al . , 2009 ) and Arch ( Chow et al . , 2010 ) , str-1 ( Troemel et al . , 1997 ) or srd-23 promoter ( Colosimo et al . , 2004 ) was used for AWB-expression , and sra-6 promoter ( Troemel et al . , 1995 ) was used for ASH-expression . Germline transformation was performed using microinjection ( Mello et al . , 1991 ) . The plasmids and strains used in this study are listed in Supplementary file 2 and 3 . In Figures 3 , 4 and 6B and Figure 6—figure supplement 1 , multiple transgenic lines were used for each type of experiment , and the different lines produced similar results . The representative transgenes ( i . e . , extra chromosomal arrays ) were used for genetic analyses in Figures 6A and 7 . For the OOSaCaBeN ( Olfactory and Optogenetic Stimulation associated with Calcium imaging on Behaving Nematode , or OSB2 ) system , we integrated an auto-tracking microscope system for calcium imaging and optogenetic manipulation with an odor-delivery subsystem ( Busch et al . , 2012; Tanimoto et al . , 2016 ) . Briefly , a wild-type C . elegans ( N2 ) on a NGM plate was placed on a motorized stage HV-STU02 ( HawkVision , Japan ) combined with an upright microscope and illuminated with infrared light . Bright field images of the animal were acquired by a charge-coupled device ( CCD ) camera at 200 Hz to regulate the motorized stage for maintaining the region-of-interest ( ROI ) of a freely moving animal in the center of the view field of the microscope ( ‘ROI-tracking’ , Video 2 ) ( Maru et al . , 2010 ) . A ROI was set around the head neuropil . The system also allowed us to maintain the centroid of a whole animal in the center of the view field ( ‘centroid-tracking’ , Video 4 ) . ROI-tracking was used for calcium imaging with a 20× objective lens , and centroid-tracking was used for wild-type behavioral analyses and optogenetic behavioral analyses with a 10× objective lens . The behaving animal was continuously exposed to an odor flowing from two syringe pumps ( Video 3 ) , which changed the odor concentration according to a predefined program . For odor delivery , 4 μM of 2-nonanone was sampled from the 50 L vaporizing tank with a 25 ml Gastight Syringe ( Hamilton , USA ) . Two such syringes were set on a syringe pump HV-SSP01 ( HawkVision , Japan ) that was controlled by the same program for the auto-tracking . Adapting the gas delivery strategy described previously ( Busch et al . , 2012 ) , one pump was used for 2-nonanone and the other one was for air . The pump speeds were programmed to deliver a constant gas flow of 8 mL/min from the end of the tube , but with varying combinations from each pump to make the temporal gradient of 2-nonanone concentration . For example , when the pump speed of 2-nonanone syringes was changed from 2 ml/min to 5 ml/min , the air syringes went from 6 ml/min to 3 ml/min during the same period . The programs of the pump speeds were designed so that the magnitude of dC/dt was similar to that which animals experienced during the odor avoidance assay in the plate ( Figure 2 ) . The actual concentration of 2-nonanone was monitored at the end of the tube by the same type of semiconductor sensor as the one in the gas chromatograph ( GC ) , and the values were recorded with a PC via a digital multimeter MAS345 ( Mastech , Hong Kong ) before and after the behavioral assays for each day . The sensor was calibrated every day with a similar method as the GC , with calibration concentrations of 0 . 5 , 1 , 2 , and 4 μM . We consider that the measured odor concentrations ( Figures 3 , 4 , 6 and 7 ) closely matched to the actual odor concentration that the animals experienced during the odor avoidance behavior ( Figures 1 and 2 ) because of the following reasons . ( 1 ) The tube end was always maintained at ~1 mm from the freely-moving animal during tracking , and the entire body of an animal was exposed to essentially uniform odor flow without significant diffusion and/or turbulence ( Figure 3—figure supplement 1A and Video 3 ) . ( 2 ) With the flow ( 8 ml/min ) , the animals exhibit robust behavioral response reproducibly through multiple trials ( Figures 3 and 4 ) while the flow itself did not affect the animal's behavior . ( 3 ) The animals likely sense the odor concentration in air phase but not in water phase ( i . e . , agar surface ) because of the high hydrophobicity of 2-nonanone ( a nine carbon ketone ) . The behavior of animals was calculated from records of displacement of the auto-tracking stage and from the position of the ROI or centroid of the view field . For ROI-tracking , the trajectory of an animal’s behavior was wavier than for centroid tracking because the ROI was usually set around the animal's head , which moves in a sinusoidal pattern ( Video 2 ) . To compensate for the wavy pattern , the x−y coordinates for ROI-tracking were calculated as a moving-average for ±10 frames at 10 Hz ( i . e . , ±1 s ) . This gave similar results to centroid-tracking on the quantitative behavioral analysis . The migratory trajectory from either of the tracking methods was sampled at each second ( 1 Hz ) , and a change in the migratory vector for 1 s larger than 90° was recognized as a turn . In the Figures , ensemble averages in each 10 s bin are shown . In Figure 4A and B , we investigated the time when the turning rate changed based on the rate of increase or decrease in odor concentration . In order to investigate the timing , it was necessary to finely set the time window . However , since a turn is an uncommon occurrence ( a turning rate of 0 . 1 is once in 10 s ) , narrowing the time window increased the variation . In order to obtain the same number of turns as the 60 s using a time window of 10 s , six times as much sampling had to be performed . Furthermore , even more samples were required for performing multiple tests . Therefore , we used the prediction interval , a criterion in the field of statistical inference . The 99% prediction interval is an interval in which future data will fall with 99% probability , if it obeys the same probability distribution as the previously observed data ( in this case , odor-zero or odor-plateau phase ) . A 100 ( 1-α ) % prediction interval on a single future observation ( Xn+1 ) from a normal distribution is given by the following formula:x¯−tα2 , n−1s1+1n≤Xn+1≤x¯+tα2 , n−1s1+1n where x¯ is the sample mean , n is the number of previously observed data , tα/2 , n−1 is the 100 ( 1-α/2 ) percentage point of a t-distribution with n−1 degrees of freedom , and s is the sample standard deviation ( Montgomery and Runger , 2002 ) . Using this criterion , we analyzed ‘timing when the unexpected value appears for the first time’ in Figure 4A and B . The details of calcium imaging with the OSB system were previously described ( Tanimoto et al . , 2016 ) . In brief , the sample was exposed to excitation light from a MiLSS ( Multi-independent Light Stimulation System , Aska Company , Japan ) ( Sakai et al . , 2013 ) . The images for GCaMP and mCherry were split and simultaneously captured side-by-side on an EM-CCD camera ImagEM with W-View system ( Hamamatsu , Japan ) . Images were taken at a 32 . 6 ms exposure time and 100 ms sampling interval with 2 × 2 binning . The cell body was tracked off-line with another custom-made program for the centering ( Video 2 ) , and signal intensities of particular regions were measured by ImageJ ( NIH ) . The data from frames where the cell body was not centered were omitted . The signal intensity of the background was subtracted from that of the cell body , and the value was moving-averaged for ±1 frames and further analysed . The average of fluorescence intensity of GCaMP during 1 min before the odor increment or decrement was defined as the baseline F0 . Because ∆F/F0 of GCaMP and the ratio between fluorescence intensities of GCaMP and mCherry ( GCaMP/mCherry: R ) exhibited similar tendencies , and because and the noise level was smaller in ∆F/F0 than in R , the data of ∆F/F0 were used in the figures . In Figure 7 , ∆R was used because the mutations in itr-1 or unc-68 could affect the baseline as well as the response calcium levels of the neurons . Also in Figure 7 , the animals were immobilized with the acetylcholine receptor agonist levamisole for high-throughput analysis , in which multiple animals were stimulated and imaged simultaneously . Even with the levamisole treatment , the responses of AWB and ASH neurons in the naive wild-type animals were essentially similar to those in the freely moving animals ( Figure 4A and B ) . Animals were raised in the presence or absence of ATR according to the previous report ( Kawazoe et al . , 2013 ) , and transferred to an NGM plate on the OSB2 system and maintained under the objective lens by auto-tracking . For ChR2 ( C128S ) experiments in the absence of a 2-nonanone stimulus ( Figure 3D ) , after 1 min without light stimulation , the animal was transiently illuminated with blue light ( 3 s ) for activation through BP460-495 and DM505 with ND25 ( ~0 . 8 mW/mm2 ) . Turning rates of 30–60 s and 65–95 s were calculated as before or after the blue light illumination , respectively . The turns of 60–65 s were not included in the calculation because blue light illumination ( 60–63 s ) appeared to somewhat affect the animals’ locomotion for a few seconds ( Ward et al . , 2008 ) . For Arch experiments in the presence of a 2-nonanone stimulus ( Figures 3F and 4C ) , green light was delivered through BP530-550 and DM570 at ~1 . 0 mW/mm2 , and turning rates were calculated . The optical filters were from Olympus . For the time-differential models of neuronal responses , the following time-differential equation was used:X ( t ) =kdC ( t ) dt where X ( t ) is neuronal response , k is the conversion factor , and C ( t ) is the measured odor concentration . The dC ( t ) /dt was calculated as the central difference of C ( t ) . This equation indicates that the neuronal response X ( t ) responds to the odor gradient dC ( t ) /dt at each time . The value of k was determined by the least squares method to fit X ( t ) to the measured ΔF/F0 in response to the odor gradients . For the time-integral models of neuronal responses , the following leaky integrator equation was used:dX ( t ) dt=kI ( t ) −1τX ( t ) where external input was given by temporal odor change; I ( t ) =dC ( t ) /dt . τ is the time constant of leaky integration . k and τ were determined by the least squares method to fit X ( t ) to the measured ΔF/F0 responded to the odor gradients . This differential equation was numerically integrated by the Euler method with a time-step of 1 s . The initial value was X ( t ) = 0 which corresponds to ΔF/F0 = 0 in the basal state . For odr-3 mutants , on the other hand , external input was I ( t ) =− ( C ( t ) −C ( t−Δt ) ) /Δt in the leaky integrator equation , and the values k , τ , and Δt were determined to fit X ( t ) to the measured ΔF/F0 of odr-3 . Estimation of intracellular calcium concentration in Figure 4—figure supplement 2 was conducted as follows: Since the relationship between fluorescence signals and calcium concentration is non-linear , a change in the neuronal activity to stimulation is properly evaluated , not by the fluorescence intensity of the calcium indicator , but by the calcium concentration itself . Taking the non-linear relationship into account , intracellular calcium concentration [Ca2+] was estimated by the Hill equation; ( F−Fmin ) / ( Fmax−Fmin ) =[Ca2+]h/ ( [Ca2+]h+Kdh ) . The F is the measured fluorescence intensity , Fmin and Fmax are the fluorescence intensities under Ca2+-free and Ca2+-saturated conditions , respectively . The h is the Hill coefficient and Kd is the dissociation constant . For GCaMP3 , the values of h and Kd were reported previously ( Akerboom et al . , 2012 ) . In each experiment , Ca2+ response to stimulation is expressed as the ratio of the fluorescence response to the basal fluorescence intensity F0 , ΔF/F0= ( F−F0 ) /F0 . By solving the Hill equation for [Ca2+] in terms of the fluorescence intensities , the following equation to calculate the intracellular calcium concentration from the measured ratio ΔF/F0 was obtained:[Ca2+]=Kd ( 1+ΔF/F0−fminfmax−1−ΔF/F0 ) 1/h where fmin=Fmin/F0 and fmax=Fmax/F0 are the minimum and maximum fluorescence intensities relative to F0 , respectively . For GCaMP3 , fmax=12fmin since the dynamic range Fmax/Fmin ( i . e . , fmax/fmin ) is reported to be ~12 fold ( Tian et al . , 2009 ) . The time delay of fluorescence response to a calcium concentration change was not taken into account since the temporal resolution of the odor concentration measurement was of the second order , while the association and dissociation time constants of GCaMP3 are of the sub-second order ( Tian et al . , 2009 ) . When X ( t ) corresponds to the calcium concentration , the basal value of X ( t ) in the steady state is not zero since the intracellular calcium concentration is not reduced to zero even in the basal state . Therefore , the leaky integrator equation was generalized as follows:dX ( t ) dt=kI ( t ) −1τ ( X ( t ) −Xbase ) where Xbase corresponds to the basal calcium concentration in the steady state and takes a positive value . For AWB and ASH neurons , unknown model parameters k , τ , fmin and Xbase were determined by the least squares method to fit X ( t ) calculated by the generalized leaky integrator equation to the calcium concentrations estimated from ΔF/F0 . Similar estimation of non-linear property of GCaMP3 has been reported previously ( Kato et al . , 2014 ) . The time-differential and time-integral models reasonably approximated the neural responses under the conditions used in this study . However , with stronger odor concentration changes , input saturation may need to be considered , in which case the input could be put through a logistic sigmoid function for example . The values of the fitting parameters are shown in Tables 1 , 2 , 3 and 4 . The previous algorithms ( Iino and Yoshida , 2009; Yamazoe-Umemoto et al . , 2015 ) were modified as follows to simulate 2-nonanone avoidance behavior ( Figure 5 ) . The parameters for simulation were based on the migratory statistics of real wild-type animals and contained no free parameters unless otherwise indicated . The model animal moved at a speed of 0 . 14 mm/s . In the low-turning state , the model animal moved forward with fluctuations in migratory direction , which was randomly chosen from the Gaussian distribution of −0 . 065 ± 5 . 14° ( mean ± SD ) . The odor signal periodically fluctuated because of the sinusoidal movement of the animal . The position of animal’s anterior end , where the sensory endings of ASH and AWB neurons are located , was calculated as a sine curve along the animal’s track . The amplitude and frequency of the sine curve was 0 . 1 mm and 0 . 5 Hz , respectively ( Kimura et al . , 2004; Shen et al . , 2012 ) . The track of the anterior end was used for the calculation of Cworm . A turn occurred based on the pirouette initiation rate of 0 . 0326/ ( 0 . 200 + exp ( −231 ×dC/dt ) ) +0 . 0260 , which is relatively constant ( ~0 . 03 s−1 ) when dC/dt <0 and increases when dC/dt >0; The dC/dt-dependency in the pirouette initiation rate was determined from the probability of pirouette initiation after 2 s of the step for real animals . The turning duration was 3 s . After a turn , the model animal was in the high-turning state and initiated a migration , whose deviation in direction from the direction just before the turn was randomly chosen from a pool of the measured values in real animals . In the high-turning state , the model animals turned at a constant rate of 0 . 2 s−1 , which results in ~95% of migratory duration shorter than the threshold value tcrit ( 13 . 1 s ) . Therefore , in the high-turning state , most of the migrations were classified as pirouettes . When the model animals happened to migrate down the gradient and experienced dC/dt <0 , their state transited from high- to low-turning according to the leaky integration of dC/dt described in the previous section . When the leaky integration of dC/dt became higher than 0 . 18 , the high turning state was switched to the low turning state . The threshold value 0 . 18 was chosen as a value similar to the one associated with the turn suppression in Figure 4B . For the ‘differentiation model , ’ dC/dt itself was used for the initiation of a low turning state instead of leaky integration , and the high turning state was switched to the low turning state when dC/dt was negative . The simulation was repeated 100 times for each model animal condition . Time was discretized into intervals with ∆t = 1 s . For the experiments with the OSB2 system , the data were obtained on multiple days from approximately 20–50 animals for each condition . We chose this sample number because a large scale behavioral analysis of C . elegans concluded that 20 animals would discriminate single SD in a behavioral phenotype at over 80% power , and 24 ± 14 ( average ± SD ) animals per condition were used in the study ( Yemini et al . , 2013 ) . For Figures 1 and 2 and Figure 1—figure supplements 1 , 100 animals were used because we investigated various aspects of behavior in detail . After the sample acquisition , the data of some animals for Figures 3 , 4 , 6 and 7 were excluded when any of the following problems were found: ( 1 ) trials interrupted by errors in auto-tracking , ( 2 ) animals with too weak intensity of basal mCherry or GCaMP3 fluorescence for off-line tracking , ( 3 ) animals with abnormal sudden transient activation of AWB during odor-zero or odor-up phases ( 8 out of 279 animals tested for AWB ) , or ( 4 ) animals with basal locomotion speeds slower than 0 . 02 mm/s ( average speed ± SD of normal animals was 0 . 15 ± 0 . 04 mm/s ) . Experimental conditions , such as the presence/absence of ATR , light stimulation , odor gradient , or different strains were randomized on a daily basis . A Kruskal-Wallis test with a post hoc Steel-Dwass test was used for multiple comparisons in Figures 3B , 4A , B and 7C , while a Mann-Whitney test was used for single comparison in Figures 1D , 3D , F and 4C , and Figure 2—figure supplement 1D ( right panel ) using R ( The R Project ) or Prism ver . 5 . 0 for Mac OSX ( GraphPad Software , San Diego , CA ) . The Mardia-Watson-Wheeler test was used for Figure 5C and Watson's U2 test was used for Figure 3—figure supplement 1B by using Oriana ver . 3 ( Kovach Computing Service , Wales , UK ) . All the statistical details are shown in Supplementary file 1 . The Bayesian information criterion ( BIC ) was used to assess mathematical model fitting in Figures 4 and 6 . In BIC , the goodness of fit for the model including a penalty term to discourage overfitting is given by the following equation:BIC=Nln ( RSSN ) +Mln ( N ) where N is the number of samples ( frames ) used for the fitting , RSS is the residual sum of squares obtained from fitting by the least squares method , and M is the number of free parameters in a given model , respectively . M = 1 for the differentiation model ( k ) , M = 2 for the leaky integration model ( k , τ ) , and M = 3 for the leaky integration model for odr-3 ( k , τ , ∆t ) . The lower BIC value means a better fitting with a model . | Animals use information from their environment to make decisions , like where to go , what to eat , and with whom to mate . This information may be changing or confusing , and decisions may be quick when the sensory information is clear , or slower when the sensory clues are muddled . Scientists often study this kind of decision-making in monkeys and rodents , but it can be hard to pinpoint the exact decision-making mechanisms because these animals have hundreds of millions neurons in their brains . Studying the mechanisms that underlie decision-making can be easier in a simpler organism with fewer neurons . A tiny roundworm called Caenorhabditis elegans is one such creature , with only 302 neurons . These worms avoid noxious odors , by first wandering around when they detect the odor , and then fleeing . About 80% of the time the worms flee in the correct direction to escape the foul smell . However , it was not clear how the worms decided which direction to flee . Now , Tanimoto , Yamazoe-Umemoto et al . show that the worms choose which direction to move by mathematically calculating information about odor concentrations . In the experiments , a robotic microscope simultaneously measured nerve activity and the worm’s behavior while an odor was presented . Specifically , the amount of calcium in the neurons was measured . The experiments showed that when the worms were wandering to determine which direction to flee the amount of calcium in the neurons changed in proportion to how much the concentration of the odor changed overtime . The experiments suggest that the animals use a mathematical process called integration to add up the changes in the concentration of the odor over time , and when the total reaches a certain threshold the animal successfully moves away from the source . Tanimoto , Yamazoe-Umemoto et al . also identified the gene that enables these calculations . More complicated animals make similar calculations that take into account environmental changes over time when making a decision . Future experiments are needed to determine if more complex animals also use the same mechanism as C . elegans , and whether the same gene is responsible . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | Calcium dynamics regulating the timing of decision-making in C. elegans |
The link between the combined action of neuromodulators in the brain and global brain states remains a mystery . In this study , using biophysically realistic models of the thalamocortical network , we identified the critical intrinsic and synaptic mechanisms , associated with the putative action of acetylcholine ( ACh ) , GABA and monoamines , which lead to transitions between primary brain vigilance states ( waking , non-rapid eye movement sleep [NREM] and REM sleep ) within an ultradian cycle . Using ECoG recordings from humans and LFP recordings from cats and mice , we found that during NREM sleep the power of spindle and delta oscillations is negatively correlated in humans and positively correlated in animal recordings . We explained this discrepancy by the differences in the relative level of ACh . Overall , our study revealed the critical intrinsic and synaptic mechanisms through which different neuromodulators acting in combination result in characteristic brain EEG rhythms and transitions between sleep stages .
During sleep , unique electrophysiological rhythms are observed in the EEG , intracellular recordings , EOG and EMG ( Steriade et al . , 1993a; Niedermeyer and da Silva , 2005 ) . These features form the basis for classification of sleep into different stages: rapid eye movement sleep ( REM ) , and stages N1 ( Stage 1 ) , N2 ( Stage 2 ) and N3 ( Stage 3 ) of non-REM sleep ( Rechtschaffen and Kales , 1968; Iber and Medicine , 2007; Silber et al . , 2007 ) . A typical night of sleep consists of 4–5 sleep cycles of transitions across stages in healthy adults . Each sleep cycle shows progression of sleep stages in the following order , N1 → N2 → N3 → N2 → REM ( Feinberg and Floyd , 1979 ) . Approximately 75% of sleep is spent in NREM stages . There is also a greater amount of N3 ( also termed Slow Wave Sleep , SWS ) earlier in the night , whereas REM stage is relatively dominant during later cycles of sleep ( Aeschbach and Borbely , 1993 ) . Each sleep stage is characterized by specific patterns of the brain electrical activity . The N1 stage consists of slow eye movements and low-amplitude low-frequency ( 4–7 Hz ) EEG activity in humans . The sleep stage N2 is dominated by sleep spindle oscillations at 10–17 Hz ( in humans ) with waxing-and-waning field potentials . Spindles last about 0 . 5–2 s and recur every 2–20 s . Similar events have been studied in non-human mammals , where their frequency may be as low as 7 Hz . During the N3 delta band ( 0 . 2–4 Hz ) , rhythms are found in the EEG ( Rechtschaffen and Kales , 1968; Amzica and Steriade , 1998 ) . This stage is dominated by the slow oscillation , consisting of active ( Up ) and silent ( Down ) cortical states alternating at frequency 0 . 2–1 Hz , that are prominent and visible in the EEG , and in the extracellular and intracellular recordings ( Steriade et al . , 1991 , 1993a; 2001 , Werth et al . , 1996; Timofeev et al . , 2000 , 2001 ) . During Up states , most cells within the cerebral cortex are relatively depolarized and may generate action potentials; during Down states , most cortical neurons are hyperpolarized and do not fire ( Steriade et al . , 1993b; Contreras and Steriade , 1995; Timofeev et al . , 2000 ) . The slow oscillation may nest faster spindles , which are commonly found during Down-to-Up state transitions ( Molle et al . , 2002; Clemens et al . , 2007 ) . During REM sleep , electrical brain activity resembles that of the awake state , including mixed signals with bursts of alpha ( 7–12 Hz ) activity ( Cantero and Atienza , 2000 ) . Precise coordination of different EEG rhythms during sleep is believed to be critical for memory consolidation which manifests in reactivation of specific neural activity patterns – sleep replay – that have been observed in different brain areas including the hippocampus , amygdala , neocortex and striatum ( Nadasdy et al . , 1999; Pennartz et al . , 2004; Euston et al . , 2007; Popa et al . , 2010; Bendor and Wilson , 2012 ) . Neuromodulators , such as acetylcholine ( ACh ) , GABA , histamine ( HA ) , serotonin ( 5-HT ) and norepinephrine ( NE ) , are known to vary significantly during sleep and awake as well as across sleep stages . Compared to the awake state , in N2 and N3 , the levels of ACh and monoamines such as HA and NE are reduced while the level of the inhibitory neurotransmitter GABA is increased ( Aston-Jones and Bloom , 1981 ) ; ( McCormick , 1992; Vazquez and Baghdoyan , 2001; Lena et al . , 2005; Vanini et al . , 2011 ) . During REM sleep , the level of ACh is increased , but monoamines and GABA are reduced ( Baghdoyan and Lydic , 2012 ) . While intracellular and synaptic targets of specific neuromodulators are somewhat known , we still lack a clear understanding of how the orchestrated action of many neuromodulators leads to the very specific types of the brain electrical activity in awake and sleep . In this study , we present a comprehensive computational model of the thalamocortical system implementing effects of neuromodulators and identify the critical intrinsic and synaptic neuronal mechanisms required to explain transitions between sleep stages within an ultradian cycle . We predict that the differences in the temporal dynamics of spindle and delta band oscillations observed in the LFP recordings of cats and mice and ECoG activity of human subjects during SWS ( N3 in humans ) could arise from the relative differences in the level of neuromodulators . We further apply the model to explain electrical activity observed in propofol-induced anesthesia . Our study predicts that even minor changes of neuromodulators may affect the properties of sleep spindles and sleep slow oscillation in NREM sleep , thus possibly affecting the dynamics of memory consolidation .
When the model implemented the intracellular and synaptic changes associated with the putative actions of ACh , HA and GABA , it produced the main features of the different sleep stages observed in human and animal studies . During the awake period , electrical activity in the cortical PY and IN and thalamic RE cells was sparse , that is mean firing was around 2–3 Hz and 4–5 Hz in PY and IN cells , respectively , while the TC neurons primarily remained silent and RE cells were busting intermittently ( Figure 2A and Figure 3A ) . To test the stability of network dynamics , we applied brief thalamic stimulation in the form of a short ( 100 ms ) DC pulse during the awake state . It triggered a transient increase of firing in cortical PY neurons and INs , after which the network returned to the baseline , as observed in experiments with a sensory input during alert awake conditions ( Bruno and Sakmann , 2006; Hengen et al . , 2016 ) . The average activity ( simulated LFP ) during awake period ( Figure 2C ) showed no oscillatory activity or large deflections from baseline , reflecting desynchronized neuronal firing . 10 . 7554/eLife . 18607 . 004Figure 2 . Network activity in awake , N2 , N3 and REM sleep stages . ( A ) Top: Activity of all classes of neurons in the thalamocortical network model ( 500 PY , 100 IN , 100 RE and 100 TC cells ) with the neuromodulators ( ACh , GABA and HA ) varying as shown in the Figure 1B–D . X-axis is time and Y-axis is neuron index . Colors indicate membrane voltage . For awake , N2 , N3 and REM stages , respectively , ACh was 100% , 80% , 50% and 115%; HA was 100% , 40% , 30% and 10%; and GABA was 100% , 115% 130% and 75% . ( B ) Membrane voltages of the representative selected neurons from the network . ( C ) LFP calculated from PY network ( mean voltage of all 500 PY neurons ) . Note spindles and slow oscillations during N2 and N3 , respectably . ( D ) Spectrogram ( based on the sliding time window FFT ) of the LFP shows activity in spindle frequency ( 8–15 Hz ) during N2 , rare alpha-burst events ( 8–15 Hz ) during REM , and slow oscillation ( 0 . 5–1 Hz ) during N3 . DOI: http://dx . doi . org/10 . 7554/eLife . 18607 . 00410 . 7554/eLife . 18607 . 005Figure 3 . Characteristic patterns of electrical activity during different sleep stages . ( A ) Top: Network activity in PY , IN , TC and RE cells shown for a 5 s time window during awake . Y-axis is neuron index . Colors indicate membrane voltage . Bottom: membrane voltages of selected neurons from the network ( neuron #250 in PY , #50 in IN , RE and TC populations ) . ( B–D ) Network activity during N2 ( B ) N3 ( C ) and REM ( D ) sleep for a 10 s time window . Legend is same as for panel A . Note a single localized in space 'alpha-burst' event in D . DOI: http://dx . doi . org/10 . 7554/eLife . 18607 . 005 The N2 stage was characterized in the model by the periods of spindles – thalamically-organized bursts of 7–15 Hz oscillations lasting 0 . 5 to 2 s each and recurring every 3–20 s ( Steriade et al . , 1993a ) . In the model , we observed a transition to N2-like activity following reduction of ACh and HA , and increase of GABA . No external stimulation was applied to initiate or maintain spindles; however , thalamic stimulation could trigger a spindle response . In this state of the model ( see N2 in Figure 2 ) , periods of 7–15 Hz oscillatory activity reappeared spontaneously and lasted for at least 2–5 s as revealed by the spectrogram ( Figure 2D ) . Membrane voltages were hyperpolarized in the PY , IN and TC cells due to the reduction of the ACh ( Figures 2B and 3B ) , similar to the intracellular recording ( Steriade et al . , 1993c ) . In agreement with the prior intracellular data ( Contreras and Steriade , 1996 ) and computational models ( Bazhenov et al . , 2000; Bonjean et al . , 2011 ) , spindles recurred every 3–10 s . When the levels of ACh and HA were reduced and GABA was increased compared to the simulations of the N2 state , N3-like activity appeared in the network ( Figures 2 and 3C ) . In this state , a stereotypical slow oscillation consisting of transitions between Up and Down cortical states at a frequency of around 0 . 5–1 Hz was observed in the LFP , similar to experimental data ( Steriade et al . , 1993c ) . Up states were characterized by the higher spiking activity in all cortical pyramidal cells , interneurons and thalamic reticular neurons , while TC cells showed a depolarized state ( similar to intracellular recordings from higher order thalamic nuclei [Sheroziya and Timofeev , 2014] ) with a few spikes usually at transition from the silent to the active state; the Down states corresponded to the periods of the network silence ( Timofeev et al . , 2000; Bazhenov et al . , 2002; Compte et al . , 2003 ) . As in vivo , faster spindle oscillations were found during Up states of the slow oscillation in the model . To model REM sleep , the level of ACh was increased slightly compared to the awake period , whereas the levels of HA and GABA were reduced . In this network state , there was an overall increase in the cortical activity with very brief periods of 7–15 Hz oscillations that appeared on the background of desynchronized low-frequency cortical firing . The clusters of synchronized firing were somewhat similar to that observed during spindle oscillations . However , they appeared to be more localized in space . In the model , these brief periods of 7–15 Hz synchronized firing ( Figure 3D ) depended on the reduced level of monoamines in the thalamus and higher excitability in the cortex . We propose that these periods of synchronized oscillations correspond to the alpha or mu rhythms ( 7–13 Hz activity ) that have been reported during REM sleep ( Cantero and Atienza , 2000 ) . In order to reveal the specific role of different neuromodulators on network dynamics , the levels of ACh , GABA and HA were varied across a wide range ( ACh: 0 to 120% , GABA: 25–225% , HA: 34 to 100% of the awake stage ) . Power was measured in the spindle ( 7–15 Hz ) and delta ( 0 . 5–4 Hz ) frequency bands , during 10 s periods for each combination of the neuromodulators . In our model , during N3 the delta band activity was dominated by slow oscillations ( 0 . 5–1 Hz ) . Synchronization among network sites was measured using the phase locking value ( PLV ) in a broadband frequency ( 0 . 5–20 Hz ) range , between the LFPs obtained by averaging membrane voltages within five groups of neurons , each comprising 100 PY neurons . Figure 4A shows the summary of the results from all simulations projected to the spindle vs . delta power plane . Several clusters became apparent . Using cluster analysis ( Gaussian mixture model ) with spindle and delta power and PLV as inputs , 10 different clusters were identified ( Akaike information criteria saturated around 10 components ) . The mean values of the clusters are indicated by the different color spheres in Figure 4A and examples of characteristic activity in each cluster are shown in Figure 4C . 10 . 7554/eLife . 18607 . 006Figure 4 . Cluster analysis of the electrophysiological and neuromodulatory spaces . ( A ) Delta ( 0 . 5–4 Hz ) power of the global LFP ( mean voltage of the 500 PY neurons ) is plotted against spindle ( 7–15 Hz ) power for different combinations of ACh , GABA and HA . Different colors represent different clusters identified based on the cluster analysis ( mixed Gaussian model-based clustering was computed with spindle and delta power and Phase Locking Value ( PLV ) as inputs; 10 different clusters were identified ) . The PLV was computed for all pairs taken from 5 LFPs ( mean voltage time series that were obtained by averaging every 100 PY neurons ) . Color spheres indicate the mean values of the clusters . ( B ) Projections of the clusters identified in panel A to the neuromodulator space of the ACh , HA and GABA . Each ellipsoid represents the approximate location of the clusters identified in A . ( C ) Characteristic electrical activity of the thalamocortical network ( top to bottom: PY , IN , TC and RE cells ) for each cluster identified in panels A and B . Cluster index is indicated above each plot and corresponds to the index in the panels A and B . DOI: http://dx . doi . org/10 . 7554/eLife . 18607 . 006 We found that identified cluster states corresponded to either well-defined characteristic types of sleep activity ( e . g . sleep spindles or slow oscillations ) or various mixed states . When spindle and delta power were both low ( e . g . clusters 3 and 4 ) , the activity resembled an awake or REM sleep . In many instances of activity in cluster 4 ( Figure 4C ) , we observed very brief and spatially localized periods of 7–15 Hz oscillations , which were similar to the alpha or mu rhythms recorded during REM sleep . For higher levels of delta power , two groups of clusters were observed: clusters corresponding to low spindle power ( e . g . cluster 2 ) and clusters where spindle power was positively correlated to the delta power ( e . g . clusters 5 , 6 ) . Cluster 2 with low-spindle power but high-delta power corresponded to the stereotypic transitions between Up and Down states ( Figure 4C ) as observed during the sleep slow oscillation in vivo , while clusters with a correlated spindle and delta power ( clusters 5 and 6 ) corresponded to the mixed states , where brief ( cluster 6 ) or long ( cluster 5 ) periods of spindles appeared between periods of slow oscillation . Finally , a distinct cluster 1 corresponded to low-delta power , but high-spindle power . This cluster represented quasi-continuous spindle activity ( Figure 4C ) . The clusters shown in the Figure 4A were identified based on the quantitative characteristics of the electrical activity in the network . In order to link them to specific levels of the neuromodulators , all detected clusters were projected to the neurochemical space . Such projection was performed by fitting an ellipsoid around the means of each cluster in the three-dimensional space of the neuromodulators ( Figure 4B ) . Thus , location of each ellipsoid provides a guide to the range of neuromodulators that produced a specific electrographic pattern . A multivariate ANOVA on the location of the different clusters in the neuromodulator space was significant ( p<10e-6 ) , suggesting that the clusters occupy different regions . Clusters 3 and 4 ( green ellipsoids in Figure 4B corresponding to the awake or REM sleep like electrical activity ) were associated with relatively high ACh levels such as that observed during awake and REM periods . Cluster 2 ( violet ellipsoid in Figure 4B corresponding to the slow oscillation ) was observed in the low ACh conditions , as found during N3 . Clusters 5 and 6 ( sleep stage 2 based on electrical activity ) corresponded to intermediate levels of the neuromodulators . Finally , cluster 1 was observed for lower HA and higher GABA levels , in agreement with the data showing that strong GABA promotes spindle activity ( Parrino and Terzano , 1996 ) . In sum , our analysis confirmed a specific role of neuromodulators in controlling the electrical activity of the thalamocortical network in different sleep and awake conditions , in agreement with past experimental studies ( Baghdoyan and Lydic , 2012; McCormick , 1992 ) . In order to further identify the critical mechanisms responsible for state transitions in the thalamocortical network , we tested the minimal changes of the neuromodulators that were sufficient to generate the essential electrographic characteristics of each sleep stage ( Figure 5 ) . Changes in the neuromodulators were restricted to either thalamus or cortex to isolate the role of different regions . Furthermore , we focused on manipulations of the ACh and HA and not GABA , since in our model GABA modulation ( within the physiological range ) alone was not able to induce any state transitions . We found that in order to induce a transition from awake to N2 , a reduction of HA in the thalamus was necessary but produced very focal and scattered spindles ( Figure 5C ) . Reduction of both ACh and HA in the thalamus ( Figure 5D ) resulted in spindle activity that was similar to that in a full model ( Figure 5A , see also Figure 3B ) . Cortical changes alone did not lead to spindles ( Figure 5B ) but were required to obtain cortical membrane potentials in the physiological range during N2 . We concluded that reduction of ACh and HA in the thalamus , along with reduction of ACh in the cortex , constitute a minimal model needed to simulate the main features of the transition from the awake state to the N2 sleep state . 10 . 7554/eLife . 18607 . 007Figure 5 . Minimal models of sleep state transitions . In all these simulations only specific indicated manipulations were performed . Other model parameters remained unchanged compared to the awake state . The time of the parameter changes is indicated by the bars at the top of each plot . Each plot shows activity of all classes of neurons in the thalamocortical network ( 500 PY , 100 IN , 100 RE and 100 TC cells ) and the LFP . ( A ) All previously indicated manipulations corresponding to the awake to N2 transition ( reduction of ACh and HA , and an increase in GABA ) were performed in the thalamic network; no cortical changes . For awake stage: AChTC , AChRE , ShiftHA , LGABA ( RE-RE , RE-TC ) were 1 . 0 , 1 . 0 , −8 . 0 and 1 . 0 respectively . For N2 stage: AChTC , AChRE , ShiftHA , LGABA ( RE-RE , RE-TC ) were 1 . 25 , 1 . 25 , −3 . 0 and 1 . 15 respectively; LACh , AChPY and LGABA ( IN-PY ) were all fixed at 1 . 0 . ( B ) All previously indicated manipulations corresponding to the awake to N2 transition ( reduction of ACh and increase in GABA ) were performed in the cortical network; no thalamic changes . For awake stage: LACh , AChPY , ShiftHA , LGABA ( IN-CX ) were 1 . 0 , 1 . 0 , -8 . 0 and 1 . 0 respectively . For N2 stage: LACh , AChPY , ShiftHAand LGABA ( IN-CX ) were 1 . 25 , 1 . 25 , -3 . 0 and 1 . 15 respectively; AChTC , AChREand LGABA ( RE-RE , RE-TC ) were fixed at 1 . 0 . ( C ) HA level was reduced in the thalamic network; no any other changes were performed . For awake and N2: only ShiftHA was changed from −8 to −3; all other variables were fixed at awake level . ( D ) HA and ACh levels were reduced in the thalamic network; no any other changes were performed . For awake stage: AChTC , AChRE , ShiftHA , LGABA ( RE-RE , RE-TC ) were 1 . 0 , 1 , 0 , −8 . 0 and 1 . 0 respectively . For N2 stage: AChTC , AChRE , ShiftHA , LGABA ( RE-RE , RE-TC ) were 1 . 25 , 1 . 25 , −3 . 0 and 1 . 15 respectively; LACh , AChPY , LGABA ( IN-CX ) were fixed at 1 . 0 . ( E ) All specific manipulations corresponding to the N2 to SWS transition ( ACh , HA and GABA levels modified ) were performed in the thalamic network; no changes in the cortex . For N2 stage: AChTC , AChRE , ShiftHA , LGABA ( RE-RE , RE-TC ) were 1 . 25 , 1 . 25 , − 3 . 0 and 1 . 15 respectively . For N3 stage: AChTC , AChRE , ShiftHA , LGABA ( RE-RE , RE-TC ) were 2 . 0 , 2 . 0 , −8 . 0 and 1 . 3 , respectively; LACh , AChPY and LGABA ( IN-PY ) were fixed at N2 stage level . ( F ) ACh level was reduced in the cortex; no any other changes were performed . For N2 and N3 stages , LACh was 1 . 0 and 1 . 25; AChPY was 1 . 0 and 1 . 25; all other variables were fixed at N2 level . ( G ) Only AMPA strength was increased in the cortex . For N2 and N3 stages , LACh was 1 and 1 . 25 , correspondingly; all other variables were fixed at N2 level . ( H ) K+ leak current was decreased in the PY neurons . For N2 and N3 stages , AChPY was 1 and 1 . 25 , correspodingly; all other variables were fixed at N2 level . DOI: http://dx . doi . org/10 . 7554/eLife . 18607 . 007 Transition from N2 to N3 could be observed when ACh was reduced in the cortex alone ( Figure 5F ) . In contrast , even when all the other proposed neuromodulatory changes ( reduction of ACh and HA , and increase of GABA ) were applied in the thalamus , the thalamocortical network did not display N3-like activity ( Figure 5E ) , suggesting that appearance of the slow oscillation during N3 requires neuromodulatory action within the cortex itself . Since reduction of the ACh has two main actions in our model , an increase in excitatory connection strength and an increase of leak currents , we examined if either of these changes alone could result in transition to the N3 . Increasing excitatory AMPA conductance alone led to an activated state ( Figure 5G ) , and an increase in the K+ leak conductance alone led to a hyperpolarized quiescent state in the network ( Figure 5H ) . Thus , both actions together were required for transition between the N2 and N3 . We next examined if our thalamocortical model implementing varying levels of neuromodulators can explain LFP data from regions of cat and mouse neocortex across sleep stages ( Figure 6 A , B ) . Surprisingly , visual scoring of cat LFPs revealed that there were only relatively short periods of the S2 ( equivalent to N2; to describe animal experiments we use here terminology common in the animal studies ) compared to the SWS ( equivalent to N3 ) . This result was consistent across animals . The power in the spindle ( 8–15 Hz ) and delta ( 0 . 2–4 Hz ) frequency bands was measured across all channels and the average power was plotted in Figure 6C and D , separately for the cortical and thalamic ( VPL ) recordings . The amplitude of the slow oscillation and spindles were both high during the S2 and SWS , and were reduced in awake and REM . This suggests that the SWS in naturally sleeping cats is marked with the relatively high-spindle activity . Slow ( 0 . 2–1 Hz ) oscillations provided the major contribution to delta band activity during SWS . To further examine the interaction between sleep spindles and slow oscillations , we plotted the power in the spindle vs delta frequency bands across different cortical LFP electrodes ( Figure 6E ) . We found a strong correlation between slow oscillation and spindle activity . In mice ( Figure 6F , G ) , a similar relationship between spindle and delta power was observed across recordings from five animals ( 3 channels per mouse ) ; a positive correlation was observed in majority of the channels ( 9 channels out of 15 channels showed significant correlation of which 7 were positively correlated [r ranged from 0 . 17 to 0 . 43; p<0 . 05] and 2 were negatively correlated [r was −0 . 41 and –0 . 43; p<0 . 05] ) . Recordings from four out of five mice revealed significant correlations . In one mouse ( Figure 6F–G ) , frontal and posterior somatosensory electrodes showed a significant positive correlation between slow oscillations and spindle activity . 10 . 7554/eLife . 18607 . 008Figure 6 . LFP recordings from cats and mice show positively correlated spindle and delta power during NREM sleep . ( A ) . In vivo LFP recordings from different locations in the cortex ( black lines ) and the thalamus ( red lines ) of a non-anesthetized cat . ( B ) Locations of the recording electrodes . ( C , D ) Power at delta frequency ( 0 . 2–4 Hz ) ( C ) and spindle frequency ( 8–15 Hz ) ( D ) measured from 5 s time windows ( 1 s sliding window ) . ( E ) Spindle power is plotted against delta power for different channels . PCA was applied prior to computing the power and the principal components correlating strongly to EOG were removed; channel data were recomputed from remaining components to remove eye movement artifacts . Each dot represents the power in spindle and delta bands measured from a 30-s period of cat recordings . Note a positive correlation between delta and spindle power when combining S2 and SWS epochs . ( F ) Top: Hypnogram of recordings in mice . Bottom: In vivo LFP recordings from different cortical locations in mice . Note lack of S2 periods in mice . ( G ) Spindle power is plotted against delta power for different channels in mice . Each dot represents the power in spindle and delta bands measured from a 30-s window of data . There is a positive correlation between delta and spindle power during SWS epochs . DOI: http://dx . doi . org/10 . 7554/eLife . 18607 . 008 Our analysis of the neuromodulator space ( Figure 4 ) suggests that a moderate reduction of ACh would lead to a network state with correlated variations in spindle and delta power ( clusters 4 , 5 and 6 in Figure 4A ) . Thus , in order to replicate our animal data ( cat and mouse LFP activity ) , a computational model with a relatively high level of ACh during SWS ( reduction to 75% of the baseline awake level ) was required ( Figure 7A ) . In contrast , in our control simulations of SWS ( N3 ) , the ACh was reduced to 57% of the baseline ( Figures 2 and 3 ) . Using the new model with the higher level of the ACh during SWS , we found a relative increase in the spindle activity during the SWS ( Figure 7A ) and a strong positive correlation between slow oscillations and spindle activity ( Figure 7B , right ) , as was observed in the animal LFP data ( Figure 7B , left ) . 10 . 7554/eLife . 18607 . 009Figure 7 . Model predicts temporal characteristics of the LFP activity in cats and mice . ( A ) The network model implementing moderate ACh reduction ( to 75% of the baseline awake level ) reproduces sleep stage transitions with pattern of activity similar to that recorded from cats and mice . ( B ) Spindle power ( 7–15 Hz ) averaged across all electrodes is plotted against delta band power ( 0 . 5–4 Hz ) for different sleep stages . PCA was applied to remove eye movement artifacts in cat recordings . Right: Spindle power vs . delta power based on the model simulations . Average activity within groups of 100 neurons was used as an estimate of the LFP . Each dot represents the power in spindle and delta bands measured from 2 s windows of data obtained from the model simulations . DOI: http://dx . doi . org/10 . 7554/eLife . 18607 . 009 Next , ECoG recordings from humans during different sleep stages were examined for the relationship between delta ( 0 . 01–2 Hz ) and spindle ( 9–17 Hz ) power ( Figure 8A , B ) . ( Note , these slightly different frequency bands are conventional in human studies . ) In contrast to the pattern found in cats and mice , there was a significant negative correlation between the spindle and the delta power observed in successive 30 s epochs obtained during the N2 and N3 ( Figure 8C and D ) in humans . Generally , the slow oscillation was higher during N3 than during N2 , whereas spindle activity was higher in N2 than in N3 , resulting in a negative correlation across epochs . The average Pearson’s correlation coefficient across 42 channels was −0 . 2595 ( sd = 0 . 2281 ) ( Figure 8E ) . Fifteen out of 42 electrodes showed a significant negative correlation ( p<0 . 001 ) , and no electrodes showed a significant positive correlation . We also observed periods of increased power in the 7–15 Hz frequency band during REM and awake suggesting transient periods of alpha ( or mu ) rhythm . To simulate the negative correlation between sleep slow oscillation and spindle activities , we again applied results from our analysis of the neuromodulator space ( Figure 4 ) . When the ACh level was reduced to 45% of the baseline during the N3 ( compared to 57% used in the control model , see Figures 2 , 3 ) , the network activity revealed a negative correlation between the delta and spindle frequency bands ( Figure 8 F and G ) . These findings suggest that a relatively low level of the ACh during the N3 could possibly explain the negative correlation between the slow oscillation and spindle activities as observed in human ECoG recordings . 10 . 7554/eLife . 18607 . 010Figure 8 . ECoG recordings from a human patient show negatively correlated spindle and delta power during NREM sleep . ( A ) Patient MG29 had a 64-contact grid and 4 strips implanted over the left hemisphere . After rejection of electrodes that either had poor contact or showed significant epileptic interictal activity , 42 electrode contacts remained for analysis . ( B ) Recording of patient for 9 hr beginning in the evening , during which patient was both awake and asleep . Shown are LFP over four grid electrodes , along with the right EOG channel . Beneath that is the sleep staging done in 30 s epochs . Staging was scored using information from scalp , EOG and intracranial electrodes . Data from Awake , REM , N2 and N3 were used for further analysis . A significant decrease in overall LFP amplitude was seen in REM and occasionally during awake . ( C ) Delta and spindle power over the same period . A Fast Fourier Transform with a window size of 30 s was done for delta ( band 0 . 01–2 Hz ) and spindle ( 9–17 Hz ) band frequencies to obtain their power . All 42 electrodes were averaged and are plotted here . ( D ) Delta vs . spindle bands for 30 s time epochs were plotted for individual electrodes and separated based on sleep staging . There is a negative correlation between delta ( dominated by slow oscillation ) and spindle activities for these electrodes when combining N2 and N3 epochs . ( E ) A histogram of Pearson's R correlation coefficients between delta and spindle power during N2 and N3 across all 42 electrodes . ( F ) The computational model implementing significant ACh reduction ( to 45% of the baseline awake level ) reproduces sleep stage transitions with pattern of activity similar to that recorded from humans . ( G ) Spectral analysis of model activity . Power at delta frequency ( 0 . 2–4 Hz , red ) and spindle frequency ( 8–15 Hz , blue ) were measured from FFT obtained with non-overlapping sliding 2 s windows , similar to the analysis shown in C of actual recordings . ( H ) Inverse correlation of delta and spindle power in the model during combined N2 and N3 activity , similar to human recordings in D . DOI: http://dx . doi . org/10 . 7554/eLife . 18607 . 010 We next applied our model to explain the features of the electrical activity induced by propofol anesthesia . In humans , propofol has shown to increase 8–15 Hz oscillations frontally and to reduce alpha oscillations posteriorly ( Murphy et al . , 2011; Purdon et al . , 2013 ) . Propofol application also increased the slow oscillations in the delta ( 1–4 Hz ) frequency range across all regions . The onset of the loss of consciousness was closely matched by the increase of slow ( <1 Hz ) and 8–15 Hz oscillations ( Purdon et al . , 2013 ) . We examined if our model could capture these changes in the EEG due to the action of propofol . Propofol is a GABA agonist and is known to increase the decay time constant of the GABA-A mediated inhibitory post-synaptic potential ( Kitamura et al . , 2003 ) . Multiple lines of evidence suggest that propofol also acts by reducing the action of ACh and HA . Indeed , propofol is shown to decrease the level of ACh in frontal cortex ( Kikuchi et al . , 1998 ) and to attenuate ACh receptor responses ( Flood et al . , 1997; Murasaki et al . , 2003 ) . Increasing ACh transmission prevents the action of propofol in humans ( Meuret et al . , 2000 ) . Propofol inhibits the activity of the tuberomammillary nucleus ( Nelson et al . , 2002 ) , although its action on ventrolateral preoptic nucleus neurons ( Liu et al . , 2013 ) leading to reduction of HA . When these effects of ACh and HA were implemented in our model , as described before , and the decay time constant of GABA was increased , we observed a large increase in slow oscillations as well as oscillatory activity in 8–15 Hz frequency band ( Figure 9A ) . In comparison to the model of natural sleep ( Figure 9C ) , there was elevated 8–15 Hz power in the propofol condition ( Figure 9D ) . This is consistent with our recordings from cats ( not shown ) , where we found that 8–15 Hz power under propofol was high compared to natural sleep or ketamine-xylazine anesthesia . These results are also consistent with the previous computational models of propofol application , where the increase in decay time constant of IPSPs led to increase in 8–15 Hz oscillations ( Ching et al . , 2010 ) . However , in our model , when only the time constant of the GABA-A IPSP was increased compare to the awake condition , the model failed to generate slow oscillations or the 8–15 Hz synchronized activity ( Figure 9B ) . Overall , our study suggests that the known action of propofol in vivo may require its indirect effect on both the ACh and HA neuromodulatory systems . 10 . 7554/eLife . 18607 . 011Figure 9 . Mechanism of action of propofol induced anesthesia . ( A ) Model of propofol anesthesia implementing inhibition of ACh and HA release and increase in the decay time constant of the GABAergic IPSPs . From top to bottom: Spatio-temporal pattern of activity in PY neurons , average activity in PY network ( simulated LFP ) , band-filtered ( 7–15 Hz ) PY LFP . The GABA decay time constant was increased by 150%; ACh and HA were reduced to the same level as in simulations of the natural sleep in cats . Note significant amount of spindle activity . ( B ) The model where only the GABA time constant was increased by 150% . Note an almost complete absence of spindle activity . ( C ) Network activity during simulated natural SWS in cat model . Note decreased spindle activity compared to the propofol simulations in panel A . ( D ) Power in 0 . 2–4 Hz and 8–15 Hz bands for all three conditions . The network was divided into 10 groups of 50 neurons . Membrane voltages were averaged within each group , then FFT was used to estimate power in each group . Bar height indicates average across 10 groups , error bar indicates standard deviation across 10 groups . DOI: http://dx . doi . org/10 . 7554/eLife . 18607 . 011 Experimental studies suggest that the concentration of extracellular GABA in the neocortex is higher during NREM sleep and it is reduced during REM sleep compared to the awake state ( Vanini et al . , 2012 ) . Increase of extracellular GABA may reflect an increase of synaptic inhibition . Indeed , during slow wave sleep , synaptic excitation and inhibition are balanced ( Rudolph et al . , 2007; Haider et al . , 2006 ) . Since decrease of ACh during stage N2/N3 sleep is known to increase excitatory connections ( Gil et al . , 1997 ) , synaptic inhibition and thus phasic GABA release may be also increased , which would then be reflected in elevated extracellular GABA . The effect of increase of synaptic inhibition was implemented in our baseline model as reported above . Nevertheless , the origin of the change in the extracellular GABA is still not fully understood . Increase of extracellular GABA may increase tonic inhibition but not necessarily be associated with increase of the phasic GABA release . Further , several studies revealed co-release of GABA with ACh and HA ( Saunders et al . , 2015; Yu et al . , 2015 ) which suggests a decrease of GABA during NREM sleep while the observations from microdialysis experiments ( Vanini et al . , 2012 ) suggest an increase of GABA during NREM sleep . Therefore , in our study , we also tested models with ( a ) no change in the level of the GABA release; ( b ) increase of tonic inhibition reflecting increase of the extracellular GABA , based on the observations from microdialysis experiments ( Vanini et al . , 2012 ) ; and ( c ) tonic inhibition proportional to the ACh and HA levels , based on the evidence of the co-release of GABA with ACh ( Saunders et al . , 2015 ) and HA ( Yu et al . , 2015 ) . In all these conditions ( Figure 1 ) , we found transitions between sleep stages similar to those reported in the baseline model . However , the model implementing co-release of GABA with ACh had poorly formed spindle activity during N2 and elevated alpha activity during REM sleep . Future experiments are required to distinguish between changes in tonic and phasic inhibition to match specific features of the synchronized oscillatory activity across sleep stages . 10 . 7554/eLife . 18607 . 012Figure 10 . Effect of GABA on the network state transitions . ( A ) Both phasic and tonic GABA conductances were fixed for entire simulation . Note the state transitions similar to the baseline model ( Figure 2 ) . ( B ) Tonic GABA conductance ( conductance of the miniature IPSPs ) was varied based on the measured GABA levels in microdialysis experiments ( mIPSPs were modified as 115% , 130% and 75% of awake stage for N2 , N3 and REM stages , respectively ) , while phasic GABA conductance was remained fixed . ( C ) Assuming co-release of ACh with GABA , tonic GABA conductance was proportionally varied with ACh concentration ( mIPSP for GABA conductance was scaled as 85% , 70% and 125% of awake stage for N2 , N3 and REM stages respectively ) . Transition between sleep stages was observed in all these conditions , but spindles were less synchronous in C; in A and B alpha-bursts were less common . DOI: http://dx . doi . org/10 . 7554/eLife . 18607 . 012
The functional state and the patterns of electrical activity of human and animal brain are continuously influenced by a broad range of neuromodulators ( McCormick , 1992; Steriade et al . , 2001; Baghdoyan and Lydic , 2012 ) . Changes in the level of neuromodulators have been correlated with sleep induction as well as transitions between characteristic EEG sleep patterns ( McCormick , 1992; McCormick and Bal , 1997; Steriade et al . , 2001 ) . However , the mechanisms whereby specific cellular and synaptic changes triggered by the combined action of neuromodulators result in transitions between sleep stages and their precise electrical characteristics remain poorly understood . In this new study , we used biophysically realistic computational models of the thalamocortical system to identify the minimal set of cellular and network level changes , linked to the putative action of the neuromodulators , that is sufficient to explain the characteristic neuronal dynamics during sleep as well as the transitions between primary sleep stages and from sleep to awake state . Our study predicts a critical role of the neuromodulators in controlling the precise spatio-temporal characteristics of neuronal synchronization that manifest in major sleep rhythms – sleep spindles and slow oscillations . Our results are consistent with in vivo intracellular and LFP recordings from animals and ECoG data from human subjects , as reported in this study . Previous studies identified intrinsic and synaptic mechanisms involved in generating specific types of sleep rhythms . The survival of the slow oscillation ( <1 Hz global electrical activity found during deep ( N3 ) sleep ) after extensive thalamic lesions in vivo ( Steriade et al . , 1993b ) , its existence in cortical slice preparations ( Sanchez-Vives and McCormick , 2000 ) , and the absence of the slow oscillations in the thalamus of decorticated cats ( Timofeev and Steriade , 1996 ) point to the primarily intracortical origin for this rhythm . ( Note that thalamus can be actively involved in synchronization of the SO [Lemieux et al . , 2014] ) . In vivo and in vitro studies suggest that the minimal substrate accounting for spindle oscillations ( 7–15 Hz recurrent oscillatory activity found during stage 2 ( N2 ) of NREM sleep ) is the interaction between thalamic reticular and relay cells ( Steriade and Deschénes , 1984; Steriade et al . , 1985; Steriade and Llinas , 1988 , 1990; von Krosigk et al . , 1993 ) . Previous computational studies proposed a minimal set of the mechanisms sufficient to model sleep spindles and slow oscillations ( Destexhe et al . , 1996; Bazhenov et al . , 1998 , 1999; Timofeev et al . , 2000; Bazhenov et al . , 2002; Compte et al . , 2003; Hill and Tononi , 2005; Bonjean et al . , 2011 ) . An increase in the K+ leak current was identified as a critical component for the transition between awake and slow wave sleep ( Bazhenov et al . , 2002; Hill and Tononi , 2005 ) , and it was predicted that synchronization of the slow oscillation depends on cortico-cortical connections ( Hill and Tononi , 2005 ) . Several phenomenological and reduced mathematical models have been proposed to test the effects of neuromodulators on the sleep-wake cycle ( McCarley and Hobson , 1975; Robinson et al . , 2011; Schellenberger Costa et al . , 2016 ) . Nevertheless , none of these previous models examined transitions among all major sleep stages based on biophysical mechanisms . In this study , we found that reduction in HA and ACh in the thalamus was required for induction of N2-like activity , characterized by recurrent spindles with characteristics consistent with animal and human data . Other neuromodulatory changes ( such as change of GABA conductance ) were not sufficient alone to induce N2-like activity , but played a role in determining the power and the synchrony of oscillations . Reduction in the ACh levels within the cortical network was identified to be a minimal requirement for the transition from the N2 ( spindles ) to the N3 ( slow oscillation ) . Our study of the minimal models revealed the importance of the cell-type-specific action of the neuromodulators . Indeed , in order to model sleep stage transitions , the crucial changes were in HA , primarily acting on thalamic neurons , and in ACh , primarily acting within a cortex . All the network rhythms reported in this study resulted from intrinsic ( autonomous ) dynamics of the thalamocortical network; no external stimulation was applied to entrain any of the oscillatory activities in the model . By parametrically varying the levels of the ACh , HA and GABA , we observed a wider range of activities in the thalamocortical network than have been previously reported in the computational models , but which do correspond to those that have been observed experimentally . While we found network states dominated by either spindles or slow oscillations , under certain conditions ( such as only moderate reduction of the ACh level ) we also observed mixed states combining these two major rhythms . Such mixed states may occur in vivo under normal conditions ( Aeschbach and Borbely , 1993; Muller et al . , 2006 ) ; our recent behavior study suggests that the strengths of the phase amplitude coupling between the spindle and the slow oscillation during NREM sleep correlates with memory consolidation ( Niknazar et al . , 2015 ) . We speculate that the mixed states may become prevalent in some pathological conditions such as in Alzheimer’s disease where sleep is altered and changes to the neuromodulatory system are reported ( Wulff et al . , 2010 ) . Furthermore , our results on the putative relationship between levels of neuromodulators and characteristic sleep electrical activities could be potentially relevant to understanding the changes observed with aging . Indeed , the level of the HA in cerebrospinal fluid is known to be elevated with aging ( Prell et al . , 1988 ) , where spindle power is known to be reduced ( Crowley et al . , 2002 ) . In our model , the HA level is one of the major predictors of spindle power . Thus , we can speculate that an increase of HA with age could contribute to the decline of the spindle power . In our study , the effects of neuromodulators were limited by changing the following four model properties – maximal conductances of the K+-leak , Ih , GABA and AMPA currents . We found that specific combinations of changes of these core parameters were sufficient to cause transitions between sleep stages . We also tested a more detailed model , implementing the effects of ACh and HA on the other K+ currents as well as on the persistent Na+ and potassium M-currents ( Constanti and Galvan , 1983; McCormick and Williamson , 1989; Mittmann and Alzheimer , 1998 ) . We observed that this extended model qualitatively captured results similar to the minimal model . In contrast , after fixing the K+-leak , Ih , GABA and AMPA currents but leaving the action of neuromodulators on the other channels , the model failed to show transitions between sleep stages , suggesting that the cellular and synaptic properties identified in our study are critical in determining patterns of electrical activity in the thalamocortical network . In developing our model , we assumed that the levels of ACh , GABA and HA were different between N2 ( Stage 2 ) and N3 ( Stage 3 or SWS ) . These assumptions are consistent with data on spiking activity in the Basal Forebrain region ( Aston-Jones and Bloom , 1981; Szymusiak et al . , 2000 ) , suggesting a difference in ACh between the N2 and N3 . Another assumption used in this study was that ACh modulates the strength of the excitatory AMPA synapses in the intracortical and thalamocortical circuits . Indeed , ACh suppresses the spread of excitation in vivo ( Kimura et al . , 1999 ) and reduces evoked responses ( Chauvette et al . , 2012 ) . In slice experiments , ACh activation of muscarinic receptors suppressed both intracortical and thalamocortical excitatory connections , while nicotinic activation led to suppression of the intracortical connections and enhancement of the thalamocortical connections ( Gil et al . , 1997; Hsieh et al . , 2000 ) . By comparing recordings from animals ( cats and mice LFP ) and human ECoG data across sleep stages , we found a significant difference in the interaction between major sleep rhythms ( spindles and slow oscillation ) during NREM sleep . In cats and mice , there was a strong positive correlation between spindle and delta power , while in humans there was a negative correlation between oscillations in these frequency bands . This difference may reflect the prominence of N2 ( stage 2 ) in humans , in whom N2 has the longest duration of any sleep stage . In contrast , in cats , N2 is very rare compared to N3 . In humans , N2 is mainly characterized by spindles with relatively short periods of slow oscillation or isolated K-complexes . ( K-Complexes are large downward deflections in EEG often followed by spindle activity and is one of the electrophysiological markers of N2 ( Loomis et al . , 1937; Amzica and Steriade , 1997; Cash et al . , 2009 ) . In N3 , spindles continue to occur in conjunction with the slow oscillation , but the EEG is dominated by slow activity . Our model predicts that this inter-species difference can be explained by a difference in the relative level of ACh during NREM sleep . This result is consistent with microdialysis measurements in cats where ACh was present during SWS albeit at low concentrations compared to wakefulness ( Marrosu et al . , 1995 ) . Surprisingly , we found that even minor changes of neuromodulators may lead to significant changes in the characteristics of brain electrical activity during sleep . Our model predicts that fluctuations of the level of ACh over time may explain the appearance of short periods of slow oscillations in N2 in humans ( Achermann and Borbely , 1997 ) or modulate gamma activity during Up-states of the slow oscillation ( Mena-Segovia et al . , 2008 ) . We hypothesize that transient activities such as K-Complexes during N2 sleep or PGO waves during REM stage may arise from fluctuations in ACh and/or other neuromodulators . Ultimately , our study predicts that transient changes in the level of neuromodulators may affect the interaction and characteristic properties of sleep rhythms and may thus affect functional outcomes of sleep , such as the impact of sleep on memory and learning . | There are several stages of sleep that cycle repeatedly through the night with each producing distinctive patterns of electrical activity in the brain . It is thought that these patterns may help us to remember things that have happened throughout the day . Cells in parts of the brain called the hypothalamus and the brainstem control transitions between sleep stages . They regulate the release of chemicals known as neuromodulators in many parts of the brain , including the cortex and thalamus , which play the roles in memory and learning . Researchers now know how the neuromodulators influence the properties of individual brain cells . However , it is not clear how coordinated action of many neuromodulators result in the patterns of electrical activity seen in the brain during each stage of sleep . Krishnan et al . used a computer model to investigate how three of these neuromodulators – acetylcholine , histamine and GABA – shift electrical activity in the brain between sleep stages . The computer model was able to recreate the network of brain cells in the cortex and thalamus and how this network responds to the changes in the levels of neuromodulators . The study found that simultaneous and balanced changes of acetylcholine , histamine , and GABA work together to shift the brain between the stages of sleep and to initiate patterns of the brain electrical activity specific to the different sleep stages . Krishnan et al . predict that the relative differences in the level of acetylcholine in the brains of humans , cats and mice may explain why different species have different patterns of electrical activity during sleep . The study also found that an anesthetic drug called propofol may induce sleep-like patterns of electrical activity in the human brain by affecting the levels of all three of the neuromodulators . More studies are needed to look at how the networks of cells in the cortex and thalamus communicate with the brainstem , and how changes in the levels of neuromodulators affect memory and learning . | [
"Abstract",
"Introduction",
"Results",
"Discussion"
] | [
"neuroscience"
] | 2016 | Cellular and neurochemical basis of sleep stages in the thalamocortical network |
Most bacteria use an indirect pathway to generate aminoacylated glutamine and/or asparagine tRNAs . Clinical isolates of Mycobacterium tuberculosis with increased rates of error in gene translation ( mistranslation ) involving the indirect tRNA-aminoacylation pathway have increased tolerance to the first-line antibiotic rifampicin . Here , we identify that the aminoglycoside kasugamycin can specifically decrease mistranslation due to the indirect tRNA pathway . Kasugamycin but not the aminoglycoside streptomycin , can limit emergence of rifampicin resistance in vitro and increases mycobacterial susceptibility to rifampicin both in vitro and in a murine model of infection . Moreover , despite parenteral administration of kasugamycin being unable to achieve the in vitro minimum inhibitory concentration , kasugamycin alone was able to significantly restrict growth of Mycobacterium tuberculosis in mice . These data suggest that pharmacologically reducing mistranslation may be a novel mechanism for targeting bacterial adaptation .
The long treatment duration of regimens for tuberculosis are thought in part to be due to phenotypic resistance ( tolerance ) of a subpopulation of genetically susceptible bacteria to antibiotic-mediated killing ( Gold and Nathan , 2017; Maisonneuve and Gerdes , 2014; Lewis , 2010; Wakamoto et al . , 2013 ) . A better mechanistic understanding of antibiotic tolerance is required to rationally devise regimens that may reduce tuberculosis regimen duration . Multiple mechanisms have been proposed for how mycobacteria tolerate antibiotics , including non-replicating persistence ( Saito et al . , 2017; Gold and Nathan , 2017 ) , antibiotic efflux ( Adams et al . , 2011 ) and phenotypic variation in cell-size ( Rego et al . , 2017; Richardson et al . , 2016 ) . We proposed that in mycobacteria , increased specific errors in gene translation – mistranslation – led to intracellular protein variants of the drug target of rifampicin , RpoB , which resulted in phenotypic resistance to rifampicin ( Javid et al . , 2014; Su et al . , 2016 ) . Clinical isolates with mutations in the essential amidase gatCAB that mediates variation in cellular mistranslation rates had both increased mistranslation and rifampicin tolerance , suggesting that this is a clinically relevant mode of antibiotic tolerance ( Su et al . , 2016 ) . The indirect aminoacylation pathway is present in the majority of bacterial species ( with the exception of some proteobacteria such as Escherichia coli ) , all archaea , some mitochondria and other organelles ( Sheppard and Söll , 2008 ) . Bacteria lacking specific glutamine and/or asparagine tRNA synthetases instead utilize a non-discriminatory glutamyl- ( asparaginyl ) synthetase that forms misacylated Glu-tRNAGln and Asp-tRNAAsn aminoacyl complexes , respectively ( Curnow et al . , 1997; Rathnayake et al . , 2017 ) . These misacylated complexes are specifically recognized by GatCAB and amidated to the cognate Gln-tRNAGln and Asn-tRNAAsn aminoacyl tRNAs , thereby preserving the fidelity of the genetic code . We recently identified that in mycobacteria , strains with mutations in gatA causing partial loss of function are not only viable , but can be isolated from patient samples ( Su et al . , 2016 ) . These strains have much higher rates of specific mistranslation – of glutamine to glutamate , and asparagine to aspartate – since a proportion of misacylated Glu-tRNAGln and Asp-tRNAAsn complexes are not fully converted to the cognate aminoacyl forms before taking part in translation at the ribosome . Importantly , wild-type GatCAB could also be limiting . Wild-type mycobacteria flow-sorted for lower GatCAB expression had both higher mistranslation rates and rifampicin tolerance ( Su et al . , 2016 ) , suggesting that targeting the indirect tRNA aminoacylation pathway may present a novel and attractive means for increasing mycobacterial rifampicin susceptibility . Here , we identify the natural product kasugamycin as a small molecule that can specifically decrease mistranslation due to the indirect tRNA aminoacylation pathway . At sub-inhibitory concentrations , kasugamycin , but not another aminoglycoside streptomycin can increase mycobacterial rifampicin susceptibility both in vitro and in animal infection .
We hypothesized that a small molecule that could specifically decrease mycobacterial mistranslation would result in increased susceptibility to rifampicin . GatCAB-mediated mistranslation is not due to ribosomal decoding errors – but rather due to misacylated Glu-tRNAGln and Asp-tRNAAsn complexes taking part in translation ( Su et al . , 2016 ) . In addition to other reported activities in E . coli ( Lange et al . , 2017; Müller et al . , 2016; Kaberdina et al . , 2009; Moll and Bläsi , 2002 ) , the aminoglycoside kasugamycin decreased ribosomal misreading of mRNA ( van Buul et al . , 1984 ) , but it was not known if it could also decrease errors due to translation of misacylated tRNAs , as the indirect tRNA aminoacylation pathway is not present in E . coli . We tested whether kasugamycin could increase fidelity of misacylated tRNA-mediated mycobacterial mistranslation using a gain-of-function genetic reporter ( Figure 1A ) . Kasugamycin at sub-inhibitory doses ( Supplementary file 1 ) increased translational fidelity in both the model organism Mycobacterium smegmatis ( Msm ) and pathogenic Mycobacterium tuberculosis ( Mtb ) ( Figure 1B , C and Figure 1—figure supplement 1 ) . Importantly , kasugamycin decreased mistranslation in mycobacterial strains with mutated gatA that have extremely high misacylated-tRNA-mediated mistranslation due to partial loss of GatCAB function ( Su et al . , 2016 ) – Figure 1D , verifying that it was effective in increasing discrimination against translation of misacylated Glu-tRNAGln and Asp-tRNAAsn aminoacyl-tRNAs . To further test whether kasugamycin was able to decrease Asp-tRNAAsn misacylation-mediated mistranslation , we developed a hybrid cell-free translation system . A commercially available E . coli coupled transcription-translation system was supplemented with a non-discriminatory aspartyl-synthetase ( Ruan et al . , 2008 ) that was able to misacylate E . coli tRNAAsn ( see Materials and methods ) . Addition of the non-discriminatory synthetase markedly increased mistranslation , as measured by the Nano-luciferase-GFP gain-of-function reporter . Kasugamycin , at concentrations that did not decrease GFP signal , was able to reduce mistranslation-induced Nluc signal ( Figure 1E ) , confirming that kasugamycin could increase ribosomal discrimination of misacylated tRNA . One of the activities of kasugamycin in E . coli is the inhibition of the translation of canonical mRNA transcripts ( i . e . those with a 5’ UTR including a Shine-Dalgarno sequence ) , but not leaderless transcripts lacking a 5’ UTR ( Kaberdina et al . , 2009; Moll and Bläsi , 2002 ) , although permissive translation of leaderless transcripts was not universal ( Schuwirth et al . , 2006 ) . We wanted to test whether selective inhibition of translation of canonical but not leaderless transcripts with kasugamycin was evident in mycobacteria , especially since mycobacteria have many annotated leaderless transcripts ( Shell et al . , 2015; Cortes et al . , 2013 ) . We constructed a reporter strain of M . smegmatis that expressed two fluorescent proteins , GFP and mCherry from the same basic promoter ( Psmyc ) , but the promoter driving mCherry resulted in a leaderless transcript ( Figure 1—figure supplement 2A and Materials and methods ) . The translation inhibitor chloramphenicol inhibited translation of both fluorescent proteins , but kasugamycin at 1500 µg/ml , more than 10 times higher concentrations than required to decrease mistranslation , did not significantly attenuate translation of either fluorescent protein ( Figure 1—figure supplement 2B ) . Kasugamycin has also been reported to generate novel 61S ribosomes in E . coli ( Kaberdina et al . , 2009 ) , but our attempts to isolate such structures from M . smegmatis were unsuccessful ( not shown ) . Kasugamycin acts on the ribosome as an inhibitor of 30S initiation ( Wilson , 2014 ) . Could other translation inhibitors – either targeting 30S initiation or other steps of translation also decrease mistranslation of misacylated tRNAs ? We tested edeine , another 30S initiation inhibitor , and chloramphenicol , an inhibitor of peptide-bond formation ( Wilson , 2014 ) , both at sub-inhibitory concentrations with our gain-of-function mistranslation reporter . Intriguingly , edeine and kasugamycin , but not chloramphenicol could decrease mistranslation rates ( Figure 1—figure supplement 3 ) , suggesting that inhibition of 30S initiation might interfere with ribosomal discrimination of misacylated tRNAs by a yet to be characterized mechanism . We had previously showed that mistranslation generated via the indirect tRNA aminoacylation pathway in mycobacteria played an important role in rifampicin phenotypic resistance ( Su et al . , 2016; Javid et al . , 2014 ) . We therefore hypothesized that since kasugamycin could reduce mistranslation generated by this pathway , it would be able to reduce rifampicin tolerance . In keeping with our hypothesis , kasugamycin reduced mycobacterial rifampicin phenotypic resistance in wild-type ( Figure 2A , B ) and high mistranslating gatA mutant mycobacterial strains ( Figure 2C ) . We then tested the ability of kasugamycin to enhance rifampicin-mediated killing of mycobacteria in axenic culture . Kasugamycin had no effect on mycobacterial growth , but addition to rifampicin significantly enhanced killing and sterilization of mycobacterial cultures ( Figure 2D ) . Kasugamycin-mediated decrease in rifampicin tolerance was not due to its activity as an aminoglycoside – known protein translation inhibitors . The aminoglycoside streptomycin increases mistranslation ( Leng et al . , 2015 ) . Plating of Msm on rifampicin-agar in the presence of sub-inhibitory concentrations of streptomycin increased the number of phenotypically-resistant colonies ( Figure 2—figure supplement 1 ) . Furthermore , in keeping with our observations with regard to 30S initiation inhibition and mistranslation , plating Msm on rifampicin-agar with sub-inhibitory concentrations of edeine but not chloramphenicol , decreased the number of phenotypically resistant colonies ( Figure 2—figure supplement 2 ) , confirming the link between reducing mistranslation rates and decreasing rifampicin tolerance . To further verify that kasugamycin’s effects on rifampicin susceptibility are due to its specific activity in decreasing mycobacterial mistranslation , we used a mycobacterial strain with a single point mutation in RpoB – strain Msm-RpoB-N434T . This strain has lower tolerance to rifampicin since a critical rifampicin-binding residue could no longer be mistranslated via the indirect pathway ( Su et al . , 2016 ) . Msm-RpoB-N434T was less rifampicin tolerant than its parent strain , but kasugamycin was less potent at decreasing rifampicin tolerance further ( Figure 2E ) . Since other reported activities of kasugamycin ( Lange et al . , 2017; Müller et al . , 2016; Kaberdina et al . , 2009 ) would not be affected by a single point mutation in the rpoB gene , we conclude that the major mechanism by which kasugamycin increased rifampicin susceptibility is by decreasing mistranslation-induced protein variants . In E . coli , antibiotic resistance is preceded by tolerance ( Levin-Reisman et al . , 2017 ) . Kasugamycin but not streptomycin given alongside rifampicin pre-treatment significantly reduced the likelihood of resistance following high-dose rifampicin challenge ( Figure 2F ) , suggesting that kasugamycin may limit development of de novo resistance . To test activity in vivo , we needed to establish whether kasugamycin had favorable tolerability and pharmacokinetics in an animal model , and if so , was kasugamycin administration with rifampicin able to boost killing of M . tuberculosis . We characterized the pharmacokinetics of parenterally administered kasugamycin and streptomycin in mice . Both agents showed dose-dependent concentration-time profiles in plasma , with rapid clearance ( Figure 3—figure supplement 1 ) . Administration of the maximum tolerated daily dose of kasugamycin – 400 mg/kg – resulted in a Cmax/EC ( mistranslation ) 50 ( peak plasma concentration divided by the half-maximal effective in vitro dose of kasugamycin in reducing mistranslation rates – Figure 1C ) of ~2 . 5 and time over EC50 of 1 hr or only 4% of the dosing interval . Nevertheless , co-administration of kasugamycin resulted in an astonishing 30-fold boosting of rifampicin killing of Mtb in mouse lungs ( Figure 3A ) . However , concomitant dosing of rifampicin and kasugamycin was poorly tolerated , even in the absence of tuberculosis ( Figure 3—figure supplement 2 ) . Histopathological examination of organs did not reveal a specific cause ( not shown ) . Lower doses of kasugamycin co-administered with rifampicin were ineffective at enhancing rifampicin activity in vivo ( not shown ) . In vitro , pre-treatment of axenic mycobacterial cultures with kasugamycin but not streptomycin decreased rifampicin tolerance , to a lesser degree ( Figure 3—figure supplement 3 ) . We thus opted for successive administration of high-dose kasugamycin intermittently with rifampicin in mice , which was well-tolerated . To specifically exclude the observed activity as being due to aminoglycoside-mediated inhibition of protein synthesis or post-antibiotic effects , we also included streptomycin-treated arms ( Figure 3B ) . Since the driver of aminoglycoside efficacy is Cmax/MIC ( peak plasma concentration divided by in vitro minimum inhibitory concentration ) ( Scaglione and Paraboni , 2006 ) , we selected a streptomycin dose of 3 mg/kg , achieving a Cmax/MIC of 7 , while the Cmax/MIC of kasugamycin was 0 . 8 ( Supplementary file 1 ) , thus avoiding underestimating streptomycin’s activity . Sequential treatment with rifampicin and kasugamycin , but not streptomycin led to significantly enhanced killing of Mtb in mouse spleens but not lungs ( Figure 3C , Figure 3—figure supplement 4 ) , which was not explained by differences in bulk-tissue PK ( Figure 3—figure supplement 5 ) . Thus , despite limited drug exposure resulting in effective concentrations achieved for only a small fraction of the treatment period , the in vitro effects on rifampicin tolerance were recapitulated in mice .
There are multiple non-redundant models proposed for antibiotic tolerance ( Abel Zur Wiesch et al . , 2015; Aldridge et al . , 2014; Balaban et al . , 2013 ) . In all these models , tolerance is mediated by generation of phenotypic heterogeneity within bacterial populations . Generation of stochastic errors during gene translation – mistranslation – is more prevalent , and occurs at higher rates , than previously appreciated ( Mohler and Ibba , 2017; Ribas de Pouplana et al . , 2014 ) and is one mechanism by which bacteria may generate considerable phenotypic heterogeneity . We show here that it is possible to identify a small molecule that can specifically decrease mistranslation rates , and hence increase antibiotic susceptibility . There has been great interest in identifying small molecules that specifically target antibiotic-tolerant mycobacteria ( Darby and Nathan , 2010; Zheng et al . , 2014; Grant et al . , 2013; Zheng et al . , 2017; Sukheja et al . , 2017; Alumasa et al . , 2017; Wang et al . , 2013; Vilchèze et al . , 2017 ) . These molecules have for the most part been identified via in vitro phenotypic screens that are specific for non-replicating persistence , although novel ‘target-specific’ , whole-cell screens are proving highly useful in identifying compounds with activity against mycobacteria in vitro that are specific for certain stress-adaptation pathways ( Zheng et al . , 2017 ) . With one exception ( Wang et al . , 2013 ) , however , none of these candidates , and the pathways that they target , have been validated within an animal model of infection . Kasugamycin is unique among aminoglycosides in its ability to decrease translational error – all other aminoglycosides increase mistranslation ( Leng et al . , 2015; Ribas de Pouplana et al . , 2014; van Buul et al . , 1984 ) . In addition to its known effects in reducing ribosomal decoding errors , we have shown that kasugamycin can increase discrimination against physiologically misacylated tRNAs . It had been previously demonstrated that the ribosome has some proof-reading functionality beyond Watson-Crick base-pairing . This was limited to rejection of peptides formed from incorrect codon·anti-codon base-pairs ( Zaher and Green , 2009 ) . Misacylated tRNAs would still form cognate codon·anti-codon pairs at the ribosome , and would therefore not be rejected by such a mechanism . Rejection of misacylated aminoacyl tRNA formation had previously been described at the aminoacyl synthetase stage ( Ibba and Söll , 1999 ) , or by discrimination by EF-Tu ( LaRiviere et al . , 2001 ) . Kasugamycin’s binding to the E . coli ribosome is close to the A794 and G926 residues ( E . coli numbering ) of 16S rRNA ( Schuwirth et al . , 2006 ) . Given that these residues are universally conserved , it is likely that kasugamycin’s binding to mycobacterial and other bacterial ribosomes is in a similar location . Therefore , our data suggest that kasugamycin-bound ribosomes also possess a hitherto unknown ability to discriminate against misacylated EF-Tu·aminoacyl-tRNA complexes . Since edeine , another 30S initiation inhibitor , but not inhibitors of elongation ( streptomycin ) or peptide-bond formation ( chloramphenicol ) could also decrease mistranslation from misacylated tRNAs , this suggests a conserved mechanism , potentially involving translation initiation , directly or indirectly , in discrimination of physiologically misacylated tRNAs . When kasugamycin and rifampicin were given daily , there was significant potentiation in mouse lungs after 2 weeks of treatment ( Figure 3A ) , but with significant toxicity . However , in the better-tolerated alternate dosing schedule ( Figure 3B ) , potentiation was seen only in mouse spleens , not lungs ( Figure 3C ) . These differences could not be explained by differences in kasugamycin bulk tissue distribution ( Figure 3—figure supplement 5 ) . Even within a single organ/tissue , there can be significant heterogeneity in drug penetration and distribution ( Prideaux et al . , 2015 ) , which we did not measure , and such heterogeneity of distribution in either rifampicin or kasugamycin or both may explain the differences observed . Furthermore , a greater proportion of M . tuberculosis may be resident in macrophages in lungs compared with spleens , and aminoglycosides have poor intra-cellular penetration ( Brezden et al . , 2016 ) . As such , under this dosing schedule kasugamycin and rifampicin may be targeting the same mycobacterial subpopulation within spleens but different subpopulations within lungs , potentially explaining the observations . The maximum dose of kasugamycin that could be administered to mice was limited due to toxicity and pharmacokinetics . At the maximum tolerated dose ( 400 mg/kg , once daily ) , the peak plasma concentration of 300 µg/ml failed to reach the in vitro minimum inhibitory concentration of 400 µg/ml . Intriguingly , these limitations allowed us to identify the relative in vivo potency of kasugamycin compared with conventional anti-microbials . Most anti-tuberculous drugs require peak plasma concentrations orders of magnitude greater than in vitro MIC for measurable efficacy ( Mitchison , 2012; Pasipanodya and Gumbo , 2011; Pasipanodya et al . , 2013 ) . By contrast , the bacteriostatic kasugamycin given alone , administered intermittently ( 10 doses in 30 days ) to infected mice , and with a Cmax/MIC <1 , and at plasma concentrations <25% of MIC for 99% of the dosing interval was able to restrict Mtb growth in vivo . Streptomycin , a bactericidal aminoglycoside , given at far greater equivalent doses , had no effect . Mycobacteria increase specific mistranslation rates under conditions such as nutrient limitation and low pH ( Javid et al . , 2014 ) , which mimic potential in vivo environments . These data suggest the efficacy of kasugamycin may not be as a conventional anti-microbial , but possibly by targeting bacterial adaptation to the host via reducing mistranslation . There are several alternative mechanisms by which kasugamycin alone may have led to bacterial growth restriction in vivo . Kasugamycin’s biological activities differ with other aminoglycosides in additional ways than its contrasting effects on translational fidelity . Most aminoglycosides inhibit translocation during protein synthesis , whereas kasugamycin inhibits translation initiation by blocking the mRNA channel in the small ribosomal subunit ( Schluenzen et al . , 2006; Schuwirth et al . , 2006 ) . The reported structures of kasugamycin bound to the ribosome suggest that during 70S ( leaderless ) initiation , there is less steric hindrance of mRNA passage than in canonical initiation ( Schluenzen et al . , 2006 ) , potentially explaining why kasugamycin is permissive for translation of some , but not all ( Schuwirth et al . , 2006 ) leaderless transcripts ( Kaberdina et al . , 2009; Moll and Bläsi , 2002 ) . With regard to potentiation of rifampicin in vitro , our data strongly suggest that kasugamycin is acting by reducing mistranslation generated by the indirect tRNA aminoacylation pathway , and hence protein variants that mediate rifampicin tolerance ( Figure 2E ) . However , although we did not find evidence that in mycobacteria kasugamycin selectively blocks translation of canonical but not leaderless mRNA transcripts ( Figure 1—figure supplement 2 ) , or form alternate ribosomes ( Kaberdina et al . , 2009 ) ( not shown ) , we cannot formally exclude the possibility that these mechanisms play a role in kasugamycin’s activity in restricting M . tuberculosis growth in mice in the absence of rifampicin . Other potential mechanisms may be that kasugamycin is concentrated in macrophages , such that the intracellular concentration exceeded the MIC . Although we did not measure the intra-macrophage concentration of kasugamycin , almost all aminoglycosides enter cells poorly due to their polar structure ( Brezden et al . , 2016 ) . Synergistic drug combinations have the potential to radically improve treatment for tuberculosis . New methods for modelling synergy from purely empirical in vitro measurements can identify novel combinations ( Cokol et al . , 2017 ) . More rational approaches can rescue current drugs that have limitations due to emergence of resistance or toxicity ( Blondiaux et al . , 2017 ) . Our data suggest that targeting mycobacterial mistranslation may be a generally effective strategy , and not only in the context of potentiating rifampicin activity . Since most bacteria , with the notable exception of E . coli and a few other proteobacteria utilize the indirect tRNA pathway for synthesis of aminoacylated glutamine and/or asparagine tRNAs ( Curnow et al . , 1997 ) , targeting adaptive mistranslation ( Ribas de Pouplana et al . , 2014 ) may be a useful strategy in the treatment of diverse bacterial infections .
Mycobacterium smegmatis mc2-155 and its derivatives were cultured in Middlebrook 7H9 ( BD DifcoTM ) broth supplemented with 0 . 2% glycerol , 0 . 05% Tween-80 , 10% Albumin-dextrose-salt ( ADS ) , and on Luria-Bertani agar ( LB agar ) for plate assays . The high mistranslating strain HWS . 4 with a mutation , gatA-V405D and the M . smegmatis strain with a point mutation in rpoB Msm-rpoB-N434T are previously described ( Su et al . , 2016 ) . Mycobacterium tuberculosis-H37Rv was cultured in Middlebrook 7H9 broth with 0 . 2% glycerol , 0 . 05% Tween-80 , 10% OADC ( oleic acid , albumin , dextrose and catalase ) , and in Middlebrook 7H11 agar ( BD Difco ) supplemented with OADC for plate assays . Rifampicin , kasugamycin and streptomycin were purchased from Sigma . Rifampicin was dissolved in dimethyl sulphoxide ( DMSO ) ; kasugamycin and streptomycin were dissolved in water , and filter sterilized . The in vitro MICs of kasugamycin , streptomycin and rifampicin for M . smegmatis and M . tuberculosis are given in Supplementary file 1 . For in vitro experiments , kasugamycin and streptomycin doses were selected that had no effects on growth either in axenic culture or plating . Doses are given in the Figure legends . The only exception is Figure 1—figure supplement 2 – testing whether kasugamycin could inhibit translation of canonical/leaderless transcripts , when a dose of 1500 µg/ml was chose: this was 50x higher than a dose that could decrease mistranslation , and was close to the MIC . For the in vivo experiments , the maximum tolerated dose ( 400 mg/kg daily ) of kasugamycin was given to mice . The streptomycin dose was calculated to represent 9x higher equivalent dose than kasugamycin ( by Cmax/MIC ) so as to not underestimate its effects , but not so high that streptomycin’s known bactericidal activity might interfere with interpretation of the data . Rifampicin ( 10 mg/kg ) is a standard dose used in most in vivo experiments . The nano-luciferase gene ( nluc ) sequence was obtained from Promega and optimized to accommodate mycobacterial codon usage preference and synthesized ( Genscript ) . The nano-luciferase gene was fused downstream of the secretion signal sequence ( first forty amino acids ) of the secreted mycobacterial antigen 85A/B . The fused product was PCR amplified and cloned into the pJet1 . 2 cloning vector ( Thermo Fisher Scientific ) and verified by Sanger sequencing . Site directed mutagenesis was used to create a D140N mutation in nano-luciferase , and the mutation was verified by Sanger sequencing . The mutated nluc gene containing the secretion signal was then cloned into the tetracycline-inducible pUVtetOR vector having a hygromycin-resistant gene cassette . After sequence verification , the recombinant plasmid was transformed into M . smegmatis mc2-155 and M . tuberculosis-H37Rv using standard methodology . Codon optimized green fluorescence protein ( GFP ) sequence was cloned into tetracycline inducible pMC1s vector , which has kanamycin-resistant gene cassette , and electroporated into M . smegmatis mc2-155 and M . tuberculosis-H37Rv containing the nano-luciferase ( N-luc ) reporter . The basis of the N-luc/GFP assay is a gain of Nano-luciferase ( N-luc ) signal , similar to previously published methods ( Javid et al . , 2014; Kramer and Farabaugh , 2007 ) . The D140N mutation in N-luc caused approximately 100-fold reduction in activity compared with the wild-type enzyme . Mistranslation specifically of aspartate ( D ) for asparagine ( N ) in a subset of newly translated polypeptides would result in regaining of wild-type N-luc activity . Both N-luc and GFP were induced by the same tetracycline-responsive promoter , therefore total N-luc activity was corrected by dividing by total GFP fluorescence to account for variation in gene expression between samples . The N-luc/GFP ratio gave only a relative , not absolute , indication of specific mistranslation rates . M . smegmatis mc2-155 containing the N-luc and GFP reporter plasmids was grown in 7H9 broth containing hygromycin ( 50 µg/ml ) and kanamycin ( 20 µg/ml ) to late log phase ( OD600 = 2 . 0 ) at 37˚C , after which anhydrotetracycline ( ATc , 50 ng/ml ) was added to induce N-luc expression . The bacterial culture was then immediately aliquoted into a 96-well plate ( 100 µl , that is approximately 6 × 107 cells per well ) , different concentrations of kasugamycin or water control were added , and the cultures incubated for 16 hr at 37˚C with shaking . The cultures were then transferred to a 96-well black plate and GFP fluorescence measured . The plate was then centrifuged ( 4000 rpm for 10 min ) , and supernatants transferred to a 96-well white luminescence plate . The nano-luciferase assay was performed using Nano-Glo luciferase assay kit ( Promega ) , and luminescence measured by the same machine . Relative mistranslation rates for M . tuberculosis H37Rv was measured in the same way , with minor modification . The H37Rv strain containing the N-luc-D140N and GFP reporter plasmids was induced with ATc ( 100 ng/ml ) for 2 days before measuring the N-luc and GFP signals . Mistranslation measurement using the Renilla-Firefly dual luciferase was performed as described previously ( Su et al . , 2016 ) . Briefly , M . smegmatis mc2-155 strains harboring the reporters were grown till stationary phase . The cultures were diluted 20 times in fresh 7H9 medium , and expression of the dual luciferase was induced with 50 ng/ml anhydrotetracycline ( ATc ) . After 6 hr , the bacterial cells were lysed and luciferase activities measured by dual luciferase assay kit ( Promega ) . Measurements of fluorescence/luminescence of M . smegmatis were performed on a Fluoroskan Ascent FL Fluorimeter and Luminometer ( Thermoscientifc ) , and for M . tuberculosis ( and the Edeine/Chloramphenicol experiments ) on a Biotek Synergy H1 plate reader ( Fisher Scientific ) . The different instruments used was largely responsible for the differences in arbitrary unit ( AU ) values for mistranslation rates between the two species . Mistranslation rates were calculated as previously ( Su et al . , 2016 ) . Tests of difference of means were performed by two-tailed Student’s t-test . The Deinococcus radiodurans non-discriminatory aspartyl synthetase Dr AspRS2 is able to misacylate E . coli asparagine tRNA with aspartate ( Ruan et al . , 2008 ) ( i . e . form Asp-tRNAAsn – the mycobacterial non-discriminatory enzyme does not recognize E . coli tRNA , not shown ) , and was therefore used in the E . coli cell-free translation system . Codon optimized Dr AspRS2 gene containing 6xHis-tag at the 5´end was synthesized ( Genewiz ) and cloned into XbaI and EcoRI restriction sites of pET28a ( + ) vector and transformed into E . coli BL21 ( DE3 ) . Expression of Dr AspRS2 was induced by adding 1 mM IPTG . After one hour of induction at 37°C , the cells were harvested and Dr AspRS2 was purified by Ni-NTA affinity chromatography using standard methods . The final protein concentration was determined by Bradford reagent ( Bio-Rad ) . The activity of the enzyme was confirmed by 3H-Aspartate labeling of E . coli tRNAAsn ( not shown ) . For cell-free measurement of mistranslation , a reporter expressing mutated N-luc ( D140N ) linked with wild-type GFP by a GGSGGG linker was generated . The mutated n-luc was produced by site directed mutagenesis and linked to WT gfp by Gibson assembly . The nluc linked gfp was then cloned into pIVEX vector ( 5 Prime ) for in vitro coupled transcription-translation ( IVT ) . The reporter measured relative mistranslation of aspartate for asparagine by gain-of-function Nluc activity/GFP fluorescence , as above . The coupled transcription-translation IVT reaction was carried out using an E . coli T7 S30 Extract System for Circular DNA kit ( Promega ) following manufacturer’s instructions with the following modifications . Since E . coli lacks the indirect tRNA aminoacylation pathway , the reaction mix was spiked with the non-discriminatory Dr AspRS2 to form misacylated Asp-tRNAAsn complexes that could take part in translation . 2 µM Dr AspRS2 or reaction buffer was added to the IVT reaction mix . To determine if kasugamycin could increase ribosomal discrimination of misacylated Asp-tRNAAsn-mediated translational errors , different concentrations of kasugamycin or carrier were added to the reaction mix and incubated on ice for 10 min before addition of the reporter template . Once the DNA template was added , the tube was mixed thoroughly and incubated at 37°C for 2 hr before the reaction was quenched by placing on ice for 5 min . The Nluc activity and GFP fluorescence was measured as above . A dual-fluorescent reporter was constructed to measure translation of leaderless and canonical transcripts . The Psmyc promoter was cloned from plasmid PML1357 ( Huff et al . , 2010 ) ( Addgene ) and its transcription start site mapped by 5’ RACE ( not shown ) . The gene for mCherry was fused directly to the transcription start site , forming a leaderless expressed gene , and gfp was cloned 3’ to a canonical Shine-Dalgarno sequence ( Figure 1—figure supplement 2 ) . Both fluorescent proteins had a C-terminal tag , AAV , which decreases stability and half-life of expressed proteins by targeting them for protease-mediated degradation ( Andersen et al . , 1998 ) , thereby allowing monitoring of translation in real time . The two expression cassettes were subcloned into plasmid PSE100 and transformed into wild-type M . smegmatis . The transcription start sites of the two fluorescent promoters was verified by 5’RACE and was as predicted ( not shown ) . Biological triplicates of the strain were grown to log phase , and then back-diluted to lag phase . After 4 hr growth , antibiotics ( or carrier ) were added to cultures . OD600 , green and red fluorescence were measured by a Varioskan FLASH ( Thermo ) instrument . Rifampicin tolerance ( phenotypic resistance ) on agar medium was measured as previously described ( Su et al . , 2016 ) . Stationary phase M . smegmatis mc2-155 cultures were serially diluted , and spread on LB-agar plates containing rifampicin ( 50 µg/ml ) , with or without kasugamycin/streptomycin . The plates were incubated at 37˚C for 5 – 7 days after which the number of colony-forming units ( cfu ) were counted . The number of bacterial cells in the inoculum was calculated by plating serial dilutions of the culture on antibiotic-free LB-agar plates and counting total plated cfu . For RSPR analysis of M . tuberculosis H37Rv , bacteria were spread on 7H11 agar medium containing rifampicin ( 0 . 2 µg/ml ) with or without kasugamycin ( 31 µg/ml ) , and the plates were incubated for 6 weeks . For antibiotic pre-treatment experiments , M . smegmatis mc2-155 was grown in 7H9 broth containing kasugamycin ( 50 µg/ml ) or streptomycin ( 0 . 25 µg/ml ) and then spread on LB agar containing rifampicin . In all cases , doses of kasugamycin or streptomycin or other antibiotics were selected that did not by themselves decrease plating efficiency . Tests of difference of means were performed by two-tailed Student’s t-test . M . smegmatis mc2-155 was grown overnight in 7H9 ( OD600 = 0 . 6 ) broth , and approximately 5 × 106 cells were inoculated into fresh 7H9 broth containing different concentrations of rifampicin ( 0 , 10 , 20 , 50 µg/ml ) , with or without kasugamycin ( 50 µg/ml ) . At different time points , aliquots were removed from each culture , cells were washed once and 10-fold dilutions spread onto LB agar medium with sterile glass beads . The number of viable bacteria at each time point was calculated from the resulting number of colonies . M . smegmatis mc2-155 was grown to early stationary phase in 7H9 medium containing DMSO , rifampicin ( 1 µg/ml ) , or rifampicin with kasugamycin ( 50 µg/ml ) or streptomycin ( 0 . 2 µg/ml ) for 3 hr . After pre-exposure , cells were washed twice in PBS and 100 µl aliquots ( 2 . 5 × 107 cells ) were transferred to wells in 96-well plates ( 240 wells/group in total , experiments conducted with three independent cultures over three separate days and results pooled ) containing 7H9 and rifampicin ( 100 µg/ml ) , which was selective for bona fide genetic resistance to the antibiotic . After 4 days of incubation , the number of rifampicin resistant cultures was observed by the presence of turbid growth . The ‘resistance ratio’ was calculated as the percentage of wells with turbid ( resistant ) cultures divided by the total number of wells per condition . Statistical analysis was performed by Fischer’s exact test ( GraphPad Prism ) to compare conditions . All mouse experiments were approved by the Institutional Animal Care and Use Committee of the New Jersey Medical School , Rutgers University , Newark , NJ , under protocol number 15114D1018 . For PK analysis , 8- to 10-week-old female BALB/c mice were injected intramuscularly with 100 mg/kg , 200 mg/kg or 400 mg/kg kasugamycin dissolved in 0 . 9% saline . Separate analysis for single-dose injections confirmed that intra-peritoneal and intra-muscular injection had exactly the same PK profiles ( not shown ) . Blood was collected by tail snip after 15 and 30 mins , 1 , 3 , 5 and 8 hr following kasugamycin injection . Plasma was separated by centrifuging the blood at 5000 rpm for 5 min . Of plasma sample , 10 µl was extracted by adding 10 µl of acetonitrile/water ( 1:1 ) and 100 µl of acetonitrile: methanol ( 1:1 ) containing 10 ng/ml of verapamil ( used as internal standard to correct for differences in injection volume that may happen during HPLC run ) , and plasma kasugamycin levels determined by high-pressure liquid chromatography coupled with tandem mass spectrometry ( LC/MS/MS ) . The LC/MS/MS was performed on a Sciex Applied Biosystems Qtrap 4000 triple quadrupole mass spectrometer coupled to an Agilent 1260 HPLC system . Chromatography for kasugamycin was performed on a Cogent Diamond Hydride column ( 2 . 1 × 50 mm , particle size 4 µm ) using a normal phase gradient elution . The gradient used 0 . 1% formic acid in Milli-Q deionized water and 0 . 1% formic acid in acetonitrile . Kasugamycin and verapamil were ionized using ESI-positive mode ionization and monitored using masses 380 . 17/112 . 10 and 455 . 4/156 . 2 , respectively . Standard curve and quality control solutions were created by diluting 1 mg/ml of DMSO stocks of kasugamycin in acetonitrile/water ( 1:1 ) . 10 µl of each dilution was added to 10 µl of drug-free plasma ( Bioreclamation ) and 100 µl of acetonitrile: methanol ( 1:1 ) containing 10 ng/ml verapamil . These standards were extracted as mentioned above . For PK analysis of streptomycin , ( 8-10 ) week-old female BALB/c mice were intraperitoneally injected with 10 mg/kg , 20 mg/kg and 50 mg/kg of streptomycin dissolved in 0 . 9% saline , and blood was collected by tail snip after 15 and 30 mins , 1 , 3 , 5 and 8 hr . To 10 µl of plasma samples , 15 µl of extraction solution ( 35% trichloroacetic acid in water ) was added . Then 10 µl of internal standard ( 20 µg/ml amikacin in water ) and 70 µl of water was added to the mixture , and centrifuged at 3000 rpm for 5 min at 10˚C . Of the extract , 70 µl was transferred to an analysis plate , and 3 µl was injected on LC/MS/MS for analysis . Agilent Zorbax SB-C8 , 4 . 6 × 75 mm , 3 . 5 µm was used as liquid chromatography column and Sciex Applied Biosystems Qtrap 4000 mass spectrometer was used for analysis . This experiment was approved by the Institutional Animal Care and Use Committee of Tsinghua University under protocol number 17-BJ2 . Six-week-old female BALB/c mice were given 10 mg/kg rifampicin orally by gavage and/or different doses ( 50 mg/kg and 400 mg/kg ) of kasugamycin intraperitoneally daily for a week . Bodyweights of the mice were monitored daily . Mice were euthanized if they showed signs of visible distress or if they lost >20% of initial bodyweight . All mouse infection and treatment experiments were approved by the Institutional Animal Care and Use committee of Rutgers University . Initial sample size calculation was calculated to have 80% power to detect a 0 . 5 log CFU reduction based on inter-animal variability established through a large number of similar animal efficacy experiments . For the subsequent animal experiments , sample size was determined from preliminary data . Nine-week-old female BALB/c mice were infected with M . tuberculosis H37Rv with a Glas-Col inhalation system . An inoculum of 1 × 107 cfu/ml bacteria was added to the nebulizer , and the initial bacterial lung load was determined by sacrificing four mice after 3 hr of infection . After 1 week , four mice were sacrificed to determine the bacterial burden in the lungs before start of treatment . The mice were treated daily with 10 mg/kg rifampicin orally ( po ) alone , or with 400 mg/kg kasugamycin intra-peritoneally ( ip ) alone , or with a combination of rifampicin ( 10 mg/kg ) and kasugamycin ( 400 mg/kg ) or carrier controls for 2 weeks . The mice were then sacrificed , the lungs harvested and homogenized; and the homogenates spread on Middlebrook 7H11 agar supplemented with OADC . For the intermittent rifampicin-kasugamycin treatment procedure ( Figure 2B ) , 10- to 12-week-old female BALB/c mice were infected with M . tuberculosis H37Rv , with 1 × 107 cfu/ml added to the nebulizer , and the lung bacterial load was determined by sacrificing four mice after 3 hr of infection . After 2 weeks of infection , four mice were sacrificed to determine bacterial burden in the lungs before start of treatment . The mice were given the following antibiotic regimen: The control group received PBS ip for 4 days , then water po for 5 days . From there on , they received PBS for 2 days followed by water for 5 days; and this cycle was repeated until the end of the treatment . This treatment procedure was used for the kasugamycin and streptomycin groups as well , with PBS being replaced by 400 mg/kg kasugamycin and 3 mg/kg streptomycin , respectively . The same procedure was also applied for the rifampicin only group , with water being replaced by 10 mg/kg rifampicin . Mice that received two antibiotics were given 400 mg/kg kasugamycin or 3 mg/kg streptomycin for 4 days , and then 10 mg/kg rifampicin for 5 days . This was followed by kasugamycin or streptomycin for 2 days and rifampicin for 5 days; and the cycle repeated until the end of the experiment . After 44 days of infection , mice were euthanized , lungs and spleens of the mice were harvested and homogenized , and the homogenates were spread on Middlebrook 7H11 agar supplemented with OADC . Statistical analysis of differences in means between groups was performed by one-way ANOVA followed by Tukey’s post-hoc correction for multiple samples ( GraphPad Prism ) . Tests of significance are stated in the relevant methods sections and figure legends . All experiments were performed at least three times independently except for the edeine/chloramphenicol experiments ( performed twice independently ) and animal experiments . Figure 2A is representative of an experiment performed twice . For Figure 2B , data from two independent experiments are presented except for the Streptomycin arm , which was performed once . | A bacterium called Mycobacterium tuberculosis is responsible for nearly 98% of cases of tuberculosis , which kills more people worldwide than any other infectious disease . This is due , in part , to the time it takes to cure individuals of the disease: patients have to take antibiotics continuously for at least six months to eradicate M . tuberculosis in the body . Bacteria , like all cells , make proteins using instructions contained within their genetic code . Cell components called ribosomes are responsible for translating these instructions and assembling the new proteins . Sometimes the ribosomes produce proteins that are slightly different to what the cell’s genetic code specified . These ‘incorrect proteins’ may not work properly so it is generally thought that cells try to prevent the mistakes from happening . However , scientists have recently found that the ribosomes in M . tuberculosis often assemble incorrect proteins . The more mistakes the ribosomes let happen , the more likely the bacteria are to survive when they are exposed to rifampicin , an antibiotic which is often used to treat tuberculosis infections . This suggests that it may be possible to make antibiotics more effective against M . tuberculosis by using them alongside a second drug that decreases the number of ribosome mistakes . Chaudhuri , Li et al . investigated the effect of a drug called kasugamycin on M . tuberculosis when the bacterium is cultured in the lab , and when it infects mice . The experiments found that Kasugamycin decreased the number of incorrect proteins assembled by the M . tuberculosis bacterium . When the drug was present , rifampicin also killed M . tuberculosis cells more efficiently . Furthermore , in the mice but not the cell cultures , kasugamycin alone was able to restrict the growth of the bacteria . This implies that M . tuberculosis cells may use ribosome mistakes as a strategy to survive in humans and other hosts . When it was given with rifampicin , kasugamycin caused several unwanted side effects in the mice , including weight loss; this may mean that the drug is currently not suitable to use in humans . Further studies may be able to find safer ways to decrease ribosome mistakes in M . tuberculosis , which could speed up the treatment of tuberculosis . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2018 | Kasugamycin potentiates rifampicin and limits emergence of resistance in Mycobacterium tuberculosis by specifically decreasing mycobacterial mistranslation |
Treatment of EGFR-mutant lung cancer with erlotinib results in dramatic tumor regression but it is invariably followed by drug resistance . In characterizing early transcriptional changes following drug treatment of mutant EGFR-addicted cells , we identified the stem cell transcriptional regulator SOX2 as being rapidly and specifically induced , both in vitro and in vivo . Suppression of SOX2 sensitizes cells to erlotinib-mediated apoptosis , ultimately decreasing the emergence of acquired resistance , whereas its ectopic expression reduces drug-induced cell death . We show that erlotinib relieves EGFR-dependent suppression of FOXO6 , leading to its induction of SOX2 , which in turn represses the pro-apoptotic BH3-only genes BIM and BMF . Together , these observations point to a physiological feedback mechanism that attenuates oncogene addiction-mediated cell death associated with the withdrawal of growth factor signaling and may therefore contribute to the development of resistance .
The invariable development of drug resistance presents a critical challenge to the success of targeted cancer therapies ( Jänne et al . , 2005; O'Hare et al . , 2006; Poulikakos and Rosen , 2011 ) . Several mechanisms leading to such acquired resistance have been identified in patients with EGFR-mutant non-small cell lung cancer ( NSCLC ) treated with small molecule EGFR inhibitors such as erlotinib . Following dramatic initial tumor shrinkage , tumor regrowth is most frequently associated with the emergence of a secondary genetic change , the T790M ‘gatekeeper’ mutation within the EGFR kinase domain , which restores ATP binding in the presence of drug ( Pao et al . , 2004; Kobayashi et al . , 2005; Yun et al . , 2007 ) . In other cases , amplification of related receptor tyrosine kinases ( e . g . , MET ) or mutational activation of downstream kinases ( e . g . , BRAF , PIK3CA ) may bypass the effect of EGFR inhibition ( Engelman et al . , 2007; Sequist et al . , 2011; Ohashi et al . , 2012 ) . Different drug resistance mechanisms may coexist within different metastatic lesions of individual patients . Recently , clinical trials involving rebiopsy of tumor lesions at the earliest sign of drug resistance have also revealed phenotypic conversions that may contribute to drug resistance , including activation of epithelial-to-mesenchymal transition ( EMT ) and the remarkable trans-differentiation of lung cancers from adenocarcinoma to small cell histologies ( Thomson et al . , 2005; Witta et al . , 2006; Sequist et al . , 2011 ) . While some well-defined resistance mechanisms , such as the T790M-EGFR gatekeeper mutation and MET amplification , may be addressed using second line targeted drugs , the plasticity of cancer cell adaptation to disrupted oncogenic signaling poses a major challenge to the long-term success of these promising therapies . In vitro modeling of acquired resistance to EGFR inhibitors has raised the possibility that a transient so-called ‘drug-tolerant’ state may precede the development of mutationally defined , heritable drug resistance ( Sharma et al . , 2010 ) . By analogy with bacterial models of antibiotic resistance , such an intermediate state may be unstable , but enable treated cells to survive in the presence of drug long enough to acquire mutations that ultimately confer sustained drug resistance ( Balaban et al . , 2004 ) . In PC9 mutant , EGFR-addicted lung cancer cells , EGFR inhibition triggers apoptosis in the vast majority of cells in vitro , uncovering approximately 0 . 3% that are drug tolerant , quiescent , and expressing the stem cell marker CD133 and the histone H3K4 demethylase KDM5A ( Sharma et al . , 2010 ) . These drug tolerant cells readily revert to a drug-sensitive state following removal of the EGFR inhibitor , and their emergence in vitro is suppressed by treatment with an EGFR inhibitor combined with inhibitors of either histone deacetylases ( HDACs ) or the IGF-1 receptor . While this intermediate resistance mechanism remains to be validated in the clinical setting , it raises the possibility of suppressing pre-conditions that favor the acquisition of drug resistance , in order to circumvent the challenge of treating multiple established drug-resistant pathways . Beyond the selection of cancer cell populations with transient drug-resistant phenotypes , recent studies of targeted cancer drugs have defined more rapid signaling feedback loops that modulate the cellular response to growth factor inhibition . For instance , acute loss of ERK signaling triggered by RAF or MEK inhibitors in BRAF mutant melanoma cells relieves ERK-dependent inhibition of RAS and CRAF , whose activation through ErbB receptor signaling may lead to paradoxical proliferative signals ( Pratilas et al . , 2009; Paraiso et al . , 2010; Lito et al . , 2012 ) . Similarly , in BRAF mutant colorectal cancers , feedback activation of EGFR-dependent signaling attenuates the consequences of mutant BRAF inhibition , suppressing the apoptotic effect ( Corcoran et al . , 2012; Prahallad et al . , 2012 ) . In addition to signaling feedback loops , transcriptional outputs that generally limit cell proliferation have also been implicated following disruption of EGFR activity , including the expression of transcriptional repressors , regulators of mRNA stability and microRNAs ( Kobayashi et al . , 2006; Amit et al . , 2007; Avraham et al . , 2010 ) . Here , we screened for early , unique transcriptional changes following erlotinib treatment in mutant EGFR-addicted cells , identifying highly specific induction of SOX2 , a master transcriptional regulator required for embryonic stem cell maintenance . SOX2 represses the expression of pro-apoptotic molecules that mediate death following oncogene withdrawal in these cells . The induction of SOX2 results from the activation of FOXO6 , a forkhead family transcription factor , following EGFR inhibition . Knockdown or ectopic expression of SOX2 modulates the degree of apoptosis observed following oncogene withdrawal and promotes drug resistance , pointing to a novel homeostatic mechanism that may contribute to cellular adaptation to the withdrawal of growth factor signaling , which underlies most approaches to targeted cancer therapy .
To interrogate the transcriptional response to EGFR inhibition , we used HCC827 lung cancer cells , harboring an amplified mutated EGFR allele ( in-frame deletion of 15 nucleotides in exon 19 ) and displaying exquisite sensitivity to the EGFR inhibitor erlotinib . Cell cultures were treated in triplicate with 1 µM erlotinib for 6 hr , followed by mRNA isolation and whole transcriptome analysis ( Affymetrix U133 Plus 2 . 0 expression arrays ) ( Rothenberg , 2015 ) . A total of 35 genes showed >fourfold change in expression ( FDR <0 . 05 ) , including 22 downregulated and 13 upregulated transcripts ( represented by 48 unique probe sets; Figure 1—figure supplement 1A ) . Among induced transcripts , SOX2 was unique in the specificity and rapidity of its induction following EGFR inhibition ( Figure 1 , Figure 1—figure supplement 1B ) . Thus , SOX2 was strongly induced in three mutant EGFR-addicted lung cancer cell lines ( HCC827 , PC9 , H3255 ) following treatment with physiologically relevant concentrations of erlotinib ( 0 . 1 µM ) , but not when these cells were treated with comparably effective doses of cytotoxic chemotherapy ( Figure 1A , B and Figure 1—figure supplement 2A ) . SOX2 was also not induced in other oncogene-dependent models , such as ALK-translocated lung cancer cells treated with crizotinib , HER2-amplified breast cancer cells exposed to lapatinib or BRAF-mutant melanoma cells treated with AZD6244 ( Figure 1A and Figure 1—figure supplement 2B ) . Consistent with its dependence on suppression of mutant EGFR signaling in the context of EGFR ‘addiction’ , SOX2 was not induced following erlotinib treatment of H1975 cells , which harbor both an EGFR activating mutation and the T790M gatekeeper mutation that confers resistant to erlotinib; or in H1650 cells with mutated EGFR that are relatively resistant to the effects of EGFR inhibition in part through genetic loss of PTEN ( Figure 1—figure supplement 2B ) ( Sos et al . , 2009 ) . However , treatment of H1975 cells with the L858R/T790M mutation-selective inhibitor WZ4002 resulted in SOX2 induction ( Figure 1—figure supplement 2B , right ) ( Zhou et al . , 2009 ) . In cells that show erlotinib-mediated induction of SOX2 , siRNA-mediated knockdown of EGFR also led to strong induction of SOX2 ( in the absence of erlotinib ) , confirming the specificity of the drug effect ( Figure 1C ) . Simultaneous treatment of cells with actinomycin D and erlotinib suppressed the induction of SOX2 , consistent with a primary effect of EGFR inhibition in increasing SOX2 transcript levels ( Figure 1—figure supplement 2C ) . 10 . 7554/eLife . 06132 . 003Figure 1 . SOX2 transcript is specifically induced by erlotinib in EGFR-mutant and addicted lung cancer cell lines . ( A ) Cell lines were treated with an inhibitor of the driving oncogenic lesion for 24 hr ( erlotinib for EGFR-mutant , lapatinib for HER2-amplified and crizotinib for ALK-translocated cells ) , followed by isolation of total RNA and quantitative PCR for SOX2 transcript . ( B ) PC9 and HCC827 cells were treated with different agents , followed by quantitative PCR for SOX2 . The IC50 for PC9 of erlotinib , taxol , AZD6244 , and GDC0941 is 0 . 05 , 0 . 005 , 5 , and >10 μM; for HCC827 , 0 . 1 , 0 . 01 , >10 , and 1 µM ( data not shown ) . ( C ) PC9 cells were transfected with control siRNA or siRNA targeting EGFR . 48 hr after transfection , the levels of SOX2 and EGFR were determined by qPCR . ( D ) PC9 and HCC827 cells were treated continuously with 0 . 1 µM erlotinib for 9 days , with fresh media/drug added every 3 days . SOX2 level at each time point was analyzed by qPCR ( left ) or immunoblot ( right ) . All qPCR data are displayed as mean Ct value ( normalized to GAPDH and untreated cells ) of 3–6 replicates −/+ SEM , with data in ( C ) normalized to untreated siCTRL cells . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 00310 . 7554/eLife . 06132 . 004Figure 1—figure supplement 1 . Gene expression profiling after erlotinib treatment . ( A ) Heat map showing the list of fourfold significantly ( FDR <0 . 05 ) upregulated ( green ) or downregulated ( red ) transcripts in HCC827 cells with erlotinib treatment . ( B ) Time course of changes in transcript levels for erlotinib-responsive genes . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 00410 . 7554/eLife . 06132 . 005Figure 1—figure supplement 2 . Effect of various treatments on SOX2 expression in different cell contexts . ( A ) HCC827 and PC9 cells . The erlotinib curve is the same as in Figure 1B . ( B ) Left panel , H1650 cells ( EGFR DEL15 activating but IC50 >1 µM ) and WM164 ( BRAF V600E ) show minimal induction following treatment with erlotinib or the MEK inhibitor AZD6244 , respectively . Data for HCC827 cells are from Figure 1A . Right panel , H1975 cells , possessing an EGFR activating L858R mutation and a T790M erlotinib-resistance gatekeeper mutation , do not induce SOX2 with erlotinib treatment ( 1 µM ) but do with the EGFR/T790M selective inhibitor WZ4002 ( 1 µM ) . ( C ) Effect of actinomycin D on induction of SOX2 transcript in HCC827 cells treated with erlotinib . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 00510 . 7554/eLife . 06132 . 006Figure 1—figure supplement 3 . Overlap of differentially expressed genes . Erlotinib treatment of EGFR-mutant cells was compared to MEK inhibitor treatment of BRAF-mutant melanoma cells lines and CL-387 , 785 ( irreversible EGFR inhibitor ) treatment of EGFR-mutant lung cancer cells . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 00610 . 7554/eLife . 06132 . 007Figure 1—figure supplement 4 . Time course of SOX2 induction by quantitative immunofluorescence microscopy . The data for the 24 hr time points are the same as in Figure 2A . p-values are shown for the comparison of mean SOX2 fluorescence of each treated population to DMSO ( Student's t-test , unequal variances , N = 341–3485 , % SOX2+ is shown ) . Source data are included as Figure 2—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 007 Other transcripts induced or repressed following erlotinib treatment of mutant EGFR-addicted cells were not selective to EGFR signaling . Downregulated genes included known direct transcriptional targets of ERK signaling ( CCND1 , FOSL1 , EGR1 , IER3 , IL-8 ) and shared feedback inhibitors of receptor tyrosine kinase ( RTK ) signaling ( DUSP6 ) ( Amit et al . , 2007 ) . This gene set overlaps with genes known to be differentially expressed following treatment of BRAF-mutant melanoma cells with a MEK inhibitor and exposure of EGFR-mutant lung cancer cells to an irreversible EGFR inhibitor ( Figure 1—figure supplement 3 ) ( Kobayashi et al . , 2006; Pratilas et al . , 2009 ) . In addition to SOX2 , the 12 other transcripts induced by EGFR inhibition included genes encoding metabolizing enzymes ( CYP1B1 , CYP1A1 ) that are normally induced by treatment with a variety of small chemical entities , genes that we also found to be equally well induced by inhibition of downstream signaling pathways ( MEK/MAPK inhibition using AZD6244 and PI3K/mTOR inhibition using BEZ235 ) , one transcript ( CCNG2 ) previously described in another EGFR model ( Kobayashi et al . , 2006 ) and a long noncoding RNA ( NEAT1 ) . Taken all together , these results indicate that suppression of EGFR signaling in mutant EGFR-addicted lung cancer cells is highly specific in triggering transcriptional induction of SOX2 . SOX2 encodes a master transcriptional regulator , implicated in stem cell maintenance and iPS cell generation . It is also required for upper aerodigestive tract development , and is known to be amplified in a subset of esophageal and squamous lung cancers , although it has not been previously implicated in lung adenocarcinomas , including the subset driven by mutant EGFR ( Ellis et al . , 2004; Gontan et al . , 2008; Bass et al . , 2009 ) . Remarkably , SOX2 expression following exposure of HCC827 cells to erlotinib was transient , peaking at 24 hr after exposure to therapeutic levels of the drug ( Figure 1D ) . Thereafter , SOX2 expression returned to basal levels despite continued erlotinib treatment in surviving cells ( Figure 1D , Figure 1—figure supplement 4 ) . The level of SOX2 induction in cultured cells exposed to erlotinib showed considerable heterogeneity , with a subset of cells ( ∼20% , with some experimental variability ) expressing high levels ( Figure 2A ) . The SOX2+ fraction was not increased by higher drug dosage , beyond that required for full inhibition of EGFR ( Figure 2—figure supplement 1 ) . Given the link between SOX2 expression and cellular reprogramming , we first asked whether cells with the high SOX2 expression represent a subset with stem cell markers . However , SOX2 expression did not correlate with expression of the putative stem cell markers CD133 , CD44 , CD24 , OCT-4 , or KLF-4 ( Figure 2—figure supplement 2 ) nor did microarray-based expression profiling of high SOX2-sorted cells identify a stem-like signature ( data not shown ) . Nonetheless , SOX2-expressing cells had a very low proliferative index , as measured by Ki67 staining ( 0 . 5% of Ki67+/SOX2+ vs 51% Ki67+/SOX2− HCC827 cells at baseline [p = 0 . 015]; and 0 . 15% Ki67+/SOX2+ vs 6 . 4% Ki67+/SOX2− cells following erlotinib [p < 0 . 0001] ) ( Figure 2C , Figure 2—figure supplement 3 ) . 10 . 7554/eLife . 06132 . 008Figure 2 . Induction of SOX2 in erlotinib-treated cells . ( A ) Left , HCC827 ( upper ) or PC9 ( lower ) cells were treated with 0 . 1 µM erlotinib for 24 hr , followed by immunofluorescence staining using an antibody to SOX2 and DAPI . For erlotinib-treated cells ( middle and right pairs of images ) , the heterogeneity in induced SOX2 levels per cell in each population is indicated by dashed outlines indicating DAPI+ nuclei lacking SOX2 , white arrows for nuclei with low ( but detectable ) SOX2 and green arrows for nuclei with high SOX2 . Right , distribution of SOX2 fluorescence in each sample . Mean fluorescence counts for each cell were quantitated and normalized to exposure time as described in ‘Materials and methods’ . p < 0 . 0001 for the comparison of erlotinib-treated cells to DMSO ( Student's t-test , unequal variances , N = 1219–3485 , means are 0 . 005/0 . 04 for untreated/treated HCC827 and 0 . 001/0 . 008 for untreated/treated PC9 , % SOX2+ is shown ) . Source data are included as Figure 2—source data 1 , 2 . ( B ) Induction of SOX2 in erlotinib-retreated cells . HCC827 cells were treated with 1 . 0 µM erlotinib for 24 hr ( 75% cell killing ) , followed by removal of drug , replating of cells after 7 days of recovery and retreatment with the same concentration of erlotinib . This protocol was repeated , and then cells were treated a third time with erlotinib for 24 hr and analyzed by immunofluorescence microscopy using antibodies to SOX2 and DAPI . The increase in SOX2-positive cells was highly significant for both erlotinib pretreated and untreated cells , but no enrichment was observed as a consequence of pretreatment p < 0 . 0001 for the comparison of erlotinib-treated cells to DMSO ( Student's t-test , unequal variances , N = 1106–2143 , means are 0 . 005/0 . 17 for untreated/treated-parental , 0 . 007/0 . 2 for untreated/treated-pretreated , % SOX2+ is shown ) . Source data are included as Figure 2—source data 3 . ( C ) Immunofluorescence analysis of single colonies formed from single clones of HCC827 cells stained for DAPI ( blue ) , SOX2 ( green ) , and Ki67 ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 00810 . 7554/eLife . 06132 . 009Figure 2—source data 1 . Raw immunofluorescence data for quantitation of SOX2 staining in HCC827 cells with erlotinib treatment in Figure 2A , and SOX2+ Ki67 staining in Figure 2—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 00910 . 7554/eLife . 06132 . 010Figure 2—source data 2 . Raw immunofluorescence data for quantitation of SOX2 staining in PC9 cells with erlotinib treatment in Figure 2A and Figure 1—figure supplement 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 01010 . 7554/eLife . 06132 . 011Figure 2—source data 3 . Raw immunofluorescence data for quantitation of SOX2 staining in HCC827 cells recovered after retreatment ( x2 ) with erlotinib , compared to previously untreated , in Figure 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 01110 . 7554/eLife . 06132 . 012Figure 2—source data 4 . Raw immunofluorescence data for quantitation of SOX2 staining in HCC827 and PC9 cells with increasing dose of erlotinib in Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 01210 . 7554/eLife . 06132 . 013Figure 2—source data 5 . Raw immunofluorescence data for quantitation of SOX2 staining in PC9 cells recovered after retreatment ( x2 ) with erlotinib , compared to previously untreated cells , in Figure 2—figure supplement 4A . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 01310 . 7554/eLife . 06132 . 014Figure 2—source data 6 . Raw immunofluorescence data for quantitation of phospho-EGFR ( pY1068 ) in parental and erlotinib-resistant PC9 cells in Figure 2—figure supplement 4B . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 01410 . 7554/eLife . 06132 . 015Figure 2—figure supplement 1 . Increasing the dose of erlotinib does not significantly increase the fraction of SOX2+ cells . HCC827 ( left ) and PC9 ( right ) cells were treated for 24 hr with the indicated dose of erlotinib , followed by immunofluorescence microscopy with antibodies to SOX2 and DAPI . The distribution of SOX2+ cells is shown . N = 607–1169 for HCC827 cells , 2746–6818 for PC9 , % SOX2+ is shown . Note: increased cell death at the highest dose of erlotinib precludes accurate determination of SOX2+ PC9 cells . Source data are included as Figure 2—source data 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 01510 . 7554/eLife . 06132 . 016Figure 2—figure supplement 2 . Stem cell markers do not colocalize with SOX2+ cells . Immunofluorescence microscopy was carried out on erlotinib-treated HCC827 ( upper panels ) or PC9 ( lower panels ) cells using antibodies to SOX2 and various stem cell markers . CD133 ( upper row of panels ) could be detected in a rare population of PC9 cells ( but not HCC827 cells ) that was clearly mutually exclusive from SOX2+ cells , while CD44 and MYC were expressed in the majority of cells , irrespective of SOX2 expression . Neither membrane localization of CD24 ( rightmost panel , middle rows ) nor expression of OCT4 and KLF4 ( data not shown ) could be detected in either cell line . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 01610 . 7554/eLife . 06132 . 017Figure 2—figure supplement 3 . SOX2 is expressed most highly in nonproliferative cells . Left panels , HCC827 cells were treated for 24 hr with 0 . 1 µM erlotinib , followed by immunofluorescence microscopy with antibodies to SOX2 , Ki-67 , and DAPI . The left three pairs of panels show HCC827 cells at 20× magnification , the right pair of panels shows a doublet of cells at 60× magnification . Right panel , the distribution of nuclear Ki-67 mean fluorescence in the SOX2− and SOX2+ cells in DMSO-treated ( green ) and erlotinib treated ( red ) cells is displayed . p = 0 . 015 and p < 0 . 0001 for the comparison of each SOX2+ population to the corresponding SOX2− population ( Student's t-test , unequal variances , N = 31–1834 , means of Ki67 fluorescence for SOX2 −/+ cells are 0 . 13/0 . 1 for DMSO-treated and 0 . 03/0 . 007 for erlotinib-treated ) . Source data are included as Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 01710 . 7554/eLife . 06132 . 018Figure 2—figure supplement 4 . Stochastic induction of SOX2 by erlotinib in PC9 cells . ( A ) Retreatment of PC9 cells after a period of recovery does not increase the fraction of cells capable of inducing SOX2 . Left panels , images of cells stained for SOX2 ( green ) and DAPI ( blue ) . Right panels , p < 0 . 0001 for the comparison of erlotinib-treated cells to DMSO ( Student's t-test , unequal variances , N = 3834–7951 cells , means of SOX2 fluorescence are 0 . 002/0 . 04 for untreated/treated-parental and 0 . 001/0 . 03 for untreated/treated-pretreated , % SOX2+ is shown ) . Note: the apparent decrease in SOX2+ fraction with retreatment is likely the result of partial selection for erlotinib resistance with serial retreatment , as shown in ( B ) below . Source data are included as Figure 2—source data 5 . ( B ) Induction of SOX2 in cells with acquired resistance to erlotinib . PC9 cells were made resistant to erlotinib by continuous culture in the presence of 0 . 1 µM drug for 30 days . Upper panel , qPCR for SOX2 expression was performed on lysates prepared from cells at the indicated time points . On day 30 , resistant cells were replated in the absence of drug and then retreated with increasing concentrations of erlotinib on day 31 . Data are shown as mean Ct of 4 replicates ( normalized to ACTB and untreated cells ) −/+ SEM . The first six data points are the same as in Figure 1D . Middle panels , parental and resistant PC9 cells were treated with increasing concentrations of erlotinib for 6 hr , followed by quantitative immunofluorescence analysis for pEGFR . N = 2073–6297 cells . Source data are included as Figure 2—source data 6 . Lower panels , representative images show strong decrease in pEGFR ( red ) with 1 µM erlotinib in parental ( left ) but not resistant ( right ) cells . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 01810 . 7554/eLife . 06132 . 019Figure 2—figure supplement 5 . The highest induction of SOX2 in individually isolated subclones of EGFR-mutant cells occurs in a subset of cells , as for the parental cells . ( A ) Immunofluorescence analysis of colonies formed from single HCC827 cells for DAPI ( blue ) , SOX2 ( green ) , and Ki67 ( red ) . ( B ) Similar analysis of DAPI ( blue ) and SOX2 ( green ) in PC9 cells . For HCC827 cells , because the cloning efficiency is extremely low ( <1% ) , a single cell suspension of HCC827 cells was plated at low density and allowed to form discrete colonies . PC9 cells were cloned by limiting dilution , and individual subclones were expanded prior to analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 01910 . 7554/eLife . 06132 . 020Figure 2—figure supplement 6 . KDM5A is not induced following treatment of PC9 cells with erlotinib for 24 hr . Data are shown as mean Ct ( normalized to GAPDH and untreated cells ) of 3 replicates −/+ SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 020 To test whether the heterogeneous induction of SOX2 following EGFR inhibition represents a stochastic event or a heritable property shared by a subset of the parental population , we treated cells sequentially with pulses of erlotinib . Retreatment of cells after a period of recovery produced similar heterogeneity of SOX2+ cells as the initial treatment , pointing to the absence of enrichment for highly inducible cells ( Figure 2B and Figure 2—figure supplement 4A ) . In addition , SOX2 could still be induced in cells made resistant to erlotinib , but only at the much higher doses of drug required to fully inhibit EGFR in resistant cells ( Figure 2—figure supplement 4B ) . Finally , cloning of 5–6 individual HCC827 and PC9 cells consistently generated mixed populations , including high level SOX2 inducers together with non-expressing cells , demonstrating that this heterogeneity is likely stochastic , rather than heritable ( Figure 2C and Figure 2—figure supplement 5 ) . Thus , erlotinib treatment of EGFR-mutant cells results in transient and heterogeneous induction of SOX2 , with a stochastic distribution , integrally tied to inhibition of EGFR , in which the cells with the highest expression have a low proliferative index . To test the physiological significance of SOX2 induction following withdrawal of mutant EGFR signaling , we first made use of mouse tumor models . The effectiveness of EGFR inhibitors in treating patients with EGFR-mutant NSCLC is well modeled in mouse xenograft assays , where oral administration of erlotinib for a few days is sufficient to cause massive regression of established tumors . We generated PC9 cell-derived subcutaneous tumors in nude mice and treated these with a single oral dose of 100 mg/kg erlotinib when the tumors had reached approximately 500 mm3 , harvesting tumors 24 hr after treatment . Immunohistochemical ( IHC ) analysis revealed minimal SOX2 expression in mock-treated xenografts ( mean 0 . 2 SOX2+ nuclei per field ) , but clearly increased ( and heterogeneous ) SOX2 positive cells after a single dose of erlotinib ( mean: 7 . 4 SOX2+ nuclei per field , N = 147–151 fields , p < 0 . 0001 ) ( Figure 3 ) . Similar studies in HCC827 cell-derived xenografts revealed low ( but detectable ) levels of SOX2 expression in mock-treated tumors; again , a single oral dose of erlotinib increased both the number of SOX2-positive cells and the level of SOX2 expression per nucleus ( Figure 3—figure supplement 1 ) . Thus , in a physiological setting that mimics the initial therapeutic response to EGFR inhibitors in EGFR-mutant NSCLC , treated cancer cells rapidly induce SOX2 . 10 . 7554/eLife . 06132 . 021Figure 3 . SOX2 is induced by erlotinib in a subset of EGFR-mutant cells in vivo . Nude mice were xenografted subcutaneously with PC9 cells and treated with a single oral dose of erlotinib ( 100 mg/kg ) ( or carrier ) when the tumors had reached ∼500 mm3 . Tumors were harvested 24 hr after treatment , and immunohistochemistry for SOX2 was carried out on formalin-fixed , paraffin embedded tumor specimens . The panels show automated scoring of SOX2+ nuclei ( brown ) as described in ‘Materials and methods’ . The actual IHC images for the areas indicated by rectangles are shown magnified to the right . p < 0 . 0001 for the comparison erlotinib-treated vs control ( Student's t-test , unequal variances , N = 147–151 fields from 4 xenografts , mean SOX2+ nuclei/10× field −/+ SEM is shown for each treatment ) . Source data are included as Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 02110 . 7554/eLife . 06132 . 022Figure 3—source data 1 . Number of SOX2+cells per field for quantitation of SOX2 staining in PC9 cell xenografts in Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 02210 . 7554/eLife . 06132 . 023Figure 3—source data 2 . Raw absorbance data for quantitation of SOX2 staining in HCC827 cell xenografts in Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 02310 . 7554/eLife . 06132 . 024Figure 3—figure supplement 1 . Erlotinib treatment results in induction of SOX2 in vivo . Quantitative immunohistochemistry for SOX2 was performed in FFPE sections from mice xenografted subcutaneously with HCC827 cells and treated with a single oral dose of erlotinib ( red in dot plot ) or carrier ( green ) . Because HCC827 cells possess a degree of basal SOX2 in vivo , the average mean fluorescence of tumor cells from each xenograft was used to determine significance . p < 0 . 0001 for the comparison of erlotinib-treated vs control ( Analysis of Variance to control for the variation among individual mice , N = 12 , 893–16 , 140 cells ) . Source data are included as Figure 3—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 024 While established cancer cell lines , such as PC9 and HCC827 , recapitulate the phenomenon of oncogene addiction , patient-derived cell lines directly cultured from biopsies may be more representative of heterogeneous primary cultures ( Crystal et al . , 2014 ) . Such biopsies are typically obtained at the time of disease progression , where defining drug resistance mechanisms may shape further therapy . We therefore analyzed short-term cultures of EGFR-mutant cells derived by re-biopsy of two patients who had initially responded to erlotinib therapy but subsequently developed progressive disease due to the acquisition of a T790M gatekeeper mutation . SOX2 induction was absent following treatment with erlotinib , which was ineffective in inhibiting EGFR in these resistant patient-derived cells ( Figure 4 ) . However , the novel ‘third line’ irreversible , EGFR-mutant-specific inhibitor WZ4002 demonstrated potent EGFR inhibition in these cells , along with induction of SOX2 ( Figure 4 ) . Thus , SOX2 induction is consistently observed following acute withdrawal of EGFR signals in cancer models as well as in patient-derived cells that exhibit oncogene dependence on the EGFR pathway . 10 . 7554/eLife . 06132 . 025Figure 4 . SOX2 is induced by therapy targeting the resistance genotype in cell lines derived by rebiopsy of patients . Short-term cultures of tumor cells derived from patients at the time of acquired resistance ( both tumors EGFR genotype exon 19 deletion + T790M ) were treated with the indicated agents for 24 hr , followed by isolation of total RNA and qPCR for SOX2 transcript ( left panels ) or quantitative immunofluorescence analysis after staining with antibodies to SOX2 ( middle panels ) . The effect of each treatment on downstream signaling was determined by immunoblot analysis with the indicated antibodies ( right panels ) . For qPCR , data are displayed as the mean of 4 replicates −/+ SEM . For histograms , p-values are shown for the comparison of each treatment to DMSO ( Student's t-test , unequal variances , N = 229–1808 , means are 0 . 001/0 . 0014/0 . 0054 for DMSO/erlotinib/WZ4002-treated MGH134 and 0 . 003/0 . 003/0 . 011 for DMSO/erlotinib/WZ4002-treated MGH141 , % SOX2+ is shown ) . Source data are included as Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 02510 . 7554/eLife . 06132 . 026Figure 4—source data 1 . Raw immunofluorescence data for quantitation of SOX2 staining with different treatments in patient-derived tumor cells . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 026 To examine the consequence of ectopic SOX2 expression , we generated HCC827 cells with a lentiviral-driven , doxycycline-inducible construct . Careful titration of doxycycline-permitted induction of ectopic SOX2 to physiologic levels in the absence of erlotinib , comparable at the single cell level to endogenous SOX2 induction in the presence of erlotinib , though in a larger fraction of cells ( Figure 5A and Figure 5—figure supplement 1A ) . As expected , treatment of control HCC827 cells with erlotinib led to the inhibition of EGFR signaling , as measured by decreased phosphorylation of EGFR , AKT , and ERK , and caused dramatic apoptosis , as determined by increased PARP and caspase-3 cleavage ( Figure 5A , lanes 1–2 ) . In contrast , expression of ectopic SOX2 significantly decreased erlotinib-mediated apoptosis , resulting in decreased PARP and caspase-3 cleavage ( Figure 5A , lane 2 vs 4 ) . Exogenous SOX2 itself did not alter basal or erlotinib-inhibited phosphorylation of EGFR , AKT , or ERK ( Figure 5A , lanes 3–4 ) . 10 . 7554/eLife . 06132 . 027Figure 5 . Induction of SOX2 protects cells from erlotinib-induced apoptosis . ( A ) HCC827 cells were stably transduced with a doxycycline inducible epitope-tagged SOX2 lentiviral expression vector . Doxycycline was added for 3 hr ( ‘SOX2-tag’ ) and then removed prior to the addition of DMSO or 0 . 1 µM erlotinib for 24 hr , followed by immunoblot of protein lysates with the indicated antibodies . Exogenous SOX2 migrates more slowly than the endogenous protein due to the presence of the tag . ( B ) HCC827 ( left ) or PC9 ( right ) cells were transfected with control siRNA or siRNA targeting SOX2 . 24-hr after transfection , DMSO or 0 . 1 µM erlotinib was added . The effect of SOX2 knockdown was assessed by immunoblot analysis of protein lysates with the indicated antibodies after overnight treatment . ( C ) PC9 cells were stably transduced with a tagged SOX2 lentiviral vector in which silent mutations were introduced into the target site for the most potent siRNA against SOX2 . Cells were transfected with the indicated siRNAs , treated with doxycycline followed by erlotinib as in ( A ) , and protein lysates were analyzed by immunoblot with the indicated antibodies after overnight treatment . The increased PARP and caspase-3 cleavage observed when erlotinib treatment is combined with siRNA targeting SOX2 ( lane 4 ) is suppressed by siRNA-resistant , exogenous SOX2 ( lane 6 ) . ( D ) PC9 and HCC827 cells were transfected in 96-well plates and treated 24 hr later with a dilution series of erlotinib , followed by Syto-60 assay . Data are displayed as the mean of 3–5 replicates −/+ SEM . p = 0 . 001 ( PC9 ) and 0 . 04 ( HCC827 ) for the comparison of mean IC50 for siCTRL vs siSOX2 ( Student's t-test , unequal variances ) . ( E ) Preventing SOX2 induction using siRNA decreases the development of acquired erlotinib resistance . PC9 cells were transfected with control siRNA or siRNA targeting SOX2 , followed by treatment after 24 hr with 1 . 0 µM erlotinib . Erlotinib-containing medium was renewed every 3 days , and plates were fixed and stained with Crystal Violet at the indicated times . The left panels demonstrate the absence of toxicity following transfection with siRNA targeting SOX2 in the absence of erlotinib . Middle panels demonstrate cell loss after 3 days of treatment , due to erlotinib-induced apoptosis; at higher magnification , more control cells remain attached than cells transfected with siRNA against SOX2 . Right panels show colonies of proliferating cells after 2 weeks of continuous erlotinib treatment . p < 0 . 0001for the number of cells per 20× field ( N = 33 fields per sample ) or the number of colonies per plate ( N = 9 plates per sample from three independent experiments ) , for siRNA targeting SOX2 vs control cells ( Student's t-test , unequal variances ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 02710 . 7554/eLife . 06132 . 028Figure 5—source data 1 . Raw immunofluorescence data for quantitation of SOX2 staining in HCC827 cells with inducible SOX2 in Figure 5—figure supplement 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 02810 . 7554/eLife . 06132 . 029Figure 5—source data 2 . Raw immunofluorescence data for quantitation of SOX2 and cleaved caspase-3 costaining in PC9 cells transfected with siCTRL or siSOX2 in Figure 5—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 02910 . 7554/eLife . 06132 . 030Figure 5—figure supplement 1 . Induction of exogenous SOX2 . ( A ) Quantitative immunofluorescence analysis confirms that ectopic SOX2 expression with a short pulse of doxycycline increases the fraction of SOX2+ cells without significantly increasing the amount of SOX2 per cell . The same conditions were used as in Figure 5A p < 0 . 0001 , for the comparison of erlotinib-treated to DMSO ( Student's t-test , unequal variances , N = 830–1115 cells , means of SOX2 fluorescence are 0 . 0007/0 . 03 for untreated/treated cells without exogenous SOX2 and 0 . 03/0 . 06 for untreated/treated cells with exogenous SOX2 , % SOX2+ is shown ) . Source data are included as Figure 5—source data 1 . ( B ) Left , HCC827 cells were stably transduced with a tetracycline-inducible SOX2 expression vector ( without an epitope tag ) . Cells were treated with carrier ( lanes 1–4 ) or 0 . 1 µg/ml doxycycline ( lanes 5–8 ) for 3 hr prior to addition of 0 . 1 µM erlotinib , with doxycycline left in the cell culture media during erlotinib treatment to ensure strong induction of exogenous SOX2 . Protein lysates were prepared at 3-hr intervals , and immunoblot analysis was carried out with the indicated antibodies . The increase in the intensity of the SOX2 band in lane 1–4 with time is due to induction of endogenous SOX2 by erlotinib; in lanes 5–8 , strong induction of exogenous SOX2 with increasing time in doxycycline masks the endogenous protein . Right , the levels of each BIM isoform were quantitated using ImageJ software and normalized to GAPDH and time zero , confirming suppression of their induction by ectopic SOX2 . ( C ) Physiologic levels of ectopic SOX2 decrease BIM induction by erlotinib . Cells were treated with erlotinib for 12 or 24 hr , −/+ doxycycline to induce tagged SOX2 , followed by immunoblot of lysates . Here , doxycycline was added for 3 hr and then removed prior to addition of erlotinib , to maintain ectopic SOX2 induction at physiologic levels during erlotinib treatment . Therefore , ectopic SOX2 is ( slightly ) apparent at time zero when erlotinib is added ( lane 4 ) , peaks 12 hr later ( lane 5 ) and has decreased at 24 hr ( lane 6 ) . However , decreased BIM levels , together with decreased cleaved PARP/caspase-3 ( lanes 5–6 vs 2–3 ) , are apparent throughout . The use of a lower percentage acrylamide gel does not permit BIM L and S to be distinguished from BIM EL . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 03010 . 7554/eLife . 06132 . 031Figure 5—figure supplement 2 . SOX2 expression modulates erlotinib-induced apoptosis . Left panel , quantitative immunofluorescence analysis showing expression of SOX2 ( x-axis ) and cleaved caspase-3 ( y-axis ) in PC9 cells transfected with siCTRL ( blue ) or siSOX2 ( red ) and treated with erlotinib ( N = 2452–3792 ) . Knockout of SOX2 results in decreased SOX2 expression and increased cleaved caspase-3 . Right panels , representative immunofluorescence images from erlotinib-treated cells showing DAPI ( blue ) , SOX2 ( red ) , and cleaved-caspase-3 ( green ) staining . Source data are included as Figure 5—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 03110 . 7554/eLife . 06132 . 032Figure 5—figure supplement 3 . The effect of siRNA targeting SOX2 is specific . ( A ) , PC9 cells were transfected with two different siRNA duplexes targeting SOX2 ( or control siRNA ) , followed by addition of DMSO or 0 . 1 µM erlotinib for 24 hr and immunoblot of protein lysates with the indicated antibodies . ( B ) , the degree of knockdown of SOX2 was quantitatively assessed by qPCR . Data are shown as mean Ct ( normalized to GAPDH and untreated siCTRL cells ) of 3 replicates −/+ SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 03210 . 7554/eLife . 06132 . 033Figure 5—figure supplement 4 . Quantitation of the effect of SOX2 knockdown on BIM levels . Each BIM isoform ( EL , L , and S ) was normalized to the GAPDH loading control and untreated , siCTRL cells . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 033 To determine the functional significance of endogenous SOX2 induction , we next used siRNAs to block its induction in erlotinib-treated cells . While both HCC827 and PC9 cells are highly sensitive to EGFR inhibition at baseline , SOX2 knockdown further increased erlotinib-induced apoptosis , as determined by PARP and caspase-3 cleavage assays and by cell enumeration ( Figure 5B , D and Figure 5—figure supplement 2 ) . The apoptotic effect of the most potent siRNA was rescued by expression of an ectopic siRNA-resistant SOX2 construct ( Figure 5C ) and individual siRNAs-induced apoptosis in proportion to the degree of knockdown ( Figure 5—figure supplement 3 ) , confirming the specificity of the effect . As with expression of exogenous SOX2 , endogenous SOX2 suppression itself did not have a consistent effect on EGFR signaling , as measured by phosphorylation of EGFR , AKT , or ERK ( Figure 5B ) . Given the effect of transient SOX2 induction in acutely modulating cell survival following erlotinib treatment of EGFR-mutant cells , we tested whether this translates into a longer term impact on acquired drug resistance . We used the well-characterized PC9 cell model of EGFR-mutant NSCLC to transfect siRNAs against SOX2 ( the same duplex validated by rescue in Figure 5C ) or control , followed by continuous exposure to 1 . 0 µM erlotinib ( Figure 5E ) . As expected , SOX2 knockdown alone had no significant toxicity , while erlotinib treatment led to massive cell death at 3 days ( Figure 5E , left and middle images ) . At high magnification , many more individual surviving cells remained in control transfected cells than after transfection with siRNA targeting SOX2 ( Figure 5E , middle insets ) . With continued incubation in erlotinib , drug-resistant colonies emerged in control transfected cultures at higher frequency than in those treated with siRNA targeting SOX2 ( Figure 5E , right ) . Consistent with the short-term duration of siRNA effectiveness , these results suggest that preventing the short-term induction of SOX2 allows fewer cells to survive the initial exposure to erlotinib , preventing an adaptation response that ultimately delays the emergence of erlotinib-resistant colonies . The mechanisms underlying the cell death response of cancer cells following withdrawal of oncogene-addicting signals are complex but appear to result in part from activation of the pro-apoptotic proteins BIM and PUMA ( Costa et al . , 2007; Gong et al . , 2007; Bean et al . , 2013 ) . To search for additional targets that mediate SOX2's anti-apoptotic effect , we screened for altered expression of multiple pro-apoptotic and anti-apoptotic proteins in cells with overexpression of ectopic SOX2 , following treatment with erlotinib . In both HCC827 and PC9 EGFR-mutant lung cancer cells , three of 14 BH3-domain containing proteins tested showed strongly reduced mRNA induction by erlotinib in the presence of ectopic SOX2: BIM , BMF , and HRK ( but not PUMA ) ( Figure 6A and Figure 6—figure supplement 1 ) . All three of these are known to induce apoptosis to different degrees , depending on the type of apoptotic stimulus . The effect of ectopic SOX2 on BIM protein levels was confirmed using two different expression strategies ( Figure 5—figure supplement 1B , C ) . Ectopic expression of SOX2 , both at high and physiological levels , attenuated the erlotinib-induced induction of SOX2 . In contrast , SOX2 knockdown coincident with erlotinib treatment led to an overall increase in BIM levels , although this effect was attenuated when averaged across a cell population with heterogeneous induction of endogenous SOX2 ( Figure 5—figure supplement 4 ) . 10 . 7554/eLife . 06132 . 034Figure 6 . Erlotinib-induced SOX2 directly regulates expression of BIM and BMF . ( A ) The levels of transcripts for each of the indicated BH3 domain-containing proteins was assessed by quantitative PCR at multiple time points after erlotinib treatment in uninduced PC9 cells ( blue lines ) and in cells in which expression of SOX2 was induced with doxycycline ( red lines ) , which was not removed prior to erlotinib addition in order to further increase SOX2 levels , as shown in Figure 5—figure supplement 1B . The y-axis maximum for all graphs is set to 4 except for HRK ( y maximum = 11 ) and BMF ( y maximum = 110 ) . Data are displayed as mean Ct of 4 replicates ( normalized to untreated cells and GAPDH ) −/+ SEM . ( B ) Upper panel , ChIP seq demonstrates SOX2 binding to BIM and BMF . HCC827 cells were treated with 0 . 1 µM erlotinib for 24 hr , followed by chromatin immunoprecipitation ( ChIP ) using anti-SOX2 antibody and ChIP Seq as described in ‘Materials and methods’ . ChIP seq signal tracks are displayed . Lower panel , HCC827 cells were left untreated or were treated with 0 . 1 µM erlotinib for 24 hr , followed by chromatin immunoprecipitation using anti-SOX2 antibody or IgG as a negative control , and qPCR with control primers or primers within the peaks indicated by the gray boxes in the ChIP seq tracks . Data are displayed as mean Ct value ( normalized to input chromatin ) of 4 replicates −/+ standard error , p-values for the comparison of untreated vs erlotinib treated cells are shown ( Student's t-test , unequal variances ) . ( C ) Knockdown of BIM and BMF decreases apoptosis . Left panel , PC9 cells were transfected with siRNA constructs targeting BIM and BMF ( alone or together ) or a control siRNA . 24 hr after transfection , cells were treated with DMSO or 0 . 1 µM erlotinib . 24 hr after treatment , protein lysates were prepared , and immunoblot was performed with the indicated antibodies . Right panel , the efficiency of knockdown of BMF was confirmed by qPCR . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 03410 . 7554/eLife . 06132 . 035Figure 6—figure supplement 1 . Effect of SOX2 overexpression on apoptotic regulators . Induction of BIM , BMF , and HRK following erlotinib treatment of HCC827 cells was assessed as in Figure 6A . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 035 To determine whether BIM and BMF are endogenous SOX2 target genes , we carried out chromatin immunoprecipitation sequencing ( ChIP seq ) experiments , using erlotinib-treated HCC827 cells . Strong SOX2-binding peaks were present within ∼2 kB of the transcriptional start sites ( TSS ) of both BIM and BMF genes ( for BMF , the peak spans the TSS; for BIM , it is located within the first intron ) ( Figure 6B , upper ) . Importantly , ChIP qPCR demonstrated significant enhancement of SOX2 binding with erlotinib treatment at both peaks compared to untreated cells , suggesting that binding is functionally significant ( Figure 6B , lower ) . Consistent with their functional roles , knockdown of either BIM or BMF ( but not HRK ) decreased erlotinib-induced apoptosis , with combined BIM and BMF knockdown displaying an additive effect ( Figure 6C ) . The role of BMF could be even more important , since BMF knockdown increases BIM levels , which may blunt the anti-apoptotic effect of BMF loss ( Figure 6C , left ) . Together , these results suggest that SOX2 induction following erlotinib exposure in EGFR-mutant cells suppresses transcriptional induction of the BH3-only BIM and BMF genes , which contribute to apoptosis following oncogene withdrawal . To search for mediators of SOX2 induction , we explored the Molecular Signatures and TRANSFAC databases for transcription factor target sequences within the promoters of the 12 highest erlotinib-induced genes ( Wingender et al . , 2000; Subramanian et al . , 2005 ) . Several binding motifs for FOXO proteins were highly significantly enriched ( q-value = 0 . 003 or less ) : for SOX2 , multiple sites were present within 2 kb of the transcriptional start site ( Figure 7A and Figure 7—figure supplement 1 ) . Expression of all of the FOXO family members was detectable at baseline in HCC827 cells and erlotinib treatment ( 8 hr ) was associated with a 1 . 6–4 . 4-fold induction ( Figure 7B ) , as well as with loss of the AKT-mediated inhibitory N-terminal threonine phosphorylation of the FOXO proteins ( Figure 7—figure supplement 2A ) . 10 . 7554/eLife . 06132 . 036Figure 7 . SOX2 expression in EGFR-mutant cells is regulated by FOXO6 . ( A ) Putative FOXO protein binding sites within the promoter of SOX2 , identified using TRANSFAC and Zhang et al . , 2011 . ( B ) HCC827 cells were transfected with control siRNA or siRNA targeting the indicated FOXO proteins ( alone or FOXOs 1 , 3a and 4 in combination ) . 72-hr after transfection , DMSO or 0 . 1 µM erlotinib was added for 8 hr , and the levels of SOX2 and FOXO mRNAs were determined by qPCR . Data are shown as mean Ct ( normalized to GAPDH and untreated siCTRL cells ) of 3 replicates −/+ SEM . Only knockdown of FOXO6 results in significantly decreased induction of SOX2 mRNA by erlotinib compared to siCTRL cells . p < 0 . 0001 , other siFOXOs are without significant decrease ( Student's t-test , unequal variance ) . Although siRNA pools targeting FOXOs 3a and 4 also decrease FOXO1 , the lack of a SOX2 effect with specific FOXO1 knockdown argues against their role in regulation of SOX2 . ( C ) The effect of FOXO6 knockdown on induction of SOX2 in HCC827 cells is shown by immunofluorescence after staining of cells with SOX2 and DAPI ( upper panels ) and quantitated for knockdown of all of the FOXO isoforms ( lower panel ) . Only knockdown of FOXO6 significantly decreases induction of SOX2 by erlotinib compared to siCTRL cells . p < 0 . 0001 , other siFOXOs are without significant decrease ( N = 766–1027 cells , Student's t-test , unequal variances ) . Source data are included as Figure 7—source data 1 . Knockdown efficiency is demonstrated in Figure 7B . ( D ) Multiple different siRNAs effectively targeting FOXO6 block erlotinib-mediated induction of SOX2 . HCC827 ( upper ) or PC9 ( lower ) cells were transfected with control siRNA or four different siRNA duplexes targeting FOXO6 , treated with 0 . 1 µM erlotinib for 24 hr and mRNA was analyzed by qPCR . Data are shown as mean Ct ( normalized to ACTB and untreated siCTRL cells ) of 4 replicates −/+ SEM . p < 0 . 0001 for the comparison of each FOXO6 siRNA to siCTRL for both SOX2 and FOXO6 ( Student's t-test , unequal variances ) . The effect of each siRNA on the levels of the other FOXO isoforms is shown in Figure 7—figure supplement 2C . ( E ) Knockdown of FOXO6 has minimal effects on other downstream components of EGFR signaling . HCC827 cells were transfected with siRNA targeting FOXO6 and treated with 0 . 1 µM erlotinib overnight followed by immunoblot analysis of protein lysates with the indicated antibodies . Knockdown of FOXO6 is demonstrated in Figure 7—figure supplement 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 03610 . 7554/eLife . 06132 . 037Figure 7—source data 1 . Raw immunofluorescence data for quantitation of SOX2 staining with different FOXO protein knockdown in Figure 7C . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 03710 . 7554/eLife . 06132 . 038Figure 7—source data 2 . Raw immunofluorescence data for quantitation of SOX2 and FOXO6 costaining in HCC827 cells in Figure 7—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 03810 . 7554/eLife . 06132 . 039Figure 7—figure supplement 1 . Recurrent FOXO binding sites in erlotinib-induced genes . ( A ) MSigDB/TRANSFAC output for the 12 genes most highly upregulated by erlotinib ( FDR <0 . 05 ) . ( B ) Although binding sites for FOXF2 are also enriched , knockdown of FOXF2 does not decrease erlotinib-mediated induction of SOX2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 03910 . 7554/eLife . 06132 . 040Figure 7—figure supplement 2 . FOXO6 uniquely regulates SOX2 expression . ( A ) Same lysates as Figure 7E , showing immunoblot for FOXO proteins . ( B ) Similar data as in Figure 7B , but shown after immunoblot of protein lysates with the indicated antibodies . Immunoblot for FOXO4 was consistently unable to detect a band of the correct size ( 65 kD ) . ( C ) Effect of individual siRNA duplexes targeting FOXO6 on the levels of other FOXO isoforms . Data are presented as in Figure 7D . Individual siFOXO6 siRNA duplexes −01 and −02 do not alter the levels of FOXOs 1 , 3a or 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 04010 . 7554/eLife . 06132 . 041Figure 7—figure supplement 3 . Distribution of FOXO6 vs SOX2 nuclear staining . ( A ) HCC827 cells were left untreated or were treated with 0 . 1 µM erlotinib for 24 hr . Cells were stained with goat anti-SOX2 and rabbit anti-FOXO6 primary antibodies , followed by anti-goat-Alexa Fluor 488 ( green ) and anti-rabbit-Alexa Fluor 647 ( red ) secondary antibodies ( and DAPI in blue ) . FOXO6 appears to colocalize ( yellow in the leftmost panels ) with SOX2 in cells with the highest expression of the latter , especially in erlotinib-treated cells ( arrows ) . ( B ) Quantitative immunofluorescence analysis demonstrates a positive correlation between FOXO6 and SOX2 nuclear fluorescence in individual cells ( Correlation coefficient R untreated/treated = 0 . 7/0 . 5 , N = 1700/835 cells ) . Source data are included as Figure 7—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 04110 . 7554/eLife . 06132 . 042Figure 7—figure supplement 4 . Assessing the role of previously identified regulators on erlotinib-induced expression of SOX2 . ( A ) Pre-treatment of PC9 cells with FGF10 has minimal effects on SOX2 induction by erlotinib . ( B ) The addition of exogenous Wnt3A has no effect on induction of SOX2 by erlotinib . ( C ) The beta-catenin pathway does not regulate SOX2 expression . HCC827 cells were stably transduced with inducible lentiviral constructs expressing a dominant negative TCF4 transgene ( DN TCF4 ) or the constitutively activated S33Y variant of Beta-Catenin ( S33Y B-Cat ) , and with a lentiviral TOP FLASH reporter . Left , the expected activity of each transgene was confirmed by TOP FLASH luciferase assay in the absence or presence of the GSK3 inhibitor/Beta-Catenin activator BIO ( and −/+ erlotinib ) . Representative wells after luciferase imaging are shown above the graph . Right , the effect of each transgene on the levels of SOX2 induction by erlotinib compared to control ( GUS ) cells is minimal . ( D ) Knockdown of TTF1 ( NKX2 . 1 ) with siRNA has minimal effects on the degree of SOX2 induction by erlotinib . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 042 Given the evidence of erlotinib-mediated FOXO activation and its potential regulation of SOX2 , we tested the consequence of siRNA-mediated knockdown of each gene family member , alone and in combination . Knockdown of FOXO6 using pooled siRNA constructs , but not the other FOXO proteins ( individually or simultaneously ) , dramatically reduced erlotinib-mediated induction of SOX2 ( Figure 7B , C and Figure 7—figure supplement 2B ) . The effect of FOXO6 knockdown on SOX2 was evident using multiple individual siRNAs targeting FOXO6 in both HCC827 and PC9 cells ( Figure 7D; effect on other FOXO isoforms is shown in Figure 7—figure supplement 2C ) . Although some individual siRNAs targeting FOXO6 had off target effects on other FOXOs , direct targeting of FOXOs 1 , 3a , and 4 had no effect on SOX2 expression , further supporting the specificity of the FOXO6 effect ( Figure 7B ) . The effect of FOXO6 knockdown on SOX2 was not associated with any consistent effect on other aspects of EGFR signaling , although a moderate decrease in phospho-EGFR was observed in some experiments without significant differences in phospho-AKT or phospho-ERK ( Figure 7E ) . Notably , FOXO6 expression was also heterogeneous and partially colocalized with SOX2 expression among populations of both untreated and treated cells ( Figure 7—figure supplement 3 ) . FOXO6 differs from other FOXO proteins in that even the inactive protein is localized in the nucleus ( Jacobs et al . , 2003; van der Heide et al . , 2005 ) . However , it shares the FOXO protein AKT-dependent inhibitory phosphorylation , whose suppression following repression of mutant EGFR signaling may in part explain the erlotinib-mediated FOXO6 activation . Virtually all EGFR-mutant lung cancer cell lines established from patients who have not been treated with erlotinib are highly sensitive to this drug , although a few cells lines appear to be intrinsically resistant . The HCC2935 human lung cancer cell line is remarkable for harboring a characteristic oncogene-addicting EGFR mutation ( exon 19 deletion ) yet having unexplained resistance to erlotinib ( Figure 8A ) , including absence of the common T790M gatekeeper mutation within EGFR and no amplification of the MET bypass signaling pathway ( Zheng et al . , 2011 ) . Notably , SOX2 expression at baseline is detectable in 90% of HCC2935 cells ( compared to 3% of HCC827 and <1% PC9 cells ) , and its expression per cell is further increased upon EGFR inhibition ( Figure 8B ) . To test whether increased SOX2 contributes to decreased erlotinib sensitivity in HCC2935 cells , we knocked down SOX2 using siRNA . A striking increase in erlotinib cytotoxicity was evident following SOX2 suppression ( IC50 0 . 8 µM for siCTRL cells with no further toxicity up to 10 µM , IC50 0 . 1 µM for siSOX2 cells ) , associated with higher levels of BIM and increased PARP and caspase-3 cleavage ( Figure 8A , C ) . Thus , increased baseline SOX2 contributes to erlotinib resistance in these EGFR-mutant cells . 10 . 7554/eLife . 06132 . 043Figure 8 . Knockdown of SOX2 sensitizes HCC2935 cells to erlotinib-induced apoptosis . ( A ) HCC2935 cells were transfected with siCTRL or siSOX2 48 hr prior to erlotinib addition and assayed for cytoxicity 48 hr later with Syto-60 . Data are displayed as the mean of 5 replicates −/+ SEM . The IC50 is 0 . 8 µM for siCTRL and 0 . 1 µM for siSOX2 cells ( calculated by four parameter logistic sigmoidal fit ) . p = 0 . 003 for the comparison of mean IC50 for siCTRL vs siSOX2 ( Student's t-test , unequal variances ) . ( B ) Upper panels , images of untreated and erlotinib-treated HCC2935 cells , demonstrating SOX2 expression in the majority of cells . Lower panels , the distribution of SOX2 in HCC2935 was determined by quantitative immunofluorescence microscopy . p < 0 . 0001 for the comparison of mean SOX2 fluorescence in untreated vs treated cells ( Student's t-test , unequal variances , N = 3342/1181 , means for SOX2 fluorescence are 0 . 17/0 . 24 for untreated/treated cells , % SOX2+ is shown ) . Source data is included as Figure 8—source data 1 . ( C ) HCC2935 cells were transfected with control siRNA or siRNA targeting SOX2 . 48 hr after transfection , DMSO or 1 . 0 µM erlotinib was added . The effect of SOX2 knockdown was assessed by immunoblot analysis of protein lysates with the indicated antibodies after overnight treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 04310 . 7554/eLife . 06132 . 044Figure 8—source data 1 . Raw immunofluorescence data for quantitation of SOX2 staining in HCC2935 cells in Figure 8B . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 044
The regulation of SOX2 expression by FOXO6 , whose activation is normally repressed by EGFR signaling , is consistent with the critical role played by FOXO proteins as integrators of cellular signaling pathways . The best studied isoform , FOXO1 , is highly expressed in ES cells and has been implicated in the maintenance of pluripotency through activation of SOX2 transcription ( Zhang et al . , 2011 ) . All FOXO proteins bind to similar DNA sequences , with isoform-specific activity presumably conferred by cellular and promoter context ( Furuyama et al . , 2000 ) . Indeed , all four FOXO isoforms are expressed in EGFR-mutant lung cancer cells , transcriptionally induced following EGFR inhibition and phosphorylated at homologous Serine residues , yet only FOXO6 regulates SOX2 in these cells . Given variability in knockdown efficacy ( Figure 7B , column 6 ) , we cannot exclude some contribution from the other FOXO family members on SOX2 expression , but FOXO6 has the dominant effect in the cells tested . Activation of FOXO6 may occur through both AKT dependent and independent pathways ( Jacobs et al . , 2003; van der Heide et al . , 2005 ) , and indeed we observed that treatment with PI3K inhibitors alone is insufficient for induction of SOX2 ( Figure 1B ) . Other pathways that have been implicated in SOX2 regulation in the developing lung , including FGF10 , WNT/beta-Catenin signaling , and TTF1 ( Que et al . , 2007; Gontan et al . , 2008; Hashimoto et al . , 2012 ) , had relatively modest effects on its induction following erlotinib treatment in EGFR-mutant cancer cells ( Figure 7—figure supplement 4 ) , pointing to FOXO6 as the dominant pathway in this model of oncogene-dependent signaling . Our study extends the pro-apoptotic signals implicated in withdrawal of mutant EGFR signaling to include BMF , in addition to BIM and PUMA ( Gong et al . , 2007; Bean et al . , 2013 ) . In contrast to the latter , BMF binds with significant affinity to a subset of BCL-2 family members ( BCL-2 , BCL-xL , and BCL-w ) ( Chen et al . , 2005; Kuwana et al . , 2005 ) , yet it clearly contributes to the apoptotic response to erlotinib ( Figure 6 ) . The direct binding by SOX2 of the BIM and BMF genes is consistent with the ability of the four reprogramming factors ( SOX2 , OCT-4 , KLF-4 , and c-MYC ) to bind to the promoters of several anti-apoptotic genes ( including BMF ) which are induced early during the reprogramming process ( Kim et al . , 2008; Soufi et al . , 2012 ) . Interestingly , control of SOX2 expression and its modulation of apoptosis may differ in EGFR-mutant lung cancer , compared with other forms of NSCLC . In a recent study of lung cancers with wild-type EGFR , the activation of EGFR upregulated SOX2 , thereby decreasing apoptosis through BCL2L1 , a phenomenon that was notably absent in EGFR-mutant lung cancers ( Chou et al . , 2013 ) . A similar pathway was reported in a prostate cancer cell line , also with wild-type EGFR ( Rybak and Tang , 2013 ) . Thus , an EGFR-SOX2-BCL2L1 pathway may be implicated in cancer cells with wild-type EGFR , whereas the EGFR-FOXO6-SOX2-BIM/BMF pathway we describe is specific for cells that are dependent on mutant EGFR signals for their survival . The heterogeneous induction of SOX2 within a clonally derived cell population could reflect the existence of an intrinsic subpopulation with heritable traits , as recently proposed in medulloblastoma and skin carcinoma ( Boumahdi et al . , 2014; Vanner et al . , 2014 ) . Alternatively , it could result from stochastic variation between cells in the activity of cellular signaling pathways . We favor the latter model , given the absent coexpression of SOX2 with putative stem cell markers , the failure to enrich for SOX2 positive cells following repeated erlotinib treatments , and the regeneration of heterogeneous SOX2 inducibility following single cell cloning experiments . A stochastic cell killing model for TRAIL-induced apoptosis of HeLa cells was recently described , in which naturally occurring cell-to-cell variation in the levels or activity of upstream signaling proteins leads to differential induction of mitochondrial membrane permeability and apoptosis only in some cells ( Spencer et al . , 2009 ) . Critically , the ability of siRNA targeting SOX2 to substantially decrease the number and rate at which resistant subclones of EGFR-mutant cells emerge following continuous erlotinib treatment suggests that , despite its heterogeneous , transient , and stochastic expression , SOX2 contributes to the emergence of stably acquired resistance ( Figure 5E ) . In these respects , our observations are reminiscent of transient drug-tolerant persister cells ( DTPs ) , also observed following erlotinib treatment ( Sharma et al . , 2010 ) . However , there are also significant differences between the mechanisms underlying DTPs and those observed here . DTPs emerge following longer erlotinib exposure at much higher concentrations , and they constitute a lower percentage of cells within the population . Their expression of the stem cell marker CD133 and the chromatin remodeling protein KDM5A was not reproduced by SOX2+ cells ( Figure 2—figure supplements 2 , 6 ) . Furthermore , DTPs are detectable in many cancer cell lines following treatment with multiple cytotoxic and targeted agents , whereas induction of SOX2 appears to be strictly limited to targeted EGFR inhibition in cells addicted to mutant EGFR signaling . SOX2 induction may thus be a specific and early signaling response to EGFR withdrawal , which contributes to increased cell survival , thus enhancing the likelihood of epigenetic events leading to DTPs and ultimately to stable genetic mechanisms of acquired drug resistance . Our results with HCC2935 cells ( Figure 8 ) further suggest that in a subset of cases , high basal SOX2 may even blunt the initial signaling response to EGFR inhibitors . In summary , the rewiring of cellular signaling pathways driven by a dominant mutationally activated kinase underlies oncogene addiction in cancer , providing powerful opportunities for targeted therapy . At the same time , feedback pathways that attenuate these signaling readouts may play a major role in enhancing the development of drug resistance . The EGFR-FOXO6-SOX2 signaling pathway regulating expression of the BIM and BMF apoptotic factors thus identifies a feedback loop that may attenuate the effectiveness of anti-EGFR therapy in cancer and contribute toward the ultimate development of drug resistance ( Figure 9 ) . The contribution of key embryonic regulators such as SOX2 points to the conservation of critical developmental pathways in cancer cells , which modulate their response to the disruption of oncogenic signals . 10 . 7554/eLife . 06132 . 045Figure 9 . Model of SOX2 feedback signaling pathway . In untreated cells , mutant EGFR drives cell survival by activating downstream signaling pathways , including PI3K and MAPK , which inhibit apoptosis through transcriptional and post-transcriptional effects on BH3-domain proteins , including pro-apoptotic BIM and BMF . In most cells ( red lines ) , erlotinib treatment results in EGFR inhibition , inhibition of downstream signaling and increased pro-apoptotic proteins , leading to apoptosis . The high SOX2 induced by erlotinib through activation of FOXO6 in some cells ( blue lines ) counteracts the pro-apoptotic effects of EGFR inhibition , sufficiently decreasing the levels of BIM and BMF to delay the apoptotic response . DOI: http://dx . doi . org/10 . 7554/eLife . 06132 . 045
All cell lines were grown in RPMI ( GIBCO ) with 10% FBS and were obtained from the ATCC or the Massachusetts General Hospital Center for Molecular Therapeutics , which performs routine cell line authentication testing by SNP and STR analysis . Erlotinib , AZD6244 , BEZ235 , crizotonib , lapatinib , WZ4002 ( Selleckchem , Houston , TX ) , and BIO ( Sigma , St . Louis , MO ) were dissolved in DMSO . HCC827 cells were plated in triplicates and treated the following day for 6 hr with DMSO , erlotinib , AZD6244 , and BEZ235 ( each 1 µM ) or for 0 , 3 , 6 , 12 , and 24 hr with erlotinib ( single samples for each time point ) . Total mRNA was isolated using the RNeasy Mini Kit ( Qiagen ) . Generation of cRNA , hybridization to GeneChip Human Genome U133 Plus 2 . 0 mRNA expression arrays and array scanning were done according to the manufacturer's standard protocols ( Affymetrix , Inc . ) . Raw Affymetrix CEL files were converted to a single value for each probe set using Robust Multi-array Average ( RMA ) and normalized using quantile normalization . Quality control was performed using the distribution analysis , correlation and principle variance components analysis functions in Jmp Genomics ( SAS Institute ) . Individual genes with statistically significantly altered expression after treatment ( compared to untreated cells ) were identified using ANOVA after model adjustment for multiple hypothesis testing across LSMeans differences using the False Discovery Method of Benjamini and Hochberg ( FDR ) with Alpha set to 0 . 05 . The complete data set is available at the NCBI Gene Expression Omnibus ( http://www . ncbi . nlm . nih . gov/geo/ ) , Accession GSE51212 . For RT-qPCR , 1 µg mRNA was converted to cDNA using the First-strand cDNA Synthesis Kit ( GE Healthcare , Pittsburgh , PA ) . cDNA was analyzed on a 7500 Real Time PCR System ( Applied Biosystems ) using TaqMan Gene Expression Master Mix and TaqMan gene expression assays ( with GAPDH or ACTB as control , individual assays listed in Supplementary file 1—Thermo Fisher Scientific , Grand Island , NY ) . For ChIP-qPCR , immunoprecipitated chromatin ( see below ) was analyzed on the same system , using paired DNA PCR primers ( Supplementary file 1 ) and Power SYBR Green PCR Master Mix ( Applied Biosystems ) . Immunoblotting was performed using standard methods . After treatment with the indicated drugs , cells were washed with cold PBS and lysed in buffer containing 20 mM Tris pH 7 . 5 , 150 mM NaCl , 100 mM MgCl2 , 1% Nonidet P-40 and 10% glycerol supplemented with HALT protease and phosphatase inhibitor cocktail ( Thermo Fisher Scientific ) using a Q800R sonicator ( Qsonica , Newtown , CT ) . Lysates were centrifuged at 16 , 000×g for 5 min at 4°C . Protein concentrations were determined by BCA assay ( Thermo Fisher Scientific ) . Proteins were resolved by SDS-PAGE and transferred to a polyvinylidenes difluoride membranes ( Biorad ) using the Transblot Turbo Transfer System ( Biorad , Hercules , CA ) . Immunoblotting was performed per each antibody manufacturer's specifications . Antibodies used were pEGFR ( Y1068 ) , EGFR , pAKT ( S473 ) , AKT , pERK ( T202/Y204 ) , ERK , cleaved PARP , cleaved caspase-3 , SOX2 , BIM , phospho-FOXO1 ( T24 ) /FoxO3a ( T32 ) , FOXO1 , FOXO3 ( Cell Signaling Technology , Beverly , MA ) , GAPDH ( EMD Millipore , Billerica , MA ) , Ki67 ( Epitomics , Burlingame , CA ) , phospho-FOXO6 ( S184 ) ( Abcam , Cambridge , MA ) , FOXO6 ( Proteintech , Chicago , IL ) , CD133 , GKLF , OCT4 , MYC ( Santa Cruz , Dallas , TX ) , CD44 and CD24 ( BD Biosciences , San Jose , CA ) , and TY1 ( Diagenode , Denville , NJ ) . ChIP assays were carried out using approximately 5–10 × 106 HCC827 cells , following the procedures described previously ( Mikkelsen et al . , 2007; Ku et al . , 2008 ) . In brief , chromatin from formaldehyde-fixed cells was fragmented to a size range of 200–700 bases with a Branson 250 sonifier . Solubilized chromatin was immunoprecipitated overnight with goat anti-SOX2 antibody or goat IgG as a negative control ( both antibodies are from R&D Systems , Minneapolis , MN ) . Antibody–chromatin complexes were pulled down with protein G-Dynabeads ( Thermo Fisher Scientific ) , washed , and then eluted . After crosslink reversal , RNase A , and proteinase K treatment , immunoprecipitated DNA was extracted with the Agencourt AMPure XP PCR Purification Kit ( Beckman Coulter , Brea , CA ) . ChIP DNA was quantified with Qubit ( Thermo Fisher Scientific ) . 5 ng purified DNA ( immunoprecipiated chromatin and input controls ) were used to prepare Illumina compatible sequencing libraries for sequencing using the MiSeq Desktop Sequencer ( Illumina , San Diego , CA ) . ChIP seq reads were aligned to the hg19 reference genome using BWA ( Li and Durbin , 2009 ) . Aligned reads were extended to 200 bp to approximate fragment sizes , and then 25-bp resolution density maps were derived by counting the number of fragments overlapping each position , using IGV tools ( Robinson et al . , 2011 ) . The density maps were normalized to 5 million reads , and IGV was used to visualize ChIP seq coverage maps ( Thorvaldsdottir et al . , 2013 ) . For immunofluorescence ( IF ) analysis , cells plated in chamber slides were fixed with 4% formaldehyde , permeabilized/blocked with 5% normal goat serum/0 . 3% Triton X-100 and then incubated overnight at 4°C with antibody to SOX2 ( either rabbit—Cell Signaling Technology or goat—R&D systems ) . SOX2 staining was visualized with appropriate Alexa Fluor conjugated secondary antibodies ( Jackson Immunoresearch , West Grove , PA ) . Ki-67 costaining was performed with antibody to Ki-67 conjugated to Alexa Fluor 488 ( Epitomics ) . Nuclei were visualized with DAPI . Immunohistochemistry ( IHC ) of SOX2 on formalin-fixed , paraffin embedded tumor tissue was performed according to the antibody manufacturer's suggested protocol by the Specialized Histopathology Laboratory of the Massachusetts General Hospital using the SignalStain Boost IHC detection reagent ( Cell Signaling Technology ) . All IF/IHC samples were imaged and quantitated using the Vectra Automated Multispectral Imaging System ( PerkinElmer , Waltham , MA ) . Images were initially scanned at 4× magnification and then multiple high-powered fields were automatically acquired . The emission spectra of each fluorophore/IHC stain was computationally unmixed by preparing matched single stained samples . Unmixed images were segmented to identify individual cell nuclei based on DAPI or hematoxylin signal , and the mean nuclear signal for each fluorophore/IHC stain was calculated using inForm Advanced Image Analysis Software ( PerkinElmer ) . For some samples , in addition to analyzing the mean signal for each stain on every cell analyzed , individual nuclei were scored as positive or negative for SOX2 using the scoring function in inForm to set a fixed threshold for SOX2 signal based on the presence or absence of any degree of SOX2 staining . For fluorescence images , when samples to be compared were acquired using different exposure times , data were normalized to exposure time using the normalized counts setting/function in inForm . Cells were plated in multiple replicate wells at 2500 cells per well in 96-well format , treated 24 hr after plating and analyzed 72 hr later . Wells were fixed with 4% formaldehyde and stained with Syto60 red fluorescence nucleic acid stain ( Thermo Fisher Scientific ) for 1 hr at room temperature . After washing each well three times with water , the fluorescence of each well was analyzed using a Spectramax M5 plate reader ( Molecular Devices , Sunnyvale , CA ) with excitation 630 nm , emission 697 nm , and cutoff 695 nm , background corrected by subtracting the mean signal from empty wells and normalized to the mean value of untreated wells . All Dharmacon ON-TARGET plus siRNA pools were purchased from Thermo Fisher Scientific . Sequences are included in Supplementary file 1 . Cells were plated at 12 , 500–25 , 000 per ml in 96-well plates ( 0 . 2 ml ) , 8-chamber slides ( 0 . 5 ml ) , 6-well plates ( 3 ml ) , or 60 mm plates ( 6 ml ) in antibiotic-free medium and transfected the following day with each siRNA ( 12 . 5 nM final concentration ) with the Dharmafect I transfection reagent ( 2 µl/200 μl for PC9 and 4 µl/200 μl for HCC827 and HCC2935 ) according to the manufacturer's standard protocol . Media were changed 6–24 hr after transfection , and isolation of mRNA and protein lysates and immunofluorescence analysis was carried out 48–96 hr after transfection ( and treatment with erlotinib ) . Lentiviral-inducible expression constructs containing SOX2 ( wild type and epitope tagged ) or WNT pathway mutants under the control of a doxycycline-inducible promoter or control vector containing the GUS gene were constructed by subcloning each ORF into the pInducer 20 lentiviral vector ( Meerbrey et al . , 2011 ) using the Gateway cloning system ( Thermo Fisher Scientific ) . This vector also contains a neomycin resistance cassette . Lentiviral expression constructs were cotransfected into 293T cells with the HIV-1 packaging construct pCMVdeltaR8 . 91 and the VSV-G envelope construct pMD . G using TransIT-LT-1 transfection reagent ( Thermo Fisher Scientific ) . Viral supernatants were collected at 48 and 72 hr post transfection in DMEM with 30% FCS , filtered through 0 . 45 μm syringe filters to remove cell debris and stored in aliquots at −80°C . Transductions were carried out in the presence of 8 μg/ml Polybrene ( Sigma ) by spinoculation at 1200×g and at 32°C for 60 min in a Sorvall Legend RT table-top centrifuge . Viral supernatant was exchanged for fresh media 24 hr after spinoculation . To generate stable cDNA expressing cell lines , cells were selected in G418 . Tetracycline-reduced FBS ( Clontech , Mountain View , CA ) was substituted for all media for cells transduced with the pInducer 20 vectors . To induce expression of SOX2 or WNT pathway constructs , doxycycline ( Sigma ) at 0 . 1 µg/ml was added to all cultures at least 3 hr prior to addition of erlotinib . Stable cell lines expressing the TOP-FLASH reporter were generated by transducing cells with 7TFP recombinant lentiviruses ( Fuerer and Nusse , 2010 ) , and luciferase assay was performed as previously described ( Singh et al . , 2012 ) . Briefly , Rediject D-Luciferin Ultra ( Perkin Elmer ) was added in 0 . 2 ml fresh media ( 1–200 dilution ) to each well of cells in a 96-well plate and incubated for 15 min at 37°C . Luciferase activity was imaged with the IVIS Lumina II In Vivo Imaging System ( Perkin Elmer ) . The radiance of each well was determined using Living Image 4 . 2 software ( Perkin Elmer ) , background corrected by subtracting the mean signal from empty wells and normalized both to the relative cell number of each well as determined by Syto60 assay and the resulting normalized mean value of untreated wells . All mouse studies were carried out according to Institutional IUCAC guidelines . Mice were treated by oral gavage with a single 100 mg/kg dose of erlotinib when subcutaneous tumors had reached ∼500 mm3 in sizes ( ∼21–28 days ) . PC9 and HCC827 xenograft tumors were harvested 21 hr after erlotinib treatment . All statistical analyses , including number of replicates and statistical method used , are included in the relevant figure legends . | Tumors can form when cells gain mutations in genes that enable them to grow and divide rapidly . In some human lung cancers , genetic mutations are found in a gene that makes a protein called EGFR . This protein encourages cells to divide and the mutations can lead to the cancer cells producing more EGFR , or producing a form of the protein that is more active . Treating these cancers with a drug called erlotinib inhibits EGFR and makes the tumors shrink dramatically , but the tumors will usually re-grow because any tumor cells that survive often become resistant to the drug . There are several ways that the tumor cells can become resistant , which makes the task of developing a solution to this problem more difficult . It has been suggested that the tumor cells may enter a temporary ‘drug-tolerant’ state that helps them to survive and makes it more likely that they will develop resistance to the drug . However , it is not clear how this drug-tolerant state might work . To address this question , Rothenberg et al . examined which genes are switched on ( or ‘expressed’ ) in tumor cells with a mutant version of EGFR after they were treated with the erlotinib drug . The experiments show that a gene called SOX2 is expressed in these cells . Cells that had lower levels of SOX2 expression were more sensitive to the effects of the drug and fewer cells developed resistance . On the other hand , cells that had higher levels of SOX2 expression were less sensitive to the drug and resistance was more likely to develop . A protein called FOXO6—which is usually suppressed by EGFR—activates the SOX2 gene in these cells . Therefore , using erlotinib to inhibit EGFR to kill the cancer cells increases the activity of FOXO6 , which in turn promotes the survival of some of the cells by activating the SOX2 gene . A better understanding of the ways in which cancer cells adapt to erlotinib and other drugs may help us to design more effective treatments with better outcomes for patients . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cancer",
"biology"
] | 2015 | Inhibition of mutant EGFR in lung cancer cells triggers SOX2-FOXO6-dependent survival pathways |
The Na+/K+-pump maintains the physiological K+ and Na+ electrochemical gradients across the cell membrane . It operates via an 'alternating-access' mechanism , making iterative transitions between inward-facing ( E1 ) and outward-facing ( E2 ) conformations . Although the general features of the transport cycle are known , the detailed physicochemical factors governing the binding site selectivity remain mysterious . Free energy molecular dynamics simulations show that the ion binding sites switch their binding specificity in E1 and E2 . This is accompanied by small structural arrangements and changes in protonation states of the coordinating residues . Additional computations on structural models of the intermediate states along the conformational transition pathway reveal that the free energy barrier toward the occlusion step is considerably increased when the wrong type of ion is loaded into the binding pocket , prohibiting the pump cycle from proceeding forward . This self-correcting mechanism strengthens the overall transport selectivity and protects the stoichiometry of the pump cycle .
The Na+/K+-pump is a primary active membrane transporter present in nearly all animal cells . It belongs to the P-type ATPase family , which utilizes the energy released from ATP hydrolysis to move ions against their concentration gradients across a membrane barrier . The ion species transported by the pump under physiological conditions are Na+ and K+ . Like many other membrane transporters , the Na+/K+-pump works according to an 'alternating-access' ion transport mechanism , with the bound ions accessible from only one side of the membrane at a time . The consensus scheme of the pump cycle is known as the 'Albers-Post' cycle ( Albers , 1967; Post et al . , 1969 ) . It involves two major conformations , E1 and E2 , with inward- and outward-facing ion binding sites , respectively . In each cycle , the E1 conformation binds three Na+ from the cytosol and exports them using the energy from ATP hydrolysis . After external release of Na+ , binding of extracellular K+ promotes the structural transition to the occluded E2 state , which finally imports two K+ as binding of ATP returns the pump to the E1 conformation . E1/E2 conformational change and the accompanied electrogenic active transport are facilitated by the phosphorylation and dephosphorylation of an aspartate residue in the cytoplasmic domain , which is a hallmark of the P-type ATPase family ( Axelsen et al . , 1998 ) . The presence of the Na+/K+-pump is essential for excitability and secondary active transport . More than 40% of the energy produced in mammals is consumed by the Na+/K+-pump ( Milligan et al . , 1985 ) . Although it is a machine designed for such a precise and important function , it has been shown that many cations , including congeners of K+ and some organic cations , can bind to the same sites used by the pump to bind and transport K+ ions ( Mahmmoud et al . , 2015; Ratheal et al . , 2010 ) . Competitive binding between Na+ and other cations at the cytoplasmic side of the membrane has also been observed ( Schneeberger et al . , 2001 ) . An unsolved puzzle , therefore , is how the Na+/K+-pump is able to recognize and accept two K+ from the extracellular matrix , where Na+ concentration is much higher , and how it selectively binds Na+ from the cytoplasm to keep the pump cycle running forward . Structural studies have provided tremendous insights into the function of the Na+/K+-pump , which consists of two obligatory subunits , α ( catalytic ) and β ( auxiliary ) , and sometimes a tissue specific regulatory subunit from the FXYD protein family ( Geering , 2006; Mercer et al . , 1993 ) . The transmembrane region of the α-subunit contains the ion binding sites within its 10 helices ( called M1-M10 ) . Recent crystal structures of the pump in its E1 and E2 states reveal the locations of the three Na+ binding sites in E1 and the two K+ binding sites in E2 ( Kanai et al . , 2013; Laursen et al . , 2013; Morth et al . , 2007; Shinoda et al . , 2009 ) . From structural alignments based on the heavy atom positions in the binding sites , it becomes clear that the binding pocket harboring sites I and II in E1 overlaps with those in E2 ( Figure 1 ) . Site III is only formed in E1 . It is located between the transmembrane helices M5 , M6 , and M8 and is thought to exclusively bind Na+ but appears to catalyze H+ transport ( Poulsen et al . , 2010; Ratheal et al . , 2010 ) in a manner that presents a complex dependence on the concentrations of Na+ , K+ , and H+ ( Mitchell et al . , 2014 ) . Previous studies have indicated that protonation at the ion binding pocket may play a role in the selectivity of these sites for K+ when the pump is in its E2 state ( Yu et al . , 2011 ) . Biochemical assays on the mutant D926N , which is often used as a surrogate for protonated D926 , also show that it induces distinct effects on Na+ and K+ binding ( Einholm et al . , 2010; Jewell-Motz et al . , 1993 ) , suggesting a potential change in its protonation state upon the E1/E2 transition . Taken together , the evidence , although indirect , suggests the possibility of an E1 specific protonation state that favors Na+ binding to the pump . The nature of this protonation state is , however , unclear . 10 . 7554/eLife . 16616 . 003Figure 1 . The ion binding sites in ( A ) Na3·E1· ( ADP·Pi ) ( PDBID 3WGV ) and ( B ) E2 ( K2 ) ( PDBID 2ZXE ) states . Only the transmembrane helices M4 , M5 , M6 , M8 , and M9 from the α subunit are shown . Residues in the binding site are highlighted in stick presentation with those protonatable colored in yellow . Na+ ( yellow ) and K+ ( magenta ) ions are in spheres . The binding site number indices are presented on top of the ions . The view is from the extracellular side towards the intracellular side . The figure is produced with PyMOL ( DeLano , 2002 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16616 . 003 In the present study , we started with the recently published crystal structure of the Na+/K+-pump in a partially occluded Na3·E1 ( ADP·Pi ) state and used molecular dynamics ( MD ) simulations to show that there is a correlation between the binding pocket protonation and the Na+ selectivity in E1 . The binding sites in Na3·E1 ( ADP·Pi ) were tested one by one by 'alchemically transforming' the bound Na+ into a K+ while keeping the other two sites occupied by Na+ . A similar approach was previously used to study the selectivity in other conformational states of the pump ( Yu et al . , 2011 ) ( see also Materials and methods section ) . The results show that the Na+ selectivity at all three sites is realized only when a specific set of binding pocket residues are protonated . This set of residues is , by comparison , different from that in the K+ selective E2 state . The implication is that the protonation state and the selectivity of the pump are tightly coupled; when the pump undergoes a transition between E1 and E2 , a protonation state switch occurs . The present findings also show that the effective selectivity of the pump is reinforced by a self-correcting mechanism , which prevents the occlusion step on either side of the membrane if the wrong type of ions were loaded into the binding sites . This mechanism ensures that counterproductive transport cycles do not occur .
The crystal structures of the pump in its E1 ( PDBID 3WGV [Kanai et al . , 2013] ) and E2 ( PDBID 2ZXE [Shinoda et al . , 2009] ) conformations show that there is a large structural overlap between the Na+ and K+ binding pockets . The sites I and II are in the main binding chamber formed by helices M4 , M5 , and M6 , while site III , located in between M5 , M6 , and M8 ( Figure 1 ) , is Na+ exclusive and appears in E1 only . The coordination of Na+ and K+ at sites I and II are similar . Many residues are found to coordinate both Na+ and K+ at these two sites in E1 and E2 ( Table 1 ) . Although the structural difference between the ion binding pockets in E1 and E2 may not be particularly large , the local physicochemical environment can display considrable variations . In particular , the latter could affect the pKa and the protonation states of the six protonatable residues in the binding pocket ( Figure 1 ) . Their pKa values calculated using PROPKA3 . 1 ( Olsson et al . , 2011 ) are listed in Table 2 . 10 . 7554/eLife . 16616 . 004Table 1 . Atoms coordinating the binding site ions in the crystal structures and from the MD simulations . O is the backbone carbonyl oxygen atom . OG and OG1 are the hydroxyl oxygen atoms in serine and threonine . OD1 and OD2 are the carboxyl oxygen atoms in asparate . OE1 and OE2 are the carboxyl oxygen atoms in glutamate . OH2 is the water oxygen . DOI: http://dx . doi . org/10 . 7554/eLife . 16616 . 004E1Site ISite IISite IIIx-rayMDx-rayMDx-rayMDA323OT772OG1V322OE779OE1Y771OY771OE779OE1T772OV325OD804OD1T774OT774OD808OD1N776OD1E327OE2D808OD1Q923OE1Q923OE1D808OD1D804OD1WaterOH2D926OD1D808OD2WaterOH2E2Site ISite IIx-rayMDx-rayMDT772OS775OGV322OA323OS775OGN776OD1V325OV325ON776OD1D804OD2E779OE2E779OE1D804OD2D804OD2D804OD110 . 7554/eLife . 16616 . 005Table 2 . pKa values of binding site titratable residues calculated from the crystal structures . The crystal structure resolution is given below the PDB ID . DOI: http://dx . doi . org/10 . 7554/eLife . 16616 . 005E1E23WGV 2 . 8 Å4HQJ 4 . 3 Å2ZXE 2 . 4 Å3B8E 3 . 5 ÅD8045 . 9 ( 6 . 2 ) *11 . 1 ( 11 . 2 ) 3 . 70 . 8 ( 2 . 1 ) D8083 . 5 ( 3 . 1 ) 3 . 7 ( 3 . 7 ) 5 . 86 . 8 ( 6 . 6 ) D9266 . 4 ( 7 . 2 ) 5 . 6 ( 5 . 6 ) 8 . 97 . 4 ( 8 . 4 ) E32711 . 0 ( 11 . 3 ) 5 . 7 ( 5 . 6 ) 8 . 310 . 8 ( 9 . 9 ) E7799 . 9 ( 8 . 4 ) 7 . 4 ( 7 . 3 ) 10 . 79 . 6 ( 8 . 0 ) E9549 . 6 ( 9 . 7 ) 9 . 2 ( 9 . 2 ) 10 . 310 . 7 ( 10 . 3 ) *If two chains are present in the same asymmetric unit , the pKa of the same residue in the other chain is shown inside the parenthesis . It is important to note that the pKa values calculated with empirical methods like PROPKA are sensitive to local structural perturbations . Even when the structures assume the identical conformational state and are taken from the same asymmetric unit from a crystal , the pKa values of the same residue can differ by more than one pH unit . For example , the pKa of E779 is 9 . 9 in chain A of the crystal structure 3WGV but the value is 8 . 4 in chain C , yet , the structural difference between the two chains is minimal ( Table 2 ) . Because of this , only the structures with the highest resolution for the E1 and E2 states of the pump , 3WGV and 2ZXE , were used to guide the protonation state assignments in order to avoid false assignments of protonation state originated from structural uncertainty in lower resolution crystal structures . Interestingly , the protonation states of E327 and D804 assigned based on the 3WGV and 4HQJ structures are reversed . According to the PROPKA results , E327 appears to be deprotonated and D804 protonated in 3WGV , and vice versa in 4HQJ . While this inconsistency could be due to the lower resolution in the crystal structure of 4HQJ , it is worth noting that these two residues are in close proximity from one another suggesting that a proton could be passed back and forth between them in the E1 state . Previous calculations have indicated that a specific set of residues has to be protonated for the E2 binding sites to be K+ selective ( Yu et al . , 2011 ) . Although the protonation states of D926 and E954 were not considered in the previous study because they lie outside of the two K+ binding sites , they are likely to be protonated in E2 according to the predicted pKa ( Table 2 ) . The binding pocket protonation derived from the crystal structure of the pump in its partially occluded Na3·E1 ( ADP·Pi ) state differs from that in E2 . According to their pKa values , D804 , D808 , and D926 should be deprotonated and the glutamates , E327 , E786 , and E954 , should be protonated . Arguments based on pKa values predicted by PROPKA , however , have to be taken with caution , because the empirical method is highly sensitive to the local structural variation . For example , a deprotonated acidic side chain could have a PROPKA predicted pKa value at slightly higher than 7 because of the structural snapshot used to make the prediction . To seek a more robust assessment of these factors , molecular dynamics ( MD ) simulations were conducted to examine the structural stability of the binding pocket for different protonation configurations . One goal is to determine the protonation states of D804 and D926 , both of which have a predicted pKa value within 1 . 5 pH unit to 7 . The protonation state of D808 was also investigated since it affects the K+ selectivity of the binding sites in the E2 ( K2 ) state . A total of eight MD simulations were carried out ( Table 3 ) , starting from all possible protonation state configurations accessible by these three residues . The protocol for setting up these systems is given in details in the Materials and methods section . 10 . 7554/eLife . 16616 . 006Table 3 . Summary of the all-atom simulation systems and the FEP/H-REMD reduced systems . The binding site residues E327 , E779 , and E954 were kept protonated in all the systems . DOI: http://dx . doi . org/10 . 7554/eLife . 16616 . 006SystemsBinding site residuesSimulation time ( ns ) MDFEP/H-REMDWildtypeE1_S0D804-D808-D926-1402 × 128E1_S1D804pD808-D926-3002 × 128E1_S2D804-D808pD926-1392 × 128E1_S3D804-D808-D926p5032 × 128E1_S4D804pD808pD926-94E1_S5D804pD808-D926p87E1_S6D804-D808pD926p277E1_S7D804pD808pD926p141E2_S0D804-D808pD926-1932 × 128E2_S1D804-D808pD926p3502 × 128P-E2_S0D804-D808pD926p1002 × 128MutantsE1_S1MD804ND808-D926-402 × 128E1_S2MD804-D808ND926-402 × 128E1_S3MD804-D808-D926N402 × 128E2_S1MD804ND808-D926p402 × 128E2_S2MD804-D808ND926p402 × 128E2_S3MD804-D808-D926N402 × 128P-E2_S1MD804ND808-D926p402 × 128P-E2_S2MD804-D808ND926p402 × 128 Figure 2 shows the binding pocket conformation in the E1 systems ( Table 3 ) at the end of the all-atom MD simulations . Out of the three aspartates when two or more are protonated the binding pocket becomes unstable , showing a large deviation from the crystal structure either with one of the bound Na+ expelled from its binding site ( systems E1_S5-7 ) , or with a Cl- ion attracted into the binding pocket ( system E1_S4 ) . These scenarios are not likely to happen in the normal function cycle of the pump . On the other hand , when the number of protonated aspartates is less than two the binding pocket remains structurally close to that in the crystal structure ( systems E1_S0-3 ) . The heavy atom root mean squared deviations ( RMSD ) between the crystal structure and the structure snapshots from the last 50 ns of the simulations are less than 1 . 6 Å . 10 . 7554/eLife . 16616 . 007Figure 2 . Comparison of snapshots at the end of the MD simulations ( green ) and the crystal structure of Na3·E1· ( ADP·Pi ) ( PDBID 3WGV ) ( white ) . The binding site residues are shown in stick presentation and the ions are shown as spheres . Binding site Na+ ions from the MD simulation snapshots are in yellow , and the crystal Na+ are in orange . A Cl- ion has entered the binding site in system E1_S4 during the simulation and is shown in green . The view is from the extracellular side towards the intracellular side . The figure is produced with PyMOL ( DeLano , 2002 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16616 . 007 Analysis of the structural stability of the binding sites based on the MD simulations indicates that all of the four protonation states producing stable ion binding pockets may coexist when the pump is in the Na3·E1 ( ADP·Pi ) state . The relative population of these states in the state Na3·E1 ( ADP·Pi ) , however , would differ from one another . The overall selectivity of the binding sites has contributions from all the states , and the predominant protonation state configuration should produce binding sites that are Na+ selective . The determination of this protonation state configuration starts from the protein-membrane systems generated by the restraint-free MD simulations . A reduced structural model of the binding pocket is derived from each of these systems ( Table 3 ) and the binding free energy differences between Na+ and K+ ( ΔΔGNa→K ) at the binding sites are calculated ( See Materials and methods ) . The value of ΔΔGNa→K reflects the affinity difference between K+ and Na+ binding as ΔΔGNa→K=ΔGk−ΔGNa=RT ln ( KD , K/KD , Na ) . The results are plotted as the logarithm of the affinity ratio , ln ( KD , K/KD , Na ) , in Figure 3 . The values of ΔΔGNa→K are shown in Table 4 . 10 . 7554/eLife . 16616 . 008Figure 3 . Ion binding sites selectivity characterized by ln ( KD , K/KD , Na ) in states Na3E1· ( ADP·Pi ) ( blue ) and E2 ( K2 ) ( red ) . Sites I ( square ) , II ( circle ) , and III ( triangle ) are distinguished by their shapes . Values from the previous calculations with a smaller reduced region are shown as empty symbols . All the binding site glutamates ( i . e . , E327 , E779 , and E954 ) are kept protonated . The protonation states of the binding site aspartates are indicated below the plot . DOI: http://dx . doi . org/10 . 7554/eLife . 16616 . 00810 . 7554/eLife . 16616 . 009Table 4 . The binding free energy difference ( ΔΔGNa→K ) at all the binding sites calculated from FEP/H-REMD simulations . The energy values are in kcal/mol . DOI: http://dx . doi . org/10 . 7554/eLife . 16616 . 009SystemsBinding sitesIIIIIIWildtypeE1_S00 . 1 ± 0 . 14 . 5 ± 0 . 11 . 8 ± 0 . 1E1_S13 . 7 ± 0 . 21 . 7 ± 0 . 14 . 7 ± 0 . 1E1_S2-3 . 5 ± 1 . 03 . 8 ± 0 . 32 . 9 ± 0 . 6E1_S3-1 . 5 ± 0 . 24 . 7 ± 0 . 4-0 . 5 ± 0 . 3E2_S0-4 . 4 ± 0 . 3-6 . 1 ± 0 . 0E2_S1-3 . 5 ± 0 . 1-6 . 1 ± 0 . 1P-E2_S0-1 . 0 ± 0 . 60 . 7 ± 0 . 4MutantsE1_S1M1 . 4 ± 0 . 15 . 7 ± 0 . 02 . 4 ± 0 . 1E1_S2M4 . 3 ± 0 . 76 . 6 ± 0 . 65 . 2 ± 0 . 9E1_S3M-0 . 6 ± 0 . 11 . 7 ± 0 . 10 . 0 ± 0 . 0E2_S1M-3 . 3 ± 0 . 3-5 . 7 ± 0 . 0E2_S2M-5 . 4 ± 0 . 1-6 . 4 ± 0 . 1E2_S3M-3 . 8 ± 0 . 1-7 . 2 ± 0 . 1P-E2_S1M0 . 8 ± 0 . 91 . 9 ± 0 . 4P-E2_S2M1 . 6 ± 0 . 51 . 9 ± 0 . 4 As shown in Figure 3 , the protonation state of the aspartates greatly affects both the E1 and E2 binding sites selectivity . The protonation state at E2 ( K2 ) binding pocket proposed in the earlier study ( Yu et al . , 2011 ) is confirmed by the present calculations , in which the reduced system includes a larger region of the protein and more residues at the outskirt of the main binding pocket , including D926 and E954 . It is evident that a different set of residues is protonated in the Na+ selective E1 as compared to in the K+ selective E2 ( Figure 3 ) . While the glutamates ( i . e . , E327 , E779 , and E954 ) remain protonated in both the Na3·E1 ( ADP·Pi ) and E2 ( K2 ) states , the binding pocket aspartates take on opposite protonation states . In Na3·E1 ( ADP·Pi ) D804 is protonated and both D808 and D926 are deprotonated , and their protonation states are reversed in E2 ( K2 ) . Free energy perturbation ( FEP ) calculations on the protonation/deprotonation of D804 estimated a pKa value of 9 . 2 ( See Materials and methods ) , further supporting its protonated form under physiological conditions in E1 . The impact of charge-neutralizing mutations of the key aspartate residues on the binding site selectivity was examined . D to N mutations have previously been used in experimental studies as a strategy to ascertain the possible charged state of these residues . The results of the free energy calculations , summarized in Figure 4 , indicate that all the three sites remain Na+ selective in the partly occluded Na3·E1 ( ADP·Pi ) in both D804N and D808N , whilst D926N causes the sites to lose most of the Na+ selectivity . Using Na+ dependent phosphorylation and ATP binding assays , it was shown that the cytoplasmic binding affinity for both Na+ and K+ is reduced in D804N and D808N ( Jorgensen et al . , 2001; Pedersen et al . , 1997 ) . The D804N mutation affects the cytoplasmic K+ and Na+ binding differently . The mutation’s impact on cytoplasmic Na+ binding is not as prominent as on K+ binding and D804N would appear more Na+ selective than the wildtype . This is reflected in the calculation results . The KD , K/KD , Na at sites I and III are reduced by a factor of ~50 from 476 . 6 and 2523 . 3 in the wildtype to 10 . 3 and 54 . 6 in the D804N mutant , but the sites are still Na+ selective . The selectivity of site II is increased to a much larger extent , from 17 . 0 in the wildtype to 13359 . 7 in the mutant . This is an ~1000 fold increase in selectivity and it implies that D804N is likely more Na+ selective than the wildtype pump . The D808N mutant also becomes more Na+ selective compared to the wildtype pump . The most dramatic increase in Na+ selectivity happens at site II as in the case of D804N . Interestingly , site I , which prefers to bind K+ in the wildtype with the protonated D808 ( Figure 3 ) , is now Na+ selective in D808N . It could be that the spatial packing is the major contributing factor to the site I ion selectivity . In the calculations , both sites II and III lose their Na+ selectivity in the D926N mutant . The shifting of selectivity towards K+ at site III in this mutant is a bit surprising , and worth commenting on . There are two possible explanations . First , substituting a negatively charged carbonyl oxygen with a bulkier but neutral –NH2 at the D926 sidechain could prevent entrance of an ion to this site . Thus , the simulated conformation with an ion at this site could be energetically inaccessible . In other words , this is a metastable state with the absolute free energy of Na+ and K+ binding to this conformation equally prohibitive . An alternative explanation is that the D926N mutation alters the available conformational space accessible by helix M5 and this allows K+ to go into site III as suggested in reference ( Kanai et al . , 2013 ) . The binding of this K+ , however , prevents the further binding of Na+ , and results in the compromised Na+ selectivity in this mutant . Even though the calculations show D926N with decreased Na+ selectivity , they do not explain why experimentally the D926N mutant has compromised Na+ binding ability in the absence of K+ ( Einholm et al . , 2010 ) . The loss in selectivity seen in this D926N mutant , however , could account for the strong inhibition by high K+ on Na+/K+-ATPase activity ( cf . Figure 2A in reference [Einholm et al . , 2010] ) . 10 . 7554/eLife . 16616 . 010Figure 4 . Charge-neutralizing mutations and their impact on binding site selectivity . The wildtype protein is colored black and the mutations in E1 ( blue ) and E2 ( red ) are colored differently . Sites I ( square ) , II ( circle ) , and III ( triangle ) are distinguished by their shapes . The empty symbols represent values calculated from the outward facing P-E2 model . DOI: http://dx . doi . org/10 . 7554/eLife . 16616 . 010 All the D to N mutations tested for the occluded E2 ( K2 ) state appear to have little impact on the K+ selectivity , in apparent contradiction with the experimental observations showing that both D804N and D808N display decreased selectivity for external K+ ( Kuntzweiler et al . , 1996; Pedersen et al . , 1997 ) . In a previous computational study , the D808N mutation was also shown to minimally affect the K+ selectivity in E2 ( K2 ) . Discrepancies between the calculations and the selectivity inferred from biochemical experiments have been noted previously , though the reason was not clear . A plausible explanation could be that the experimentally measured selectivity is an outcome from the relevant states including both P-E2 and E2 ( K2 ) . However , any contribution from the P-E2 state to the observed selectivity was not taken into account by the calculations because of the missing crystal structure of the outward-facing state . To test this idea , we generated a model structure of the P-E2 state based on the homolog structures from the Ca2+ SERCA pump ( see below ) . This model was then used to calculate the ΔΔGNa→K in the wildtype and the mutant pumps . The results show that both sites I and II have increased preference for external Na+ in the P-E2 state of D808N ( Figure 4 ) , which explains previous discrepancies between the calculations and the experiments . Using the crystal structures of the Ca2+ SERCA pump as templates and a coarse grained transition pathway calculation method , ANMpathway ( Das et al . , 2014 ) , we generated structural models of the Na+– and K+–loaded outward-facing P-E2 state ( see Materials and methods ) . The models appear to be structurally similar to the recently published P-E2 like structures of the Na+/K+-pump ( Gregersen et al . , 2016; Laursen et al . , 2013 ) and the vanadate-Inhibited , P-E2 mimic of the Ca2+ SERCA pump ( Clausen et al . , 2016 ) . The MD simulations of the models are also able to predict with considerable accuracy the experimentally measured gating charge upon ion binding ( Castillo et al . , 2015 ) . The structural transition pathways leading to these models provide a view of the intermediate structures along the pump cycle . Using these models , we calculated the ΔΔGNa→K for the binding site of the intermediate state structures . The entire atomistic protein-membrane systems were used in the calculations . The results are presented in Figure 5 and Table 5 . The calculations are valid as the ΔΔGNa→K values mirror those computed from the reduced systems with the same protonation states ( Figure 3 ) . 10 . 7554/eLife . 16616 . 011Figure 5 . The binding site ion selectivity along the pump cycle . Sites I ( square ) , II ( circle ) , and III ( triangle ) are distinguished by their shapes . Different colors indicate whether there are two ( red ) or three ( blue ) sites that are included in the calculations . The conformational states are numbered and stamped along the pump cycle in the top panel . The protonation states of the aspartates are indicated below . DOI: http://dx . doi . org/10 . 7554/eLife . 16616 . 01110 . 7554/eLife . 16616 . 012Table 5 . Selectivity in the form of at the binding sites along the pump cycle from state P-E2·K2 to P-E2·Na3 . The energies are in kcal/mol . DOI: http://dx . doi . org/10 . 7554/eLife . 16616 . 012*p/--/pp/pp/pp/ps/--/ss/pp/ss/s ( 1 ) P-E2 . K2-0 . 7 ± 1 . 00 . 1 ± 0 . 5-1 . 0 ± 0 . 60 . 7 ± 0 . 4-0 . 6 ± 0 . 7 ( 2 ) Intermediate0 . 0 ± 0 . 41 . 8 ± 1 . 3-0 . 5 ± 0 . 3-5 . 1 ± 0 . 3-4 . 0 ± 0 . 9 ( 3 ) E2 ( K2 ) -4 . 9 ± 0 . 8-3 . 0 ± 0 . 5-5 . 0 ± 0 . 2-4 . 2 ± 0 . 1-9 . 3 ± 0 . 5 ( 4 ) Intermediate0 . 6 ± 0 . 7-0 . 8 ± 0 . 32 . 0 ± 0 . 3-2 . 7 ± 0 . 2-1 . 6 ± 0 . 4 ( 5 ) K2 . E11 . 5 ± 0 . 31 . 1 ± 0 . 21 . 1 ± 0 . 1-3 . 8 ± 0 . 3-1 . 2 ± 0 . 5 ( 6 ) K2 . E1*0 . 6 ± 0 . 50 . 7 ± 0 . 41 . 3 ± 0 . 2-0 . 4 ± 0 . 80 . 1 ± 0 . 7*s/-/--/s/-s/s/-s/s/-s/s/-s/s/ss/s/ss/s/ss/s/ss/s/ss/s/ss/s/sp/-/--/p/-p/s/-s/p/-p/p/-p/s/ss/p/ss/s/pp/p/sp/s/ps/p/pp/p/p ( 7 ) Na3·E1 ( ADP·Pi ) -1 . 0 ± 0 . 51 . 7 ± 0 . 20 . 7 ± 0 . 72 . 2 ± 0 . 23 . 0 ± 1 . 00 . 8 ± 0 . 52 . 6 ± 0 . 36 . 0 ± 0 . 32 . 1 ± 1 . 55 . 8 ± 1 . 17 . 4 ± 0 . 87 . 5 ± 2 . 0 ( 8 ) Intermediate2 . 9 ± 1 . 0-0 . 5 ± 1 . 21 . 5 ± 1 . 62 . 6 ± 1 . 32 . 5 ± 1 . 33 . 0 ± 0 . 70 . 6 ± 0 . 20 . 8 ± 0 . 54 . 0 ± 0 . 82 . 8 ± 1 . 12 . 7 ± 0 . 53 . 5 ± 1 . 5 ( 9 ) P-E2·Na32 . 3 ± 0 . 31 . 6 ± 0 . 52 . 1 ± 0 . 32 . 4 ± 0 . 34 . 6 ± 0 . 51 . 4 ± 0 . 31 . 7 ± 0 . 3-2 . 2 ± 0 . 33 . 1 ± 0 . 8-0 . 6 ± 0 . 6-0 . 8 ± 0 . 6-1 . 2 ± 1 . 2*The top and bottom rows represent the starting and ending binding site ion configurations . A 'p' represents a K+ ( potassium ) ion and an 's' represents a Na+ ( sodium ) ion . The binding sites I , II , and III in this order are separated by '/' . The results reveal a fascinating feature of the binding site selectivity along the pump cycle . Initially , the two K+ binding sites in the outward-facing model P-E2·K2 ( state ① in Figure 5 ) appear to be non-selective or only weakly K+ selective . However , the selectivity for K+ over Na+ increases as the pump occludes to form the K+–bound occluded state E2 ( K2 ) ( state ③ ) . After the dephosphorylation of P-E2 , the binding pocket opens up to the cytoplasmic side . Accompanying this structural transition is the selectivity reversal at site I , switching the site from K+ to Na+ selective ( ③ to ④ in Figure 5 ) . The protonation state change further shifts the ion selectivity at the sites in E1 towards Na+ . Changing the protonation state in K2E1 ( ④ to ⑤ in Figure 5 ) to the protonation state dominant in Na3·E1 ( ADP·Pi ) , i . e . , state ⑥ in Figure 5 with protonated D804 and deprotonated D808 and D926 , further reduces the K+ selectivity at both sites I and II . When the pump is in this state , site I is Na+ selective and site II is only weakly K+ selective . Together , the E2 to E1 structural transition and the protonation state change promote the release of K+ to the cytoplasmic side . Transient changes in the binding pocket protonation state upon occlusion/deocclusion are possible and could lead to variations in the magnitude of ΔΔGNa→K for the intermediate states . Nonetheless , the general trend in free energy changes upon occlusion/deocclusion should remain because the ΔΔGNa→K of the occluded state E2 ( K2 ) has a much larger magnitude than that of the open-access outward- and inward-facing states . The free energy calculations corroborate the notion that the two K+ binding from the extracellular side is sequential and possibly cooperative . The sequential binding of extracellular K+ was initially demonstrated by Forbush ( Forbush , 1987 ) . Recently using crystallography and isotopic measurements Ogawa et al . presented strong evidence that the first K+ binds at site I and the second K+ at site II ( Ogawa et al . , 2015 ) . Whether there is cooperativity upon extracellular K+ binding , however , is not clear from the crystal structures . Two electrode voltage clamp experiments have shown that the two extracellular K+ binding events are relatively independent in the absence of extracellular Na+ , but there is positive cooperativity of K+ binding when extracellular Na+ ions are present ( Jaisser et al . , 1994 ) . The current ΔΔGNa→K calculations of the two ion bound P-E2 ( Table 5 ) provide an interpretation of this phenomenon . With a K+ ion at site I , site II is relatively nondiscriminatory ( ΔΔGNa→K = 0 . 7 kcal/mol , p/p to p/s in state ① P-E2 . K2 , Table 5 ) , but it becomes much more Na+ selective ( ΔΔGNa→K = 2 . 4 kcal/mol , s/s to s/p in state ⑨ P-E2·Na3 , Table 5 ) when there is a Na+ ion bound at site I . In the presence of extracellular Na+ , K+ ions have to first compete with Na+ to bind at site I in order for the subsequent K+ to bind , therefore resulting in the observed binding cooperativity in the presence of extracellular Na+ . A similar phenomenon is seen in the Na+ branch of the pump cycle . Initially , site I in the inward-facing Na3·E1 ( ADP·Pi ) structure ( state ⑦ ) is weakly Na+ selective when the other two sites are filled with Na+ . However , the intermediate state ⑧ during the occlusion process shows that its site I is highly discriminating against K+ with a ΔΔGNa→K of 2 . 9 kcal/mol ( Figure 5 and Table 5 ) . Therefore , if a K+ is bound in this site , the free energy barrier toward occlusion would be much higher than when a Na+ ion was bound and the subsequent occlusion is not likely . In the case when a K+ ion replaces a Na+ at site II or III , although the energy barrier of occlusion is not as prominent , such a binding pocket ion configuration is energetically less favorable than the native configuration and its appearance is unlikely in the first place . Unlike the K+ branch , the ΔΔGNa→K calculations along the Na+ branch offers limited insights on the sequence of Na+ binding from the cytoplasmic side . The results are indicative of the selectivity at the sites , but not the absolute binding affinity . Based on the calculations alone , it is not possible to determine which site is the first to bind cytoplasmic Na+ . It could be site III , as suggested by Kanai et al . , and this prepares the other two sites for the following Na+ binding ( Kanai et al . , 2013 ) . Or , alternatively , the first two Na+ bind at sites I and II in the main binding pocket and the last Na+ enters , takes over site II , and pushes the two previously bound ions further , so that the ions in sites II and I now move to sites I and III , in a process reminiscent of the “knock-on” mechanism occurring in potassium channels ( Hodgkin et al . , 1955 ) . A direct assessment of these proposed binding sequences will require further experiments .
The results from the MD simulations support the notion that the protonation state of the binding pocket and its selectivity are closely related . Because the binding pocket in the Na+/K+-pump displays considerable flexibility , it is worth pausing to reflect on the possible mechanism that underlies this relation . While the selectivity of a very rigid binding site is first and foremost predetermined by its atomic geometry , the selectivity of a flexible binding site is strongly affected by local ion-ligand and ligand-ligand interactions . In such flexible systems , selectivity is controlled by the both number and the physicochemical properties of ion-coordinating ligands ( Yu et al . , 2010 ) . For example , high-field ligands , such as deprotonated acidic side chains tend to favor Na+ binding and protonation revert those to low-field ligands , which tend to favor K+ . Hence , in the Na+/K+-pump protonation is exploited to modulate selectivity by altering the electrostatic properties of several of residues in the binding pocket . This is also consistent with the results from our previous study on the ion selectivity in E2 ( K2 ) , which concluded that changes in the electrostatic properties of the protonatable residues was the likely mechanism responsible for the K+ selectivity in the E2 state of the pump ( Yu et al . , 2011 ) . Even though the local structural rearrangements at the binding sites are small upon the change in protonation state , it is enough to change the physical properties of the coordinating ligand , thus giving rise to discernable differences in the ion selectivity . The crystal structures of the Na+/K+-pump in its Na3 . E1· ( ADP·Pi ) and E2 ( K2 ) states show similarity in their binding pocket configurations ( Figure 1 ) , including the coordination patterns of the bound ions at sites I and II ( Table 1 ) . The heavy atom RMSD between the binding pocket residues is 2 . 5 Å and the structural difference remains after hundreds of ns of simulations . Empirical pKa and FEP calculations based on the MD simulation equilibrated structures indicate that the binding pocket glutamates ( i . e . , E327 , E779 , and E954 ) are likely protonated in both Na3 . E1· ( ADP·Pi ) and E2 ( K2 ) , although previous mutagenesis experiments showed that charge-neutralizing mutation E327Q affects pump function , possibly by altering ion-pump interactions and the kinetics of the occlusion/deocclusion reactions along the pump cycle ( Jorgensen et al . , 2001; Kuntzweiler et al . , 1995; Li et al . , 2006; Nielsen et al . , 1998 ) . The calculations also suggest that the protonation states of D804 , D808 , and D926 are different in Na3·E1· ( ADP·Pi ) and E2 ( K2 ) . Among the three aspartates only D804 is protonated in order for all three sites to stay Na+ selective in Na3·E1· ( ADP·Pi ) . The E1 binding pocket devoid of ions carries a net charge of -2 , consistent with that deduced from previous fluorescence studies ( Schneeberger et al . , 1999 ) . Mutagenesis experiments are also consistent with the unlikely protonation of D808 and D926 when the pump is in E1 trying to bind cytoplasmic Na+ ( Jewell-Motz et al . , 1993; Pedersen et al . , 1997 ) . The protonation states of these three residues are reversed in E2 ( K2 ) with D804 deprotonated and the other two protonated , a result that is supported by a previous computational study ( Yu et al . , 2011 ) . The different protonation states for E1 and E2 also agree well with the 'four-site' model proposed by Skou and Esmann more than 30 years ago with the K+-bound E2 state carrying two sidechain protons ( H2EK2 ) and the Na+-bound E1 state carrying only one ( HENa3 ) ( Skou et al . , 1980 ) . It is worth pointing out that the second proton ( i . e . , the proton on D926 ) in the K+ bound E2 state must come from and return to the same side of the membrane during the pump cycle so that the net charge moved per cycle by the pump is one . It seems unlikely that this proton could come from the extracellular side because altering extracellular pH would then protonate or deprotonate D926 , causing major changes in Na+ binding affinity and the maximum pumping turnover rate . This is not observed over an external pH range 9 . 6 to 5 . 6 ( Mitchell et al . , 2014; Vasilyev et al . , 2004; Yu et al . , 2011 ) . Therefore , it is more likely that the D926 proton comes from the cytoplasmic side . This is supported by MD simulations of the P-E2·Na2 and Na2·E1· ( ADP·Pi ) revealing the existence of aqueous pathways connecting the cytoplasm and D926 ( Figure 6 ) . One of the water pathways is located between the helices M5 , M7 , and M8 ( Figure 6B ) , similar to the C-terminal proton pathway previously proposed ( Poulsen et al . , 2010 ) . A proton could enter through this pathway to protonate D926 and then leave through the same pathway during the E2 to E1 transition , or through an alternative path passing the main binding pocket along with the dissociating ions as seen in the Na2·E1· ( ADP·Pi ) simulation ( Figure 6A ) . 10 . 7554/eLife . 16616 . 013Figure 6 . Water pathways from the cytoplasm to D926 in ( A ) Na2·E1· ( ADP·Pi ) and ( B ) P-E2·Na2 . The top ( top ) and side ( bottom ) views are shown . D926 are shown in sphere representation . Water path connecting the cytoplasm and the D926 are in surface representation colored in blue . Na+ ( yellow ) are shown as spheres . DOI: http://dx . doi . org/10 . 7554/eLife . 16616 . 013 Under physiological conditions , the challenge faced by the Na+/K+-pump is to effectively pick out the correct ion species from a solution much more concentrated with other types of ions . Even when the binding affinity of the ion species being transported is a few orders of magnitude higher than that of the other ion ( ΔΔGbind = 1 to 3 kcal/mol ) , the advantage in binding is undermined by the higher concentration of the competing ion . The pump overcomes this problem by raising the free energy barrier for the occlusion step in the presence of incorrect ions in the binding pocket , thus preventing the faulty transport ( Figure 5 and Table 5 ) . This self-correcting mechanism makes sure that the intracellular and the extracellular gates do not close ( i . e . , occlude ) , which is required for the pump cycle to go forward , unless the correct ion configuration is present at the binding pocket . Therefore , this is an inherent part of the ping-pong mechanism the pump uses to transport ions with high selectivity . Although other energetic barriers , like pathways the ions use to travel to the binding site , could contribute to the ion binding specificity , it is not likely the case here as both Na+ and K+ can access the binding pocket in the simulations of the E1 and P-E2 states without ions bound . The concept of a self-correcting pumping mechanism sheds new light on a number of puzzling functional observations . For instance , it explains why E1 displays a strong apparent affinity for K+ in competitive binding assays ( Schneeberger et al . , 2001 ) , even though the pump still binds and occludes Na+ from the cytoplasmic side . The calculations show that all the sites in the Na3·E1· ( ADP·Pi ) state are selective for Na+ , likely because this is a partially occluded state . A wrong combination of binding site ions would not have been occluded and reached such a state . In a fully inward-open conformation , the selectivity of the sites ought to be fairly weak . The high affinity for K+ observed in E1 in these experiments is due to the backward occlusion leading to the K+-bound state E2 ( K2 ) . The proposed self-correcting mechanism parallels the suggestion based on experiments that the phosphorylation and occlusion of E1 requires 3 Na+ bound , and this increases the apparent affinity for Na+ in the normal pump cycle ( Schneeberger et al . , 2001 ) . The self-correcting mechanism can also account for the different effects of Na+ and K+ congeners on the release rate of the occluded 86Rb+ ( a K+ congener ) to the extracellular side upon the “backdoor” phosphorylation in the presence of Pi reported by Forbush ( Forbush , 1987 ) . This backward deocclusion is accelerated in the presence of Na+ , and it follows a single phase , while a much slower deocclusion process with a second slow phase is observed in the presence of K+ or Rb+ . According to the self-correcting mechanism , it is likely that upon the replacement of one of the two occluded 42K+ or 86Rb+ by cold K+ at one site , the pump succeeded to occlude again , resulting in the second slow phase . These explanations are sensible , in turn lending support to the proposed self-correcting mechanism . Given that the selectivity in the outward/inward facing states is after all not as strong as previously thought , such a mechanism must be in place . This analysis suggests that an important component of the overall selectivity emerges from the increased free energy barrier associated with the occlusion process while the wrong type of ion is bound to a site . Although the pump would be kinetically more efficient by preventing the wrong ions from reaching the binding sites to begin with , the free energy calculations do not support the notion of highly selective binding sites for the open-access states . While such mechanism may seem inefficient because the pump must try to figure out that there is an issue with selectivity only after binding of several ions of the wrong type , it is important to realize that the Na+/K+-pump is not a particular fast molecular machine . The estimated turnover rate is less than 100 per second and decreases even further as the transmembrane voltage becomes more negative ( Heyse et al . , 1994 ) . The implication is that the system has plenty of time , at the molecular level , to function near thermodynamic equilibrium . In other words , the pump has not been evolutionarily optimized to be a particularly fast molecular machine , but an energetically efficient one . Thus , even though the idea of a self-correcting ion selectivity mechanism seems counterintuitive and inefficient , it is consistent with the physical conditions under which the pump has to operate . In summary , the present study highlights the tight coupling between the Na+/K+ selectivity of the binding sites , the protonation state of the coordinating residues , and the conformational state of the pump . The important functional consequence of such tight coupling is the necessity to have the correct type and number of ions in the binding pocket for the pumping cycle to proceed forward toward the occlusion step . Because the ion binding selectivity is strongly dependent upon the protein conformation , a self-correcting mechanism counteracting the effect of the ion concentration in the environment ensures an efficient function of the pump .
The crystal structures , 3WGV ( Kanai et al . , 2013 ) and 2ZXE ( Shinoda et al . , 2009 ) , representing the Na+/K+-pump in its Na3·E1 ( ADP·Pi ) and E2 ( K2 ) states were used to build the simulation systems . The structure 3WGV contains two copies of the pump assembly in the asymmetric unit . Since the structural variations between the two assemblies are minimal , only the copy including chains A , B , and G was kept . Several small molecule ligands were co-crystalized with the pump in 3WGV , including an ADP , a AlF4− ion , an oligomycin A , two Mg2+ , three cholesterol , four Na+ , and five 1 , 2-diacyl-sn-glycero-3-phosphocholine molecules for each copy of the pump structure . Among these ligands , oligomycin A was not included in the simulation system . The ion was replaced by a PO43− and POPC molecules were used in place of the 1 , 2-diacyl-sn-glycero-3-phosphocholine . The structure of 2ZXE also contains small molecule ligands including cholesterol , K+ , Mg2+ , and MgF42− . Similarly , a PO43− was used to replace the MgF42− ion . Since the Na+/K+-pump crystal structures were obtained from different organisms , the residue numbers differ slightly . For the sake of convenience , the numbering scheme in the newly resolved 3WGV from pig kidney was adopted in this study . After removing irregularities from the PDB files , the Na+/K+-pump subunits were capped with acetylated N-termini and amidated C-termini . The ectodomain of the β-subunit was not included in simulation systems to reduce the computational cost . The orientation was chosen to be the same as in the OPM database ( Lomize et al . , 2006 ) . At this stage , the protonation states of the binding site residues were assigned with the PATCH command in CHARMM ( Brooks et al . , 2009 ) . Eight different protonation states were considered in Na3·E1 ( ADP·Pi ) state systems and both the protonated and charged D926 were included in the E2 ( K2 ) state systems . This resulted in a total of ten systems before proceeding to the next step . Table 3 shows the system names and the associated binding site protonation . We used the membrane builder module in CHARMM-GUI ( Jo et al . , 2007 , 2008 , 2009; Wu et al . , 2014 ) to generate POPC bilayers around the pump structures . Once this was completed , each protein-membrane complex was solvated by an equal molar mixture of KCl and NaCl . The final system had a combined cation concentration of 0 . 3 M ( i . e . , [K+] = [Na+] = 0 . 15 M ) . At the end the E1 system was 84 × 110 × 158 Å3 in size and contained ~138 , 000 atoms , while the dimension of the E2 system was 85 × 109 × 105 Å3 with ~134 , 000 atoms . Each completed system was subjected to a 675-ps equilibration with reducing restraints on the heavy atoms to relax the initially uncorrelated components , followed by a 10-ns unrestrained pre-production using the simulation package NAMD2 . 9 ( Phillips et al . , 2005 ) . After the systems were well equilibrated , they were simulated longer using the special-purpose supercomputer Anton ( Shaw et al . , 2009 ) , which is designed for long time scale MD simulations . Experimentally asparagine and glutamine are used as surrogates for protonated aspartate to study the effect of protonation . The effects of these charge-neutralizing mutants on the selectivity of the pump can be investigated computationally . The most interesting mutations in the context of this study are the single D to N mutations at the binding pocket , including residues D804 , D808 , and D926 . The mutations were made on the crystal structures by replacing the proton on the OD2 atom in the aspartate with an –NH2 amine group . D804N and D808N mutations were also made in the outward facing P-E2 structural model ( see below ) to study how they affect the external K+ binding . The protonation states of the other titratable residues were determined with PROPKA . Each of these systems was equilibrated without any restraints for 40 ns . The system snapshot with the least structural deviation of the pump to the averaged structure during the simulation was used to generate the reduced system . The mutant systems are shown in Table 3 . The absolute free energy of an ion i binding to a binding site inside a protein has the following form , ( 1 ) ΔGi , bind= ( Gi , intsite−Gi , intbulk ) +[−kBTIn ( FtC∘ ) −Gi , transsite] . The difference in the first term , Gi , intsite−Gi , intbulk , represents the nonbonded interaction component of the free energy change upon moving the ion from the bulk solution to the binding site . The subtraction in the second term , −kBTIn ( FtC∘ ) −Gi , transsite , reflects the lost of translational freedom . The translational freedom factor Ft in bulk solution can be evaluated analytically under a rigid rotor approximation ( Deng et al . , 2006 ) . Its final form depends on the force constants and the equilibrium values in the distance and angle restraints applied on the ion and the surrounding protein atoms . Based on Equation ( 1 ) , the selectivity of a binding site can be expressed as the binding free energy difference of two ion species . For example , the binding free energy change upon changing ion i to j at the binding site is ( 2 ) ΔΔGi→j= ( Gj , intsite−Gj , intbulk ) − ( Gi , intsite−Gi , intbulk ) + ( Gi , transsite−Gj , transsite ) =ΔGi→jsite−Gi→jbulk−Gi→jtrans . A negative value of ΔΔGi→j indicates that ion j binds more favorably than ion i and the site is j selective . There are three terms to be evaluated in Equation ( 2 ) . ΔGi→jsite and ΔGi→jtrans are calculated using the reduce binding site model , while ΔGi→jbulk is computed using a water sphere with the impact from the bulk solution factored in with a boundary potential ( see below ) . To generate the reduced binding site , the all-atom system prepared according to the procedures above was divided into an inner region and an outer region . The inner region was defined as residues and water molecules within 15 Å to the center of mass of the bound ions . Everything within this region was treated explicitly . An extended inner region was specified by a 3-Å thick shell continuing from the inner region outwards to create a smooth spherical dielectric cavity . Water molecules in this region were removed and their impact on the binding site was included implicitly . The coordinates of atoms in these extended region and those linked to them within three atomic bonds in the inner region were held fixed during the simulations . The rest of the system was considered the outer region , in which the atoms were removed and their impact on the inner region atoms was represented by the General Solvent Boundary Potential ( GSBP ) in the form of a solvent-shielded static field and a solvent-induced reaction field ( Im et al . , 2001 ) . The reaction field arising from changes in charge distribution in the inner region was expressed in terms of a generalized multipole expansion using 11 spherical harmonic functions . Both the solvent-shielded static field and the reaction field matrix were independent of the inner region configuration , and therefore were calculated only once before further simulations using the finite-difference Poisson−Boltzmann ( PB ) method with the PBEQ module ( Im et al . , 1998 ) in CHARMM . In these calculations a dielectric constant of 1 was used for the inside of the protein within the inner and outer regions , whereas the rest of the system had a dielectric constant of 80 . The atomic Born radii used in the PB calculations were determined by free energy calculations in explicit solvent ( Nina et al . , 1997 ) . All these reduced systems were further equilibrated for 200 ps using Langevin dynamics at 303 . 15 K with a friction coefficient of 5 ps−1 assigned to all non-hydrogen atoms . All bonds involving hydrogen atoms were fixed with the SHAKE algorithm ( Ryckaert et al . , 1977 ) . Nonbonded interactions within 14 . 5 Å were accounted for explicitly , while everything else beyond this distance was treated with an extended electrostatics method ( Stote et al . , 1991 ) . The simulation program CHARMM ( Brooks et al . , 2009 ) was used for the equilibration . Table 3 lists all the reduced systems . It is known that the binding free energy calculated from FEP/MD simulations suffers from convergence issues because the residue sidechains at the binding site only sample a few of the several accessible rotameric states . This problem can be alleviated by introducing a replica-exchange scheme aiming at enhancing the sampling of sidechains ( Jiang et al . , 2009 , 2010 ) . This scheme allows exchange between the neighboring λ windows . Each λ window is coupled with a set of replica systems with a boosting potential of increasing strength used to accelerate the inter-conversion between sidechain rotameric states . We employed this hybrid FEP/H-REMD scheme implemented in the REPDSTR module in CHARMM ( Jiang et al . , 2010 ) to calculate the ΔΔGi→j . The boosting potential for the χ1 sidechain torsion of a binding site residue was obtained by fitting the potential of mean force as a function of the torsion χ , 𝒲 ( χ ) , with a cosine Fourier series in the form of ( 3 ) UBP ( χ ) =∑n=1nKn{1+cos[n ( χ−χ0 , n ) ]} . The angle χ1 is dihedral formed by the bonded atoms N , CA , CB , and CG . The total number of the cosine terms , N , varies from 3 to 6 , depending on which one produce a better fit to the 𝒲 ( χ ) . A residue is considered in the binding site if any of its sidechain heavy atoms is within 4 . 5 Å to the bound Na+ or K+ . Table 6 lists all the binding site residues included in the boosting potential calculations . 10 . 7554/eLife . 16616 . 014Table 6 . Binding site residues and the fitted parameters ki and χ0 , i . DOI: http://dx . doi . org/10 . 7554/eLife . 16616 . 014k1χ0 , 1k2χ0 , 2k3χ0 , 3k4χ0 , 4k5χ0 , 5k6χ0 , 6M4E327p-0 . 93253 . 111 . 05561 . 081 . 84462 . 05-0 . 31256 . 440 . 39289 . 040 . 22582 . 28M5Y771-2 . 1962 . 511 . 21989 . 272 . 4660 . 910 . 66240 . 850 . 72144 . 750 . 63950 . 42T774-4 . 2535 . 822 . 437101 . 33 . 60261 . 711 . 12844 . 861 . 07191 . 960 . 85776 . 09S7753 . 068-26 . 370 . 12438 . 671 . 29162 . 22N7762 . 664-166 . 230 . 91887 . 151 . 70962 . 85-0 . 5371 . 79-0 . 49550 . 76-0 . 40675E779p-2 . 50829 . 171 . 79288 . 012 . 59162 . 280 . 48724 . 810 . 47673 . 740 . 46253 . 74M6D804-3 . 895-172 . 950 . 48980 . 041 . 89760 . 56-0 . 19782 . 91-0 . 27182 . 85-0 . 19379 . 47D804p2 . 968-157 . 980 . 53281 . 761 . 83560 . 45-0 . 24731 . 05-0 . 29689 . 84-0 . 27883 . 37N8042 . 981-160 . 180 . 371841 . 45159 . 42D808-3 . 215-159 . 940 . 2157 . 871 . 74459 . 13D808p-2 . 29855 . 971 . 20466 . 882 . 4462 . 530 . 74637 . 740 . 71555 . 740 . 62763 . 72N808-2 . 34429 . 851 . 77163 . 62 . 65464 . 990 . 91439 . 190 . 7358 . 370 . 6361 . 7M8Q923-2 . 94333 . 42 . 21787 . 952 . 67161 . 480 . 60374 . 20 . 53530 . 650 . 55144 . 23D926-4 . 361-169 . 10 . 45495 . 511 . 90260 . 16-0 . 28392 . 02-0 . 30688 . 18-0 . 21585 . 37D926p-2 . 70428 . 741 . 47793 . 382 . 74461 . 080 . 81658 . 990 . 85258 . 370 . 80958 . 93N9263 . 674-157 . 180 . 47989 . 171 . 37761 . 78-0 . 62582 . 5910 . 7554/eLife . 16616 . 015Figure 7 . Fitting the potential of mean force 𝒲 ( χ ) ( red ) with the boosting potential , UBP ( χ ) ( black ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16616 . 015 We performed umbrella sampling simulations to obtain the 𝒲 ( χ ) for each binding site residue . First , the entire transmembrane helix containing the residue of interest was taken from the crystal structure with its orientation kept the same as in the OPM database . This was to provide a proper secondary structure environment . Next , we replaced the sidechains of all the other residues on the helix to hydrogen atoms to remove their steric effects . An implicit planar membrane model , EEF1/IMM1 ( Lazaridis , 2003; Schneeberger et al . , 1999 ) was used in place of a membrane bilayer to solvate the helix . 72 umbrella windows along χ were set up with a window spacing of of 5° . A harmonic bias potential was applied in each window with a force constant of 100 kcal/mol/rad2 . After the windows were generated , each of them was simulated for 50 ns at 303 . 15 K using Langevin dynamics with a 2 fs time-step . All bond lengths involving hydrogen were fixed with the SHAKE algorithm ( Ryckaert et al . , 1977 ) . The cutoff distance for nonbonded interactions was set to 11 Å . The simulation program CHARMM ( Brooks et al . , 2009 ) was used to conduct the simulations . The resulting distributions of χ were unbiased using the weighted histogram analysis method ( WHAM ) ( Kumar et al . , 1992 ) . Figure 7 shows the 𝒲 ( χ ) and the fitted curves . The fitting parameters and are given in Table 6 . An appropriate boosting potential can be easily applied using the CONS DIHE command in CHARMM , with the sign in front of UBP ( χ ) reversed . This will effectively cancel out the potential barrier between the rotamers of a given residue sidechain . Before computing ΔGi→jsite and ΔGi→jtrans , restraints were set up to restrict the translational freedom of the bound ion of interest . First , three points within the protein region were picked out . They were used along with the position of the bound ion to set up the restraints . We employed the same protocol as in the Ligand Binder module ( Jo et al . , 2013 ) in CHARMM-GUI to select the three protein anchor points , p1 , p2 , and p3 . The relative position of the bound ions to the protein can be defined by the distance r from l1 to p1 , the angle θ between l1 , p1 and p2 , and the torsion ψ along l1 , p1 , p2 , and p3 . The translational restraint potential took the form of ( 4 ) utrans=12[kr ( r−r0 ) 2+kθ ( θ−θ0 ) 2+kψ ( ψ−ψ0 ) 2] . The force constant regarding distance ( i . e . , kr ) was set to 10 kcal/mol/Å2 , and the rest of the force constants were 200 kcal/mol/rad2 . The equilibrium values in utrans were taken from the 200 ps equilibration of the reduced binding site . ΔGNa→Ksite was computed with a set of FEP/H-REMD simulations . The translational restraint , utrans , was applied to restrict the translational freedom of the ion . The Na+ ion was changed alchemically to a K+ with an alchemical coupling factor , λ . 16 λ windows were used . Each λ window included 8 replicas with the strength of boosting potential scaled from 0 to 1 . Exchange attempts were made every 0 . 2 ps and were only allowed between neighboring replicas with different boosting potential strengths in the same λ window and between the neighboring λ windows with zero boosting potential . A total of 128 replica systems were included in the calculations . Each replica system was simulated for 2 ns . To compute ΔGNa→Ktrans at a given binding site , two sets of simulations were set up , one with Na+ at the binding site and the other with K+ for the calculation ofΔGNa , transsite and ΔGK , transsite . The translational and orientational restraints were decoupled gradually ( Beglov et al . , 1994 ) using a coupling factor λ . The λ window and the boosting potential setup were similar to those used in the ΔGNa→Ksite calculations . All the FEP/H-REMD calculations were performed using the REPDSTR module in CHARMM ( Jiang et al . , 2010 ) . The output energies from the zero boosting potential systems were collected and processed using WHAM ( Kumar et al . , 1992 ) . To compute the standard error of ΔΔG , we divided the trajectories into 10 blocks . The standard deviations of the averaged ΔΔG from all the blocks are computed and reported in Tables 4 and 5 and Figures 3–5 . Cautions should be taken when comparing the calculated and experimentally measured ΔΔG , because the latter usually contains contributions from multiple states of the pump while the calculated ΔΔG is done using a single state . The bulk system was generated by building a water sphere of 10 Å radius and centering it at the origin . The water sphere was made form pre-equilibrated water boxes with TIP3P water molecules . A Na+ ion was placed at the center of the sphere . Any water molecules overlapping with the ion were deleted . The influence of the remaining bulk was approximated by the spherical solvent boundary potential ( Beglov et al . , 1994 ) . The system was equilibrated for 200 ps at 303 . 15 K . Other simulation options were kept the same as described in the reduced binding site system . During the equilibration the position of the ion was restrained using a weak harmonic bias potential with a force constant of 0 . 5 kcal/mol/Å2 . Once equilibrated , the Na+ was gradually changed to a K+ using the PERT module in CHARMM with 11 λ windows . The simulation time for each λ window was 1 ns . The resulting ΔGNa→Kbulk is 18 . 34 kcal/mol after unbiasing the energy outputs with WHAM . The calculated ΔGNa→K for all the binding sites in the systems summarized in Table 3 are shown in Table 4 . To further confirm the protonation state of D804 , we evaluated its pKa shift with additional simulations in explicit solvent using the following formula: ( 5 ) ΔpKa= ( ΔGsitedeprot−ΔGbulkdeprot ) /2 . 303kBT . ΔGsitedeprot and ΔGsitedeprot are the free energy change of aspartate deprotonation in the protein environment and in bulk water , respectively . The reduced system of E1_S1 ( Table 3 ) was used to compute the deprotonation free energy of D804 at the ion binding site ( ΔGsitedeprot ) . To calculate the ΔGbulkdeprot , an aspartic acid residue with an acetylated N-terminus and an amidated C-terminus was put into a pre-equilibrated water sphere of 15 Å radius . As in the calculations of ΔGNa→Kbulk and ΔGK→Nabulk , the impact of the bulk solution beyond the current system was incorporated by the spherical solvent boundary potential ( Beglov et al . , 1994 ) . Alchemical FEP calculations were carried out in both systems . The λ windows were evenly spaced to gradually deprotonate the aspartate . The numbers of windows used in the ΔGsitedeprot and ΔGbulkdeprot calculations were 24 and 10 respectively . Each window was equilibrated for 200 ps and contained 5 ns of sampling . The calculated ΔGsitedeprot and ΔGbulkdeprot are −44 . 8 kcal/mol and −51 . 9 kcal/mol , respectively . Using Equation ( 5 ) , the ΔpKa is 5 . 1 , meaning the pKa of D804 is right shifted by 5 . 1 pK unit when it is in the Na+/K+ pump ion binding site . The final calculated pKa of D804 is 4 . 1 + 5 . 1 = 9 . 2 with 4 . 1 being the pKa value of an aspartate in bulk water ( Berg et al . , 2002 ) , reinforcing the conclusion that D804 is protonated in the E1 state of the pump . Three conformational transition pathways were generated based on the shared homology between the Na+/K+-pump and the Ca2+ SERCA pump . One of such pathways connects states E2 ( K2 ) and P-E2·K2 . The second connects states E2 ( K2 ) and K2·E1 , and the third describes the transition between states Na3·E1-P and P-E2·Na3 . First , A Cα-atom-only transition pathway for each of these was generated using the ANMPathway online server ( http://anmpathway . lcrc . anl . gov/anmpathway . cgi ) ( Das et al . , 2014 ) based on the SERCA pump crystal structures including 3B9B ( Olesen et al . , 2007 ) , 1WPG ( Toyoshima et al . , 2004 ) , and 1VFP ( Toyoshima et al . , 2004 ) . Since both the Na+/K+-pump and the Ca2+ SERCA pump are P-type ATPases , it is reasonable to assume that they have similar pumping mechanisms and they share the same set of states along the pump cycle . Although crystal structures of the Na+/K+-pump are scarce , multiple high-resolution structures of the SERCA pump in different states are available ( Karlsen et al . , 2016 ) . Among them , 3B9B represents the outward facing P-E2 state . All the other SERCA pump structures were aligned based on their transmembrane region Cα positions to the Na+/K+-pump and the Cα RMSD were computed to find those that resemble the structural states captured by the Na+/K+-pump crystal structures 3WGV ( Kanai et al . , 2013 ) and 2ZXE ( Shinoda et al . , 2009 ) . The two SERCA pump structures representing states Na3·E1ATP ( 3WGV ) and E2 ( K2 ) ( 2ZXE ) are 1VFP and 1WPG , respectively . Each generated coarse-grained pathway is made of a sequence of structural snapshots containing Cα atoms only . These snapshots are called images and are distributed at equal RMSD intervals . The images from the coarse-grained pathways were then used in the all-atom targeted molecular dynamics ( TMD ) simulations to guide each transition . The starting system in the first and second pathway TMD simulations was taken from the well-equilibrated MD simulation system E2_S1 starting from the crystal structure 2ZXE . The third pathway was realized using the MD simulation system E1_S1 built from the crystal structure 3WGV , with the catalytic D369 phosphorylated . The protocol of the TMD simulations followed those published before ( Castillo et al . , 2015 ) Binding site ion selectivity was evaluated for nine systems representing different stages along the pump cycle . Among these systems are well-defined states: P-E2·K2 , E2 ( K2 ) , K2·E1 , Na3·E1 ( ADP·Pi ) , and P-E2·Na3 . The structures used for them are ① the generated P-E2·K2 model , ③ the equilibrated crystal structure 2ZXE , the equilibrated crystal structure 3WGV with ⑦ bound Na+ and with ⑤ bound Na+ replaced by K+ at sites I and II , and ⑨ the generated P-E2·Na3 model . Several intermediate states were also included . Intermediate state ② is between states ① P-E2·K2 and ③ E2 ( K2 ) , taken from the TMD simulations of the first path after image 35 . State ④ was taken at image 33 along the transition from state ③ E2 ( K2 ) to ⑤ K2·E1 . Along the same vein , state ⑧ represented the intermediate at image 29 between states ⑦ Na3·E1 ( ADP . Pi ) and ⑨ P-E2·Na3 . The intermediates were taken near the midpoints of the transitions . An additional state ⑥ K2·E1* with altered binding site residue protonation was also included to reflect the protonation state change upon the E2 to E1 transition . All the states included in the selectivity calculations and their relative placement along the pump cycle can be seen in Figure 5 . To evaluate the selectivity at a given binding site in a system , three simulations were performed with slightly varied Lennard-Jones ( LJ ) parameters for the ion of interest . In the first simulation , the parameters of the ion were unaltered . A linear interpolation of the LJ and the NBFIX parameters between Na+ and K+ was made . In the second simulation , the LJ and NBFIX parameters of the ion were replaced by those of a Na/K hybrid at the middle point of the interpolation . In the last simulation , the ion’s parameters were completely changed to represent the other ion type , either K+ or Na+ , depending on the starting ion type . 10 ns trajectories were generated for each simulation using the simulation package OpenMM6 . 2 ( Eastman et al . , 2013 ) . Energies were calculated from each trajectory using two parameter sets . The difference was fed into the WHAM equation ( Kumar et al . , 1992 ) and the free energy changes upon mutating the ion to the intermediate as well as upon mutating the intermediate to the other ion type were solved self-consistently . The sum of the two free energies gives the ΔGsite . The same strategy was used to compute the ΔGbulk in water ( ΔGbulk kcal/mol ) . The binding free energy difference is given by ΔΔGbind=ΔGsite−Gbulk . The selectivity of binding site with different occupancies was also included in the calculations . The ΔΔGbind are shown in Table 5 . The initial relaxation and the restraint-free equilibration of the membrane pump systems were performed using the NAMD2 . 9 ( Phillips et al . , 2005 ) simulation package with the input scripts from the CHARMM-GUI membrane builder module . The NAMD simulation temperature was set to 303 . 15 K using Langevin Dynamics with a damping coefficient of 10 ps−1 during the relaxation and 1 ps−1 during the restraint-free equilibration . The van der Waals interactions were smoothly switched off at 10–12 Å by a force switching function ( Steinbach et al . , 1994 ) and the electrostatic interactions were calculated using the particle-mesh Ewald method with a mesh size of ~1 Å . After a short equilibration with NAMD , each all-atom system listed in Table 3 were subjected to a hundred ns scale long production without any restraints using the special-purpose supercomputer Anton ( Shaw et al . , 2009 ) . The volume of the periodic cell was kept constant and the temperature was set to 303 . 15K using the Nosé-Hoover thermostats ( Martyna et al . , 1992 ) . The lengths of all bonds involving hydrogen atoms were constrained using M-SHAKE ( Kräutler et al . , 2001 ) . The cutoff of the van der Waals and short-range electrostatic interactions was set to an optimal value suggested by the Anton guesser script , guess_chem . Long-range electrostatic interactions were evaluated using the k-space Gaussian split Ewald method ( Shan et al . , 2005 ) with a 64 × 64 × 64 mesh . The integration time step was 2 fs . The r-RESPA integration method ( Tuckerman et al . , 1992 ) was employed and long-range electrostatics were evaluated every 6 fs . Simulations used to compute ion selectivity along the pump cycle were conducted using the simulation package OpenMM6 . 2 ( Eastman et al . , 2013 ) . The constant pressure and temperature ( NPT ) ensemble was used for these simulations . The pressure was maintained at 1 atm with a Monte Carlo barostat , which attempts to adjust the system volume every 0 . 2 ps . The Langevin dynamics algorithm with a 1 . 0 ps−1 friction coefficient was used to hold the simulation temperature , which is at 303 . 15 K . The lengths of all bonds involving hydrogen were fixed and the integration time step was set to 2 fs . A force switching function was applied from 10 to 12 Å to gradually turn off the van der Waals interactions and the particle-mesh Ewald method with an error tolerance of 0 . 0005 was used to evaluate the electrostatic interactions . The same force field parameters were used in all other simulations in this study . The PARAM27 all-atom force field of CHARMM ( MacKerell et al . , 1998 ) with a modified version of dihedral cross-term correction ( Mackerell et al . , 2004 ) was used for the protein and the C36 lipid force field ( Klauda et al . , 2010 ) was used for POPC . Water molecules were modeled with the TIP3P potential ( Jorgensen et al . , 1983 ) . | A protein called the sodium-potassium pump resides in the membrane that surrounds living cells . The role of this protein is to 'pump' sodium and potassium ions across the membrane to help restore their concentration inside and outside of the cell . About 25% of the body's energy is used to keep the pump going , rising to nearly 70% in the brain . Problems that affect the pump have been linked to several disorders , including heart , kidney and metabolic diseases , as well as severe neurological conditions . The sodium-potassium pump must be able to effectively pick out the correct ions to transport from a mixture of many different types of ions . However , it was not clear how the pump succeeds in doing this efficiently . Rui et al . have now used a computational method called molecular dynamics simulations to model how the sodium-potassium pump transports the desired ions across the cell membrane . The pump works via a so-called 'alternating-access' mechanism , repeatedly transitioning between inward-facing and outward-facing conformations . In each cycle , it binds three sodium ions from the cell’s interior and exports them to the outside . Then , the pump binds to two potassium ions from outside the cell and imports them inside . Although the bound sodium and potassium ions interact with similar binding sites in the pump , the pump sometimes preferentially binds sodium , and sometimes potassium . The study performed by Rui et al . shows that this preference is driven by how protons ( hydrogen ions ) bind to the amino acids that make up the binding site . The simulations also suggest that the pump uses a ‘self-correcting’ mechanism to prevent the pump from transporting the wrong types of ions . When incorrect ions are present at the binding sites , the pump cycle pauses temporarily until the ions detach from the pump . Only when the correct ions are bound will the pump cycle continue again . In the future , Rui et al . hope to use long time-scale molecular dynamics simulations to show the conformational transition in action . In addition , the 'self-correcting' mechanism will be directly tested by letting the wrong and correct ions compete for the binding sites to see whether the pump will transport only the correct ions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics",
"computational",
"and",
"systems",
"biology"
] | 2016 | The selectivity of the Na+/K+-pump is controlled by binding site protonation and self-correcting occlusion |
Catechol dehydroxylation is a central chemical transformation in the gut microbial metabolism of plant- and host-derived small molecules . However , the molecular basis for this transformation and its distribution among gut microorganisms are poorly understood . Here , we characterize a molybdenum-dependent enzyme from the human gut bacterium Eggerthella lenta that dehydroxylates catecholamine neurotransmitters . Our findings suggest that this activity enables E . lenta to use dopamine as an electron acceptor . We also identify candidate dehydroxylases that metabolize additional host- and plant-derived catechols . These dehydroxylases belong to a distinct group of largely uncharacterized molybdenum-dependent enzymes that likely mediate primary and secondary metabolism in multiple environments . Finally , we observe catechol dehydroxylation in the gut microbiotas of diverse mammals , confirming the presence of this chemistry in habitats beyond the human gut . These results suggest that the chemical strategies that mediate metabolism and interactions in the human gut are relevant to a broad range of species and habitats .
The human gastrointestinal tract is one of the densest microbial habitats on Earth . Possessing 150-fold more genes than the human genome , the trillions of organisms that make up this community ( the human gut microbiota ) harbor metabolic capabilities that expand the range of chemistry taking place in the body ( Koppel et al . , 2017; Qin et al . , 2010; Sender et al . , 2016 ) . Microbial metabolism affects host nutrition and health by breaking down otherwise inaccessible carbohydrates , biosynthesizing essential vitamins , and transforming endogenous and exogenous small molecules into bioactive metabolites ( Koppel and Balskus , 2016 ) . Gut microbial activities can also vary significantly between individuals , affecting the toxicity and efficacy of drugs ( Zimmermann et al . , 2019; Koppel et al . , 2018; Gopalakrishnan et al . , 2018; Wallace et al . , 2010; Haiser et al . , 2013 ) , susceptibility to infection ( Buffie et al . , 2015; Devlin and Fischbach , 2015 ) , and host metabolism ( Yao et al . , 2018; Romano et al . , 2017 ) . To decipher the biological roles of gut microbial metabolism , it is critical that we uncover the enzymes responsible for prominent transformations . This will not only increase the information gained from microbiome sequencing data but may also illuminate strategies for manipulating and studying microbial functions . Yet , the vast majority of gut microbial metabolic reactions have not yet been linked to specific enzymes . A prominent but poorly understood gut microbial activity is the dehydroxylation of catechols ( 1 , 2-dihydroxylated aromatic rings ) , a structural motif commonly found in a diverse range of compounds that includes dietary phytochemicals , host neurotransmitters , clinically used drugs , and microbial siderophores ( Wilson et al . , 2016; Ozdal et al . , 2016; Yang et al . , 2007 ) ( Figure 1A ) . Discovered over six decades ago , catechol dehydroxylation is a uniquely microbial reaction that selectively replaces the para hydroxyl group of the catechol with a hydrogen atom ( Scheline et al . , 1960 ) ( Figure 1A ) . This reaction is particularly challenging due to the stability of the aromatic ring system . Prominent substrates for microbial dehydroxylation include the drug fostamatinib ( Sweeny et al . , 2010 ) , the catecholamine neurotransmitters norepinephrine and dopamine ( Smith et al . , 1964; Sandler et al . , 1971 ) , the phytochemicals ellagic acid ( found in nuts and berries ) , caffeic acid ( a universal lignin precursor in plants ) , and catechin ( present in chocolate and tea ) ( Peppercorn and Goldman , 1972; Cerdá et al . , 2005; Takagaki and Nanjo , 2010 ) ( Figure 1B ) . Dehydroxylation alters the bioactivity of the catechol compound ( Kim et al . , 2016; Ryu et al . , 2016 ) and produces metabolites that act both locally in the gut and systemically to influence human health and disease ( Sweeny et al . , 2010; Ryu et al . , 2016; Kang et al . , 2016; Pietinen et al . , 2001; Mabrok et al . , 2012; Maini Rekdal et al . , 2019; Singh et al . , 2019 ) . However , the gut microbial enzymes responsible for catechol dehydroxylation have remained largely unknown . We recently reported the discovery of a catechol dehydroxylating enzyme from the prevalent human gut Actinobacterium Eggerthella lenta . This enzyme participates in an interspecies gut microbial pathway that degrades the Parkinson’s disease medication L-dopa by catalyzing the regioselective p-dehydroxylation of dopamine to m-tyramine ( Maini Rekdal et al . , 2019 ) . To identify the enzyme , we grew E . lenta strain A2 with and without dopamine and used RNA sequencing ( RNA-seq ) to find genes induced by dopamine . Only 15 genes were significantly upregulated in the presence of dopamine , including a putative molybdenum-dependent enzyme that was induced >2500 fold . Hypothesizing this gene encoded the dopamine dehydroxylase , we purified the enzyme from E . lenta and confirmed its activity in vitro . Dopamine dehydroxylase ( Dadh ) is predicted to bind bis-molybdopterin guanine nucleotide ( bis-MGD ) , a complex metallocofactor that contains a catalytically essential molybdenum atom ( Hille et al . , 2014 ) . Our previous work illuminated a role for Dadh in dopamine metabolism by pure strains and complex communities . Here , we sought to explore the substrate scope of Dadh and its broader role in catechol dehydroxylation by the gut microbiota .
Because the human gut microbiota metabolizes a range of catecholic compounds ( Figure 1B ) , we first investigated whether the recently discovered Dadh possessed promiscuous dehydroxylase activity . We evaluated the reactivity of natively purified E . lenta A2 Dadh towards a panel of established or potential host- and diet-derived catechol substrates ( Supplementary file 1a and Figure 1—figure supplement 1 ) . This enzyme displayed a narrow substrate scope , metabolizing only dopamine and the structurally related neurotransmitter norepinephrine , which differ only by the presence of a benzylic hydroxyl group ( Figure 1C ) . To identify the elements necessary for substrate recognition by Dadh , we profiled its activity towards synthetic and commercially available dopamine analogs ( Figure 1D , Figure 1—figure supplement 1 , and Supplementary file 1b ) . We found that Dadh tolerated only minor modifications to the dopamine scaffold , including a single N-methylation and the presence of additional hydroxyl groups on the aromatic ring ( Figure 1E ) . The catechol moiety was absolutely necessary for activity , and dehydroxylation required that at least one hydroxyl group be in the para position relative to the aminoethyl substituent . These data demonstrated that Dadh specifically recognizes the catecholamine scaffold . This result prompted us to explore whether the transcriptional regulation of Dadh displayed similar specificity . Thus , we cultured E . lenta A2 in the presence of a subset of the dopamine analogs that we had tested in the previous experiment , measured dehydroxylation using liquid chromatography-mass spectrometry ( LC-MS ) , and profiled the global transcriptome using RNA-seq . We found that the regulation of dadh was also specific for the catecholamine scaffold ( Figure 1F , Supplementary file 1c ) . While the catecholamines dopamine and norepinephrine induced dadh expression and were dehydroxylated by E . lenta , analogs lacking the catechol ( analog 1 in Figure 1D ) or having a shorter side chain ( analog 9 in Figure 1D ) did not induce a transcriptional or metabolic response ( Figure 1F , Supplementary file 1c ) ( Maini Rekdal et al . , 2019 ) . Together with our biochemical results , these transcriptional data suggest that Dadh may have evolved for the purpose of catecholamine neurotransmitter metabolism in E . lenta . We propose that dopamine is an endogenous substrate of this enzyme , because it was the best substrate both in vitro and in vivo , induced the highest levels of expression in E . lenta , and is produced at substantial levels within the human gastrointestinal tract ( Eisenhofer et al . , 1997 ) . In addition to uncovering a preference for the catecholamine scaffold , the substrate scope of Dadh reveals potential mechanistic distinctions between this enzyme and the only other biochemically characterized reductive aromatic dehydroxylase , 4-hydroxybenzoyl Coenzyme A ( CoA ) reductase ( 4-HCBR ) ( Unciuleac et al . , 2004 ) . 4-HCBR is a distinct molybdenum dependent-enzyme containing a monomeric molybdopterin co-factor that uses a Birch reduction-like mechanism to remove a single aromatic hydroxyl group from 4-hydroxybenzoyl CoA . While 4-HCBR requires an electron-withdrawing thioester group to stabilize radical anion intermediates ( Unciuleac et al . , 2004 ) , Dadh does not require an electron-withdrawing substituent and can tolerate additional electron-donating hydroxyl groups ( Figure 1D and E , analogs 11–13 ) . We preliminarily propose a mechanism for Dadh in which the dopamine p-hydroxyl group coordinates to the molybdenum center . This could be followed by tautomerization of the m-hydroxyl group to a ketone with protonation of the adjacent carbon atom . Oxygen atom transfer to molybdenum could be accompanied by rearomatization , providing the dehydroxylated product ( Figure 1—figure supplement 2 ) . Our proposal is consistent with the postulated mechanisms of other oxygen transfer reactions catalyzed by bis-MGD enzymes ( Tenbrink et al . , 2011; Hille et al . , 2014 ) . The specificity of Dadh for dopamine suggested this metabolic activity might have an important physiological role in E . lenta . We noted the chemical parallels between catechol dehydroxylation and reductive dehalogenation , a metabolic process in which halogenated aromatics serve as alternative electron acceptors in certain environmental bacteria ( Holliger et al . , 1998 ) . This insight inspired the hypothesis that dopamine dehydroxylation could serve a similar role in gut bacteria . While we observed no growth benefit when E . lenta was grown in complex BHI medium containing dopamine ( Figure 2—figure supplement 1 ) , we found that including dopamine in a minimal medium lacking electron acceptors ( basal medium ) increased the endpoint optical density of E . lenta cultures ( Figure 2A ) . This growth-promoting effect was only observed in dopamine-metabolizing E . lenta strains , as non-metabolizing strains that express an apparently inactive enzyme ( Maini Rekdal et al . , 2019 ) did not gain a growth advantage ( Figure 2A and Figure 2—figure supplement 2 ) . The effect of dopamine on E . lenta contrasts with recent studies of digoxin , a drug that is reduced by E . lenta without impacting growth in the same medium ( Koppel et al . , 2018 ) . We further investigated the relationship between dopamine and bacterial growth in the metabolizing strain E . lenta A2 . The growth increase observed in response to dopamine was dose-dependent ( Figure 2—figure supplement 3 ) , mirrored the effects of the known electron acceptors DMSO and nitrate ( Koppel et al . , 2018; Sperry and Wilkins , 1976 ) , and did not derive from the product of dopamine dehydroxylation , m-tyramine ( Figure 2B ) . Additionally , the growth benefit was directly tied to dopamine dehydroxylation . Inclusion of tungstate in the growth medium , which inactivates the big-MGD cofactor of Dadh , blocked metabolism and inhibited the growth increase . In contrast , inclusion of molybdate in the growth medium did not impact growth or metabolism ( Figure 2B and Figure 2—figure supplement 4 ) . Molybdate and tungstate alone did not impact E . lenta A2 growth in the basal medium ( Figure 2—figure supplement 5 ) . Taken together , these results indicate that active metabolism of dopamine provides a growth advantage to E . lenta , likely by serving as an alternative electron acceptor . We next examined whether dopamine could promote E . lenta growth in microbial communities . First , we competed dopamine metabolizing and non-metabolizing E . lenta strains in minimal medium . E . lenta is genetically intractable , preventing the use of engineered plasmids encoding defined fluorescence or antibiotic resistance as markers of strain identity . Instead , we took advantage of intrinsic differences in tetracyline ( Tet ) resistance to differentiate the closely related strains in pairwise competitions ( Bisanz et al . , 2018 ) . Inclusion of dopamine in growth medium significantly increased the proportion of the metabolizer relatively to the non-metabolizer in this competition experiment ( p<0 . 001 , two-tailed unpaired t-test ) ( Figure 2C and Figure 2—figure supplement 6 ) . This was driven by the growth increase of the metabolizer rather than a decrease in the non-metabolizer ( Figure 2C and Figure 2—figure supplement 6 ) . Next , we explored the impact of dopamine on Tet-resistant E . lenta in the presence of a defined bacterial community representing the major phylogenetic diversity in the human gut ( Figure 2D and Supplementary file 1d ) ( Devlin et al . , 2016; Romano et al . , 2015 ) . We found that including dopamine in the medium boosted the growth of metabolizers by an order of magnitude while non-metabolizing strains did not gain a growth advantage ( Figure 2D ) . Finally , we evaluated the impact of dopamine on E . lenta strains present in complex human gut microbiotas . We cultured fecal samples from 24 unrelated subjects ex vivo in the presence and absence of dopamine and used qPCR to assess the abundance of E . lenta and dadh . We found that both dadh and E . lenta significantly increased by an order of magnitude in cultures containing dopamine ( p<0 . 005 , two-tailed unpaired t-test ) ( Figure 2E ) ( Figure 2—figure supplement 7 ) . Finally , we amplified the full length dadh gene from these cultures and sequenced the region harboring the SNP that distinguishes metabolizing and non-metabolizing strains ( Maini Rekdal et al . , 2019 ) . These assays indicated that the increase in dadh abundance in the complex communities was accompanied by a shift from a mixture of inactive and active dadh variants to a dominance of the metabolizing R506 variant ( p<0 . 01 , Fisher’s exact test ) ( Figure 2F ) . Finally , we noticed in these growth assays that a small number of samples did not display an increase in E . lenta or dadh abundance ( n = 4 and n = 3 samples , respectively ) ( Figure 2E and Figure 2—figure supplement 7 ) . While the factors influencing this outcome are unclear , they could include the possibility that these specific communities support the growth of E . lenta in other ways that such that dopamine metabolism does not provide any additional advantage , that these samples contain inhibitory factors , or that organisms not targeted by our primers were responsible for metabolism . Altogether , these results are consistent with the hypothesis that dopamine dehydroxylation can increase the fitness of metabolizing E . lenta strains in microbial communities . Having uncovered Dadh’s specialized role in gut bacterial dopamine metabolism , we sought to identify additional gut bacterial strains and enzymes that could dehydroxylate other catechol substrates . Among human gut bacteria , only Eggerthella and closely related members of the Actinobacteria phylum have been reported to perform catechol dehydroxylation . For example , Eggerthella metabolizes dopamine ( Maini Rekdal et al . , 2019 ) and ( + ) -catechin ( Takagaki and Nanjo , 2015 ) , while related Gordonibacter species dehydroxylate ellagic acid ( Selma et al . , 2014 ) and didemethylsecoisolariciresinol ( dmSECO ) , an intermediate in the multi-step biosynthesis of the anti-cancer metabolite enterodiol ( Bess et al . , 2020 ) . These reports suggest that Actinobacteria could be a promising starting point to identify new dehydroxylating strains and enzymes . Thus , we screened a library of related gut Actinobacteria ( Bisanz et al . , 2018 ) ( n = 3 replicates for each strain ) for metabolism of a range of compounds relevant in the human gut , including plant- and host-derived small molecules , bacterial siderophores , and FDA-approved catecholic drugs ( Wilson et al . , 2016; Ozdal et al . , 2016; Yang et al . , 2007 ) ( Supplementary file 1e ) ( Figure 3A ) . We initially used a colorimetric assay that detects catechols to assess metabolism , which allowed us to rapidly screen for potential catechol depletion across the collection of 25 strains . We observed complete depletion of several host and diet-derived catechols in this initial screen ( Figure 3—figure supplement 1 ) . We chose to focus on the dehydroxylation of hydrocaffeic acid , ( + ) -catechin , and DOPAC for further characterization , repeating the incubations with these compounds and using LC-MS/MS to confirm the production of dehydroxylated metabolites . This analysis showed that both DOPAC and hydrocaffeic acid are directly dehydroxylated by members of this library , while ( + ) -catechin undergoes benzyl ether reduction followed by dehydroxylation into derivative 2 , as has been observed previously ( Takagaki and Nanjo , 2015 ) ( Figure 3 ) . While ( + ) -catechin metabolism has been previously linked to Eggerthella ( Takagaki and Nanjo , 2015 ) , the dehydroxylation of DOPAC and hydrocaffeic acid has only been previously observed by complex gut microbiota communities ( Scheline et al . , 1960; Peppercorn and Goldman , 1972 ) . The variability in these activities across closely related gut bacterial strains suggests that distinct enzymes might dehydroxylate different catechols . We next sought to determine the molecular basis of the dehydroxylation reactions examined above . To test the hypothesis specific rather than promiscuous enzymes were involved , we first established that dehydroxylation is an inducible activity in Gordonibacter and Eggerthella strains ( Figure 4—figure supplement 1 ) . This allowed us to use the dehydroxylase activity of cell lysates as a proxy for transcriptional induction and a means of examining dehydroxylase activity . We grew E . lenta A2 in the presence of ( + ) -catechin , hydrocaffeic acid , and dopamine , and grew G . pamelaeae 3C in the presence of DOPAC . We then screened each anaerobic lysate for its activity towards all of these substrates . Consistent with our prediction , each lysate quantitively dehydroxylated only the catechol substrate with which the strain had been grown . While the E . lenta lysates did not display any promiscuity ( Figure 4A ) , cell lysate from G . pamelaeae grown in the presence of DOPAC displayed reduced activity ( <45% conversion ) towards hydrocaffeic acid , which structurally resembles DOPAC ( Figure 4B ) . Overall , these results suggest that different catechol substrates induce the expression of distinct dehydroxylase enzymes that are specific in their activity and transcriptional regulation . We expected that these enzymes would likely resemble Dadh , as only molybdenum-dependent enzymes are known to catalyze aromatic dehydroxylation ( Maini Rekdal et al . , 2019; Hille et al . , 2014; Unciuleac et al . , 2004 ) . To identify the molecular basis of ( + ) -catechin and hydrocaffeic acid dehydroxylation in E . lenta A2 , we turned to RNA-seq , the approach that we used previously to identify the dopamine dehydroxylase ( Maini Rekdal et al . , 2019 ) . We grew E . lenta A2 to early exponential phase and then added each catechol substrate , harvesting the cells after ~1 . 5 hr of induction . Hydrocaffeic acid and ( + ) -catechin each upregulated a number of genes ( 21 and 43 , respectively ) , including two predicted molybdenum-dependent enzymes ( Supplementary files 2a and 2b ) . While one of these predicted molybdenum-dependent enzymes was among the highest upregulated genes in response to each substrate ( 450-fold upregulated in response to catechin , >2000 fold with hydrocaffeic acid ) , the expression of the other enzyme was only increased 3-fold relative to the vehicle . Thus , we propose that the most highly upregulated molybdenum-dependent enzyme in each dataset is the most reasonable candidate dehydroxylase . The candidate hydrocaffeic acid dehydroxylase ( Elenta-A2_02815 , named hcdh ) shares 35 . 3% amino acid identity with Dadh , while the candidate ( + ) -catechin dehydroxylase ( E . lenta-A2_00577 , named cadh ) shares 50 . 9% amino acid identity with Dadh ( Supplementary files 2a-2c ) . To evaluate the involvement of a molybdenum enzyme in each dehydroxylation reaction , we cultured the genetically intractable E . lenta A2 in the presence of tungstate ( Maini Rekdal et al . , 2019; Rothery et al . , 2008 ) . As with dopamine dehydroxylation , tungstate inhibited dehydroxylation of ( + ) -catechin and hydrocaffeic acid by E . lenta A2 without inhibiting growth in the rich BHI medium , suggesting these activities are indeed molybdenum dependent ( Figure 4—figure supplements 2 and 3 ) . Tungstate did not inhibit benzyl ether reduction of ( + ) -catechin , indicating this step is likely performed by a distinct enzyme ( Figure 4—figure supplement 3 ) . Finally , we found that the overall distribution of the genes encoding these the putative hydrocaffeic acid and ( + ) -catechin dehydroxylating enzymes across closely related Eggerthella strains correlated with metabolism of each substrate ( Figure 4C ) . For example , all Eggerthella strains except AN5LG harbored the putative hydrocaffeic acid dehydroxylase and could dehydroxylate this substrate . Similarly , carriage of the putative catechin dehydroxylase correlated with ( + ) -catechin metabolism , except for in the case of strain AB12#2 , which did not encode for the enzyme but still had low metabolism ( <10% ) ( Figure 4C and Figure 4—source data 1 ) . This suggests that another enzyme might metabolize ( + ) -catechin in this strain . Overall , these data suggest that Eggerthella uses distinct molybdenum-dependent enzymes dehydroxylate hydrocaffeic acid and ( + ) -catechin . We next sought to identify the enzyme responsible for DOPAC dehydroxylation in G . pamelaeae 3C . We added DOPAC to G . pamelaeae 3C cultures at mid-exponential phase and harvested cells after 3 hr of induction when the cultures had reached early stationary phase . In this experiment , G . pamelaeae 3C upregulated 100 different genes , including four distinct molybdenum-dependent enzyme-encoding genes ( Supplementary file 2d ) . One of these genes ( C1877_13905 ) was among the highest upregulated genes across the dataset ( >1700 fold induced ) . To further explore the association between this gene and DOPAC dehydroxylation , we repeated the RNA-seq experiment , growing G . pamelaeae 3C in the presence of DOPAC from the time of inoculation and then harvesting cells in mid-exponential phase as soon we could detect metabolism ( 12 hr of growth ) . In this experiment , the same molybdenum-dependent enzyme-encoding gene ( C1877_13905 ) that was highly upregulated in our first experiment ( Supplementary file 2e ) was among the highest upregulated genes . The only two other molybdenum-dependent enzymes induced in this experiment were expressed at an order of magnitude lower levels ( <2 . 5-fold induced ) . We propose that the molybdenum-dependent enzyme encoded by C1877_13905 is a likely candidate DOPAC dehydroxylase . This assignment is also supported by comparative genomics . First , carriage of C1877_13905 ( named dodh ) correlated with DOPAC dehydroxylation among members of our gut Actinobacterial library ( Figure 4C ) . Consistent with our lysate assays , those organisms harboring this gene also had activity towards hydrocaffeic acid , which could explain the pattern of hydrocaffeic acid metabolism across the gut Actinobacterial library ( Figure 4C ) . Finally , the functionally annotated gene most similar to the candidate DOPAC dehydroxylase is Cldh ( 45% amino acid ID ) , a Gordonibacter enzyme recently implicated in the dehydroxylation of the lignan dmSECO ( Bess et al . , 2020 ) ( Supplementary file 2c ) . Though this functional assignment awaits biochemical confirmation , we propose that the highest upregulated enzyme across our two independent datasets is the DOPAC dehydroxylase . Interestingly , unlike with the Eggerthella dehydroxylases , tungstate did not inhibit dehydroxylation of DOPAC by G . pamelaeae ( Figure 4—figure supplement 3 ) . This may be explained if the dehydroxylating enzyme can use both molybdenum and tungsten for catalysis , as is seen in certain closely related enzymes ( Rosner and Schink , 1995 ) . To biochemically validate one of our candidate dehydroxylases , we adapted the native purification protocol used for Dadh to fractionate the hydrocaffeic acid dehydroxylase activity from E . lenta A2 cell lysates . This yielded an active fraction that contained five major bands as assessed by SDS-PAGE ( Figure 4—figure supplement 4 ) and quantitatively dehydroxylated hydrocaffeic acid into m-hydroxyphenylacetic acid under anaerobic conditions ( Figure 4D ) . We confirmed that the band with the apparent correct size ( Figure 4—figure supplement 4 ) contained the proposed hydrocaffeic acid dehydroxylase ( Hcdh , Elenta-A2_02815 ) ( Figure 4—figure supplement 5 and Supplementary file 2f ) using proteomics . We also performed an additional set of enzyme assays using this preparation to evaluate the substrate scope of Hcdh . Consistent with our experiments in cell lysates ( Figure 4A ) , we observed dehydroxylation only of hydrocaffeic acid and not of dopamine , ( + ) -catechin , or DOPAC ( Figure 4E ) . These data biochemically link the newly identified hcdh gene to hydrocaffeic acid dehydroxylation , further supporting the proposal that different enzymes dehydroxylate distinct catechol substrates . Despite being expected to perform the same type of chemical reaction , the putative catechol dehydroxylases from E . lenta and G . pamelaeae differ in sequence identity , genomic context , and predicted subunit composition ( Supplementary file 3a and Figure 4—figure supplement 6 ) . The dopamine , catechin and hydrocaffeic acid dehydroxylases from E . lenta ( dadh , cadh and hcdh , respectively ) are likely membrane-bound complexes as they co-localize with genes encoding an electron shuttling 4Fe-4S ferredoxin and a putative membrane anchor ( Figure 4—figure supplement 6 ) ( Rothery et al . , 2008; Rothery and Weiner , 2015 ) . These enzymes all carry a Twin-Arginine-Translocation ( TAT ) signal sequence , suggesting they are exported from the cytoplasm before the signal sequence is cleaved off . We found no peptide coverage of the TAT signal sequence in the proteomics experiment that identified E . lenta Hcdh , further confirming that this sequence is cleaved in the mature protein as in other membrane-anchored moco enzymes ( Figure 4—figure supplement 5 ) ( Iobbi-Nivol and Leimkühler , 2013 ) . In contrast , dodh and similar enzymes from G . pamelaeae do not harbor a TAT signal sequence , are smaller than the E . lenta enzymes , and co-localize with a gene predicted to encode a small electron shuttling 4Fe-4S protein , suggesting they are likely soluble protein complexes ( Figure 4—figure supplement 6 ) . These putative G . pamelaeae dehydroxylases are also encoded adjacent to members of the Major Facilitator Superfamily , transporters that may import or export the catechol substrates or dehydroxylated metabolites ( Figure 4—figure supplement 6 ) . Altogether , these data indicate the existence of distinct subtypes of molybdenum-dependent catechol dehydroxylases . Our finding that catechol dehydroxylases and their associated metabolic activities are variably distributed among closely related gut Actinobacteria made us wonder whether human gut microbial communities would harbor similar genetic and metabolic diversity . To address this , we first searched >1800 publicly available human gut metagenomes ( Nayfach et al . , 2015 ) for dadh , hcdh , cadh , dodh , and the recently identified cldh ( Bess et al . , 2020 ) genes . Although found at generally low abundances , these catechol dehydroxylases were widely but variably distributed across these metagenomes . Dadh and hcdh were the most prevalent ( in >70% and>90% of individuals , respectively ) , followed by cadh ( 30% ) , dodh ( 20% ) , and cldh ( 25% ) ( Figure 5A ) . Notably , the prevalence of the different genes in metagenomes is consistent with their distribution among individual human gut Actinobacterial isolates ( Figure 4C ) . To assess the presence of catechol dehydroxylation in complex gut microbiotas , we incubated fecal samples from unrelated humans ( n = 12 ) ex vivo with hydrocaffeic acid , ( + ) -catechin , and stable-isotope deuterium-labeled dopamine and DOPAC and analyzed dehydroxylation by LC-MS/MS . In this experiment , we observed dehydroxylation of dopamine , hydrocaffeic acid , and DOPAC across the majority of subjects , indicating that metabolic activities of low-abundance gut Actinobacteria are indeed prevalent ( Figure 5B–D ) . However , catechol metabolism varied between compounds and subjects , with some individuals metabolizing all compounds and some metabolizing none ( Figure 5B–D ) . ( + ) -Catechin was depleted without production of the corresponding dehydroxylated metabolites , consistent with this compound undergoing a wide range of metabolic reactions in complex communities ( Figure 5—source data 1 ) ( Takagaki and Nanjo , 2013; van't Slot and Humpf , 2009; Aura et al . , 2008 ) . To investigate whether metabolic variability correlated with the presence of specific dehydroxylase enzymes , we further investigated DOPAC metabolism . We separated the 12 samples into a group of 9 metabolizers and three non-metabolizers ( in which no biological replicate displayed dehydroxylation activity ) . qPCR enumeration in these cultures revealed that the abundance of the candidate DOPAC dehydroxylase gene dodh discriminated metabolizing and nonmetabolizing subjects ( p<0 . 001 , unpaired t-test ) ( Figure 4E ) , and correlated significantly with dehydroxylation activity within the nine metabolizers ( Pearson’s correlation , r = 0 . 73 , R2 = 0 . 53 , p<0 . 05 ) ( Figure 4F ) . Altogether , these data are consistent with our previous finding that dadh SNP status correlates with dopamine metabolism in human gut microbiotas ex vivo ( Maini Rekdal et al . , 2019 ) and suggest that the candidate dehydroxylases may be active in complex gut communities . We next investigated the relationship of catechol dehydroxylases to other characterized molybdenum-dependent enzymes . These enzymes bear no sequence homology to the only other biochemically characterized aromatic dehydroxylase , 4-HCBR; whereas 4-HCBR belongs to the xanthine oxidase family of molybdenum-dependent enzymes , the catechol dehydroxylases belong to the bis-MGD family of molybdenum-dependent enzymes , suggesting independent evolutionary origins ( Hille et al . , 2014; Unciuleac et al . , 2004; Tenbrink et al . , 2011 ) . Further phylogenetic analysis revealed that catechol dehydroxylases form a unique clade within the bis-MGD enzyme family , clustering away from pyrogallol hydroxytransferase ( Pht ) , the only other bis-MGD enzyme known to modify the aromatic ring of a substrate ( Messerschmidt et al . , 2004 ) ( Figure 6 and Supplementary file 3b ) . The catechol dehydroxylases are instead most closely related to acetylene hydratase , an enzyme that adds water to acetylene to provide a carbon source for the marine Proteobacterium Pelobacter acetylenicus ( Figure 6 ) ( Tenbrink et al . , 2011; Rosner and Schink , 1995; Schoepp-Cothenet et al . , 2012 ) . A sequence similarity network ( SSN ) analysis using sequences of bis-MGD enzymes revealed distinct clusters of catechol dehydroxylases , further suggesting these enzymes are functionally different from known family members ( Figure 7—figure supplement 1 ) . The clustering of the dehydroxylases in the SSN did not simply reflect the phylogeny of the organisms because additional sequences from both Eggerthella and Gordonibacter were found in clusters containing distinct , biochemically characterized enzymes ( Figure 7—figure supplement 2 ) . In addition , we found that the two catechol dehydroxylase-containing clusters also harbored sequences from organisms other than Eggerthella and Gordonibacter ( Figure 7—figure supplement 2 ) . Based on these data , we propose that catechol dehydroxylases are a distinct group of molybdenum-dependent enzymes . To assess the diversity of putative dehydroxylases , we queried the NCBI nucleotide database and our collection of Actinobacterial genomes for homologs of the Eggerthella and Gordonibacter enzymes . Phylogenetic analyses of the resulting sequences revealed a large diversity of putative dehydroxylases , including numerous uncharacterized enzymes encoded in individual Gordonibacter and Eggerthella genomes ( Figure 7 ) . This highlights that catechol dehydroxylases likely have diversified within these closely related gut Actinobacteria , that individual gut Actinobacteria can likely metabolize a range of different catechols , and that many substrate-enzyme pairs remain to be discovered . Our analysis also revealed that catechol dehydroxylases are not restricted to human-associated Actinobacteria and are instead part of a larger group of bis-MGD enzymes present in diverse bacteria and even Archaea ( Figure 7 ) . These organisms come from mammal-associated , plant-associated , soil , and aquatic habitats . Notable organisms encoding putative dehydroxylases include soil-dwelling Streptomycetes ( Wu et al . , 2017; Huang et al . , 2012 ) , the industrially important anaerobe Clostridium ljungdahlii ( Köpke et al . , 2010 ) , and a large number of anaerobic bacterial genera known for their ability to degrade aromatic compounds , including Azoarcus , Thauera , Desulfobacula , Geobacter , Desulfumonile , and Desulfitobacterium ( Figure 7 ) ( DeWeerd et al . , 1991; Wöhlbrand et al . , 2013; Butler et al . , 2007; Wagner et al . , 2012; Cole et al . , 1995; Fernández et al . , 2014; Molina-Fuentes et al . , 2015; Villemur et al . , 2006; Pacheco-Sánchez et al . , 2019a ) . The presence of similar enzymes in gut and environmental microbes likely reflects the availability of catechol substrates in many different environments ( Figure 7 ) . As the vast majority of dehydroxylase homologs remain uncharacterized , it is difficult to assign the biochemical activities of the major clades and define the characteristic features of these enzymes . However , we are confident that at least some portion of the sequences captured in this analysis are true catechol dehydroxylases . First , we found that representative sequences from across our phylogenetic tree are more closely related to acetylene hydratase and the Gordonibacter and Eggerthella dehydroxylases than to any other member of the bis-MGD enzyme family , indicating shared evolutionary origins ( Supplementary file 3c and Figure 7—figure supplements 3 and 4 ) . Moreover , recent genetic studies have implicated several homologs from environmental bacteria in catechol dehydroxylation . For instance , a putative dehydroxylase is present in Streptomyces biosynthetic gene clusters that produce the potent anti-tumor compounds yatakemycin and CC-1065 ( Wu et al . , 2017; Huang et al . , 2012 ) ( Figure 7 ) . Gene knock-out and complementation studies revealed this enzyme is essential for CC-1065 production and likely catalyzes reductive dehydroxylation of a late-stage biosynthetic intermediate ( Wu et al . , 2017 ) . Another homolog is present in the 3 , 5-dihydroxybenzoate ( 3 , 5-DHB ) degradation operon within the anaerobic soil Proteobacterium Thaeura aromatica ( Figure 7 ) . Strains lacking this enzyme exhibit impaired growth on 3 , 5-DHB as a sole carbon source , suggesting a possible role for this enzyme in metabolizing the one of the two catecholic intermediates involved in this pathway ( Molina-Fuentes et al . , 2015; Pacheco-Sánchez et al . , 2019a; Pacheco-Sánchez et al . , 2019b ) . Based on this analysis , we conclude that the catechol dehydroxylases harbor vast uncharacterized diversity that contributes to both primary and secondary metabolic pathways in habitats beyond the human gut . Our phylogenetic analysis suggested that catechol dehydroxylase activity is present in a range of microbial habitats , making us curious whether we could detect this metabolism in additional microbial communities . As a first step , we explored catechol dehydroxylation by gut microbiotas of non-human mammals . We assembled a panel of gut microbiota samples from 12 different mammals representing diverse phylogenetic origins and diets ( three individuals per mammal ) ( Reese et al . , 2018; Reese et al . , 2019 ) ( Figure 8 and Figure 8—figure supplement 1 ) . We cultured these gut communities anaerobically ex vivo , assessed metabolism using a colorimetric assay , and confirmed potential hits using LC-MS/MS ( Figure 8—figure supplement 1 ) . We observed catechol dehydroxylation across the gut microbiotas of mammals spanning different diets and phylogenies ( Figure 8 ) . Hydrocaffeic acid dehydroxylation occurred in >50% of species , while dopamine and ( + ) -catechin metabolism were observed in 5/12 and 4/12 animals , respectively ( Figure 8 ) . DOPAC was only metabolized by the rat gut microbiota , which was the only community that had activity towards all compounds tested . While a larger sample size is required to reach clear conclusions about possible links between metabolism of specific catechols and individual mammal gut microbiotas , our results clearly demonstrate that catechol dehydroxylation is found in distantly related mammal gut microbiotas that have large differences in species composition and gene content ( Reese et al . , 2018; Reese et al . , 2019; Coelho et al . , 2018 ) . This finding further reinforces the relevance of catechol dehydroxylation to variety of different microbial habitats .
For many decades the human gut microbiota has been known to dehydroxylate catechols , but the molecular basis of this enigmatic transformation has remained largely unknown . In this study , we characterized the specificity and regulation of a gut bacterial enzyme that dehydroxylates dopamine ( Dadh ) . We then used this knowledge to identify candidate enzymes that dehydroxylate additional host-and plant-derived small molecules . Together , the catechol dehydroxylases represent a previously unappreciated group of molybdenum-dependent enzymes that is present in diverse microbial phyla and environments . Our studies of Dadh revealed a high specificity for catecholamines , supporting the hypothesis that the physiological role of this enzyme is to enable neurotransmitter metabolism by E . lenta . This idea is also consistent with recent observations of gut bacteria using specific neurotransmitters for growth ( Strandwitz et al . , 2019 ) . To our knowledge , Dadh is the first catecholamine-metabolizing enzyme from a human gut commensal . However , interactions between catecholamines and intestinal pathogens are well-characterized and have long been known as key players in virulence and infection ( Freestone et al . , 2007; Lyte and Ernst , 1992 ) . Whereas pathogenic organisms such as Escherichia coli , Yersinia enterocolitica , and Salmonella enterica require the intact catechol group of dopamine and norepinephrine to sequester iron and boost growth ( Freestone et al . , 2007; Lyte and Ernst , 1992; Rooks et al . , 2017; Dichtl et al . , 2019 ) , we propose that E . lenta uses these molecules as electron acceptors . Thus , Dadh might represent a novel strategy by which gut bacteria take advantage of catecholamines present in the gastrointestinal tract ( Eisenhofer et al . , 1997; Eisenhofer et al . , 1996 ) . Understanding the interplay between pathogenic and commensal interactions with catecholamines is an intriguing avenue for further research . In addition to characterizing Dadh , we discovered candidate dehydroxylases that metabolize ( + ) -catechin , hydrocaffeic acid , and DOPAC . We also partially purified the hydrocaffeic acid dehydroxylase to confirm its involvement in this reaction . Further biochemical studies are important for validating the activities of the remaining enzymes , but our preliminary data support a working model in which catechol dehydroxylation is performed by distinct enzymes that are specialized for individual substrates . We identified large numbers of uncharacterized dehydroxylases encoded within individual Eggerthella and Gordonibacter genomes ( Figure 7 ) , hinting at an expansion of this group of enzymes among human gut Actinobacteria . While it remains to be seen whether these uncharacterized enzymes are also specific for distinct substrates , this type of diversification of closely related enzymes indicates a potentially important role for catechol dehydroxylation in the human gut microbiota . Expansion of enzyme families within specific clades of gut microbes is well-characterized in the context of polysaccharide metabolism . For example , individual human gut Bacteroides strains isolates harbor hundreds of polysaccharide utilization loci but upregulate only a subset of genes in response to distinct substrates ( Martens et al . , 2011; Rogowski et al . , 2015; Ndeh et al . , 2017; Larsbrink et al . , 2014 ) . This transcriptional regulation and biochemical specificity enables utilization of various host- or plant-derived carbon sources depending on their availability ( Hehemann et al . , 2010; Desai et al . , 2016; Sonnenburg et al . , 2010 ) . The diversity of catechol dehydroxylases might have evolved in a similar manner , providing a biochemical arsenal that enables Actinobacteria to use a range of different electron acceptors whose availability depends on the diet and/or physiology of the host . Identifying the substrates of uncharacterized catechol dehydroxylases could shed light on the adaptation of gut organisms to small molecules produced and ingested by the host . In addition to uncovering the diversity of catechol dehydroxylases , our study illustrates that the chemical strategies used to enable microbial survival and interactions in the human gut may be relevant to a broad range of species and habitats . While mammalian gut microbiomes have previously been compared in terms of gene content and species composition ( Reese et al . , 2019; Coelho et al . , 2018; Youngblut et al . , 2019 ) , our study provides functional evidence for conservation of specific gut microbial metabolic pathways across distinct hosts . While this hints at potentially important roles for catechol dehydroxylation across mammalian gut communities , the distribution of putative dehydroxylases among environmental microbes suggests this chemistry is present in many additional microbial habitats . This reinforces findings from studies of additional gut microbial enzymes . For example , gut microbial carbohydrate-degrading enzymes and glycyl radical enzymes , which play important roles in degrading diet-derived polysaccharides , amino acids , and osmolytes in the human gut , are also found in environmental isolates ( Ndeh et al . , 2017; Levin et al . , 2017; Craciun and Balskus , 2012; Peck et al . , 2019 ) . Enzyme discovery in the human gut microbiota not only has implications for improving human health and disease , but also for discovering novel catalytic functions and metabolic pathways broadly relevant to microbial life . Our study now sets the stage for further investigations of the chemical mechanisms and biological consequences of catechol dehydroxylation in the human body and beyond . More broadly , our study underscores how enzyme discovery can help to dissect the metabolic diversity of gut microbial strains and communities . Although previous studies had linked certain dehydroxylation reactions to individual gut Actinobacteria ( Maini Rekdal et al . , 2019; Takagaki and Nanjo , 2015; Selma et al . , 2014; Bess et al . , 2020 ) , we have found that specific catechol dehydroxylases are variably distributed among closely related strains and human gut metagenomes . These findings reinforce the idea that gut microbial phylogeny is often not predictive of functional capabilities ( Koppel et al . , 2018; Maini Rekdal et al . , 2019; Levin et al . , 2017; Craciun and Balskus , 2012; Peck et al . , 2019; Martínez-del Campo et al . , 2015 ) . Additionally , we noticed that the prevalence of the different dehydroxylation reactions among human and animal gut microbiota samples reflected their distribution among individual Actinobacterial strains , with hydrocaffeic acid metabolism being the most prevalent across all strains and species , and DOPAC dehydroxylation being the least prevalent . This may suggest that the strain-level variability in dehydroxylases is important for metabolism both within humans and other mammalian species . While the evolutionary forces shaping the distribution of specific dehydroxylases within gut bacterial strains and complex gut communities remain unknown , our data provide a starting point for further understanding the effects of catechol dehydroxylation on both gut microbiota and host . Finally , our findings provide a framework for linking metabolic transformations performed by complex gut microbial communities to individual strains , genes , and enzymes . Our broad exploration of a class of metabolic transformations contrasts with the more common focus on metabolism of individual drugs or dietary compounds ( Maini Rekdal et al . , 2019; Koppel et al . , 2018; Levin et al . , 2017; Craciun and Balskus , 2012; Peck et al . , 2019; Martínez-del Campo et al . , 2015; Williams et al . , 2014; Yan et al . , 2018 ) . This functional group-focused approach may greatly increase the efficiency with which we can link metabolic activities to microbial genes and enzymes . We envision that related experimental workflows could find broad utility in the discovery of gut microbial enzymes catalyzing other widespread , biologically significant reactions , including reductive metabolism of additional functional groups that are prevalent in diverse molecules encountered by the gut microbiota ( Koppel et al . , 2017 ) .
The following chemicals were used in this study: tetracycline ( Sigma Aldrich , catalog# 87128–25G ) , p-tyramine ( Sigma Aldrich , catalog# T2879-1G ) , DL-3 , 4-Dihydroxymandelic acid ( Carbo Synth , catalog# FD22118 ) , protocatechuic Acid ( Millipore Sigma , catalog# 37580–25 G-F ) , DL norepinephrine ( Millipore Sigma , catalog# A7256-1G ) , L-norepinephrine ( Matrix Scientific , catalog# 037592–500 MG ) L-epinephrine ( Alfa Aesar , catalog# L04911 . 06 ) , DL-epinephrine ( Sigma Aldrich , catalog# E4642-5G ) , 3 , 4-dihydroxyphenylacetic acid ( Millipore Sigma , catalog# 850217–1G ) , 3 , 4-dihydroxyhydrocinnamic acid ( hydrocaffeic acid ) ( Millipore Sigma , catalog# 102601–10G ) , caffeic acid ( Millipore Sigma , catalog# C0625-2G ) , ( + ) -catechin hydrate ( Millimore Sigma , catalog# C1251-5G ) , ( +/– ) -catechin hydrate ( Millipore Sigma , catalog# C1788-500MG ) , ( – ) -Epicatechin ( Millipore Sigma , catalog# E1753-1G ) , L- ( - ) -a-Methyldopa ( Chemcruz , catalog# sc-203092 ) , 2 , 3-dihydroxybenzoic acid ( Millipore Sigma , catalog# 126209–5G ) , R- ( – ) -apomorphine hydrochloride hemihydrate ( Sigma Aldrich , catalog# A4393-100MG ) , hydroxytyrosol ( Ava Chem Scientific , catalog# 2528 ) , enterobactin ( generous gift from Prof . Elizabeth Nolan , MIT ) , fenoldopam mesylate ( Sigma Aldrich , catalog# SML0198-10MG ) , 5-hydroxydopamine ( Sigma Aldrich , catalog# 151564–100G ) , 6-hydroxydopamine ( Sigma Aldrich , catalog #H4381-100MG ) , 3-methoxytyramine ( Sigma Aldrich , catalog# M4251-100MG ) , 3 , 4-dihydroxybenzylamine ( Sigma Aldrich , catalog# 858781–250 MG ) , N-methyldopamine ( Santa Cruz Biotechnology , catalog# sc-358430A ) , 4- ( 2-aminoethyl ) benzene-1 , 3-diol ( Enamine , catalog # EN300-65185 ) , m-tyramine ( Chemcruz , catalog# sc-255257 ) , 3-hydroxyphenylacetic acid ( Sigma Aldrich , catalog# H49901-5G ) , 3-hydroxyphenylpropionic acid ( Toronto Research Chemicals , catalog# H940090 ) , L-dopa ( Oakwood Chemical , catalog# 358380–25 g ) , dopamine ( Sigma-Aldrich , catalog# PHR1090-1G , or Millipore Sigma , catalog# H8502-25G ) , m-tyramine ( Santa Cruz Biotechnology , catalog# sc-255257 ) , carbidopa ( Sigma-Aldrich , catalog# PHR1655-1G ) , L-arginine ( Sigma-Aldrich , catalog# A5006-100G ) , sodium molybdate ( Sigma-Aldrich , catalog # 243655–100G ) , sodium tungstate ( 72069–25G ) , SIGMAFAST protease inhibitor tablets ( Sigma-Aldrich , catalog#: S8830 ) , benzyl viologen ( Sigma-Aldrich , catalog# 271845–250 mg ) , methyl viologen ( Sigma-Aldrich , catalog# 856177–1 g ) , diquat ( Sigma-Aldrich , catalog# 45422–250 mg ) , sodium dithionite ( Sigma-Aldrich , catalog# 157953–5G ) , 3 , 4-dihydroxyphenylacetic acid ( ring-d3 , 2 , 2-d2 , 98% ) ( Cambridge Isotope Laboratories , catalog #DLM-2499–0 . 01 ) , dopamine HCl ( 1 , 1 , 2 , 2-d4 , 97–98% ) ( Cambridge Isotope Laboratories , catalog #DLM-2498–0 . 1 ) . LC-MS grade acetonitrile and methanol for LC-MS analyses were purchased from Honeywell Burdick and Jackson or Sigma-Aldrich . Brain Heart Infusion ( BHI ) broth was purchased from Beckton Dickinson ( catalog# 211060 ) or from VWR ( catalog# 95021–488 ) . All bacterial culturing work was performed in an anaerobic chamber ( Coy Laboratory Products ) under an atmosphere of 10% hydrogen , 10% carbon dioxide , and nitrogen as the balance , unless otherwise noted . Hungate tubes were used for anaerobic culturing unless otherwise noted ( Chemglass , catalog# CLS-4209–01 ) . All lysate work and biochemical experiments were performed in an anaerobic chamber ( Coy Laboratory Products ) situated in a cold room at 4°C under an atmosphere of 10% hydrogen and nitrogen as the balance . Gut Actinobacterial strains were grown on BHI containing 1% arginine ( w/v ) to obtain isolated colonies for culturing . All genomic DNA ( gDNA ) was extracted from bacterial cultures using the DNeasy UltraClean Microbial Kit ( Qiagen , catalog # 12224–50 ) according to the manufacturer’s protocol . Method A: Samples were analyzed using an Agilent technologies 6410 Triple Quad LC/MS and a Dikma Technologies Inspire Phenyl column ( 4 . 6 × 150 mm , 5 μm; catalog #81801 ) . The flow rate was 0 . 5 mL min−1 using 0 . 1% formic acid in water as mobile phase A and 0 . 1% formic acid in acetonitrile as mobile phase B . The column temperature was maintained at room temperature . The following gradient was applied: 0–2 min: 0% B isocratic , 2–9 min: 0–10% B , 9–11 min: 10–95% B , 11–15 min: 95% B isocratic , 15–18 min: 95–0% B , 18–21 min: 0% B isocratic . For mass spectrometry , the source temperature was 300°C , and the masses of dopamine ( precursor ion m/z = 154 . 3 , daughter ion m/z = 137 . 3 ) , and tyramine ( precursor ion m/z = 138 . 3 , daughter ion m/z = 121 . 3 ) were monitored at a collision energy of 15 mV and fragmentor setting of 135 in positive MRM mode . Method B: Samples were analyzed using an Agilent technologies 6410 Triple Quad LC/MS and a Thermo Scientific Acclaim Polar Advantage II column ( 3 µM , 120A , 2 . 1*150 mm , product #: 063187 ) . The flow rate was 0 . 2 mL min−1 using 0 . 1% formic acid in water as mobile phase A and methanol as mobile phase B . The following gradient was applied: 0–4 min: 50% B isocratic , 4–7 min: 50–99% , 7–9 min: 99–50% , 9–13 min: 50% B isocratic . For mass spectrometry , the source temperature was 300°C , and the masses of trihydroxydopamine ( precursor ion m/z = 170 . 3 , daughter ion m/z = 153 . 3 ) , dopamine ( precursor ion m/z = 154 . 3 , daughter ion m/z = 137 . 3 ) , phenylethylamine ( precursor ion m/z = 122 . 3 , daughter ion m/z = 105 . 2 ) , and tyramine ( precursor ion m/z = 138 . 3 , daughter ion m/z = 121 . 3 ) were monitored at a collision energy of 15 mV and fragmentor setting of 135 in positive MRM mode . Method C: Samples were analyzed using an Agilent technologies 6530 Accurate-Mass Q-TOF LC/MS and a Dikma Technologies Inspire Phenyl column ( 4 . 6 × 150 mm , 5 μm; catalog #81801 ) . The flow rate was 0 . 4 mL min−1 using 0 . 1% formic acid in water as mobile phase A and 0 . 1% formic acid in acetonitrile as mobile phase B . The column temperature was maintained at room temperature . The following gradient was applied: 0–2 min: 5% B isocratic , 2–25 min: 0–95% B , 25–30 min: 95% B isocratic , 30–40 min: 95–5% B . For the MS detection , the ESI mass spectra data were recorded in positive mode for a mass range of m/z 50 to 3000 . A mass window of ±0 . 005 Da was used to extract the ion of [M+H] . Method D: Samples were analyzed using an Agilent technologies 6530 Accurate-Mass Q-TOF LC/MS and a Dikma Technologies Inspire Phenyl column ( 4 . 6 × 150 mm , 5 μm; catalog #81801 ) . The flow rate was 0 . 4 mL min−1 using 0 . 1% formic acid in water as mobile phase A and 0 . 1% formic acid in acetonitrile as mobile phase B . The column temperature was maintained at room temperature . The following gradient was applied: 0–2 min: 5% B isocratic , 2–25 min: 0–95% B , 25–30 min: 95% B isocratic , 30–40 min: 95–5% B . For the MS detection , the ESI mass spectra data were recorded in negative mode for a mass range of m/z 50 to 3000 . A mass window of ±0 . 005 Da was used to extract the ion of [M+H] . Method E: Samples were analyzed using an Agilent technologies 6410 Triple Quad LC/MS and a Thermo Scientific Acclaim Polar Advantage II column ( 3 µM , 120A , 2 . 1*150 mm , product #: 063187 ) . The flow rate was 0 . 2 mL min−1 using 0 . 1% formic acid in water as mobile phase A and methanol as mobile phase B . The following gradient was applied: 0–4 min: 50% B isocratic , 4–7 min: 50–99% , 7–9 min: 99–50% , 9–13 min: 50% B isocratic . For mass spectrometry , the source temperature was 300°C , and the masses of catechin ( precursor ion m/z = 289 . 2 , daughter ion m/z = 109 . 1 ) , benzyl ether reduced catechin ( precursor ion m/z = 291 . 2 , daughter ion m/z = 123 . 1 ) , benzyl ether reduced , dehydroxylated catechin ( precursor ion m/z = 275 . 2 , daughter ion m/z = 107 . 1 ) were monitored at a collision energy of 15 mV and fragmentor setting of 135 in negative MRM mode . Method F: Samples were analyzed using an Agilent technologies 6410 Triple Quad LC/MS and a Thermo Scientific Acclaim Polar Advantage II column ( 3 µM , 120A , 2 . 1*150 mm , product #: 063187 ) . The flow rate was 0 . 2 mL min−1 using 0 . 1% formic acid in water as mobile phase A and methanol as mobile phase B . The following gradient was applied: 0–4 min: 50% B isocratic , 4–7 min: 50–99% , 7–9 min: 99–50% , 9–13 min: 50% B isocratic . For mass spectrometry , the source temperature was 300°C , and the masses of hydrocaffeic acid ( precursor ion m/z = 181 . 2 , daughter ion m/z = 137 . 2 ) , hydroxyphenylpropionic acid ( precursor ion m/z = 165 . 1 , daughter ion m/z = 121 . 2 ) , DOPAC ( precursor ion m/z = 167 . 2 , daughter ion m/z = 123 . 2 ) , and hydroxyphenylacetic acid ( precursor ion m/z = 151 . 3 , daughter ion m/z = 107 . 3 ) were monitored at a collision energy of 15 mV and fragmentor setting of 135 in negative MRM mode . Method G: Samples were analyzed using an Agilent technologies 6410 Triple Quad LC/MS and a Thermo Scientific Acclaim polar advantage II column ( 3 µM , 120A , 2 . 1*150 mm , product #: 063187 ) . The flow rate was 0 . 2 mL min−1 using 0 . 1% formic acid in water as mobile phase A and methanol as mobile phase B . The following gradient was applied: 0–4 min: 50% B isocratic , 4–7 min: 50–99% , 7–9 min: 99–50% , 9–13 min: 50% B isocratic . For mass spectrometry , the source temperature was 275°C , and the masses of norepinephrine ( precursor ion m/z = 170 . 1 , daughter ion m/z = 152 . 1 ) and octopamine ( precursor ion m/z = 154 . 2 , daughter ion m/z = 136 . 1 ) were monitored at a collision energy of 5 mV and fragmentor setting of 135 in positive MRM mode . Method H: Samples were analyzed using an Agilent technologies 6410 Triple Quad LC/MS and a Thermo Scientific Acclaim Polar Advantage II column ( 3 µM , 120A , 2 . 1*150 mm , product #: 063187 ) . The flow rate was 0 . 2 mL min−1 using 0 . 1% formic acid in water as mobile phase A and methanol as mobile phase B . The following gradient was applied: 0–4 min: 50% B isocratic , 4–7 min: 50–99% , 7–9 min: 99–50% , 9–13 min: 50% B isocratic . For mass spectrometry , the source temperature was 275°C , and the masses of caffeic acid ( precursor ion m/z = 179 . 2 , daughter ion m/z = 135 . 2 ) and coumaric acid ( precursor ion m/z = 163 . 3 , daughter ion m/z = 119 . 2 ) were monitored at a collision energy of 5 mV and fragmentor setting of 135 in negative MRM mode . Method I: Samples were analyzed using an Agilent technologies 6410 Triple Quad LC/MS and a Thermo Scientific Acclaim Polar Advantage II column ( 3 µM , 120A , 2 . 1*150 mm , product #: 063187 ) . The flow rate was 0 . 2 mL min−1 using 0 . 1% formic acid in water as mobile phase A and methanol as mobile phase B . The following gradient was applied: 0–4 min: 50% B isocratic , 4–7 min: 50–99% , 7–9 min: 99–50% , 9–13 min: 50% B isocratic . For mass spectrometry , the source temperature was 300°C , and the masses of dihydroxybenzoic acid ( precursor ion m/z = 153 . 1 , daughter ion m/z = 137 . 1 ) and hydroxybenzoic acid ( precursor ion m/z = 137 . 1 , daughter ion m/z = 93 . 2 ) were monitored at a collision energy of 15 mV and fragmentor setting of 135 in negative MRM mode . Method J: Samples were analyzed using an Agilent technologies 6410 Triple Quad LC/MS and a Thermo Scientific Acclaim Polar Advantage II column ( 3 µM , 120A , 2 . 1*150 mm , product #: 063187 ) . The flow rate was 0 . 2 mL min−1 using 0 . 1% formic acid in water as mobile phase A and methanol as mobile phase B . The following gradient was applied: 0–4 min: 50% B isocratic , 4–7 min: 50–99% , 7–9 min: 99–50% , 9–13 min: 50% B isocratic . For mass spectrometry , the source temperature was 300°C , and the masses of norepinephrine ( precursor ion m/z = 184 . 1 , daughter ion m/z = 166 . 1 ) and dehydroxynorepinephrine ( precursor ion m/z = 168 . 1 , daughter ion m/z = 150 . 1 ) were monitored at a collision energy of 15 mV and fragmentor setting of 135 in positive MRM mode . Method K: Samples were analyzed using an Agilent technologies 6410 Triple Quad LC/MS and a Thermo Scientific Acclaim Polar Advantage II column ( 3 µM , 120A , 2 . 1*150 mm , product #: 063187 ) . The flow rate was 0 . 2 mL min−1 using 0 . 1% formic acid in water as mobile phase A and methanol as mobile phase B . The following gradient was applied: 0–4 min: 50% B isocratic , 4–7 min: 50–99% , 7–9 min: 99–50% , 9–13 min: 50% B isocratic . For mass spectrometry , the source temperature was 300°C , and the masses of dihydroxybenzylamine ( precursor ion m/z = 140 . 3 , daughter ion m/z = 123 . 2 ) and hydroxybenzylamine ( precursor ion m/z = 124 . 3 , daughter ion m/z = 107 . 2 ) were monitored at a collision energy of 15 mV and fragmentor setting of 135 in positive MRM mode . Method L: Samples were analyzed using an Agilent technologies 6410 Triple Quad LC/MS and a Thermo Scientific Acclaim Polar Advantage II column ( 3 µM , 120A , 2 . 1*150 mm , product #: 063187 ) . The flow rate was 0 . 2 mL min−1 using 0 . 1% formic acid in water as mobile phase A and methanol as mobile phase B . The following gradient was applied: 0–4 min: 50% B isocratic , 4–7 min: 50–99% , 7–9 min: 99–50% , 9–13 min: 50% B isocratic . For mass spectrometry , the source temperature was 300°C , and the masses of 3-aminotyramine ( precursor ion m/z = 153 . 3 , daughter ion m/z = 136 . 2 ) and 3-aminophenylethylamine ( precursor ion m/z = 137 . 3 , daughter ion m/z = 120 . 2 ) were monitored at a collision energy of 15 mV and fragmentor setting of 135 in positive MRM mode . Method M: Samples were analyzed using an Agilent technologies 6410 Triple Quad LC/MS and a Thermo Scientific Acclaim Polar Advantage II column ( 3 µM , 120A , 2 . 1*150 mm , product #: 063187 ) . The flow rate was 0 . 2 mL min−1 using 0 . 1% formic acid in water as mobile phase A and methanol as mobile phase B . The following gradient was applied: 0–4 min: 50% B isocratic , 4–7 min: 50–99% , 7–9 min: 99–50% , 9–13 min: 50% B isocratic . For mass spectrometry , the source temperature was 300°C , and the masses 3-methoxytyramine ( precursor ion m/z = 151 . 1 , daughter ion m/z = 91 . 1 ) and 3-methoxyphenylethylamine ( precursor ion m/z = 135 . 1 , daughter ion m/z = 75 . 1 ) were monitored at a collision energy of 15 mV and fragmentor setting of 135 in positive MRM mode . Method N: Samples were analyzed using an Agilent technologies 6410 Triple Quad LC/MS and a Thermo Scientific Acclaim Polar Advantage II column ( 3 µM , 120A , 2 . 1*150 mm , product #: 063187 ) . The flow rate was 0 . 2 mL min−1 using 0 . 1% formic acid in water as mobile phase A and methanol as mobile phase B . The following gradient was applied: 0–4 min: 50% B isocratic , 4–7 min: 50–99% , 7–9 min: 99–50% , 9–13 min: 50% B isocratic . For mass spectrometry , the source temperature was 300°C , and the masses 3-hydroxytyrosol ( precursor ion m/z = 153 . 2 , daughter ion m/z = 123 . 1 ) and tyrosol ( precursor ion m/z = 137 . 2 , daughter ion m/z = 107 . 1 ) were monitored at a collision energy of 15 mV and fragmentor setting of 135 in negative MRM mode . Method O: Samples were analyzed using an Agilent technologies 6410 Triple Quad LC/MS and a Thermo Scientific Acclaim Polar Advantage II column ( 3 µM , 120A , 2 . 1*150 mm , product #: 063187 ) . The flow rate was 0 . 15 mL min−1 using 0 . 1% formic acid in water as mobile phase A and methanol as mobile phase B . The following gradient was applied: 0–4 min: 50% B isocratic , 4–7 min: 50–99% , 7–9 min: 99–50% , 9–13 min: 50% B isocratic . For mass spectrometry , the source temperature was 300°C , and the masses of d5-DOPAC ( precursor ion m/z = 172 . 2 , daughter ion m/z = 128 . 2 ) and d5-hydroxyphenylacetic acid ( precursor ion m/z = 156 . 3 , daughter ion m/z = 113 . 3 ) were monitored at a collision energy of 15 mV and fragmentor setting of 135 in negative MRM mode . Method P: Samples were analyzed using an Agilent technologies 6410 Triple Quad LC/MS and a Thermo Scientific Acclaim Polar Advantage II column ( 3 µM , 120A , 2 . 1*150 mm , product #: 063187 ) . The flow rate was 0 . 2 mL min−1 using 0 . 1% formic acid in water as mobile phase A and methanol as mobile phase B . The following gradient was applied: 0–4 min: 50% B isocratic , 4–7 min: 50–99% , 7–9 min: 99–50% , 9–13 min: 50% B isocratic . For mass spectrometry , the source temperature was 300°C , and the masses of d4-dopamine ( precursor ion m/z = 158 . 3 , daughter ion m/z = 141 . 3 ) , and d4-tyramine ( precursor ion m/z = 142 . 3 , daughter ion m/z = 125 . 3 ) were monitored at a collision energy of 15 mV and fragmentor setting of 135 in positive MRM mode . The colorimetric assay for dopamine dehydroxylation was based on the Arnow test ( Arnow , 1937 ) . Briefly , 50 µL of 0 . 5 M aqueous HCl was added to 50 µL of culture supernatant . After mixing , 50 µL of an aqueous solution containing both sodium molybdate and sodium nitrite ( 0 . 1 g/mL each ) was added , which produced a yellow color . Finally , 50 µL of 1 M aqueous NaOH was added followed by pipetting up and down to mix . This allowed the characteristic pink color to develop . Absorbance was measured at 500 nm immediately using a Synergy HTX Multi-Mode Microplate Reader ( BioTek ) or SPECTROstar Nano ( BMG LABTECH ) . Active fractions from the size exclusion chromatography described above were combined and then diluted in 20 mM Tris pH 8 containing 250 mM NaCl to a final enzyme concentration of 0 . 1 µM . The enzyme mixture was transferred the wells of a 96 well plate , for a final volume of 50 µL in each well ( VWR , catalog# 82006–636 ) . 1 µL of substrate ( in water , or 50:50 water:DMF for caffeic acid and catechin substrates ) was then added at a final concentration of 500 µM . Following this , 1 µL of a solution containing electron donors ( final concentration 1 mM each of methyl viologen , 1 mM diquat dibromide , 1 mM benzyl viologen , all dissolved in water ) and 2 µL of sodium dithionite ( 2 mM final concentration , dissolved in water ) were added . The resulting solution was mixed by pipetting and the 96-well plate was then sealed tightly with an aluminum seal . The enzyme assay mixtures were left at room temperature in an anaerobic chamber for 22 hr to allow dehydroxylation to proceed . The enzyme reaction mixtures were quenched by bringing the samples out of the anaerobic chamber and freezing at –20°C . These mixtures were then diluted 1:10 with LC-MS grade methanol and analyzed by LC-MS/MS . For the screen with physiologically relevant catechol substrates , samples containing caffeic acid were analyzed using Method H , hydrocaffeic acid and DOPAC were analyzed using Method F , catechin was analyzed using Method E , protocathecuic acid was analyzed using Method I , epinephrine was analyzed using Method J , norepinephrine was analyzed using Method G , and ellagic acid was analyzed using Method D . For the screen with dopamine analogs , all monohydroxylated , dihydroxylated , and trihydroxylated phenylethylamine analogs were analyzed using method B , N-methyldopamine was analyzed using Method C , methoxytyramine was analyzed using Method M , dihydroxybenzylamine was analyzed using Method K , hydroxytyrosol was analyzed using Method N , and aminotyramine was analyzed using Method L . Cells were cultured in 96-well plates and all experiments were performed anaerobically . The strains screened for dopamine dehydroxylation have been previously described ( Koppel et al . , 2018; Bisanz et al . , 2018 ) . E . lenta A2 was inoculated from a single colony into 10 mL of BHI liquid medium and grown for 48 hr at 37°C to provide turbid starter cultures . These were diluted 1:10 in triplicate into 200 µL of fresh BHI medium containing 500 µM substrate ( p-tyramine , dopamine , 3 , 4-dihydroxybenzylamine , or DL-norepinephrine ) . These cultures were grown for 48 hr at 37°C . Cultures were harvested by centrifugation at 4000 rpm for 10 min , and the supernatants were diluted 1:10 with LC-MS grade methanol . Samples containing dopamine or p-tyramine were analyzed using Method B , norepinephrine was analyzed using Method G , dihydroxybenzylamine was analyzed using Method K . We repeated the setup previously used in the RNA-sequencing experiment with dopamine ( Maini Rekdal et al . , 2019 ) . Turbid 48 hr starter cultures of E . lenta in BHI medium were inoculated 1:100 into 5 mL of BHI medium containing 1% arginine and 10 mM formate , and cultures were grown at 37°C anaerobically . When the cultures reached OD600 = 0 . 200 , hydrocaffeic acid , ( + ) -catechin , p-tyramine , 3 , 4-dihydroxybenzylamine , DL-norepinephrine , or N-methyldopamine were added at final concentrations of 500 µM to triplicate cultures . All compounds except for ( + ) -catechin were dissolved in water; ( + ) -catechin was dissolved in DMF . Control cultures contained vehicle ( water or DMF ) . Cultures were harvested when they reached OD600 = 0 . 500 . They were centrifuged for 15 min at 4000 rpm , and cell pellets were re-suspended in 500 µL Trizol reagent ( ThermoFisher , catalog#: 15596026 ) . Total RNA was isolated by first bead beating to lyse cells and then using the Zymo Research Direct-Zol RNA MiniPrep Plus kit ( Catalog # R2070 ) according to the manufacturer’s protocol . Illumina cDNA libraries were generated using a modified version of the RNAtag-Seq protocol ( Shishkin et al . , 2015 ) . Briefly , 500 ng of total RNA was fragmented , depleted of genomic DNA , and dephosphorylated prior to its ligation to DNA adapters carrying 5’-AN8-3’ barcodes with a 5’ phosphate and a 3’ blocking group . Barcoded RNAs were pooled and depleted of rRNA using the RiboZero rRNA depletion kit ( Epicentre ) . These pools of barcoded RNAs were converted to Illumina cDNA libraries in three main steps: ( i ) reverse transcription of the RNA using a primer designed to the constant region of the barcoded adaptor; ( ii ) addition of a second adapter on the 3’ end of the cDNA during reverse transcription using SmartScribe RT ( Clonetech ) as described ( Shishkin et al . , 2015 ) ; ( iii ) PCR amplification using primers that target the constant regions of the 3’ and 5’ ligated adaptors and contain the full sequence of the Illumina sequencing adaptors . cDNA libraries were sequenced on Illumina HiSeq 2500 . For the analysis of RNAtag-Seq data , reads from each sample in the pool were identified based on their associated barcode using custom scripts , and up to one mismatch in the barcode was allowed with the caveat that it did not enable assignment to more than one barcode . Barcode sequences were removed from the first read as were terminal G’s from the second read that may have been added by SMARTScribe during template switching . Reads were aligned to the Eggerthella lenta A2 genome using BWA ( Li and Durbin , 2009 ) and read counts were assigned to genes and other genomic features using custom scripts . Differential expression analysis was conducted with DESeq2 ( Love et al . , 2014 ) and/or edgeR ( Robinson et al . , 2010 ) . Method 1 ( compound added at mid-exponential phase ) : Turbid 48 hr starter cultures of G . pamelaeae 3C grown in BHI medium were inoculated 1:100 into triplicate Hungate tubes containing 20 mL BHI medium with 10 mM formate . When cultures reached OD600 = 0 . 110 , DOPAC ( 0 . 5 mM final ) or vehicle ( water ) was added to the cultures . The cultures were then grown at 37°C anaerobically and harvested when they reached OD600 = 0 . 185 . They were centrifuged , and cell pellets were re-suspended in 500 µL Trizol reagent ( ThermoFisher , catalog#: 15596026 ) . Method 2 ( compound added at the beginning of growth ) : Turbid 48 hr starter cultures of G . pamelaeae 3C grown in BHI medium were inoculated 1:100 into triplicate hungate tubes containing 20 mL BHI with 10 mM formate and DOPAC ( 0 . 5 mM final ) or vehicle ( water ) . These cultures were then left to grow at 37°C anaerobically . When cultures reached OD600 = 0 . 110 , they were harvested . They were centrifuged , and cell pellets were re-suspended in 500 µL Trizol reagent ( ThermoFisher , catalog#: 15596026 ) . RNA extraction and sequencing: this was performed using the exactly same setup as described above , except the reads were aligned to the genome of Gordonibacter pamelaeae 3C . Cells were cultured in Hungate tubes and all experiments were performed anaerobically . E . lenta A2 was inoculated from a single colony into 10 mL BHI liquid medium and grown for 48 hr at 37°C to provide turbid starter cultures . These were diluted 1:100 in triplicate into 5 mL BHI medium containing either 0 . 5 mM dopamine or vehicle . Growth was assessed by measuring the optical density at 600 nm using a Genesys 20 spectrophotometer ( Thermo Scientific ) . The medium was prepared as described previously , with minor modifications ( Maini Rekdal et al . , 2019 ) . A 100-fold stock solution of salts was first prepared by dissolving 100 g NaCl , 50 g MgCl2•6H2O , 20 g KH2PO4 , 30 g NH4Cl , 30 g KCl , 1 . 5 g CaCl2 × 2H2O in 1 L of water . Then , 10 mL of this solution was added to 1 L of water containing 1 g yeast extract ( Beckton Dickinson #288260 ) , 1 g tryptone ( Beckton Dickinson #21175 ) , and 0 . 25 mL of 0 . 1% resazurin ( dissolved in MilliQ water ) . This medium was autoclaved . Following autoclaving , the medium was left to cool for 15 min in an atmosphere of air ( outside the anaerobic chamber ) . After cooling , the following components were added using sterile technique: 10 mL of ATCC Trace element mix ( ATCC , catalog# MD-TMS ) , 10 mL of Vitamin Supplement ( ATCC , catalog# MD-VS ) , solid NaHCO3 ( SIGMA , 2 . 52 g , to give 30 mM ) and solid L-cysteine HCl ( SIGMA , 63 mg , to give 0 . 4 mM ) . The medium had a final pH of 7 . 2–7 . 3 . The medium was then sparged with nitrogen gas ( for how long ) and was brought into the anaerobic chamber to equilibrate for at least 30 hr prior to use . In all experiments utilizing the basal medium , except for those experiments performed with Gordonibacter pamelaeae 3C or the screen for catechol metabolism by mammalian gut microbiota samples , sodium acetate was added at a final concentration of 10 mM at the time of bacterial inoculation . In experiments performed with Gordonibacter pamelaeae 3C , sodium formate was added at a final concentration of 10 mM . In the ex vivo experiments with the mammalian gut microbiota , neither acetate nor formate were added to the basal medium . Cells were cultured in hungate tubes and all experiments were performed anaerobically . E . lenta strains were inoculated from single colonies into 10 mL of BHI liquid medium and grown for 48–72 hr at 37°C to provide turbid starter cultures . These were diluted 1:100 in triplicate into 5 mL of basal medium containing 10 mM acetate and either 1 mM dopamine ( in water ) or vehicle ( water ) . If applicable . molybdate ( 0 . 5 mM ) , tungstate ( 0 . 5 mM ) , DMSO ( 14 mM ) , or nitrate ( 1 mM ) were added at the time of inoculation . Cultures were grown anaerobically for 36–72 hr at 37°C . Endpoint growth was assessed by measuring the optical density at 600 nm using a Genesys 20 spectrophotometer ( Thermo Scientific ) . Catechol dehydroxylation was assessed at the end of growth in culture supernatants using the colorimetric method . Cells were cultured in hungate tubes and all experiments were performed anaerobically . E . lenta strains W1BHI6 ( Tet resistant non-metabolizer , and Valencia ( Tet sensitive metabolizer ) were inoculated from single colonies into individual tubes containing 10 mL of BHI liquid medium and grown for 48 hr at 37°C to provide turbid starter cultures . For the competition experiment , 50 µL of each starter culture of the two competing strains was combined in triplicate in 5 mL of basal medium containing 10 mM acetate and either 1 mM dopamine or vehicle ( water ) . Following inoculation , cultures were grown anaerobically for 72 hr at 37°C . At the end of the incubation , growth of E . lenta was assessed . Cultures were serially diluted in PBS under anaerobic conditions , and 8 µL of each serial dilution ( 10−1 through 10−7 ) was plated onto BHI plates containing 1% arginine ( w/v ) with and without 10 µg/mL Tetracycline using a spot plating method . Plates were grown at 37°C for 72 hr following by counting of colonies . To calculate the proportion of metabolizer in the W1BHI6/Valencia competition experiment , we selected a dilution where distinct colonies were clearly visible ( 10−4-10−5 ) and counted the number of colonies growing on the BHI 1% arginine Tetracycline plates ( W1BHI6 ) as well as the colonies growing on the BHI 1% arginine plates ( Both Valencia and W1BHI6 ) . To get the number of metabolizer ( Valencia ) colonies , we subtracted the number of Tetraycline resistant colonies from the colonies on the no Tetracycline plate . Cells were cultured in hungate tubes and all experiments were performed anaerobically . E . lenta strains , as well as Enterococcus faecalis OGR1F , Escherichia coli MG1655 , Bacteroides fragilis ATCC 25285 , Clostridium sporogenes ATCC 15579 , Edwarsiella tarda ATCC 23685 , were inoculated from single colonies into individual tubes containing 10 mL of BHI liquid medium and grown for 48–72 hr at 37°C to provide turbid starter cultures . Growth was assessed by measuring the optical density at 600 nm using a Genesys 20 spectrophotometer ( Thermo Scientifc ) . These starter cultures were then diluted to a final OD600 of 0 . 100 in BHI medium anaerobically . The defined community was created by combining equal volumes of all strains ( after diluting each culture to OD600 = 0 . 100 ) except for E . lenta . The community was then inoculated 1:100 in triplicate into 5 mL basal medium containing 10 mM acetate and either 1 mM dopamine or vehicle . E . lenta strains were then added by diluting the E . lenta starter cultures ( normalized to OD600 = 0 . 100 ) 1:50 into the tubes containing the defined community . Cultures were then grown anaerobically for 72 hr at 37°C . At the end of the incubation , growth of E . lenta was assessed . Cultures were serially diluted in PBS under anaerobic conditions , and 8 µL of each serial dilution ( 10−1 through 10−7 ) was plated onto BHI plates containing 1% arginine ( w/v ) and 10 µg/mL Tetracycline ( spot plating method ) . Plates were grown at 37°C for 72 hr following by counting of colonies . The human fecal samples used in this study have been previously described ( Maini Rekdal et al . , 2019 ) . To prepare them for culturing , all samples were resuspended anaerobically in anaerobic PBS at a final concentration of 0 . 1 g/mL . The mixture was vortexed to produce a homogenous slurry and was then left for 30 min to let particulates settle . Aliquots of the supernatant were dissolved 50:50 with 40% glycerol and flash-frozen in liquid nitrogen , creating slurries that were used for anaerobic culturing of human fecal samples . Slurries were stored at –80°C and were thawed anaerobically at room temperature at the time of use . This procedure was performed in an anaerobic chamber ( Coy Laboratory Products , atmospheric conditions: 20% CO2 , 2–2 . 5% H2 , and the balance N2 ) —equilibrating media and consumables to the atmosphere prior to use—until centrifugation , which was performed using a benchtop centrifuge . The 96-well plates used in this experiment were purchased from VWR ( catalog# 10861–562 ) . Into the wells of flat-bottom 96-well plates , 100 μL of BHI medium supplemented with L-cysteine-HCl ( 0 . 05% , w/v ) , L-arginine ( 1% , w/v ) , and sodium formate ( 10 mM ) ( referred to here as BHI++ ) were aliquoted . Seed cultures were prepared by inoculating wells , in triplicate , with Actinobacterial strains that were cultured on BHI++ agar plates . Additional wells served as sterile controls . Plates were sealed with tape and incubated at 37°C for 12 to 18 hr to afford dense cultures . Next , 99 μL of BHI++ medium containing 500 μM of compound were aliquoted into the wells of a 96-well plate . To these wells , 1 μL of dense seed culture ( or sterile control ) was added . Plates were sealed and incubated at 37°C for 24 or 48 hr . Plates were then centrifuged at 2000 rpm for 10 min at 4°C , and the supernatant was aspirated and transferred to a fresh 96-well plate . An aliquot ( 35 μL ) of supernatant was then immediately screened via the catechol colorimetric assay ( described above in ‘Colorimetric assay for catechol detection’ ) . Absorbance was immediately measured at 500 nm using a plate reader ( Spectrostar Nano , BMG LABTECH ) . A standard curve ( 2-fold serial dilutions , 1000–15 . 6 μM in BHI++ ) was simultaneously prepared , developed , and analyzed using the conditions listed above . The catechol concentrations in bacterial cultures were normalized to the sterile control . To confirm metabolism of ( + ) -catechin , DOPAC , and hydrocaffeic acid , the incubations were repeated following the same procedure with minor modifications . Strains were grown in BHI for 48 hr anaerobically at 37°C . Cultures were harvested by centrifugation and were then analyzed by LC-MS . To prepare samples for LC-MS , 20 µL of the culture supernatant was diluted 1:10 with 180 µL of methanol , followed by centrifugation at 4000 rpm for 10 min to pellet particulates , salts , and proteins . 50 µL of the resulting supernatant was then transferred to a 96-well plate and 5 µL of the supernatant was injected onto the instrument using Method E for catechin and Method F for hydrocaffeic acid and DOPAC . Following this screen , select strains were re-cultured to confirm absence/presence of metabolism . The following stock solutions were used in the screens: dihydroxymandelic acid ( 50 mM in water ) , dopamine ( 50 mM in water ) , protocatechuic acid ( 50 mM in ethanol ) , L-dopa ( 50 mM in 0 . 5 M HCl ) , norepinephrine ( 50 mM in 0 . 5 M HCl ) , epinephrine ( 50 mM in 0 . 5 M HCl ) , DOPAC ( 50 mM in 0 . 5 M HCl ) , Hydrocaffeic acid ( 50 mM in ethanol ) , caffeic acid ( 50 mM in ethanol ) , ( + ) -catechin ( 50 mM in ethanol ) , ( +/– ) -catechin ( 50 mM in ethanol ) , ( – ) -epicatechin ( 50 mM in DMSO ) , methyldopa ( solid dissolved directly into the media at 0 . 5 mM final concentration ) , carbidopa ( solid dissolved directly into the media at 0 . 5 mM final concentration ) , dihydroxybenzoic acid ( 50 mM in methanol ) . Hydroxytyrosol ( 50 mM in water ) , enterobactin ( 10 mM in DMSO ) , apomorphine ( 50 mM in DMSO ) . Cells were cultured in hungate tubes and all experiments were performed anaerobically . E . lenta A2 was inoculated from a single colony into 10 mL BHI liquid medium and grown for 48 hr at 37°C to provide turbid starter cultures . These were diluted 1:100 in triplicate into 5 mL of BHI medium containing either 0 . 5 mM hydrocaffeic acid ( in water ) , 0 . 5 mM ( + ) -catechin ( in DMF ) , or vehicle ( water or DMF ) . After 48 hr of anaerobic growth at 37°C , cultures were harvested by centrifugation and were then analyzed by LC-MS . To prepare samples for LC-MS , 20 µL of the culture supernatant was diluted 1:10 with 180 µL of methanol , followed by centrifugation at 4000 rpm for 10 min to pellet particulates , salts , and proteins . 50 µL of the resulting supernatant was then transferred to a 96-well plate and 5 µL of the supernatant was injected onto the instrument using Method E for catechin and Method F for hydrocaffeic acid . Cells were cultured in hungate tubes and all experiments were performed anaerobically . G . pamelaeae 3C was inoculated from a single colony into 10 mL of BHI liquid medium and grown for 48 hr at 37°C to provide turbid starter cultures . These were diluted 1:100 in triplicate into 5 mL of BHI medium containing 10 mM formate and 0 . 5 mM DOPAC or vehicle ( water ) . After 72 hr of anaerobic growth at 37°C , the cultures were harvested by centrifugation and were then analyzed by LC-MS . To prepare samples for LC-MS , 20 µL of the culture supernatant was diluted 1:10 with 180 µL of methanol , followed by centrifugation at 4000 rpm for 10 min to pellet particulates , salts , and proteins . 50 µL of the resulting supernatant was then transferred to the LC-MS 96-well plate and 5 µL of the supernatant was injected onto the instrument using Method F for DOPAC . To characterize the distribution of Cadh , Hcdh , Dodh among our gut Actinobacterial strain library , we performed a tBLASTn search . We queried the genomes for Cadh , Hcdh , Dodh and used 90% coverage , 75% amino acid identity , and e-value = 0 as the cutoff for assessing the presence of each dehydroxylase . Active fractions from the size exclusion chromatography described above were combined and then diluted 1:5 in 20 mM Tris pH 8 containing 250 mM NaCl . The enzyme mixture was transferred the wells of a 96 well plate , for a final volume of 50 µL in each well ( VWR , catalog# 82006–636 ) . 1 µL of substrate ( in water for hydrocaffeic acid , dopamine , and DOPAC , or 50:50 water:DMF for ( + ) -catechin ) was then added at a final concentration of 500 µM . Following this , 1 µL of a solution containing methyl viologen ( 1 mM final concentration ) and 2 µL of sodium dithionite ( 2 mM final concentration , dissolved in water ) were added . The resulting solution was mixed by pipetting and the 96-well plate was then sealed tightly with an aluminum seal . The enzyme assay mixtures were left at room temperature in an anaerobic chamber for 26 hr to allow dehydroxylation to proceed . The enzyme reaction mixtures were quenched by bringing the samples out of the anaerobic chamber and freezing at –20°C . These mixtures were then diluted 1:10 with LC-MS grade methanol and analyzed by LC-MS/MS . Samples containing hydrocaffeic acid and DOPAC were analyzed using Method F , catechin was analyzed using Method E , and dopamine was analyzed using Method C . To generate estimates of the enzyme prevalence in human populations , the amino acid sequences of Cldh , Dodh , Dadh , Hcdh , and Cadh were searched against a non-redundant gut microbiome gene catalogue using BLASTP with a minimum 70% percent identity , query coverage and target coverage and used to extract per-sample gene abundances from a collection of human metagenomes ( 10 . 1093/bioinformatics/btv382 ) . High identity matches were obtained for all queries ( 81 . 7% , 81 . 7% , 100% , 99 . 5% , and 99 . 9% respectively over >99% target coverage ) with a second lower-identity match observed for Hcdh ( 75 . 7% ) for which abundances were summed with the higher identity hit . Next , to account for repeated sampling of individuals , the median gene abundance across a subject’s samples was calculated and carried forward for prevalence estimates leading to a total of 1872 human subjects considered ( Nayfach et al . , 2015 ) . Prevalence was then calculated as a rolling function of minimum abundance . For phylogenetic analysis of the bis-MGD enzymes , we gathered sequences that have been previously used to study the evolution of bis-MGD enzymes ( Schoepp-Cothenet et al . , 2012 ) . However , we also added sequences to capture additional diversity of biochemically characterized bis-MGD enzymes that were not included in the original tree described in Schoepp-Cothenet et al . ( 2012 ) . In particular , we performed a pBLAST search in Uniprot using perchlorate reductase ( Uniprot ID# PCRA_DECAR ) , ethylbenzene dehydrogenase ( Uniprot ID# Q5NZV2_AROAE ) , acetylene hydratase ( Uniprot ID# AHY_PELAE ) , and pyrogallol transhydroxylase ( Uniprot ID# PGTL_PELAC ) as the queries , and collected sequences with 85–90% amino acid ID . In addition , we added the sequences of Dadh , Hcdh , Cadh from E . lenta A2 , and Dodh and Cldh ( Bess et al . , 2020 ) from G . pamelaeae 3C . The sequences were combined with those reported in Schoepp-Cothenet et al . ( 2012 ) and were aligned in Geneious ( version 11 ) using MUSCLE . We subsequently used FastTree ( standard settings , 20 rate categories of sites ) to create a maximum likelihood tree . The tree files were uploaded to the Interactive Tree of Life web server ( https://itol . embl . de/ ) to annotate the trees ( Letunic and Bork , 2016 ) . A SSN was generated using the EFI-EST tool ( http://efi . igb . illinois . edu/efi-est/ ) on July 15 2017 ( Gerlt et al . , 2015 ) . In particular , we generated an SSN of the molybdopterin dinucleotide binding domain enzyme superfamily ( PF01568 ) , including sequences between 600 and 1400 amino acids in length and using an initial alignment score of e-150 . Nodes represented sequences with 75% amino acid identity . The SSN was imported into Cytoscope v 3 . 2 . 1 and visualized with the ‘Organic layout’ setting . The alignment score cutoff was increased to e-167 until the groups of biochemically characterized enzymes included in our phylogenetic analysis ( Figure 6 ) separated from each other into putatively isofunctional clusters . To identify additional diversity beyond the newly identified putative dehydroxylases from this study , we created a database containing putative homologs from a collection of 26 previously sequenced Actinobacterial genomes ( Bisanz et al . , 2018 ) , as well as from genomes publicly available through NCBI . First , the Eggerthella lenta A2 dopamine dehydroxylase ( Dadh ) protein sequence was used as the query sequence for a tBLASTn search of 26 previously sequenced Actinobacterial genomes ( Bisanz et al . , 2018 ) ( April 23 , 2019 ) . The genomes were loaded in Geneious ( version 11 ) and hits with an amino acid ID of >30% and e-value of e-34 were considered potential dehydroxylase hits and were saved . This cutoff was chosen because sequences captured within this window more closely resembled the acetylene hydratase and Dadh than to any other biochemically characterized moco enzyme , as assessed by percent amino acid identity . In addition , we used the representative Gordonibacter enzyme Cldh as a separate query to identify the more distantly related , smaller enzymes from Gordonibacter that were not detected when using the large , multi-subunit Dadh as the query . Specifically , we used tBLASTn to search the 26 Actinobacterial genomes for the Gordonibacter pamelaeae 3C Cldh protein sequence . Hits from Paraeggerthella hongongensis , Gordonibacter pamelaeae 3C and Gordonibacter sp . 28C , the only organisms containing these smaller Cldh-like enzymes in our collection , were saved . Again , amino acid ID of >30% and e-value of e-34 were considered potential hits because sequences captured within this window more closely resembled the acetylene hydratase and Dadh than to any other biochemically characterized moco enzyme , as assessed by percent amino acid identity . The hits from our searches with Cldh and Dadh were combined into a preliminary database in Geneious . To expand the sequence diversity within this database , we used Cldh and Dadh as queries for two separate tBLASTn searches in NCBI ( nucleotide collection ) . To ensure that we captured diversity beyond human gut microbes , we excluded Gordonibacter and Eggerthella as organisms in the tBLASTn searches for Cldh and Dadh queries , respectively . For the two searches , sequences of >29% amino acid ID and e-value of e-55 were considered potential dehydroxylase hits . This was a more conservative cutoff than we used with human Actinobacteria and was selected based on the observation that the pBLAST alignment of Dadh and Cldh has an e value of e-45% and 29% amino acid ID . The sequences retrieved from NCBI were added to the database already containing the hits from searches of the 26 Actinobacterial genomes . In addition , we added the biochemically characterized E . coli bis-MGD enzyme DMSO reductase ( DmsA , Uniprot ID#P18775 ) to this database as the outgroup . This sequence was also used as the root of the tree . For phylogenetic analysis of these sequences , we first aligned sequences in Geneious using MUSCLE and removed sequences that were 95% identical to each other ( considered duplicates ) . After deleting these duplicate sequences , we re-aligned the sequences using MUSCLE ( standard settings ) and subsequently used FastTree ( standard settings , 20 rate categories of sites ) to create a maximum likelihood tree . The tree files were uploaded to the Interactive Tree of Life web server ( https://itol . embl . de/ ) to annotate the trees ( Letunic and Bork , 2016 ) . Once we had constructed the two trees described above ( Figures 6 and 7 in the main text ) and uncovered dehydroxylase homologs in gut and environmental bacteria , we wanted to explore the phylogenetic relationship between these enzymes and the broader bis-molybdopterin guanine dinucleotide ( bis-MGD ) enzyme family . To do this , we added the representative sequences from Figure 7—figure supplement 3 and Supplementary file 3c to the sequence database already described in ‘Phylogenetic analysis of relationship between catechol dehydroxylases and other characterized members of the bis-molybdopterin guanine dinucleotide enzyme family’ . Using MUSCLE , we aligned the newly added sequences with the sequences represented on the tree ( Figure 6 in main text ) . We then used FastTree ( standard settings , 20 rate categories of sites ) in Geneious ( version 11 ) to generate the tree seen in Figure 7—figure supplement 4 . The tree files were uploaded to the Interactive Tree of Life web server ( https://itol . embl . de/ ) to annotate the trees ( Letunic and Bork , 2016 ) . The collection of fecal samples from mammals ( n = 12 different species , n = 3 individuals per species ) has been previously described ( Reese et al . , 2018 ) . To prepare these samples for culturing , all samples were resuspended anaerobically in pre-reduced PBS at a final concentration of 0 . 1 g/mL . The mixture was vortexed to produce a homogenous slurry and was then left for 30 min to let particulates settle . Aliquots of the supernatant were dissolved 50:50 with 40% glycerol in water and flash-frozen in liquid nitrogen , creating slurries . These slurries were stored at –80°C and were defrosted anaerobically at room temperature at the time of use . Mammalian fecal slurries were prepared as described above and were defrosted by incubation at room temperature at the time of use . 20 µL of each slurry was then combined with 980 µL of basal medium containing 500 µM each of dopamine , ( + ) catechin , DOPAC , or hydrocaffeic acid . Each individual sample was grown in one well , with the n = 3 individual samples for each animal serving as the biological replicates . Control wells contained compound but no bacteria . Samples were grown anaerobically at 37°C for 96 hr in a 96-well plate ( Agilent Technology , catalog# A696001000 ) . Following growth , we first assessed the total microbial growth by measuring the OD600 in a plate reader ( BioTek Synergy HTX ) . Cultures were then harvested by centrifugation and 50 µL of supernatant was transferred to a new 96-well plate , at which time the catechol colorimetric assay was used to assess total dehydroxylation by the complex microbial community . Samples that had potential catechol depletion as assessed by the colorimetric assay were then further analyzed by LC-MS . To prepare samples for LC-MS , 20 µL of the culture supernatant was diluted 1:10 with 180 µL of methanol , followed by centrifugation at 4000 rpm for 10 min to pellet particulates , salts , and proteins . 50 µL of the resulting supernatant was then transferred to the LC-MS 96-well plate and 5 µL of the supernatant was injected onto the instrument using Method A for dopamine , Method E for catechin , and Method F for DOPAC and hydrocaffeic acid . The mammalian phylogenetic tree was generated using the Automatic Phylogenetic Tree Generator ( aptg , version 0 . 1 . 0 ) script in R ( version 3 . 5 . 1 ) . Mammals not part of the aptg database were added manually to the tree using additional information about the mammalian phylogeny as a reference ( Reese et al . , 2019 ) . The mammalian icons were adapted under a Creative Commons license ( https://creativecommons . org/licenses/by/3 . 0/ ) at phylopic ( http://phylopic . org ) , including Alpaca logo ( made by Steven Traver ) , Bison ( Lukasiniho ) . Cow ( Steven Traver ) , Dog ( Tracy A Heath ) , Fox ( Anthony Caravaggi ) , Guinea pig ( Zimices ) , Mouse ( Madeleine Price Ball ) , Pig ( Steven Traver ) , Rabbit ( Steven Traver ) , Rabbit ( Steven Traver ) , Rat ( Rebecca Groom ) , Sheep ( Zimices ) , and Wolf ( Tracy A . Heath ) . | Inside the human gut there are trillions of bacteria . These microbes are critical for breaking down and modifying molecules that the body consumes ( such as nutrients and drugs ) and produces ( such as hormones ) . Although metabolizing these molecules is known to impact health and disease , little is known about the specific components , such as the genes and enzymes , involved in these reactions . A prominent microbial reaction in the gut metabolizes molecules by removing a hydroxyl group from an aromatic ring and replacing it with a hydrogen atom . This chemical reaction influences the fate of dietary compounds , clinically used drugs and chemicals which transmit signals between nerves ( neurotransmitters ) . But even though this reaction was discovered over 50 years ago , it remained unknown which microbial enzymes are directly responsible for this metabolism . In 2019 , researchers discovered the human gut bacteria Eggerthella lenta produces an enzyme named Dadh that can remove a hydroxyl group from the neurotransmitter dopamine . Now , Maini Rekdal et al . – including many of the researchers involved in the 2019 study – have used a range of different experiments to further characterize this enzyme and see if it can break down molecules other than dopamine . This revealed that Dadh specifically degrades dopamine , and this process promotes E . lenta growth . Next , Maini Rekdal et al . uncovered a group of enzymes that had similar characteristics to Dadh and could metabolize molecules other than dopamine , including molecules derived from plants and nutrients in food . These Dadh-like enzymes were found not only in the guts of humans , but in other organisms and environments , including the soil , ocean and plants . Plant-derived molecules are associated with human health , and the discovery of the enzymes that break down these products could provide new insights into the health effects of plant-based foods . In addition , the finding that gut bacteria harbor a dopamine metabolizing enzyme has implications for the interaction between the gut microbiome and the nervous system , which has been linked to human health and disease . These newly discovered enzymes are also involved in metabolic reactions outside the human body . Future work investigating the mechanisms and outputs of these reactions could improve current strategies for degrading pollutants and producing medically useful molecules . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2020 | A widely distributed metalloenzyme class enables gut microbial metabolism of host- and diet-derived catechols |
Improving in one aspect of a task can undermine performance in another , but how such opposing demands play out in single cells and impact on fitness is mostly unknown . Here we study budding yeast in dynamic environments of hyperosmotic stress and show how the corresponding signalling network increases cellular survival both by assigning the requirements of high response speed and high response accuracy to two separate input pathways and by having these pathways interact to converge on Hog1 , a p38 MAP kinase . Cells with only the less accurate , reflex-like pathway are fitter in sudden stress , whereas cells with only the slow , more accurate pathway are fitter in increasing but fluctuating stress . Our results demonstrate that cellular signalling is vulnerable to trade-offs in performance , but that these trade-offs can be mitigated by assigning the opposing tasks to different signalling subnetworks . Such division of labour could function broadly within cellular signal transduction .
Cells must adapt to changes in their environment and to do so specialise their response to the nature of the signal being detected . Improving performance in one task , however , often undermines performance in another ( Pareto , 1896 ) . In engineering , for example , it is well-known that fast responses have lower accuracy and that higher amplifications can cause overshooting and unintended oscillations ( Astrom and Murray , 2008 ) . At the cellular level , we expect signal transduction has evolved to reduce such trade-offs because performance in , for example , both speed and accuracy are likely to be under selection ( Shoval et al . , 2012; Lan et al . , 2012; Siggia and Vergassola , 2013 ) . In particular , stress responses can not only be a life-and-death situation where a too slow response is fatal , but also often lead to the consumption of substantial cellular resources so that cells must accurately coordinate their response with the stress ( López-Maury et al . , 2008; Perkins and Swain , 2009 ) . To maintain accuracy , we can think of cells having to continuously match the degree of activation of signalling networks with both their internal state and with the magnitude and type of extracellular signals . By doing so , cells can then correctly ‘interpret’ the environment and launch and modify the appropriate response at the appropriate level . To understand how cells mitigate trade-offs in signalling , we turned to one of the most studied eukaryotic stress responses: hyperosmotic stress in budding yeast . Following an abrupt increase in environmental osmolarity , yeast cells can shrink in seconds ( Hohmann , 2002 ) and must therefore respond quickly . Their response though is metabolically costly , involving the synthesis of the osmoprotectant glycerol , and inaccurate hyperactivation of the signalling network can be highly deleterious ( Tao et al . , 1999; Mitchell et al . , 2015 ) . When the osmolarity of the environment increases , yeast activate a p38 kinase , Hog1 , to launch the stress response . The Hyper-Osmolarity-Glycerol ( HOG ) network has a Y-shaped structure with two input pathways both converging on Pbs2 , a MAP kinase kinase ( Figure 1A ) . One branch of the Y , the Sln1 pathway , uses a two-component phosphorelay , analogous to those in bacteria , to propagate the signal ( Posas et al . , 1996; Posas and Saito , 1998 ) ; the other , the Sho1 branch , uses protein kinases , similar to signalling in higher organisms ( Posas and Saito , 1997; Tatebayashi et al . , 2006 ) . Once activated , Pbs2 in turn activates Hog1 via phosphorylation ( Brewster et al . , 1993 ) . Similarly to MAP kinases in mammalian cells , Hog1 can then translocate into the nucleus . Upon activation , Hog1 causes an increase in the intracellular concentration of glycerol , yeast’s main osmoprotectant , in two ways ( Saito and Posas , 2012 ) : first , through cytosolic changes , such as diverting glycolysis towards synthesizing glycerol and closing channels that export glycerol , and , second , through altering gene expression to increase the numbers of enzymes involved in glycerol synthesis . As levels of intracellular glycerol increase , water returns to the cell , and the cellular volume expands . This increase in volume reduces signalling through the HOG network and the levels of activation of Hog1 . 10 . 7554/eLife . 21415 . 003Figure 1 . The signalling network in budding yeast that responds to hyperosmotic stress has two input pathways , both activating Pbs2 and Hog1 , and its response can be quantified in single cells using the nuclear localisation of Hog1 . ( A ) Two input branches regulate the activity of the Hog1 kinase . The Sln1 ( green ) branch is a bacterial-like phosphorelay . The Sho1 ( red ) branch is a MAP kinase cascade , which is tethered to the membrane by the sensors Sho1 and Msb2 . ( B ) A schematic showing the operation of the ALCATRAS microfluidic device ( Crane et al . , 2014 ) . Single cells are confined between PDMS traps ( blue ) and exposed to changes in osmolarity . To ensure all strains experience identical environments , they are loaded into separate chambers of the same device . ( C ) When exposed to hyperosmotic stress , cellular volume shrinks , Hog1 undergoes nuclear translocation , and cells arrest . Growth typically resumes once the volume has recovered . ( D , E ) A reduction in volume causes Hog1 to translocate within minutes and the recovery of the volume correlates with levels of nuclear Hog1 . The average of the cell population ( n=356 ) and three single-cell traces selected at random are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 003 Despite these discoveries , the advantage of having two input pathways in the HOG network is still not understood ( Tanaka et al . , 2014; Brewster and Gustin , 2014 ) . The two branches of the Y may sense stress differently ( Reiser et al . , 2003; Tanaka et al . , 2014 ) and are known to operate at different time-scales ( Maeda et al . , 1995; Hersen et al . , 2008 ) : the Sln1 branch being faster than the Sho1 branch . These different response times imply that each pathway has the potential to respond distinctly to input signals ( Behar et al . , 2007 ) and hence generate distinct dynamics of volume recovery . Mutants that have only one branch of the Y have been created ( Maeda et al . , 1995; Reiser et al . , 2003; Hersen et al . , 2008; Macia et al . , 2009; Schaber et al . , 2012; English et al . , 2015 ) , but there is no reported phenotype for strains having only the Sln1 branch , and the Sho1 pathway is often considered redundant ( Klipp et al . , 2005; Muzzey et al . , 2009 ) . We hypothesized that the Y-shaped structure could allow the cell to respond to stress with both speed and accuracy . Our approach was to characterize the behaviour of both Hog1 and cellular volume at the single-cell level in the wild-type and in mutants with only one of the input pathways . With both types of single-cell measurements , we can quantify accuracy by the statistical dependency between the dynamics of cellular volume and the dynamics of nuclear Hog1 . We show that each input pathway specializes to a particular task and that by having the two pathways the wild-type is both fast and accurate over a wide range of dynamic environments .
In steps of hyperosmotic stress , by far the most common type of input so far investigated ( Saito and Posas , 2012 ) , the fast ( Sln1 only ) mutant has been reported to perform almost identically to wild-type . We first verified that the two mutants , each with one of the branches of the Y , behave as expected ( Maeda et al . , 1995; Hersen et al . , 2008; Macia et al . , 2009 ) . Indeed , in steps , the mean response of Hog1 in the fast mutants is equivalent to wild-type cells , but the slow mutant has typically longer response times and a lower maximum level of nuclear localization ( Figure 2A ) . 10 . 7554/eLife . 21415 . 004Figure 2 . For hyperosmotic stress , accuracy can be quantified as the statistical dependency between the dynamics of Hog1 and the dynamics of volume recovery . ( A ) Characterization of the wild-type ( WT ) and mutant strains in response to a 1 M sorbitol step . Colours here and in all following figures: blue ( WT ) ; green ( fast mutant—ssk1Δ ) ; red ( slow mutant—ste11Δ ) . Mean responses are shown and error bars are SEM . See also Video 1 . ( B ) Normalized response from wild-type cells to illustrate the degree of matching between the time of adaptation of Hog1 ( the time for nuclear Hog1 to undergo a 85% decrease from its maximum ) and the time of volume recovery ( the time for the volume to undergo a 85% increase from its minimum ) . ( C ) Accuracy is the correlation between the adaptation times and is lowest for the fast mutant in late stages of the volume recovery ( data from six experiments with at least 500 cells per strain; Figure 2—figure supplement 1 ) . Error bars are 95% confidence intervals for the mean calculated by bootstrapping . ( D ) Adaptation of Hog1 in single cells becomes less sensitive to the magnitude of the stress in the fast mutant . The mutual information between the distributions of adaptation times of Hog1 and the magnitude of the steps from four experiments shows that the fast mutant becomes the least informative late in adaptation explaining the drop in correlation in C . Error bars are 95% credible intervals for the mean calculated by bootstrapping . Differences between strains are therefore at a 5% significance level when the error bars do not overlap . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 00410 . 7554/eLife . 21415 . 005Figure 2—figure supplement 1 . The Hog1 and volume response for wild-type and mutants in steps . ( A ) Step data used in Figure 2C with the sorbitol concentrations given above each panel . Average and SEM error bars are shown . Numbers of cells are listed in order of wild-type , ste11∆ , ssk1∆ for each experiment ( n = 78 , 112 , 94 for 0 . 2 M; n = 116 , 140 , 87 for 0 . 4 M; n = 105 , 123 , 113 for 0 . 6 M ; n = 82 , 81 , 87 for 0 . 8 M; n = 192 , 148 , 125 for 1 . 0 M; n = 133 , 89 , 94 for 1 . 2 M ) . ( B ) Average volume traces for the experiments shown in A . Error bars are SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 00510 . 7554/eLife . 21415 . 006Figure 2—figure supplement 2 . Distributions of the adaptation time of Hog1 for different step inputs . Single-cell distributions of the adaptation time of Hog1 ( time to adapt to 85% of the maximum value ) for step inputs of 0 . 6 , 0 . 8 , 1 and 1 . 2 M sorbitol . Adaptation times were found from the experiments of Figure 2—figure supplement 1 and the distributions used to calculate the mutual information in Figure 2D . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 00610 . 7554/eLife . 21415 . 007Video 1 . Nuclear localization of Hog1-GFP in a step of 1M sorbitol for the wild-type and two mutants ( related to Figure 2 ) . Overlay of DIC and fluorescence microscopy images showing cells trapped between two pillars in the ALCATRAS microfluidics device . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 007 Considering the reduction in cellular volume , the wild-type and fast response are again almost identical on average , and the longer response time of the slow mutant is reflected in a slower volume recovery ( Figure 2A ) , particularly in larger steps ( Figure 2—figure supplement 1 ) . In all strains , the mean volume and the mean level of nuclear Hog1 simultaneously go to zero ( Figure 2A ) . At the single-cell level , however , this picture changes ( Figure 2B ) . We quantified the degree to which the cellular response matches the cellular volume—the accuracy of the response—by the statistical dependency ( the Pearson correlation ) between the time of adaptation of Hog1 and the time of adaptation of the volume . Cells differ in their internal states , for example in their intracellular levels of glycerol , and so the recovery of the volume reports the extent of the subjective stress experienced by each cell . To normalize between cells and for the magnitude of the step , the correlation is calculated for different fractions of the recovery of the volume ( Figure 2C ) . Although the accuracy of all strains increases as the volume recovers , it is the wild-type and the slow mutant that behave similarly , and the correlation for the fast mutant remains consistently lower than the wild-type . This discrepancy between the mean ( Figure 2A ) and the single-cell results ( Figure 2C ) implies that the Hog1 behaviour in the fast mutant during adaptation is more noisy than the wild-type . The variation can be quantified by the statistical dependency ( the mutual information ) between the adaptation times of Hog1 in single cells and the magnitude of the stress ( Figure 2D ) . A higher mutual information implies that there is less overlap between the distributions of adaptation times for each stress ( Figure 2—figure supplement 2 ) and therefore that the adaptation time of a typical Hog1 response is different for different levels of stress . As Hog1 adapts , Hog1 in the fast mutant becomes less informative on the level of stress and its distribution of adaptation times is broader than the wild-type for some stresses . The two pathways therefore have contrasting behaviors: the slow pathway has a slower mean response of Hog1 but is almost as accurate as the wild-type at long times , and the fast pathway although responding the same as the wild-type on average is inaccurate at the single-cell level . Our results suggest that maintaining accuracy is principally addressed by the slow pathway , which best correlates the dynamics of Hog1 with the dynamics of the cellular volume in individual cells . The adaptation of Hog1 and the adaptation of the volume are connected by negative feedback ( Muzzey et al . , 2009 ) . This feedback acts through intracellular glycerol . Higher intracellular concentration of glycerol cause water to move into the cell and the resulting increase in volume reduces the level of activation of the HOG network . The rate of increase in glycerol is expected to depend on the time-integral of nuclear Hog1 ( Muzzey et al . , 2009; English et al . , 2015 ) , and the feedback is therefore called integral feedback ( Astrom and Murray , 2008 ) . We reasoned that if the slow mutant is able to gradually increase its accuracy ( Figure 2C ) then the slow pathway should be more sensitive to the integral feedback . Indeed , the slow pathway unlike the fast pathway is known to have multiple types of osmo-sensors ( Tanaka et al . , 2014 ) and so may better sense the increase in volume resulting from the increase in glycerol . To determine the sensitivity of each pathway to the feedback , we exogenously perturbed the level of activation of the network to measure the extent to which each pathway can compensate for the perturbation . If Hog1 activity in the nucleus is reduced and the network is sensitive to the integral feedback , the system will compensate by increasing the time spent by Hog1 in the nucleus ( Figure 3A ) ( Mettetal et al . , 2008; Zi et al . , 2010 ) . We therefore expect the wild-type and the slow mutant , but not the fast , to maintain accuracy in compromised networks . 10 . 7554/eLife . 21415 . 008Figure 3 . The slow pathway specializes in matching Hog1 dynamics to volume recovery by being most sensitive to the network’s integral feedback . ( A ) Before reaching Hog1 , the signals from each pathway are transduced through Pbs2 , and we perturb the HOG network by controlling PBS2 expression through a TET inducible promoter . Reduced induction of PBS2 compromises the network and decreases the maximum activity of Hog1 . This reduction in activity should be compensated by the system’s integral feedback lengthening the nuclear residence of Hog1 ( box inset ) . ( B ) Measuring relative to unperturbed Pbs2 , under-expression of Pbs2 reduces the amplitude of mean Hog1 nuclear localization , but increases its adaptation time on average in wild-type cells ( 1 M step; arrows indicate time for 85% adaptation ) . ( C , D ) Mean Hog1 dynamics for the slow and fast mutants show that only the slow mutant extends the adaptation time of Hog1 like the WT . Insets: Mean ratio for three experiments of the adaptation time of Hog1 to the adaptation time of the volume in single cells . Error bars are SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 00810 . 7554/eLife . 21415 . 009Figure 3—figure supplement 1 . Only the wild-type and slow mutant can compensate when Pbs2 levels are reduced by increasing the adaptation time of Hog1 . ( A ) Average volume as a function of time in the three strains for different levels of Pbs2 repression . The time to recover the volume is extended in all strains for reduced levels of Pbs2 . ( B ) Distributions of the ratio between the adaptation time of Hog1 and the adaptation time of the volume in single cells for different levels of Pbs2 . Each distribution comprises at least 200 cells from three independent experiments ( 27 experiments in total ) . The median is indicated by a coloured line on the x-axis . For the fast mutant , reduced levels of Pbs2 result in premature adaptation of Hog1 relative to the volume recovery ( p<10−6: two-sided Wilcoxon rank sum test for equal medians and indicated by asterisks ) . For wild-type and the slow mutant , the ratio of the time of adaptation of Hog1 adaptation to the time of adaptation of the volume is not significantly affected by reduced Pbs2 levels ( p>0 . 1 ) . ( C ) The correlation between adaptation time of Hog1 and the time for volume recovery as a percentage of volume recovery calculated by pooling together the data from B ( cf . Figure 2C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 009 By exogenously decreasing levels of the MAP kinase kinase Pbs2 , which lies downstream of both pathways ( Figure 3A ) , we perturbed the network’s activity and showed that the slow pathway is most sensitive to the integral feedback . Decreasing expression of PBS2 reduces the maximum level of Hog1 localization but increases its adaptation time in both the wild-type and slow mutant ( Figure 3B and C ) . In contrast , for the fast pathway , there is little increase in the adaptation time ( Figure 3D ) . Similarly , there is a corresponding change in the speed of the response of the slow , but not the fast , pathway indicating that the slow pathway is better coupled to the dynamics of the integral feedback . Both the ratio of adaptation time of Hog1 to the adaptation time of the volume and the accuracy decreases significantly only for the fast mutant ( Figure 3D inset with p<10−6 using a two-sided Wilcoxon rank sum test for equal medians calculated by pooling distributions from single cells ) . To reduce metabolic costs , cells should respond accurately to stress , matching the dynamics of their response with the dynamics of the stress and of their internal states . Our results are consistent with this task being principally performed by the slow pathway because of its stronger coupling to the system’s integral feedback . The insensitivity of the adaptation time of Hog1 and the sensitivity of the adaptation time of the volume for the fast mutant in these experiments ( Figure 3D; Figure 3—figure supplement 1 ) suggest the existence of a mechanism within the fast pathway that decreases signalling of Hog1 before the volume completely recovers . This additional adaptation within the pathway to a step of stress implies that the fast pathway has the potential to respond to the time-derivative of its input ( Block et al . , 1983; Behar et al . , 2007; Alon , 2007 ) . A derivative response , referred to as derivative action , is often used in engineering to improve performance by predicting the future behaviour of the input ( Astrom and Murray , 2008 ) . A system that only senses the time-derivative of an input should continually respond to an input that ramps linearly from low to high values because such an input has a constant time-derivative ( Block et al . , 1983 ) . We therefore exposed cells to inputs where stress increases gradually ( Figure 4A ) . 10 . 7554/eLife . 21415 . 010Figure 4 . A component of the fast pathway that responds to the time-derivative of the input enables its high speed . ( A ) The Hog1 trajectory in the fast mutant overshoots the Hog1 trajectories of both the wild-type and the slow mutant in ramp inputs ( two examples with different slopes of approximately 0 . 03 M min-1 and 0 . 06 M min-1 ) . The mean response is shown and error bars are SEM . ( B ) Distributions of response times relative to the wild-type for six different ramps ( Figure 4—figure supplement 1 ) shows that the fast mutant is even quicker than the wild-type on average ( p-value <10−6 using a t-test for distributions with at least 600 cells per strain ) . ( C ) The average amplitude of the Hog1 response for the fast mutant consistently overshoots the wild-type for ramp inputs , which responds linearly to the slope of the ramp . ( D ) An input with a fluctuating time-derivative shows the average Hog1 response of the fast mutant consistently over-shooting the wild-type . Errors are SEM . ( E ) The average of the single-cell cross-correlations of the trajectories of Hog1 with the trajectory of the ( smoothed ) time-derivative of the input shows that the high correlation of the wild-type comes from the fast and not the slow pathway ( average of three independent experiments with fluctuating ramps and error bars as SD; p-value <10−6 using a t-test on pooled single-cell data from the three experiments ) . ( F ) The mutual information between the time-derivative of the input in D and the level of Hog1 at each time point shows that the fast mutant best predicts the time-derivative ( at 5% significance level calculated using credible intervals of the median ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 01010 . 7554/eLife . 21415 . 011Figure 4—figure supplement 1 . The Hog1 response of the fast mutant in ramps of stress indicates derivative action . ( A ) Ramp data from Figure 2D showing the consistent overshoot of the wild-type Hog1 by the fast mutant . The sorbitol concentration was calculated from the fluorescent signal of the cy5 dye ( black dotted lines and right y-axis ) and a linear approximation is shown by the orange lines . Numbers of cells are listed in order of wild-type , ste11∆ , ssk1∆ for each experiment ( n = 201 , 195 , 192 for 0 . 026 M/min; n = 112 , 198 , 97 for 0 . 03 M/min; n = 193 , 226 , 200 for 0 . 035 M/min; n = 175 , 164 , 144 for 0 . 041 M/min; n = 134 , 86 , 60 for 0 . 055 M/min; n = 148 , 187 , 186 for 0 . 071 M/min ) . ( B ) The cross-correlation between the single-cell trajectories of Hog1 and the time-derivative of the input in the fluctuating ramp of Figure 4D . The derivative was smoothed using a first order filter and the correlation is plotted as a function of the smoothing parameter α . Cross-correlation for three experiments were calculated in total and the average is shown in Figure 4E . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 011 In contrast to steps , ramp inputs reveal a substantial phenotypic difference between the fast mutant and the wild-type . We observe a systematic increase in the amplitude of Hog1 in the fast mutant over that of the wild-type and the mutant responds quicker ( Figure 4A and B; Figure 4—figure supplement 1 ) . Indeed , the mean wild-type response is closer in amplitude to the slow mutant , particularly during adaptation ( Figure 4A ) . The overshoot of the fast mutant ( Figure 4C ) is consistent with stronger derivative action in the fast pathway . Nevertheless , Hog1 in the fast mutant adapts despite the ramp in stress , and so the fast pathway must not only have derivative action but also respond to other aspects of the input . Indeed , linearizing a mathematical model of adaptation with negative feedback ( Behar et al . , 2007 ) shows that the network performs a time-derivative of the input in parallel with a proportional response . We note that the wild-type response is generated by interactions between the two pathways and is not an average of their responses ( Figure 4A ) : the maximum amplitude of the wild-type strain increases linearly with the gradient of the ramp although the response of strains with each individual pathway does not ( Figure 4C ) . To test further the existence of derivative action in the fast pathway , we exposed cells to fluctuating ramps , which have varying time-derivatives ( Figure 4D ) , and determined if the response of Hog1 in each strain best correlated with either the input or the time-derivative of the input . As expected , both the fast mutant and the wild-type have the highest statistical dependency with the ( smoothed ) time-derivative ( Figure 4E and F; Figure 4—figure supplement 1 ) . The cellular response to osmotic stress should be sufficiently fast to enable survival . Our results are consistent with this task being addressed principally by the fast pathway , which initiates a ‘knee-jerk’ , reflex-like response , partly through derivative action , that can overshoot and be too fast in comparison to the wild-type in ramps of stress . Together our results indicate contrasting roles for the two input pathways: the slow pathway provides accuracy at the expense of speed ( Figure 3 ) ; the fast pathway provides speed at the expense of accuracy ( Figure 4 ) . Further , the response to ramps of stress implies the outputs of the two pathways do not always sum to give the wild-type response ( Figure 4A ) . Building on previous work ( Mettetal et al . , 2008; Muzzey et al . , 2009 ) , we developed a modular mathematical model of the network with the aim of highlighting general principles governing how cells might balance two opposing tasks . Control theory is a natural framework to describe the modulation of cellular responses ( Yi et al . , 2000; El-Samad et al . , 2005 ) , and correspondingly we present the model as a block diagram ( Figure 5A; Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 21415 . 012Figure 5 . A mathematical model with interactions between the pathways can describe the behaviours of the wild-type and mutants . ( A ) A block diagram of a modular model of the HOG network . The slow pathway responds to the error , the difference between the intracellular and extracellular osmolarity; the fast pathway responds to both the error and the time-derivative of the input u ( the extracellular osmolarity ) . These pathways mutually inhibit each other and then activate Hog1 . The rate of change of glycerol is determined by both the time-integral of Hog1 and by the level of activation of a Hog1-independent pathway that responds proportionally to the error . The accumulation of glycerol determines the intracellular osmolarity and the network’s negative feedback . ( B , C ) Predictions of the wild-type and the two mutants in steps ( 0 . 6 M ) and ramps ( 0 . 03 M min−1 ) of stress . The inset shows the contributions of the fast and slow pathways ( dotted ) to the wild-type response . Mutations remove cross-inhibition between the pathways causing the behaviour of the mutants to be different from the behaviour of the corresponding pathway in the wild-type . ( D , E ) Predictions of glycerol show that all strains initially over- or under-shoot the long-term behaviour ( dotted black line ) . The fast mutants overshoots in both cases . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 01210 . 7554/eLife . 21415 . 013Figure 5—source data 1 . Parameters for the mathematical model of the HOG network . Values for the zero-order gain and time-constants of the different components of the block diagram shown in Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 01310 . 7554/eLife . 21415 . 014Figure 5—figure supplement 1 . Modular model of the HOG network including derivative action in the fast pathway and the interactions between the fast and slow pathways . The slow pathway is modelled as a first-order filter of the error . The fast pathway also has a first-order filter of the error and derivative action: an amplified and filtered time-derivative of the input . Cross-inhibition between the two pathways consists of negative feedback of the filtered output of the slow pathway on the fast pathway and negative feedback of the filtered derivative action of the fast pathway on the slow pathway . Hog1 is the sum of the outputs of both pathways after the cross-inhibition and feeds into an integrator that is amplified and filtered before feeding into glycerol . The Hog1-independent pathway responds proportional to the error and directly feeds into glycerol . Glycerol negatively feeds back on the input to give the error . All parameter values are Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 014 We modelled the Hog1 response as the output of the cell’s ’controller’ and the production of glycerol as the process being controlled . The integral feedback , which exists because glycerol affects the volume , makes the system a closed loop , and the network as whole acts to reduce the ‘error’—the difference between the intracellular osmolarity ( predominately determined by glycerol ) and the extracellular osmolarity ( the stress or input ) ( Klipp et al . , 2005; Muzzey et al . , 2009 ) . The slow pathway passes the error through a low-pass filter ( the time-scale of this filter is determined by the adaptation time of the pathway ) and then responds proportionally to the filtered error . Low-pass filtering in this pathway has been observed previously ( Hersen et al . , 2008 ) . In contrast , the activation of the fast pathway has two sources: it both responds to the error and to the time-derivative of the input . The error passes through a low-pass filter that has a higher cut-off frequency than the filter for the slow pathway ( Hersen et al . , 2008 ) . The overall ( zero-frequency ) gain of the pathway is higher than the gain of the slow pathway , and the response of the pathway is therefore faster . To generate the wild-type behaviour , the two input pathways inhibit each other ( Figure 5A ) , but the model is agnostic to the biochemical details of this inhibition . A possible mechanism is competition between the pathways for Pbs2 ( both pathways activate Pbs2 via phosphorylation of the same two residues [Hohmann , 2002] ) . For example , each pathway may be able to sequester Pbs2 from the other . This sequestering could potentially arise either from the different spatial locations of the receptors—Sln1 is observed throughout the plasma membrane , but Sho1 localizes to sites of polarized growth ( Raitt et al . , 2000; Reiser et al . , 2000 ) —or from different allosteric states of Pbs2 ( Monod et al . , 1965 ) . Indeed , from our data ( Figure 3B ) , the amount of Pbs2 is limiting because the response of Hog1 changes if the levels of Pbs2 are reduced . Nevertheless , the inhibition could also be indirect: for example , the slow pathway could positively feedback on the fast pathway , which in turn inhibits the slow pathway . The output of this biochemical controller is Hog1 , whose activation is determined by the cross-inhibition between the input pathways . Hog1 feeds into an integrator , which determines the levels of glycerol and gives the system integral feedback ( Muzzey et al . , 2009 ) . This integrator could be a long-lived gene product whose transcription is activated by Hog1 and whose levels are therefore proportional to the total amount of time that Hog1 spends in the nucleus . Finally , Hog1-independent mechanisms are know to regulate the early accumulation of glycerol ( Brewster et al . , 1993 ) . For example , the Fps1 channels , which export glycerol through the plasma membrane , are not only controlled by Hog1 , but also have additional regulation ( Luyten et al . , 1995; Ahmadpour et al . , 2016 ) . Their fast closure gives an initial boost of glycerol ( Petelenz-Kurdziel et al . , 2013 ) , which is important in all three mutants because the glycerol produced from nuclear Hog1 accumulates relatively slowly , over the time-scale of the integrator . We include such mechanisms as an additional input pathway that responds proportionally to the error and directly controls glycerol ( Muzzey et al . , 2009 ) . We note that if the error becomes negative , this pathway reduces intracellular glycerol and therefore partly describes the effects of open Fps1 channels . The model captures the differences between the mutants and the wild-type we observe in both steps ( Figure 5B ) and ramps ( Figure 5C ) and provides insight into how two opposing tasks can be implemented in the network by having specialized subnetworks . Analogous specialization is believed to occur , for example , in the establishment of polarity in yeast where the speeds of activating and de-activating of the relevant signalling network are made distinct by having two separate positive feedbacks ( Brandman et al . , 2005 ) . Using the model , we can understand the architecture of the HOG network as a means to provide both speed and accuracy . A fast response requires a high gain , but increasing gain typically comes with a reduction in structural stability ( Astrom and Murray , 2008 ) . Within the model ( Figure 5D and E ) , this instability can manifest as the levels of glycerol overshooting and potentially oscillating ( Schaber et al . , 2012 ) . Uncontrolled production of glycerol decreases the accuracy of the response by causing a mismatch between the dynamics of Hog1 and the dynamics of the volume . Derivative action is a well-known way to increase gain while maintaining structural stability ( Astrom and Murray , 2008 ) , and derivative action in the fast pathway enables that pathway’s high gain . The derivative action is open loop , responding to the input not the error , and therefore reduces the coupling of the fast pathway to the integral feedback , undermining the network’s accuracy . The intrinsic time-scale of the derivative action in the fast pathway is highlighted in the Pbs2 mutants where a reduction in Pbs2 decreases the gain of only those elements of the fast pathway that are sensitive to the error . Consequently , the derivative action principally determines the adaptation time of Hog1 ( Figure 3D ) . A further signature of the intrinsic time-scale is also present in the non-monotonic character of the mutual information between the adaptation time of Hog1 and the level of stress in both the fast mutant and the wild-type ( Figure 2D ) . This behaviour implies that the derivative action is typically strong enough to dominate the Hog1 response in steps and causes cells to adapt appropriately to the level of shock , at least early in the response . The presence of the slow pathway allows the network to maintain accuracy despite the open-loop component of the fast pathway . The slow pathway has a lower gain and is therefore slower but more stable ( Figure 3B and C ) . Further , it only responds to the error and can compensate for inaccuracies generated by the fast pathway because it is more sensitive to the integral feedback . In steps , the wild-type and the fast mutant are initially driven by the derivative action and the quicker response of the fast pathway allows inhibition of the slow pathway . Consequently , the fast pathway dominates the wild-type response ( Figure 5B ) . At later times when the response from the derivative action falls , the differences between the accuracy of the wild-type , which benefits from the additional stabilising effects of the slow pathway , and the fast mutant become apparent . In ramps , the activation of the fast pathway by the derivative of the input is again important giving the overshoot observed in the fast mutant , but the maximum of its contribution is smaller and the initial inhibition of the slow pathway is lower ( wild-type Hog1 begins as the average of the two mutants ) . Once the pathways activate further , they both inhibit each other and both control the response and can reduce the levels of Hog1 in the wild-type below either mutant ( Figure 5C ) . The two mutants perform better at different tasks , responding at different speeds and with different levels of accuracy , and if these tasks are important for the cell we expect that the mutations may come with a fitness cost , although potentially only in particular environments . We therefore measured cell viability for stress with three different types of dynamics: steps , linear ramps , and fluctuating ramps . Bulk fitness has been previously measured in the two mutants , and growth deficiencies for the slow mutant have been observed at high osmolarity ( above approximately 0 . 3 M salt ) ( Macia et al . , 2009 ) . Such bulk measurements , based on monitoring optical density in liquid media , make varying the environmental dynamics challenging and miss single-cell events . Using ALCATRAS , we measure the number of cells that either die in the stress or never restart growth over five hours once the environment has stabilized . This direct measure allows the performance of individual cells to be evaluated ( Figure 6A ) . All strains grow similarly in rich media and any change in fitness is therefore a consequence of the osmotic stress . 10 . 7554/eLife . 21415 . 015Figure 6 . Each input pathway increases survival in specific environments , but the wild-type is the most fit in all environments being both fast and accurate . A We measure failure to grow ( inset ) by the number of cells that either did not resume the cell cycle ( arrested ) or die over the 5 hr after the stress has stabilized for three different environments: a 1M step , a 0 . 03 M min-1 linear ramp ( of 40 min length ) , and 2 hr of a fluctuating ramp . Mean and SD of 2 experiments each comprising at least 300 cells per strain . The asterisks denote significance with a p-value <10−6 calculated using a t-test and bootstrapping . See also Videos 2 and 3 . B Together our results imply that a network with only one input pathway is subject to a speed-accuracy trade-off , which the two mutants satisfy in contrasting ways and we illustrate by showing the mutants lying on the extremes of a hypothetical Pareto front ( dotted line ) . The wild-type by having two interacting pathways , each specializing to one aspect of the trade-off , escapes this constraint . Mean and 95% confidence intervals for response time and accuracy using data from Figure 2 . . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 015 Quantifying the number of unfit cells , we find that the selective advantage of having both input pathways is revealed by exposing cells to environments with a range of different dynamics: the fast mutant outperforms the slow mutant in steps of stress and the slow mutant outperforms the fast mutant in ramps of fluctuating stress ( Figure 6A; Videos 2 and 3 ) . 10 . 7554/eLife . 21415 . 016Video 2 . Survival of wild-type and mutant cells following a step of 1 M sorbitol ( related to Figure 6 ) . A representative field of view ( DIC channel ) showing cells trapped in the ALCATRAS microfluidics device . White arrows indicate cells that are either arrested or dead . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 01610 . 7554/eLife . 21415 . 017Video 3 . Survival of wild-type and mutant cells following a fluctuating ramp of sorbitol from 0 to 1 . 2 M over approximately 2 hr ( related to Figure 6 ) . A representative field of view ( DIC channel ) showing cells trapped in the ALCATRAS microfluidics device . White arrows indicate cells that are either arrested or dead . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 017 For steps , a fast response is a priority , and although the slow pathway is more accurate , the corresponding greater degree of overshoot of glycerol in the fast pathway , as expected by our model ( Figure 5D ) , does not substantially affect survival of the fast mutant . In the model , the overshoot is counteracted by export through the Fps1 channels , and so the fast mutant pays a greater metabolic cost by exporting more glycerol . This cost does not affect survival but may change other correlates with fitness , such as lag time . In ramps , where the environment continually changes , the derivative action in the fast pathway does not eventually decrease like it does in steps and the overshoot of glycerol correspondingly has a longer life-time in the fast mutant ( Figure 5E ) and is therefore likely to be more deleterious . An indication of deleterious effects in higher gradients can be seen in the amplitude of Hog1 ( Figure 4C ) : although the amplitude of the wild-type Hog1 increases linearly with the gradient of the ramp and is therefore likely to be informative about this gradient , the saturating response of the fast mutant implies that its amplitude of Hog1 will often be similar , at least for inputs with higher gradients . Individual cells in the fast mutant may not then be able to match their response to the magnitude of the stress . In the fluctuating ramps , accuracy is a priority . In the fast mutant , the derivative action depends only on the current value of the input and so the fast mutant responds almost anew each time the stress increases regardless of the cell’s internal state ( levels of glycerol ) . The resulting overshooting of glycerol is therefore compounded because of the long life-times of the overshoots and the ramp’s greater duration . The slow mutant , in contrast , can modulate its response both by the cell’s internal state and by the history of the stress because of its greater sensitivity to the integral feedback . With its two input pathways , the wild-type performs both tasks and is both fast and accurate , always having the highest probability of survival .
We have thus shown that trade-offs in performance can undermine signalling in a single input pathway with either speed being sacrificed for accuracy or vice versa , but that by having two input pathways , each specializing to particular task , signalling networks can mitigate these trade-offs ( Figure 6B ) . In the HOG network , the Sln1 pathway is a fast reflex-like response that provides the speed necessary to survive sudden shocks but at the expense of accuracy , and alone can cause adaptation of Hog1 before recovery of the cellular volume . Consistent with this observation , Macia et al . report that fast mutants have a shorter duration of Hog1 phosphorylation compared to wild-type cells ( Macia et al . , 2009 ) . In contrast , the Sho1 pathway provides accuracy at the expense of speed and by specializing to sensing the integral feedback coming from volume recovery is more sensitive to the cell’s internal state and the history of the extracellular stress . This behaviour is consistent with earlier speculations that the Sho1 branch primarily monitors osmotic changes during normal growth ( Hohmann , 2002 ) . If the integral feedback is to allow recovery of the volume , the network must remember the cellular volume before the stress ( Astrom and Murray , 2008 ) , and the Sho1 pathway interacts with the actin cytoskeleton ( Tanaka et al . , 2014 ) , which might allow information from cell morphology and growth to be integrated with activation of Hog1 . The two input pathways have been reported to have different thresholds of activation ( Macia et al . , 2009 ) , but our data and modelling points towards a re-interpretation in terms of different gains for the pathways . For all the steps and ramps of stress that we consider , we observe a response from both mutants , and so any differences in thresholds must be small ( less than 0 . 2 M sorbitol in steps and 0 . 03 M min-1 in ramps ) . The advantage of multiple thresholds might be to increase the network’s dynamic range , but given that both thresholds can only be small , such an increase is unlikely to be substantial in the HOG network . Our data points towards it being the interaction between the two pathways that increases the dynamic range: we observe that only the wild-type response increases linearly with the gradient of a ramp of stress ( Figure 4C ) . A potentially alternative architecture of the HOG signalling network is to have a single fast input pathway controlling the integral feedback . Such a network , however , would not only have structural instabilities in its dynamics because of the high gain necessary for high speed , but also would be more likely to become insensitive to the cell’s internal state for sufficiently high stress . In large stress , all Hog1 molecules can become activated and the output of the HOG controller is then saturated . This saturation will happen for shocks of smaller magnitude for systems with high gain and causes a loss of accuracy because the system is then in open loop and is unable to exploit the integral feedback . Saturated activity of Hog1 should generate maximum production of glycerol and so faster recovery of cellular volume , but , once the volume has recovered , the level of glycerol synthesis will be too high for the level of stress and there will be a fitness cost . Having a slow pathway that inhibits the fast pathway helps prevent saturation of Hog1 activity and so increases sensitivity to the integral feedback for higher levels of stress . We have developed a block diagram model of the HOG network , but a caveat is that , although the model is therefore modular , it is also agnostic to the biochemical details of both the interaction between the two input pathways and the mechanism allowing the fast pathway to respond to the time-derivative of the input . Competing for Pbs2 is one possible means of cross-inhibition between the pathways , but multiple feedbacks , both positive and negative , exist within the HOG network ( Hao et al . , 2007; Macia et al . , 2009; Sharifian et al . , 2015; English et al . , 2015 ) , and a feedback-based interaction is possible . That biochemistry can be used to measure a time-derivative on a time-scales as fast as seconds is well established ( Block et al . , 1983 ) , and , in analogy with bacterial chemotaxis , we expect that upstream signalling in the fast pathway encodes a short-term memory of the level of the input to allow comparison of the current level to a value in the past . More generally , our results confirm the importance of using inputs with varying dynamics to uncover the logic behind cellular signalling ( Nurse , 2008; Alexander et al . , 2009 ) . In the wild , organisms are exposed to signals with a wider range of temporal behaviours then the constant inputs typically studied in the laboratory ( López-Maury et al . , 2008 ) , and signal transduction is likely to have evolved to allow organisms to differentiate between such signals or at least between classes of signals ( Bowsher and Swain , 2014; Tkačik and Bialek , 2016 ) . Although such work is still in its infancy , dynamic inputs have been successfully used to understand signalling responses in , for example , bacteria ( Young et al . , 2013 ) , yeast ( Hao et al . , 2013 ) , and mammalian cells ( Kellogg and Tay , 2015 ) , and , with the ease of use of microfluidics ( Bennett and Hasty , 2009 ) , we believe should become commonplace . In conclusion , we have shown that cellular signalling is vulnerable to fundamental trade-offs in performance , but that these vulnerabilities can be overcome by distributing tasks to different parts of the network and integrating together the outputs of this division of labour . We therefore expect such improvements in performance by the specialization of subnetworks to different tasks to exist broadly within cellular signal transduction .
All strains ( Table 1 ) were constructed using PCR-based genomic integration and were validated by colony PCR . For inducible expression of PBS2 , we used the Tet-off system ( Bellí et al . , 1998 ) , for which doxycycline causes repression . We PCR-amplified the kanMX4-tTA-PtetO7 from plasmid pCM225 ( Bellí et al . , 1998 ) and inserted to substitute 200 bp upstream of the PBS2 ORF . Correct insertion was verified by colony PCR . The mutants showed equivalent growth from wild-type strains in XY media with 2% glucose over 24 hr ( data not shown ) . 10 . 7554/eLife . 21415 . 018Table 1 . Saccharomyces cerevisiae strains used . All strains are in the S288C background . DOI: http://dx . doi . org/10 . 7554/eLife . 21415 . 018StrainGenotypeSourceBY4741MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0EUROSCARFBY4742MATα his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0EUROSCARFSL364MATa , leu2Δ0 , lys2Δ0 , HOG1-GFP::HIS3 , HTB2-mCherry::URA3gift - P . HersenSL373MATa , met15Δ0 , HOG1-GFP-HIS3 , HTB2-mCherry::URA3 , ste11::LEU2gift - P . HersenSL268MATa , leu2Δ0 , lys2Δ0 , HOG1-GFP-HIS3 , HTB2-mCherry::URA3 , ssk1::KANMX6This studySL395MATa , leu2Δ0 , lys2Δ0 , HOG1-GFP::HIS3 , HTB2-mCherry::URA3 , P PBS2Δ::KANMX4-tTA-P tetO7This studySL396MATa , met15Δ0 , HOG1-GFP::HIS3 , HTB2-mCherry::URA3 , P PBS2Δ::KANMX4-tTA-P tetO7 , ste11::LEU2This studySL442MATa , lys2Δ0 , HOG1-GFP-HIS3 , HTB2-mCherry-URA3 , P PBS2Δ::kanMX4-tTA-P tetO7 , ssk1::HphThis study To model the HOG network we adopted a modular approach based on control theory ( Astrom and Murray , 2008 ) and used Matlab’s modelling platform Simulink . The architecture of the model is shown in Figure 5—figure supplement 1 . Building on previous studies ( Muzzey et al . , 2009 ) , we developed a linear model in which the different components of the system are first-order linear time-invariant systems characterized by a transfer function kτs+1 with a zero-frequency gain of k and a time-constant of τ . The error is defined as the difference between the external input , u ( t ) ( the extracellular osmolarity ) , and the internal state , g ( the intracellular concentration of glycerol ) . The slow pathway responds to the error; the fast pathway responds to the error and additionally has a derivative component that depends directly on the input u ( t ) . The two pathways inhibit each other at different time-scales: τfast and τslow . The output of the two pathways , after the cross-inhibition , is added to generate the wild-type response of Hog1 , which affects levels of glycerol through an integrator . Finally , the Hog1-independent pathway responds proportionally to the error and feeds directly into glycerol . To simulate the mutants , we remove one input pathway and the cross-inhibition between the pathways . To parameterize the model , we used parameter optimization in the Simulink platform to simultaneously fit the response of the mutants to a step ( 0 . 6 M ) and a ramp ( 0 . 03 M min-1 ) . We then incorporated the cross-inhibition between the two pathways and fitted the inhibition parameters , τfast and τslow , to the wild-type response . A Simulink file is available as supplemental material . Data shown in the figures is freely accessible at dx . doi . org/10 . 7488/ds/2043 . | The faster we do tasks the harder it is to do them well . For example , when we wish to judge if , say , a cup , is too hot , we first quickly withdraw our hand after touching it: we know that the cup is hot but not how much . Next we hold a finger against the cup to accurately judge its temperature . Such speed-accuracy trade-offs are studied widely in fields ranging from neuroscience to engineering , but their consequences for single cells are unknown . This is despite the fact that when cells are exposed to stress they must respond both quickly ( to survive ) and accurately ( to reduce how many resources they consume ) . One way of stressing yeast cells is to place them in a syrupy substance called sorbitol . This causes the cells to lose water , shrink in size , and launch a stress response to regain volume . If the cells respond inappropriately to the situation , they may die . The signalling network that produces the stress response is unusual in that it has a Y-shaped structure , where the two ‘arms’ of the Y are the input pathways . Although it was known that one input pathway responds to stress faster than the other , the advantages of having two inputs in the signalling network were not understood . Granados , Crane et al . thought that the differences in speed and the Y-shaped structure could allow the cell to respond to stress with both speed and accuracy . To investigate this theory , Granados , Crane et al . used a microscope to study individual yeast cells that had been exposed to sorbitol . Combining these results with a mathematical model of the cell signalling network revealed that a mutant yeast cell that only has one of the input pathways specializes in speed but is inaccurate , similar to a reflex-like response . In contrast , a mutant with only the other pathway specializes in accuracy , being slower but matching the level of the cell’s response to the level of stress placed on it . This trade-off is reflected in rates of cell survival: the first mutant survives best in sudden shocks of stress; the second mutant survives best in gradually increasing stress . Normal yeast cells that have both input pathways survive more often than either mutant . Overall , the results presented by Granados , Crane et al . reveal principles behind cellular decision-making that should hold true in more complex organisms and could be exploited by synthetic biologists to programme cells with new behaviours . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Methods",
"and",
"materials"
] | [
"computational",
"and",
"systems",
"biology"
] | 2017 | Distributing tasks via multiple input pathways increases cellular survival in stress |
Understanding how cellular identity naturally interconverts with high efficiency and temporospatial precision is crucial for regenerative medicine . Here , we revealed a natural midgut-to-renal lineage conversion event during Drosophila metamorphosis and identified the evolutionarily-conserved homeodomain protein Cut as a master switch in this process . A steep Wnt/Wingless morphogen gradient intersects with a pulse of steroid hormone ecdysone to induce cut expression in a subset of midgut progenitors and reprogram them into renal progenitors . Molecularly , ecdysone-induced temporal factor Broad physically interacts with cut enhancer-bound Wnt pathway effector TCF/β-catenin and likely bridges the distant enhancer and promoter region of cut through its self-association . Such long-range enhancer-promoter looping could subsequently trigger timely cut transcription . Our results therefore led us to propose an unexpected poising-and-bridging mechanism whereby spatial and temporal cues intersect , likely via chromatin looping , to turn on a master transcription factor and dictate efficient and precise lineage reprogramming .
Classical regenerative strategies are facing challenges , including difficulties associated with the acquisition , delivery and integration of proper cell types into a complex milieu of tissues . In vivo lineage reprogramming , conversion of a highly specialized cell into the desired cell identity , has therefore emerged as an alternative and promising regenerative strategy ( Heinrich et al . , 2015; Jopling et al . , 2011 ) . However , molecular mechanisms underlying in vivo lineage conversion remained obscure ( Heinrich et al . , 2015; Jopling et al . , 2011 ) . Due to its high efficiency and temporospatial precision , rare naturally-occurring lineage reprogramming events provide powerful model systems for elucidating the molecular basis of cell plasticity and identity switch . Here , we revealed a unique naturally-occurring midgut-to-renal lineage reprogramming event at the onset of Drosophila metamorphosis . Drosophila excretory system , so-called Malpighian tubules , are two pairs of tubules converge through common ureters onto midgut-hindgut junction ( Figure 1—figure supplement 1A ) ( Denholm and Skaer , 2009; Dow , 2009; Singh et al . , 2007 ) . Each pair of renal tubules can be mainly divided into three segments: ureter , lower tubule and upper tubule ( Figure 1A , B and Figure 1—figure supplement 1A ) ( Singh et al . , 2007; Sözen et al . , 1997 ) . The ureter can be further divided into lower and upper regions ( Figure 1B ) . Renal stem cells ( RSCs ) were found to be dispersed in the adult ureter and lower tubule regions ( Figure 1B ) ( Singh et al . , 2007 ) but not in the larval renal tubules , raising the question of how the adult RSCs emerge in development . Earlier work ( Takashima et al . , 2013 ) and our independent observations found that adult RSCs are likely to be derived from progenitors within the midgut region . Midgut progenitors ( MPs ) and renal progenitors ( RPs ) , although both express Snail-type transcription factor Escargot ( Esg ) , are distinct populations of precursor cells in terms of lineage composition and functionality: midgut progenitors/stem cells undergo asymmetric cell divisions to self-renew and meanwhile differentiate into hormone/peptide-secreting enteroendocrine ( EE ) cells and nutrient-absorbing enterocytes ( ECs ) ( Micchelli and Perrimon , 2006; Ohlstein and Spradling , 2006 ) ; in contrast , renal progenitors undergo asymmetric , self-renewing divisions to give rise to principal cells that mediate organic cation and solute transport ( Singh et al . , 2007 ) . Intriguingly , we observed that , during metamorphosis , a small subset of Esg+ progenitors appeared to migrate away from the midgut and onto the renal tubules ( Figure 1C–E ) , where they terminally differentiated into new Cut+principal cells ( arrowheads in Figure 1D ) , replacing the old Cut- principal cells in the lower ureter region ( arrowheads in Figure 1C ) ( Takashima et al . , 2013 ) . However , it remains enigmatic when , where and how the pool of Esg+ midgut progenitors is selected and converted into renal identity during metamorphosis . To ascertain this midgut-to-renal lineage conversion event and , more importantly , to probe its underlying regulatory mechanisms and molecules , we carried out a genome-wide unbiased RNAi-based genetic screen ( Xu et al . , unpublished ) and identified the homeodomain protein Cut as a master switch dictating this lineage reprogramming event . Cut is an evolutionarily-conserved homeodomain-containing transcription factor that has been shown to regulate various developmental events in Drosophila and mammals , including sensory organ identity specification and dendritic morphogenesis in peripheral nervous system , dorsal-ventral boundary formation in the fly wings , projection neuron dendritic targeting , as well as patterning and growth during fly airway remodeling ( Becam et al . , 2011; Blochlinger et al . , 1988; Bodmer et al . , 1987; Cubelos et al . , 2010; Grueber et al . , 2003; Komiyama and Luo , 2007; Ludlow et al . , 1996; Pitsouli and Perrimon , 2013; Rodríguez-Tornos et al . , 2016 ) . Here , we show that a steep Wnt/Wingless ( Wg ) morphogen gradient ( Clevers and Nusse , 2012; Loh et al . , 2016 ) at the midgut-hindgut boundary intersects with a pulse of the steroid hormone ecdysone at the onset of metamorphosis ( Yamanaka et al . , 2013 ) to induce cut expression in a subset of midgut progenitors and reprogram them into renal progenitors . Mechanistically , the Wg morphogen gradient , through its pathway effector TCF/β-catenin , determines the pool of future renal progenitors , presumably by poising a distal cut enhancer for timely activation . On the other hand , the hormone ecdysone-induced BTB-Zinc finger protein Broad determines the timing of lineage conversion by physically interacting with enhancer-bound TCF/β-catenin complex and likely bridging the distal enhancer and promoter region of cut through its self-association . Such long-range enhancer-promoter looping could subsequently trigger timely cut transcription . Thus , integration of spatial and temporal cues by a master cell identity switch , likely through a chromatin looping mechanism , orchestrates natural lineage reprogramming with temporal and spatial precision .
To identify key regulators governing midgut-to-renal lineage conversion , we carried out a genome-wide RNAi-based screen . We used temperature-sensitive , midgut and renal progenitor-specific esg-Gal4 , UAS-CD8-GFP , tub-Gal80ts system to drive RNAi expression , transferred animals from permissive temperature ( 18°C ) to restrictive temperature ( 29°C ) at mid third instar larval stage , approximately 48 hr before the lineage reprogramming event occurs , and analyzed the renal phenotypes at early adult stages . Intriguingly , we found that midgut progenitor-specific knockdown of the transcription factor Cut resulted in lack of the entire lower ureter region ( brackets in Figure 1—figure supplement 1B ) and appearance of extra Esg- diploid cells along the renal tubules ( arrowheads in Figure 1—figure supplement 1B ) . Such striking renal tubule phenotypes of cut-RNAi prompted us to carefully investigate the expression pattern of Cut in adult intestine and renal tubules . Cut has previously been found to be highly expressed in two types of post-mitotic , polyploid cells in the Drosophila digestive-excretive system: the acid-secreting copper cells in the middle midgut region ( Strand and Micchelli , 2011 ) and the fluid-balancing principal cells within renal tubules ( Singh et al . , 2007; Singh et al . , 2011 ) . Surprisingly , we detected moderate Cut expression in Esg+ diploid cells along adult renal tubules ( arrowheads in Figure 1F , F’ ) . Knockdown or overexpression of Cut within adult renal stem cells , by esg-Gal4 , diminished or elevated the Cut expression levels respectively ( Figure 1—figure supplement 1C ) , indicating that such Cut expression is specific . In contrast , Cut was not detectable in adult intestinal stem cells or enteroblasts ( EBs ) in midgut ( arrowheads in Figure 1G , G’ ) . To further confirm the specific expression of Cut in renal stem cell lineages , we induced wild-type MARCM ( mosaic analysis with repressible cell marker ) clones ( Lee and Luo , 2001 ) at larval stages and analyzed Cut expression pattern of clones at early adult stage . Using the MARCM system , the homozygous clones are produced upon mitotic recombination and are positively labeled by GFP . MARCM clones induced at the lower ureter region ( bracket in Figure 1H ) contained both diploid cells expressing moderate levels of Cut ( white arrowheads in Figure 1I , I’ ) and polyploid cells expressing high levels of Cut ( yellow arrowheads in Figure 1I , I’ ) , indicating that new Cut+ principal cells are derived from Cut+ renal progenitors ( Figure 1B , D , I ) . By contrast , MARCM clones in the midgut region , composed of diploid ISCs ( cyan arrowheads in Figure 1I ) , EBs , EEs and polyploid ECs ( pink arrowheads in Figure 1I ) , did not exhibit Cut expression . Therefore , distinct from Cut- intestinal stem cells ( ISCs ) that differentiate into Cut- EEs and ECs in the midgut ( Figure 1J ) , Cut+ renal stem cells ( RSCs ) differentiate into Cut+ principal cells ( PCs ) on the renal tubules ( Figure 1K ) . Taken together , we identified Cut as a specific marker that distinguishes adult RSCs from ISCs . Utilizing Cut as a cell identity marker , we studied the emergence , proliferation , migration and differentiation of RPs during metamorphosis . Cut was undetectable in any Esg+ progenitors in third instar larval ( L3 ) midguts ( Figure 2A ) . At the onset of metamorphosis ( 0 hr after puparium formation ( APF ) ) , a few clusters of Esg+ progenitors in closest proximity to the midgut-hindgut boundary started to exhibit Cut expression ( white arrowheads in Figure 2B ) . At 0 . 5 hr APF , 20–30 Esg+ progenitors in 5–6 clusters specifically expressed Cut ( yellow bracket in Figure 2C ) . By 1 hr APF , midgut progenitor islands merged together and the Cut+ progenitors aligned along the midgut-hindgut border in 1–3 rows ( yellow bracket in Figure 2D ) . Starting at 3 hr APF , Esg+ Cut+ progenitors migrated across the midgut border and entered the renal tubules ( yellow bracket in Figure 2E ) . Cut- principal cells in the lower ureter region were engulfed and expelled into the pupal midgut ( pink arrowheads in Figure 2E ) ( González-Morales et al . , 2015 ) . Cut+ progenitors in turn occupied the lower ureter region ( Figure 2F ) and differentiated into Esg- Cut+ new principal cells ( cyan arrowheads in Figure 2G ) . The dynamic changes in the number of Cut+ progenitors and the emergence , proliferation , migration and differentiation of RPs during metamorphosis are schematically presented in Figure 2H and I respectively . At the onset of metamorphosis , Cut+ progenitors expressed midgut progenitor marker Delta ( Dl ) ( arrowheads in Figure 2—figure supplement 1A , B ) , indicating that Cut+ progenitors were transited from MP to RP characteristics . Interestingly , immediately after the Cut+ progenitors migrated onto renal tubules , they started to turn off Dl expression ( arrowheads in Figure 2—figure supplement 1C ) . By 9 hr APF , Cut+ progenitors along the renal tubules completely shut down Dl expression ( arrowheads in Figure 2—figure supplement 1D ) . Therefore , Cut+ progenitors are progressively reprogrammed from MPs to RPs . The highly specific and restrictive induction of Cut expression in RPs right before their migration onto renal tubules prompted us to investigate whether Cut plays a critical role in the identity switch and/or migration of future RPs . Upon cut knockdown by esg-Gal4 , the migration of the small subset of Esg+ progenitors onto renal tubules remained relatively normal ( Figure 3A–D’ ) , indicating that Cut is dispensable for progenitor mobility . To investigate whether Cut is important for the identity switch of Esg+ progenitors , we performed lineage-tracing experiments based on the G-TRACE ( the Gal4 Technique for Real-time and Clonal Expression ) system ( Evans et al . , 2009 ) . We used temperature-sensitive esg-Gal4 , UAS-CD8-GFP , tub-Gal80ts system to drive FLP ( flippase ) expression , transferred animals from permissive temperature ( 18°C ) to restrictive temperature ( 29°C ) at late third instar larval stage and analyzed the renal phenotypes at early adult stages ( Figure 3A , B ) . Upon esg-Gal4-driven expression of FLP recombinase , a transcriptional stop cassette flanked by FRT sites is excised , resulting in lacZ expression in Esg+ progenitors and all their subsequent daughter cells ( lineage expression; Figure 3A ) . Meanwhile , GFP reveals the real-time expression of esg-Gal4 ( Figure 3A ) . In wild type renal tubules , Esg- lacZ+ cells derived from Esg+ lacZ+ RPs ( white arrowheads in Figure 3C , C’ ) were found in the lower ureter region ( brackets in Figure 3C , C’ ) . These Esg- lacZ+ cells were polyploid , new Cut+ principal cells ( white arrowheads in Figure 3C , C’ ) , in accordance with our earlier notion that Cut- principal cells are replaced with RP-derived new Cut+ principal cells during metamorphosis . cut knockdown by esg-GAL4 caused the majority of flies to die at late pupal or early adult stages ( 90 . 3% , n = 567 ) . The adult escaper flies lacked the whole lower ureter region ( Figure 3D , D’ ) , hinting that maldevelopment of the renal tubules contributed to the lethality of the animals . In contrast to the wild type control ( Figure 3C , E ) , Esg- lacZ+ cells derived from cut-RNAi RPs were dispersed throughout the ureter and lower tubule regions ( white arrowheads in Figure 3D’ ) . Strikingly , these Esg- lacZ+ cells were diploid , Cut- ( Figure 3D , D’ ) , and expressed Prospero ( Pros ) ( Figure 3F , F’ ) , a typical marker for EE cells . These results strongly suggested that , upon cut depletion , Esg+ progenitors migrate normally onto the renal tubules , yet fail to switch into renal identity and differentiate into midgut cells along renal tubules . Importantly , upon cut depletion , Esg+ progenitors along renal tubules behaved essentially the same as MPs at pupal stages ( Guo and Ohlstein , 2015 ) : ( 1 ) they produced ectopic Pros+ diploid cells along renal tubules at the same developmental stages as wild type MPs giving rise to EEs in midgut ( arrowheads in Figure 3—figure supplement 1A , B ) ; ( 2 ) The ectopic Pros+ cells derived from them expressed additional EE markers such as exocytosis regulator Rab3 ( Dutta et al . , 2015; Patel et al . , 2015 ) , synaptic protein Bruchpilot ( Brp ) ( Zeng et al . , 2013 ) , and secretory neuropeptide hormone Allatostatin A ( Ast A ) ( Beehler-Evans and Micchelli , 2015 ) ( yellow arrowheads in Figure 3—figure supplement 1C–E ) , indicating that they are bona fide EE cells; ( 3 ) they underwent asymmetric cell divisions , resulting in unidirectional Notch activation in progenitors ( open arrowheads in Figure 3—figure supplement 2A–D ) and asymmetric localization of Pros to the basal context ( Figure 3—figure supplement 2E ) ; and ( 4 ) their cell fate decisions were tightly regulated by the strength of Notch signaling ( Figure 3—figure supplement 2F–H ) . To further confirm the notion as proposed above , we carried out MARCM clonal analysis . GFP-marked clones were induced at second instar stage and analyzed at mid pupal ( Figure 3G , H ) or early adult stages ( Figure 3I , J ) . Pros+ EEs or Pdm1+ polyploid ECs were present in wild type MARCM clones in the midgut ( white arrowheads in Figure 3G , I ) but not renal tubule regions ( pink arrowheads in Figure 3G , I ) . In sharp contrast , both EEs and ECs were found within cut mutant MARCM clones in the renal tubule regions ( yellow arrowheads in Figure 3H , J ) . Moreover , as a consequence of cut depletion from Esg+ progenitors , both the total length of the ureter region ( Figure 3K ) and the number of principal cells in ureter ( Figure 3L ) were sharply reduced . Taken together , our results clearly demonstrated that Cut is necessary for MP-to-RP identity switch . To test whether Cut is sufficient in dictating midgut-to-renal lineage conversion , we overexpressed Cut in MPs and analyzed their lineage progression using MARCM clonal analysis . In wild type MARCM clones derived from single MPs , GFP+ polyploid cells were Pdm1+ ECs ( white arrowheads in Figure 3M ) . In contrast , MARCM clones derived from Cut-overexpressing MPs mainly contained Pdm1- Cut+ polyploid cells ( yellow arrowheads in Figure 3M ) , strongly suggesting that MPs were converted into RPs that in turn differentiated into principal cells in midgut region . Collectively , our observations reinforced the idea that Cut acts as a master switch in dictating natural midgut-to-renal progenitor identity conversion ( Figure 3N ) . We next sought to identify the signaling cues that induced Cut expression in the specific subsets of MPs . The spatially restricted distribution of Cut+ MPs hinted that Cut expression might be induced by a morphogen gradient emanating from proximal cells . After examining the distribution pattern of various signaling molecules or receptors at the onset of metamorphosis , we found that , in accordance with previous observations ( Fox and Spradling , 2009; Takashima et al . , 2008; Tian et al . , 2016 ) , the Wnt/Wingless ( Wg ) ligand ( Clevers and Nusse , 2012; Loh et al . , 2016 ) was expressed in a narrow zone of 2–3 rows of cells right at the midgut-hindgut boundary , starting from early larval stages ( Figure 4A , B and Figure 4—figure supplement 1A , B ) . At the onset of metamorphosis , 2–3 clusters of progenitors in closest proximity to the stripe of Wg-producing cells started to turn on Cut expression ( arrowheads in Figure 4C and Figure 4—figure supplement 1C ) . Around 0 . 5 hr APF , Cut was expressed in 5–6 progenitor islands adjacent to the Wg-secreting cells ( arrowheads in Figure 4D ) . Some Cut+ progenitor islands were not in direct contact with the Wg-producing band ( arrowheads in Figure 4D ) , suggesting the existence of Wg morphogen gradient . At 1 hr APF , peripheral cells surrounding progenitor islands ( Mathur et al . , 2010 ) partially opened up , allowing Cut+ progenitor islands to merge with each other ( Figure 4E ) . A clear boundary between Cut+ and Cut- progenitors could be delineated ( dashed lines in Figure 4E and Figure 4—figure supplement 1D ) , suggesting the spatial precision of the inductive cue . Wg signaling activity , as faithfully reflected by the Fz3-GFP reporter ( Sivasankaran et al . , 2000; Tian et al . , 2016 ) , was only detectable in a few rows of MPs proximal to Wg-secreting cells ( closed arrowheads in Figure 4F ) , further supporting Wg signaling as a spatial cue in selecting future RPs . If Wg signaling provides an inductive cue , we reasoned that downregulation of Wg signaling in future RPs should abolish Cut expression . Indeed , specific knockdown of Wg within Wg-producing cells at midgut-hindgut boundary , by wg-Gal4 ( Alexandre et al . , 2014 ) , resulted in diminished Cut expression in Esg+ progenitors ( arrowheads in Figure 4G , H ) . Furthermore , inhibition of Wg signaling through overexpression of either a dominant-negative form of the Wnt pathway effector TCF ( ΔN-TCF ) ( van de Wetering et al . , 1997 ) or a constitutively-active form of the Wnt pathway inhibitor GSK3β/Shaggy ( Sgg-CA ) ( Bourouis , 2002 ) in Esg+ progenitors completely abolished Cut induction ( brackets in Figure 4I ) . Consistently , MARCM clonal analysis revealed that , upon depletion of the Wnt/Wg pathway positive component Disheveled ( Dsh ) , ectopic Pros+ EEs appeared in renal tubes ( yellow arrowheads in Figure 4—figure supplement 1E ) . Therefore , Wg signaling is essential for the MP-to-RP identity switch . We next investigated why Cut was expressed in only a small subset of Esg+ progenitors by probing the competence of MPs to respond to Wg signaling . Upon overexpression of Wg ligand in all MPs , high levels of Cut expression were detected in all Esg+ progenitors dispersed throughout the midgut ( arrowheads in Figure 4J ) . As a consequence , elevated number of Cut+ progenitors migrated onto renal tubules ( Figure 4—figure supplement 1F , G ) . In accordance , activation of Wg signaling in all MPs through overexpression of a constitutively-active form of the Wnt pathway effector β-catenin/Armadillo ( Arm-S10 ) ( Pai et al . , 1997 ) exhibited similar effects as Wg overexpression ( arrowheads in Figure 4J ) , demonstrating that all MPs are competent to respond to Wg signaling . Taken together , our data strongly support the notion that a steep Wg morphogen gradient provides a spatial cue to precisely select the pool of future RPs during metamorphosis ( Figure 4K ) . To find out to what extent the spread of Wg is required for Cut induction , we employed a membrane-tethered form of Wg , wg ( KO; NRT-Wg ) ( Alexandre et al . , 2014 ) . While membrane-tethered Wg sufficed to control fly patterning and growth ( Alexandre et al . , 2014 ) , it failed to induce Cut expression in future RPs ( Figure 4—figure supplement 1H ) , indicating that the spread of Wg is necessary for activating Wg signaling in these progenitors . The subset of MPs in close proximity to Wg-producing cells do not turn on Cut expression until the onset of metamorphosis , raising the question of how Cut induction is temporally controlled . Since the expression levels of wg-Gal4 , wg-lacZ or Wg signaling reporter Fz3-GFP at the midgut-hindgut boundary remained relatively constant from second instar larval to early pupal stages ( Figure 4A–F and Figure 4—figure supplement 1A–D ) , temporal cue ( s ) other than Wg ligand triggers Cut expression in future RPs at the onset of metamorphosis . The induction of Cut coincided with the pulse of the steroid hormone ecdysone released from the ring glands ( Ou and King-Jones , 2013; Yamanaka et al . , 2013 ) , hinting that the ecdysone hormone may serve as a temporal cue ( Praggastis and Thummel , 2017; Uyehara et al . , 2017 ) . Indeed , Cut induction was abolished upon Esg+ progenitor-specific expression of a dominant-negative form of the Ecdysone receptor ( EcR-DN ) ( Brown et al . , 2006 ) ( Figure 5A ) . Since EcR was widely expressed in all cell types in the midgut at different developmental stages ( Figure 5—figure supplement 1A , B ) , we considered the possibility that the strong pulse of ecdysone at metamorphosis was translated into a temporal patterning of early response genes downstream of EcR . Supporting this notion , Esg+ progenitor-specific depletion of the Broad complex ( Br-C ) ( Figure 5A ) , a crucial early response gene of the ecdysone signaling ( Fletcher and Thummel , 1995; Karim et al . , 1993 ) , phenocopied the effects of EcR-DN . In comparison , Cut expression was normally turned on in future RPs upon downregulation of E74 and E75 , the other two well-characterized ecdysone early response genes ( Figure 5—figure supplement 1C ) . These observations indicated that the steroid hormone ecdysone executed its control on Esg+ progenitor identity through specific downstream effector Br-C . Strongly supporting this idea , the temporally dynamic expression pattern of Br-C coincides with the strong pulse of ecdysone during metamorphosis: Br-C protein expression was barely detectable in early third instar larvae , progressively increased starting mid third instar larval stage and peaked at the onset of metamorphosis , followed by a gradual decline ( Figure 5B and Figure 5—figure supplement 1D–H ) . Together , our results revealed an ecdysone-EcR-Br regulatory axis , which induces Cut expression and dictates progenitor identity switch with temporal precision . The above findings identified the Wnt/Wg ligand and the steroid hormone ecdysone as spatial and temporal cues respectively in turning on the master identity switch Cut . This leaves us with the important question of how the spatial and temporal signaling is integrated at molecular level . To address this question , we first carried out epistatic analysis . Depletion of Br-C abolished ectopic Cut expression induced by Wg or Arm-S10 overexpression in MPs ( Figure 5C and Figure 5—figure supplement 1I ) , demonstrating that Br-C acts downstream of or in parallel with the Wg pathway transcription activation complex TCF/Arm in cut induction . Furthermore , Wg pathway activity , as indicated by the Fz3-GFP reporter , was highly responsive to a reduction in Wg signaling , but remained unaltered upon downregulation of ecdysone signaling ( Figure 5D ) . This rules out the possibility of a general modulation of the Wg signaling output by the ecdysone pathway . Therefore , it is likely that Wg and ecdysone signaling converge on the control of cut expression in a gene-specific manner ( Figure 5E ) . Given that Br itself is a BTB-ZF transcription factor , it may physically associate with TCF/Arm to form a transcription activation complex and synergistically trigger cut transcription . To test this idea , we first carried out coimmunoprecipitation ( coIP ) assays . The Br-C gene locus encodes four distinct splicing isoforms , Br-Z1 , Br-Z2 , Br-Z3 and Br-Z4 , which share a common N-terminal core domain but have distinct C-terminal zinc-finger domains ( Figure 6A ) ( Mugat et al . , 2000 ) . Indeed , we found that Arm could be specifically coimmunoprecipitated with Br-C isoforms from HEK293T cell extracts and exhibited relatively strong binding affinity to Br-TNT-Z1 , Br-Z2 and Br-Z4 ( Figure 6B ) . The results described above prompted us to probe the functional significance of the physical interaction between Br-C and TCF/Arm . While overexpression of Br-Z1 , Br-Z2 or TCF alone was barely able to precociously induce cut transcription at 9 hr before puparium formation ( BPF ) ( induction rate of 0 , 36 and 0% respectively; n = 11–12; arrowheads in Figure 6C ) , coexpression of Br-Z1 or Br-Z2 with TCF dramatically enhanced Br activity in premature induction of cut expression ( induction rate of 70 and 100% respectively; n = 9–17; arrowheads in Figure 6C ) , providing compelling evidence for a combinatorial regulation of cut transcription by Br and Arm/TCF . Simultaneous overexpression of TCF and either Br-Z3 or Br-Z4 failed to induce premature cut expression , highlighting the importance of the isoform-specific C-terminal domain for Br functionality . We next examined the temporal expression pattern of Br-Z1 and Br-Z2 . While Br-Z2 , as stained by our newly-raised antibody ( Figure 6—figure supplement 1A ) , exhibited prominent and specific expression in Esg+ progenitors at 0 hr APF , Br-Z1 was undetectable in the midgut region ( Figure 6D and Figure 6—figure supplement 1B ) . Significantly , the temporal expression pattern of Br-Z2 in Esg+ progenitors fully recapitulated that of Br-C ( Figure 6E ) . Therefore , our results clearly indicated that , although Br-Z1 has moderate ability to precociously induce cut transcription , Br-Z2 is the most likely isoform that governs Cut induction at the onset of metamorphosis . We next sought to investigate how Br-Z2 synergized with TCF/Arm to induce precocious cut expression . We first carried out coimmunoprecipitation assay and found that both TCF and Arm were specifically coimmunoprecipitated with Br-Z2 from 293 T cell extracts ( Figure 6F , G ) . To confirm the physical interaction between Arm and Br-Z2 in the nucleus , we next performed proximity ligation assay ( PLA ) , which detects protein-protein interaction in situ with high specificity ( Söderberg et al . , 2006 ) ( Figure 6H ) . Strong PLA signal was detected in the nuclei of S2 cells coexpressing Br-Z2 and Myc-tagged Arm ( Arm-Myc ) ( Figure 6I , J ) . By contrast , PLA signal was barely detectable in S2 cells expressing Br-Z2 or Arm-Myc alone or coexpressing Br-Z2 and Myc-tagged ELL , a subunit of the transcription regulatory complex SEC ( Super Elongation Complex ) ( Figure 6I , J ) ( Liu et al . , 2017 ) . These results clearly demonstrated that Arm and Br-Z2 physically interact within the nucleus . Furthermore , our detailed domain-mapping analysis revealed that the C-terminal domain but neither the ZF nor the BTB domain was crucial for Br-Z2 to physically interact with TCF ( Figure 6F , K ) . Together , our results strongly suggested that Br-Z2 forms a transcription activation complex with TCF/Arm in inducing cut expression in future RPs . We next sought to identify cis-regulatory elements of cut that confer its response to Wg and ecdysone signaling in future RPs . Since a cut-lacZ reporter harboring a well-characterized enhancer upstream of cut promoter ( Jack et al . , 1991; Jia et al . , 2016 ) did not exhibit expression in RPs during metamorphosis , we systematically screened a series of cut enhancers . Out of 22 cut enhancer-Gal4 driver lines from the Janelia Gal4 collection , in which Gal4 is expressed under the control of cut enhancer fragments ( Pfeiffer et al . , 2008 ) ( Figure 7—figure supplement 1A ) , we identified one line , R35B08-Gal4 , which drove UAS-CD8-RFP expression specifically in future RPs at metamorphosis ( Figure 7—figure supplement 1B ) . R35B08 is a previously-uncharacterized 3 . 2 kilobases ( kb ) enhancer fragment in the second intron of cut , approximately 50 kb downstream of cut promoter ( cut-intron2-enhancer in Figure 7A ) . The temporal and spatial expression pattern of cut-intron2-enhancer-GFP , a GFP reporter that we generated for this intronic enhancer , was essentially identical to that of endogenous Cut protein ( Figure 7B ) , indicating that the temporospatial induction of cut in future RPs is regulated at transcriptional level and primarily , if not solely , through this distal enhancer . Consistent with this idea , cut-intron2-enhancer-GFP expression in future RPs was highly responsive to alterations in Wg or ecdysone signaling ( Figure 7C , D ) . The long distance between the intron2-enhancer and the promoter of cut suggested that chromatin looping might juxtapose the distal enhancer with cut promoter , crucial for cut induction . We therefore considered the tantalizing scenario whereby Br-Z2 self-association promotes the enhancer-promoter communication and cut transcription , based on the following observations: ( 1 ) cut intron2-enhancer contains closely-spaced putative TCF- and Br-Z2-binding sites ( Figure 7E ) ( Archbold et al . , 2014; Chang et al . , 2008; von Kalm et al . , 1994 ) ; ( 2 ) cut promoter region harbors putative Br-Z2-binding site but not TCF-binding site ( Figure 7E ) ; and ( 3 ) Br-Z2 contains BTB domain at its N-terminus ( Figure 6D ) , which is likely to mediate protein dimerization or oligomerization ( Perez-Torrado et al . , 2006 ) . To test this looping hypothesis , we first investigated whether TCF and Br-Z2 binds to their putative bindings site in the cut locus . Indeed , our electromobility shift assay ( EMSA ) results demonstrated a direct and sequence-specific binding of TCF and Br-Z2 to their putative binding sites in the cut promoter or intron2-enhancer region ( Figure 7F ) . Next , we assessed whether Br-Z2 can self-associate . Our coIP data clearly showed that Br-Z2 formed protein dimer in a BTB domain-dependent manner ( Figure 7G ) . Furthermore , deletion of only five amino acids ( 5AA; aa 46–50 ) in the BTB domain was sufficient to completely abolish the ability of Br-Z2 to self-associate ( Figure 7H , I ) . Finally , we assayed the functional significance of Br-Z2 protein dimerization . Deletion of the whole BTB domain or only five amino acids in this domain ( aa 46–50 ) abolished the activity of Br-Z2 to precociously induce cut expression within future RPs ( Figure 7J ) , indicating that the ability to form protein dimer per se is crucial for Br-Z2 to control cut transcription . ZF domain-deleted form of Br-Z2 also failed to prematurely induce cut expression ( Figure 7J ) , demonstrating that the sequence-specific DNA-binding ability is equally important for Br-Z2 to dictate cut transcription . Taken together , our results identified the homeodomain protein Cut as a master switch that converts MPs into RPs in the right place at the right time ( Figure 8A ) . RPs in turn migrate onto renal tubules and differentiate into renal cells ( Figure 8A ) . When Cut is depleted , Esg+ progenitors migrate normally , yet fail to switch identity and differentiate into midgut cells along renal tubules ( Figure 8A ) . Mechanistically , the temporal and spatial signals inducing cut transcription in future RPs seem to intersect by facilitating enhancer-promoter looping of cut: At the onset of metamorphosis , the pulse of hormone ecdysone induced peak expression of Br . Br in turn acts as a transcription activator through its physical interaction with TCF/Arm and meanwhile likely serves as a looping factor juxtaposing the TCF/Arm-bound enhancer with cut promoter , triggering timely cut transcription . ( Figure 8B ) . One important predication of this model is that cohesin ( Dorsett , 2011; Dorsett and Merkenschlager , 2013; Dowen and Young , 2014 ) and its loading factor Nipped-B , which facilitates and stabilizes enhancer-promoter looping ( Rollins et al . , 1999 ) , are essential for cut induction in future RPs . Indeed , cut expression at metamorphosis was markedly decreased upon depletion of cohesin subunit Stromalin ( SA ) or Nipped-B ( Figure 8C ) . Significantly , in sharp contrast , the expression of Fz3-GFP ( Figure 8D ) or cut-intron2-enhancer-GFP ( Figure 8E ) remained unaltered upon downregulation of cohesin or Nipped-B , indicating that cohesin is specifically required for chromatin looping-dependent cut transcription in future RPs .
Here we revealed a naturally-occurring midgut-to-renal lineage conversion event at the onset of Drosophila metamorphosis . Compared with experimentally induced reprogramming , natural reprogramming events in physiological settings are relatively rare , yet much more efficient , predictable and robust ( Gettings et al . , 2010; Jarriault et al . , 2008; Red-Horse et al . , 2010; Schaub et al . , 2015 ) . The lineage reprogramming process unveiled in our studies represents a unique natural lineage conversion event in that ( 1 ) The cell identity switch occurs between organ-specific progenitors , not fully-differentiated cells; ( 2 ) the reprogramming event takes place at postembryonic stages , when cells are much less plastic; and ( 3 ) the cell identity fully converts from one organ-specific characteristics to another . Thus , this midgut-to-renal lineage conversion event provides a previously unexplored physiological context for elucidating the detailed molecular mechanisms underlying cell plasticity . Our results further show that the homeodomain protein Cut is a master cell identity switch controlling the natural conversion of midgut progenitors into renal identity . Cut was originally identified as a binary identity switch between subtypes of neurons in the peripheral nervous systems ( PNS ) of Drosophila ( Bodmer et al . , 1987 ) . Compelling evidence demonstrated that Cut is both necessary and sufficient in specifying neuronal identities in fly PNS ( Blochlinger et al . , 1991; Bodmer et al . , 1987 ) . Therefore , Cut dictates cell identity switch within or across organ boundary . We reason that Cut might be particularly suitable for being a master cell identity switch in diverse biological contexts . Firstly , cut is an unusually large gene harboring a long and complex enhancer region spanning more than 150 kb . Such a long enhancer region can be subdivided into small segments responsive to different stimuli or signals in diverse tissues or organs at distinct developmental stages . Therefore , akin to neurons with extensive and complex arbors , the extra-long and segmented enhancer region of cut receives and integrates diverse input signals , and drives Cut expression with high temporospatial precision . Secondly , Cut , as a homeobox transcription factor , may intrinsically possess the ability to specify and confer organ or tissue identities , analogous to the classic homeotic genes such as Antennapedia , by simultaneously erasing old cell identities and writing new ones . Acquisition and switch of distinct cell identities are precisely and tightly controlled in both space and time ( Erclik et al . , 2017; Red-Horse et al . , 2010; Schaub et al . , 2015 ) . Yet the molecular basis underlying the integration of spatial and temporal signals remain poorly understood . We posit that two molecular mechanisms may underlie the intersection of temporal and spatial cues: ( 1 ) the spatial cues restrict the expression/activity of the temporal pathway component ( s ) , or vice versa . This strategy has been well exemplified in spatiotemporal control of border cell migration in the Drosophila ovary ( Jang et al . , 2009 ) and spatial restriction of neural competence ( Huang et al . , 2014 ) ; and ( 2 ) the temporal and spatial pathways converge to induce the expression of a specific set of target genes . The molecular basis of the latter scenario remains enigmatic . Interestingly , our results revealed that the spatial and temporal cues might intersect by an unexpected ‘poising and bridging’ mechanism to dictate the midgut-to-renal progenitor identity switch ( Figure 8B ) . In this model , the spatial transcription factors ( TF ) bind a distal enhancer and prime it for a timely response to the temporal cues , whereas the temporal TF acts as a bridging factor that binds to both the distal enhancer and the promoter region and induces enhancer-promoter looping through its self-association ( Figure 8B ) . Meanwhile , the temporal TF is likely to also act as a transcription activator by physically interacting with the spatial TFs . Importantly , since protein dimerization or oligomerization occurs only when the protein concentration rises above certain threshold ( Marianayagam et al . , 2004 ) , such a protein-dimerization-based regulatory mechanism is ideal for integrating and translating gradual changes in temporal and spatial signaling strength into a timely and all-or-none biological event such as cell identity switch . Although long-range chromatin looping has been found at numerous gene loci ( Ghavi-Helm et al . , 2014; Levine et al . , 2014 ) , the identity and mode of action of looping factors under developmental or physiological settings remain unclear . Our findings shed mechanistic insights into how a temporal factor might bridge the distal enhancer and the promoter of a master gene via protein dimerization in development . In light of recent studies implicating dimerization of TFs , such as CTCF , Yin Yang one and LDB1 , in the organization of chromatin architecture either globally ( Phillips and Corces , 2009; Weintraub et al . , 2017 ) or locally ( Deng et al . , 2012 ) , it is conceivable that TF dimerization or multimerization might represent a precise and prevailing mechanism establishing chromatin loops in space and time .
Fly culcture and crosses were performed according to standard procedures . Drosophila stocks used in this study include: UAS-Cut ( #36496; Bloomington stock center ( BDSC ) ) ; FRT19A , cutC145 ( #36496; BDSC ) ; UAS-Wg-HA ( #5918; BDSC ) ; UAS-Arm-S10 ( #4782; BDSC ) ; UAS-cut-RNAi ( #33967; BDSC ) ; UAS-br-RNAi ( #27272; BDSC ) ; esg-Gal4 , tubP-Gal80ts , UAS-GFP ( Biteau et al . , 2008; Micchelli and Perrimon , 2006 ) ; wg-Gal4 ( Alexandre et al . , 2014 ) ; wg ( KO; NRT–Wg ) ( Alexandre et al . , 2014 ) ; UAS-TCF-ΔN ( #4785; BDSC ) ; wg-lacZ ( #1672; BDSC ) ; Rab3-GFP ( #62541; BDSC ) ; UAS-CD8-RFP ( #27392 , BDSC ) ; cut-intron2-Gal4 ( #49818 , BDSC ) ; NRE-GFP ( #30727; BDSC ) ; UAS-Sgg . S9A ( #5255; BDSC ) ; UAS-EcR-DN ( #6872; BDSC ) ; UAS-wg-RNAi ( #33902; BDSC ) ; UAS-Br-Z1 ( #51190; BDSC ) ; UAS-Br-Z3 ( #51192; BDSC ) ; UAS-Br-Z4 ( #51193; BDSC ) ; FRT19A , dsh3 ( #6331; BDSC ) ; UAS-TCF ( #4838; BDSC ) ; UAS-Nipped-B-RNAi ( #32406; BDSC ) ; UAS-SA-RNAi ( #33395; BDSC ) ; UAS-Eip74EF-RNAi ( #29353; BDSC ) ; UAS-Eip75B ( #26717; BDSC ) ; UAS-white-RNAi ( #33623; BDSC ) ; Act5C-FRT>stop>FRT-lacZ . nls ( #6355; BDSC ) . wg-Gal4 transgenic flies were generated by integrating the Gal4 containing plasmid RIVGal4 ( Baena-Lopez et al . , 2013 ) into the attP site of wg flies ( Alexandre et al . , 2014 ) . To generate Fz3-pH-Stinger and cut-intron2-enhancer-pH-Stinger constructs , a 2 , 318 bp genomic DNA fragment ( −2324 to −7 ) from frizzled3 gene region and a 3 , 157 bp genomic DNA fragment from cut gene region were PCR amplified and inserted into the pH-Stinger vector respectively . Transgenic lines of these reporters were generated by site-specific integration into the attP2 landing site on the third chromosome using standard phiC31 transformation methods . All transgenic plasmids were verified by DNA sequencing before germline transformation . The PCR primers are listed as below: fz three reporter Fw: 5’-GAACGAAAGAGTTGGCAGAGAG−3’; fz three reporter Rv: 5’-GCTTAGTGGGTTTCAGGAGG−3’; ct reporter Fw: 5’-GCCGAGATGCGGTAGTAAAACG−3’; ct reporter Rv: 5’-CTGTTTGTTTCTGGCGAGCTTA −3’ . To generate UAS-FLAG-Br-Z2-FL; UAS-FLAG-Br-Z2-ΔBTB; UAS-FLAG-Br-Z2-Δ5AA or UAS-FLAG-Br-Z2-ΔZF transgenic lines , full length or truncated version of FLAG-Br-Z2 was inserted into pUAST vector , followed by phiC31-mediated integration into the attP2 landing site . For Coimmunoprecipitation experiments , Arm-HA in pcDNA3 . 1 ( Invitrogen ) was made by cloning Arm-S10 from genome DNA extracted from UAS-Arm-S10 transgenic fly line ( #4782; BDSC ) , followed by replacing Myc tag with HA tag ( YPYDVPDYA ) . FLAG-tagged Br isoforms in pcDNA3 . 1 were constructed by cloning cDNA of each isoform from corresponding br isoform-specific transgenic lines using genomic DNA PCR . aa 191–514; aa 1–190; aa 32–97 , aa 439–492 and aa 46–50 of Br-Z2 were deleted to make Br-Z2-N , Br-Z2-C , Br-Z2-ΔBTB , Br-Z2-ΔZF and Br-Z2-Δ5AA respectively . pCMV-Myc-TCF is a generous gift from Dr . Esther M Verheyen . To generate MARCM clones shown in Figure 1H–I’ , Figure 3I , J or Figure 4—figure supplement 1E , larvae were heat-shocked at 37°C for 6 times for 40 min each time successively at 24 , 28 , 48 , 52 , 72 , 76 hr after-larvae-hatching ( ALH ) and further aged at 25°C before dissection at early adult stage ( 2–3 days after eclosion ) . To generate MARCM clones shown in Figure 3G , H , larvae were heat-shocked at 37°C for 6 times for 60 min each time successively at 24 , 28 , 48 , 52 , 72 , 76 hr after-larvae-hatching ( ALH ) and further aged at 25°C before dissection at 72 hr APF . To generate MARCM clones shown in Figure 3M , third instar larvae were heat-shocked at 37°C for 40 min and farther aged at 25°C before dissection at 96 hr APF . The lineage tracing experiments as shown in Figure 3A–F’ were performed by crossing the esg-Gal4 , Gal80ts , UAS-CD8-GFP; UAS-w-RNAi or esg-Gal4 , Gal80ts , UAS-CD8-GFP; UAS-cut-RNAi flies with UAS-FLP; Act5C-FRT>stop>FRT-lacZ . nls flies . Embryos were collected and kept at 18°C . Late third instar larvae were shifted to 29°C until dissection at early adult stage ( 2–3 days after eclosion ) . For intestine-renal tubule immunostaining , samples were dissected in Schneider’s insect medium ( Sigma-Aldrich ) and proceeded as previously described ( Lin et al . , 2008 ) with modifications . Briefly , samples were fixed in 4% formaldehyde/1xPBS/n-heptane ( v/v/v = 1:1:2 ) for 15 min at room temperature , washed with 100% methanol and gradually rehydrated in 75 , 50 and 25% PBST ( 1xPBS plus 0 . 1% Triton X-100 ) /methanol mix . Samples were washed several times with PBST , blocked in blocking solution ( 1% BSA in PBST ) for 1 hr , followed by incubating with appropriate primary antibody overnight at 4°C . After incubation with secondary antibodies according to standard procedures , samples were mounted in Vectashield ( Vector Laboratories ) . For DNA staining , Hoechst ( Life Technologies ) was added in the wash step with a dilution of 1:3000 . Images were obtained on a Leica TCS SP8 AOBS confocal microscope and were processed with Adobe Photoshop . Primary antibodies used in this study were chicken anti-GFP ( 1:2000 , Abcam ) , mouse anti-Pros ( 1:100 , Developmental Studies Hybridoma Bank [DSHB] ) , mouse anti-Cut ( 2B10 ) ( 1:100 , DSHB ) , mouse anti-Broad-core ( 25E9 ) ( 1:200 , DSHB ) , mouse anti-Nc82 ( 1:100 , DSHB ) , mouse anti-AstA ( 1:100 , DSHB ) , rabbit anti-Pdm1 ( 1:1000 , a generous gift from Dr . Xiaohang Yang ) , mouse anti-β-galactosidase ( 1:100 , DSHB ) , rabbit anti-β-galactosidase ( 1:1000 , Cappel ) , rabbit anti-phospho-Histone H3 ( 1:1000 , Upstate ) . Human embryonic kidney HEK293T cells ( ATCC , RRID: CRL-3216 obtained from Dr . Hong Wu’s laboratory , Peking University , and authenticated by ATCC ) were maintained in DMEM medium ( Invitrogen ) supplemented with 10% FBS at 37 ˚C and 5% CO2 . DNA transfection was performed using a standard polyethylenimine ( PEI ) protocol . Drosophila S2 ( Schneider 2 ) cells ( DGRC , Cat#: S2-DGRC; obtained from Dr . Alan Jian Zhu’s laboratory , Peking University , and authenticated by DGRC ) were cultured at 25°C in Schneider’s Drosophila medium Drosophila Medium ( Invitrogen ) supplemented with 10% FBS , 100 U/ml penicillin and 100 mg/ml streptomycin . DNA transfection of S2 cells were carried out using Effectene Transfection Reagent ( QIAGEN ) according to manufacturer’s protocol . Both cell lines used in this study have been tested for and confirmed to be negative for mycoplasma contamination , using short tandem repeat ( STR ) profiling technique . Coimmunoprecipitation assays in HEK293T cell extracts were performed as previously described ( Liu et al . , 2017 ) . Briefly , 48 hr after transfection , HEK293T cells were harvested , washed and resuspended in lysis buffer [50 mM Tris-HCl ( pH 8 . 0 ) ; 120 mM NaCl; 5 mM EDTA; 1% NP-40; 10% glycerol; protease inhibitor cocktail ( Sigma-Aldrich ) ; 2 mM Na3VO4] and kept on ice for 20 min . Cell extracts were sonicated with Bioruptor Plus ( Biosense ) at 4°C with low power for 5 cycles of 10 s on/10 s off . The cell extracts were clarified by centrifugation , and proteins immobilized by binding to anti-FLAG M2 ( Sigma-Aldrich ) affinity gel for 4 hr at 4°C . Beads were washed and proteins recovered directly in SDS-PAGE sample buffer . Rabbit anti-Flag ( Cell Signaling Technology ) , Rabbit anti-c-myc ( Cell Signaling Technology ) or rabbit anti-HA ( Santa Cruz Biotechnologies ) were used for western blot analysis . Isoform-specific rabbit anti-Br-Z2 antibody was generated in this study [GST fusion of Br-Z2 aa 432–514 , affinity purified ( Abclonal Biotech . ) ] and used at 1:40 for immunostaining . The DNA-binding domains ( DBD ) of TCF ( aa 271–408 ) and Br-Z2 ( aa 432–514 ) were cloned into pGEX-6P-1 . GST-tagged protein was purified by ProteinIso GST resin ( Transgen Biotech ) through column buffer ( 25 mM HEPES ( pH 7 . 6 ) , 150 mM NaCl , 10% glycerol , 1 mM EDTA , 10 mM β-mercaptoethanol , 2 mM PMSF ) . After washing , protein was eluted with 100 mM glutathione in column buffer . Protein concentrations were measured by Coomassie stained gels . Known concentrations of BSA ( bovine serum albumin ) were used as a standard . Increasing concentration of purified GST-tagged protein and 10 fmol biotin-labelled double-stranded DNA substrate were incubated at 25°C in 20 μl reaction buffer ( 20 mM HEPES ( pH 7 . 9 ) , 50 mM KCl , 0 . 1 mM EDTA , 2 mM DTT , 6 mM MgCl2 , 0 . 1 mg/ml BSA , 50 ug/ml poly ( dI-dC ) , 5% glycerol ) for 45 min . The reaction mixture was loaded and resolved in 8% TBE gel . Amounts of recombinant protein used per reaction were as follows: 1 . 2–7 . 2 ug GST-TCF-DBD ( E probe ) ; 0 . 01–0 . 2 ug GST-Br-Z2-DBD ( P probe ) and 0 . 2–1 ug GST-Br-Z2-DBD ( E1 probe ) . Duolink in situ PLA was performed with Drosophila S2 cells according to manufacturer’s instructions ( DUO92101; Sigma-Aldrich ) . Briefly , after transfection and fixation , S2 cells were incubated with primary antibodies at RT for 3 hr , followed by incubation with Duolink PLA probes ( 1:12 ) at 37°C for 1 hr , ligation at 37°C for 1 hr and amplification at 37°C for 2 hr . Primary antibodies used were rabbit anti-Myc ( 1:200 ) and mouse-anti-Br-core ( 1:150 ) . Length of ureter was measured with Leica Application Suite 2 . 6 . 3 from Leica Microsystems . For quantification of the intensity of antibody staining , images were taken with the same confocal settings and the mean fluorescence intensity was measured with Histogram function of Adobe Photoshop . Unpaired Student’s t-tests were used for statistical analysis between two groups . | As an embryo develops , an organism transforms from a single cell into an organized collection of different cells , tissues and organs . Regulated by genes and messenger molecules , non-specialized cells known as precursor cells , move , divide and adapt to produce the different cells in the adult body . However , sometimes already-specialized adult cells can acquire a new role in a process known as lineage reprogramming . Finding ways to artificially induce and control lineage reprogramming could be useful in regenerative medicine . This would allow cells to be reprogrammed to replace those that are lost or damaged . So far , scientists have been unable to develop a clear view of how lineage reprogramming happens naturally . Here , Xu et al . identified a cell-conversion event in the developing fruit fly . As the fly larva develops into an adult , a group of cells in the midgut reprogramme to become renal cells – the equivalent to human kidney cells . The experiments revealed that a combination of signals from a cell messenger system important for cell specialization ( called Wnt ) and the hormone that controls molting in insects , activate a gene called cut , which controls the midgut-to-renal lineage reprogramming . Together , Wnt and the hormone ensure that cut is activated only in a small , specific group of midgut precursor cells at a precise time . The reprogrammed cells then move into the excretory organs , the renal tubes , where they give rise to renal cells . Midgut precursor cells in which cut had been experimentally removed , still traveled into the renal tubes . However , they failed to switch their identity and gave rise to midgut cells instead . Further examination revealed that both Wnt and the ecdysone hormone are needed to activate the cut gene . This is probably achieved by creating loops in the DNA to bring together the two distantly located key regulatory elements of cut gene expression . If this mechanism can be seen in other contexts it may be possible to adapt it for medical purposes . The ability to reprogramme groups of cells with high specificity could transform medicine . It would make it easier for our bodies to regenerate and repair . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2018 | Temporospatial induction of homeodomain gene cut dictates natural lineage reprogramming |
The apical inflammatory cytokine TNF regulates numerous important biological processes including inflammation and cell death , and drives inflammatory diseases . TNF secretion requires TACE ( also called ADAM17 ) , which cleaves TNF from its transmembrane tether . The trafficking of TACE to the cell surface , and stimulation of its proteolytic activity , depends on membrane proteins , called iRhoms . To delineate how the TNF/TACE/iRhom axis is regulated , we performed an immunoprecipitation/mass spectrometry screen to identify iRhom-binding proteins . This identified a novel protein , that we name iTAP ( iRhom Tail-Associated Protein ) that binds to iRhoms , enhancing the cell surface stability of iRhoms and TACE , preventing their degradation in lysosomes . Depleting iTAP in primary human macrophages profoundly impaired TNF production and tissues from iTAP KO mice exhibit a pronounced depletion in active TACE levels . Our work identifies iTAP as a physiological regulator of TNF signalling and a novel target for the control of inflammation .
The cytokine TNF controls numerous important biological processes ( e . g . inflammation , fever , apoptosis , necroptosis , cachexia , tumorigenesis , viral infection , insulin signaling ) and is heavily implicated in inflammatory disease ( Brenner et al . , 2015 ) . Anti-TNF biologics are the highest-selling drugs internationally and there is intense interest in how TNF signaling is regulated ( Brenner et al . , 2015 ) . TNF is expressed by a range of cells including macrophages , lymphocytes , natural killer cells , endothelial cells and microglia and is synthesized as a type II transmembrane protein with a cytosolic domain of 76 amino acids that assembles into a trimer ( Locksley et al . , 2001 ) . The capacity of TNF to trigger such pleiotropic biological outcomes is determined by its ability to activate two distinct receptors ( Locksley et al . , 2001 ) . Generally , TNFRI activation is associated with induction of acute or chronic inflammatory responses , or cell death , whereas TNFRII mediates pro-survival signals and has been associated with the tolerogenic properties of regulatory T cells ( Kleijwegt et al . , 2010; Richter et al . , 2012 ) . Additional complexity is imposed by the form of TNF that engages in signalling . The transmembrane form of TNF ( mTNF ) triggers juxtacrine signalling , while TNF released as a soluble form ( sTNF ) , drives paracrine signalling . Notably , whereas TNFRI can be activated by both soluble and membrane TNF , TNFRII is activated efficiently only by mTNF ( Grell et al . , 1995 ) . Whereas mTNF is sufficient for the development and maintenance of some lymphoid tissues , soluble TNF is important for acute and chronic inflammation . Specifically , mice unable to produce soluble TNF resist endotoxic shock and exhibit reduced severity in experimental autoimmune encephalomyelitis models ( Ruuls et al . , 2001 ) . Membrane TNF is also insufficient to rescue the defects in primary B cell follicle formation observed in TNF KO mice , but mediates protective immune responses to intracellular bacterial infection ( Ruuls et al . , 2001; Alexopoulou et al . , 2006 ) . The ability to engage biological outcomes that require TNFRI versus TNFRII , ( and the capacity to control the physical distance over which signaling is effective ) therefore , critically depends on the ability to release soluble TNF from the cell surface . This is catalyzed by the protease TACE ( TNF α Converting Enzyme ) ( Horiuchi et al . , 2007; Peschon et al . , 1998 ) , also called ADAM17 ( A Disintegrin And Metalloprotease ) ( Gooz , 2010; Zunke and Rose-John , 2017 ) . Crucially , TACE imposes an additional layer of versatility and regulation to TNF signaling , since in addition to cleaving TNF , both TNFRs are also physiological TACE substrates . Hence , TACE is a master orchestrator of TNF signaling , tuning signaling to fit a panoply of biological roles ranging from inflammatory responses to immune tolerance . TACE also has significant biological importance beyond TNF signaling since it cleaves other prominent substrates , including the activating ligands of the EGFR ( Epidermal Growth Factor Receptor ) , an important pathway that drives growth control , tissue repair and immune responses . Given its ability to elicit potent biological responses , it is unsurprising that TACE is stringently regulated ( Murphy , 2009; Grötzinger et al . , 2017 ) . A major control point in TACE regulation involves its trafficking within the secretory pathway ( Schlöndorff et al . , 2000 ) . TACE is synthesized in the endoplasmic reticulum ( ER ) as a catalytically inactive precursor . For TACE to become proteolytically active , it must undergo a maturation step—removal of its prodomain—which is catalysed by proprotein convertases in the trans-Golgi ( Schlöndorff et al . , 2000 ) . The exit of TACE from the ER and its trafficking to the cell surface requires regulatory proteins called iRhoms ( Adrain et al . , 2012; McIlwain et al . , 2012; Li et al . , 2015 ) . Hence , iRhom KO mice , or cells in which iRhoms are ablated , lack TACE activity ( Adrain et al . , 2012; Li et al . , 2015; Christova et al . , 2013 ) . Mice null for iRhom2 , whose expression is enriched in myeloid cells , cannot secrete TNF ( Adrain et al . , 2012; McIlwain et al . , 2012; Siggs et al . , 2012 ) . An important checkpoint to license TACE activity involves its stimulation on the cell surface by agents including phorbol esters , Toll-like receptor agonists and G-protein coupled receptor ligands ( Grötzinger et al . , 2017; Arribas et al . , 1996; Hall and Blobel , 2012; Brandl et al . , 2010; Wetzker and Böhmer , 2003 ) . Importantly , as well as controlling TACE trafficking , iRhoms exist in a molecular assembly with TACE on the cell surface—the ‘sheddase complex’ which is central to stimulation of TACE sheddase activity . Within the sheddase complex , iRhom proteins serve as a platform that senses and transduces TACE-activating stimuli . These agents provoke the MAP kinase-dependent phosphorylation of the iRhom2 cytoplasmic tail , which in turn triggers the recruitment of 14-3-3 proteins . This enforces the detachment of TACE from iRhom2 ( or triggers a conformational change within the sheddase complex ) which is required to facilitate TACE’s ability to cleave its substrates , including TNF ( Grieve et al . , 2017; Cavadas et al . , 2017 ) . Hence , iRhoms are allosteric regulators of TACE’s proteolytic activity as well as acting as trafficking factors . As iRhom and TACE form an intrinsic complex , their trafficking itinerary and fate within the secretory pathway must presumably be interdependent . In spite of the importance of trafficking for TACE regulation ( Schlöndorff et al . , 2000; Adrain et al . , 2012; Dombernowsky et al . , 2015 ) , surprisingly little is known about the machinery that controls TACE , or iRhom , trafficking to/from the plasma membrane . An exception is PACS-2 ( Phosphofurin Acidic Cluster Sorting Protein 2 ) , a protein that colocalizes with mature TACE in endocytic compartments ( Dombernowsky et al . , 2015 ) and controls its endocytic recycling . Ablation of PACS-2 in cells impairs the cell surface availability of TACE , reducing substrate cleavage ( Dombernowsky et al . , 2015 ) . However , PACS-2 has a relatively modest impact on TACE biology in vivo ( Dombernowsky et al . , 2017 ) , suggesting the possibility of unidentified trafficking regulators that may act separately from , or redundantly with , PACS-2 . As iRhoms form functionally important complexes with cell surface TACE ( Grieve et al . , 2017; Cavadas et al . , 2017; Maney et al . , 2015 ) , modulation of iRhom trafficking in the endocytic pathway has the potential to act as a regulatory mechanism that controls TNF secretion . It has been shown that not only TACE ( Doedens and Black , 2000; Lorenzen et al . , 2016 ) , but also iRhoms ( Grieve et al . , 2017; Cavadas et al . , 2017 ) are endocytosed and degraded in lysosomes , but the machinery involved in maintaining stable cell surface levels of the sheddase complex is unknown . Here we identify a novel protein that we name iTAP ( iRhom Tail-Associated Protein ) that is essential for the control of the stability of iRhom2 and TACE on the plasma membrane . Ablation of iTAP triggers the mis-sorting of iRhom2 , and consequently , TACE , to lysosomes , where they are degraded . Consistent with this , loss of iTAP results in a dramatic reduction in TACE activity and TNF secretion . Our work reveals iTAP as a key physiological regulator of TNF release .
To identify novel regulators of mammalian iRhoms 1 and −2 , we adopted an immunoprecipitation/mass spectrometry ( IP/MS ) approach described in our previous work ( Cavadas et al . , 2017 ) . As shown in Figure 1A , we generated a panel of HEK 293ET cell lines stably expressing HA-tagged forms of full-length iRhom1 , iRhom2 , or the iRhom1 N-terminal cytoplasmic tail only . To focus only on proteins that bind selectively to iRhoms , we included the related rhomboid-like proteins , Rhbdd2 , RHBDD3 , Ubac2 , as specificity controls ( Figure 1A ) . As expected , only immunoprecipitates ( IPs ) from cells expressing full-length HA-tagged iRhom1 or iRhom2 captured endogenous TACE , confirming the validity of the approach ( Figure 1B ) . To identify novel iRhom-binding proteins , we subjected IPs from these cells to mass spectrometry . This analysis revealed , in multiple replicate experiments , peptides from a largely uncharacterized protein , called FRMD8 ( FERM Domain-containing protein 8 ) , in IPs of iRhom1 or iRhom2 , but not in control IPs of Rhbdd2 , RHBDD3 and Ubac2 ( Figure 1C ) . Furthermore , FRMD8 was found in IPs from cells expressing the N-terminus of iRhom1 ( Figure 1A , C ) , suggesting that it was recruited to the iRhom cytoplasmic tail ( R-domain ) , an important regulatory region ( Grieve et al . , 2017; Cavadas et al . , 2017; Maney et al . , 2015; Hosur et al . , 2014 ) . In light of this , we named the novel protein iTAP ( ‘iRhom Tail Associated Protein’ ) . A closer inspection of the iTAP sequence revealed that it encodes a FERM ( band 4 . 1/Ezrin/Radixin/Moesin ) domain ( Chishti et al . , 1998 ) ( Figure 1D ) . Proteins containing FERM domains fulfil many important roles , including signaling , organization of the cell cortex and its mechanical properties and cell surface stabilization ( anchoring ) of membrane proteins or phospholipids ( McClatchey , 2014; Fehon et al . , 2010; Hoover and Bryant , 2000; Baines et al . , 2014; Moleirinho et al . , 2013 ) . The well-characterized FERM domain contains three distinct lobes that together resemble a three-leaf clover ( Pearson et al . , 2000 ) . However , unlike most FERM domain-containing proteins , but similar to its paralog KRIT1—an adaptor protein in the cerebral cavernous malformation pathway that regulates the establishment of vasculature ( Pal et al . , 2017 ) , iTAP contains only the central ( FERM-M ) lobe ( Figure 1D ) . iTAP orthologs are present in metazoans , including Drosophila and Danio ( Kategaya et al . , 2009 ) . The iTAP protein is expressed broadly and is co-expressed with TACE and iRhom1 or iRhom2 in a range of tissues relevant for TACE biology ( Figure 1—figure supplement 1A , B ) . Independent immunoprecipitation experiments verified that iTAP binds specifically to both iRhom1 and iRhom2 , but not to the related rhomboid pseudoproteases Ubac2 and Rhbdd2 ( Figure 2A ) . iTAP binds the cytoplasmic tail of iRhoms , since a mutant containing only the cytoplasmic tail of iRhom1 bound iTAP robustly , whereas a mutant lacking all of the iRhom2 cytoplasmic tail ( ∆Nterm ) failed to bind ( Figure 2A ) . By contrast , removal of the iRhom homology domain ( IRHD ) , the luminal globular domain between transmembrane helices 1 and 2 ( Figure 2B ) from iRhom2 had no impact on iTAP recruitment ( Figure 2A ) . These data indicate that iTAP is specifically recruited to the cytoplasmic tail of iRhoms . Notably , when iTAP-FLAG , but not a panel of control proteins , was immunoprecipitated from cell lysates , we detected the binding of endogenous iRhom1 and iRhom2 to iTAP ( Table 1 ) . These data confirm that iTAP is a specific endogenous interactor of both iRhom paralogs in mammals . To delineate the region within the cytoplasmic tail of iRhom that iTAP binds to , we divided the cytoplasmic tail into subdomains ( Figure 2B ) and , initially , created a series of sequential truncations within the iRhom2 cytoplasmic tail ( Figure 2C ) . Then , we created more focussed deletions within the area identified in the previous experiment ( Figure 2D ) . This revealed that amino acids 191–271 of the iRhom2 tail contain the main determinant for iTAP binding ( Figure 2C , D ) . Further studies are required to assess whether the specific binding of iTAP to iRhom is direct , or via an intermediary . We next examined the cellular localization of GFP-tagged iTAP in fixed and permeabilized mammalian cells . As shown in Figure 2—figure supplement 1A , iTAP-GFP exhibited a powdery staining in the cytoplasm and nucleus . When co-expressed with mCherry-tagged iRhom2 , iTAP was recruited to areas of iRhom2 staining ( Figure 2—figure supplement 1B ) . To determine the functional importance of iTAP binding to iRhom , we used CRISPR to ablate iTAP in HEK 293ET cells , which was confirmed by the lack of iTAP protein expression ( Figure 3A ) . As TACE trafficking and cell surface stimulation depends on iRhoms , we examined the ability of WT versus iTAP-null cells to support release of TACE substrates . Notably , the PMA-induced shedding of a panel of chimeric alkaline phosphatase ( AP ) TACE substrates ( Sahin et al . , 2004 ) , including EGFR ligands and TNF , was substantially impaired in iTAP KO cells ( Figure 3B ) . This shedding defect was rescued by the expression of an iTAP cDNA , confirming that the loss of iTAP was directly responsible for defective TACE activity ( Figure 3C , D ) . To test the hypothesis that the basis for the shedding defects was in fact reduced TACE proteolytic activity in iTAP KO cells , we assayed TACE enzymatic activity directly using a peptide substrate ( Figure 3—figure supplement 1 ) . As expected , TACE immunoprecipitates ( Figure 3—figure supplement 1A ) from iTAP KO cells exhibited substantially depleted levels of TACE activity ( Figure 3—figure supplement 1B ) , confirming that the loss of iTAP specifically impairs TACE rather than affecting its substrates . Our previous studies have shown that iRhom proteins are highly specific regulators of TACE that do not affect the trafficking/activity of related proteases in the ADAM metalloprotease family , including ADAM10 , the closest relative of TACE ( Adrain et al . , 2012; Christova et al . , 2013 ) . To examine whether iTAP was similarly dedicated specifically to the iRhom/TACE pathway we examined whether cleavage of the EGFR ligands EGF and BTC , which are cleaved specifically by ADAM10 ( Sahin et al . , 2004 ) , was affected by loss of iTAP . Notably the cleavage of these ADAM10 substrates was unaffected ( Figure 3E ) while the release of a model secreted substrate was similarly unimpaired in iTAP KO cells ( Figure 3F ) . These data confirm that iTAP is a highly specific regulator of the TACE pathway that does not affect secretion in general . To investigate how loss of iTAP affected TACE so profoundly , we examined the maturation status of TACE , a readout for its trafficking and activation status ( Adrain et al . , 2012 ) in iTAP KO cell lines ( Figure 4A ) . As a positive control , we included lysates from iRhom1/iRhom2 double knockout MEFs which completely lack mature TACE ( Christova et al . , 2013 ) . Although a few experiments showed an overall reduction in the TACE levels in iTAP KO cells , the consistent and most pronounced phenotype , found in all iTAP-null cell lines , was a dramatic depletion of mature TACE , identified by its faster migration pattern ( Figure 4B ) . As TACE is heavily glycosylated , to confirm this observation more clearly , we treated lysates with the deglycosylating enzymes Endoglycosidase-H ( Endo-H ) , which removes high mannose N-linked glycans added in the ER , but not complex N-linked glycans found in the later secretory pathway , versus PNGase F , which deglycosylates both ( Figure 4C , D ) . This confirmed that iTAP KO cell lines were substantially depleted of mature TACE ( Figure 4B , D , E ) , which could be rescued specifically by iTAP overexpression in iTAP KO cells ( Figure 4E ) . Overexpression of iTAP in WT cells also modestly enhanced mature TACE levels ( Figure 4E ) and densitometric analysis confirmed once again that the loss , or reintroduction , of iTAP most profoundly affected mature TACE levels ( Figure 4E , graphs ) . A clear prediction from these experiments is that iTAP-null cells should lack mature , cell surface TACE , explaining the basis of the proteolytic defects observed ( Figure 3 , Figure 3—figure supplement 1A , B ) . We tested this hypothesis in experiments with a non-cell permeable biotinylated cross-linker , which revealed drastically reduced mature TACE levels on the cell surface ( Figure 4F ) . Finally , as predicted by stringent specificity of iTAP for TACE , the maturation of other related ADAM proteases was unimpaired ( Figure 4G ) . Together these data confirm that iTAP is a dedicated regulator of the iRhom/TACE sheddase complex . The observation that iTAP-null cells exhibited drastically depleted mature TACE levels could be explained by two potential mechanisms . First , loss of iTAP , which binds to iRhom2 ( Figures 1–2; Table 1 ) , could impair ER exit of the iRhom/TACE complex , causing a failure in TACE maturation , as observed in iRhom KO cells ( Adrain et al . , 2012; Christova et al . , 2013 ) . Alternatively , as TACE undergoes constitutive recycling from the plasma membrane ( Dombernowsky et al . , 2015 ) and iRhom2 traffics to the cell surface and enters the endolysosomal pathway ( Grieve et al . , 2017; Cavadas et al . , 2017; Maney et al . , 2015 ) , iTAP could stabilize iRhom/TACE complexes on the plasma membrane or within the endocytic pathway . Given the established role of FERM domain proteins in stabilizing proteins on the cell cortex , this second possibility seemed plausible . To investigate the impact of iTAP on iRhom2 stability , we first used RAW 264 . 7 cells . These macrophage-like cells express high levels of endogenous iRhom2 , making its detection more feasible than in HEK 293ET cells or MEFs . Strikingly , endogenous iRhom2 was depleted in iTAP KO RAW 264 . 7 cells ( Figure 5A ) , indicating that iTAP is essential to maintain iRhom2 stability . Consistent with this , in HEK 293ET cells , iTAP transient overexpression increased steady state levels of overexpressed iRhom2-HA and enhanced the half-life ( see graph , Figure 5B ) of the protein , during a timecourse of cycloheximide ( CHX ) treatment , used to block additional protein synthesis ( Figure 5B ) . This experiment , in which iRhom2 was expressed from an artificial promoter , indicates that the impact of iTAP on iRhom2 levels is independent of transcription . As anticipated by these results , transiently overexpressed iRhom2-HA was also destabilized in iTAP KO cells ( Figure 5—figure supplement 1A ) . Consistent with the ability of iTAP to impact profoundly on iRhom2 stability , we observed striking colocalization between mCherry-iRhom2 and iTAP-GFP , when they were co-expressed in HeLa cells , as judged by Pearson’s correlation and Manders’ overlap coefficients ( Figure 5C ) . These colocalization data indicate that iRhom and iTAP interact in multiple compartments ( Figure 5C ) , including the ER and plasma membrane . Two major possibilities exist to explain the decreased half-life of iRhom2 in iTAP KO cells: in the absence of iTAP , iRhom2 may be degraded in the early secretory pathway ( the ER ) , or in the late secretory pathway ( lysosomes ) . To address this , we used Endo-H deglycosylation to discriminate between the impact upon iTAP overexpression on the Endo-H-sensitive pool of iRhom2 in the ER , versus the Endo-H-insensitive fraction that has entered the later secretory pathway . As anticipated ( Zettl et al . , 2011 ) , most overexpressed iRhom2 is Endo-H sensitive , hence still located within the early secretory pathway ( Figure 5D ) . Strikingly , the co-expression of iTAP increased the overall levels of iRhom2 , but selectively enriched the post-ER fraction of iRhom2 ( Figure 5D ) , suggesting a disproportionate impact on the form of iRhom2 that had traversed to the late secretory pathway . To obtain additional insights , we used a binding assay , coupled to deglycosylation analysis , to compare which species of iRhom2 bind to iTAP . Cells were treated with or without DSP , a thiol-reducible cell-permeable cross-linker , to covalently trap complexes in situ ( Adrain et al . , 2012 ) , enabling us to discriminate between interaction in vivo , compared to potential adventitious binding post-lysis . After immunoprecipitation , samples were treated with DTT to reverse the cross-linking . Notably , compared to IPs done without cross-linking , iTAP IPs from cross-linked cells showed a clear enrichment for post-ER ( Endo-H insensitive ) iRhom2 ( Figure 5E ) , although the ER-localized form of iRhom2 was also readily detected . Taken together with the colocalization data ( Figure 5C ) , we propose that the loading of iTAP onto the sheddase complex occurs already in the ER but the binding is sustained throughout the sheddase complex’s itinerary in the late secretory pathway , where iTAP’s affinity for iRhom2 appears to be higher . These data are consistent with the finding that iTAP selectively affects mature TACE ( Figure 4 ) . Finally , ruling out a requirement for iTAP in controlling the ER-to-Golgi trafficking of iRhom2 , we found that the ER exit of iRhom2 was unimpaired in iTAP KO cells , judged by the presence of Endo-H-resistant iRhom2 ( Figure 5F; Figure 5—figure supplement 1B ) . Hence , although iTAP binds to the sheddase complex in the early secretory pathway , its impact appears to be more decisive in the late secretory pathway . As ablation of iTAP results in dramatically reduced iRhom2 levels ( Figure 5A ) and iTAP increases the abundance of post-ER iRhom2 ( Figure 5D , E ) , we hypothesized that iTAP controls the stability of iRhom2 in the late secretory pathway . Consistent with this premise , cell surface biotinylation experiments revealed that iTAP expression increased the steady state levels of cell surface iRhom2 , and prolonged its cell surface stability , when CHX was used to block additional protein synthesis ( Figure 5G ) . In further agreement , the co-expression of iTAP with GFP-iRhom2 in MEFs enhanced the amount of GFP-iRhom2 detected on the plasma membrane ( Figure 5H ) , supporting the premise that iTAP promotes the cell surface stability of iRhom2 . Together , our data indicate that iTAP’s primary function is to stabilize the sheddase complex on the cell surface , reconciling the pronounced loss of mature TACE in iTAP KO cells and the increased binding propensity of iTAP for post-ER iRhom2 . We next addressed the functional basis for the pronounced loss of iRhom2 and mature TACE in iTAP KO cells . To test the hypothesis that loss of iTAP triggers the degradation of iRhom2 and TACE in lysosomes , the major degradative compartment in the late secretory pathway , we examined the localization of mCherry-iRhom2 in WT or iTAP KO HeLa cells derived by CRISPR ( Figure 6—figure supplement 1 ) . As shown in Figure 6A , in WT HeLa cells , mCherry-iRhom2 did not co-stain appreciably with lysosomes . By sharp contrast , iTAP ablation resulted in a pronounced co-localization of iRhom2 with the lysosomal marker LAMP2 ( Lysosomal-Associated Membrane Protein 2 ) , indicating mis-sorting of iRhom2 to lysosomes . This phenotype was specific since the co-transfection of iTAP-GFP into iTAP KO HeLas rescued the aberrant accumulation of mCherry-iRhom2 in lysosomes ( Figure 6B ) , resulting in the marked co-localization of iRhom and iTAP observed previously ( Figure 2—figure supplement 1B; Figure 5C ) . Consistent with these observations , a panel of lysosomotropic drugs that inhibit lysosomal proteolysis by impairing lysosomal acidification , rescued iRhom2 stability in iTAP KO cells ( Figure 6C ) . The rescue of mature TACE under similar conditions was more modest ( Figure 6D and data not shown ) , perhaps because of the slow trafficking time of TACE in the secretory pathway ( Schlöndorff et al . , 2000 ) and because iTAP acts directly on iRhoms . Consistent with the notion that iRhom2 can influence the routing of TACE into lysosomes , we found that overexpressed TACE-GFP was only recruited into lysosomes when sufficient iRhom2 was co-overexpressed ( Figure 6E ) . In conclusion , our data reveal that when the normal stoichiometric ratio of iTAP to iRhom2 is disrupted ( e . g . upon iRhom2 overexpression or iTAP ablation ) , iRhom2 is mis-sorted into the lysosome , then degraded . This highlights an important physiological role for iTAP in maintaining the cell surface stability of the iRhom2/TACE sheddase complex . iTAP is expressed in a range of mouse tissues relevant to iRhom and TACE biology ( Peschon et al . , 1998; Li et al . , 2015 ) ( Figure 1—figure supplement 1 ) . As the experiments conducted so far focused on transformed cell lines , we next examined the physiological importance of iTAP at the organismal level . To examine the role of iTAP in mice , we generated a mutant in which the first coding exon ( exon 2 ) of the Frmd8 ( iTAP ) gene was deleted by CRISPR ( Figure 7A , Figure 7—figure supplement 1 ) . MEFs isolated from iTAP KO embryos lacked iTAP protein expression , confirming the successful targeting of the Frmd8 gene ( Figure 7B ) . As anticipated , MEFs from two independent iTAP KO embryos exhibited the characteristic pronounced depletion of mature TACE levels ( Figure 7C ) . Focussing next on potential phenotypes of the iTAP–null mouse mutants themselves , we harvested tissues from iTAP KO mice to assess the maturation status of TACE ( Figure 7D ) . Significantly , with the possible exception of skin , where iTAP loss may be mitigated by other molecules , we observed a substantial depletion in the relative proportion of mature TACE in a range of iTAP KO mouse tissues , and in primary macrophages isolated from the bone marrow of iTAP KOs . These data reinforce the notion that iTAP is an important physiological regulator of the iRhom/TACE/TNF axis in vivo , making it important to dissect fully , in future , the organismal role of iTAP . As the PMA-stimulated release of a chimeric alkaline phosphatase-tagged TNF was impaired in iTAP KO cells ( Figure 3B ) , we hypothesized that iTAP was an important physiological regulator of TNF secretion . To test this , we isolated primary monocytes from peripheral human blood , then induced the differentiation of these cells to primary human macrophages . Notably , the stimulated release of endogenous TNF in response to lipopolysaccharide in these cells was profoundly impaired , when iTAP expression was ablated by specific siRNAs ( Figure 7E ) . As expected , secretion of IL-6 and IL-8 , which is TACE-independent , was unaffected ( Figure 7E ) . Our data indicate that iTAP is an essential physiological regulator of TNF secretion in primary human macrophages , the principal source of secreted TNF in vivo .
Our work identifies iTAP as an important physiological regulator of the iRhom2/TACE sheddase complex , which is essential for the secretion of TNF and for a panoply of other substrates including ligands of the epidermal growth factor receptor . Our current and previous data ( Cavadas et al . , 2017 ) suggest that trafficking to , and degradation within , the lysosome is a default itinerary incurred by iRhom2 , and that iRhom2 potentially encodes the determinants that lead to the default trafficking of iRhom2 and TACE to the lysosome . We now show that iTAP is essential to stabilize iRhom2 on the cell surface , preventing the routing of the sheddase complex to the lysosome , and licensing TACE to cleave its substrates for signaling ( summarized in Figure 8 ) . iTAP , hence , emerges as an important regulator of inflammation and growth factor signaling , during development , normal physiology , infection and inflammatory disease . An obvious question concerns the extent to which the established features and roles of FERM domain proteins apply to iTAP and hence to the regulation of the iRhom/TACE pathway . A general theme is that FERM-domain proteins connect the cytoplasmic tails of cell surface client proteins to the cortical actin cytoskeleton to enhance their stability ( Hoover and Bryant , 2000; Baines et al . , 2014; Moleirinho et al . , 2013 ) . While iTAP binds to the cytoplasmic tails of iRhoms , which are found on the plasma membrane , our preliminary experiments failed to detect robust binding of iTAP to actin ( Figure 8—figure supplement 1A ) . Besides , we have not identified predicted actin binding motifs in the C-terminus of iTAP . Future experiments will be required to determine precisely how iTAP stabilizes iRhom and TACE in the late secretory pathway . Some FERM-domain proteins are implicated in endosomal sorting , the process whereby endocytosed proteins are sorted in early endosomes , for routing to the multi-vesicular body , lysosome , trans-Golgi network , recycling endosome or , alternatively , ‘fast’ recycling back to the cell surface ( Cullen , 2008 ) . Analogous to the degradation of iRhom2 in iTAP KO cells , loss of Snx17 , which binds to the cytoplasmic tail of β1 integrins , results in a failure in their endocytic recycling , leading to their degradation in lysosomes ( Böttcher et al . , 2012 ) . Notably , the iRhom2 cytoplasmic tail contains two motifs , NxxY and NPxY ( Figure 8—figure supplement 1B ) that are the consensus endocytic signals recognized by a subset of FERM-domain containing sorting nexins , involved in endocytic recycling . However , our preliminary experiments in which we have mutated those motifs to alanines ( AAAA ) , show that they appear not to be required for iTAP/iRhom2 binding ( Figure 8—figure supplement 1C ) . Moreover , although sorting nexins are intimately connected with the endocytic/recycling machinery , our preliminary experiments detect no obvious colocalization of iTAP with early endosomes ( Figure 8—figure supplement 1D ) . Endocytic sorting is however only one theme within the wider FERM biology . Future studies will be required to clarify the relationship between iTAP and the trafficking machinery , to map the vesicular itinerary taken by iRhom/TACE complexes , and to establish the precise basis of the mis-sorting defect in iTAP-null cells . It will also be interesting to reconcile the role of iTAP in the control of iRhom/TACE stability , versus that of PACS-2 , which binds directly to TACE ( Dombernowsky et al . , 2015 ) . iTAP and PACS-2 both impact on TACE stability , but iTAP can presumably only influence TACE stability indirectly via iRhoms . This is relevant because TACE stimulants trigger detachment of TACE from iRhom2 on the cell surface ( Grieve et al . , 2017 ) , a mechanism important for facilitating access of TACE to its substrates ( Cavadas et al . , 2017 ) . As iRhom and TACE are uncoupled at a crucial stage during signaling , their degradative fates could also be separated , leaving open the possibility that iTAP and PACS-2 may govern different stages in TACE’s trafficking lifecycle . The TNF ( and EGFR ) pathway ( s ) are very stringently regulated by positive and negative feedback ( Avraham and Yarden , 2011; Wallach , 2016; Vereecke et al . , 2009 ) . Considering the significant impact that iTAP has on TACE biology , it is tempting to speculate that feedback control over the signaling pathways that culminate in TNF or EGFR ligand release , could be governed by controlling the interaction between iTAP and iRhom , or by modulating the stability of iTAP itself . Our preliminary experiments suggest that the phosphorylation of iRhom2 at residues required for the stimulation of TACE activity ( Cavadas et al . , 2017 ) does not appear to influence iTAP binding ( data not shown ) , but future studies are required to investigate more widely the possibility of iTAP regulation by stimuli relevant to TACE biology . In addition to our extensive evidence indicating the requirement for iTAP for normal TACE function in human and mouse cells , we show that mature TACE levels are dramatically depleted in tissues from iTAP KO mice , including macrophages ( Figure 7D ) . This reinforces the notion that iTAP is an important physiological regulator of the sheddase complex at the organismal level , in multiple tissues . Notably , whereas ADAM17 homozygous mutant mice exhibit perinatal lethality ( Peschon et al . , 1998 ) , iTAP KO mice are born at near-normal mendelian ratios ( Table 2 ) , reach adulthood , appear superficially healthy and are fertile . These data indicate that although loss of iTAP indeed has a profound impact on the levels of mature TACE , the fraction remaining of mature TACE is sufficient to ensure normal mouse development . Several hypomorphic TACE mutant mice have been studied , including ADAM17ex/ex mice , which were generated by the insertion of a new exon containing an in-frame stop codon , flanked by weak splice donor/acceptor sites inside the Adam17 ( TACE ) locus ( Chalaris et al . , 2010 ) . 95% of the TACE mRNA produced contains the mutant exon , resulting in a dramatic reduction in TACE levels ( Chalaris et al . , 2010 ) . Notably , these animals are born at normal mendelian ratios but are highly susceptible when challenged to an experimental model of colitis ( Chalaris et al . , 2010 ) . This suggests that while traces of TACE can mitigate against lethality , they are not sufficient to prevent disease challenge . Hence , it will be important to dissect fully , in future , the organismal role of iTAP , particularly within the context of disease . Inhibiting TACE activity has been the subject of considerable pharmaceutical interest for decades , but attempts have failed , often because of cytotoxicity caused by unintended collateral targeting of ADAMs and matrix metalloproteases , that share active site architectures related to TACE ( Murumkar et al . , 2010 ) . As iTAP has no apparent impact on other ADAMs , the blockade of the iRhom:iTAP interaction may be an interesting potential therapeutic approach to attenuate TACE activity during disease . Such an approach would obviate the concern of collateral targeting of other metalloproteases . Although iTAP ablation at the cellular level has a potent impact on TACE substrate cleavage , at the organismal level the impact is significantly less severe than the lethal phenotype of TACE KO mice ( Peschon et al . , 1998 ) . This implies that it may be possible to target iTAP to reduce TACE activity sufficiently to achieve a therapeutic impact in diseased tissues , without impinging on the normal physiological roles of TACE , which are sustained with minimal TACE levels in ADAM17ex/ex mutants ( Chalaris et al . , 2010 ) and presumably our iTAP KO mice .
C-terminally triple HA-tagged versions of human iRhom1 , the cytoplasmic N-terminus of iRhom1 ( amino acids 1–404 ) , mouse iRhom2 , mouse Rhbdd2 , human RHBDD3 and mouse Ubac2 were cloned into the lentiviral expression plasmid pLEX-MCS , using Gibson cloning . These plasmids were used only for the respective mass spectrometry experiments . C-terminally triple FLAG-tagged iTAP and mouse iRhom2-HA were cloned into the retroviral pM6P vector ( a kind gift of Felix Randow ) using Gibson assembly . The N-terminal truncations of iRhom2 used in Figure 2 were cloned into a modified version of the lentiviral expression vector pLEX-MCS in which the puromycin resistance cassette was replaced with a blasticidin resistance gene . The packaging vectors for the production of retrovirus or lentivirus were described previously ( Cavadas et al . , 2017 ) . A Cherry-tagged iRhom2 plasmid previously described ( Luo et al . , 2016 ) was used only for the experiments in Figure 2—figure supplement 1 . LAMP1-mCherry was subcloned from Addgene plasmid #45147 ( a gift from Amy Palmer ) into pM6P vector with zeocin resistance ( prepared by replacing the blasticidin resistance gene in the original vector with zeocin resistance ) . Mouse eGFP-iRhom2 and mouse TACE-GFP were cloned using Gibson assembly into pM6P . HisD vector . Mouse TACE-TagRFP was cloned using Gibson assembly into pM6P . Blast , using Addgene plasmid #42635 ( a gift from Silvia Corvera ) as a source of TagRFP DNA and a mTACE-GFP plasmid from Jürgen Scheller as a source of mTACE cDNA . Murine iTAP-mCherry ( iTAP cDNA from Origene Technologies ) and mCherry-iRhom2 were cloned using Gibson assembly into pM6P . Blast . Human iTAP was inserted with standard cloning techniques into the eGFP containing pIC111 vector ( pIC111 is gift from Iain Cheeseman and Arshad Desai; Addgene plasmid # 44435 ) . Alkaline phosphatase-tagged TACE substrates , a gift of Shigeki Higashiyama were described previously ( Sahin et al . , 2004 ) . V5-tagged ADAM expression plasmids and secreted luciferase construct were described previously ( Christova et al . , 2013 ) . CRISPR plasmids are described below . Human Tumor Necrosis Factor ( TNF ) containing an N-terminal FLAG tag , was cloned into pCR3 by standard techniques . Flag-tagged SREBP2 was a gift of Larry Gerace and STING-FLAG a gift of Lei Jing . In Figure 8—figure supplement 1C , iRhom2-HA in a modified version of pEGFP-N1/non EGFP was used as a template for Quick-Change mutagenesis resulting in the constructs iRhom2 NPAY >AAAA , iRhom2 NRSY >AAAA and iRhom2 Double NxxY >AAAA . Our work involved the use of cell lines . Routine testing for mycoplasma revealed mycoplasma negative status . None of our lines are identified on the list of commonly misidentified cell lines provided by ILCAC . We have noted the sources ( e . g . ATCC accession number ) of our lines in the Key resources table . HEK 293ET , RAW 264 . 7 , L929 , MEF and HeLa cell lines were maintained under standard conditions in Dulbecco's Modified Eagle Medium ( DMEM ) -high glucose supplemented with fetal bovine serum . Bone Marrow Derived Macrophages ( BMDM ) were isolated from 8 week old mice and cultured as previously described ( Adrain et al . , 2012 ) . Embryonic fibroblasts were generated from E14 . 5 embryos and immortalized using lentiviral transduction of SV40 virus large T antigen . Primary human peripheral blood mononuclear cells ( PBMC ) were purified from donor whole blood using the Ficoll-Hypaque gradient method as described previously ( Henry and Martin , 2017 ) . After overnight plastic adherence in heat-inactivated serum containing medium , non-adherent cells were removed and remaining cells were washed three times in PBS . Macrophage differentiation was induced using recombinant human macrophage-colony stimulating factor ( M-CSF , 100 ng/ml ) over five-seven days during cell culture in RPMI supplemented with 10% FCS . Primary human macrophages ( 5 × 105 ) were nucleofected with 200 nM of each siRNA ( control NS oligo , MWG Eurofins - 5'- guuccugagccuggacuac −3'; iTAP oligo #1 , Santa Cruz - catalog code sc-96500; iTAP oligo #2 , GE Dharmacon - M-018955-01-0005; TACE oligo #1 , Santa Cruz - sc-36604; TACE oligo #2 , GE Dharmacon - M-003453-01-0005 ) in nucleofection buffer ( 5 mM KCl , 15 mM MgCl2 , 20 mM HEPES , 150 mM Na2HPO4 [pH 7 . 2] ) using Amaxa Nucleofector ( program Y-010 ) . Cells were plated in 6-well plates ( 2 × 105 cells/well ) or in 24-well plates ( 1 × 105 cells/well ) and 48 hr after nucleofection were stimulated with lipopolysaccharide ( LPS ) . After 18 hr , cell culture supernatants were collected and clarified by centrifugation for 5 min at 800 x g . Cytokines and chemokine concentrations were measured from clarified cell culture supernatants using specific ELISA kits obtained from R and D Biotechne systems ( human TNF – DY210; human IL-6 – DY206; IL-8 – DY208 ) . HEK 293ET cells ( 1 × 106 ) were transfected with pCL ( -Eco , or 10A1 ) packaging plasmids ( Naviaux et al . , 1996 ) plus pM6P . BLAST empty vector ( kind gift of F . Randow , Cambridge , UK ) or pM6P containing the cDNA of human or mouse iTAP or mouse iRhom2 . WT or iTAP KO HEK 293ET cells were transduced with the viral supernatant supplemented with polybrene 8 µg/mL , and selected with blasticidin ( 8 µg/mL ) to generate stable cell lines . To transduce RAW 264 . 7 or L929 , lentiviruses were prepared and concentrated as follows: HEK 293FT cells ( 24 × 106 ) were transfected with pMD-VSVG envelope plasmid , psPAX2 helper plasmid and pLentiCas9-blast or empty vector pLentiGuide-puro or pLentiGuide-puro containing iTAP targeted gDNAs . The viral supernatants were concentrated 300-fold using ultracentrifugation ( 90 , 000 g ) at 4°C for 4 hr , followed by re-suspension in 0 . 1% BSA in PBS . Cells were transduced with the concentrated virus , supplemented with 8 μg/ml polybrene . For CRISPR-mediated knockout of iTAP in human cells , gRNAs targeting exons common to all transcripts: the first coding exon 5’-GCCCCGCTGAGCGATCCCAC-3’ or coding exon 4 of FRMD8 ( iTAP ) 5’-ACGTGTTCTTCCCAAAGCGG-3’ were cloned into pLentiCRISPR v2 ( Addgene plasmid # 52961 ) , a gift from Feng Zhang . For the ablation of iTAP in human cells , HEK 293ET cells ( 2 . 5 × 104/ cm2 ) were transfected using Fugene with pLentiCRISPR v2 empty vector , or either of the pLentiCRISPR-derived sgRNA plasmids described above . The next day , the cells were selected with puromycin ( 8 µg/mL ) for 3 days until mock transfected cells were eliminated . Cells were expanded and single-cell sorted by FACS or serial dilutions on 10 cm culture plates . To screen for the presence of indels in clones , genomic DNA was extracted from each clone and a 200 bp region flanking the site targeted by the gRNA was amplified for exon 1 ( forward = 5’-CCTCCAGCCCCCCATCCCTGGCTC-3’; reverse = 5’-GCCAGAGCTACTTCTCCAGGGCTGGGG-3’ ) or exon 4 ( forward = 5’-TCGGGAGAGGGGAGGGCTAAGCAG-3’; reverse = 5’-GGGCAAGGTGCGAATGTCCAGGGGTC-3’ ) . Clones with mutant alleles were selected and the original PCR fragments amplified were isolated and sequenced via TOPO TA cloning . The selected clones ‘KO A’ and ‘KO B’ which contain indels in all alleles of FRMD8 , were then confirmed for loss of iTAP at the protein level by immunoprecipitation and subsequent western blot with an anti-FRMD8 antibody . For the ablation of iTAP in HeLa cells , an alternative approach was used: HeLas were transiently transfected with a pool of 3 gRNA plasmids ( PX330 , Zhang lab , Addgene 42230 [Cong et al . , 2013] ) ( TGACGTGCTGGTATACCTAG; GGAACGTGTTCTTCCCAAAG; GGCACTTGAGGAGATAGGCG ) specifically targeting exons 2 , 4 and 6 of the human FRMD8 gene respectively , in conjunction with a plasmid encoding puromycin resistance ( pEGFP-C1 from Clontech expressing a GFP-tagged puromycin N-acetyl-transferase ) . After puromycin selection , the efficient knockout of iTAP in the bulk population was confirmed by immunoblotting the lysates with iTAP antibodies , and by the significant depletion of mature TACE ( Figure 6—figure supplement 1 ) . To ablate iTAP in mouse cells , gRNAs targeting the first coding exon ( 5’-TTCGGTGGGACCGCTCCGCA-3’ ) or second coding exon ( 5’-GCACTACTGTATCATCCGCC-3’ ) were cloned into pLentiGuide-Puro ( Addgene plasmid # 52963 ) and used in combination with pLentiCas9-Blast ( Addgene plasmid # 52962 ) ; both gifts of Feng Zhang . For transfection of the gRNAs , 2 × 105 L929 or 5 × 105 RAW 264 . 7 cells were transduced with 40 µl ( RAW 264 . 7 ) or 20 µl ( L929 ) of 300-fold concentrated lentivirus encoding pLentiCas9-Blast ( Addgene #52962 ) and selected with blasticidin ( 4 µg/mL , RAW 264 . 7 and 8 µg/mL , L929 ) . The Cas9-expressing lines were then transduced with the pLentiGuide-Puro sgRNA plasmds targeting the first or second coding exons of mouse Frmd8 ( iTAP ) . Following selection with puromycin ( 4 µg/mL , RAW 264 . 7 and 7 µg/mL L929 ) the cells were single clone sorted by FACS . To screen for iTAP KO clones , genomic DNA was extracted from each clone and PCR used to amplify a 200 bp region flanking the guide sequence ( exon 1: forward = 5’TTGAGAGCTTGAGGAGACCA-3’; reverse 5’-CAGGCTGGAACCAAAGAGTTC-3’ ) ; exon 2: forward = 5’-GGAAATGCTGATTGGACCTC-3’; reverse 5’-CCTGCTGCCAGACCTTACCC-3’ ) . Clones with mutant alleles were identified as described for human cells . Experiments with mice were performed in accordance with protocols approved by the Ethics Committee of the Instituto Gulbenkian de Ciência and the Portuguese National Entity ( DGAV-Direção Geral de Alimentação e Veterinária ) and with the Portuguese ( Decreto-Lei no . 113/2013 ) and European ( directive 2010/63/EU ) legislation related to housing , husbandry , and animal welfare . Generation of iTAP mutant mice iTAP mutant mice were generated via CRISPR/Cas9 as previously described ( Wang et al . , 2013; Casaca et al . , 2016 ) . In brief , two gRNA´s ( 5’-CAGCCGAGTGCAGATCGGGT-3’ and 5’-GTGGCGGACTCAGAAATCAA-3’ ) were designed to introduce a deletion of the first coding exon ( exon 2 ) of the mouse Frmd8 gene . Oligos encoding the gRNA were inserted into the plasmid pgRNAbasic ( Casaca et al . , 2016 ) , which contains a T7 promoter . The linearized vector was used as template for the production of sgRNAs , produced by in vitro transcription using the MEGAshortscript T7 Kit ( Life Technologies ) . RNA was cleaned using the MEGAclear kit ( AM1908 , Life Technologies ) . Cas9 mRNA was produced by in vitro transcription using the mMESSAGE mMACHINE T7 Ultra Kit ( Life Technologies ) and plasmid pT7-Cas9 as a template ( Casaca et al . , 2016 ) . Cas9 mRNA ( 10 ng/ml ) , plus the sgRNAs ( 10 ng/ml ) were injected into the pronuclei of fertilized C57 BL/6 oocytes using standard procedures ( Hogan et al . , 1994 ) . Deletions were assessed by PCR from tail genomic DNA using primers 5′-CCCGACTTGTTTGGCCATTTC-3′ or 5’-CGGGGCCTCGGGTTTG-3’ ( forward ) and 5′-TGGGACAAAGGAAGTGGTGCC-3′ ( reverse ) . The deletion was confirmed by direct sequencing and TOPO-cloning followed by sequencing . These primers ( along with 5’-ACTTTCACCCTACACATTTG-3’ 5’-AGTCCGCCACATCTAAAC-3’ for better amplification of WT alleles ) were also used for genotyping mice and embryos of the iTAP KO line . HeLa ( 5 × 104 cells/well ) were plated on coverslips and transfected with iTAP-GFP ( 600 ng ) or iRhom2-Cherry ( 600 ng ) , either alone or in combination . After 24 hr , cell supernatant was removed and cells were washed three times with PBS ( 2 mL ) . Cells were fixed with 3% paraformaldehyde for 10 min . Cells were washed again three times with PBS ( 2 mL ) followed by permeabilisation with 0 . 15% TX100 for 15 min . Cells were blocked with 2% BSA ( in PBS , pH 7 . 2 ) for 1 hr to reduce non-specific binding of antibodies . Specific primary antibodies against Calnexin ( Cell Signaling , C5C9 ) and Golgi GM130 ( BD , 610823 ) were diluted 1:100 in 2% BSA . Primary antibodies were incubated for 2 hr at room temperature . Cells were washed three times with PBS ( 2 mL ) . Cells were incubated with the relevant rhodamine red-conjugated secondary antibody ( Alexa Fluor ) diluted 1:1000 in 2% BSA for 1 hr at room temperature . Cells were washed again with PBS , followed by incubation with Hoechst ( Sigma ) for 10 min . Coverslips were mounted on slides with 5 μ∧ of Slow Fade ( Molecular Probes ) . For mitotracker staining , cells were transfected with iTAP-GFP ( 600 ng ) , as described previously . After 24 hr , cells were treated with Mitotracker-Red ( 50 nM ) for 15 min at 37°C , followed by fixation with 3% paraformaldehyde . Nuclei were stained with Hoechst . Cells were visualised and analysed using a laser scanning confocal microscope ( Olympus FV1000 ) using a 488 nm Argon laser ( green fluorescence ) , a 543 nm HeNe laser ( red fluorescence ) and a 405 nm LD laser . Confocal images were acquired using Fluroview 1000 V . 1 software . To investigate the lysosomal mis-sorting of mCherry-iRhom2 in iTAP deficient HeLa cells , 1 µg mCherry-iRhom2 or 500 ng mCherry-iRhom2 and 500 ng iTAP-GFP were transfected into parental and iTAP KO HeLa cells using Fugene 6 ( Promega ) . 48 hr post transfection , cells were fixed in 4% PFA in PBS , permeabilized with 0 . 2% Saponin in PBS , blocked in 0 . 1% Saponin/1% BSA in PBS and incubated with an antibody against the lysosomal marker LAMP2 and appropriate secondary antibodies . The cells expressing mCherry-iRhom2 and iTAP-GFP were also co-stained under the same conditions with the early endosome marker EEA1 to visualize a potential localization to endosomes . iRhom1/iRhom2 DKO MEFs expressing eGFP-miRhom2 alone or with mouse iTAP-mCherry ( Figure 5H ) , or WT MEFs stably expressing eGFP-miRhom2 , mTACE-GFP , mTACE-TagRFP , or LAMP1-mCherry delivered by retroviral transduction ( using pM6P derivatives ) in the indicated combinations were plated ( 5 × 104 per well ) on 4-chamber glass-bottomed dishes ( In Vitro Scientific , D35C4-20-1 . 5-N ) 24 hr prior to imaging , in the presence or absence of 10 µM Chloroquine as indicated . Cells were imaged on a laser scanning confocal microscope Zeiss LSM 780* using the 40x/1 . 2 M27 W Korr C-Apochromat objective and a 488 or 561 nm excitation wavelength . Cells were washed twice in cold PBS before incubation in 0 . 2 mg/mL DSP for 45 min . The cross-linker was aspirated off and the cell monolayers were washed three times for 10 min in ice cold PBS containing 50 mM Tris , pH 8 . 0 to quench any remaining cross-linker . Subsequently the cells were lysed in Triton X-100 lysis buffer ( 150 mM NaCl , 50 mM Tris-HCl , protease inhibitors , pH 7 . 4 and 10 mM 1 , 10-phenanthroline ) . Post-nuclear supernatants were supplemented to contain 0 . 1% SDS and 0 . 25% sodium deoxycholate . HEK 293ET cells expressing the indicated plasmids were lysed for 10 min on ice in TX-100 lysis buffer ( 1% Triton X-100 , 150 mM NaCl , 50 mM Tris-HCl , pH 7 . 4 ) containing complete protease inhibitor cocktail ( Roche ) , and 10 mM 1 , 10-phenanthroline ( to inhibit TACE autoproteolysis ) unless otherwise indicated . Post-nuclear supernatants were pre-cleared with unconjugated magnetic beads or agarose at 4°C for 60 min with rotation , followed by capture on anti-HA magnetic beads or anti-FLAG respectively for 90 min . Beads were washed 3–5 times , for 10 min , at 4°C in the same Triton X-100 lysis buffer supplemented with NaCl to 300 mM . Samples were eluted with 1 . 5 x SDS-PAGE sample buffer and incubated at 65°C for 15 min before loading . HEK 293ET cells were stably transduced with lentiviruses encoding pLEX empty vector , or pLEX derivatives containing HA-tagged iRhom1 , iRhom2 , iRhom1 N terminus , Rhbdd2 , RHBDD3 , Ubac2 . Live cells were washed twice with ice-cold PBS , then left untreated or treated with the crosslinker DSP ( 0 . 2 mg/mL ) as described below . Lysates were clarified , then pre-cleared with irrelevant control antibodies conjugated to magnetic beads for 60 min at 4°C with rotation . After saving ‘input’ samples , the lysates were incubated with anti-HA resin for 90 min at 4°C with rotation . Subsequently , the precipitated beads were washed four times in the respective buffers indicated above . One quarter of the beads were reserved for SDS-PAGE analysis , whereas three-quarters of the precipitated beads were resuspended in UREA buffer ( 8M Urea , 4% CHAPS , 100 mM DTT , 0 . 05% SDS ) . For MS analysis , immunoprecipitates were enzymatically digested on 3 kD MWCO filters ( Pall Austria Filter GmbH ) using an adaption of the FASP protocol as described previously ( Bileck et al . , 2014; Slany et al . , 2016 ) . After pre-concentration of the samples , protein reduction and alkylation was performed , then trypsin was added and incubated at 37°C for 18 hr . The digested peptide samples were dried and stored at −20°C then later reconstituted in 5 µl 30% formic acid ( FA ) containing 10 fmol each of 4 synthetic standard peptides and further dilution with 40 µl mobile phase A ( 98% H2O , 2% ACN , 0 . 1% FA ) , as described previously ( Bileck et al . , 2014; Wiśniewski et al . , 2009 ) . LC-MS/MS analyses were performed using a Dionex Ultimate 3000 nano LC-system coupled to a QExactive orbitrap mass spectrometer equipped with a nanospray ion source ( Thermo Fisher Scientific ) . For LC-MS/MS analysis , 5 µl of the peptide solution were loaded and pre-concentrated on a 2 cm x 75 µm C18 Pepmap100 pre-column ( Thermo Fisher Scientific ) at a flow rate of 10 µl/min using mobile phase A . Following this pre-concentration , peptides were eluted from the pre-column to a 50 cm x 75 µm Pepmap100 analytical column ( Thermo Fisher Scientific ) at a flow rate of 300 nl/min and further separation was achieved using a gradient from 7 to 40% mobile phase B ( 80% ACN , 20% H2O , 0 . 1% FA ) over 85 min including column washing and equilibration steps . For mass spectrometric analyses , MS scans were accomplished in the range from m/z 400–1400 at a resolution of 70000 ( at m/z = 200 ) . Subsequently , data-dependent MS/MS scans of the eight most abundant ions were performed using HCD fragmentation at 30% normalized collision energy and analyzed in the orbitrap at a resolution of 17500 ( at m/z = 200 ) . Protein identification was achieved using Proteome Discoverer 1 . 4 ( Thermo Fisher Scientific , Austria ) running Mascot 2 . 5 ( Matrix Science ) . Therefore , raw data were searched against the human proteome in the SwissProt Database ( version 11/2015 with 20 . 193 entries ) with a mass tolerance of 50 ppm at the MS1 level and 100 mmu at the MS2 level , allowing for up to two missed cleavages per peptide . Further search criteria included carbamidomethylation as fixed peptide modification and methionine oxidation as well as protein N-terminal acetylation as variable modifications . Lysates from HEK 293ET cells transfected with either empty vector , iTAP-FLAG , TNF-FLAG , STING-FLAG or SREBP2-FLAG and subjected to an immunoprecipitation with anti-FLAG M2 Affinity Gel . The beads were digested with mass spectrometry-grade porcine trypsin ( Promega ) 10 ng/µl , in 2M urea , 50 mM tris-HCL pH7 . 5 and 1 mM DTT , overnight at 37°C . The peptides were alkylated with Iodoacetamide ( Sigma ) , desalted using Empore Octadecyl C18 extraction disks and analysed on a Q-Exactive + mass spectrometer coupled to a nano uHPLC ( Thermo Fisher ) . Analysis was performed with MaxQuant 1 . 5 . 8 . 3 software . The abundance of the different interactors was determined with the average of protein peptides detected in each sample . Biotinylation was performed as previously described for BMDM with small modifications ( Adrain et al . , 2012 ) . RAW 264 . 7 macrophages or iRhom2-HA HEK 293ET ( 1 . 5 × 106 cells , six well plates ) were moved to a cold room ( at 4°C ) , washed with ice-cold PBS pH 8 . 0 for 10 min , incubated with ( 1 mg/mL ) Sulfo-NHS-LC-Biotin in PBS pH 8 . 0 , according to the manufacturer’s instructions . Following quenching with 50 mM Tris in PBS , cells were lysed for 10 min with TX-100 lysis buffer ( 1 , 10-phenathroline , protease inhibitors , 50 mM Tris ) , then biotinylated surface proteins from post-nuclear supernatants were captured on neutravidin agarose resin at 4°C overnight . The resin was washed three times , 10 min , with TX-100 lysis buffer containing 300 mM NaCl at 4°C . Samples were eluted with 1 . 5 x SDS-PAGE sample buffer and incubated at 65°C for 15 min , before loading . To improve the detection of TACE , cells were lysed in TX-100 lysis buffer supplemented with 1 mM EDTA , 1 mM MnCl2 , 1 mM CaCl2 and glycoproteins were captured using Concanavalin A ( ConA ) Agarose . Beads were washed twice in the same buffer and eluted by heating for 15 min at 65°C in sample buffer supplemented with 15% sucrose or for 5 min at 95°C in sample buffer supplemented with 15% sucrose and 1x Glycoprotein denaturation buffer ( NEB ) . Post-nuclear lysate supernatants or denatured lysates , are denatured at 65°C for 15 min in the presence of 1x Glycoprotein Denaturing buffer ( NEB ) . Endo-H and PNGase F reactions are set up according to manufacturer’s instructions for 1 hr at 37°C . Shedding assays were performed using previously described plasmids encoding alkaline phosphatase-tagged EGFR ligands: Transforming Growth Factor α ( TGFα ) , Amphiregulin ( AREG ) , Epiregulin ( EPIREG ) , Heparin Binding-Epidermal Growth Factor ( HB-EGF ) , Epidermal Growth Factor ( EGF ) and Betacellulin ( BTC ) or Tumor necrosis factor ( TNF ) ( Sahin et al . , 2006; Zheng et al . , 2002 ) . HEK 293ET ( 3 × 105 in six well-plates ) were transfected with 1 μg cDNA of AP-substrates and 6 μL PEI . 48 hr later , cells were washed three times with serum free media before incubation for 1 hr in 1 mL Optimem ( containing the vehicle of the drug in next step ) ( for basal shedding ) , followed by 1 hr with 1 mL Optimem containing 1 μM PMA ( Phorbol 12-myristate 13-acetate ) or 2 . 5 μM IO ( Ionomycin; for stimulated shedding ) . Supernatants from each incubation step were collected . Following , the cells were washed in ice-cold PBS three times and lysed in Triton X-100 buffer described previously ( unshed material ) . Supernatants and lysates were centrifuged on a bench-top centrifuge at top speed for 10 min to remove cells and cell debris . Supernatants and lysates were incubated with the Alkaline Phosphate ( AP ) substrate p-nitrophenyl phosphate ( pNPP ) at room temperature . AP activity measured using a 96-well plate spectrophotometer ( 405 nm ) . Results are presented as PMA or IO ‘shedding over total’ calculated by the formula ( 'cleared stimulated shedding’ ) / ( 'cleared stimulated shedding’ plus’ unshed’ ) . ‘Cleared stimulated shedding’ denotes ‘stimulated shedding’ minus ‘basal shedding’ . The release into the medium of a secreted form of luciferase containing a signal peptide was assessed as a control , as described ( Christova et al . , 2013 ) . The assay was performed as previously described ( Adrain et al . , 2012 ) . In brief , 8 million HEK 293ET cells were lysed for 10 min on ice in 1% Triton X-100 , 150 mM NaCl , 50 mM Tris-HCl , pH 7 . 4 containing complete protease inhibitors cocktail ( Roche ) . Importantly , all steps were performed in the absence of the metalloprotease inhibitor 1 , 10-phenanthroline to preserve TACE activity . As previously described ( Schlöndorff et al . , 2000 ) when lysates are made in the absence of 1 , 10-Phenathroline , TACE autocatalytically cleaves off its cytoplasmic tail , result in loss of the epitope detected by the rabbit polyclonal antibody ( Ab39162 , Abcam ) used for western blotting ( Adrain et al . , 2012 ) . Therefore , mouse anti-TACE ( 9301 , R and D ) , which recognizes an epitope within the ectodomain , was used for immunoprecipitations , to ensure that TACE was captured regardless of autocatalysis . Mature TACE without its cytoplasmic domain retains proteolytic activity in vitro ( Adrain et al . , 2012 ) . Anti-TACE or mouse IgG ( anti-GFP; mock ) antibodies were incubated overnight with lysates , followed by capture of the immunocomplexes with anti-mouse magnetic beads . Immunoprecipitates were mixed with the fluorogenic TACE substrate peptide ( ENZO Life Sciences , BML-P235-0001 ) and fluorescence was measured over 3 hr on a Victor three plate reader at 37°C , according to manufacturer instructions . As the western blots used the rabbit polyclonal antibody ( Ab39162 , Abcam ) , immunoprecipitates consequently show no evidence of mature TACE . Protein was extracted from mouse tissues by lysing in a modified RIPA buffer ( 150 mM NaCl , 50 mM Tris-HCl , pH 7 . 4 , 1 mM EDTA , 1% Triton X-100 , 1% Na Deoxycholic Acid , 0 . 1% SDS containing protease inhibitors and 10 mM 1 , 10-phenanthroline ) . Homogenates were clarified and normalized , then incubated with ConA resin , as described above . Semi-quantitative densitometric analysis on scanned images from western blot exposures was performed with Fiji software , measuring at least three independent experiments . Results represent mean values ± standard deviation . Half-life was calculated using GraphPad Prism software . Alignments were performed using Geneious software using the CLUSTALW algorithm . All statistical analyses were done with two-sample two tail unpaired t-tests assuming unequal variances , using excel software . In Figure 3 , comparisons were performed after transforming the raw values into their relative ( fold change ) values to the WT sample . The Pearson’s correlation and Mander’s colocalization coefficient were calculated after appropriate thresholding with the Volocity ( Perkin Elmer ) image analysis package . Thresholds were applied evenly across conditions . For the statistical analysis of the Pearson’s correlation , the Pearson’s taken from at least 20 individual cells acquired over two independent experiments were subjected to unpaired , two tailed t-tests in Excel . P-Values above 0 . 05 were considered as not significant . | Inflammation forms part of the body's defense system against pathogens , but if the system becomes faulty , it can cause problems linked to inflammatory and autoimmune diseases . Immune cells coordinate their activity using specific signaling molecules called cytokines . For example , the cytokine TNF is an important trigger of inflammation and is produced at the surface of immune cells . A specific enzyme called TACE is needed to release TNF , as well as other signaling molecules , including proteins that trigger healing . Previous work revealed that TACE works with proteins called iRhoms , which regulate its activity and help TACE to reach the surface of the cell to release TNF . To find out how , Oikonomidi et al . screened human cells to see what other proteins interact with iRhoms . The results revealed a new protein named iTAP , which is required to release TNF from the surface of cells . It also protects the TACE-iRhom complex from being destroyed by the cell’s waste disposal system . When iTAP was experimentally removed in human immune cells , the cells were unable to release TNF . Instead , iRhom and TACE travelled to the cell's garbage system , the lysosome , where the proteins were destroyed . Removing the iTAP gene in mice had the same effect , and the TACE-iRhom complex was no longer found on the surface of the cell , but instead degraded in lysosomes . This suggests that in healthy cells , the iTAP protein prevents the cell from destroying this protein complex . TNF controls many beneficial processes , including fighting infection and cancer . However , when the immune system releases too many cytokines , it can lead to inflammatory diseases or even cause cancer . Specific drugs that target TNF are not always effective administered on their own , and sometimes , patients stop responding to the drugs . Since the new protein iTAP works as a switch to turn TNF release on or off , it could provide a target for the development of new treatments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2018 | iTAP, a novel iRhom interactor, controls TNF secretion by policing the stability of iRhom/TACE |
BRCA1 plays a critical role in homology-directed repair ( HDR ) of DNA double strand breaks , and the repair defect of BRCA1-mutant cancer cells is being targeted with platinum drugs and poly ( ADP-ribose ) polymerase ( PARP ) inhibitors . We have employed relatively simple and sensitive assays to determine the function of BRCA1 variants or mutants in two HDR mechanisms , homologous recombination ( HR ) and single strand annealing ( SSA ) , and in conferring resistance to cisplatin and olaparib in human cancer cells . Our results define the functionality of the top 22 patient-derived BRCA1 missense variants and the contribution of different domains of BRCA1 and its E3 ubiquitin ligase activity to HDR and drug resistance . Importantly , our results also demonstrate that the BRCA1-PALB2 interaction dictates the choice between HR and SSA . These studies establish functional and mutational landscapes of BRCA1 for HDR and therapy resistance , while revealing novel insights into BRCA1 regulatory mechanisms and HDR pathway choice .
Germline , heterozygous mutations in BRCA1 confer high risk of breast and ovarian cancer development in an autosomal dominant fashion ( Couch et al . , 2014; Fackenthal and Olopade , 2007 ) . BRCA1 has been implicated in numerous cellular processes including DNA repair , cell cycle checkpoints , centrosome duplication , and transcriptional regulation , etc . ( Deng , 2006; Mullan et al . , 2006; Roy et al . , 2012 ) . Ever since BRCA1 was found to localize to discrete nuclear foci and colocalize with the recombination enzyme RAD51 ( Scully et al . , 1997 ) , its function in homologous recombination ( HR ) -based repair of DNA double strand breaks ( DSBs ) has been a subject of intense study ( Moynahan and Jasin , 2010 ) . Tumors arising from BRCA1 mutation carriers usually show loss of the wild-type ( wt ) allele , which renders tumor cells biallelically null for the gene . It is generally believed that genome instability resulting from the DNA repair defect following the loss of BRCA1 is a driver of tumor development ( Li and Greenberg , 2012; Venkitaraman , 2014 ) . Importantly , the very DNA repair defect that leads to tumor development is also an ‘Achilles’ Heel’ of the resulting tumor cells , which can be selectively killed by suitable DNA-damaging agents that target HR defect , such as platinum drugs and poly ( ADP-ribose ) polymerase ( PARP ) inhibitors ( Lord and Ashworth , 2016 ) . The human BRCA1 gene consists of 24 exons encoding a large polypeptide of 1863 amino acid residues . BRCA1 contains a RING domain at the N terminus and a tandem BRCT domain at the C terminus ( Figure 1A ) . Two nuclear localization signals ( NLSs ) facilitate the localization of BRCA1 primarily to the nucleus ( Chen et al . , 1996 ) , whereas a nuclear export signal ( NES ) can mediate its cytoplasmic export ( Rodríguez and Henderson , 2000 ) . The RING domain of BRCA1 binds to a similar domain in its close partner BARD1 ( Wu et al . , 1996 ) , leading to the formation of a stoichiometric complex that possesses substantial ubiquitin E3 ligase activity ( Hashizume et al . , 2001; Ruffner et al . , 2001 ) . At the same time , binding of BARD1 to BRCA1 shields the NES and helps retain BRCA1 in the nucleus ( Fabbro et al . , 2002 ) . The BRCT domain directly binds , in a phosphorylation-dependent manner , to at least three other proteins , namely BRIP1 , CtIP and Abraxas , all of which have function in DNA repair ( Huen et al . , 2010; Jiang and Greenberg , 2015 ) . Additionally , BRCA1 contains a highly conserved coiled-coil ( CC ) motif that directly binds PALB2 , the partner and localizer of BRCA2 ( Xia et al . , 2006 ) , which links BRCA1 and BRCA2 in the HR pathway ( Sy et al . , 2009; Zhang et al . , 2009a , 2009b ) . 10 . 7554/eLife . 21350 . 003Figure 1 . Sequence alterations generated in BRCA1 and their effects on protein-protein interactions and HR . ( A ) Domain structure of BRCA1 and the binding sites for its interacting partners . NES , nuclear export signal; NLS , nuclear localization signal . ( B ) Effects of the BRCA1 variants on the binding of BARD1 , PALB2 , BRIP1 and Abraxas . The 3xMyc-tagged BRCA1 proteins were transiently expressed in 293T cells and IPed with an anti-Myc antibody . WCE , whole cell extract . ( C ) Schematic of the HR reporter assay . The DR-GFP reporter contains two defective copies of the GFP gene , one disrupted by an I-SceI site and the other lacking a promoter . I-SceI cutting of the first copy generates a DSB , and repair by HR with the second copy as a template leads to restoration of a functional GFP gene . ( D ) HR activities of the variants relative to the wt BRCA1 protein . Data shown are the means from two to seven independent experiments for each variant or mutant . Error bars represent standard deviations ( SDs ) . The grey bars indicate variants that are among the top 20 missense variants but are already present in BRCA1 cDNAs obtained from three independent sources . The calculated cutoff threshold is indicated by horizontal lines . **p<0 . 01 . See Figure 1—source data 1 for details . ( E ) Levels of BRCA1 protein following knockdown and re-expression . Cells treated with a control siRNA ( NSC1 ) were used as a control for the endogenous protein abundance . DOI: http://dx . doi . org/10 . 7554/eLife . 21350 . 00310 . 7554/eLife . 21350 . 004Figure 1—source data 1 . HR activities of the BRCA1 variants and mutants analzyed in this study . See texts and materials and methods for details . DOI: http://dx . doi . org/10 . 7554/eLife . 21350 . 00410 . 7554/eLife . 21350 . 005Figure 1—figure supplement 1 . Capacity of the BRCA1 variants to bind BARD1 , PALB2 , CtIP and BRIP1 . The 3xMyc-tagged BRCA1 proteins were transiently expressed in 293T cells and IPed with an anti-Myc antibody . Note that the amount of CtIP co-IPed with BRCT mutants R1699Q and A1708E was the same as the background level in the vector lane . DOI: http://dx . doi . org/10 . 7554/eLife . 21350 . 00510 . 7554/eLife . 21350 . 006Figure 1—figure supplement 2 . Expression levels of BRCA1 BRCT missense and truncating mutants in U2OS/DR-GFP cells first depleted of the endogenous BRCA1 and then transfected with cDNA expressing the mutants . Cells untreated with any siRNA were used as a control for the endogenous protein abundance ( spliced from the same gel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21350 . 006 The Breast Cancer Information Core ( BIC ) database was the first major public database into which patient-derived mutations and variants were deposited . It contains a total of 15 , 311 entries of BRCA1 sequence alterations , among which 6133 ( 40% ) are frameshift or protein-truncating mutations , which can generally be classified as pathogenic . Importantly , 4577 ( 30% ) are missense variants , most of which are considered to be of ‘unknown clinical importance’ , or commonly known as variants of unknown ( or uncertain ) clinical significance ( VUSs ) , the interpretation of which remains a challenge ( Couch et al . , 2014 ) . Recently , more data have been deposited into the ClinVar database , which now includes all BIC cases along with ones from multiple other sources . Based largely on genetic and epidemiological evidence , a large number of VUSs has been designated by ClinVar as ‘benign’ . However , the majority of the variants have not been functionally characterized . A number of studies have been conducted to assess the functions of select BRCA1 VUS in transcription , DNA repair , and in supporting the viability of mouse embryonic stem ( ES ) cells ( Bouwman et al . , 2013; Chang et al . , 2009; Millot et al . , 2012; Ransburgh et al . , 2010; Towler et al . , 2013 ) . However , data on the impact of VUS on drug sensitivity have only begun to emerge . In particular , a recent study used a recombinase-mediated cassette exchange ( RMCE ) approach to profile 74 patient-derived missense variants for their activities in conferring resistance to cisplatin and a PARP inhibitor in mouse ES cells , and it found that all functionally deleterious variants were confined to the RING and BRCT domains ( Bouwman et al . , 2013 ) .
To establish an initial functional landscape of BRCA1 mutations for HR-mediated DSB repair , we constructed 32-patient-derived missense variants along the entire length of the protein ( Table 1 ) . The list included 11 variants with known or partially known functional consequences: two in the RING domain ( C61G and C64R ) that disrupt BARD1 binding , three in the CC motif ( M1400V , L1407P and M1411T ) that affect PALB2 binding and six in the BRCT domain ( S1655F , C1697R , R1699Q , A1708E , S1715R and M1775R ) that abrogate BRCA1 interaction with Abraxas , BRIP1 and CtIP ( Figure 1B and Figure 1—figure supplement 1 ) . Also included were the top 20-patient-derived missense variants in the BIC database ( 21 in total due to a tie at the 20th place ) and the top 22 in the ClinVar database ( Table 1 ) . Additionally , we included four other VUSs , L668F , S1101N , S1140G and T1561I , which have now been recorded between 16 and 40 times in ClinVar . As a control for the BRCT missense variants , we also generated a patient-derived frameshift mutation ( 5055delG or p . Val1646Serfs ) that truncates the protein immediately before the BRCT domain . 10 . 7554/eLife . 21350 . 007Table 1 . BIC database and ClinVar reports of patient-derived BRCA1 missense variants characterized in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 21350 . 007HGVS cDNABIC designationBIC entry countBIC clinical importanceClinVar individualsClinVar designationAlign-GVGD gradeIARC classificationc . 4837A>GS1613G248No1319BenignC01c . 181T>GC61G239Yes323PathogenicC655c . 2612C>TP871L211No1348BenignC01c . 3113A>GE1038G182No1237BenignC01c . 3548A>GK1183R164No1215BenignC01c . 4039A>GR1347G161Unknown712BenignC01c . 1067A>GQ356R155Unknown315BenignC01c . 3024G>AM1008I139Unknown146BenignC01c . 2521C>TR841W119Unknown128BenignC151c . 4883T>CM1628T96Unknown491BenignC01c . 1487G>AR496H86Unknown90BenignC01c . 736T>GL246V80Unknown83BenignC01c . 3119G>AS1040N68Unknown141BenignC01c . 4910C>TP1637L67Unknown76BenignC01c . 2077G>AD693N65No229BenignC01c . 4956G>AM1652I61Unknown94BenignC01c . 2315T>CV772A60Unknown62BenignC01c . 4535G>TS1512I58No72BenignC01c . 1456T>CF486L56Unknown58BenignC01c . 536A>GY179C54Unknown59BenignC351c . 1648A>CN550H54Unknown59BenignC01c . 5123C>AA1708E46Yes80PathogenicC655c . 5324T>GM1775R31Unknown43PathogenicC455c . 3418A>GS1140G29Unknown40BenignC01c . 4682C>TT1561I26Unknown33ConflictingC0N/Ac . 2002C>TL668F25Unknown28BenignC01c . 3302G>AS1101N14Unknown16BenignC01c . 190T>CC64R12Unknown15ConflictingC65N/Ac . 5096G>AR1699Q11Unknown22ConflictingC355c . 5145C>GS1715R#5Unknown5PathogenicC655c . 4964C>TS1655F3Unknown4ConflictingC25N/Ac . 5098T>CC1697R3Unknown7ConflictingC65N/Ac . 4198A>GM1400V1Unknown1UncertainC0N/Ac . 4220T>CL1407P1Unknown1UncertainC65N/Ac . 4232T>CM1411T1Unknown1UncertainC65N/A5055delGV1646Sfs5Yes6Pathogenic--Align-GVGD grade: C0 to C65 denote increasing likelihood of a variant to cause damage ( to protein function ) . IARC ( ENIGMA ) classification: 5 , Definitely pathogenic; 4 , Likely pathogenic; 3 , Uncertain , 2 , Likely not pathogenic or of little clinical significance; 1 , Not pathogenic or of no clinical significance . # c . 5145C>G , c . 5145C>A and c . 5143A>C have all been reported to generate BRCA1-S1715R . Data shown are up to date as of March 21 , 2017 . Notably , our sequencing analysis revealed P871L , E1038G , K1183R and S1613G in the ‘wild-type’ BRCA1 cDNAs from multiple independent sources . Indeed , they are considered as neutral/benign by both BIC and ClinVar . Among the rest of the top 22 variants in ClinVar , C61G and A1708E are considered pathogenic , and the rest are all considered benign largely based on genetic or epidemiological evidence . Similarly , L668F , S1101N , S1140G and T1561I are now also considered as benign . Note that this study was initiated in 2010 largely based on BIC , in which most of the above are now still listed as VUS . To measure the activity of BRCA1 variants in HR , we re-developed a previously described HR assay based on a ‘protein replacement’ strategy ( Ransburgh et al . , 2010; Sy et al . , 2009; Xia et al . , 2006 ) ( Figure 1C ) . In this assay , the endogenous BRCA1 was first depleted in U2OS/DR-GFP cells ( Nakanishi et al . , 2005; Xia et al . , 2006 ) using an siRNA targeting the 3’-UTR of the gene , and various BRCA1 proteins were then expressed from cDNAs ( lacking the 3’-UTR ) co-transfected with a plasmid expressing I-SceI , which cleaves the reporter to induce DSB formation and subsequent repair . To ensure uniformity of BRCA1 knockdown , siRNA transfections were performed in 10 cm plates and the cells were harvested , mixed and reseeded into six-well plates before the second transfection . The averaged expression levels of exogenous BRCA1 was three to five times that of the endogenous protein ( Figure 1E ) , suggesting a significant degree of overexpression . However , the variants were expressed at levels comparable to that of the wt BRCA1 ( not shown ) , with the exception being the BRCT point mutants , among which all but R1699Q showed much lower expression levels ( Figure 1—figure supplement 2 ) . Under the setting used , C61G and C64R reduced BRCA1 HR activity by ~5-fold to a level slightly above the vector control ( Figure 1D and Table 2 ) . Similarly , a profound impact on HR was observed with mutations in the BRCT domain . Consistent with a previous report ( Sy et al . , 2009 ) , variants L1407P and M1411T , which abrogate or substantially impair PALB2 binding , reduced HR by ~4 . 5-fold , whereas M1400V , which moderately impairs PALB2 binding ( Figure 1B , lane 15 ) , caused a ~ 25% decrease in the apparent HR activity . In contrast , all variants that are neutral for protein-protein interaction at the RING , coiled-coil and BRCT domains showed greater than 85% of the HR activity of the wt protein . Using variants that have been functionally characterized previously as controls , our assay was found to be both sensitive and specific , and the cutoff threshold of the assay was calculated to be 54 . 3–54 . 7% , which would define M1400V and all the protein-protein interaction neutral variants as being functionally benign . However , on an individual basis , the modest decrease of HR activity in the case of M1400V is statistically significant . 10 . 7554/eLife . 21350 . 008Table 2 . Comparison of results on HR activity and drug response of the BRCA1 variants analyzed in this study obtained from previous and this studies . DOI: http://dx . doi . org/10 . 7554/eLife . 21350 . 008DomainHR activity ( % ) Cisplatin responseOlaparib responseSy et al . , 2009Ransburgh et al . ( 2010 ) Towler et al . ( 2013 ) Bouwman et al . ( 2013 ) Lu et al . , 2015This studyBouwman et al . ( 2013 ) This studyBouwman et al . ( 2013 ) This studyVectorN/A~10~9~20~1817 . 3SSSSWT98100100100100100RRRRC61GRING-~17--23 . 621 . 2SS-SC64RRING-----22 . 3-S-SY179C--~95-157 . 192 . 7-R-RL246V-----91 . 5RR-RQ356R-----95 . 5-R-RF486L----16095 . 8-R-RR496H-----95 . 9-R-RN550H----90 . 888 . 4-R-RL668F----96 . 893 . 6RR-RD693N-----111 . 5RR-RV772A----110 . 484 . 5-R-RR841W-----98 . 2-R-RP871L-----wt-wt-wtM1008I-----99 . 2RR-RE1038G-----wt-wt-wtS1040N-----98 . 0-R-RS1101N-----97 . 3RR-RS1140G-----106 . 2RR-RK1183R-----wt-wt-wtR1347G-----106 . 5-R-RM1400VCC56----74 . 8RSRSL1407PCC24----24SSSSM1411TCC25----26RSRSS1512I-----95 . 3-R-RT1561I----133 . 2102 . 3-R-RS1613G-----wt-wt-wtM1628T----107107 . 6RR-RP1637L----98 . 899 . 5-R-R5055△GBRCT-----22 . 5-S-SM1652IBRCT-----106 . 2RR-RS1655FBRCT---~30-8 . 4SS-SC1697RBRCT-----8 . 0-S-SR1699QBRCT---~45-16 . 9SSSSA1708EBRCT-----10 . 7-S-SS1715RBRCT-----9 . 3-S-SM1775RBRCT36-~6--10 . 2-S-SR , resistant; S , sensitive . See Figure 1—source data 1 , Figure 2—source data 1 and Figure 3—source data 1 for details . Given the proposed role of BRCA1 on DSB end resection ( Bunting et al . , 2010 ) , the activities of all the above variants in SSA , a deletion-causing repair pathway that also utilizes resected DNA ends , were measured . For this purpose , we used a similar protein replacement strategy with another U2OS-based reporter cell line containing an integrated SA-GFP reporter ( Stark et al . , 2004 ) . Consistent with a previous report that Brca1-deficient mouse cells show reduced SSA ( Stark et al . , 2004 ) , depletion of BRCA1 in the human cells resulted in a reduction of the ( already low ) basal SSA activity ( Figure 2A ) , and re-expression of wt BRCA1 restored the SSA activity ( Figure 2B ) . As reported previously that pathogenic missense mutations in the RING domain can cause defects in both HR and SSA ( Towler et al . , 2013 ) , we found that the BARD1-binding mutants ( C61G and C64R ) showed greatly reduced SSA activities ( Figure 2B ) . At the same time , all the BRCT mutants were completely defective . In contrast , the three mutations that impact PALB2 binding all resulted in increased SSA activity . In particular , the L1407P and M1411T mutations elevated SSA by ~2 . 5-fold . This finding suggests that BRCA1 interactions with BARD1 and BRCT-binding partners are both required for its function in the resection step of the DSB repair , whereas the BRCA1-PALB2 interaction functions downstream to promote HR and reduce SSA . 10 . 7554/eLife . 21350 . 009Figure 2 . The BRCA1-PALB2 interaction suppresses SSA . ( A ) Effect of BRCA1 depletion on SSA . Data shown are the means of the four data points obtained from two independent experiments each performed in duplicates . Error bars represent SDs . ***p<0 . 001 . ( B ) SSA activities of the BRCA1 variants relative to the wt protein . A schematic of the SA-GFP reporter assay is shown at the upper left corner . Data shown are the means ± SDs from two to five independent experiments for each variant or mutant . *p<0 . 05 . See Figure 2—source data 1 for details . ( C ) Effects of BRCA2 and PALB2 depletion on SSA . Data shown are the means ± standard errors of mean ( SEMs ) from four independent experiments . *p<0 . 05; **p<0 . 01 . ( D ) Requirement of RAD52 for SSA upregulation following BRCA2 and PALB2 depletion . Data shown are the means ± SEMs from three independent experiments . *p<0 . 05; **p<0 . 01 . ( E ) BRCA1 and BRCA2-binding defects of PALB2-L21A , L35A and A1025R mutants . The FLAG-HA-tagged mutants were transiently expressed in 293T cells and IPed with anti-FLAG M2 agarose beads . ( F–G ) HR ( F ) and SSA ( G ) activities of PALB2-L21A , L35A and A1025R mutants . Data shown are the means ± SDs from three to five independent experiments for each mutant . ***p<0 . 001 . ( G ) Levels of PALB2 protein following knockdown and re-expression in U2OS/DR-GFP cells . Cells untreated with any siRNA were used as a control for the endogenous protein abundance . DOI: http://dx . doi . org/10 . 7554/eLife . 21350 . 00910 . 7554/eLife . 21350 . 010Figure 2—source data 1 . SSA activities of the BRCA1 variants and mutants analyzed in this study . See texts and materials and methods for details . DOI: http://dx . doi . org/10 . 7554/eLife . 21350 . 010 It has been reported that loss of BRCA2 increases SSA in mouse cells ( Stark et al . , 2004; Tutt et al . , 2001 ) , and a recent report showed that some PALB2 heterozygous human lymphoblastoid cell lines have increased SSA activity ( Obermeier et al . , 2016 ) . To better understand the role of PALB2 and BRCA2 in this process , we depleted the two proteins in parallel in the U2OS/SA-GFP reporter cells and measured the effect . As shown in Figure 2C , depletion of either protein caused substantial upregulation of SSA , with BRCA2 siRNAs stimulating SSA by 1 . 8–5 . 4 fold and PALB2 siRNAs 5 . 6–7 . 2 fold . Initially , two siRNAs for each gene were used ( #1949 and #9025 for BRCA2 and #1493 and #2693 for PALB2 ) . The number of BRCA2 siRNAs were later increased to six due to the observation that siRNAs #1949 and #9025 showed a large difference in the fold change of SSA they induced . With the exception of siRNA #11170 , all BRCA2 siRNAs showed similar and effective knockdown . Although we cannot explain why siRNA #1949 stood out as the sole outlier in terms of SSA induction , we did notice that this siRNA had the least negative effect on cell morphology and growth rate ( data not shown ) . The level of RAD51 was reduced by all BRCA2 siRNAs , which could be explained by potentially reduced stability in the absence of BRCA2 , a ‘carrier’ of RAD51 that can simultaneously bind six molecules of the latter ( Jensen et al . , 2010; Liu et al . , 2010 ) and thus could potentially serve as a stabilizer as RAD51 . However , there is no correlation between the levels of RAD51 reduction and SSA increase among all siRNAs used ( Figure 2C ) , suggesting that the partial loss of RAD51 is not a significant cause of SSA upregulation . RAD52 has been shown to possess strand annealing activity in vitro ( Mortensen et al . , 1996 ) and to promote SSA in vivo ( Stark et al . , 2004 ) . To test if RAD52 is required for the increased SSA in the absence of BRCA2 and PALB2 , we silenced its expression , either alone or in combination with BRCA2 and PALB2 , and measured the effect on SSA efficiency . As shown in Figure 2D , strong depletion of RAD52 ( by siRNA #1972 ) reduced basal SSA , whereas relatively mild depletion of RAD52 ( by siRNA #2569 ) showed little effect . Strong RAD52 depletion greatly impeded the increase of SSA elicited by both BRCA2 and PALB2 loss , while the relatively weaker depletion of RAD52 also reduced the upregulation of SSA upon loss of either BRCA2 or PALB2 , albeit to lesser extents . To further elucidate the role of PALB2 in the DSB repair pathway choice , we employed point mutants that are defective for BRCA1 binding ( L21A and L35A ) or BRCA2 binding ( A1025R ) ( Figure 2D ) ( Oliver et al . , 2009; Sy et al . , 2009 ) . As expected , and in contrast to wt PALB2 , which effectively restored HR in reporter cells depleted of the endogenous protein , the BRCA1- and BRCA2-binding mutants all showed dramatically reduced HR activity ( Figure 2E ) . Re-expression of wt PALB2 in the SA-GFP reporter cells depleted of the endogenous PALB2 significantly suppressed the greatly enhanced SSA activity , whereas neither the BRCA1-binding mutants nor the BRCA2-binding mutant showed such ability ( Figure 2F ) . Very recently , we reported the first patient-derived missense pathogenic mutation in PALB2 , L35P , that disrupts the binding of BRCA1 ( Foo et al . , 2017 ) . In a way virtually identical to L35A , this mutant also failed to suppress SSA in the above setting ( data not shown ) . The suppression by wt PALB2 was moderate , which could be due to that the depletion of PALB2 might have caused certain secondary effects that cannot be immediately restored upon sudden re-expression of the protein . Taken together , our results establish that PALB2 controls the HR/SSA pathway choice following resection and that the interactions between PALB2 with both BRCA1 and BRCA2 are required for this function . To profile the functional impact and therapeutic relevance of BRCA1 variants , we developed a relatively simple drug sensitivity assay ( Figure 3A–B ) using the BRCA1-deficient , triple negative MDA-MB-436 breast cancer cells ( Elstrodt et al . , 2006 ) . Briefly , MDA-MB-436 cells were transfected with cDNA constructs expressing wt BRCA1 or the variants and then subjected to selection with either Geneticin ( G418 ) alone ( for transfection efficiency ) or a combination of G418 and either olaparib , a potent PARP inhibitor ( PARPi ) , or cisplatin . Colonies were allowed to grow for a period of 3–4 weeks . Under the condition used , we never obtained a single olaparib-resistant colony in untransfected or vector transfected cells; in only one ( out of about ten ) occasion , four cisplatin-resistant colonies were obtained from vector transfected cells . These results allow us to rule out any meaningful contribution of the endogenous truncated BRCA1 to our readout . Thus , MDA-MB-436 is a suitable cell line for BRCA1 functional analysis using our protocol . 10 . 7554/eLife . 21350 . 011Figure 3 . Abilities of the wt and mutant or variant BRCA1 proteins to confer cisplatin and olaparib resistance . ( A ) Schematic of the colony formation assay . The BRCA1 mutant MDA-MB-436 breast cancer cells were transfected with BRCA1 expression plasmids , reseeded and then selected with G418 alone or G418 with cisplatin or olaparib . Cells were stained with crystal violet 3–4 weeks after selection . ( B ) Representative crystal violet-stained plates . ( C ) Relative activities of wt and mutant or variant BRCA1 proteins to support colony formation in the presence of cisplatin or olaparib . Data shown are the means ± SDs from two to three independent experiments for each variant or mutant . **p<0 . 01 . Horizontal lines represent the cutoff thresholds for cisplatin ( blue ) and olaparib ( orange ) . See Figure 3—source data 1 for details . DOI: http://dx . doi . org/10 . 7554/eLife . 21350 . 01110 . 7554/eLife . 21350 . 012Figure 3—source data 1 . Abilities of the BRCA1 variants and mutants to support olaparib and cisplatin resistance in MDA-MB-436 cells . See texts and materials and methods for details . DOI: http://dx . doi . org/10 . 7554/eLife . 21350 . 012 Resistance to PARPi was found to largely correlate with the HR activity of each variant ( Figure 3C ) . Specifically , the BARD1-binding mutants ( C61G and C64R ) , the PALB2-binding mutants ( M1400V , M1407P and M1411T ) and the BRCT mutants all demonstrated virtually no growth under PARP inhibition , whereas other VUSs were fully or largely resistant . Notably , BRCA1-M1400V , which showed only moderately impaired binding to PALB2 and 73% of the HR activity of the wt protein ( Figure 1B , D ) , failed to confer PARPi resistance . The pattern of cisplatin resistance also largely reflected the HR activities of the variants , as the RING , CC and BRCT mutants all showed hypersensitivity , while other variants conferred greater than ~90% colony-forming ability of the wt protein . These data clearly demonstrate the correlation between HR activity , but not SSA activity , and drug resistance . They also underscore the importance of the RING , CC and BRCT domains and their corresponding binding partners for BRCA1 function in DNA repair . To understand the cause of the functional deficiency of the mutants more fully , we analyzed their subcellular localization in U2OS cells . Although the endogenous BRCA1 is almost exclusively localized in the nucleus , transiently expressed exogenous BRCA1 can be found in the cytoplasm in some cells , with the extent depending on the transfection condition . This is commonly explained by the insufficient amount of the endogenous BARD1 , which is thought to shield the NES of BRCA1 thereby retaining BRCA1 in the nucleus ( Fabbro et al . , 2002 ) . Indeed , BRCA1-C61G , which does not bind BARD1 , showed a more diffuse localization pattern than the wt protein ( Figure 4A ) . The PALB2-binding mutant M1411T behaved like the wt protein , as reported before ( Sy et al . , 2009 ) . Interestingly , patient-derived mutations that truncate BRCA1 at the BRCT domain have been reported to cause BRCA1 to localize primarily in the cytoplasm ( Rodriguez et al . , 2004 ) , even though the mutants retain intact NLSs and can still bind BARD1 . The localization defect of the BRCT mutants cannot be rescued by Leptomycin B ( LMB ) , which blocks CRM1-dependent export but can instead be rescued by co-expression of BARD1 ( Rodriguez et al . , 2004 ) . In this vein , we tested the above aspects for the BRCT missense mutations and obtained the same results ( Figure 4B–E ) . We consider the BRCT point mutants to be different from truncating mutants for two reasons: first , the majority of the these mutants were expressed at much lower levels than the wt protein presumably due to destabilization , whereas the truncating mutants were not ( Figure 1—figure supplement 2 ) , although frameshift mutations often cause nonsense-mediated decay ( NMD ) of mRNA in vivo; second , the missense mutants may still bind known interaction partners with low affinity or even certain unknown factors that may affect their localization . Collectively , the data lend strong support to the notion that the BRCT mutants may be unable to enter the nucleus and further suggest that the BRCT domain may bind a heretofore unidentified factor that promotes BRCA1 nuclear entry , with BARD1 directly or indirectly involved in the process . 10 . 7554/eLife . 21350 . 013Figure 4 . Rescue of the localization defect but not HR activity of BRCA1 BRCT missense mutants by BARD1 . ( A ) Immunofluorescence staining of wt BRCA1 and representative RING , CC and BRCT mutants . 3xMyc-tagged BRCA1 expression constructs were transiently transfected into U2OS/DR-GFP cells first depleted of endogenous BRCA1 , cells were fixed 48 hr after transfection , and the proteins were stained with an anti-Myc antibody . ( B ) Quantification of the subcellular distribution of wt BRCA1 and a panel of BRCT mutants . cDNA constructs were transfected into U2OS cells and the tagged BRCA1 proteins were stained as in ( A ) . ( C ) Effect of Leptomycin B ( LMB ) on the localization of wt BRCA1 and BRCA1-M1775R . The BRCA1 proteins were expressed and stained as in ( A ) . LMB ( 50 ng/ml ) was added 48 hr after transfection , and cells were incubated with LMB for 12 hr prior to fixation . Staining of endogenous cyclin B in untransfected U2OS cells was used as a positive control . ( D–E ) Rescue of the localization defect of BRCA1 BRCT mutants by BARD1 . 3xMyc-tagged BRCA1 and FLAG-HA-tagged BARD1 were transiently co-expressed in U2OS cells , cells were fixed 48 hr after transfection , and the proteins were stained with Myc and HA antibodies , respectively . Panel D shows the staining patterns of wt and two representative mutants of BRCA1 and the co-expressed BARD1 . Panel E shows the quantification of the results . ( F ) No rescue of the HR defect of the BRCA1 BRCT mutants by BARD1 co-expression . U2OS/DR-GFP cells were depleted of the endogenous BRCA1 for 48 hr and then co-transfected with BRCA1 and BARD1 expression constructs . GFP-positive cells were scored another 52 hr later . The values were normalized against that of wt BRCA1 , which was set as 1 . Data shown are means ± SDs from three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 21350 . 013 To learn to what extent the lack of HR activity of the BRCT mutants is due to mislocalization , we tested their activities when co-expressed with exogenous BARD1 . Co-expression of BARD1 slightly increased the HR activity of some of the mutants , but the overall effect was modest at best ( Figure 4F ) . For example , mutant S1655F showed comparable , if not better , localization than the wt protein ( Figure 4E ) , but still had a threefold lower HR activity . These data suggest that the lack of HDR activity of the BRCT mutants is due to both their localization defect and , perhaps more importantly , their inability to interact with one or more of the three binding partners , namely Abraxas , CtIP and BRIP1 , or with another binding partner yet to be identified . The BRCA1/BARD1 heterodimer exhibits robust ubiquitin E3 ligase activity in vitro ( Hashizume et al . , 2001; Ruffner et al . , 2001 ) . An artificial I26A mutation was isolated in a screen for residues in the RING domain that abrogate E3 ligase activity without affecting BARD1 binding ( Brzovic et al . , 2003 ) . It was first thought that the E3 ligase activity may play an important role in DNA repair and/or tumor suppression; however , mouse cells carrying a I26A knockin allele did not show any significant defect in HR activity ( Reid et al . , 2008 ) , and the mutant protein was still able to suppress mammary tumor development in Brca1 conditional knockout models ( Shakya et al . , 2011 ) . To our surprise , I26A substantially reduced , by about three fold , the coIP of endogenous BARD1 with ectopically expressed BRCA1 ( Figure 5A ) , indicating that this mutation significantly weakens the BRCA1-BARD1 interaction . Thus , I26A is a complex mutation affecting at least two different BRCA1 properties . Functionally , I26A reduced the HR activity of BRCA1 by ~30% ( Figure 5B ) ; although the apparent HR defect caused by I26A was mild , the mutation caused a 67% reduction in colony formation in the presence of cisplatin and virtually a complete loss of resistance to olaparib ( Figure 5C ) . 10 . 7554/eLife . 21350 . 014Figure 5 . Effects of BRCA1 sumoylation and E3 ligase activity on HR and drug resistance . ( A ) Effect of I26A and K119R mutations on BRCA1 binding to BARD1 and other interacting partners . The proteins were transiently expressed in 293T cells and IPed with anti-Myc . ( B ) Quantification of the BARD1-binding capacity of the BRCA1 mutants . Data shown are the means ± SDs of the ratios of BARD1 and BRCA1 band intensities from four independent experiments . **p<0 . 01 . ( C ) HR and SSA activities of the BRCA1 mutants relative to that of the wt protein . Data shown are the means ± SDs from two to six independent experiments for each mutant . ***p<0 . 001 . See Figure 1—source data 1 and Figure 2—source data 1 for details . ( D ) Abilities of the BRCA1 mutants to confer resistance to cisplatin and olaparib . Data shown are the means ± SDs from three independent experiments for both mutants . **p<001 . See Figure 3—source data 1 for details . DOI: http://dx . doi . org/10 . 7554/eLife . 21350 . 014 It has been reported that BRCA1 can be sumoylated and this modification promotes the E3 ligase activity of the BRCA1/BARD1 complex ( Morris et al . , 2009 ) . The same study identified lysine 119 ( K119 ) as a critical residue for BRCA1-SUMO conjugation and the promotion of its E3 ligase activity . Thus , we generated and analyzed a K119R mutation in parallel with the I26A mutant . K119R had no effect on BRCA1 HR , nor did it affect cisplatin resistance ( Figure 5B–C ) ; however , colony formation in the presence of olaparib was reduced by ~30% , suggesting a possible HR-independent role of BRCA1 sumoylation for PARPi resistance . To systematically assess the contribution of various BRCA1 structural elements in HDR and drug resistance , we generated a series of 10 overlapping deletions ( Figure 6A ) . Interestingly , BD1 , which lacks both the RING domain and the NES , was expressed at a much higher level , and despite the complete lack of the RING domain , it still associated with a small but significant amount of BARD1 ( Figure 6B ) ; it showed ~50% activity for both HR and SSA ( Figure 6C ) , which is significantly higher than the activities of either C61G or C64R mutants ( Figure 1D ) . As expected , BD8 , which lacks the CC motif , was unable to bind PALB2 , and BD10 lacking the BRCT domain failed to bind Abraxas and BRIP1 . In line with observations made with the point mutants ( Figures 1D and 2A ) , BD8 showed no HR activity and a threefold increase in SSA , while BD10 displayed similar HR and SSA activities as the vector . Notably , BD4 , lacking both NLSs , still retained ~75% activity in both HR and SSA assays , indicating that a significant amount of BRCA1 can still be recruited into the nucleus . In contrast , BD7 , which retains all known functional domains , showed only ~50% of HR and SSA activities , suggestive of the possible existence of a novel element in the deleted region that is important for DNA repair . 10 . 7554/eLife . 21350 . 015Figure 6 . Functionalities of BRCA1 deletion mutants in HR , SSA and drug resistance . ( A ) Schematic of wt BRCA1 and 10 overlapping deletions generated for this study . ( B ) Capacity of the deletion mutants in binding key interacting partners . The proteins were transiently expressed in 293T cells and IPed with anti-Myc . ( C ) HR and SSA activities of the deletion mutants . See Figure 1—source data 1 and Figure 2—source data 1 for details . ( D ) Levels of cisplatin and olaparib resistance conferred by the deletion mutants . Values presented are means ± SDs from two to four independent experiments for each mutant . See Figure 3—source data 1 for details . DOI: http://dx . doi . org/10 . 7554/eLife . 21350 . 015 Consistent with their lack of HR activity , BD8 and BD10 were completely or nearly completely unable to confer resistance to either drug ( Figure 6D ) . Although BD1 , BD4 and BD7 all displayed 50–75% apparent HR activity in the reporter assay , they supported no more than 20% colony formation . Also , BD9 only supported ~50% of colony formation despite having no defect in the aforementioned protein interactions and possessing greater than 80% apparent HR and SSA activities . These observations demonstrate a positive but non-linear correlation between apparent HR activity and drug resistance . The non-linearity could be due to different effects of the deletions on protein folding , stability , conformation , posttranslational modifications and dynamics in protein-protein interactions; it may also reflect the differences between the two cell lines ( U2OS/DR-GFP vs MDA-MB-436 ) , time frames used or the different nature of I-SceI and drug-induced DNA breaks . Surprisingly , BD6 was twice as effective as the wt protein in supporting colony formation in the presence of olaparib but less effective in the presence of cisplatin , indicating a significant difference in the mechanisms of resistance to the two drugs despite a common requirement for HR .
In this study , we systematically assessed HR and SSA activities of 32 natural variants , including 11-patient-derived missense mutations that affect known protein-protein interactions and the top 22-patient-associated missense variants , as well as 10 overlapping deletion mutations that span the entire length of the protein . Our data clearly demonstrate that the HR function of BRCA1 requires the RING , CC and BRCT domains . Interestingly , deletion of residues 1056–1357 ( BD7 ) caused ~50% reduction in both HR and SSA activity without affecting the binding of any of the above known partners . This segment of BRCA1 does not contain any recognizable domains and has not been implicated in HR before , suggesting the existence of a novel binding partner or regulatory mechanism . Together , these results establish a functional landscape of BRCA1 for its HDR function . The RING domain of BRCA1 binds to the RING domain of BARD1 , and the BRCA1/BARD1 heterodimer formation is critical for their stability , their E3 ubiquitin ligase activity and the nuclear retention of BRCA1 ( Fabbro et al . , 2002; Hashizume et al . , 2001; Ruffner et al . , 2001 ) . Conditional knock out of either Brca1 or Bard1 in murine mammary epithelial cells led to the development of mammary carcinomas that are indistinguishable from each other ( Shakya et al . , 2008 ) , and the BRCA1 C61G mutant that abrogates BARD1 binding failed to suppress mammary tumor development in mice ( Drost et al . , 2011 ) . Moreover , the I26A mutation that abolishes its E3 ligase activity did not affect HR activity or mitomycin C ( MMC ) sensitivity ( Reid et al . , 2008 ) , nor did it affect tumor suppression in mice ( Shakya et al . , 2011 ) . These findings suggested that the BARD1-binding function of the RING domain , rather than its E3 ligase activity , plays a key role in BRCA1 HR and tumor suppression activity . In this regard , our findings on the I26A mutation are worth noting . First , it caused a significant impairment in BARD1 binding ( Figure 5A ) ; second , it moderately reduced HR efficiency ( Figure 5B ) ; and third , it caused sensitivity to both cisplatin and olaparib under the condition used ( Figure 5C ) . The reduced BARD1 binding is consistent with the fact that I26A mutant mouse cells had substantially reduced amount of both BRCA1 and BARD1 ( Reid et al . , 2008 ) . Also , a recent report showed that HeLa cells expressing only BRCA1-I26A were similarly sensitive to olaparib as BRCA1 knockdown cells ( Densham et al . , 2016 ) . Thus , it is safe to conclude that the I26A mutation affects HR and drug resistance in human cells . The discrepancy in observed HR activities and drug sensitivities in mouse ES cells and human cancer cells may be explained by the different tolerance for partial loss of BARD1 binding in the two systems or the presence of other mutations in cancer cells . At the same time , although I26A mutant mouse cells showed no defect in apparent HR activity or MMC resistance , reduced gene targeting efficiency and increased chromosomal abnormalities after MMC treatment were noted ( Reid et al . , 2008 ) , indicative of a possible defect in HR in mouse cells as well . Despite the discrepancy , it should be noted that our results do not contradict the argument that the E3 ligase activity of BRCA1 is dispensable for tumor suppression ( Shakya et al . , 2011 ) . Mutations of the BRCT domain dramatically abolish HR activity , and the mechanism is also complex . First , this domain directly interacts with at least three proteins , Abraxas , a scaffold protein ( Wang et al . , 2007 ) , BRIP1 , a DNA helicase ( Cantor et al . , 2001 ) , and CtIP , an endonuclease involved in resection ( Sartori et al . , 2007 ) . It has recently been shown that CtIP binding to BRCA1 is dispensable for CtIP-mediated DNA resection ( Polato et al . , 2014; Reczek et al . , 2013 ) , yet to what extent BRCA1 HR activity depends on Abraxas and BRIP1 remains unclear . Second , mutation of the BRCT domain causes a major defect in the nuclear accumulation of BRCA1 ( Figure 4A–B ) , supporting a key role of the BRCT domain for BRCA1 nuclear entry . Our finding that BARD1 can significantly rescue nuclear localization of BRCT mutants but not their HR defects ( Figure 4F ) also supports an important and direct role of the BRCT domain for BRCA1 DNA repair activity . Collectively , our data demonstrate the dual role of the BRCT domain and underscore a complex regulation of BRCA1 localization and repair function by the RING and BRCT domains and their binding partners ( Figure 7A ) . 10 . 7554/eLife . 21350 . 016Figure 7 . Summary models of the regulation and function of BRCA1 in HR and SSA . ( A ) Roles of BRCA1 structural elements and binding partners on its nuclear localization and HDR ( HR and SSA ) activities . In brief , the NLSs and BRCT domain both promote BRCA1 nuclear entry , whereas the NES mediates BRCA1 export from the nucleus . BARD1 bound to the RING domain shields the NES of BRCA1 thereby promoting its nuclear retention . The RING and BRCT domains are required for both HR and SSA , whereas the CC domain promotes HR but inhibits or prevents SSA through its binding to PALB2 . ( B ) A model of the BRCA1-PALB2-BRCA2 axis in the regulation of HR and SSA . Following DSB formation , the BRCA1/BARD1 complex was recruited to DNA damage sites by Abraxas and perhaps another yet to be defined factor ( s ) . The presence of BRCA1 at damage sites promotes resection and inhibits non-homologous end joining ( NHEJ ) , at least in part , by counteracting the resection-inhibiting activity of the 53BP1-RIF1 complex . At the same time , BRCA1 helps recruit the PALB2/BRCA2 complex , which displaces RPA from the resected ssDNA ends and loads RAD51 to initiate HR . When PALB2 or BRCA2 is lost or when the direct interactions in the BRCA1-PALB2-BRCA2 axis are disrupted , RAD52 binds to resected ends and mediates SSA when homologous sequences are available , leading to genomic deletions . It is unclear whether PALB2 or BRCA2 can directly suppress the binding of RAD52 to ssDNA or inhibiting its strand annealing activity . DOI: http://dx . doi . org/10 . 7554/eLife . 21350 . 016 Consistent with early observations made in mouse cells ( Stark et al . , 2004 ) , BRCA1 depletion reduced the efficiency of both HR and SSA , whereas depletion of BRCA2 led to severe loss of HR ( data not shown ) but increased SSA ( Figure 2B ) . As efficient resection is necessary for both repair mechanisms , the data is consistent with the notion that BRCA1 plays a key role in resection ( Bunting et al . , 2010 ) . Importantly , we found that depletion of PALB2 showed similar , if not more dramatic , induction of SSA as did BRCA2 depletion ( Figure 2B ) . Moreover , all three point mutations in BRCA1 that affect PALB2 binding led to increased SSA usage ( Figure 2A ) , and two different PALB2 mutants with greatly impaired BRCA1-binding capacity both failed to suppress the upregulation of SSA in PALB2-depleted cells ( Figure 2F ) . These results clearly demonstrate that the direct interaction between BRCA1 and PALB2 is required for suppressing , or at least preventing , SSA , while promoting HR ( Figure 7A–B ) . As such , our results support two critical roles of BRCA1 function in DSB repair , promoting resection and recruiting/stimulating PALB2 , which then channels the ssDNA down the HR path ( Figure 7B ) . Based on the available data , PALB2 is the key partner of BRCA1 specifically for its HR function and the primary switch point , acting upstream of BRCA2 , for HDR pathways following resection . VUSs are commonly found during clinical genetics tests; however , their biological and clinical significance is often difficult to ascribe with confidence , and this uncertainty poses significant challenges for both clinicians and patients . Our results demonstrate that relatively common VUSs outside the RING and BRCT domains do not significantly affect the HDR activity of BRCA1 . Our findings are mostly consistent with several previous studies ( Table 2 ) , including the above noted mouse ES cell-based study that covered 6 of the top 22 variants ( Bouwman et al . , 2013 ) ; on the rare PALB2-binding mutants L1407P and M1411T , however , our results showed that they are severely defective in HR ( Figure 1D ) and conferring drug resistance ( Figure 3C ) , whereas their effects in mouse cells were less pronounced . Moreover , M1400V , which was found to be neutral in the mouse cell system , shows clear sensitivity to olaparib treatment in the human cell line . This discrepancy could again be due to a difference in the thresholds of human and mouse cells to tolerate partial loss of BRCA1-PALB2 interaction or the presence of other mutations in the human cancer cells . Platinum-based therapeutics have been commonly used for treatment of ovarian cancer , and olaparib has recently been approved for the treatment of BRCA1/2 mutant ovarian cancer , with multiple trials on breast and other cancers ongoing or recently completed . These agents generate lesions that block or collapse DNA replication forks , leading to DSBs that require HR and therefore BRCA1/2 for repair ( Lord and Ashworth , 2016 ) . Under the condition used , olaparib appeared to have a better selectivity in killing MDA-MB-436 cells expressing mutant BRCA1 proteins . Although this could be due to a slightly lower than optimal concentration of cisplatin used , the possibility that cisplatin is intrinsically less selective than olaparib cannot be ruled out . Based on data from this study and other recent studies , patients with relatively common missense mutations in the RING and BRCT domains , such as C61G and A1708E , respectively , or rare mutations in the CC motif , such as L1407P or M1411T , are likely to respond to platinum- or PARPi-based therapies , unless resistance mechanisms have already occurred prior to treatment or are induced by the treatment ( Lord and Ashworth , 2013 ) . However , except for C61G and A1708E , all other top 22 missense variants are fully or mostly functional both in HR and in conferring cisplatin and olaparib resistance ( Figure 3C ) , which corroborates with clinical genetics evidence to further cast doubts on their pathogenicity and diminish the likelihood of a positive response of the patients to the above regimens . Based on the results of our deletion analyses and other available data , it is unlikely for any VUS outside the RING , CC and BRCT domains to be strongly pathogenic or to elicit high sensitivity to platinum or PARPi therapies ( unless it destabilizes the transcript or protein ) . However , the region deleted in BD7 ( residues 1056–1 , 307 ) may harbor VUS that have intermediate effects .
U2OS/DR-GFP HR reporter cells were described before ( Nakanishi et al . , 2005; Xia et al . , 2006 ) . U2OS/SA-GFP SSA reporter cells ( Gunn and Stark , 2012 ) were a gift from Dr . Jeremy Stark . U2OS and 293T cell lines were purchased from American Type Culture Collection ( ATCC ) . These four cell lines were cultured in Dulbecco’s Modified Eagle’s Medium ( DMEM ) supplemented with 10% fetal bovine serum ( FBS ) and 1x Penicillin-Streptomycin . MDA-MD-436 cells was cultured in DMEM/F12 ( 1:1 ) supplemented with 10% FBS and 1x Penicillin-Streptomycin . All cells were cultured in a humidified chamber with 5% CO2 at 37°C . Mycoplasma was not tested , but the cells have been cultured in the presence of Plasmocin ( ant-mpt , InvivoGen ) to eliminate potential mycoplasma contamination . No commonly misidentified cell lines were used . All BRCA1 expression constructs were based on pcDNA-3xMyc-BRCA1 ( Chen et al . , 1998 ) . The pOZ-FH-C1-PALB2 vector expressing FLAG-HA-tagged PALB2 was described before ( Ma et al . , 2012 ) . The pOZ-FH-N-BARD1 expressing FLAG-HA-tagged BARD1 was described before ( Greenberg et al . , 2006 ) . All mutations or deletions were generated through site-directed mutagenesis following the QuikChange protocol ( Agilent Technologies ) . For testing protein-protein interaction , BRCA1 expression constructs were transfected into 293T cells plated at a density of 5 × 105 cells per well in six-well plates and transfected with 2 μg of plasmid per well using FuGENE HD or X-tremeGENE 9 XP . Cells were collected 30 hr after transfection and lysed with 350 µl of NETNG-300 ( 300 mM NaCl , 1 mM EDTA , 20 mM Tris-HCl , 0 . 5% Nonidet P-40% and 10% Glycerol ) containing Complete protease inhibitor cocktail ( Roche ) . The 3xMyc-tagged BRCA1 proteins were IPed for 3–4 hr with anti-Myc ( 9E10 , Covance ) and protein A-agarose beads ( Roche ) . For western blotting analyses , proteins were resolved on 4–12% Tris-Glycine SDS gels , transferred onto nitrocellulose membranes and probed with the following antibodies- Myc ( 9E10 , Covance ) , PALB2 ( M11 ) ( Xia et al . , 2006 ) , BARD1 ( H300 , Santa Cruz ) , BRIP1 ( a gift from Dr . Sharon Cantor , University of Massachusetts Medical School ) , Abraxas ( ab139191 , AbCam ) . The secondary antibodies used were Horseradish peroxidase ( HRP ) -conjugated sheep anti-mouse IgG ( NA931V , GE Healthcare ) and donkey anti-rabbit IgG ( NA9340V , GE Healthcare ) . Immobilon Western Chemiluminescent HRP Substrate ( Millipore ) was used to develop the blots . To measure the HR and SSA activities of the BRCA1 and PALB2 variants , U2OS/DR-GFP and U2OS/SA-GFP cells were plated in 10 cm dish at 1 . 2 × 106 cells per dish and allowed to adapt overnight . Cells were transfected with an siRNA targeting BRCA1 3’-UTR using Lipofectamine RNAiMax ( Life Technologies ) ( 10 nM final concentration of siRNA and 12 µl of RNAiMax per plate ) . Cells were split into six-well plates ( 200 , 000 cells per well ) 30 hr after transfection . After another 18 hr , cells were co-transfected with BRCA1 or PALB2 expression constructs ( 1 µg ) and the I-SceI expression plasmid pCBAsce ( 1 . 5 µg ) using 6 µl of X-tremeGene 9 ( Roche ) . Cells were harvested 54 hr post the second transfection , and GFP-positive cells were counted by fluorescence-assisted cell sorting ( FACS ) . The sense strand sequence of the siRNA is GGAUCGAUUAUGUGACUUAdTdT . To measure SSA activity after BRCA2 , PALB2 or RAD52 knockdown with different siRNAs , U2OS/SA-GFP cells were seeded into six-well plates at density of 175 , 000 cells per well and transfected with siRNAs ~18 hr later using using Lipofectamine RNAiMax ( Life Technologies ) ( 10 nM final concentration of each siRNA and 3 µl of RNAiMax per well ) . Media were refreshed 24 hr post transfection , and cells were transfected again with pCBAcse with X-tremeGene 9 another 24 hr later ( 48 hr after the first transfection ) . Cells were collected and subjected to FACS analysis ~52 hr after the second transfection . See Supplementary file 1 for the sequences of the siRNAs used . Cells were seeded onto coverslips in 12-well plates the day before transfection . Cells were transfected with 1 µg BRCA1 expression constructs using 2 . 5 µl of FuGENE 6 or X-tremeGENE 9 ( Roche ) . At 48 hr after transfection , cells were fixed with 3% ( w/v ) paraformaldehyde ( in PBS with 300 mM sucrose ) for 10 min at room temperature , permeabilized with 0 . 5% Triton X-100 ( in PBS ) and then sequentially incubated with primary and secondary antibodies ( diluted in PBS containing 5% goat serum ) for 1 hr each at 37°C . Each of the above steps was followed by three PBS washes . After staining , coverslips were mounted onto glass slides with VECTASHIELD Mounting Medium with DAPI ( Vector Labs ) and observed using a fluorescent microscope . The primary antibodies used were anti-Myc ( 9E10 , Covance ) and anti-BRCA1 ( #07–434 , Millipore ) . MDA-MB-436 cells were seeded into six-well plates at a density of 1 × 106 cells per well the day before transfection . Cells were transfected with 1 µg of empty vector or 2 µg of BRCA1 expression vectors using 6 µl of X-tremeGENE 9 ( Roche ) . Cells were trypsinized 36 hr after transfection and reseeded into 10 cm plates at 100 , 000 cells per plate in a volume of 10 ml . Another 16 hr later , 1 ml of 10X G418 ( 6 mg/ml dissolved in the same culture medium ) or 10X G418 containing 2 . 2 μM cisplatin or olaparib was added to each plate . The final concentrations of G418 , cisplatin and olaparib were 550 µg/ml , 200 nM and 200 nM , respectively . All experiments were performed in duplicate plates for each construct . For cells selected with G418 alone , 15 days after selection , one plate was trypsinized and cells counted with a Vi-CELL Cell Counter ( Beckman Coulter ) to determine viable cell number , and the other plate was stained with Crystal Violet ( 0 . 5% w/v in 95% ethanol , 5% acetic acid ) to count the number of colonies . For cells selected with both G418 and cisplatin , both plates were stained 21 days after selection . For cells subjected to G418 and olaparib double selection , both plates were stained 28 days after selection . Colonies with approximately 50 cells or more were counted , and the numbers were normalized against the number of viable cells on corresponding plates with G418 alone . The vast majority of constructs were measured by three or more independent experiments ( each with duplicate sets of plates ) , with the only exceptions being when the first two experiments produced nearly identical results or were carried out using two independent plasmids . We assessed the sensitivity and specificity of our assays using the Receiver Operating Characteristics ( ROC ) analysis to simultaneously capture the maximum sensitivity and specificity of the assays with 95% confidence interval as previously described ( Guidugli et al . , 2013 ) . The sensitivity of each assay was defined as the ability to detect variants retaining wt activity of BRCA1 in the respective assays and specificity the ability to detect variants showing loss of BRCA1 function . For the calculation of specificity and sensitivity of our assays , we compared the values of benign variants ( Align-GVD grade of C0 and IARC classification of 1 ) and pathogenic variants ( Align-GVD grade of C35-C65 and IARC classification of 5 ) that have been confirmed by previous functional studies ( Tables 1 and 2 ) . We also included values of the wt cDNA in the benign group and the vector as well as a bona fide pathogenic mutation 5055delG ( p . Val1646Serfs or L1657STOP ) in the pathogenic group . Threshold of each assays were defined individually . To rule out variability in the assays , we also calculated the sensitivity and specificity using the highest observed values for the pathogenic variants and the lowest observed values for the benign variants with the cut off used to define either the upper or lower limit of the threshold . For each of the assays , dotted lines were drawn to represent the cut off threshold that captures the specificity and sensitivity of the assays at 100% , respectively . Analysis was not performed for the SSA assay as elevated SSA activity is associated with PALB2-binding mutants , which has yet to be defined , while loss of SSA is associated with BRCT and RING domain mutations . All analyses were conducted using GraphPad Prism 5 . 0 ( GraphPad Software ) . Raw data were normalized against values of wt BRCA1 , wt PALB2 , control siRNA ( NSC1 ) or the empty vector , where applicable . Statistical significance was calculated using Students’ t test . p-Values smaller than 0 . 05 were considered significant . | Genes are the instruction manuals of life and contain the information needed to build the proteins that keep cells alive . Over time , genes can accumulate errors or mutations and eventually become faulty , which can lead to diseases like cancer . Sometimes mutations can be passed on through generations and increase the chances of getting cancer . The BRCA1 gene , for example , provides instructions for making a protein that helps to repair or remove damaged DNA and stops cells from growing uncontrollably . When the BRCA1 gene becomes faulty , cells could continue to grow with damaged DNA . This makes it more likely for cancer to develop , especially breast cancer and ovarian cancer . However , not all changes in BRCA1 gene cause the protein to become faulty or lead to cancer . In fact , about 30% of BRCA1 gene changes identified by genetic tests are referred to as ‘variants of uncertain clinical significance’ , meaning that it is not clear if these variants are indeed mutations that could affect the clinical outcome of the people that carry them . Software predictions based largely on patient data have categorized many of these variants as not cancer-causing , but the majority still need to be experimentally tested and confirmed . Many studies have tried to determine the effect of selected variants on the BRCA1 protein , but a complete picture remains lacking . Now , Anantha et al . have tested the top 22 common variants in the BRCA1 gene , some of which had known effects and some did not . The study tested how these variants affect the ability of the protein to repair damaged DNA and the efficacy of chemotherapies targeting cancer cells with a DNA repair defect . The experiments revealed that three specific parts of the protein must remain intact in order for the protein to carry out this activity , i . e . mutations that affect these three areas are likely to cause cancer and also make cancer cells vulnerable to these chemotherapies . Anantha et al . also generated a series of 10 artificially shortened BRCA1 proteins , each missing a specific part , to determine the possible effects of other variants in those missing parts . Together the findings reveal previously unknown effects of certain variants that are commonly seen in cancer patients as well new insights into how the BRCA1 protein repairs DNA . The next step will be to assess rarer variants where little data is available . A better understanding of how these variants affect DNA repair and drug response will help to improve the genetic counseling and treatment of patients with breast cancer and ovarian cancer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"cancer",
"biology"
] | 2017 | Functional and mutational landscapes of BRCA1 for homology-directed repair and therapy resistance |
Immersive virtual reality ( VR ) enables naturalistic neuroscientific studies while maintaining experimental control , but dynamic and interactive stimuli pose methodological challenges . We here probed the link between emotional arousal , a fundamental property of affective experience , and parieto-occipital alpha power under naturalistic stimulation: 37 young healthy adults completed an immersive VR experience , which included rollercoaster rides , while their EEG was recorded . They then continuously rated their subjective emotional arousal while viewing a replay of their experience . The association between emotional arousal and parieto-occipital alpha power was tested and confirmed by ( 1 ) decomposing the continuous EEG signal while maximizing the comodulation between alpha power and arousal ratings and by ( 2 ) decoding periods of high and low arousal with discriminative common spatial patterns and a long short-term memory recurrent neural network . We successfully combine EEG and a naturalistic immersive VR experience to extend previous findings on the neurophysiology of emotional arousal towards real-world neuroscience .
Emotions are subjective , physiological , and behavioural responses to personally meaningful external stimuli ( Mauss and Robinson , 2009 ) or self-generated mental states ( e . g . , memories; Damasio et al . , 2000 ) and underlie our experience of the world ( James , 1884; James , 1890; Seth , 2013 ) . Emotions are crucial for physical and mental health ( Gross and Muñoz , 1995 ) and their investigation has long been at the core of experimental psychology ( Wundt and Judd , 1897 ) . Dimensional accounts conceptualize emotions along the two axes of valence and arousal ( Duffy , 1957; Kuppens et al . , 2013; Russell , 1980; Russell and Barrett , 1999; Wundt and Judd , 1897 ) : valence differentiates states of pleasure and displeasure , while emotional arousal describes the degree of activation or intensity that accompanies an emotional state . [Different types of arousal have been proposed and investigated , such as sexual , autonomic , emotional ( Russell , 1980 ) ; also in the context of altered states of consciousness , for example , through anaesthesia or sleep . They may share psychological ( e . g . , increase in sensorimotor and emotional reactivity; Pfaff et al . , 2012 ) and physiological aspects ( e . g . , sympathetic activation ) but are not synonymous . We here explicitly refer to arousal in the context of the subjective experience of emotions] . Emotions have been linked to activity in the autonomic ( ANS ) and the central nervous system ( Dalgleish , 2004 ) . It has thereby been difficult to consistently associate individual , discrete emotion categories with specific response patterns in the ANS ( Kragel and Labar , 2013; Kreibig , 2010; Siegel et al . , 2018 ) or in distinct brain regions ( Lindquist et al . , 2012; but Saarimäki et al . , 2016 ) . Rather , emotions seem to be dynamically implemented by sets of brain regions and bodily activations that are involved in basic , also non-emotional psychological operations ( i . e . , ‘psychological primitives’; Lindquist et al . , 2012 ) . In this view , humans are typically in fluctuating states of pleasant or unpleasant arousal ( ‘core affect’; Russell and Barrett , 1999; Lindquist , 2013 ) , which can be influenced by external stimuli . Emotional arousal could thereby be a ‘common currency’ to compare different stimuli or events ( Lindquist , 2013 ) and represent fundamental neural processes that underlie a variety of emotions ( Wilson-Mendenhall et al . , 2013 ) . It can fluctuate quickly – on the order of minutes ( Kuppens et al . , 2010 ) or seconds ( Mikutta et al . , 2012 ) – and has been connected to ANS activity , as measured by pupil diameter ( Bradley et al . , 2008 ) or skin conductance ( Bach et al . , 2010 ) . At the brain level , emotional arousal was linked to lower alpha power , particularly over parietal electrodes ( Luft and Bhattacharya , 2015; Koelstra et al . , 2012 ) . The parieto-occipital alpha rhythm , typically oscillating in the frequency range of 8–13 Hz , is the dominant EEG rhythm in awake adults with eyes closed ( Berger , 1929 ) , where it varies with vigilance ( Olbrich et al . , 2009 ) . However , in tasks of visual processing ( i . e . , with eyes open ) , parieto-occipital alpha power was linked to active attentional processes ( e . g . , distractor suppression; Kelly et al . , 2006; Klimesch , 2012 ) or , more generally , to functional inhibition for information gating ( Jensen and Mazaheri , 2010 ) . Physiologically , alpha oscillations were associated with large-scale synchronization of neuronal activity ( Buzsáki , 2006 ) and metabolic deactivation ( Moosmann et al . , 2003 ) . In sum , bodily responses interact in complex ways across situations , and activity in the brain is central for emotions and their subjective component ( Barrett , 2017; Seth , 2013 ) . As arousal is a fundamental property not only of emotions but of subjective experience in general ( Adolphs et al . , 2019 ) , an investigation of its neurophysiology , reflected in neural oscillations , is essential to understanding the biology of the mind . Studies that investigated emotions or emotional arousal in laboratory environments typically used static images . For example , more emotionally arousing relative to less emotionally arousing ( e . g . , neutral ) pictures were associated with an event-related desynchronization , that is , a decrease in the power of alpha oscillations in posterior channels ( De Cesarei and Codispoti , 2011; Schubring and Schupp , 2019; but Uusberg et al . , 2013 ) . In a study , in which emotional arousal was induced through pictures and music , blocks of higher emotional arousal were associated with decreased alpha power compared to blocks of lower emotional arousal ( Luft and Bhattacharya , 2015 ) . However , emotion-eliciting content that is repeatedly presented in trials creates an artificial experience for participants ( Bridwell et al . , 2018 ) ; it hardly resembles natural human behaviour and its ( neuro- ) physiology , which unfolds over multiple continuous timescales ( Huk et al . , 2018 ) . Moreover , such presentations lack a sense of emotional continuity . External events often do not appear suddenly but are embedded in an enduring sequence , in which emotions build up and dissipate . Real-life scenarios also include anticipatory aspects where emotional components can be amplified or even suppressed , thus rendering the relationship between the corresponding neuronal events and subjective experience more complex than the one typically studied with randomized or partitioned presentations of visual or auditory stimuli . Virtual reality ( VR ) technology – particularly immersive VR , in which the user is completely surrounded by the virtual environment – affords the creation and presentation of computer-generated scenarios that are contextually rich and engaging ( Diemer et al . , 2015 ) . As more naturalistic ( i . e . , dynamic , interactive , and less decontextualized ) experiments allow to study the brain under conditions it was optimized for ( Gibson , 1978; Hasson et al . , 2020 ) , their findings may more readily generalize to real-world circumstances and provide better models of the brain ( Matusz et al . , 2019; Shamay-Tsoory and Mendelsohn , 2019 ) . In this study , we aimed to link subjective emotional arousal with alpha power in a naturalistic setting . Participants completed an immersive VR experience that included virtual rollercoaster rides while their EEG was recorded . They then continuously rated their emotional arousal while viewing a replay of their previous experience ( McCall et al . , 2015 ) . To tackle the challenges of data acquired in naturalistic settings and with continuous stimuli , we made use of recent advances in signal processing and statistical modelling: spatial filtering methods ( originally developed for brain-computer interfaces [BCIs]; Blankertz et al . , 2008 ) have recently gained popularity in cognitive neuroscience ( Cohen , 2018; Zuure and Cohen , 2020 ) , where they have been used to analyze continuous data collected in naturalistic experiments , for example , to find inter-subject correlations in neuroimaging data of participants watching the same movie ( Dmochowski et al . , 2012; Gaebler et al . , 2014 ) . For the present experiment , two spatial filtering methods were applied to link alpha power and subjective emotional arousal: source power comodulation ( SPoC; Dähne et al . , 2014 ) and common spatial patterns ( CSP; Blankertz et al . , 2008; Ramoser et al . , 2000 ) . SPoC is a supervised regression approach , in which a target variable ( here: subjective emotional arousal ) guides the extraction of relevant M/EEG oscillatory components ( here: alpha power ) . SPoC has been used to predict single-trial reaction times from alpha power in a hand motor task ( Meinel et al . , 2016 ) , muscular contraction from beta power ( Sabbagh et al . , 2020 ) , and difficulty levels of a video game from theta and alpha power ( Naumann et al . , 2016 ) . CSP is used to decompose a multivariate signal into components that maximize the difference in variance between distinct classes ( here: periods of high and low emotional arousal ) . CSP thereby allows optimizing the extraction of power-based features from oscillatory signals , which can then be applied for training classifiers to solve binary or categorical prediction problems . CSP is being used with EEG for BCI ( Blankertz et al . , 2008 ) or to decode workload ( Schultze-Kraft et al . , 2016 ) . In addition to M/EEG-specific spatial filtering methods , non-linear machine learning methods are suited for the analysis of continuous , multidimensional recordings from naturalistic experiments . Deep neural networks transform high-dimensional data into target output variables ( here: different states of emotional arousal ) by finding statistical invariances and hidden representations in the input ( Goodfellow et al . , 2016; LeCun et al . , 2015; Schmidhuber , 2015 ) . For time-sequential data , long short-term memory ( LSTM ) recurrent neural networks ( RNNs ) are particularly suited ( Greff et al . , 2017; Hochreiter and Schmidhuber , 1995; Hochreiter and Schmidhuber , 1997 ) . Via nonlinear gating units , the LSTM determines which information flows in and out of the memory cell in order to find long- and short-term dependencies over time . LSTMs have been successfully applied for speech recognition ( Graves et al . , 2013 ) , language translation ( Luong et al . , 2015 ) , or scene analysis in videos ( Donahue et al . , 2015 ) , but also to detect emotions in speech and facial expressions ( Wöllmer et al . , 2010; Wöllmer et al . , 2008 ) or workload in EEG ( Bashivan et al . , 2016; Hefron et al . , 2017 ) . In comparison to other deep learning methods , LSTMs are ‘quick learners’ due to their efficient gradient flow and thus suitable for the continuous and sparse data recorded under naturalistic stimulation with VR . The present study tested the hypothesis of a negative association between parieto-occipital alpha power and subjective emotional arousal under dynamic and interactive stimulation . Combining immersive VR and EEG , this study aimed to ( 1 ) induce variance in emotional arousal in a naturalistic setting and ( 2 ) capture the temporally evolving and subjective nature of emotional arousal via continuous ratings in order to ( 3 ) assess their link to oscillations of brain activity in the alpha frequency range . The link between subjective emotional arousal and alpha power was then tested by decoding the former from the latter using the three complementary analysis techniques SPoC , CSP , and LSTM .
Forty-five healthy young participants were recruited via the participant database at the Berlin School of Mind and Brain ( an adaption of ORSEE; Greiner , 2015 ) . Previous studies on the relationship between emotional arousal and neural oscillations reported samples of 19–32 subjects ( e . g . , Koelstra et al . , 2012; Luft and Bhattacharya , 2015 ) . We recruited more participants to compensate for anticipated dropouts due to the VR setup and to ensure a robust estimate of the model performances . Inclusion criteria were right-handedness , normal or corrected-to-normal vision , proficiency in German , no ( self-reported ) psychiatric , or neurological diagnoses in the past 10 years , and less than 3 hr of experience with VR . Participants were requested to not drink coffee or other stimulants 1 hr before coming to the lab . The experiment took ~2 . 5 hr , and participants were reimbursed with 9€ per hour . They signed informed consent before their participation , and the study was approved by the Ethics Committee of the Department of Psychology at the Humboldt-Universität zu Berlin . The experiment was conducted in a quiet room , in which the temperature was kept constant at 24°C . After each VR experience , participants watched a 2D recording ( recorded using OBS Studio , https://obsproject . com/ ) of their experience on a virtual screen ( SteamVR’s ‘view desktop’ feature ) , that is , without removing the HMD . They recalled and continuously rated their emotional arousal by turning a dial ( PowerMate USB , Griffin Technology , Corona , CA; sampling frequency: 50 Hz ) , with which they manipulated a vertical rating bar , displayed next to the video , ranging from low ( 0 ) to high ( 100 ) in 50 discrete steps ( McCall et al . , 2015; see Figure 1B ) . The exact formulation was ‘When we show you the video , please state continuously how emotionally arousing or exciting the particular moment during the VR experience was’ ( German: ‘Wenn wir dir das Video zeigen , gebe bitte durchgehend an , wie emotional erregend , bzw . aufregend der jeweilige Moment während der VR Erfahrung war’ ) . To present the playback video and the rating bar , a custom script written in Processing ( v3 . 0 ) was used . Participants came to the lab and filled in the pre-test questionnaires . After the torso and limb electrodes had been attached , participants completed a heartbeat guessing task ( Schandry , 1981 ) to assess inter-individual differences in interoceptive accuracy ( the results of peripheral physiology and interoception will be reported elsewhere ) . Then , the EEG cap was attached , and the HMD was carefully placed on top of it . To prevent or minimize ( e . g . , movement-related ) artefacts , customized cushions were placed below the straps of the VR headset to reduce the contact with the EEG sensors . In addition , the VR experience took place while seated and without full body movements ( participants were asked to keep their feet and arms still during the recordings ) . A white grid was presented in the HMD to ensure that the participants’ vision was clear . They then completed a 10 min resting-state phase ( 5 min eyes open , 5 min eyes closed ) , before experiencing the first VR episode , which consisted of the two virtual rollercoaster rides and the intermediate break: first the ‘Space’ and then , after the break , the ‘Andes’ rollercoaster . In the subsequent rating phase , they recalled and continuously rated their emotional arousal while viewing a 2D recording of their experience . Importantly , each participant completed the VR episode ( plus rating ) twice: once while not moving the head ( nomov condition ) and once while freely moving the head ( mov condition ) during the VR experience . The sequence of the movement conditions was counterbalanced across participants ( n = 19 with nomov condition first ) . At the end of the experiment , participants completed two additional questionnaires ( the SUS and the questionnaire on subjective feelings of presence and valence during the virtual rollercoaster rides ) before they were debriefed . To exclude effects related to the on- or offset of the rollercoasters , data recorded during the first and the last 2 . 5 s of each rollercoaster were removed and the inter-individually slightly variable break was cropped to 30 s . The immersive VR experience that was analysed thus consisted of two time series of 270 s length each per participant ( nomov and mov ) . Emotional arousal ratings were resampled to 1 Hz by averaging non-overlapping sliding windows , yielding one arousal value per second . For the classification analyses , ratings were divided by a tertile split into three distinct classes of arousal ratings ( low , medium , high ) per participant . For the binary classification ( high vs . low arousal ) , the medium arousal ratings were discarded . Our main hypothesis was that EEG-derived power in the alpha frequency range allows the discrimination between different states of arousal . To calculate alpha power , we adopted spatio-spectral decomposition ( SSD; Nikulin et al . , 2011 ) which extracts oscillatory sources from a set of mixed signals . Based on generalized eigenvalue decomposition , it finds the linear filters that maximize the signal in a specific frequency band and minimize noise in neighbouring frequency bands . Preprocessing with SSD has been previously shown to increase classification accuracy in BCI applications ( Haufe et al . , 2014a ) . The alpha frequency range is typically fixed between 8 and 13 Hz . The individual alpha peak frequency , however , varies intra- and inter-individually , for example , with age or cognitive demand ( Haegens et al . , 2014; Mierau et al . , 2017 ) . To detect each participant’s individual peak of alpha oscillations for the SSD , ( 1 ) the power spectral density ( PSD ) of each channel was calculated using Welch’s method ( segmentlength=5s∗samplingfrequency [i . e . , 250 Hz] with 50% overlap ) in MATLAB ( pwelch function ) . ( 2 ) To disentangle the power contribution of the 1/f aperiodic signal from the periodic component of interest ( i . e . , alpha ) , the MATLAB wrapper of the FOOOF toolbox ( v0 . 1 . 1; Haller et al . , 2018; frequency range: 0–40 Hz , peak width range: 1–12 Hz , no minimum peak amplitude , threshold of two SDs above the noise of the flattened spectrum ) was used . The maximum power value in the 8–13 Hz range was considered the individual alpha peak frequency αi , on which the SSD bands of interest were defined ( bandpass signal αi ± 2 Hz , bandstop noise αi ± 3 Hz , bandpass noise αi ± 4 Hz ) . The entire procedure was separately applied to the nomov and the mov condition to account for potential peak variability ( Haegens et al . , 2014; Mierau et al . , 2017 ) . SSD was then computed based on these peaks . A summary of the resulting individual alpha peak frequencies can be found in Figure 2—source data 1 . Figure 2 shows the averaged power spectrum across all participants and electrodes . A clearly defined peak in the alpha frequency range is discernible for both conditions ( nomov , mov ) as well as for states of high and low emotional arousal . To test the hypothesis that alpha power in the EEG negatively covaries with the continuous level of subjective emotional arousal , SPoC ( as described in Dähne et al . , 2014; NB: throughout the paper , ‘SPoC’ refers to SPoCλ ) was applied to EEG data composed of the selected SSD components and filtered around the central individual alpha peak . Formally , SPoC is an extension of CSP ( see below ) for regression-like decoding of a continuous target variable . The information contained in the target variable is used to guide the decomposition of neural components that is correlated or anti-correlated with it ( Dähne et al . , 2014 ) . SPoC has been shown to outperform more conventional approaches to relate neural time series to continuous behavioural variables ( e . g . , correlating power extracted in sensor space and/or after blind source separation methods ) , which also suffer from additional drawbacks ( e . g . , lack of patterns’ interpretability and lack of adherence to the M/EEG generative model; for details , see Dähne et al . , 2014 ) . The supervised decomposition procedure takes the variable z as target , which comprises the continuous arousal ratings ( normalized and mean-centred; 270 s per participant ) . To reach the same temporal resolution as z ( i . e . , 1 Hz ) , EEG data were epoched into 270 consecutive segments of 1 s length . For a specific epoch e , the power of an SPoC component ( s^=WTX , where WT corresponds the transpose of the unmixing matrix W and X to the data matrix of e in SSD space ) can be approximated by the variance of its signal within that time interval ( Var[s^] ( e ) ; Dähne et al . , 2014 ) . SPoC was separately applied to each participant , producing a number of components equal to the number of previously selected SSD components . The stability and significance of the extracted components was tested with a permutation approach ( 1000 iterations ) : z values were shuffled to create a surrogate target variable with randomized phase but same auto-correlation ( Theiler et al . , 1992; adapted from the original SPoC function: https://github . com/svendaehne/matlab_SPoC/blob/master/SPoC/spoc . m; Dähne , 2015 , Dähne et al . , 2014 ) . In accordance with the primary objective of SPoC to reconstruct the target variable z , lambda values ( λ , i . e . , optimization criterion of SPoC: component-wise covariance between z and alpha power; sorted from most positive to most negative ) and corresponding Pearson correlation values ( r ) between z and the estimated zest ( obtained via zest=VarWiTXe ) were then calculated for each iteration to generate a naive probability density function ( i . e . , null hypothesis distribution ) and to estimate the probability that the correlation value that was calculated with the original target variable z was obtained by chance . Of note , zest denotes the power time course of the spatially filtered signal that maximally covaries with the behavioural variable z . Depending on i ( i . e . , from which side of the SPoC unmixing matrix the component is chosen ) , zest will be ( maximally ) positively ( left side of the matrix ) or ( maximally ) negatively ( right side of the matrix ) correlated with z . Given our main hypothesis of an inverse relationship between alpha power and self-reported emotional arousal , we therefore only retained , for each participant , the component with the most negative ( precisely: ‘smallest’ ) lambda value λ ( disregarding the p-value to avoid circularity; Kriegeskorte et al . , 2009 ) , corresponding to the last column of the unmixing matrix W . In line with our hypothesis , single participants’ p-values were then obtained by computing the number of permuted r values that were smaller than the one estimated with SPoC . Crucially , since the extracted linear spatial filters W cannot be directly interpreted ( Haufe et al . , 2014b ) , topographical scalp projection of the components are represented by the columns of the spatial patterns matrix A obtained by inverting the full matrix W ( Figure 6 ) . To further test the hypothesis of a link between alpha power and subjective emotional arousal , we aimed to distinguish between the most and the least arousing phases of the experience by using features of the alpha bandpower of the concurrently acquired EEG signal . We followed an approach which has successfully been used in BCI research to discriminate between event- or state-related changes in the bandpower of specific frequency ranges in the EEG signal: the common spatial patterns algorithm specifies , by means of a generalized eigenvalue decomposition , a set of spatial filters to project the EEG data onto components whose bandpower maximally relates to the prevalence of one of two dichotomous states ( Blankertz et al . , 2008; Ramoser et al . , 2000 ) . In our case , we were interested in distinguishing moments that had been rated to be most ( top tertile ) and least arousing ( bottom tertile ) . Using the EEGLAB extension BCILAB ( RRID:SCR_007013 , v1 . 4-devel; Kothe and Makeig , 2013 ) , data of the selected SSD components , bandpass filtered around the individual alpha peak ±2 Hz , were epoched in 1 s segments . This sample length was chosen to enable the extraction of neural features and corresponding changes in the subjective experience , while maximizing the number of samples from the sparse datasets . Epochs with mid-level arousal ratings ( middle tertile ) were discarded , yielding 180 epochs ( 90 per class ) for each subject ( per movement condition ) . To assess the classification performance , a randomized 10-fold cross-validation procedure , a common solution for sparse training data ( Bishop , 2006 ) , was used . Per fold , a CSP-based feature model was calculated on the training data by decomposing the signal of the selected SSD components according to the CSP algorithm . A feature vector comprising the logarithmized variance of the four most discriminative CSP components ( using two columns from each side of the eigenvalue decomposition matrix as spatial filters ) was extracted per epoch . Data from the training splits were used to train a linear discriminant analysis ( LDA ) on these feature vectors ( Fisher , 1936 ) . Covariance matrices used for calculating the LDA were regularized by applying the analytic solution implemented in BCILAB ( Ledoit and Wolf , 2004 ) . The LDA model was then used to classify the feature vectors extracted from the epochs in the test split to predict the according arousal label . Average classification accuracy ( defined as 1−misclassificationrate ) over the 10 folds was taken as the outcome measure to assess the predictive quality of the approach . To allow a spatial interpretation of the projections , like with the SPoC components , the spatial patterns of the two most extreme CSP components ( associated with the largest and smallest eigenvalue ) that were used to calculate the feature vectors for the linear classification were plotted in Figure 6 ( normalized and averaged across subjects per condition ) and Figure 6—figure supplement 1 ( per single subject and condition ) . Source localized patterns are shown in Figure 9 . For non-stationary , auto-correlated time-series data , randomized cross-validation can inflate the decoding performance ( Roberts et al . , 2017 ) . To assess and minimize this possibility , we tested whether a blocked cross-validation , which preserves temporal neighbourhood structures among samples , changes the classification results of the CSP analysis . To ensure balanced classes in the training set , the ‘synthetic minority oversampling technique’ , which oversamples the less frequently represented class , was applied ( Chawla et al . , 2002; as implemented in Larsen , 2021 ) . The test set was left unbalanced as oversampling of test data can invalidate the assessment of model performance ( Altini , 2015 ) , and the area under the curve of the receiver operating characteristic ( ROC-AUC ) was used as a performance measure . To avoid homogeneous test sets ( i . e . , with samples from only one target class ) , which ( 1 ) would occur in many subjects after ‘conventional’ chronological cross-validation and ( 2 ) would preclude ROC-AUC calculation , a ‘sub-blocked’ cross-validation was used: for each subject , the dataset was split into three sub-blocks of equal length , which were then used to stratify the data assignment for a ( sub-blocked ) chronological 10-fold cross-validation . In this design , each fold consists of a concatenation of equally sized stretches of consecutive data samples taken from each of the sub-blocks: for example , to build the validation set in the first fold [x1 , x2 , x3] , with xi being the n first samples from the ith sub-block where n is the total number of samples in the dataset divided by 10 * 3 ( number of folds * number of sub-blocks ) . Thereby the temporal neighbourhood structure among data samples is largely preserved when splitting them into training and testing sets . The ( smaller ) test set is still sampled from different parts of the experience , which decreases the risk of obtaining homogeneous test sets ( e . g . , only ‘low arousing’ sections ) . Exact low-resolution tomography analysis ( eLORETA , RRID:SCR_007077; Pascual-Marqui , 2007 ) was used to localize the sources corresponding to the component extracted via SPoC and CSP . Our pipeline was based on the work of Idaji et al . , 2020 , who customized the eLORETA implementation of the M/EEG Toolbox of Hamburg ( https://www . nitrc . org/projects/meth/ ) . Our forward model was constructed via the New York Head model ( Haufe et al . , 2014b; Haufe and Ewald , 2019; Huang et al . , 2016 ) with approximately 2000 voxels and by using 28 out of 30 scalp electrodes ( TP9 and TP10 were removed because they are not contained in the model ) . Crucially , we focused on dipoles perpendicular to the cortex . eLORETA was then used to construct a spatial filter for each voxel from the leadfield matrix , and respective sources were computed by multiplying the resultant demixing matrix with the spatial patterns A of the selected SPoC and CSP components . Inverse modelling was computed separately per participant and condition before it was averaged for each condition across all subjects ( Figure 9 ) . Deep learning models have become a useful tool to decode neural information ( e . g . , Agrawal et al . , 2014; Khaligh-Razavi and Kriegeskorte , 2014 ) . Applying a deep learning approach to the time series of EEG recordings ( e . g . , Bashivan et al . , 2016 ) can be achieved using LSTM RNNs ( Hochreiter and Schmidhuber , 1995; Hochreiter and Schmidhuber , 1997 ) . With their property to store and control relevant information over time , they can find adjacent as well as distant patterns in ( time ) sequential data . The LSTM analysis was implemented in the Python-based ( RRID:SCR_008394 ) deep learning library TensorFlow ( RRID:SCR_016345 , v1 . 14 . 0; Google Inc , USA; Abadi et al . , 2015; Zaremba et al . , 2015 ) . Deep learning models usually have a high variance due to random weight initialization , architectural choices , and hyperparameters ( HPs; Geman et al . , 1992; but see Neal et al . , 2019 ) . We here used a two-step random search ( Bergstra and Bengio , 2012 ) strategy in order to find optimal HPs , to reduce the model variance and make the search computationally feasible . First , a broad random search was applied on a random subset of 10 subjects ( 20 random combinations ) in each condition . Then , the two best HPs per subject were taken and applied to the datasets of all subjects . Due to time constraints and computational complexity , the HP search was limited to a predefined range of settings and the model architecture was constrained to maximal two LSTM layers followed by maximal two fully connected layers ( FC; Hefron et al . , 2017; see Figure 4 ) . Each layer size lsl varied between 10 and 100 nodes ( lsl ∈ 10 , 15 , 20 , 25 , 30 , 40 , 50 , 65 , 80 , 100 ) , and a successive layer needed to be equal or smaller in size ( bottleneck architecture ) . The output of each layer was squashed through either rectified linear units or exponential linear units , which both allow for steeper learning curves in contrast to conventional activation functions such as sigmoid nonlinearity ( Clevert et al . , 2016 ) . The output of the last network layer ( FCL ) was fed into a tangens hyperbolicus ( tanh ) to match the binned ratings , which were labelled with –1 or 1 , respectively . We applied a mean-squared error loss to calculate the difference between the model output ( i . e . , the prediction ) and the labels , leading to a stronger weighting of losses at the upper- or lower-class border , respectively . To control and tax too large model weights , different regularization methods ( L1 , L2 ) with different regularization strengths ( Λ ∈ 0 . 00 , 0 . 18 , 0 . 36 , 0 . 72 , 1 . 44 ) were tested . Weights were optimized using Adam ( learning rate: lr ∈ 1e–2 , 1e–3 , 5e–4 ) due to its fast convergence ( Marti , 2015; see also Ruder , 2017 ) . The number of input components ( SSD , Ncomp: N ∈ [1 , 10] ) was treated as HP . The specific Ncomp neural components were successively drawn according to the order of the SSD selection . The final dataset per subject was a three-dimensional tensor of size 270 × 250 × 10 ( epochs × samples × components ) . If less than 10 components were extracted for a given subject , the tensor was filled with zero vectors . After some test runs and visual observation of the convergence behaviour of the learning progress , training iterations were set to 20 ( i . e . , the model ran 20 times through the whole training dataset ) . The 1 s samples were fed into the LSTM in random mini-batches of size 9 ( bs = 9 ) , since training on batches allows for faster and more robust feature learning ( Ruder , 2017 ) , leading to the following input tensor at training step ts: xtrain , tsbsx250x10 . To test whether the results of the binary modelling approaches ( CSP , LSTM ) were statistically significantly above chance level , exact binomial tests were conducted per subject and experimental condition ( nomov , mov ) over all 180 epochs of the respective time series ( nomov , mov ) . To do so , for each of the binary modelling approaches ( CSP , LSTM ) , the predictions for the single epochs in the 10 test splits of the cross-validation were concatenated to a single vector . The proportion of correct and false predictions was then compared to a null model with prediction accuracy 0 . 5 ( chance level ) . To test the average ( across subjects ) classification accuracies of the binary models , we calculated one-sided one-sample t-tests , comparing the mean accuracy of the respective model for both experimental conditions against the theoretical prediction accuracy of a random classifier ( 0 . 5 ) . To test whether classification accuracies differed between the two models ( CSP , LSTM ) or between the experimental conditions ( nomov , mov ) , a repeated-measures two-way ANOVA was conducted on the accuracy scores of all subjects with preprocessed data from both conditions ( n = 18 ) . To account for potential biases due to auto-correlations in the time series which might affect the statistical evaluation of the classification model , in an additional control analysis , block permutation testing was applied to the CSP approach: to maintain a local auto-correlative structure similar to the original data in the permuted target vectors , the time series were split into 10 equally sized blocks , which were then shuffled while the internal temporal structure of each block remained intact ( Winkler et al . , 2014 ) . To test whether the actual decoding scores ( from non-permuted data ) were significantly above chance level , we assessed their percentile rank in relation to the null distributions ( 1000 permutations ) on the single-subject level . On the group level , one-sided paired t-tests were used to compare the distribution of the actual decoding results against the distribution of the means of the null distributions per subject . Due to its high computational processing cost and duration , we did not perform permutation testing for the LSTM model . For SPoC , in addition to the aforementioned within-participants permutation approach yielding a single p-value for each component , group-level statistics were assessed: the hypothesis of a negative correlation between alpha power and emotional arousal was tested with a one-sample , one-tailed t-test on the correlation values between z and zest , which assessed whether the mean correlation value per condition was significantly lower than the average of the permuted ones . The code for preprocessing of the data , the three prediction models , and the statistical evaluation is available on GitHub ( https://github . com/NeVRo-study/NeVRo; Hofmann et al . , 2021; copy archived at swh:1:rev:669d5c2d6c73cbb70422efb933916fe8304195b5 ) . The 30 s break differed from the rollercoaster rides in visual features ( e . g . , static vs . dynamic input ) and in arousal ratings , which were constantly relatively low during the break ( see Figure 5 ) . Thus , the break contributed mainly to the ‘low arousing’ class . To test whether the decoding approaches also succeed if the break section is excluded from the analysis , SPoC and CSP decoding were repeated for the data without the break , that is , the rollercoasters only ( 240 s in total ) . The LSTM approach was skipped in this control analysis due to its computational processing cost and duration , and the comparable performance with CSP in the main analysis . For the classification ( CSP ) , the tertile split on the subjective arousal ratings was recalculated such that the class of ‘low arousal’ segments now comprises the least arousing sections of the rollercoasters . We then trained and tested the SPoC and CSP models with the procedures that were used for the original dataset ( incl . the break ) . For maximal stringency , we used the sub-blocked cross-validation and block permutation approach to assess the performance of the CSP model . To test whether excluding the break changed the model performance , we compared the distributions of the decoding performance parameters ( SPoC: Pearson correlation with target; CSP: ROC-AUC ) from the data with and without the break using two-sided paired t-tests . We did this per model and movement condition .
Forty-five healthy young participants ( 22 men , M ± SD: 24 . 6 ± 3 . 1 , range: 20–32 years ) completed the experiment . Data from eight participants needed to be discarded due to technical problems ( n = 5 ) or electrode malfunctioning ( n = 1 ) ; one participant discontinued the experiment and another participant reported having taken psychoactive medication . The data from 37 participants entered the analysis ( 17 men , age: M ± SD: 25 . 1 ± 3 . 1 , range: 20–31 years ) . After quality assurance during the EEG preprocessing , data from 26 participants in the condition with no head movement ( nomov ) and 19 in condition with free head movement ( mov ) entered the statistical analyses that included EEG data . The classifier based on CSP was able to decode significantly above chance level whether a subject experienced high or low emotional arousal during a given second of the experience . On average , the classification accuracy was 60 . 83% ± 7 . 40% ( M ± SD; range: 47 . 22–77 . 78% ) for the nomov , and 60 . 76% ± 6 . 58% ( M ± SD; range: 48 . 33–71 . 67% ) for the mov condition . Both were significantly above chance level ( tnomov ( 25 ) = 7 . 47 , pnomov < 0 . 001; tmov ( 18 ) = 7 . 12 , pmov < 0 . 001 ) . At the single-subject level , the classification accuracy was significantly above chance level ( p < 0 . 05 ) for 17/26 ( 65 . 38% ) participants in the nomov , and for 12/19 ( 63 . 16% ) participants in the mov condition ( see Figure 10—source data 1 for single-participant results ) . The spatial patterns yielded by the CSP decomposition are shown in Figure 6 ( across participants ) and in Figure 6—figure supplement 1 ( individual participants ) . Corresponding alpha power sources ( located via eLORETA ) are shown in Figure 9 . To test for potential biases from the model or the data , specifically its auto-correlative properties , we ran a control analysis for CSP using sub-blocked chronological cross-validation and block permutation for statistical evaluation on the single-subject level . Also under these – more strict – evaluation criteria , the average decoding performance ( ROC-AUC ) for CSP was significantly above chance level , both in the nomov ( ROC-AUC: 0 . 61 ± 0 . 09 M ± SD , range: 0 . 42–0 . 79; t ( 25 ) = 4 . 59 , p < 0 . 001 ) and in the mov condition ( ROC-AUC: 0 . 60 ± 0 . 09 M ± SD , range: 0 . 44–0 . 74; t ( 18 ) = 3 . 27 , p < 0 . 01 ) . On the single-subject level ( as assessed by permutation tests ) , decoding performance was significantly ( p < 0 . 05 ) higher when decoding the actual , unpermuted labels compared to the block-permuted labels for 9/26 ( 34 . 62% ) participants in the nomov and 5/19 ( 26 . 32% ) participants in the mov condition . After a random search over a constrained range of HPs , we extracted the best individual HP set per subject ( see Figure 4—source data 1 for the list of best HPs per condition ) . The mean classification accuracy was 59 . 42% ± 4 . 57% ( M ± SD; range: 52 . 22–68 . 33% ) for the nomov , and 61 . 29% ± 4 . 5 % ( M ± SD; range: 53 . 89–71 . 11% ) for the mov condition . Both were significantly above chance level ( tnomov ( 25 ) = 10 . 82 , pnomov < 0 . 001; tmov ( 16 ) = 10 . 51 , pmov < 0 . 001 ) . At the single-subject level , the classification accuracy was significantly above chance level for 16/26 ( 61 . 54% ) participants in the nomov condition , and for 16/19 ( 84 . 21% ) participants in the mov condition ( same test as for CSP results; see Figure 10—source data 1 ) . As an illustration of the prediction behaviour across all three models in one participant ( with high performance for all three decoding approaches ) , see Figure 7 . Correlations of performances across models and experimental conditions are shown in Figure 10 . The ( positive ) correlation between the two binary classification approaches ( CSP , LSTM ) was significant ( after Bonferroni multiple-comparison correction ) , irrespective of the experimental condition ( nomov , mov ) , meaning that subjects who could be better classified with CSP also yielded better results in the LSTM-based classification . In a repeated-measures ANOVA testing for differences in the accuracies of the two binary classification models ( CSP , LSTM ) and the two conditions ( nomov , mov ) , none of the effects was significant: neither the main effect model ( F ( 1 , 17 ) = 0 . 02 , p = 0 . 904 ) nor the main effect condition ( F ( 1 , 17 ) = 0 . 72 , p = 0 . 408 ) or their interaction ( F ( 1 , 17 ) = 1 . 59 , p = 0 . 225 ) . For a further comparison of the performances of the classification approaches , the respective confusion matrices are depicted in Figure 8 ( average across the subjects per condition and model ) . SPoC and CSP performed significantly above chance level also when trained and tested on data without the break section . For CSP on data without the break , the average classification performance ( ROC-AUC ) was 0 . 57 ± 0 . 10 ( M ± SD; range: 0 . 28–0 . 78 ) in the nomov and 0 . 59 ± 0 . 09 ( M ± SD; range: 0 . 45–0 . 77 ) in the mov condition ( see previous paragraph for the decoding performance with the break included ) . Average model performances were still significantly above chance level ( means of the block permutation distributions on the single-subject level ) in both movement conditions ( nomov: t ( 25 ) = 2 . 89 , p < 0 . 01; mov: t ( 18 ) = 3 . 50 , p < 0 . 01 ) . On the single-subject level , the classification performance was significantly above chance level for 3/26 ( 11 . 54% ) participants in the nomov and 5/19 ( 26 . 32% ) participants in the mov condition . For SPoC on data without the break , the average Pearson correlation between z and zest ( estimated target variable ) was significantly smaller ( more negative ) than the average of single participants’ permuted correlation values for both the nomov ( M ± SD: –0 . 22 ± 0 . 08; range: –0 . 36 to –0 . 07; tnomov ( 25 ) = –3 . 17; p < 0 . 01 ) and the mov condition ( M ± SD: –0 . 21 ± 0 . 07; range: –0 . 37 to –0 . 061; tmov ( 18 ) = –2 . 53; p < 0 . 05 ) . On the single-subject level , 2/26 ( 7 . 69% ) participants for the nomov and 7/19 ( 36 . 84% ) participants for the mov condition remained statistically significant ( p < 0 . 05 ) after permutation-based tests . Removing the break from the training data overall numerically decreased the decoding performances of both models . For CSP , the decrease was significant in the nomov ( t ( 25 ) = 2 . 23 , p = 0 . 034 ) and not significant in the mov condition ( t ( 18 ) = 0 . 57 , p = 0 . 58 ) . For SPoC , the decrease ( Pearson correlation ) was not significant in both conditions ( nomov: t ( 25 ) = –1 . 66 , p = 0 . 108; mov: t ( 18 ) = –1 . 13 , p = 0 . 269 ) .
In studies with event-related stimulation or block designs , more emotionally arousing compared to less emotionally arousing images , videos , and sounds were associated with event-related decreases in alpha power , predominantly over parieto-occipital electrodes ( De Cesarei and Codispoti , 2011; Luft and Bhattacharya , 2015; Schubring and Schupp , 2019; Uusberg et al . , 2013; Koelstra et al . , 2012 ) . While such stimuli provide a high degree of experimental control in terms of low-level properties and presentation timings , the emotional experience and its neurophysiology under event-related stimulation may differ from the emotional experience in real-life settings , which is perceptually complex , multisensory , and continuously developing over time . Our results provide evidence that the neural mechanisms reflected in modulations of alpha power – particularly in parieto-occipital regions – also bear information about the subjective emotional state of a person undergoing an immersive and emotionally arousing experience . Also fMRI studies have related brain activity in parietal cortices and emotional processing ( e . g . , Lettieri et al . , 2019 ) . Our study thus suggests that findings from event-related , simplified stimulation generalize to more naturalistic ( i . e . , dynamic and interactive ) settings . Paralleling the idea of emotional arousal being a dimension of ‘core affect’ ( Russell and Barrett , 1999 ) and a psychological primitive that underlies many mental phenomena , also alpha oscillations have been connected to various psychological ‘core processes’: for instance , modulations of alpha power were linked to attention ( Van Diepen et al . , 2019 ) and memory ( Klimesch , 2012 ) . More generally , neural oscillations in the alpha frequency range were suggested to serve functional inhibition of irrelevant sensory input ( Jensen and Mazaheri , 2010; Foster and Awh , 2019 ) and to code for the location and the timing of task-relevant stimuli ( Foster et al . , 2017 ) . Such processes can be functionally linked to emotional arousal: during emotionally arousing experiences , preferred and enhanced processing of relevant sensory stimuli ( e . g . , indicating potential threats ) is an adaptive behaviour . In line with this , modulations of alpha oscillations over parietal sensors have been linked to threat processing ( Grimshaw et al . , 2014 ) . Variations in emotional arousal and alpha power may , thus , have guided attention and memory formation also in our experiment: during particularly arousing parts of the rollercoaster , participants may have directed their attention to specific parts of the visual scene , for example , to foresee the end of the looping . Moreover , our inverse modelling ( Figure 9 ) has also localized arousal-related alpha sources in sensorimotor cortices , which could correspond to somatic experiences typically associated with rollercoasters . Some of the averaged spatial patterns ( see Figures 6 and 9 ) we observed for the SPoC- and CSP-based decoding stronger absolute weights for electrodes above right – as compared to left – cortices . Since we did not hypothesize a lateralization of the alpha effects , we refrained from statistically testing differences between the hemispheres . Similar patterns of right-lateralized alpha oscillations have also been related to arousal in major depression ( Metzger et al . , 2004; Stewart et al . , 2011 ) . However , it is unclear to which extent these effects are specific to arousal , as lateralization of alpha power has also been observed in working-memory ( Pavlov and Kotchoubey , 2020 ) and resting-state studies ( Ocklenburg et al . , 2019 ) . Our results motivate experimental work that will model the link between emotional arousal and alpha oscillations by systematically varying additional variables ( e . g . , attention , sensorimotor processing ) . We argue that studying such relationships in naturalistic settings allows embracing and learning statistical interdependencies that are characteristics of the real world . More naturalistic experimental stimulation , for example , using immersive VR , allows to test the brain under conditions it was optimized for and thereby improve the discovery of neural features and dynamics ( Gibson , 1978; Hasson et al . , 2020 ) . Findings from naturalistic studies can test the real-world relevance of results obtained in highly controlled , abstract laboratory settings ( Matusz et al . , 2019; Shamay-Tsoory and Mendelsohn , 2019 ) . Challenges of using VR for more naturalistic research designs are the creation of high-quality VR content , more complex technical setups , and discomfort caused by the immersion into the virtual environment ( Pan and Hamilton , 2018; Vasser and Aru , 2020 ) . Despite the incongruence between VR rollercoaster-induced visual stimulation and vestibular signals , which may lead to motion sickness ( Reason and Brand , 1975 ) , only one of our participants stopped the experiment because of cybersickness . This low number may result from the relatively short length of the VR experience ( net length: <20 min ) and the professionally produced VR stimulation . Shorter exposure times ( Rebenitsch and Owen , 2016 ) and experiences that elicit stronger feelings of presence have been associated with lower levels of cybersickness ( Weech et al . , 2019 ) . Combining EEG with VR provides additional challenges: the signal-to-noise ratio ( SNR ) can be decreased due to mechanical interference of the VR headset with the EEG cap and due to movement artefacts when the participant interacts with the virtual environment ( e . g . , head rotations ) . To ensure high data quality , we applied multiple measures to prevent , identify , reject , or correct artefacts in the EEG signal ( see Materials and methods section for details ) . Ultimately , the performance of all three decoding models did not differ significantly for both conditions ( nomov , mov ) . We suggest that , with appropriate quality assurance during data acquisition and analysis ( leading to more data rejection/correction for mov than for nomov ) , EEG can be combined with immersive VR and free head movements . Other studies of mobile brain imaging , even recording outdoors and with full-body movements , came to similar conclusions ( Debener et al . , 2012; Ehinger et al . , 2014; Gramann et al . , 2011; Symeonidou et al . , 2018 ) . Each of the applied decoding approaches allows for different insights and interpretations , but overall , they yield converging results . SPoC and CSP share advantages that are common to most spatial filtering methods based on generalized eigenvalue decomposition , namely precise optimization policies , high speed , and interpretability . As dimensionality reduction techniques , they combine data from multiple M/EEG channels to obtain a new signal ( component ) with a higher SNR ( Lotte et al . , 2018; Parra et al . , 2005 ) . This aids maximizing the difference in the signal of interest between experimental conditions ( de Cheveigné and Parra , 2014; Rivet et al . , 2009 ) or against signals in the neighbouring frequency ranges ( Nikulin et al . , 2011 ) . The similarity between the two approaches ( SPoC , CSP ) and their interpretability becomes apparent in the resulting spatial patterns: the normalized and averaged SPoC topoplots and source localizations in both conditions ( nomov , mov ) resemble the ones extracted via CSP to maximize power for the low-arousal epochs of the experience ( Figures 6 and 9 ) . SPoC and CSP solve a similar problem here: extracting components whose power is minimal during states of high emotional arousal and maximal during states of low arousal . This indicates that SPoC and CSP exploited similar spatial informational patterns in the input data . However , the datasets handed to the SPoC and CSP models were not identical . For the CSP analysis , only the upper and lower extreme of the arousal ratings were included ( i . e . two-thirds of the data ) , while epochs with medium arousal ratings ( i . e . , one-third of the data ) were excluded , whereas SPoC was trained on the full continuous data stream . There are two potential explanations for the observation that SPoC and CSP nevertheless yield similar spatial patterns: either the most relevant information was encoded in the most extreme parts of the experience , or there is a truly linear relationship between alpha power and emotional arousal that can be queried on all parts of this spectrum ranging from low to high emotional arousal . The spatial patterns for the components gained from SSD , SPoC , and CSP exhibit discernible variance between the single subjects ( see Figure 6—figure supplement 1 ) . This can be , for example , caused by physiological differences ( e . g . , different shapes of the skull , different cortical folding ) or slightly different positioning of the EEG electrodes . The same cortical source might thereby lead to different patterns of scalp EEG in different participants . Spatial filtering procedures inverse this projection and the extracted patterns therefore also vary across subjects . Such inter-individual differences are well known for BCIs , and extensions for CSP have been suggested , which allow for a transfer of features across subjects ( e . g . , Cheng et al . , 2017 ) . To emphasize the communalities across individual patterns and indicate the cortical areas that contributed most to decoding results , we report the averaged patterns ( Figure 6 ) and the averaged results of the reconstructed cortical sources ( Figure 9 ) . To test for confounds or analytic artefacts , for example , due to auto-correlations in the data , we additionally applied ‘sub-blocked’ cross-validation for model training and block permutation for statistical evaluation . Also under these more strict evaluation conditions , the average decoding performance was significantly above chance level . It is therefore unlikely that the results can be explained solely by dependencies in the data ( e . g . , auto-correlation ) which are not explicitly modelled in the main analysis . Moreover , to test the impact of the differences between the rollercoasters and the break , for example , regarding visual dynamics and elicited emotional arousal , on the decoding performance , SPoC and CSP analyses were repeated on the data without the break . Again , the average decoding performances decreased compared to the data with the break , but remained significantly above chance level for both head movement conditions . The decrease in decoding performance with the break removed may result from ( 1 ) less training data being available and ( 2 ) a narrower range of emotional arousal values , more similar classes ( ‘high arousal’ and ‘low arousal’ ) , and therefore a more difficult distinction . We observed a high degree of variability in decoding performance across participants ( see Figure 10 ) . For example , for less than 70% ( and less than 35% with sub-blocked cross-validation and permutation testing ) of participants , CSP yielded significant results on the single-subject level . This variability reflects the difficulty of some features and classifiers to perform equally well across subjects , which has been reported in the BCI literature ( Krusienski et al . , 2011; Nurse et al . , 2015 ) . In a supplementary analysis , we compared the classification results to a less complex logistic regression model , which was directly trained on time-frequency data from electrodes in the occipital-parietal region of interest . The model performed almost on par with CSP in the mov condition but was less sensitive in the nomov condition . Linear regression on time-frequency data in sensor space also has methodological and conceptual limitations compared to SPoC and CSP , such as underestimating sources of noise , disregarding the generative model that underlies EEG data , and consequently a limited interpretability ( for details , see Dähne et al . , 2014 ) . We therefore did not include this analysis in the final report . Despite having recently gained more attention with the fast progress of deep learning ( e . g . , more efficient hardware and software implementations ) , LSTMs still need to stand up to well-established models such as CSP for EEG analysis . We found that the LSTM can extract features from neural input components that reflect changes in subjective emotional arousal and that the accuracy of its predictions in both conditions ( nomov , mov ) matched closely the ones of CSP ( see Figures 8 and 10 ) . It is noteworthy that for the CSP model , the ( LDA-based ) classification rested on narrowly defined spectral features of the signal while for the LSTM model , the input was the signal in the time domain and the feature selection process was part of the model fitting . The strong correlation between the predictions of the two models suggests that the LSTM extracts similar information as the CSP to make its prediction , namely power . Higher accuracies may be achievable with LSTM models due to their non-convex optimization landscape . However , in our two-step HP search , we found that for each subject a range of different HP settings led to similar prediction accuracies ( see Figure 4—source data 1 ) . Model ensembles , although computationally demanding , could further increase the robustness of the estimates ( Opitz and Maclin , 1999; Rokach , 2010; Dietterich , 2000 ) . Although it is often stated that deep learning models require large datasets ( for an empirical perspective , see Hestness et al . , 2017 ) , our model , with its architecture of one to two LSTM layers followed by one to two fully connected layers , converged in less than 200 training iterations on a relatively small dataset . This quick convergence is partly due to the fast gradient flow through the memory cell of the LSTM during the weight update , which is an additional advantage of the LSTM over other RNNs ( Doetsch et al . , 2014; Hochreiter and Schmidhuber , 1997 ) . Additionally , the spatial-spectral filtering in our study ( i . e . , SSD-based extraction of narrow-band alpha components ) may have eased the training of the LSTM . With more data , an LSTM could be trained on raw data or longer segments of the EEG to preserve more of the continuous structure and ultimately exploit its central property , as a dynamic model , of detecting long-term dependencies in the input . In contrast to SPoC and CSP , we did not compute explanatory topoplots or sources from the LSTM , since the analysis of predictions on input level in non-linear deep learning models constitutes a challenge in itself ( i . e . , ‘black box’ problem of deep learning ) . However , ‘explainable artificial intelligence’ ( XAI ) is an active area of research in machine learning , aiming to open this ‘black box’ . For EEG , there are attempts to create topologically informative maps in the signal space that explain the decision of simple shallow neural networks ( Sturm et al . , 2016 ) . Also for the more complex LSTM model , XAI methods were applied , for example , on text data ( Arras et al . , 2017; see also Lapuschkin et al . , 2019 ) . However , exploring and validating these approaches on our data was beyond the scope of this study . In summary , we find that SPoC , CSP , and LSTM can be used to decode subjective emotional arousal from EEG acquired during a naturalistic immersive VR experience . The source of the alpha oscillations could be localized in parieto-occipital regions . Compared to other EEG decoding paradigms ( e . g . , lateralized motor imagery; Herman et al . , 2008 ) , the accuracy of our models was relatively low . This may be a consequence of ( 1 ) the fast-changing events in the VR experience ( particularly the rollercoasters ) , ( 2 ) the asynchronicity of the two data streams as participants retrieved their emotional states from memory in retrospective ratings , ( 3 ) the generally high inter-individual variability in the interpretability of subjective self-reports ( Blascovich , 1990 ) , and ( 4 ) the ‘single-trial’ study design and its relatively short time series . With respect to ( 1 ) – ( 3 ) , people’s memory for feelings and events is susceptible to distortions and biases ( Kaplan et al . , 2016; Levine and Safer , 2002 ) . Following McCall et al . , 2015 , we elicited the memory recall by showing participants an audiovisual replay of their experience from their own perspective in the VR headset while recording continuous ratings . This aimed to minimize biases related to the point of view ( Berntsen and Rubin , 2006; Marcotti and St Jacques , 2018 ) or timescale ( e . g . , Fredrickson and Kahneman , 1993 ) during recall ( as discussed in McCall et al . , 2015 ) . Lastly , while our research aimed to explore the role of the alpha frequency band in the appraisal of emotional arousal ( see Introduction ) , higher frequencies could carry additional information about the phenomenon leading to better model predictions . However , higher frequency bands also include non-neural ( e . g . , muscle activity-related ) signals , limiting the interpretability of those results . Our study has limitations that need to be considered when interpreting the results: while being engaging , emotionally arousing and tolerable for the subjects , the commercial content used for stimulation did not provide access to the source code in order to control and extract stimulus features ( e . g . , height or speed of the rollercoasters ) . In general , creating high-quality VR content is a challenge for research labs , but there are recent efforts to provide toolboxes that facilitate customized VR development and scientific experimentation in VR ( e . g . , Grübel et al . , 2017; Brookes et al . , 2020 ) . The length of the experience was chosen to minimize habituation to the stimulus and inconvenience caused by the recording setup ( EEG electrodes and VR headset ) . This led to relatively short recording times per subject and condition . Data sparsity , however , is challenging for decoding models , which need a sufficient amount of data points for model training and evaluation , where especially larger training sets lead to more robust predictions ( Hestness et al . , 2017 ) . We used cross-validation , which is commonly applied in scenarios of limited data , to achieve a trade-off between training and validation data ( Bishop , 2006 ) . Nevertheless , the models and results can be expected to perform more robustly with more training data . We here confirm findings from static stimulation under more naturalistic conditions . To systematically investigate differences between approaches , a study with a within-subject design would be required . We hope that our study provides a stepping stone and motivation in this direction . Finally , emotional arousal is a multi-faceted mind-brain-body phenomenon that involves the situated organism and its interaction with the environment . The training data for multivariate models such as the LSTM can include other modalities , such as peripheral physiological ( e . g . , HR , GSR ) or environmental ( e . g . , optical flow ) features . Naturalism can be further increased by sensorimotor interaction ( beyond head movements ) in immersive VR ( McCall et al . , 2015 ) or by mobile EEG studies in real-world environments ( Debener et al . , 2012 ) , which , however , poses further challenges to EEG signal quality ( Gwin et al . , 2010 ) . We conclude that different levels of subjectively experienced emotional arousal can be decoded from neural information in naturalistic research designs . We hope that combining immersive VR and neuroimaging not only augments neuroscientific experiments but also increases the generalizability and real-world relevance of neuroscientific findings . | Human emotions are complex and difficult to study . It is particularly difficult to study emotional arousal , this is , how engaging , motivating , or intense an emotional experience is . To learn how the human brain processes emotions , researchers usually show emotional images to participants in the laboratory while recording their brain activity . But viewing sequences of photos is not quite like experiencing the dynamic and interactive emotions people face in everyday life . New technologies , such as immersive virtual reality , allow individuals to experience dynamic and interactive situations , giving scientists the opportunity to study human emotions in more realistic settings . These tools could lead to new insights regarding emotions and emotional arousal . Hofmann , Klotzsche , Mariola et al . show that virtual reality can be a useful tool for studying emotions and emotional arousal . In the experiment , 37 healthy young adults put on virtual reality glasses and ‘experienced’ two virtual rollercoaster rides . During the virtual rides , Hofmann , Klotzsche , Mariola et al . measured the participants' brain activity using a technique called electroencephalography ( EEG ) . Then , the participants rewatched their rides and rated how emotionally arousing each moment was . Three different computer modeling techniques were then used to predict the participant’s emotional arousal based on their brain activity . The experiments confirmed the results of traditional laboratory experiments and showed that the brain’s alpha waves can be used to predict emotional arousal . This suggests that immersive virtual reality is a useful tool for studying human emotions in circumstances that are more like everyday life . This may make future discoveries about human emotions more useful for real-life applications such as mental health care . | [
"Abstract",
"Introduction",
"Materials",
"and",
"methods",
"Results",
"Discussion"
] | [
"neuroscience"
] | 2021 | Decoding subjective emotional arousal from EEG during an immersive virtual reality experience |
The timing and accuracy of perceptual decision-making is exquisitely sensitive to fluctuations in arousal . Although extensive research has highlighted the role of various neural processing stages in forming decisions , our understanding of how arousal impacts these processes remains limited . Here we isolated electrophysiological signatures of decision-making alongside signals reflecting target selection , attentional engagement and motor output and examined their modulation as a function of tonic and phasic arousal , indexed by baseline and task-evoked pupil diameter , respectively . Reaction times were shorter on trials with lower tonic , and higher phasic arousal . Additionally , these two pupil measures were predictive of a unique set of EEG signatures that together represent multiple information processing steps of decision-making . Finally , behavioural variability associated with fluctuations in tonic and phasic arousal , indicative of neuromodulators acting on multiple timescales , was mediated by its effects on the EEG markers of attentional engagement , sensory processing and the variability in decision processing .
The speed and accuracy with which humans , as well as non-human animals , respond to a stimulus depends not only on the characteristics of the stimulus , but also on the cognitive state of the subject . When drowsy , a subject will respond more slowly to the same stimulus compared to when she is attentive and alert . Central arousal also fluctuates across a smaller range during quiet wakefulness , when the subject is neither drowsy or inattentive , nor overly excited or distractible . Although these trial-to-trial fluctuations can impact on behavioural performance during decision-making tasks ( Aston-Jones and Cohen , 2005 ) , it is largely unknown how arousal modulates the underlying processes that support decision formation . Perceptual decision-making depends on multiple neural processing stages that represent and select sensory information , those that process and accumulate sensory evidence , and those that prepare and execute motor commands . Variability in central arousal could affect any one or potentially all of these processing stages , which in turn could influence behavioural performance . The neuromodulatory systems that control central arousal state , such as the noradrenergic ( NA ) locus coeruleus ( LC ) and the cholinergic basal forebrain ( BF ) , have also been suggested to drive fluctuations in endogenous activity linked to changes in cortical ( de ) synchronization , that is cortical state ( Harris and Thiele , 2011; Lee and Dan , 2012 ) , and are linked to cognitive functions such as attention ( Thiele and Bellgrove , 2018 ) , both known to affect information processing and behavioural performance . These modulatory systems have both tonic and phasic firing patterns that are recruited on different timescales and support different functional roles ( Aston-Jones and Cohen , 2005; Dayan and Yu , 2006; Parikh et al . , 2007; Parikh and Sarter , 2008; Sarter et al . , 2016 ) . Tonic changes in neuromodulator activity occur over longer timescales that can span multiple trials , whereas fast ( task-evoked ) recruitment through phasic activation occurs on short enough timescales to influence neural activity and behavioural decisions within the same trial ( Aston-Jones and Cohen , 2005; Bouret and Sara , 2005; Dayan and Yu , 2006; Parikh et al . , 2007 ) . Pupil diameter correlates strongly with a variety of measurements of cortical state and behavioural arousal ( Eldar et al . , 2013; Reimer et al . , 2014; McGinley et al . , 2015b; McGinley et al . , 2015a; Vinck et al . , 2015; Engel et al . , 2016 ) , and can thus be considered a reliable proxy of central arousal state . Indeed , there is a strong correlation between pupil size and activity in various neuromodulatory centres that control arousal ( Aston-Jones and Cohen , 2005; Gilzenrat et al . , 2010; Murphy et al . , 2014a; Varazzani et al . , 2015; Joshi et al . , 2016; Reimer et al . , 2016; de Gee et al . , 2017 ) . Both baseline pupil diameter , reflecting tonic activity levels in neuromodulatory centres ( tonic arousal ) , and task-evoked pupil diameter changes ( phasic arousal ) , have been related to specific neural processing stages of perceptual decision making . Baseline pupil diameter correlates with sensory sensitivity ( McGinley et al . , 2015a; McGinley et al . , 2015b ) and is predictive of behavioural performance during elementary detection tasks ( Murphy et al . , 2011; McGinley et al . , 2015a ) . Pupil diameter also changes phasically in the course of a single decision ( Beatty , 1982a; de Gee et al . , 2014; de Gee et al . , 2017; Lempert et al . , 2015; Murphy et al . , 2016; Urai et al . , 2017 ) , and has been related to specific elements of the decision making process , such as decision bias ( de Gee et al . , 2014; de Gee et al . , 2017 ) , uncertainty ( Urai et al . , 2017 ) , and urgency ( Murphy et al . , 2016 ) . This suggests that these neuromodulatory systems do not only dictate network states ( through tonic activity changes ) , but that they are recruited throughout the decision making process ( Cheadle et al . , 2014; de Gee et al . , 2014; de Gee et al . , 2017 ) . Although both baseline pupil diameter and the phasic pupil response have been associated with specific aspects of decision-making , the relationship between pupil-linked arousal and the electrophysiological correlates of decision-making are largely unknown . Recently developed behavioural paradigms have made it possible to non-invasively study the individual electroencephalographic ( EEG ) signatures of perceptual decision-making described above ( O'Connell et al . , 2012; Kelly and O'Connell , 2013; Loughnane et al . , 2016; Loughnane et al . , 2018; Newman et al . , 2017 ) . In these paradigms , participants are required to continuously monitor ( multiple ) stimuli for subtle changes in a feature . Because stimuli are presented continuously , target onset times ( and locations ) are unpredictable , and sudden stimulus onsets are absent , eliminating sensory evoked deflections in the EEG traces . These characteristics allow for the investigation of the gradual development of build-to-threshold decision variables as well as signals that code for the selection of relevant information from multiple competing stimuli , a critical feature of visuospatial attentional orienting that impact evidence accumulation processes ( Loughnane et al . , 2016 ) . Here , we asked how arousal influences EEG signals that relate to each of the separate processing stages described above . Specifically , we tested the effects of pupil-linked arousal on pre-target preparatory parieto-occipital α-band activity , associated with fluctuations in the allocation of attentional resources ( Kelly and O'Connell , 2013 ) ; early target selection signals measured over contra- and ipsilateral occipital cortex , the N2c and N2i ( Loughnane et al . , 2016 ) ; perceptual evidence accumulation signals measured as the centroparietal positivity ( CPP ) , which is a build-to-threshold decision variable demonstrated to scale with the strength of sensory evidence and predictive of reaction time ( RT ) ( O'Connell et al . , 2012; Kelly and O'Connell , 2013 ) ; and motor-preparation signals measured via contralateral β-band activity ( Donner et al . , 2009; O'Connell et al . , 2012 ) . Of these signals , we extracted specific characteristics such as the latency , build-up rate and amplitude , and tested whether these were affected by pupil-linked arousal . Additionally , because the variance and response reliability of the membrane potential of sensory neurons varies with pupil diameter ( Reimer et al . , 2014; McGinley et al . , 2015a ) , we also investigated whether arousal affected the inter-trial phase coherence ( ITPC ) , a measure of across trial consistency in the EEG signal , of the N2 and the CPP . We found that both baseline pupil diameter as well as the pupil response were predictive of behavioural performance , and that this relationship was best described by non-monotonic , but not U-shaped , second-order polynomial model fits . Furthermore , we found that both tonic and phasic arousal bore a predictive relationship with the neural signals coding for baseline attentional engagement , early target selection , decision processing as well as the preparatory motor response . Although neural activity representing all these stages varied with changes in arousal , unique variability in task performance due to tonic arousal ( baseline pupil diameter ) could only be explained by the amplitude of target selection signals and the consistency of the CPP , reflecting decision processing . In contrast , variability due to phasic arousal ( pupil response ) was explained by pre-target α-band activity as well as the consistency of the CPP .
We first investigated the relationship between trial-by-trial pupil dynamics and behavioural performance . As stimuli were presented well above perceptual threshold , our subjects performed at ceiling ( mean , 98 . 7%; range: 92–100% , Newman et al . , 2017 ) . We therefore focused on RT and the RT coefficient of variation ( RTcv ) , a measure of performance variability calculated by dividing the standard deviation in RT by the mean ( Bellgrove et al . , 2004 ) , rather than accuracy . We found that baseline pupil diameter displayed a non-monotonic relationship with both measures of behavioural performance ( RT χ2 ( 1 ) =8 . 84 , p=0 . 003; RTcv χ2 ( 1 ) =4 . 43 , p=0 . 035 ) . Neither effects were , however , significantly U-shaped ( Figure 1B ) . Rather , RT was slower on trials with higher baseline arousal levels . The pupil diameter response , on the other hand , displayed a non-monotonic ( but not U-shaped ) relationship with RT ( χ2 ( 1 ) =51 . 89 , p<0 . 001 ) and an inverse linear relationship with RTcv ( χ2 ( 1 ) =45 . 94 , p<0 . 001 ) . For both measures , best performance was found on trials with the largest pupil responses ( Figure 1C ) . This relationship remained very similar when trial-by-trial fluctuations in the pupil response that are due to variability in the amplitude or phase of the baseline pupil diameter were not removed ( Figure 1—figure supplement 5 ) . We furthermore repeated the sequential regression analysis in single-trial , non-binned data , in which we additionally controlled for time-on-task effects , confirming that these effects were not dependent on the binning procedure ( Supplementary file 1 ) . Additionally , we noticed that when we band-pass filtered the pupil diameter , rather than low-pass filtered , the relationship between baseline pupil diameter and task performance was not significant ( Figure 1—figure supplement 6 ) . This suggests that slow fluctuations in baseline pupil diameter ( <0 . 01 Hz ) are driving the effect on task performance . Having established a relationship between task performance and both tonic and phasic modes of central arousal state , we next focused on the relationship between these pupil dynamics and the neural signatures underpinning target detection on this perceptual decision making task ( Loughnane et al . , 2016; Newman et al . , 2017 ) . During decision making , perceptual evidence has to be accumulated over time . This accumulation process has long been related to build-to-threshold activity in single neurons in parietal cortex ( Gold and Shadlen , 2007 ) ; but see Latimer et al . , 2015 , Latimer et al . , 2016; Shadlen et al . , 2016 ) . The centro-parietal positivity ( CPP ) measured from scalp EEG exhibits many of these same properties , including a representation of the accumulation of sensory evidence towards a decision bound ( O'Connell et al . , 2012; O'Connell et al . , 2018; Kelly and O'Connell , 2013 ) . Because in this study we used relatively strong sensory evidence ( 50% coherence ) , it is possible that subjects may not have relied upon any temporal integration of this motion signal to reach a decision . Rather , variability in RT could be brought about by variation in the onset transient of target selection due to the temporal and spatial uncertainty of the target stimulus . On single trials , decision formation could be a step-like signal that averaged across trials looks like an accumulate-to-bound signal ( Latimer et al . , 2015 ) . Although we cannot discount this possibility , aligning the visual early target selection signals ( N2c ) to response reveals a much lower signal amplitude compared to aligning it to target onset ( Figure 2—figure supplement 1 ) . This indicates that there is no fixed delay between target selection and the response , and that there is variability in the duration of the sustained period of this task . This variation could indicate different trial-to-trial strategies ( e . g . comparing motion in one stimulus against the stimulus on the other side of the screen ) , or in addition some variability in accumulation rate . Because of this uncertainty , we refer to the functional significance of the CPP as decision processing . Here we tested the relationship between the pupil diameter response and the onset , build-up rate , amplitude and consistency ( ITPC ) of the CPP ( Figure 2 ) . We found that the onset latency of the CPP , defined as the first time point that showed a significant difference from zero for 15 consecutive time points , displayed an inverse monotonic relationship with the size of the pupil response ( χ2 ( 1 ) =5 . 60 , p=0 . 018 ) , such that the fastest onsets were found for the largest pupil response bins ( Figure 2A ) . Likewise , the build-up rate ( χ2 ( 1 ) =4 . 45 , p=0 . 035 ) , but not the amplitude ( p=0 . 15 ) , of the CPP varied with the pupil response , displaying the steepest slope on trials with the largest pupil dilations . Because the membrane potential of sensory neurons shows the least variance and highest response reliability at intermediate baseline pupil diameter ( McGinley et al . , 2015a ) , we additionally investigated the ITPC , a measure of across trial consistency , of the CPP . We computed ITPC with a single-taper spectral analysis in a 512 ms sliding window computed at 50 ms intervals , with a frequency resolution of 1 . 95 Hz ( Materials and methods ) . Based on the stimulus-locked grand average time-frequency spectrum , we selected a time ( 300–550 ms ) and frequency window ( <4 Hz ) for further statistical analyses ( Figure 2C ) . We found an approximately linear relationship between pupil diameter response and the consistency of the CPP signal ( χ2 ( 1 ) =41 . 79 , p<0 . 001 ) , indicating that the CPP signal is less variable for larger pupil response bins ( Figure 2D ) . This relationship was also present when we aligned the CPP to the response ( Figure 2—figure supplement 2 ) , indicating that this effect is unlikely to solely reflect variability in the onset of the CPP . Thus , we found that the size of the pupillary response was predictive of both the onset latency , build-up rate as well as the ITPC of the CPP . Moreover , the relationship with the neural parameters of the CPP resembled the relationship between the pupil response and behavioural performance ( Figure 1C ) . Large pupil dilations were predictive of faster responses , earlier CPP onset latencies , as well as steeper build-up rates and more consistent CPP . Next , we asked whether other stages of information processing underpinning perceptual decision making also varied with the pupil response . We next investigated pre-target preparatory α-band power ( 8–13 Hz ) , a sensitive index of attentional deployment that has been shown to vary with behavioural performance . Specifically , previous studies have found higher pre-target α-band power preceding trials with longer RT , and suggested that fluctuations in α-power may reflect an attentional influence on variability in task performance ( Ergenoglu et al . , 2004; van Dijk et al . , 2008; O'Connell et al . , 2009; Kelly and O'Connell , 2013 ) . We first verified the relationship between α-band power and behavioural performance by binning the data into five bins according to α-band power and performing the same sequential regression analysis as described above ( Figure 3A ) . We replicated previous findings ( Kelly and O'Connell , 2013 ) and found an approximately linear relationship between α-band power and RT ( χ2 ( 1 ) =25 . 27 , p<0 . 001 ) but not RTcv ( p=0 . 48 ) . In line with previous research ( Hong et al . , 2014 ) , pupil diameter response was inversely related to α-band power ( Figure 3B ) , displaying an approximately linear relationship ( χ2 ( 1 ) =28 . 24 , p<0 . 001 ) , suggesting that pre-target attentional engagement is related to phasic arousal . We next focused on response-related motor activity in the form of left hemispheric β-power ( LHB ) . LHB decreases before a button press and has been shown to reflect the motor-output stage of perceptual decision making , but also to trace decision formation , reflecting the build-up of choice selective activity ( Donner et al . , 2009 ) . Here we investigated the LHB amplitude and build-up rate preceding response ( Figure 3C ) . We found that neither LHB amplitude ( p=0 . 63 ) nor LHB slope ( p=0 . 20 ) varied with phasic arousal , suggesting that phasic arousal does not influence the build-up rate of choice-related activity over motor cortex . Next we investigated the N2 ( Figure 3D–F ) , a stimulus-locked early target selection signal that has been shown to predict behavioural performance and modulate the onset and build-up rate of the CPP ( Loughnane et al . , 2016 ) . Because of the spatial nature of the task , we analysed the negative deflection over both the contra- ( N2c ) and ipsi-lateral ( N2i ) hemisphere , relative to the target location . The pupil response was not predictive of any aspect of the N2 . Specifically , phasic arousal was not predictive of N2c latency ( p=0 . 82 ) or amplitude ( p=0 . 64 ) , nor did we find any relationship between the pupil response and the N2c ITPC ( p=0 . 14 ) . Likewise , the pupil response was not predictive of N2i latency ( p=0 . 64 ) , amplitude ( p=0 . 11 ) or ITPC ( p=0 . 87 ) . We found that pupil-linked phasic arousal was predictive of specific neural signals at multiple information processing stages of perceptual decision making . To test which of these signals explained unique variability in behavioural performance across the five pupil response bins and subjects , the neural signals were added to a linear mixed effects model predicting either RT or RTcv with their order of entry determined hierarchically by their temporal order in the decision-making process . This allowed us to test whether each successive stage of neural processing would improve the fit of the model to the behavioural data , over and above the fit of the previous stage . Compared to the baseline model predicting RT with pupil bin , the addition of pre-target α-power significantly improved the model fit ( χ2 ( 1 ) =10 . 30 , p<0 . 001 ) . None of the measures of early target selection improved the fit of the model; neither N2c latency ( χ2 ( 1 ) =0 . 14 , p=0 . 70 ) or amplitude ( χ2 ( 1 ) =0 . 94 , p=0 . 33 ) , nor N2i latency ( χ2 ( 1 ) =2 . 39 , p=0 . 12 ) or amplitude ( χ2 ( 1 ) =2 . 39 , p=0 . 12 ) . We found that both the addition of CPP onset ( χ2 ( 1 ) =8 . 24 , p=0 . 004 ) as well as the build-up rate ( χ2 ( 1 ) =4 . 90 , p=0 . 027 ) significantly improved the model fit . Whereas the addition of CPP amplitude did not ( χ2 ( 1 ) =1 . 43 , p=0 . 23 ) , the addition of CPP ITPC substantially improved the fit of the model ( χ2 ( 1 ) =19 . 25 , p<0 . 001 ) . Neither the LHB build-up rate nor amplitude improved the fit of the model ( LHB build-up rate χ2 ( 1 ) =0 . 02 , p=0 . 88; amplitude χ2 ( 1 ) =0 . 64 , p=0 . 42 ) . Overall , this model suggested that pre-target α-power , CPP onset , build-up rate and ITPC exert partially independent influences on RT . Because some variables were highly correlated ( e . g . CPP onset and ITPC ) we used an algorithm for forward/backward stepwise model selection ( Venables and Ripley , 2002 ) to test whether each neural signal indeed explained independent variability that is not explained by any of the other signals . This procedure eliminated CPP onset ( F ( 1 ) = 0 . 06 , p=0 . 80 ) and build-up rate ( F ( 1 ) = 1 . 86 , p=0 . 17 ) from the final model . Thus , only pre-target α-power and CPP ITPC significantly improved the model fit for predicting RT . These two variables were forced into one linear mixed effects model predicting RT ( Statistical analyses ) , and comparison to a baseline model revealed a good fit ( χ2 ( 2 ) =38 . 61 , p<0 . 001 ) . The fixed effects of the model ( the neural signals ) explained 15 . 8% of the variability in RT ( marginal r2 ) across the five pupil response bins , and together with the random effects ( across subject variability ) it explained 92 . 6% of the variability ( conditional r2 ) . We performed the same hierarchical regression analysis to see which neural signals explained variability in RTcv . We summarised the results of this analysis in Supplementary file 2 , and report the most important results here . The hierarchical regression analysis revealed that both CPP onset and CPP ITPC improved the model fit , but eliminated CPP onset after the forward/backward model selection . Consequently , CPP ITPC was the only variable that exerted independent influence on RTcv . Comparison against a baseline model revealed a significant fit ( χ2 ( 1 ) =15 . 36 , p<0 . 001 ) that had a marginal r2 of 16 . 0% and a conditional r2 of 45 . 9% . To test whether our assumptions about the temporal order of the neural signals influenced these results , we fitted a model in which all EEG signatures were added at the same time and investigated their coefficients . This analysis did not identify any additional neural components to those that were found using the hierarchical regression analysis ( Supplementary file 3 ) . Table 1 shows the final parameter estimates for the neural signals that significantly predicted variability in RT or RTcv that is due to variability in phasic arousal . From this analysis we can conclude that CPP ITPC was the strongest predictor for RT and the only predictor for RTcv . These results provide novel insight into the mechanism by which the neuromodulators that control arousal can influence behaviour . The impact of these modulators on decision-making is thus mainly mediated by their effects on the consistency in decision formation . Next , we turn to tonic arousal and its relationship to these same EEG components of perceptual decision-making . Figure 4 illustrates the relationship between baseline pupil diameter and the CPP . Unlike the pupil response , baseline pupil diameter was not predictive of the onset ( p=0 . 20 ) or build-up rate ( p=0 . 12 ) , but it displayed an inverse relationship with both the amplitude ( χ2 ( 1 ) =7 . 09 , p=0 . 01 ) and the consistency of the CPP ( χ2 ( 1 ) =9 . 34 , p=0 . 002 ) . In line with previous research that revealed increased variability in the rate of evidence accumulation during periods with larger baseline pupil diameter ( Murphy et al . , 2014b ) , we found an inverse , approximately linear , relationship in which higher baseline pupil diameter displayed lower EEG signal consistency ( Figure 4D ) . Thus , states of higher arousal are characterized by less consistency , that is more variability , in decision processing . Additionally , states of higher tonic arousal also display lower task performance ( Figure 1C ) , indicating that higher variability in decision processing ( due to higher tonic arousal ) can affect behavioural performance . We found a relationship between baseline pupil diameter and specific characteristics of multiple neural processing stages of perceptual decision-making . Specifically , as observed before ( Hong et al . , 2014 ) , pre-target alpha power ( Figure 5A ) varied with baseline pupil diameter in a non-monotonic , but not inverted-U shaped , manner ( χ2 ( 1 ) =4 . 49 , p=0 . 034 ) . This suggests that with higher tonic arousal , alpha activity is higher ( or less desynchronised ) . Next , we tested whether baseline pupil diameter was predictive of EEG characteristics representing motor output ( Figure 5B ) . We found an inverse relationship with LHB build-up rate ( χ2 ( 1 ) =10 . 99 , p<0 . 001 ) , decreasing with larger baseline pupil diameter , but we did not find a relationship with LHB amplitude ( p=0 . 34 ) . Lastly , we investigated whether baseline pupil diameter affected early target selection signals , the N2 ( Figure 5C–D ) . Previous studies have revealed that baseline pupil diameter affected the size and variability of neural responses to visual and auditory stimuli ( Reimer et al . , 2014; McGinley et al . , 2015a ) . Here we found that baseline pupil diameter was not predictive of the peak latency of the N2c ( p=0 . 75 ) , but that it did display a monotonic relationship with the N2c amplitude ( χ2 ( 1 ) =13 . 72 , p<0 . 001 ) . Trials with larger baseline pupil diameter displayed smaller N2c amplitudes , suggesting that higher arousal has a negative impact on sensory encoding . N2c ITPC did not vary with baseline pupil diameter ( p=0 . 25 ) , and nor did N2i ITPC ( p=0 . 33 ) , N2i latency ( p=0 . 78 ) or amplitude ( p=0 . 06 ) . We thus found that , similar to the phasic pupil diameter response , baseline pupil diameter is predictive of specific characteristics of each of the processing stages of perceptual decision-making . Next , we investigated which of these components explained unique variance in task performance across pupil size bins . We again performed the same hierarchical regression analysis as described above , to see which of the neural signals explained unique variability in task performance associated with tonic arousal . The full results of this analysis are summarised in Supplementary file 4 . Here we discuss the main findings . After the application of a forward/backward model selection algorithm ( Venables and Ripley , 2002 ) , N2c amplitude and CPP ITPC were the only parameters that were predictive of RT ( Table 1 ) . These variables were forced into one regression model predicting RT , and comparison against a baseline model with baseline pupil diameter as a factor revealed a significant fit ( χ2 ( 2 ) =31 . 6 , p<0 . 001 ) with a marginal ( conditional ) r2 of 4 . 1% ( 94 . 4% ) . This same hierarchical regression procedure revealed that CPP ITPC was the only EEG component that explained unique variability in RTcv ( Table 1 ) . Comparison against a baseline model also led to a significant fit ( χ2 ( 1 ) =26 . 83 , p<0 . 001 ) , with a marginal ( conditional ) r2 of 11 . 9% ( 44 . 5% ) . None of the other EEG parameters that were excluded from the final model due to potential false assumptions about their temporal order revealed significant coefficients in a multilevel model analysis in which all components were added simultaneously ( Supplementary file 5 ) . Thus , additional to an effect of N2c amplitude on RT , the consistency of the CPP was the only stage of information processing that explained unique within and across-subject variability in task performance associated with changes in baseline pupil diameter .
Although the CPP has previously been found to reflect the accumulation of evidence ( O'Connell et al . , 2012; Kelly and O'Connell , 2013; Loughnane et al . , 2016; Loughnane et al . , 2018; Newman et al . , 2017 ) , as discussed in the results section , our task design does not allow us to unequivocally relate the CPP to a specific characteristic of the decision making process such as evidence accumulation . Because of the temporal and spatial uncertainty of the target stimulus , rather than accumulating evidence over an extended period of time on trials with slower RT , target onset transients could be delayed or subjects could be employing different strategies for motion detection on different trials ( e . g . verifying the presence of coherent motion in one stimulus versus the other stimulus ) . Decision signals such as the CPP or LHB could on single trials behave analogously to a step-like signal that across trials seems to be accumulating to a threshold ( Latimer et al . , 2015; but see Shadlen et al . , 2016 ) , potentially supported by neural mechanisms in V4 that increase their activity transiently in response to changes in motion coherence ( Costagli et al . , 2014 ) . Although we cannot discount that subjects use different strategies on different trials , previous studies in which subjects were required to monitor either one or multiple dot kinematograms revealed no differences in either RT or hit rate ( Loughnane et al . , 2016 ) and both the early target selection signals and the CPP scaled with the percentage of coherently moving dots ( Loughnane et al . , 2016 ) . We additionally showed here that there is no fixed delay between target selection and response ( Figure 2—figure supplement 1 ) and that there is thus variability in the duration of the sustained period of the task . Any relationship between arousal and the CPP is therefore not solely the result of fluctuations in the latency of the target onset transient . We estimated the variability in phasic arousal using the amplitude of the task-evoked pupil diameter . Because of the sluggish nature of the pupil diameter response , pupil dilation after target onset likely reflects a combination of specific aspects of phasic arousal such as a response to target onset , decision formation as well as a motor response . Here we aimed to disentangle these different components by applying a general linear model on a single trial basis . First we determined the fit for various models for each subject across trials ( Figure 1—figure supplement 1 ) , after which we applied the best ( across subject ) fitting model to each individual trial . We addressed the reliability of the estimation of each of the temporal components by comparing their relationship to behavioural performance to those of other measures of the amplitude of the pupil diameter response ( Figure 1—figure supplement 2 ) , excluding trials with high VIF values , and orthogonalizing the predictors ( Figure 1—figure supplement 3 ) , and comparing the results from the single trial parameter estimation to those of groups of trials binned by RT ( Figure 1—figure supplement 4 ) . These results revealed that we can reliably estimate the target onset component , but that the estimation of the sustained component might not be as straightforward ( Figure 1—figure supplement 3 ) . Although the current measure of the pupil response to target onset is unlikely to be completely independent of the estimation of the sustained component , inclusion of this predictor increased the fit of the model and captured variability in the pupil time course likely to reflect the influence of phasic arousal specific to decision formation . This reduced the influence of this sustained part of the arousal response on the estimation of the target onset component ( Figure 1—figure supplement 3 ) . To the extent that we could reliably estimate the amplitude of the target onset component , we investigated its relationship to the behavioural and neural signatures of perceptual decision-making . Larger target onset responses , presumably reflecting a phasic response in neuromodulatory brainstem centres , were predictive of faster and less variable RT ( Figure 1C ) , faster onset , larger build-up rates and higher consistency of the CPP ( Figure 2 ) , as well as lower pre-target occipital α-power ( Figure 3 ) . These results can be interpreted in light of the relationship between pupil dilations and the activity in brain areas such as the LC or BF ( Aston-Jones and Cohen , 2005; Gilzenrat et al . , 2010; Varazzani et al . , 2015; Joshi et al . , 2016; Reimer et al . , 2016; de Gee et al . , 2017 ) . Direct electrophysiological recordings from the LC have revealed a positive correlation between LC phasic activity and behavioural performance on elementary target detection tasks ( Aston-Jones et al . , 1994; Aston-Jones et al . , 1997; Rajkowski et al . , 1994; Rajkowski et al . , 2004 ) . Likewise , cue detection is enhanced on trials with a larger cholinergic response ( Parikh et al . , 2007 ) , and previous studies have found that large pupil responses were predictive of higher behavioural performance ( Beatty , 1982b; but see Kristjansson et al . , 2009 ) , and decreased decision bias ( de Gee et al . , 2017 ) . Additionally , poor performance upon pupil constrictions is in line with studies showing that sensory target detection is suboptimal when a transient LC or BF response is absent ( Aston-Jones et al . , 1994; Rajkowski et al . , 1994; Parikh et al . , 2007; Gritton et al . , 2016 ) . Moreover , naturally occurring pupillary constrictions are preceded by transient activity decreases in the LC ( Joshi et al . , 2016 ) , and are associated with increased synchronization of cortical activity , a signature of cortical down states , as well as suboptimal processing of visual stimuli ( Reimer et al . , 2014 ) . Our results suggest that event-related pupillary constrictions could be associated with similar neural mechanisms . Trials with large pupil responses , and better task performance , were preceded by lower pre-target occipital α-power , that is more α desynchronization ( Figure 3 ) . In line with these results and previous studies ( Kelly and O'Connell , 2013 ) , lower pre-target α-power itself was predictive of higher task performance . Fluctuations in α synchronization have previously been related to variation in both arousal and attentional deployment ( Ergenoglu et al . , 2004; van Dijk et al . , 2008; O'Connell et al . , 2009; Kelly and O'Connell , 2013; Newman et al . , 2016 ) , often interpreted as a neurophysiological correlate of cortical excitability . Here , on trials with both higher phasic arousal and more α desynchronization , behavioural performance was better . This could indicate that fluctuations in phasic arousal and attentional engagement rely on similar neuromodulatory mechanisms . We additionally found that larger pupil responses were predictive of earlier onset latencies , faster build-up and higher consistency of the CPP signal ( Figure 2 ) . Thus the effects of the fluctuations in phasic arousal and attentional deployment on task performance are likely mediated by their effect on decision signals , and insofar as the CPP represents evidence accumulation ( see above ) , these fluctuations could influence the build-to-threshold dynamics during perceptual decision-making . We found a non-monotonic relationship between baseline pupil diameter and task performance ( Figure 1B ) . This relationship was , however , not significantly U-shaped , but rather we found slower RT with higher baseline pupil diameter . This effect was moreover only observed when the pupil diameter data was not high-pass filtered ( Figure 1—figure supplement 6 ) , indicating that slow changes ( <0 . 01 Hz ) in pupil diameter are driving the effects on task performance . In line with previous research ( Hong et al . , 2014 ) , out of all the investigated EEG components , only pre-target α power displayed a small non-monotonic relationship with baseline pupil diameter . Approximately linear relationships were found with N2c amplitude , LHB build-up rate , as well as an inverse relationship with CPP amplitude and ITPC . Of these , only N2c amplitude and CPP ITPC explained within and across subject variability in task performance ( Table 1 ) . It thus seems that the effects of tonic arousal on task performance are mainly driven by an approximately linear relationship with target selection and consistency of decision formation . These results appear at odds with a U-shaped relationship as predicted by the adaptive gain theory ( Aston-Jones and Cohen , 2005 ) , and found during auditory target detection tasks ( Murphy et al . , 2011; McGinley et al . , 2015a ) . One potential reason that we did not find a U-shaped relationship with task performance , is that we might not have observed the full range of possible baseline pupil diameter values , and thus not the full range of possible tonic arousal levels . Trials were presented in blocks of 18 , after which subjects were allowed to take a short break , preventing them from becoming overly drowsy or too distracted . However , depending on the behavioural paradigm and task demands , the relationship between central arousal , performance and neural activity may take different forms ( McGinley et al . , 2015b ) . Membrane potential recordings from sensory and association areas , as well as direct electrophysiological recordings from neuromodulatory brainstem centres during decision-making tasks , are needed to gain further insight in the exact mechanisms that drive the relationship between cortical state , sensory encoding , decision formation and task performance . During epochs of quiet wakefulness , membrane potential fluctuations of neurons in visual , somatosensory and auditory cortex are closely tracked by baseline pupil diameter ( Reimer et al . , 2014; McGinley et al . , 2015a ) . These fluctuations in subthreshold membrane potential are characteristic of changing cortical state . Small pupil diameter is characterized by prominent low-frequency ( 2–10 Hz ) and nearly absent high-frequency oscillations ( 30–80 Hz ) , whereas larger pupil diameter is characterized by reduced low-frequency , but increased high-frequency oscillations ( McGinley et al . , 2015a; McGinley et al . , 2015b ) . Thus , the average subthreshold membrane potential is most stable during intermediate pupil diameter , when neither low nor high-frequency components predominate . States of lower variability are furthermore characterized by more reliable sensory responses , higher spike rates , increased neural gain and better behavioural performance ( Reimer et al . , 2014; McGinley et al . , 2015a; McGinley et al . , 2015b ) . In addition to activity in early sensory areas , there is some evidence that activity in higher-order association areas is also more reliable with intermediate arousal . During auditory target detection , human subjects displayed the least variable RT at intermediate baseline pupil diameter , as well as the highest amplitudes of the P3 component elicited by task-relevant stimuli ( Murphy et al . , 2011 ) . Here we found that the consistency of the CPP was the main EEG predictor of variability in task performance associated with both tonic and phasic arousal . For tonic arousal , our findings are largely in line with modelling studies which suggested that higher arousal is specifically predictive of more variability in evidence accumulation ( Murphy et al . , 2014b ) . For phasic arousal , higher consistency , and thus less variability , was found for larger pupil bins , which also displayed the best behavioural performance . These results suggest that similar neural mechanisms of cortical state described for sensory cortex ( Reimer et al . , 2014; McGinley et al . , 2015b; McGinley et al . , 2015a; Vinck et al . , 2015 ) might also affect neurons in higher-order association areas ( e . g . parietal cortex ) and thereby influence evidence accumulation and task performance . Simultaneous pupil diameter and membrane potential recordings in parietal cortex during decision-making are needed to confirm this hypothesis . In the present study , we used a paradigm in which two stimuli were continuously presented and target occurrence was both spatially and temporally unpredictable . Successful target detection thus relied on locating and selecting sensory evidence from multiple sources of information . Loughnane et al . ( 2016 ) have shown that early target selection signals , which occur contralateral to the target stimulus ( N2c ) , modulate sensory evidence accumulation and behavioural performance . Although previous studies have characterised the dependence of the quality of sensory responses on fluctuations in cortical state , as measured by baseline pupil diameter ( Reimer et al . , 2014; McGinley et al . , 2015a; Vinck et al . , 2015 ) , to the best of our knowledge , the influence of pupil-linked arousal on target selection signals has not been described before . Here , we showed that early target selection signals are modulated by tonic arousal such that larger baseline pupil diameter was predictive of smaller N2c amplitudes ( Figure 5C ) . Moreover , the amplitude of the N2c also explained unique variability in task performance across pupil bins and subjects ( Table 1 ) . At first glance it seems counterintuitive that target selection signal amplitudes are decreased , whereas visual encoding in early visual cortex is enhanced on trials with larger baseline pupil diameter ( Vinck et al . , 2015 ) , or during pupil dilation ( Reimer et al . , 2014 ) . These differences could be due to differences in the nature of the recordings , as these previous studies used invasive electrophysiology and calcium imaging whereas we used scalp EEG , limiting especially the spatial resolution of our analyses that might be necessary to elucidate these effects ( e . g . single neuron orientation tuning ) . Alternatively , they could constitute differential effects of arousal on visual encoding and target selection . More likely , however , they are due to specific task demands , in particular our use of multiple simultaneously presented competing stimuli . Indeed , there is some evidence that an increase in arousal , as measured by pupil diameter , can increase the ability of a distractor to disrupt performance on a Go/No-Go task in non-human primates ( Ebitz et al . , 2014 ) . At high arousal levels , performance might thus be negatively affected when the task requires the successful suppression of distracting information , that is with higher arousal it is more difficult to focus on the task at hand ( Aston-Jones and Cohen , 2005; McGinley et al . , 2015b ) . On the current task , it might thus be more difficult to select and process information from one of the two competing stimuli during states of high arousal , leading to reduced N2c amplitude as well as reduced performance . As in previous studies ( Gilzenrat et al . , 2010; Murphy et al . , 2011; de Gee et al . , 2014 ) , we found a negative correlation between baseline pupil diameter and the size of the pupillary response . Both measures were predictive of task performance as well as a unique , but overlapping , set of EEG signatures of perceptual decision-making . Because of the overlap in their effects on these EEG markers , in particular pre-target α power and CPP ITPC , it is possible that both ( in part ) reflect the same component of central arousal state . Although we removed ( via linear regression ) the variance in the pupil response that is due to fluctuations in the amplitude and the phase of the baseline pupil diameter , some variability in the baseline pupil diameter might not be fully dissociable from the pupil response , and both might thus reflect a noisy measure of tonic arousal . This interpretation is further supported by the finding that the relationship between the pupil response and task performance did not substantially change regardless of whether variability in the pupil response due to fluctuations in baseline amplitude and/or phase was removed or not ( Figure 1—figure supplement 5 ) . Importantly , however , the dissociation in the effect of baseline pupil diameter and the pupil response on these EEG markers , such as the effect on N2c amplitude , indicates that these measures also capture independent variability in central arousal ( tonic and phasic ) predictive of distinct information processing stages of decision-making . In this study we investigated the relationship between measures of tonic and phasic pupil-linked arousal and behavioural and EEG measures of perceptual decision-making . We found that trial-to-trial variability in both tonic and phasic arousal accounted for variability in task performance and were predictive of a unique , but overlapping , set of neural metrics of perceptual decision-making . Specifically , tonic arousal exerted its influence on task performance through its effects on early target selection signals and the consistency of decision formation . Phasic arousal , on the other hand , affected behaviour through its relation with attentional engagement as well as the consistency of decision formation . These results indicate that during decision-making both tonic and phasic activity in the ( network of ) neuromodulatory centres that control central arousal can affect behaviour during perceptual decision-making . Thus , fluctuations in central arousal , mediated by neuromodulatory brainstem centres , act on multiple timescales to influence task performance through its effects on attentional engagement , sensory processing as well as decision formation .
Subjects ( n = 80 ) and methods are largely overlapping with the details and procedures described elsewhere ( Newman et al . , 2017 ) . Here we summarise details necessary to understand this study , and we also describe procedures that differ from the previous study . Participants were seated in a darkened room , 56 cm from the stimulus display ( 21 inch CRT monitor , 85 Hz 1024 × 768 resolution ) , asked to perform a continuous bilateral variant ( O'Connell et al . , 2012; Kelly and O'Connell , 2013 ) of the random dot motion task ( Newsome et al . , 1989; Britten et al . , 1992 ) . Subjects fixated on a central dot while monitoring two peripheral patches of continuously presented randomly moving dots ( Figure 1A ) . At pseudorandom times , an intermittent period of coherent downward motion ( 50% ) occurred in either the left or the right hemifield . Upon detection of coherent motion , participants responded with a speeded right-handed button press . A total of 288 trials were presented over 16 blocks ( 18 trials per block ) . Data were collected under identical experimental procedures at either Monash University , Australia , or Trinity College Dublin , Ireland . The experimental protocol was approved by each University’s human research ethics committee before testing ( Project number Monash University: 3658 , Trinity College: SPREC012014-1 ) , and carried out in accordance with approved guidelines . Informed consent was obtained from all participants before testing . Electroencephalogram ( EEG ) was recorded from 64 electrodes using an ActiveTwo ( Biosemi , 512 Hz ) system at Trinity College Dublin , Ireland or a BrainAmp DC ( Brainproducts , 500 Hz ) at Monash University , Australia . Data were processed using both custom written scripts and EEGLAB functions ( Delorme and Makeig , 2004 ) in Matlab ( MathWorks ) . Noisy channels were interpolated after which the data were notch filtered between 49–51 Hz , band-pass filtered ( 0 . 1–35 Hz ) , and rereferenced to the average reference . Data recorded using the Biosemi system were resampled to 500 Hz and combined with the data recorded with the Brainproducts system . Epochs were extracted from −800 to 2800 ms around target onset and baselined with respect to −100 to 0 ms before target onset . To minimize volume conduction and increase spatial specificity , for specific analyses the data were converted to current source density ( Kayser and Tenke , 2006 ) . We rejected trials from analyses if the reaction times were <150 or>1700 ms after coherent motion onset , or if either the EEG on any channel exceeded 100 mV , or if the subject broke fixation or blinked ( Pupillometry ) during the analysis period of the trial , the 500 ms preceding target onset ( 26 . 59 ± 2 . 94 ) for pre-target α power activity or the interval of 100 ms before target onset to 200 ms after the response ( 33 . 66 ± 3 . 95 ) . Pre-target α-band power ( 8–13 Hz ) , N2 amplitude and latency , CPP onset and build-up rate and response related β-power amplitude and build-up rate were computed largely in the same way as in Newman et al . , 2017 . Briefly , α-band power was computed over the 500 ms preceding target onset using temporal spectral evolution ( TSE ) methods ( Thut et al . , 2006 ) , and pooled over two symmetrical parietal regions of interest , using channels O1 , O2 , PO3 , PO4 , PO7 and PO8 . The N2 components were measured at electrodes P7 and P8 , ipsi- and contralateral to the target location ( Loughnane et al . , 2016; Newman et al . , 2017 ) , and the CPP was measured at central electrode Pz . These signals were aggregated to an average waveform for each pupil bin and each participant . We determined the latency of the N2c/N2i as the time point with the most negative amplitude value in the stimulus-locked waveform between 150-400/200-450 ms , while N2c/N2i amplitude was measured as the mean amplitude inside a 100 ms window centered on the stimulus-locked grand average peak ( 266/340 ms ) ( Loughnane et al . , 2016; Newman et al . , 2017 ) . Onset latency of the CPP was measured by performing running sample point by sample point t-tests against zero across each participant’s stimulus-locked CPP waveforms . CPP onset was defined as the first point at which the amplitude reached significance at the 0 . 05 level for ≥ 15 consecutive points . Because we decreased our statistical power by binning the trials into five bins ( see pupillometry ) , we did not find an onset for every bin for a subset of subjects ( baseline pupil diameter: 13 bins over 11 subjects , pupil response: 16 bins over 12 subjects ) . Because of our use of linear mixed effect analyses , these subjects could still be included in the analysis , with only the missing values being omitted . Both CPP build-up rate and amplitude were computed using the response-locked waveform of the CSD transformed data to minimize influence from negative-going fronto-central scalp potentials ( Kelly and O'Connell , 2013 ) . Build-up rate was defined as the slope of a straight line fitted to this signal in the window from −250 ms to −50 ms before response . CPP amplitude was defined as the mean amplitude within the 100 ms around the response . Response related left hemisphere β-power ( LHB , 20–35 Hz ) was measured over the left motor cortex at electrode C3 using short-time Fourier transform ( STFT ) with a 286 ms window size and 20 ms step size ( O'Connell et al . , 2012; Newman et al . , 2017 ) . We baselined LHB using an across-trial baseline for each subject . LHB amplitude was measured from the response-locked waveform in the window from −130 to −70 ms preceding the response , whereas the LHB build-up rate was defined as the slope of a straight line fitted to this same waveform in the 300 ms before the response . Inter-trial phase coherence ( ITPC ) was estimated using single-taper spectral methods from the Chronux toolbox ( Bokil et al . , 2010 ) and adapted scripts . We used a 256 sample ( 512 ms ) sliding short time window , with a step size of 25 samples ( 50 ms ) . This gave us a half bandwidth ( W ) of 1 . 95 Hz: W = ( K + 1 ) /2T , with K being the number of data tapers , K = 1 , and T ( s ) being the length of the time window . Frequencies were estimated from 0 . 1 to 35 Hz . Eye movements and pupil data were recorded using an SR Research EyeLink eye tracker ( Eye-Link version 2 . 04 , SR Research/SMI ) . Automatically identified blinks and saccades were linearly interpolated from 100 ms before to 100 ms after the event , the interpolated pupil data was then low-pass ( <6 Hz ) or band-pass ( 0 . 01–6 Hz ) filtered ( second order butterworth ) . The instantaneous phase of the pupil diameter was calculated by taking the angle of the analytic signal acquired by using the Hilbert transform of the filtered data . Epochs were extracted from −800 to 4800 ms around coherent motion onset . Trials in which fixation errors or blinks occurred within the analysis period , from 100 ms before target onset to 200 ms after response , were excluded from analysis . Fixation errors were defined as gaze deviations of more than 3° . The pupil diameter was normalized by dividing by the maximum pupil diameter on any trial in the analysis window from 100 ms before target onset to 200 ms after the response for each subject , and baselined on a single trial basis . We computed the baseline pupil diameter by averaging the pupil diameter in the 100 ms before target onset , and the baseline phase was calculated as the average phase angle in the 100 ms preceding target onset . We identified the shape of the neural input to the pupil system by applying various general linear models ( GLM ) to the pupil time course ( Hoeks and Levelt , 1993; de Gee et al . , 2014; Murphy et al . , 2016 ) with two temporal components corresponding to target and response onset ( all models ) , and a third sustained component ( models 2–9 ) for which the shape varied across eight candidate models tested previously ( Figure 5 in Murphy et al . , 2016 ) . In model 1 , only the stimulus and response onset were modelled . The sustained component in the remaining models took the shape of: ( model 2 ) a boxcar component with a constant amplitude throughout the decision interval; ( model 3 ) a linear up-ramp that grew in amplitude with increasing decision time; ( model 4 ) a ramp-to-threshold; ( model 5 ) a linear decay with a starting amplitude that was larger for slower RTs but whose amplitude always terminated at zero; ( model 6 ) a linear decay-to-threshold which began at a fixed amplitude and terminated at zero; ( model 7–9 ) versions of the boxcar , up-ramp and down-ramp models in which the sustained component was normalized by the number of the samples in that trial’s decision interval . We convolved these onset , response and/or sustained temporal components with a pupil impulse response function ( IRF ) :h ( t ) = twe−t ( w/tmax ) where w is the width ( 10 . 1 ) and tmax is the time-to-peak ( 930 ms ) of the IRF ( Hoeks and Levelt , 1993; de Gee et al . , 2014; Murphy et al . , 2016 ) . Each model was regressed onto the concatenated band-pass filtered pupil diameter time series ( from -800 ms before target onset to 2500 ms after the response ) . Bayes information criterion ( BIC ) was used to assess model fit:BIC=n+nlog2π+ log ( SSR/n ) +k+1lognwhere n is the number of samples , SSR is the residual sum of squares , and k is the number of free parameters . The goodness of fit between any two models was assessed non-parametrically by applying Wilcoxon signed rank tests to their difference score . We found that the linear up-ramp model ( model 3 ) provided the best fit to the data . Figure 1—figure supplement 1 illustrates the relative goodness-of-fit of each model , compared to the best fitting linear up-ramp model , as well as the effect size of each of the components of the linear up-ramp model . To investigate the relationship between pupil-linked arousal and behavioural performance during decision-making , we binned our behavioural and EEG data according to either the baseline pupil diameter or the post target pupil response ( see below ) into five equally sized bins ( mean 49 . 63 ± SEM 0 . 81 trials , minimum bin size = 20 trials ) ( Figure 1B & C ) . The division into five bins allowed us to investigate possible quadratic trends in the data . We used linear regression to remove the trial-by-trial fluctuations in single-trial pupil amplitudes that could be due to inter-trial interval , target side , as well as baseline pupil diameter amplitude or phase , all factors that are known to influence either the post target pupil response and/or behavioural response times ( Kristjansson et al . , 2009; de Gee et al . , 2014; Kloosterman et al . , 2015; Newman et al . , 2017 ) . To partial out the effect of phase , a circular variable , we used the sine and cosine of the phase as orthogonal , linear predictor variables ( Fisher , 1993 ) . To verify the ( absence of ) correlation between pupil baseline phase and response before and after the regression , we made use of the circstat toolbox ( Berens , 2009 ) . We estimated the task-evoked phasic arousal response according to various single-trial scalar measurements of the amplitude of the pupil response ( Figure 1—figure supplement 2 ) . The relationship between the average pupil diameter around RT and behavioural performance was best described by a non-monotonic , U-shaped , relationship ( Figure 1—figure supplement 2A ) . Because of the temporal low-pass characteristics of the peripheral pupil system ( Hoeks and Levelt , 1993 ) , trial-to-trial variation in RT can affect the measurement of the size of the pupil response . To remove the trial-to-trial fluctuations in pupil responses due to variations in RT , we removed these components via linear regression ( de Gee et al . , 2017; Urai et al . , 2017 ) . After the elimination of the contribution of RT to the pupil response , we still observed a U-shaped relationship with behavioural performance ( Figure 1—figure supplement 2B ) . This measure of the pupil response , however , likely reflects both the transient response to target onset as well as any activity that occurs thereafter ( e . g . during decision formation ) . Therefore , we aimed to isolate activity specific to the phasic response to target onset . To this end , we computed the mean , slope and linear projection ( de Gee et al . , 2014; Kloosterman et al . , 2015 ) over a 400 ms time window around the peak of the derivative of the pupil IRF ( 636 ms using the canonical IRF ) . A time-window in which activity occurring after the target onset transient is , presumably , not yet reflected . We found an inverse relationship between each of these measures and behavioural performance ( Figure 1—figure supplement 2C–E ) , with better behavioural performance for larger pupil responses . Although these results suggest that measurements of the pupil response in this time-window reflect a different component of the neural input to the pupil system than the measurements of the amplitude around RT ( Figure 1—figure supplement 2A & B ) , the use of any specific time window can be interpreted as arbitrary . To further disentangle the pupil response into separate temporal components , we applied the best-fitting GLM , the linear up-ramp model ( Figure 1—figure supplement 1 ) , to individual trials by considering each individual trial as a separate condition ( Bach et al . , 2018 ) . Because we reduced the amount of data used for the regression analysis by applying it to single-trial data , we tested whether this led to collinearity amongst the temporal components by computing the variance inflation factor ( VIF ) . Although large VIF values do not necessarily imply that no conclusions can be drawn from regression analysis ( O’brien , 2007 ) , as a rule of thumb , VIF values larger than 5 or 10 indicate that predictors are collinear ( Sheather , 2009; James et al . , 2017 ) . When applying the GLM across all trials , the average VIF values are within the range of collinearity ( Figure 1—figure supplement 3A–B ) . When we apply the same model to single trial data , however , the average VIF values are substantially higher ( Figure 1—figure supplement 3C–D ) . It seemed particularly problematic to reliably estimate the sustained and the response component as their VIF scores are larger than 10 . The target onset component , on the other hand , has an average VIF score of approximately 5 . Single-trial VIF estimates larger than five for target onset ( 39 . 34 ± 2 . 84% of trials ) were mainly found on trials with short RT ( Figure 1—figure supplement 3E ) , revealing that it is difficult to distinguish between these temporal components on short trials . The overall results were , however , not affected by these trials . Repeating the analysis when excluding trials with VIF values larger than five revealed the same relationship pattern between pupil response amplitude and behavioural performance ( Figure 1—figure supplement 3F&G ) . Sorting the pupil diameter according to the estimate of the amplitude of the sustained component , revealed that the largest sustained component occurred on trials with a small ( or absent ) response to target onset ( Figure 1—figure supplement 3H ) . Rather than solely reflecting phasic arousal during decision formation , the presence of the sustained component could , for instance , indicate a compensatory mechanism for the absence of an early target onset transient . As the relationship with behavioural performance followed a downward trend when plotted against the target onset component ( Figure 1C ) , and an upward trend when plotted against the sustained component ( Figure 1—figure supplement 3I ) , together these effects could explain the U-shaped relationship between behavioural performance and the pupil response when measured as the average pupil diameter around RT ( Figure 1—figure supplement 2A–B ) . Although a target-response onset only model was the worst fitting model across trials ( Figure 1—figure supplement 1 ) , we tested whether a target-response only model could reliably estimate the single-trial target-onset response amplitude . The relationship between this component and behavioural performance ( Figure 1—figure supplement 3K&L ) , however , strongly resembled the U-shaped relationship between behaviour and the pupil response amplitude when calculated as the mean amplitude around RT ( Figure 1—figure supplement 2A ) , a measure likely to be confounded by both RT and the neural input that occurs after the target onset transient . This supports the notion that the inclusion of a sustained component can make the estimation of the target onset component ( amongst others ) more accurate , despite the potential collinearity of these predictors . Indeed , the difference in model fit ( R2 ) is significantly larger than 0 for each individual subject ( one-sided Wilcoxon signed rank test , data not shown ) . Figure 1—figure supplement 3J illustrates the average difference in R2 values between single-trial models with and without the sustained component . Lastly , Figure 1—figure supplement 3M&N illustrate the actual pupil diameter time course and the single trial fitted pupil diameter , revealing that this model is able to capture considerable variability in the pupil diameter trace . Next , we applied the same linear up-ramp model to five subsets of trials , binned by RT ( average bin size: 50 . 01 ± 0 . 82 trials ) . This analysis revealed that the relationship between RT bin and the estimated amplitude of the target component ( Figure 1—figure supplement 4A ) follows a pattern that is highly similar to the relationship between single-trial estimates of the phasic pupil response to target onset and RT ( Figure 1C ) , further supporting the notion that the single-trial GLM approach can accurately estimate the target onset transient . We again investigated the VIF values for each of the temporal components of the model applied to the binned data . Although the sustained and response components displayed relatively large values , the target onset component was smaller than 5 . Again , large VIF values by themselves are not necessarily cause for concern , if a regression coefficient is statistically significant , even when its VIF value is large , it is significant ‘in the face of that collinearity’ ( O’brien , 2007 ) . To further exclude the possibility that large VIF values brought about these results , we repeated this analysis using the data binned according to RT in three or two bins ( Figure 1—figure supplement 4B–C ) . These analysis also revealed smaller target onset component coefficients for larger RT , with progressively lower VIF values . Finally , we investigated the relationship between RT and the target onset component after Gram-Schmidt orthogonalization of the predictors ( Figure 1—figure supplement 4D–E ) , which eliminated collinearity amongst the temporal components . After orthogonalization , we again found that the estimate of the β weights of the target onset component was inversely related to RT , both when estimated across bins of trials ( Figure 1—figure supplement 4D ) as well as when estimating this component on a single trial basis ( Figure 1—figure supplement 4E ) . Altogether , these analyses reveal that although the estimation of different temporal components contributing to a single-trial pupil diameter time course has to be done with caution , in the context of the various measures of the phasic pupil response ( Figure 1—figure supplement 2 ) and the interpretation of VIF factors ( Figure 1—figure supplement 3 & Figure 1—figure supplement 4 ) , it is possible ( in this dataset ) to extract meaningful estimates of the target onset component . We used RStudio ( RStudio Team ( 2016 ) . RStudio: Integrated Development for R . RStudio , Inc . , Boston , MA URL http://www . rstudio . com ) with the package lme4 ( Bates et al . , 2015 ) to perform a linear mixed effects analysis of the relationship between baseline pupil diameter or the pupil response and behavioural measures and EEG signatures of detection . As fixed effects , we entered pupil bin ( see Pupillometry ) into the model . As random effects , we had separate intercepts for subjects , accounting for the repeated measurements within each subject . We sequentially tested the fit of a monotonic relationship ( first-order polynomial ) against a baseline model ( zero-order polynomial ) , and a non-monotonic ( second-order polynomial ) against the monotonic fit by means of maximum likelihood ratio tests , using orthogonal polynomial contrast attributes . The behavioural or EEG measure y was modelled as a linear combination of polynomial basis functions of the pupil bins ( X ) :y ~ β0+ β1X+ β2X2with β as the polynomial coefficients . This multilevel approach was preferred over a standard repeated measures analysis of variance ( ANOVA ) , because it allowed us to test for first and second-order polynomial relationships , as well as to account for missing values in the CPP onset estimation . We used a variant of the ‘two-lines’ approach ( Simonsohn , 2017 ) , to test for the presence of ( inverted ) U-shape relationships when a second-order polynomial best fit the data . Using the same multilevel model , we fit two straight lines to the first and last set of two/three bins . For a non-monotonic relationship to be classified as U-shaped , both components needed to have significant coefficients of opposite sign . We iteratively tested the first 3 against the last 2 , the first 2 against the last 3 or the first 2 against the last 2 bins ( omitting the middle bin ) , stopping if both criteria were met ( p < 0 . 05 , Bonferroni corrected ) . To verify that the relationship between pupil diameter and task performance was not dependent on the binning procedure , we ran another regression analysis wherein we predicted single trial RT by sequentially adding the linear and quadratic coefficients for baseline pupil diameter ( BPD ) and pupil response ( PR ) :RT ~ β0+ β1BPD+ β2BPD2+ β3PR+ β4PR2with β as the polynomial coefficients . We compared the first model to a random-intercept-only model including subject ID , inter-trial interval , stimulus side , as well as the trial and block number ( to control for potential time on task effects ) , and tested the fit of subsequent models to the previous model fit . This analysis revealed a significant improvement for each step of the sequential analysis , for which the results and parameters estimates are shown in Supplementary file 1 . These analyses confirm that both the size of the baseline pupil diameter and the pupil response are predictive of task performance on a single trial basis . This relationship moreover follows a non-monotonic , quadratic , function . After testing the relationship between behavioural and neural signatures of decision-making and pupillometric measures individually , the neural signals were added sequentially into consecutive regression models predicting RT and RTcv . This model had both a random intercept for each subject , allowing for different baseline-levels of behavioural performance , as well as a random slope of pupil bin for each subject , which allowed for across-subject variation in the effect of pupil bin on behavioural performance . The hierarchical entry of the predictors allowed us to model the individual differences in behavioural performance , as a function of the EEG signals representing each temporal stage of neural processing . Starting with preparatory signals ( α-power ) , to early target selection signals ( N2 ) , to evidence accumulation ( CPP ) , to motor preparation ( LHB ) . The hierarchical addition of the predictors informed us whether each of the EEG signals reflecting successive stages of neural processing improved the fit of the model predicting behavioural data . The signals that explained unique variance were then simultaneously forced into a simplified model predicting RT or RTcv , which made it possible to obtain accurate parameter estimates not contaminated by signals that were shown not to improve model fits . Note that only subjects for which we could determine the CPP onset latency for all bins were included in this hierarchical model . For this final model , all behavioural and neural variables were scaled between 0 and 1 across subjects according to the formula:yi= ( xi−minxi ) / ( maxxi−minxi ) , where yi is the scaled variable , xi is the variable to be scaled . This scaling procedure did not change the relationship of the variable within or across subjects , but scaled all predictor variables to the same range . Again , significance values were obtained by means of maximum likelihood ratio tests . Data plotted in all figures are the mean and the standard error of the mean ( SEM ) across subjects . Linear fits are plotted when first-order fits were superior to the zero-order ( constant ) fit , quadratic fits are plotted when second-order fits were superior to the first-order fit . Raw data ( https://figshare . com/s/8d6f461834c47180a444 ) are open access and available under a Creative Commons Attribution-NonCommercial-ShareAlike 3 . 0 International Licence . Analysis scripts are freely available on github ( van Kempen , 2019; copy archived at https://github . com/elifesciences-publications/2019_pupil_decisionMaking ) . | Driving along a busy street requires you to constantly monitor the behavior of other road users . You need to be able to spot and avoid the car that suddenly changes lane , or the pedestrian who steps out in front of you . How fast you can react to such events depends in part on your brain's level of alertness , or 'arousal' . This in turn depends on chemicals within the brain called neuromodulators . Neuromodulators are a type of neurotransmitter . But whereas other neurotransmitters enable brain cells to signal to each other , neuromodulators turn the volume of these signals up or down . The activity of brain regions that produce neuromodulators varies over time , leading to changes in brain arousal . These changes take place over different time scales . Sudden unexpected events , such as those on the busy street above , trigger sub-second changes in arousal . But arousal levels also show spontaneous fluctuations over minutes to hours . We can follow these changes in real-time by looking into a participant’s eyes . This is because the brain regions that produce neuromodulators also control pupil size . Van Kempen et al . have now combined measurements of pupil size with recordings of electrical brain activity . Healthy volunteers learned to press a button as soon as a target appeared on a screen . The larger a volunteer’s pupils were before the target appeared , the more slowly the volunteer responded on that trial . Large baseline pupil size is thought to indicate a high baseline level of brain arousal . By contrast , the larger the increase in pupil size in response to the target , the faster the volunteer responded on that trial . This increase in pupil size is thought to reflect an increase in brain arousal . The recordings of brain activity provided clues to the underlying mechanisms . In trials with large baseline pupil size – and therefore high baseline arousal – the volunteers’ brains showed more variable responses to the target . But in trials with a large increase in pupil size – and a large increase in arousal – the volunteers’ brains showed less variable responses , as well as stronger signals related to attention . Neuromodulators thus act on different timescales to influence different aspects of cognitive performance , including attention and target detection . Fluctuating levels of neuromodulator activity may help explain the variability in our behavior . Monitoring pupil size is one way to gain insights into the mechanisms that bring about these changes in neuromodulator activity . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2019 | Behavioural and neural signatures of perceptual decision-making are modulated by pupil-linked arousal |
In malaria , rosetting is described as a phenomenon where an infected erythrocyte ( IRBC ) is attached to uninfected erythrocytes ( URBC ) . In some studies , rosetting has been associated with malaria pathogenesis . Here , we have identified a new type of rosetting . Using a step-by-step approach , we identified IGFBP7 , a protein secreted by monocytes in response to parasite stimulation , as a rosette-stimulator for Plasmodium falciparum- and P . vivax-IRBC . IGFBP7-mediated rosette-stimulation was rapid yet reversible . Unlike type I rosetting that involves direct interaction of rosetting ligands on IRBC and receptors on URBC , the IGFBP7-mediated , type II rosetting requires two additional serum factors , namely von Willebrand factor and thrombospondin-1 . These two factors interact with IGFBP7 to mediate rosette formation by the IRBC . Importantly , the IGFBP7-induced type II rosetting hampers phagocytosis of IRBC by host phagocytes .
Along its intraerythrocytic development , Plasmodium spp . modifies the infected erythrocyte ( IRBC ) rheology ( increased rigidity for P . falciparum IRBC , reduced rigidity but increased fragility for P . vivax IRBC ) . Such alteration increases the susceptibility of IRBC to splenic clearance ( Handayani et al . , 2009; Chotivanich et al . , 2002 ) ; however , the parasites have developed escape strategies to avoid splenic elimination . For instance , P . falciparum expresses adhesins on the IRBC that mediate adhesion to the endothelial cells , resulting in deep microvasculature sequestration ( Sherman et al . , 2003 ) . Additionally , an IRBC can bind directly to uninfected red blood cells ( URBC ) to form a ‘rosette’ ( Russell and Cooke , 2016 ) . This rosetting phenomenon has been described in all human malaria parasites ( Angus et al . , 1996 ) ; however , the functional importance of rosetting remains ambigous ( Clough et al . , 1998; Wahlgren et al . , 1989 ) . While the supposed role of rosetting in facilitation of merozoite invasion of URBC is unlikely; recent studies show that rosette formation may have a role in parasite immune-evasion ( Clough et al . , 1998; Deans and Rowe , 2006; Lee et al . , 2014; Zhang et al . , 2016; Moll et al . , 2015; Lee et al . , 2019 ) . Theoretically , the masking of rosetting IRBC with URBC may hamper IRBC recognition and therefore their clearance by the host immune system ( Moll et al . , 2015 ) . Notably , rosetting has been associated ( in some but not all studies ) with disease severity ( Treutiger et al . , 1992; Carlson et al . , 1994; Rowe et al . , 1995; Doumbo et al . , 2009; al-Yaman et al . , 1995; Ho et al . , 1991 ) . Here , we observed that the addition of leukocytes increased the rosetting rates of various P . falciparum and P . vivax isolates . We next demonstrated that IRBC stimulated monocytes to secrete products capable of stimulating rosetting , the most important being insulin growth factor binding protein 7 ( IGFBP7 ) . We further showed that IGFBP7-mediated rosetting was different from the previously described rosetting ( defined here as type I rosetting ) , where it ( we refer to this as type II rosetting ) required additional serum factors to occur , in addition to the interaction between the parasite-derived ligand on IRBC surface and the receptor on the surface of URBC . Functionally , we observed that the IGFBP7-mediated type II rosetting reduced phagocytosis by monocytes , and therefore defined a new escape mechanism for the malaria parasites .
P . falciparum- and P . vivax-IRBC formed rosettes ( Figure 1A , top panel ) in the presence of 20% autologous human serum and the extent of rosetting at baseline is variable depending on the parasite isolates ( Lee et al . , 2014 ) . When autologous blood leukocytes were incubated with clinical isolates , the rosetting rate increased by 10–40% depending on individual isolates for both parasite species ( Figure 1B ) . This effect was mediated primarily by monocytes ( Figure 1C ) . To exclude donor variability as a confounding factor , we repeated the experiment with the THP-1 cell line , which was derived from the peripheral circulation of an acute monocytic leukaemia patient ( Auwerx , 1991 ) . The rosette-stimulation mediated by THP-1 was similar to that of peripheral monocytes ( Figure 1D ) . Addition of undifferentiated THP-1 ( UT ) and macrophage-like THP-1 ( MT ) increased rosetting rates in a dose-dependent manner . MT were more potent than UT , where the difference increased with the number of cells added ( Figure 1E ) . Culture supernatants ( CS ) of both THP-1 cell types showed similar rosette-stimulation effects ( Figure 1F ) , with CSMT exerting a higher degree of rosette-stimulation than CSUT . Fractionation of CSMT into aqueous and lipid fractions revealed that the rosette-stimulating factors were in the aqueous fraction ( Figure 2A ) . Subsequently , we further fractionated the aqueous fraction into high and low molecular weight sub-fractions ( with a cut-off of 30 kDa ) . Both aqueous sub-fractions induced rosetting of P . falciparum and P . vivax ( Figure 2B and C ) , demonstrating that the rosetting stimulation was mediated by multiple secreted hydrophilic factors , predominantly of sizes ≤ 30 kDa ( particularly for P . falciparum ) . Heating of CSMT’s ≤ 30 kDa aqueous fraction at 56°C for 1 h did not completely abolish its rosette-stimulating effect ( Figure 2D ) , indicating the presence of heat-stable factors . To further investigate the potential stimulating factors , mass spectrometry analysis was performed on this fraction . We identified 694 proteins ( Supplementary file 2 ) , of which complement factor D ( CFD ) , insulin-like growth factor binding protein 7 ( IGFBP7 ) , nidogen 1 ( NID1 ) , hyaluronan-binding protein 2 ( HABP2 ) , and periostin ( OSF2 ) were shortlisted ( Supplementary file 3 ) . Apart from having high scores in the mass spectrometry analysis ( Supplementary file 2 ) , these molecules were selected because they are secreted proteins and possess cell adhesion-related properties associated with pathological changes such as vascular injuries , inflammation , cancer development , and leukocyte recruitment ( UniProt Consortium , 2015; St Croix et al . , 2000; Mambetsariev et al . , 2010; Ulazzi et al . , 2007; Schwanekamp et al . , 2016; Lindner et al . , 2005; Markiewski and Lambris , 2007 ) . Antibody neutralization of IGFBP7 significantly reduced ( by ~40% ) the rosette-stimulation by CSMT for both parasite species ( Figure 3A ) . Rosette-stimulation by CSMT was also reduced by anti-CFD ( Figure 3B ) and anti-OSF2 ( Figure 3C ) antibodies , albeit to a lesser extent . Antibodies against NID1 ( Figure 3D ) and HABP2 ( Figure 3E ) had no effect on CSMT-mediated rosette-stimulation . As anti-IGFBP7 had the largest inhibitory effect , further experiments focused on IGFBP7 . Addition of human recombinant IGFBP7 to leukocyte-free parasite culture stimulated rosette formation in a dose-dependent and satiable manner for P . falciparum and P . vivax isolates , reaching a plateau at 100 ng/ml ( Figure 4A and B ) . For P . falciparum , no significant difference in rosette-stimulatory effect between the CSMT and IGFBP7 was found ( Figure 4C ) . On the contrary , P . vivax rosette-stimulation by CSMT was significantly higher than that by recombinant IGFBP7 alone ( Figure 4D ) . The rosette-stimulating capacity of IGFBP7 was abolished by heat denaturation at 95°C for 1 h ( Figure 4E and F ) , thus eliminating the possibility that the effect was caused by a non-protein contaminant in the recombinant protein preparation . Furthermore , the IGFBP7 binding-inducing effect was only observed when URBC and IRBC were present in the culture . Incubation of URBC alone with IGFBP7 did not induce clustering effect on the cells ( Figure 4—figure supplement 1; top panel ) . IGFBP7 required a minimum of 15 min to significantly stimulate rosetting . However , the rosette-stimulating effect did not significantly increase further afterwards ( Figure 4G ) . Importantly , IGFBP7-mediated rosetting was a reversible event , where removal of the protein from the system reverted the rosetting rates to their baseline values ( rosetting rates recorded prior to IGFBP7 exposure ) as fast as 15 min post-protein removal ( Figure 4H ) . Trypsinization of IRBC abrogated the rosette-stimulating effect of IGFBP7 ( Figure 5A ) , suggesting the involvement of parasite-derived proteins expressed on the surface of IRBC . P . falciparum expressed different surface antigens including rosetting ligands such as the P . falciparum erythrocyte membrane proteins 1 ( PfEMP1 ) , STEVOR , and RIFIN ( Chan et al . , 2014 ) . The sensitivity to low level of trypsin treatment ( 10 µg/ml ) used here suggested that PfEMP1 , rather than STEVOR or RIFIN , was likely the adhesin involved ( Kyes et al . , 1999; Niang et al . , 2014 ) . We further validated this hypothesis with use of the P . falciparum SBP1-KO-CS2 line , in which the skeleton binding protein 1 ( SBP1 ) gene was knocked out . SBP1 transports PfEMP1 from the parasitophorous vacuole to the Maurer’s cleft for subsequent assembly and export to the surface of IRBC . Therefore , the SBP1-KO-CS2 line is unable to express PfEMP1 on the surface of the IRBC . However , knock out of this gene does not affect the surface expression of STEVOR and RIFIN ( Maier et al . , 2007; Chan et al . , 2016 ) . IGFBP7 increased the rosetting rate of CS2 wild-type parasite but had no effect on SBP1-KO-CS2 rosetting ( Figure 5B ) . Subsequently , we used two other clones of the P . falciparum NF54 line , NF54 VAR2CSA_WT clone and the mutant clone , NF54_T934D , whose PfEMP1 variant VAR2CSA is not exported onto the surface of the IRBCs ( Dorin-Semblat et al . , 2019 ) . The rosetting machinery of NF54_VAR2CSA_WT responded positively to the presence of IGFBP7 . On the other hand , the rosetting rates of NF54_T934D clone ( lacking PfEMP1 on the surface of IRBC ) were not significantly altered by IGFBP7 ( Figure 5C ) . Of note , IRBCs with surface expression of PfEMP1 variant VAR2CSA ( which include the CS2_WT and NF54_VAR2CSA_WT ) did not form many rosettes , which was in parallel with earlier report ( Rogerson et al . , 2000 ) . In addition , the rosettes formed by NF54_VAR2CSA_WT ( Figure 5C , inset i ) and NF54_T934D ( Figure 5C , inset ii ) were small . Expression of parasite-derived , IRBC-surface proteins such as rosetting ligands is sequential and parasite stage-specific . For example , PfEMP1 is the first rosetting ligand to be expressed on the surface of IRBC ( as early as late ring stage ) , followed by RIFIN , and finally STEVOR ( which are at the much more mature stages ) ( Kyes et al . , 2000; Lavazec et al . , 2007; Kaviratne et al . , 2002; Bachmann et al . , 2012; Niang et al . , 2009 ) . In other words , PfEMP1 is the only rosetting ligand available on the surface of late ring-IRBCs . We performed the IGFBP7-rosetting assessment on the late ring stages ( ~hour 16–26 ) of a laboratory-adapted clinical isolate ( nine replicates across three different cycles ) and found that IGFBP7 significantly increased the rates of rosette formation ( Figure 5D ) by the late ring stages ( Figure 5E ) . Taken together , these results suggest strongly that PfEMP1 is essential for IGFBP7-mediated rosetting in P . falciparum . On the other hand , we could not identify which P . vivax proteins were involved in IGFBP7-mediated rosettes as the P . vivax-IRBC membrane-associated proteins have yet to be fully characterized . IGFBP7 has a heparin binding domain . We hypothesized that it might be involved in the rosetting effect . When URBC treated with either heparinase I ( Figure 6A ) or heparinase III ( Figure 6B ) were mixed with purified P . falciparum- or P . vivax- IRBC , IGFBP7 did not induce rosetting . Heparinases also affected the size of rosettes ( numbers of URBC in a rosette ) ( Figure 6C , green arrows ) . Notably , addition of IGFBP7 induced autoagglutination-like clustering of IRBC ( which were not enzyme-treated ) ( Figure 6C , red arrow ) . IRBC-autoagglutination was absent in the control groups ( untreated and enzyme-treated groups without addition of IGFBP7 ) ( data not shown ) . Complement receptor 1 ( CR1 ) expressed on URBC is a receptor for PfEMP1 in P . falciparum rosetting ( Rowe et al . , 1997 ) . However , it did not play a significant role in IGFBP7-mediated rosetting for both parasite species . Anti-CR1 mAb reduced P . falciparum rosetting rates in the absence of IGFBP7 , confirming that this molecule is involved in the direct interaction between IRBC and URBC ( Figure 6D ) . However , anti-CR1 mAb had an insignificant effect on IGFBP7-induced rosetting in P . falciparum ( Figure 6D ) . For P . vivax , anti-CR1 antibody did not inhibit rosetting , with or without addition of IGFBP7 to the culture ( Figure 6E ) . Likewise , For P . falciparum and P . vivax ( Figure 6F and G ) , ABO blood groups did not play significant roles in IGFBP7-mediated rosetting , as well as rosetting mediated by other CSMT-derived rosette-stimulators ( Figure 6H and I ) . All the experiments described above were performed using 20% human serum-enriched medium . However , serum filtration with a 0 . 45 μm filter abolished IGFBP7-induced rosetting ( Figure 7A ) . These results indicated that other large-sized serum-derived protein aggregates or multimers might be needed for the IGFBP7-mediated rosetting effect . We hypothesized that von Willebrand factor ( VWF ) may be involved as VWF has been reported to absorb onto plasma-exposed surfaces ( Grinnell and Phan , 1983; Mannhalter , 1993 ) . Addition of anti-VWF antibody ( 25 µg/ml ) did not significantly alter the baseline rosetting rates ( Figure 7B ) . However , the presence of anti-VWF antibody prevented IGFBP7 from exerting its rosette-stimulatory effect . The specificity of rosette-inhibition by the antibodies was validated with experiments using antibody isotype controls at the same working concentration ( 25 µg/ml ) ( Figure 7—figure supplement 1A ) . When we reduced the medium’s serum enrichment to only 2% , IGFBP7 could not increase rosetting rates ( Figure 7C ) . The protein could only exert its rosette-stimulatory effect in 2% serum-enriched medium when VWF was added . Importantly , VWF by itself did not stimulate rosetting ( Figure 7C ) , disputing the possibility of this phenomenon as a non-specific adhesive effect of VWF and indicating that it is a co-factor of the IGFBP7-mediated rosetting . When the media’s serum enrichment was replaced with Albumax II ( a serum substitute ) , addition of IGFBP7 ( 100 ng/ml ) and VWF did not significantly increase rosetting rates across the VWF concentration range tested ( Figure 7D ) . This suggested the need for other serum-derived factors to mediate the IGFBP7 rosetting effect . We suspected that this second cofactor would either be needed in small quantity or present in high abundance in serum , so that even a 2% serum-supplied medium would be adequate to sustain IGFBP7-mediated rosetting . In addition , this co-factor should be able to interact with some of the players identified above ( PfEMP1 , HS , VWF ) to generate IGFBP7-mediated rosettes . One candidate was thrombospondin 1 ( TSP-1 ) as it is known to bind to PfEMP1 ( Cooke et al . , 1994; Baruch et al . , 1996 ) and VWF ( Pimanda et al . , 2004 ) . In serum-enriched medium , anti-TSP-1 antibody did not significantly alter baseline rosetting rates . Nevertheless , this antibody significantly blocked the IGFBP7-mediated rosette-stimulation ( Figure 7E ) . The specificity of rosette-inhibition by this antibody was validated with experiments using antibody isotype control ( Figure 7—figure supplement 1A ) . When TSP-1 was added to Albumax-supplemented medium , the rosetting rates were lower than those in serum-enriched medium ( Figure 7F ) . However , when added together ( IGFBP7 + VWF + TSP-1 ) to the Albumax-supplemented medium , significant rosette-stimulation was observed . A high-level TSP-1 ( 500 ng/ml ) did not induce significantly higher rosetting stimulation than the lower level TSP-1 ( 10 ng/ml ) . With the addition of VWF ( 2 IU/ml ) and TSP-1 ( 10 ng/ml ) to Albumax-supplemented medium , IGFBP7 stimulated rosette formation to an extent similar to that of serum-enriched medium . Importantly , without IGFBP7 , the presence of TSP-1 and VWF in Albumax-supplemented medium could not increase the rosetting rates , disputing the possibility of this event as a non-specific adhesive effect and reflecting their status as co-factors in IGFBP7-mediated rosetting . Lastly , we quantitated the amount of VWF needed to facilitate IGFBP7-mediated rosetting . When supplemented with IGFBP7 and TSP-1 in Albumax-supplemented medium , VWF as low as 0 . 125 IU/ml was sufficient to significantly increase rosetting rates , with an optimal increment attained at 0 . 5 IU/ml ( Figure 7G ) . In fact , the 20% serum-enriched media that we used for this study contained VWF higher than 0 . 125 IU/ml ( Figure 7—figure supplement 1B ) . P . falciparum IRBC-exposed peripheral monocytes secreted significantly more IGFBP7 than their unexposed counterparts or those exposed to URBC ( Figure 8A ) . THP-1 also secreted more IGFBP7 after IRBC exposure ( Figure 8B ) . Interestingly , parasitemia as low as 0 . 25% was sufficient to significantly stimulate THP-1 to secrete more IGFBP7 . Further increase in parasite density ( up to 16% parasitemia ) did not significantly increase IGFBP7 secretion any further . THP-1 cells were transduced with shRNA lentiviral vectors specific for IGFBBP7 and vectors specific for an unrelated protein , glycophorin C . Non-transduced cells served as wild types ( THP-1_WT ) . Via ELISA , the IGFBP7 production by THP-1 with decreased expression of IGFBP7 ( referred as IGFBP7-KD_THP-1 ) was significantly lower than those of THP-1_WT and THP-1 with decreased expression of glycophorin C ( referred to as GlyC-KD_THP-1 ) after URBC or IRBC stimulation for 18 h ( Figure 8—figure supplement 1 ) . The URBC- and IRBC- conditioned supernatants from IGFBP7-KD_THP-1 ( henceforth referred to as CSKD-U and CSKD-I , respectively ) were collected for rosetting assay . CSKD-U and CSKD-I did not significantly increase rosetting rates of the P . falciparum parasites ( Figure 8D ) , indicating that IGFBP7 is the main rosette-stimulating factor secreted by IRBC-stimulated THP-1 . To further validate this , we prepared the culture supernatants ( CS ) of THP-1_WT , IGFBP7-KD_THP-1 and GlyC-KD_THP-1 in the same the way as we prepared the CSMT and CSUT in earlier experiments where we exposed the cells to IRBC for a longer time ( three days ) and allowed the cells to grow for a longer time prior to CS harvest . IRBC addition stimulated the THP-1_WT and GlyC-KD_THP-1 to produce a higher level of IGFBP7 than their URBC-exposed counterparts , whereas levels of IGFBP7 remained low and were not significantly different between IRBC and URBC-stimulated IGFBP7-KD_THP-1 ( Figure 8—figure supplement 2A ) . Furthermore , addition of anti-IGFBP7 antibody confirmed that IGFBP7 is the major rosette-stimulating factor within the culture supernatant of THP-1_WT and GlyC-KD_THP-1 ( Figure 8—figure supplement 2B ) . Importantly , the significant difference between culture supernatant groups ‘THP-1_WT_URBC’ and ‘THP-1_WT_IRBC’ ( p=0 . 0018 ) , as well as ‘GlyC-KD_THP-1_URBC’ and ‘GlyC-KD_THP-1_IRBC’ ( p=0 . 0013 ) , but not between ‘IGFBP7-KD_THP-1_URBC’ and ‘IGFBP7-KD THP-1_IRBC’ ( p=0 . 6011 ) strongly suggests that IGFBP7 may be one of the key secreted products by the monocytic cells in response to parasite exposure , whereas other factors involved in rosette-stimulation may be secreted at baseline levels with or without the presence of the parasites ( Figure 8—figure supplement 2B ) . We hypothesized that IGFBP7-mediated rosetting could be a strategy used by the parasites to avoid phagocytosis . To test this , we performed a control experiment using Zymosan A ( a protein-carbohydrate complex prepared from yeast cell wall , commonly used in phagocytosis assays ) and showed that IGFBP7 by itself did not inhibit the phagocytic ability of THP-1 ( Figure 8E ) . Unexpectedly , incubation with IGFBP7 increased the phagocytic ability of THP-1 . We next tested the phagocytic activity of THP-1 and of human primary monocytes in the presence of IGFBP7-treated culture . As expected , the rosetting rates of the parasite increased after IGFBP7 exposure . ( Figure 8F ) . However , IRBC phagocytosis rates by both types of phagocytes were reduced significantly ( Figure 8G ) . Subsequently , we repeated this experiment with THP-1 using five different P . falciparum lines . All IGFBP7-incubated P . falciparum lines prior to THP-1 exposure formed more rosettes ( Figure 8H ) . They were significantly less phagocytosed than their non-IGFBP7-exposed counterparts ( Figure 8I ) . Individual phagocytes could engulf non-rosetting IRBC ( Figure 8J ) ; however , successful engulfment of a rosette was only observed when several phagocytes were recruited ( Figure 8K and L ) .
Rosetting is a common characteristic of late stage-IRBC in human malaria parasites , occurring frequently in P . falciparum and P . vivax ( Lee et al . , 2014 ) . It has been proposed to provide a survival advantage for the parasites ( Moll et al . , 2015 ) . Earlier studies have shown that rosetting occurs between the direct interactions of the parasite-derived ligands on the IRBC ( i . e . PfEMP1 , RIFIN , and STEVOR proteins for P . falciparum ) with various receptors on the URBC ( Lee et al . , 2014; Niang et al . , 2014; Rowe et al . , 1997; Barragan et al . , 2000; Cserti and Dzik , 2007; Chen et al . , 1998; Goel et al . , 2015 ) . Here we demonstrated the existence of a different type of rosetting , which we have called ‘type II rosetting’ that does not result from the direct interaction of IRBC with URBC . It was observed in all the P . falciparum and P . vivax isolates tested . This type II rosetting differs from the classical type I rosetting as it requires bridging by soluble mediators: IGFBP7 , VWF , and TSP-1 between a rosetting ligand on IRBC and HS expressed by URBC . The fast rosette-stimulating effect by the protein and its fast reversion after the protein removal from the culture suggest that IGFBP7 does not mediate rosetting via irreversible binding to neither the rosetting receptor nor ligand . Instead , it is more likely to be mediated by weaker forces . It also suggests that these soluble mediators need to be present at a minimum concentration for the rosettes to occur . We have shown that PfEMP1 is likely the principal P . falciparum rosetting ligand in this IGFBP7-mediated type II rosetting via usage of trypsin treatments , genetically modified P . falciparum clones that cannot surface-express PfEMP1 , and the late ring stages of P . falciparum , the stage of maturation that manages to surface-express only one rosetting ligand , PfEMP1 . However , we cannot fully dismiss the involvement of other rosetting ligands such as STEVOR or RIFIN as these proteins are encoded by multigene families and there is a lack of tools to assess and evaluate the implication of these proteins thoroughly . To our surprise , P . vivax IRBC also interacted with IGFBP7 similarly . There are no PfEMP1 orthologues in P . vivax . Based on the rosetting trend of P . vivax post-trypsin treatments , we postulate that P . vivax has multiple rosetting ligands with different trypsin sensitivities , and the one required by IGFBP7 is highly sensitive to trypsin . IGFBP7 requires the HS moieties on URBC to exert their rosette-stimulatory effect . Interestingly , removal of HS from the surface of URBC by heparinase caused the clumping of untreated IRBC ( which harboured both rosetting ligands and receptors ) when supplied with IGFBP7 . Importantly , without enzymatic interference , the presence of this protein does not induce non-specific binding of URBC to each other , or autoagglutination-like clumping of IRBC . We hypothesized that the IGFBP7-mediated binding occurs preferably between the URBC and IRBC under normal circumstances possibly as a result of electrostatic differences between the URBC and IRBC ( Tokumasu et al . , 2012 ) . IGFBP7 requires other serum factors , namely VWF and TSP-1 to exert its rosette-stimulating effect . For healthy individuals , the VWF levels in the serum range from 0 . 48 to 1 . 24 IU/ml ( median 0 . 84 IU/ml ) , whereas individuals with underlying pathological conditions have much higher levels of serum VWF ( Terpos et al . , 2013; Kastritis et al . , 2016 ) . Serum TSP-1 levels in healthy individuals vary greatly ( 0–12060 ng/ml ) ( Liu et al . , 2015a; Rouanne et al . , 2016 ) . We found that under serum-free conditions ( Albumax- supplemented medium ) , concentrations of VWF as low as 0 . 5 IU/ml and of TSP-1 at 10 ng/ml were enough to optimally facilitate IGFBP7-mediated rosette-stimulation at IGFBP7 of 100 ng/ml ( the minimum concentration needed to optimally stimulate type II rosetting ) . The rosette-stimulation by the presence of these three proteins was comparable to that of 20% serum-enriched medium supplied with IGFBP7 , reinforcing that IGFBP7 is the limiting factor for the rosette-stimulation . Based on the data presented here , we propose the following mechanism of interactions for type II rosetting ( Figure 9 ) . IGFBP7 binds to HS on URBC; the interaction between IGFBP7 and cell surface HS has been demonstrated and well-characterized ( Sato et al . , 1999 ) . IGFBP7 has also been shown to bind to the D4-CK region of VWF ( van Breevoort et al . , 2012; Lenting et al . , 2012; Bryckaert et al . , 2015 ) . Although heparin is found to interact with the A1 region of VWF ( Bryckaert et al . , 2015 ) , it is likely that HS does not interact directly with VWF ( Denis et al . , 1993 ) as we did not observe any rosetting when VWF was added alone in the absence of IGFBP7 . Therefore , the HS on URBC interacts with IGFBP7 , which also interacts with VWF . In turn , VWF interacts with PfEMP1 on IRBC via TSP-1 ( Cooke et al . , 1994; Bryckaert et al . , 2015 ) . While the extracellular domain of PfEMP1 that binds to TSP-1 has yet to be identified , it should be noted that TSP1 has been commonly associated with PfEMP1 ( Cooke et al . , 1994; Janes et al . , 2011 ) . The pervasive role of TSP-1 may explain why a wide range of clinical isolates and laboratory-adapted parasite lines were capable of responding positively to IGFBP7 addition . Induction of type II rosetting in P . falciparum and P . vivax is not attributed solely to IGFBP7 . Previously , it has been shown that CFD in the serum can stimulate rosetting ( Luginbühl et al . , 2007 ) . This molecule was also identified in our proteomic study . Experiments with anti-CFD antibodies showed that CFD could also induce type II rosetting , but to a lesser extent . The knock down of IGFBP7 expression in THP-1 demonstrated that IGFBP7 is a major monocyte-derived rosette-stimulating factor . Culture supernatant from IGFBP7-KD_THP-1 collected after 18 h of parasite exposure could not induce rosette-stimulation . Other rosette-stimulating factors may be secreted by the cells much later . Interestingly , IGFBP7 and VWF are components in Weibel-Palade bodies , the storage granules of endothelial cells ( van Breevoort et al . , 2012 ) . Future work should characterize the effect of IRBC on the secretion of IGFBP7 by endothelial cells . Of note , the reported physiologic and pathological serum concentrations of IGFBP7 vary greatly , where most of the reported normal serum IGFBP7 concentrations fall below 50 ng/ml , but the serum IGFBP7 levels in different pathological conditions ( e . g . various vascular disorders and cancers ) are higher ( as high as ~1000 ng/ml ) ( Kutsukake et al . , 2008; Shersher et al . , 2011; Liu et al . , 2015b; Shaver et al . , 2016; Barroso et al . , 2016 ) . Therefore , the working concentration of IGFBP7 in this study ( 100 ng/ml ) is still within the pathophysiologic range in clinical settings . It would be interesting to compare the serum/plasma IGFBP7 levels of uncomplicated and severe malaria patients from the same area to understand better the role of IGFBP7 in malaria pathogenesis . The importance of monocytes/macrophages in eliminating Plasmodium during its course of infection has been reported ( Groux and Gysin , 1990; Theander , 1992 ) . Peripheral monocytes ( Turrini et al . , 1992; Muniz-Junqueira and Tosta , 2009 ) , as well as the tissue-resident macrophages ( Tosta et al . , 1983; Sponaas et al . , 2009 ) , have been shown to engulf IRBC readily . To survive , parasites must counter or avoid this host’s immune responses . The ability to perceive a phagocyte’s secreted IGFBP7 as a signal of ‘approaching threats’ and to respond by rosetting may provide survival advantage to the parasites . In conclusion , the host-derived IGFBP7 is used as an ‘incoming phagocyte signal’ by the IRBC , and the IRBC in turn use this protein , along with two serum factors , VWF and TSP-1 , to stimulate rosette formation , which acts as an immune-evasion strategy by hampering phagocytosis of the IRBC . It is hoped that future clinical studies will investigate associations between IGFBP7 and malaria pathogenesis and immunity .
Malaria-infected samples were collected in Shoklo Malaria Research Unit ( SMRU ) under approved ethics: OXTREC 04–10 ( University of Oxford , UK ) ; TMEC 09–082 ( Ethics Committee , Faculty of Tropical Medicine , Mahidol University , Thailand ) . Human monocytic THP-1 cell line ( Source: ATCC ) was used in this study . The cells were tested Mycoplasma-free using the MycoAlert Plus Mycoplasma Detection Kit ( Lonza ) . Clinical isolates ( uncomplicated malaria cases ) from SMRU were recruited . Blood ( volume: 3 ml ) was collected using a BD Vacutainer with lithium heparin anticoagulant . Blood groups were determined with TransClone Anti-A and Anti-B antibodies . Blood samples were centrifuged at 1500 g for 5 min . Plasma was removed , and the buffy coat was carefully collected . A CF11-packed column was used to filter the remaining leukocytes . The parasites were matured in vitro under culture conditions of 5% haematocrit using 20% human homologous serum-enriched RPMI 1640 medium for P . falciparum , and McCoy’s 5A for P . vivax , under gas conditions of 4% CO2 and 3% O2 . Leukocytes and red blood cells ( RBC ) from clinical samples were isolated and divided into two groups . One group consisted of only RBC . In another group , RBC and leukocytes at a physiologic ratio of 500:1 were matured in vitro , prior to rosetting assay ( Lee et al . , 2014; Lee et al . , 2013; Chotivanich et al . , 1998 ) . There are different wet mount-based techniques that can be used for rosetting assay ( Lee et al . , 2014; Lee et al . , 2013; Chotivanich et al . , 1998; Udomsangpetch et al . , 1992; Udomsangpetch et al . , 1989; Treutiger et al . , 1998; Ribacke et al . , 2013; Adams et al . , 2014 ) . We compared three commonly used techniques and validated that they can be used interchangeably ( Supplementary file 7 ) , as elaborated in the following sections . Rosetting assay was conducted when 70% of the parasite population reached late stages ( late trophozoites and schizonts ) unless stated otherwise . The parasite culture suspension was stained subvitally with Giemsa ( 5% stain working concentration ) for 20 min . Subsequently , 7 . 6 µl stained suspension was pipetted onto a clean glass slide , immediately covered with a 22 × 32 mm glass cover slip . The wet mount was examined with light microscope using 1000× magnification . Rosetting rate ( percentage of rosetting IRBCs ) was defined as the percentage of IRBC ( over 200 recruited IRBC ) that form rosettes . Experiments were conducted in a blinded manner . There are different microscopy-based techniques that can be used for rosetting assay , namely the unstained , Giemsa-stained , and the fluorescent dye acridine orange ( AO ) -stained wet mounts , but none of them have been thoroughly validated and compared . When the majority of the P . falciparum culture population ( laboratory-adapted lines 3D7 , CS2-WT , FVT201 , MKK183 , WPP3065 ) reached late stages , the culture suspension was used for this experiment . For each parasite line , the culture suspension was divided into three parts . One aliquot was stained with 5% Giemsa subvitally for 15 min prior to wet mount preparation for rosetting assay . Another was stained with acridine orange ( working concentration 2 µg/ml ) for 15 min prior to wet mount for rosetting assay . The third aliquot was used for unstained wet mount preparation prior to rosetting assay . Rosetting rates were determined and compared . The late stage-IRBCs were purified with MACS-LD columns . Only the yields with IRBC purity of at least 90% were used . The purified IRBCs were divided into two groups . One was mixed with URBCs to make a 3% parasitemia packed cell mixture . The other group was used as a purified IRBC group . Packed URBCs from two healthy individuals were used as controls . Each of these groups was further divided into two parts , where one was exposed to IGFBP7 ( 100 ng/ml ) and the other acted as an IGFBP7-free control . All groups were suspended with culture medium , and incubated for 1 h at in vitro cultivation conditions prior to rosetting assay using the Giemsa-wet mount , acridine orange-wet mount , and unstained wet mount methods described in the previous paragraph . In our hands , these three techniques yielded comparable rosetting rates with or without IGFBP7 ( Supplementary file 7 ) . Importantly , in all techniques applied , IGFBP7 did not exert a clumping effect on groups consisting of only URBC and IRBC ( Figure 4—figure supplement 1 ) , demonstrating the IRBC-URBC specificity of the IGFBP7 interaction . Of note , the formation of some IRBC aggregates when enriched IRBC are cultured has been described previously by Adams et al . ( 2014 ) . However , addition of IGFBP7 did not aggravate the IRBC aggregate formation or enlarge the size of the aggregates ( data not sown ) . Thus , we have validated that the currently available microscopy-based rosetting assays have similar reliability and can be used interchangeably . Three Percoll gradients were prepared from the isotonic Percoll ( 9 parts Percoll stock + 1 part 10× PBS ) , namely 81% Percoll , 68% Percoll , and 55% Percoll . A 15 ml conical centrifuge tube was layered with 3 ml of 81% Percoll solution , followed by 3 ml of 68% Percoll solution . Blood from healthy donors was centrifuged for 5 min at 1500 g . Subsequently , three-quarters of the plasma supernatant were removed , followed by careful collection of the leukocyte-rich buffy coat layer . Uptake of RBC must be avoided as much as possible . The leukocytes were suspended with 3 ml of 55% Percoll , and carefully transferred onto the Percoll layered column prepared . The gradient tube was centrifuged at 1500 g for 20 min . The upper two-thirds of the top layer was removed . The remaining one-third ( the ‘55–68’ interface zone , i . e . monocyte-rich PBMCs ) was collected . After that , the upper two-thirds of the second layer was removed , and the ‘68–81’ interface ( neutrophil-rich PMNs ) was collected in another conical centrifuge tube . The separated monocytes and neutrophils were washed with RPMI 1640 medium . After that , trypan blue exclusion examination and Giemsa-stained blood smear examination were performed to evaluate viability , cell numbers , and purity of cell population harvested . The cells were suspended in RPMI 1640 and incubated in petri dishes for 3 h at 37°C . For the monocyte group , cells that adhered to the petri dish ( monocytes ) were retained for experiments whereas the non-adhered cells were removed from the petri dishes . P . falciparum ( three laboratory-adapted lines: 3D7 , FVT402 , FVT201 , and one clinical isolate RDM00036 ) were cultured ( 5% haematocrit , 3% parasitemia ) . Culture suspensions were incubated with or without neutrophils ( RBC: neutrophil ratio 1000: 1 ) or monocyte ( RBC: monocyte ratio 10 , 000: 1 ) from individual donors . Rosetting assay was performed afterwards . In a separate experiment , monocytes from two healthy donors were purified via Ficoll density gradient concentration method , followed by sorting of CD14+ microbeads . The impacts of purified CD14+ monocytes and CD14− peripheral blood mononuclear cells ( PBMC ) , as well as the human monocytic THP-1 cell line on rosetting rates of P . falciparum lines were assessed using laboratory-adapted P . falciparum lines ( FVT402 , 3D7 , and MKK183 ) . For each parasite line , three batches of cultures ( thawed from vials that were cryopreserved at different times ) were used for three experiment replicates . Human monocytic THP-1 cell line ( Mycoplasma-free ) was cultured [10% FBS -enriched RPMI 1640] and expanded with two methods; one being cultivated with stringent control of cell density below 106 cells/ ml as undifferentiated THP-1 ( UT ) cells . The other was allowed to replicate until the cell population reached 6 × 106 cells/ ml for differentiation into macrophage-like THP-1 ( MT ) cells . Supernatant ( 2 ml ) collected from P . falciparum culture ( rich with parasite antigens ) was added to the second group , along with interferon gamma ( IFNγ ) ( final concentration 50 ng/ml ) . Three days later , the culture medium was discarded , then 2 . 5 × 105 cells were transferred into each well of a 48-well flat bottom culture plate , whereas the remaining cells continued to be cultured in the flask . All cell cultures were replenished with fresh culture medium ( 5% FBS-enriched RPMI 1640 ) without addition of P . falciparum culture supernatant and IFNγ . Two days later , cell counts were performed . The culture supernatant ( CS ) from the 48-well plate ( 1 ml for each well ) was collected as CSMT for subsequent experiments . The preparation was repeated with the UT cells to collect CSUT . The attached , differentiated MT cells in the culture flask were harvested by removing the culture medium , followed by addition of pre-chilled 1× PBS into the culture flask and incubation on ice for 10 min . UT and MT of different cell numbers were tested in the rosetting assay in triplicate for each parasite line . CSMT and CSUT ( used in dilution equivalent to that of 1 × 106 cells ) , and IFNγ ( 50 ng/ml ) were also tested in the rosetting assay . CSMT was fractionated into lipophilic ( lipid ) and aqueous ( aq ) compartments using Folch’s chloroform-methanol extraction method ( Folch et al . , 1957 ) . Briefly , chloroform-methanol extraction mixture ( 2:1 ratio ) was prepared . Washing liquid consisted of chloroform , methanol , and distilled water in a ratio of 3:48:47 was prepared . CSMT ( 1 ml , from culture with cell density of 106 cells/ml ) was mixed with 19 ml extraction mixture . The mixed liquid was washed with 4 ml of distilled water and allowed to settle for a few minutes . Two phases of liquid formed from this . The upper portion ( around 40% of the total volume ) being the aq fraction , and the lower part being the lipid fraction . The aq fraction was collected separately . The remaining liquid was washed gently with washing mixture three times to remove the interphase . After that , the lipid phase was collected . The collected aq and lipid fractions were dried with a vacuum concentrator . After that , the pellet was suspended with 500 µl distilled water . Vortexing was applied to facilitate solubilization of the lipid pellet . These fractions were tested with rosetting assay . The aq fraction was further subjected to size-based fractionation using Vivaspin20 twin PES membrane ( 30 , 000 MWCO ) concentrator and tested with rosetting assay . A separate experiment was conducted with laboratory-adapted P . falciparum lines ( 3D7 , FVT402 , FVT201 , MKK183 ) to compare the rosette-stimulating effect of the aq ≤30 kDa fraction and the aq ≤30 kDa fraction that was heated for 1 h at 56°C . Subsequently , the aq fraction ( ≤30 kDa ) was digested for mass spectrometry analysis using an Orbitrap Fusion mass spectrometer . Samples were in-solution digested . Initial denaturation was done with 8M urea in 50 mM Tris-HCl pH 8 . 5 . Following denaturation , proteins were reduced in 25 mM Tris- ( 2-carboxyethyl ) phosphine ( TCEP ) , alkylated with 55 mM chloroacetamide ( CAA ) , and further diluted with 100 mM triethylammonium bicarbonate ( TEAB ) to achieve <1M urea concentration . Two-step enzyme digestion with lysyl endopeptidase ( LysC ) and trypsin was performed for 4 h ( 1:100 –enzyme/protein ratio ) and 18 h ( 1:100 ) , respectively . After acidification with 1% trifluoroacetic acid ( TFA ) , desalting was done using Sep-Pak C-18 columns . The organic phase was evaporated in the vacuum centrifuge . For high pH reverse phase initial separation sample was re-suspended in 10 mM ammonium formate/5% acetonitrile . Two-hundred minutes continuous gradient separation ( Solvent A: 10 mM ammonium formate pH10 . 5/Solvent B: 10 mM ammonium formate pH10 . 5/90% acetonitrile ) was performed on an ÄKTA Micro system using Gemini 5 u/C-18/110A , 150 mm × 1 mm column . Collected fractions were combined into 14 fractions , evaporated and used for mass spectrometry analysis . Mass spectrometry analysis was performed on an Orbitrap Fusion mass spectrometer coupled to nano-ultra-high-performance liquid chromatography ( UHPLC ) Easy nano liquid chromatography ( nLC 1000 system ) . Fractions were injected and separated on in-house prepared ( C-18 ReproSil Pur Basic beads 2 . 5 um ) fused silica emitter column 20 cm × 75 µm in 75 min gradient ( solvent A: 0 . 1% formic acid; solvent B: 0 . 1% formic acid/99 . 9% acetonitrile ) in data dependent mode using Orbitrap ( OT ) and Ion trap ( IT ) detectors simultaneously ( speed mode −3 s cycle ) with ion targets and resolution ( OT-MS 2xE5 , resolution 60K , OT-MS/MS 3 . 5E4 , resolution 15 k; IT-MS/MS 2E4 , Normal scan ) . Peak lists were generated with Proteome Discoverer 1 . 4 software and searches were done with Mascot 2 . 5 against forward and decoy Human-HHV4 Uniprot database ( 88 , 559 entries ) with the following parameters: precursor mass tolerance [mass spectrum ( MS ) ] 30 ppm , OT-MS/MS 0 . 06 Da , IT-MS/MS 0 . 6 Da; two miss cleavages; static modifications: carbamidomethyl ( C ) , variable modifications: oxidation ( M ) , deamidated ( NQ ) , acetyl N-terminal protein . Forward/decoy searches were used for false discovery rate ( FDR ) estimation ( FDR 1% ) . Peak lists were generated . Following data review , coupled with critical information ( i . e . the protein’s subcellular location and cellular functions ) from UNIPROT ( UniProt Consortium , 2015 ) , candidates were shortlisted for further validation . The parasite culture suspensions were incubated with CSMT and antibodies against the shortlisted proteins ( Supplementary file 3 ) and tested in rosetting assay at a final concentration of 25 µg/ml . The parasite culture suspension was divided into three groups , one served as control , the second group was added with recombinant human IGFBP7 ( final concentration 100 ng/ml ) , and the third group was mixed with CSMT ( diluted with culture medium to make the CSMT ‘working concentration’ equivalent to that of 1 × 106 cells ) . After 1 h of incubation under in vitro cultivation conditions , rosetting assay was conducted . Separately , a portion of the IGFBP7 suspension was heat-denatured at 95°C for 1 h , prior to use in rosetting assay . Parasite cultures were incubated with IGFBP7 ( working concentrations 0–25 , 000 ng/ml ) prior to rosetting assay . Time course experiments were performed with laboratory-adapted P . falciparum lines ( 3D7 , MKK183 , FVT402 , FVT201 , WPP3065 ) incubated with IGFBP7 ( 100 ng/ml ) . Rosetting assay was conducted after 5 min , using 7 µl of the suspension . The remaining suspension was kept back into the incubator . Rosetting assay was repeated at 5 min-intervals until 1 h-post-IGFBP7 exposure . Reversibility of IGFBP7-mediated rosette-stimulation was also tested . Parasite lines ( 3D7 , MKK183 , FVT402 , FVT201 , WPP3065 ) were incubated with IGFBP7 ( 100 ng/ml ) and rosetting assay was conducted with 7 µl of the suspension . The remaining suspension was centrifuged at 1500 g for 5 min . Supernatant was removed , and the pellet was washed three times with culture medium , followed by re-suspension with culture medium . Five minutes later , 7 µl of the suspension was taken for rosetting assay , subsequently repeated at 5 min-intervals up to 1 h post-IGFBP7 removal . Magnetic activated cell sorter ( MACS ) -sorted late stage-IRBC ( purity ≥95% ) were trypsinized at different working concentrations . The first group was mixed with enzyme trypsin ( final trypsin concentration of 10 µg/ml ) , and the second was mixed with enzyme trypsin ( final trypsin concentration of 1 mg/ml ) . The third served as an untreated control . The cells were incubated at 37°C for 30 min . After that , the cells were washed with serum-enriched medium three times . Each group was incubated with or without IGFBP7 ( 100 ng/ml ) prior to rosetting assay . P . falciparum line CS2 deficient of SBP1 [ ( SBP1-KO-CS2 ) , which lacks P . falciparum erythrocyte membrane protein 1 ( PFEMP1 ) on its IRBC surface] and its wild type ( CS2-WT ) counterpart were cultured as described ( Maier et al . , 2007; Chan et al . , 2016 ) . Their rosetting rates post-incubation with IGFBP7 at different concentrations were determined . P . falciparum clones NF54 VAR2CSA_WT and NF54_T934D ( cannot express PfEMP1 variant VAR2CSA on IRBC surface [Dorin-Semblat et al . , 2019] ) were cultivated . Experiments were conducted when parasite population reached late stages . For each parasite line , two conditions were applied; one was incubated with IGFBP7 ( 100 ng/ml ) , whereas the other acted as a IGFBP7-free control . Rosetting assay was conducted afterwards . Five replicates were conducted for each experiment setting . In a separate experiment , a laboratory-adapted clinical isolate from Thai-Burmese border ( NHP1106 ) was cultivated and staging of parasites was tightly synchronized . The experiment was conducted when the parasite population reached late rings ( ~hour 16–26 ) . Two settings were prepared; one was incubated with IGFBP7 ( 100 ng/ml ) and the other acted as a IGFBP7-free control . One hour of incubation under in vitro cultivation conditions was done prior to rosetting assay . Nine replicates ( across three cycles of cultivations ) were conducted . The role of heparan sulfate ( HS ) in IGFBP7-mediated rosetting was also tested . URBC ( blood group O ) were treated with heparinase I ( final working concentration of 25 µg/ml ) or heparinase III ( final working concentration of 25 µg/ml ) , with the untreated URBC served as control . The enzyme-erythrocyte mixtures and the untreated controls were incubated at 37°C for 30 min . After that , the suspension was centrifuged to remove supernatant . The treated erythrocytes were washed with 20% human serum enriched-culture medium three times . Subsequently , cells were suspended in plain culture medium . The prepared cells were kept at 4°C until use within 1 week . Late stage-IRBC were concentrated with MACS . The IRBC ( IRBC purity: 90–96% ) were divided into three groups , each mixed with the control , heparinase I-treated and heparinase III-treated URBC , respectively . The roles of complement receptor 1 ( CR1/CD35 ) and A/B blood antigens were also investigated . Recruited isolates were matured in vitro , subsequently divided into four groups . One group served as the control , another group was added with rhIGFBP7 ( final concentration 100 ng/ml ) . The third group was added with mouse anti-human CR1 ( CD35 ) IgG1 ( final concentration 25 µg/ml ) , whereas the fourth group was added with rhIGFP7 ( final concentration 100 ng/ml ) and mouse anti-human CR1 IgG1 ( final concentration 25 µg/ml ) . Rosetting assay was conducted after the incubation . Separately , the late stage-IRBCs were sorted with MACS . The sorted cells were divided into four groups , each to be mixed with URBCs of A , B , O , and AB groups , respectively . Each of the cell mixture groups was further divided into two groups , where rhIGFBP7 ( final concentration 100 ng/ml ) was added into one group and the other group served as control . Culture media enriched with 20% AB serum were used . The experiment was repeated with CSMT replacing rhIGFBP7 . An aliquot of human serum used for culture medium preparation was filtered with cellulose acetate syringe filter ( pore size 0 . 45 µm ) . The filtered fraction was used to prepare 20% filtered serum-enriched RPMI1640 medium . Packed erythrocytes from cultures ( P . falciparum laboratory-adapted lines: 3D7 , CS2-WT , FVT201 , MKK183 , WPP3065 ) were divided into two groups . The first group was suspended with 20% complete human serum-enriched RPMI 1640 ( denoted as ‘human serum’ group ) . The second group was suspended with the 20% filtered human AB serum-enriched medium ( denoted as ‘filtered human serum’ group ) . Culture was further incubated with or without IGFBP7 ( 100 ng/ml ) before rosetting assessment . Culture suspension of the laboratory-adapted P . falciparum lines ( 3D7 , MKK183 , NHP1106 , WPP3065 , WPP2803 , NHP4770 , FVT201 , FVT402 ) was centrifuged , and the packed cells were divided into groups: IGFBP7-free , anti-VWF , IGFBP7 , and IGFBP7 + anti-VWF groups . Rabbit anti-human VWF polyclonal IgG was used at a working concentration of 25 µg/ml , whereas IGFBP7 at 100 ng/ml was applied . Rosetting assay was done after incubation . In another experiment , the packed cells of parasite cultures ( 3D7 , MKK183 , NHP1106 , WPP3065 , WPP2803 , NHP4770 , FVT201 , FVT402 ) were divided into two parts , one was suspended with 20% serum-enriched RPMI1640 whereas the other group was suspended with 2% serum-enriched RPMI 1640 . Each group was further divided into four categories , that is control , IGFBP7 , VWF , and IGFBP7 + VWF . The working concentration of IGFBP7 was 100 ng/ml . For rhVWF ( referred to as VWF ) , final concentration of 1 IU/ml was used . Rosetting assay was conducted after incubation . In a separate experiment , the packed cells of cultures ( 3D7 , MKK183 , NHP1106 , WPP3065 , NHP4770 , FVT201 , FVT402 ) were suspended with 0 . 25% Albumax II ( Alb ) -enriched RPMI1640 , and divided into seven groups , each incubated with different concentrations of VWF ( 0 , 0 . 06 , 0 . 125 , 0 . 25 , 0 . 5 , 1 . 0 , 2 . 0 IU/ml ) prior to rosetting assay . The antibody blocking experiment using anti-VWF was repeated using mouse anti-human TSP-1 IgG2B in place of the anti-VWF antibody . Subsequently , experiments were conducted using rhTSP-1 ( referred as TSP-1 ) . The parasite culture packed cells were washed with plain RPMI 1640 medium twice . Each parasite line was divided into 12 groups . Ten groups were suspended with 0 . 25% Albumax-enriched medium ( referred to as ‘Alb’ in this experiment ) and the remaining two groups were suspended with 20% serum-enriched medium ( referred to as ‘20% serum’ in this experiment ) . The groups were as follows: IGFBP7-free Alb ( control ) , Alb + IGFBP7 ( 100 ng/ml ) , Alb + VWF ( 2IU/ml ) , Alb + 10 ng/1 TSP-110 ( henceforth referred to as TSP-110 ) , Alb + TSP-110 + IGFBP7 , Alb + TSP-110 + IGFBP7 + VWF , Alb + 500 ng/ml TSP-1 ( referred to as TSP-1500 ) , Alb + TSP-1500 + IGFBP7 , Alb + TSP-1500 + IGFBP7 + VWF , Alb + TSP-1500 + VWF , 20% serum , 20% serum + IGFBP7 . The working concentrations of IGFBP7 and VWF used were 100 ng/ml and 2 IU/ml , respectively . Rosetting assay was conducted after incubation . To quantitate VWF needed in IGFBP7-mediated rosetting , the parasite culture packed cells ( 3D7 , MKK183 , NHP1106 , WPP3065 , NHP4770 , FVT201 , FVT402 ) were washed with plain RPMI 1640 medium twice . Subsequently , the cells were suspended with 0 . 25% Albumax-RPMI . Each isolate was further divided into seven categories , each added with different concentrations of VWF ( 0 , 0 . 125 , 0 . 5 , 2 . 0 IU/ml ) . All these groups were given IGFBP7 ( working concentrations 100 ng/ml ) and TSP-1 ( 10 ng/ml ) . Rosetting assay was conducted after incubation . In a separate experiment , the levels of IGFBP7 of media used in the experiments ( 20% human serum-enriched RPMI media , Albumax-enriched RPMI media and plain RPMI media ) were quantitated using an ELISA kit , following instructions provided by the kit’s manufacturer . Peripheral monocytes ( CD14+ ) were purified from blood collected from five healthy donors via the Ficoll concentration method , followed by CD14+ bead purification . The purified cells were suspended in 10% FBS-enriched RPMI 1640 medium . Three wells of 96-well flat bottom microplate were allocated to cells harvested from each donor , where 1 × 105 cells were seeded into each well . One well served as plain control , whereas the other well was incubated with URBC , and the third one was incubated with the purified P . falciparum 3D7 IRBCs ( monocyte: IRBC ratio = 1:1000 ) . The cells were incubated at in vitro cultivation for 24 h . Subsequently , supernatants of the cultures were collected separately . During supernatant collection , care was taken to minimize uptake of sedimented cells ( RBCs , lysed cell products , hemoglobin may interfere with ELISA ) . Human IGFBP7 DuoSet ELISA kit was used to measure the IGFBP7 level in the supernatant of each experiment group using the manufacturer’s protocol . Measurements were done with microplate reader Tecan i-Control ( Tecan ) . The steps were repeated on THP-1 , with slight modifications , where five laboratory-adapted P . falciparum lines ( 3D7 , CS2-WT , FVT201 , MKK183 , WPP3065 ) were recruited . The mature stage-IRBCs were purified , and these purified IRBCs were then added with URBCs to make cell mixtures of parasitemia 16% , 8% , 4% , 2% , 1%–0 . 5% and 0 . 25% . Packed cells of only URBCs ( 0% parasitemia ) was used as control . The cells were added into the Lab-Tek 8-chamber-slides that were already seeded with respective cell lines ( 1 × 105 cells per well ) , making cellular suspension of 1 . 5% hematocrit . The cell mixtures were incubated for 24 h under in vitro cultivation conditions . Subsequently , supernatant was collected for ELISA analysis . During supernatant collection , care was taken to minimize uptake of sedimented cells ( RBCs , lysed cell products , hemoglobin etc . may interfere with ELISA ) . ELISA was conducted on the supernatant collected . THP-1 cells were thawed and cultured with RPMI1640 medium enriched with 10% FBS . A 96-well plate was used . Each recruited well was seeded with 1 × 104 cells . For each well , 110 µl of medium and hexadimethrine bromide ( final concentration 8 µg/ml; to enhance transduction ) were added . Lentiviral transduction particles to knockdown expression IGFBP7 ( hPGK-Puro_CMV-tGFP; SHCLNV-NM_001553 ) were added ( MOI 3 ) based on formulas provided in the kit’s user guide . On the following day , the media containing lentiviral particles were removed , and replaced with fresh medium . The next day , the transduced cells were cultivated with puromycin-added medium ( working concentration 3 µg/ml ) for selection . A small aliquot of the cells was examined with an epifluorescence microscope to check the GFP expression , which could be used for cell sorting ( Supplementary files 8–10 ) . The cells were used as ‘IGFBP7-knockdown ( KD ) THP-1’ in subsequent experiments . Late stage-IRBCs ( P . falciparum lines 3D7 , CS2-WT , FVT201 , MKK183 , WPP3065 ) were purified . The WT- and IGFBP7-KD THP-1 were used . For each cell type , two groups ( each group contains five sets , each well contained 1 × 104 cells ) were prepared . One was added with the purified IRBCs ( THP-1 to IRBC ratio of 1: 1000 ) and the other was added with URBCs from five healthy donors ( control ) . RPMI enriched with 1% serum ( to keep the viability of cells long enough for the experiment while minimizing the confounding effect on the protein quantification by the IGFBP7-KD cells ) was used . The cells were incubated for 18 h at in vitro cultivation conditions . The supernatant of the cells was collected . Care must be taken to avoid uptake of cell/cell debris . The supernatant was used for IGFBP7 quantitation using ELISA and subsequent experiments described below . The parasite culture packed cells ( P . falciparum lines 3D7 , CS2-WT , FVT201 , MKK183 , WPP3065 ) were divided into four groups . The first well was exposed to 1× PBS ( negative control ) , the second group was exposed to 100 ng/ml IGFBP7 ( positive control ) , the third group was added with similar volume of culture supernatant collected from the IGFBP7-KD-THP-1 exposed to URBCs ( CSKD-U ) , and the fourth group was added with culture supernatant collected from the IGFBP7-KD-THP-1 exposed to IRBCs ( CSKD-I ) . The suspension was topped up with 20% serum-enriched medium and incubated for 1 h prior to rosetting assay . The experiment was repeated with a control knockdown ( knockdown of Glycophorin C [Gly C] , a gene [cytogenetic location 2q14 . 3] that is different from IGFBP7 gene [cytogenetic location 4q12] and with low expression levels in monocytes ) and shRNA lentiviral vector , with slight modifications , where the approach of collecting CSMT/CSUT as described earlier was used . Quantification of IGFBP7 secretion was done . Besides , the culture supernatant was also used in rosetting assessment with laboratory-adapted P . falciparum line 3D7 coupled with use of anti-IGFBP7 antibody . THP-1 was cultivated and expanded into three batches . For each batch of culture , 1 × 106 cells were incubated with IGFBP7 ( working concentration 100 ng/ml ) for 1 h under in vitro cultivation conditions . Another set of cells acted as IGFBP7-free control . Subsequently , Zymosan A ( working concentration 10 µg/ml ) was added to both sets of cells and incubated for another hour under in vitro cultivation conditions . Supravital staining with Giemsa was done for 15 min following this . Using a wet mount technique , the percentage of THP-1 cells which has engulfed Zymosan A was determined as phagocytosis rate by recruiting 1000 THP-1 cells . The experiment was repeated with the other two batches of THP-1 culture . And all the steps were repeated another two times using THP-1 cultures thawed at different time points . P . falciparum lines ( 3D7 , MKK183 , FVT402 , FVT201 , CS2_WT ) were incubated with or without IGFBP7 in serum-enriched medium prior to rosetting assay . Subsequently , THP-1 were added . Using a wet mount technique , the IRBC phagocytosis rates were determined following the same formula to determine phagocytosis rate in the Zymosan A experiment . Prior to this , an experiment to compare IRBC phagocytosis activity of THP-1 and peripheral monocytes was conducted with a P . falciparum line ( NHP1106 ) , THP-1 and CD14+ peripheral monocytes from healthy donors , using the steps described above ( five biological replicates conducted ) . GraphPad Prism 7 . 0 was used for data analysis . For normally distributed data ( Shapiro-Wilk normality tested ) , a paired t-test was conducted for pairwise comparison . Matched measurement comparison for non-normally distributed datasets was done using Friedman test with Dunn’s multiple comparison test . To compare two sets of non-normally distributed data , a Mann-Whitney test was used . One-way ANOVA tests were conducted for grouped data set comparison . For a normally distributed dataset , Tukey’s test was applied for multiple group comparisons . Dunnett’s test was used to compare groups against a control . Two-way ANOVA was used to study the effect of multiple experiment conditions on rosetting in different parasite lines , each with different culture batches . Any p values < 0 . 05 were interpreted as statistically significant . | Malaria is a life-threatening disease transmitted by mosquitoes infected with Plasmodium parasites . Part of the parasite life cycle happens inside human red blood cells . The surface of an infected red blood cell is coated with parasite proteins , which attract the attention of white blood cells called monocytes . These immune cells circulate in the bloodstream and use a process called phagocytosis to essentially 'eat' any infected cells they encounter . However , the monocytes cannot always reach the infected cells . Some of the proteins made by the parasites make the infected red blood cells stickier than normal . This allows the infected red blood cells to surround themselves in a protective cage of uninfected red blood cells . Known as “rosettes” because of their flower-like shape , these cages seem to protect the infected cells from attack by the immune system . Lee et al . noticed that adding white blood cells to parasite-infected red blood cells made them clump together more , but it was unclear exactly how and why this happened . To find out , Lee et al . took fluid from around monocytes grown in the laboratory and added it to red blood cells infected with Plasmodium parasites . This made the cells clump together , suggesting that something in the fluid may potentially be alerting the parasites to impending immune attack . The fluid contained almost 700 different molecules , and Lee et al . narrowed down their investigations to the five most likely candidates . Interfering with the activities of these five proteins revealed that one – a protein IGFBP7 – not only alerted the parasites but also helped them to form the rosettes . It turns out that the parasites appear to use IGFBP7 like a bridge , linking it to two other human proteins to stick red blood cells together . Once the rosettes had formed , the monocytes were unable to eat the infected blood cells by themselves . Instead several monocytes had to work together as a team to consume the whole rosette . Further research is now needed to shed light on this interaction between malaria parasites and human cells . Such research would be particularly relevant in the clinical setting , since some previous studies has linked the forming of rosettes to the severity of disease for malaria . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"immunology",
"and",
"inflammation"
] | 2020 | Plasmodium-infected erythrocytes induce secretion of IGFBP7 to form type II rosettes and escape phagocytosis |
Retinal axon projections form a map of the visual environment in the tectum . A zebrafish larva typically detects a prey object in its peripheral visual field . As it turns and swims towards the prey , the stimulus enters the central , binocular area , and seemingly expands in size . By volumetric calcium imaging , we show that posterior tectal neurons , which serve to detect prey at a distance , tend to respond to small objects and intrinsically compute their direction of movement . Neurons in anterior tectum , where the prey image is represented shortly before the capture strike , are tuned to larger object sizes and are frequently not direction-selective , indicating that mainly interocular comparisons serve to compute an object’s movement at close range . The tectal feature map originates from a linear combination of diverse , functionally specialized , lamina-specific , and topographically ordered retinal ganglion cell synaptic inputs . We conclude that local cell-type composition and connectivity across the tectum are adapted to the processing of location-dependent , behaviorally relevant object features .
Theories of efficient sensory coding ( Barlow , 1961 ) often make the implicit assumption that the goal of sensory processing is a veridical representation of the external world . However , it is clear that the ultimate arbiter of efficiency is natural selection and that genetic information , developmental time , space , and material impose constraints on the design of the nervous system . Each of these evolutionary constraints has contributed to the neural implementations as we witness them in today’s animal brains , making the ultimate goal of calculating an optimization function difficult to achieve ( Chalk et al . , 2018; Dan et al . , 1996; Machens et al . , 2005; Simoncelli , 2003 ) . To understand why circuits are organized as they are and develop as they do , it is of paramount importance to identify constant and pervasive selective pressures that arise from the species-specific lifestyle of the animal . This study provides experimental support for the notion that the local statistics of the sensory environment , which changes dynamically as the animal interacts with the outside world , shape the topographic specializations of higher-order sensory and sensorimotor circuitry . For many decades the retinotectal projection of zebrafish has served as a paradigmatic example for a visual map . Retinal inputs to the tectum are ordered retinotopically such that the position of an object in the visual field matches a corresponding focus of activity in tectal space ( e . g . , Muto et al . , 2013 ) . Neighborhood relationships in the environment , as they are projected onto the two-dimensional sheet of photoreceptors in the retina , are represented by neural activity in neighboring regions of the tectum . Visual stimuli in the front of the larva are detected by temporal regions of the retina , which transmit pre-processed information via the axons of retinal ganglion cells ( RGCs ) to anterior regions of the tectum . Similarly , stimuli in the peripheral visual field behind the animal activate nasal retina and posterior tectum , respectively ( Figure 1A ) . The tectum then ultimately transforms visual information into behavioral commands ( e . g . , Helmbrecht et al . , 2018 ) . The neuropil of the larval zebrafish tectum is spatially organized along the superficial-to-deep axis into layers , ten of which are receiving input from dedicated subsets of RGCs ( Robles et al . , 2013; Robles et al . , 2014 ) . The remaining layers are innervated by axons from the somatosensory lateral line ( Thompson et al . , 2016 ) or contain dendrites and axons of interneurons and projection neurons ( Helmbrecht et al . , 2018 ) . The tectal neuropil layers are schematically depicted in Figure 1B . Recent work has revealed an enormous functional and morphological diversity of RGC types , which serve as local feature detectors for specific aspects of the visual scene , such as direction of motion , onset or offset of light , object size or chromaticity . Earlier studies have shown that individual RGCs select one layer each , in which they arborize and make synapses onto tectal dendrites ( Xiao and Baier , 2007 ) . Thus , each retinorecipient layer contains a complete , yet feature-selective , map of visual space . RGCs that respond to visual features resembling the speed and size of prey project to the most superficial layer ( SO; Semmelhack et al . , 2014; see Figure 1B ) , whereas RGCs that are specifically tuned to a rapidly expanding ( looming ) dark object , simulating an approaching predator or an obstacle on a collision course , terminate in deeper layers ( SFGS5/6; Temizer et al . , 2015; see Figure 1B ) . Asymmetries in visual feature processing have been recognized across the retina of several vertebrates ( for a recent review , see Baden et al . , 2020 ) . Prime examples for such functional specializations are the fovea of primates ( Sinha et al . , 2017 ) , the asymmetric distributions of RGC types and photoreceptors in mice ( Baden et al . , 2016; Bleckert et al . , 2014; Szatko et al . , 2020; Warwick et al . , 2018 ) and of bipolar cells , photoreceptors and RGCs in zebrafish ( Yoshimatsu et al . , 2020; Zhou et al . , 2020; Zimmermann et al . , 2018 ) . The two retinotopic dimensions of the tectum , the anterior-posterior and the dorsal-ventral axis , have so far received little attention in this regard . Zebrafish larvae do not possess a prima facie fovea , although they have evolved a high-acuity subarea in the temporal-ventral quadrant of the retina in which RGCs are more densely packed than in the periphery ( Schmitt and Dowling , 1999; Zhou et al . , 2020 ) . This region holds the image of prey in the final phase of hunting behavior and , similar to the mammalian fovea , occupies a disproportionately large area of the visual map in the tectum . Despite a wealth of data on tectal neuron morphologies ( see Figure 1B; Förster et al . , 2017; Nevin et al . , 2010; Robles et al . , 2011; Scott and Baier , 2009 ) , systematic changes in cell-type composition or connectivity along the anterior-posterior or dorsal-ventral axes of the tectum , resulting in gradients or other asymmetries of feature selectivity , have just begun to be revealed ( Wang et al . , 2020 ) . Here , we ask if such asymmetries can be predicted from first principles and related to the behavioral ecology of the zebrafish larva . As the animal interacts with a visual object through its own movements , relevant stimulus features continually change within the retinotopic coordinate frame . For example , in a typical hunting cycle , a zebrafish larva detects a prey item at a distance in its peripheral , monocular visual field ( Mearns et al . , 2020; Patterson et al . , 2013 ) . Posterior tectal circuits might therefore have evolved to respond to small-sized objects of ca . 5° and to locally compute their direction of movement . As the fish turns toward and approaches the prey , the stimulus enters the central , binocular visual field and expands to ca . 30° in visual angle ( Figure 1A ) . Activation of the anterior tectum has previously been described during this late hunting phase ( Muto et al . , 2013 ) . Neurons in the anterior tectum should therefore be tuned to larger object sizes and may rely on interocular comparisons to compute the object’s displacement from the midline . At all positions , the tectum should be able to distinguish between prey and looming threats and process them separately ( Barker and Baier , 2015 ) . The laminar segregation of functional channels , which are established by RGC inputs , is therefore expected to be maintained by tectal circuits independent of retinotopic location . Using volumetric two-photon calcium imaging to map out the feature space along the anterior-posterior retinotopic axis and across the layers of the tectum , we discovered a neural substrate for each of above predictions . Moreover , we show that the broad range of tectal responses originate to a large extent , but not exclusively , from a linear combination of functionally diverse RGC inputs . The dendrites of tectal cells are positioned in layers that predict their stimulus selectivity . We conclude that the cellular architecture underlying local processing in the tectum is adapted to the expected features of a prey object as it moves across the visual field during a hunting pursuit .
To broadly sample responses to object features , we designed a battery of simplified visual stimuli and controls . We employed two-photon calcium imaging of 5 to 7 dpf old larvae , which received monocular visual stimulation ( Figure 1C ) . At this larval stage , panneuronal expression of the nuclear-localized calcium indicator GCaMP6s ( driven by the elavl3 promoter ) labels on average 5793 ± 202 cells per tectum ( n = 10 fish; mean ± SEM ) ( Figure 1D ) . The stimulus set consisted of a moving dot of 5° ( ‘small’ ) , which approximates the size of prey at the onset of hunting behavior ( Bianco and Engert , 2015; Patterson et al . , 2013; Semmelhack et al . , 2014 ) , a moving dot of 30° ( "large" ) , which is the approximate size of prey directly before the capture strike , and an expanding disc at different velocities , which simulates an approaching object and is able to evoke escape responses ( Bhattacharyya et al . , 2017; Dunn et al . , 2016; Temizer et al . , 2015 ) . We further added controls for global luminance changes ( dark and bright ramps and flashes ) , as well moving gratings with high spatial ( 5° ) and temporal frequency as a negative control for small-dot responses ( Figure 1E; see Materials and methods ) . With this battery of visual stimuli , we obtained reproducible calcium responses in up to 30% of all tectal cells per imaging plane . We created 15 regressors for the different stimulus variants and calculated a score value for each tectal cell ( Figure 2A ) . To classify functional response types , we performed hierarchical clustering of representative response vectors obtained by affinity propagation ( see Materials and methods ) . This resulted in a dendrogram for 76 exemplars , which are representative of the 1759 sampled tectal cells in total ( Figure 2B–D , and Figure 2—figure supplement 1A ) . A silhouette analysis to validate the clustering showed that a minimal number of 14 clusters yielded an optimal classification of the data ( Figure 2—figure supplement 1B; see Materials and methods ) . To investigate the dimensional structure of the different response profiles , we performed principal component analysis on the scores for all tectal cells . Plotting the three main principal components ( PCs ) , which could explain 74 . 9% of the variance in the dataset , aligned the scores along three axes for small-dot , large-dot , and looming/luminance- ( OFF- ) responding cells ( Figure 2—figure supplement 1C and D ) . To show that the measured tectal cell responses were significantly different from chance , we shuffled the scores for each regressor 1000 times and calculated the PCs . Taking the average of the explained variance per shuffling , we consistently found a lower average explained variance , that is 57 . 6% for the three main PCs ( Figure 2—figure supplement 1D ) , indicating that tectal cells do not respond randomly to our set of stimuli . Overall , we found a broad spectrum of different response types in the tectum . Few cells responded to only one of the presented stimuli; most cells we imaged were multi-responsive ( Figure 2B ) . 43 . 6% of all cells responded to a looming stimulus ( with a score >0 . 2 ) , 41 . 1% responded to a small dot , and 33 . 1% responded to a large dot ( Figure 2D ) . Only a small number of cells responded to a bright ramp ( 2 . 7% ) or a bright flash ( 2 . 2% ) , and these cells were rarely sensitive to other stimuli . Responses to dark ramp and dark flash often coincided with each other and with responses to looming stimuli ( fast and slow ) , but rarely overlapped with responses to small or large moving dots . Responses to a slow-looming stimulus showed a gradual overlap with moving-dot responses; more than half of all cells that were sensitive to a large dot also responded to a slow-looming stimulus . The 5° grating did not trigger significant responses in the tectum , suggesting a selectivity to individual objects rather than to high spatial frequency . Next , we characterized the tuning properties of tectal cells whose somata reside inside the tectal neuropil . Superficial interneurons ( SINs ) , with cell bodies in the SO to SFGS1 neuropil layers , have previously been reported to receive size-tuned retinal inputs ( Del Bene et al . , 2010; Preuss et al . , 2014 ) . The largest fraction of SINs was mapped to our large-dot responsive cluster ( ~45%; Figure 2—figure supplement 1E ) , whereas only a small number of SINs ( ~6% ) were sensitive to a 5° dot . Neuropil interneurons ( NINs ) , residing within deeper layers of the neuropil , predominantly belong to the looming/dark ramp-responsive cluster ( ~30% ) , with about 20% of NINs responding to large dots ( Figure 2—figure supplement 1E ) . Taken together , the majority of tectal cells , both in the periventricular layer and embedded in the neuropil , respond to object motion , that is small , or large , or looming dots , sometimes in combination . A substantial fraction of cells responds to global dimming or looming ( OFF cells ) . Very few cells respond to global brightening ( ON cells ) . OFF and ON cells are largely non-overlapping with object-detecting cells . We next asked to what extent the feature selectivity of tectal neurons is inherited from retinal inputs . In our imaging setup , we applied the same battery of visual stimuli to larvae expressing cytoplasmic GCaMP6s in RGCs ( Figure 1D ) . A pixel-wise regressor and cluster analysis resulted in a dendrogram for 1157 exemplars , which were grouped into ten functional clusters ( although four RGC clusters resulted in the highest silhouette coefficient , we chose 10 clusters , for a significantly higher modeling correlation score , as shown below ) ( Figure 3A–C , and Figure 3—figure supplement 1; see Materials and methods ) . Overall , RGC responses were similar to tectal responses , but less specialized , with only few pixels responding exclusively to a single stimulus . Two thirds ( 67 . 2% ) of the pixels responded to a large dot with a score greater than 0 . 2 ( Figure 3A and Figure 3—figure supplement 1A ) . Generalized OFF responses to a dark ramp and a looming stimulus were similarly prominent . Non-intuitively , ON responses were sometimes combined with dark looming stimuli ( RGC cluster no . 2 and 9; Figure 3A and Figure 3—figure supplement 1A ) , a tuning profile we did not observe in tectal cells . Interestingly , direction-selective responses to forward- ( nasalward- ) moving stimuli , especially to a large dot , were more abundant than for the opposite direction ( Figure 3A ) . These units are expected to be activated when an object approaches from behind or when the fish turns toward an object in its peripheral visual field . We asked to what extent we could quantitatively explain the sampled tectal responses by using ganglion cell input . This analysis can distinguish between two extreme scenarios: The tectum may either be a passive relay station for RGC inputs . Or , alternatively , it may ‘re-compute’ the image based on unrelated RGC inputs . We used a simple feed-forward , linear modeling approach ( L1-regularized , Lasso ) with non-negative constraints to predict tectal cell responses by a sum of weighted RGC inputs ( Figure 4A; see Materials and methods ) . Modeling the scores for each of the 1759 tectal cells resulted in a high prediction quality ( median correlation R2score = 0 . 68 , median RMSE = 0 . 06; Figure 4B and Figure 4—figure supplement 1 ) . Similarly , we modeled the calcium transients for all tectal cells and calculated the correlation Rtrace between measured and predicted values ( Figure 4C and D ) . We also tested how a varying score threshold for the RGC responses , and thus a different number of RGC clusters would change the modeling prediction quality . We found that the best prediction of tectal calcium transients ( Rtrace ) can already be achieved by linear modeling of only four RGC clusters . Correlation for the tectal score values ( R2score ) , however , increases significantly with ten RGC clusters ( Figure 4—figure supplement 1 ) . Most tectal cell responses could be well explained by a linear combination of on average two RGC input clusters ( Figure 4D ) ;~36% of all responses could even be predicted by a single RGC input weight . However , specific tectal response features were modeled poorly: First , nearly all modeled tectal calcium traces showed responses to a large dot , owing to the high abundance of RGC responses to this stimulus ( Figure 4D ) . Second , the weak RGC responses to a moving small dot resulted in a poor prediction of the DS tectal clusters no . 2 and 6 ( Figure 4D ) . Third , modeling tectal calcium responses that are exclusive to ON or OFF stimuli was generally imperfect , and the worst correlation R2score was found for the tectal gradual OFF-selective cluster no . 11 . Our modeling results suggest that most visual representations in the tectum are directly inherited from RGC inputs . In addition , non-retinal , presumably intratectal computations add feature specificities , such as information on the direction of small moving objects , and sharpen both object-size and luminance selectivities of tectal neurons . We asked if tectal layers are distinct with respect to their feature selectivity . Along the superficial-to-deep axis , in line with previous publications , we found that RGC axons sensitive to small dots enter the tectum in superficial layers ( SO to SFGS4 ) with a peak in SFGS1/2 ( Figure 3—figure supplement 1D; Preuss et al . , 2014 ) . DS pixels were located most superficially in the posterior half of SFGS1 ( Figure 3C; Nikolaou et al . , 2012 ) . OFF-responsive axons , on the other hand , arborized in deep SFGS layers , SGC and SAC/SPV , and most extensively in SFGS5/6 ( Figure 3—figure supplement 1D; Temizer et al . , 2015 ) . To investigate if the dendrite morphologies of functionally identified tectal neurons matched these input layers , we carried out function-guided inducible morphological analysis ( FuGIMA ) of single tectal neurons ( Förster et al . , 2018 ) . We used nuclear-localized GCaMP6f ( nls-GCaMP6f ) and regressor-based analysis to identify tectal cells that belong to the three largest clusters: small-dot responsive , large-dot responsive , and OFF cells . Co-expressed photoactivatable GFP ( paGFP ) was then used to fluorescently label a cell of interest with a two-photon laser pulse directed at the soma ( Figure 5A and B ) . After allowing some time for diffusion of the activated GFP into the neurites , single cells were traced and registered to a standard brain together with RGC reference markers . This allowed us to quantify the extent of neurite arborization in each layer of the tectum ( Figure 5C–E ) . We compared our FuGIMA dataset ( n = 91 cells ) to a random collection of single tectal cells ( n = 188; Figure 5—figure supplement 1 ) , which were stochastically labeled with the BGUG method ( Xiao and Baier , 2007 ) . This analysis revealed that the three functional classes sampled branched preferentially in SFGS5/6 . In addition , we found that small- and large-dot responsive cells showed significantly denser arborizations in SO , SFGS1/2 , and SFGS3/4 compared to OFF cells . OFF cells , on the other hand , were biased to extend neurites in the SGC , the SAC , SAC/SPV and the SM ( Figure 5E ) . SM is a layer at the surface of the tectum , which is innervated by the torus longitudinalis , a higher-order visual area with strong OFF responses ( Northmore , 1984; Robles et al . , 2020 ) . SGC is a neuropil area abutting SFGS , in which multisensory information is processed . SAC is close to RGC axons that terminate in SAC/SPV and carry ambient luminance information to the tectum ( Kölsch et al . , 2020 ) . A comprehensive catalog of all identified tectal interneuron morphotypes is shown in Figure 5—figure supplement 2 . We further investigated the extent of tectal cell arborizations by measuring the arbor areas in each layer ( Figure 5—figure supplement 3A ) . We found that single cell arbors were generally small in superficial layers ( SM to SFGS3-4 ) and largest in deeper layers ( SFGS5-6 to SAC/SPV; Figure 5—figure supplement 3B ) . When comparing the ratio of deep vs . superficial arbor size of multi-stratified cells , we found morphological differences between object-motion responsive and OFF cells . While on average , small-dot responsive cells have a columnar shape , OFF cells have extended arbors in deeper layers , rendering them cone-shaped ( Figure 5—figure supplement 3C and D ) . We did not detect a systematic morphological difference between small- and large-dot responsive cells ( Figure 5F ) . Object-motion responsive and OFF cells thus target layers that match their corresponding retinal and , in the case of SM , non-retinal inputs and also differ in more subtle morphological features . Along the anterior-posterior ( A-P ) axis , we found a separation of size-selective RGC terminals . RGC axons responding to a large dot were mainly located in the anterior-dorsal quadrant of the tectal neuropil , whereas small-dot responsive pixels were found in the medial to posterior part ( Figure 3C ) . This compartmentalization is inherited by the corresponding tectal populations ( Figure 6A ) . Compared to all sampled cells , the large-dot response cluster was shifted to the anterior tectum , while cell bodies responding to small dots were biased to the posterior region . The strongest posterior bias was found for direction-selective cell bodies , responsive to a small , forward moving dot ( Figure 6A ) . We extended this analysis to our FuGIMA dataset , to quantify the extent of neurite arborizations in the neuropil . We found the same effect , that is large-dot responsive cells arborize more extensively in the anterior neuropil , compared to all sampled interneurons , while ~90% of neurites from DS small-dot responsive cells were found in the posterior half ( Figure 6B and C ) . These findings indicate a spatial gradient of sensitivity to object size , which is introduced by the topographic order of RGC inputs and inherited by the retinotopic array of tectal cells . To directly demonstrate that RGCs impose their feature selectivity onto postsynaptic tectal cells , we carried out an ablation experiment . From a previous study , we knew that small-dot responsive RGCs project specifically into SO after forming a collateral arbor in AF7 , the neuropil of the parvocellular superficial pretectal nucleus ( Semmelhack et al . , 2014 ) . By laser ablation of the RGC axon bundle that leaves AF7 , we achieved selective disruption of small-object input to the SO layer ( Figure 7A–C ) . Functional calcium imaging before and after the ablations revealed that small-dot responses were significantly diminished in tectal cells ( Figure 7D , E and G ) . In contrast , the number of looming-responsive cells in the affected tectum was not reduced , but even increased in some animals , possibly due to the loss of inhibition by the small-object-processing circuit ( see Barker and Baier , 2015; Figure 7D , F and G ) . These results indicate that RGC projections to SO are essential for tectal cells to assume their tuning to small-object motion . As the fish larva approaches a prey item , such as a paramecium or a rotifer , object size on the retina increases in visual angle . During hunting , the eyes converge and create an area of binocular overlap in the temporal retina . Convergent eye movements are accompanied by specialized turns , known as J-turns , that serve to center the prey in the visual field . Converged eyes and J-turns are characteristic of hunting episodes . We hypothesized that the large-dot responsive cells in the anterior tectum might be relevant for tracking prey at close range . To test this , we ablated between 3 and 15 single cells , which had been classified as large-dot responsive , in the right tectum ( Figure 8A ) . Prey capture behavior was then analyzed in free-swimming larvae ( Mearns et al . , 2020 ) . Following removal of large-dot responsive cells , animals spent less time with their eyes converged , indicating less time spent engaged in hunting behavior ( Figure 8B ) . In addition , their J-turns were biased to the right side , indicating defective prey detection by the left eye or right ( ablated ) tectum , respectively ( Figure 8C and D ) . Control fish , in which entirely non-responsive cells were ablated , showed no effect on prey capture behavior and were indistinguishable from untreated or agarose-embedded larvae ( Figure 8A–D and Figure 8—figure supplement 1A–C ) . Likewise , ablation of small-dot responsive cells , either in the anterior or posterior tectum did not result in significant behavioral changes ( Figure 8B–D ) . This suggests that for cells , which tile the visual field by only 5° , the ablated cell numbers were not sufficient to observe an effect on behavior . Intriguingly , we observed in our imaging experiments , that a substantial number of cells in the left tectum were responsive to prey-like stimuli presented to the left ( ipsilateral ) eye ( Figure 8—figure supplement 1D ) . These cells are probably activated by the right ( contralateral ) tectum via an intertectal commissure . We hypothesized that these cells help to sharpen responses across both tecta by suppressing background activity in the tectum that is not directly stimulated by RGC inputs . To test this hypothesis , we laser-ablated large-dot responsive cells in the anterior tectum on both sides ( Figure 8B–D ) . ( Note that these cells were identified by imaging responses in both tecta to stimulation of only the left eye . ) In bilaterally ablated animals , the tendency to increase right J-turns and reduce left J-turns in response to prey was even more pronounced than in right-tectum-only ablated larvae , supporting our hypothesis . Background suppression of the tectal activity ipsilateral to the stimulated eye might enhance activity in the contralateral , visually stimulated tectum . To investigate this possibility , we imaged the tectum of fish in which the right eye was removed ( Figure 8—figure supplement 1E ) . In these animals , we observed a significant increase in the number of large-dot responsive cells in the right , visually stimulated tectum , that is ipsilateral to the enucleated side . This result suggests that stimulus-evoked activity is normally dampened by background activity in the contralateral tectum by intertectal inhibitory connections . This background activity is suppressed , either physiologically by strong unilateral , stimulus-evoked activation of the other tectum ( Figure 8C and D ) , or experimentally by removal of its own retinal inputs ( Figure 8—figure supplement 1E ) . Taken together , our ablation results begin to reveal the logic of intertectal coordination of responses to prey in the frontal visual field .
In this study , we have discovered how the topographic layout of retinotectal circuitry is adapted to demands of the zebrafish larva’s behavioral ecology . We postulate that natural selection has favored the evolution of position-dependent specializations in the neural architecture underlying the processing of object motion as it is caused by both the prey’s and the fish’s movements . The tectum is critically involved in identification , localization , pursuit , and capture of prey ( Gahtan et al . , 2005; Semmelhack et al . , 2014 ) . In the following , we will go through circuit adaptations to each of these functions . Identifying prey , and distinguishing it from a potential predator , is critical for the larva’s survival . In previous studies , we discovered that this distinction is made by size and movement characteristics of the perceived object ( Barker and Baier , 2015 ) . Small , sideways-moving dots are readily approached ( Semmelhack et al . , 2014 ) , whereas expanding ( looming ) dots , displayed to the side of the fish while it is immobilized , are categorized as threatening and avoided by vigorous escape attempts ( Bhattacharyya et al . , 2017; Temizer et al . , 2015 ) . Here , we show that the RGC axon populations that respond to these two categories terminate in different layers of the tectum . Retinal inputs carrying prey-like signals mainly enter the tectum in layers SO and SFGS1-4 , and looming-sensitive RGC axons are largely restricted to SFGS5/6 and SGC . In addition , broadly-tuned OFF signals are transmitted to the deep retinorecipient layers SGC and SAC/SPV . These include sudden and gradual transitions from light to dark . The tectal cells that respond to these stimulus categories exhibit morphologies that match their predicted input channels , as previously shown for direction-selective RGC inputs and tectal cell dendritic arborizations ( Gabriel et al . , 2012 ) . Prey-selective neurons extend dendrite branches into the superficial layers of the tectum , while looming-sensitive neurons tend to arborize in middle to deep layers . As a general principle , most of the feature selectivities of tectal neurons are inherited from their functionally diverse RGC inputs . A simple excitation-only , feed-forward model showed that more than a third of the tectal response classes match a single RGC input class . The vast majority of the remaining responses were explained by a linear combination of two , or sometimes more , RGC inputs . In one experimentally accessible case , we could directly show that RGCs pass their small-dot responsive tuning on to downstream tectal cells . A similar modeling approach was recently performed to study functional connectivity between RGCs and the dorsolateral geniculate nucleus ( dLGN ) in the mouse thalamus ( Román Rosón et al . , 2019 ) . Analogous to our findings , the authors described a high correlation between functional dLGN in- and output , and thus a low level of signal convergence . The zebrafish tectum , however , is not merely a passive relay station for retinal inputs . First , we found that responses to large objects are markedly reduced in the tectum compared to RGCs . Second , direction selectivity to backward-moving objects is calculated de novo in the tectum . This was especially striking for the 5° dot stimulus . A circuit involving feed-forward inhibition by SINs , which suppresses tectal responses to non-preferred directions , could account for this computation ( Abbas et al . , 2017 ) . Third , a substantial number of tectal cells selectively respond to a dark ramp stimulus; such cells were not observed in our RGC dataset . Thus , tectum-intrinsic circuitry adds direction selectivity to a subset of channels and generally refines and sharpens the responses . The anatomical separation of small-dot responsive and looming-sensitive circuits probably reflects functional segregation of the two processing streams . Barker and Baier , 2015 postulated a circuit motif that implements balanced , reciprocal inhibition of the two systems driving approach vs . avoidance . Such a circuit could generate a winner-take-all mechanism capable of coordinating behavioral responses to stimuli of opposite valence . The visual system needs to rapidly distinguish between prey and threat across the entire visual field . A specialization of tectal layers for the processing of key features orthogonal to the two retinotopic axes , as reported here , seems to be an adaptive solution for that challenge . Moreover , bundling in space the visual processing of object valence , global patterns and luminance levels by laminar separation may also serve to minimize wiring lengths of the corresponding neural elements in the tectal neuropil ( Baier , 2013; Chklovskii et al . , 2002 ) . By sampling feature-selective responses along the anterior-posterior axis of the tectum , we uncovered functional specializations of tectal regions , which probably reflect systematic changes in cell-type composition and connectivity . Object translation across the visual field is caused by a combination of both the prey’s movement and the fish’s own swimming , the latter often in response to position of the prey . Larval zebrafish are able to detect a prey item at a distance of several millimeters . A typical prey object , such as a paramecium or a rotifer , of 250 µm length , which is 3 mm away , subtends a visual angle of approximately 5° . Previously , we and others had detected responses of head-fixed larvae , embedded in agarose , to virtual , high-contrast objects of 2–6° diameter ( Bianco et al . , 2011; Semmelhack et al . , 2014 ) . Moving dots of 1° rarely elicited a response . This seems to be the resolution limit of the larval fish’s visual system and is in agreement with the physical limit posed by photoreceptor spacing in the retina ( Haug et al . , 2010 ) . As the fish turns toward and approaches the prey , the prey ‘image’ slides from nasal to temporal zones of the retina and from posterior to anterior regions of the tectum ( Figure 8E ) . At the same time , the visual angle covered by the prey gradually increases . This might explain a shared sensitivity to slow-looming stimuli , which is featured by more than half of all large dot-responsive cells . Interestingly , this overlap is negligible for fast looming stimuli , what might indicate a separation of approach and avoidance circuits . The fish executes a capture strike when the prey is in the upper central field of both eyes at a distance range of 0 . 3–0 . 7 mm ( Mearns et al . , 2020 ) . This corresponds to 20–40° of visual angle . Two tasks of successful hunting , the detection of distant prey in the peripheral visual field and the fixation of prey at close range in front of the animal shortly before the capture strike , informed our choice of 5° ( ‘small’ ) and 30° ( ‘large’ ) virtual objects for our imaging experiments . We discovered two asymmetries in the retinotectal map that appear to support these two different phases of hunting behavior . First , an overrepresentation of small-dot responsive , direction-selective cells in the posterior tectum appears to be an adaptation to the preponderance of small prey objects in the peripheral field of view , whose movement is , at least initially , independent of the fish’s own . These tectal cells acquire their direction selectivity by de novo computations from size-tuned , non-DS RGC inputs ( Figure 8E ) . Determining the direction of prey by a local mechanism is particularly important for the lateral field of view , which is entirely monocular . The further away from the midline the prey’s location is the greater the turning angle that is needed to steer the fish toward its food . Fish will preferentially orient towards prey in their lateral visual field , because it gives them more time to move their body into the right position for a successful strike . Moreover , turning is energetically costly and may alert nearby predators and prey alike . For a prey object that already moves from back to front and whose image therefore slides from the nasal to the temporal retina , the turn angle will be smaller: the food may swim right in front of the fish , from where it might even be sucked into the mouth without extensive pursuit . Zebrafish larvae have been observed to use such a sit-and-wait mode of hunting ( Patterson et al . , 2013 ) . Second , the anterior tectum is enriched for large-dot responsive tectal cells ( Figure 8E ) , which appear to facilitate the initiation of prey capture-associated J-turns , as shown here by laser ablation . J-turns are fine adjustments of body posture characteristic of hunting . These cells are frequently not direction-selective and communicate via commissural connections with the contralateral tectum . Initial imaging and behavioral experiments following ablations suggest that activation of large-dot selective cells suppresses responses in the contralateral tectum . We propose that such an intertectal inhibitory mechanism helps to correct slight displacements of the prey from the midline ( see also Gebhardt et al . , 2019 ) . This signal may be transformed into fine orienting tail movements by one of the tectorecipient premotor areas in the hindbrain ( Helmbrecht et al . , 2018 ) . In conclusion , this work has revealed a neural architecture of the tectum that is well adapted to the demands of the animal’s behavioral ecology . More generally , we demonstrate that the well-studied retinotectal map is spatially organized by function along both its retinotopic and laminar axes . The visual map in the tectum is thus not a veridical , unbiased representation of all positions in visible space , but rather warped by location-dependent feature statistics . Future work will undoubtedly uncover additional adaptations and will shed light on both the proximate , developmental mechanisms and the ultimate , evolutionary forces that are shaping this important visuomotor hub in the vertebrate brain .
All animal procedures conformed to the institutional guidelines set by the Max Planck Society , and were approved by the regional government of Upper Bavaria ( Regierung von Oberbayern; approved protocols: ROB-55 . 2-1-54-2532-101-2012 and ROB-55 . 2–2532 . Vet_02-19-16 ) . To generate UAS:FuGIMA-f , paGFP ( gift from K . Svoboda , addgene no . 18697 ) and nls-GCaMP6f ( Förster et al . , 2017 ) were cloned on either side of a bidirectional 14xUAS in a Tol2 vector , featuring a transgenesis marker ( ‘bleeding heart’ , cmlc2:mCherry ) . Transgenic fish were generated using the standard Tol2 transposon system , and the highly variegated line Tg ( UAS:paGFP , nlsGCaMP6f ) mpn104 was used for experiments . For all experiments , we used 5–7 days post fertilization ( dpf ) larvae carrying mutations in the mitfa gene ( nacre ) , which were raised on a 14 hr light/10 hr dark cycle at 28°C . To record functional responses to visual stimuli of tectal cells , we used Tg ( elavl3:nls-GCaMP6s ) mpn400 fish and similarly for RGCs , we used Tg ( atoh7:Gal4-VP16 ) s1992t; Tg ( UAS:GCaMP6s ) mpn101 fish . RGC axon ablation experiments were performed in Tg ( atoh7:Gal4-VP16 ) s1992t; Tg ( UAS:mCherry ) s1984t; Tg ( elavl3:nls-GCaMP6s ) mpn400 fish , and tectal cells were ablated in Tg ( elavl3:nls-GCaMP6s ) mpn400 fish . FuGIMA experiments were performed in incrossed Et ( E1b:Gal4-VP16 ) s1101t ( =Gal4s1101t ) ; Tg ( UAS:paGFP , nlsGCaMP6f ) mpn104 ( =FuGIMA-f ) ; Tg ( elavl3:lyn-tagRFP ) mpn404 fish . Other single-cell reconstructions were generated using Et ( E1b:Gal4-VP16 ) s1013t ( =Gal4s1013t ) ; Tg ( brn3c:Gal4 , UAS:gap43-GFP ) s318t ( =BGUG ) fish . To define tectal layers , RGC expression in Tg ( isl2b:Gal4-VP16 ) zc65; Tg ( 14xUAS:EGFP ) mpn100 fish and in Tg ( Shha:GFP ) t10 fish was used . To allow registrations to a standard brain , all fish were crossed to the line Tg ( elavl3:lyn-tagRFP ) mpn404 . 5–7 dpf larvae expressing elavl3:nls-GCaMP6s were embedded in 2% low-melting-point agarose and a lethal dose of tricaine methanesulfonate ( MS-222 ) was applied . After 15 min , the tectal brain regions were imaged on a Zeiss LSM780 microscope ( voxel size: 0 . 27 × 0 . 27 × 1 . 5 µm3 ) . Images were manually segmented in Imaris ( v8 . 0 , Bitplane ) by setting pixel intensities outside of the tectum to 0 . Using ImageJ ( v1 . 52n ) , pixel intensities were inverted , images were Gaussian filtered and a classic watershed segmentation was applied ( MorphoLibJ plugin ) . ROIs smaller than 400 voxels were removed and the number of ROIs was analyzed in 3D . In vivo calcium imaging was performed on a previously described two-photon microscope ( Förster et al . , 2017 ) on 5–7 dpf transgenic zebrafish larvae expressing either cytoplasmic GCaMP6s in RGCs or nuclear-localized GCaMP6s panneuronally . Larvae were mounted in 2% low-melting-point agarose . The stimulus was projected onto a white diffusive screen using the red channel of a LED projector , in a distance of 4 cm from the larva . The projection was presented monocularly and covered ~120° of the larva’s field of view . GCaMP6 signals were recorded by scanning at 920 nm , at ~2 Hz , at a resolution of ~0 . 6 µm/pixel . The tectum was covered in depth by acquiring z-planes with a distance of ~7 µm . Visual stimulation was designed using PsychoPy2 and consisted of a dark ramp ( red to black , 3 s ) , a bright ramp ( black to red , 3 s ) , a dark flash ( red to black ) , and a bright flash ( black to red ) . This was followed by a small horizontally moving dot ( 5° , 90°/s ) in forward ( temporal to nasal ) and backward ( nasal to temporal ) directions ( two repetitions each ) , and at two elevations of the screen , first at equatorial plane and then elevated by ~20° ( two repetitions each ) . We chose dark dots on a bright ( red ) background . Published ( Antinucci et al . , 2019 ) and our own unpublished results have shown that these stimuli are efficient at eliciting hunting-like behavior in a dark 2P microscope environment , in the absence of UV stimulation ( Yoshimatsu et al . , 2020 ) . Subsequently , a big dot was moving horizontally ( 30° , 90°/s ) in forward and backward directions ( repeated twice ) , at an elevation of ~10° , thus covering the two horizontal planes of the small dot . The frequency control consisted of black gratings with a spatial frequency of 5° and a temporal frequency of 90°/s , moving in forward and backward directions ( repeated twice ) . The looming stimuli consisted of a fast ( ~60°/s , linear expansion ) and a slow-looming disc ( ~20°/s , linear expansion ) , both ending with a black screen ( two repetitions each ) . This stimulus protocol was repeated twice with a total acquisition length of 515 s . Recorded imaging data were pre-processed as described previously ( Helmbrecht et al . , 2018 ) . In brief , images were motion-corrected using the CaImAn package , uniformly filtered over three frames and the dF/F was calculated using the 5th percentile of the traces . In total 15 regressors for all stimulus components were created and convolved with a corresponding GCaMP6 kernel . Neuronal activity was analyzed pixel-wise for RGC and ROI-wise for tectal imaging data , by calculating a score of all regressors to the calcium responses of each pixel using a linear regression model of the selected response window with the regressor ( Python scikit-learn ) . For the score , the coefficient of the regression ( CR; corresponding to the dF/F ) was multiplied by the correlation value R2 . All pixels and ROIs were imaged twice using the same stimulus and the final score was calculated via a weighted average of the scores by the corresponding R2 . To determine overall response types , the scores were normalized per fish to the 99th percentile of all pixels/ROIs recorded . For the functional clustering of the responsive tectal ROIs , three fish ( 7594 ROIs ) expressing elavl3:nls-GCaMP6s were analyzed by first removing ROIs with maximum scores smaller than 0 . 2 ( 1908 ROIs remaining ) . Next , to reduce noise and to find local structure in the dataset , affinity propagation clustering ( scikit learn – preference: median of similarities ) was performed ( 151 clusters ) . Keeping clusters with at least 5 ROIs , yielded in total 80 clusters with chosen exemplars . To extract the global cluster structure , these 80 exemplars were further clustered using hierarchical clustering ( scipy . cluster ) using correlation as distance metric . Clusters with less than 20 ROIs were removed . We calculated a silhouette coefficient to validate the clustering . A distance threshold of 0 . 25 was chosen , which yielded a minimal number of clusters ( 14 ) with the highest silhouette coefficient . This finally resulted in 14 tectal cell clusters with a total of 76 exemplars and 1759 ROIs ( 92 . 2% ) . Principal component analysis ( PCA ) was performed on the score values of each stimulus for all tectal cells . Similarly to the clustering of tectal neurons , the responsive RGC pixels of one fish ( 14 planes; each 297 × 303 pixel ) expressing ath5:Gal4 UAS:GCaMP6s were analyzed by again removing pixels with maximum scores smaller than 0 . 4 ( remaining 58 , 910 pixel ) and performing affinity propagation clustering ( scikit learn – preference: median of similarities ) . Keeping clusters with at least five pixels ( 0 . 01% of all pixels ) , yielded in total 1243 clusters with chosen exemplars . These 1243 exemplars were further ordered by hierarchical clustering ( scipy . cluster ) using correlation as distance metric . Cluster with less than 589 pixels ( 1% of all pixels ) were removed . After silhouette analysis , a distance threshold of 0 . 45 was chosen , which yielded ten clusters with a total of 1157 exemplars and 55 , 153 pixels ( 93 . 6% ) . Although four RGC clusters yielded a higher silhouette coefficient , we chose ten clusters , which resulted in a significantly higher correlation value ( R2score ) for the following linear modeling analysis ( see Figure 4—figure supplement 1A ) . To quantify the number of pixels per RGC cluster in tectal compartments and layers ( Figure 3C and Figure 3—figure supplement 1D ) , we used ImageJ to manually draw ROIs and to count pixels for each compartment/lamina in each image plane . To map response types of SINs , NINs and enucleated fish , functional imaging was performed as described . ROIs were defined semi-automatically to segment only single , separated tectal cell bodies in the tectal neuropil and/or the periventricular layer . Several fish per experiment were analyzed to calculate the scores , and again pixels with maximum scores smaller than 0 . 2 were removed . A k-nearest neighbor classifier ( sklearn . neighbors . KNeighborsClassifier ) was trained on the elavl3:nls-GCaMP6s clustered ROIs ( 1759 ROIs with cluster labels , k = 10 ) and the scores of every mapped fish were assigned to the cluster dataset using either predicted labels for the ROIs distribution or probability estimates for the population distributions . The classification was cross-validated by splitting the elavl3:nls-GCaMP6s dataset into 70% training and 30% test data , which evaluated to an accuracy of 92% . A similar , pixel-wise approach was used to map the functional RGC data of two additional ath5:Gal4; UAS:GCaMP6s fish onto the ten RGC clusters by choosing k = 100 ( Figure 3—figure supplement 1C ) . To predict the tectal responses using RGC information , we applied a linear modeling approach using L1-regularized regression ( Lasso ) ( sklearn . linear_model . Lasso ) with non-negative constraint . The cost function of the Lasso is defined by:Cost=∑i=0nyi- ∑j=0mwj xij2+λ ∑j=0mwj The regularization parameter ( λ ) helps to reduce the impact of multicollinearities between the average scores of RGC classes , and the optimal λ was found by minimizing the mean squared error of a grid search on a log scale between 1e−5 and 1e−1 ( Figure 4—figure supplement 1B ) . The modeling of the scores of every single tectal neuron ( total 1759 ) was performed using the 15-dimensional average scores of the 10 defined RGC clusters , so that:PredScore TectalNeuron=b+ ∑j=0m ( RGC ) wj AvgScoreRGCj The PredScore was evaluated by calculating the R2score of the regression . To predict the calcium traces of the tectal cells , we used the resulting weights of the regression and calculated the dot product of the average RGC responses with the corresponding weights ( w ) and bias ( b ) and evaluated the result via the pearson correlation ( Rtrace ) between the predicted and measured calcium responses . The model was tested by comparing the resulted distribution of response correlations to the distribution of a random model , by choosing for every cell 1000 times random weights ( Figure 4—figure supplement 1C ) . In addition , the model was cross-validated by splitting the data into a training and test set using one of the two trials per cell , and a corresponding RMSE ( root mean squared error ) of the test dataset was calculated ( Figure 4B ) . Tectal responses in fish expressing elavl3:lyn-tagRFP and UAS:FuGIMA-f under control of Gal4s1101t were functionally imaged as described above . After image acquisition , a custom-written , regressor-based python script was used to overlay a color map of correlated pixels on the mean ∆F/F image to identify cells of functional interest . Single-cell photoactivation of paGFP was performed as previously described ( Förster et al . , 2018 ) . Typically , 2–3 photoactivation cycles were sufficient to reach the maximal fluorescence intensity in tectal interneurons . After allowing paGFP to diffuse into all neurites of the photoactivated cell for about 30–45 min , a high-resolution z-stack of the whole tectum , including both paGFP and lyn-tagRFP channels , was acquired at a confocal microscope ( LSM700 or LSM780 , Zeiss; 20x/1 . 0 NA water-dipping objective ) . Other single-cell reconstructions ( randomly-labeled tectal neurons ) were performed using the BGUG method as previously published ( Helmbrecht et al . , 2018 ) . In brief , fish expressing a highly variegated Gap43-GFP under control of the tectal Gal4s1013t line were crossed to elavl3:lyn-tagRFP fish and offspring were screened for sparse GFP expression in tectal interneurons . All individual neurons were traced semi-automatically using the software neuTube ( Build1 . 0z ) and SWC files were generated for each cell . All image registrations were performed using the Advanced Normalization Tools ( ANTs ) software ( Avants et al . , 2010 ) , and live expression of elavl3:lyn-tagRFP served as a reference channel . First , a FuGIMA standard brain was generated by mirroring all FuGIMA cells to one brain half and by subsequent registration to one exemplary lyn-tagRFP channel , which served as a template . ANTs parameters recently determined for live samples were applied ( Marquart et al . , 2017 ) . Second , this FuGIMA standard brain was registered to the zebrafish single-neuron atlas ( Kunst et al . , 2019 ) in three steps: ( 1 ) registration of the FuGIMA template to a tectal subvolume of the live lyn-tagRFP standard brain from the atlas , ( 2 ) extension to the full live standard brain volume , ( 3 ) registration of the live standard brain to the fixed standard brain of the atlas . Similarly , the BGUG dataset was first registered to its own standard brain , which was subsequently registered to the single-neuron atlas . Finally , single-neuron tracings ( SWC files ) were aligned using the antsApplyTransformToPoints function contained in the ANTsR package . For visualizations and 3D renderings , we used the web interface of the single-neuron atlas ( http://fishatlas . neuro . mpg . de/ ) . All single-neuron data from this study are publicly available through this atlas . To add landmarks for the tectal laminae , we co-registered the expression patterns of isl2b:Gal4 UAS:GFP and shh:GFP into the FuGIMA standard brain . We then used these anatomical labels , together with the software 3D slicer ( http://www . slicer . org/ ) , to manually segment the individual tectal layers . For every cell , we measured the fiber lengths in each layer and calculated the percentage of the cell's total neurite length ( proportional branch length ) using a custom-written python script . Single-cell morphological barcodes ( heatmaps ) were generated using Plotly Chart Studio ( https://plot . ly/ ) . To quantify the neurite arbor size of tectal cells , we used the 'Oblique slicer' and 'Measurement points' tools in Imaris ( v8 . 02; Bitplane ) to define and extract planar coordinates for each laminar stratification ( Figure 5—figure supplement 3A ) . The areas in µm2 were quantified using a custom-written python script . For RGC axon ablations , 6 dpf old larvae expressing mCherry in RGCs and nuclear GCaMP6s panneuronally were mounted in agarose and were intraspinally injected with alpha-bungarotoxin ( 2 mg/ml , Invitrogen , B1601 ) . Tectal cell responses were functionally imaged as described above . Subsequently , the axon bundle , which leaves AF7 for the tectal SO layer was cut at the same 2P microscope by scanning a 10 µm line ( 0 . 01 µm/pixel ) at 760 nm for 500 ms transverse to the fascicle . The laser intensity at the objective focal plane was ~30 mW . Afterwards , fish were released from agarose to recover overnight in Danieau's solution . At 7 dpf , fish were re-embedded and functional imaging of tectal cell responses was repeated . Somata signals in the tectal neuropil served as landmarks for approximate reidentification of the same imaging planes obtained at 6 dpf . Regressor analysis was described as above and cluster-color-coded responsive tectal cells were counted manually . For tectal cell ablations , 7 dpf old larvae expressing nuclear GCaMP6s panneuronally were embedded in agarose and functionally imaged at the 2P microscope . Up to 3–5 tectal cells per imaging plane ( max . 15 cells per fish ) were selected for their response type , and were ablated by 30 ms two-photon laser pulses ( 800 nm , ~35 mW ) , pointed at the nucleus . For enucleation experiments , 4 dpf old fish expressing elavl3:nls-GCaMP6s were placed in 2% low-melting agarose with 0 . 02% tricaine methanesulfonate ( MS-222 ) . The right eye was removed using custom-made micro-scalpels . Fish were allowed to recover for two days in Danieau's solution until functional imaging was performed at 6 dpf . Prior to testing prey capture behavior , larvae were allowed to feed ad libitum on paramecia from 5 to 6dpf . At 7 dpf , larvae were embedded in agarose and cells in the tectum were ablated ( see above ) . Larvae were freed from agarose and allowed to recover overnight . Prey capture behavior was tested the following day at 8 dpf . Controls groups were unembedded siblings , siblings embedded but not subject to the ablation protocol , and ‘sham’ ablated siblings . The free-swimming prey capture assay was performed as described previously ( Mearns et al . , 2020 ) . Briefly , larvae were introduced individually into an arena ( 15 × 15 × 5 mm ) with 50–100 paramecia ( Paramecium multimicronucleatum ) . Each larva was allowed to feed for 20–30 min while being recorded from above at 500 frames per second using a high-speed camera ( PhotonFocus , MV1-D1312-160-CL , Switzerland ) . In each frame of the recordings , the eyes and tail of the fish were tracked offline using custom-written Python software ( https://bitbucket . org/mpinbaierlab/mearns_et_al_2019 ) . Tail tracking was performed using background subtraction and thresholding followed by skeletonization of the largest contour in the image . Swim bouts were identified using a change point algorithm on the derivative of the tail angle with respect to time . Eye tracking was performed similarly using background subtraction , thresholding and contour detection . For each animal independently , we calculated the distribution of eye convergence angles over the experiment and used the local minimum in the resulting bimodal distribution as the prey capture threshold . Since eye convergence is a reliable indicator of prey capture in zebrafish larvae ( Bianco et al . , 2011; Patterson et al . , 2013; Mearns et al . , 2020 ) , we defined hunting events as any time the eye convergence angle was above this threshold . Initial orienting J-turns were defined as any bout where the eyes were unconverged before and converged after the bout . The bout integral was calculated by summing the tail tip angle values over the duration of the bout , with positive values indicating a rightward turn and negative values indicating a leftward turn . The direction of the turn was defined by the sign of the bout integral ( positive for right , negative for left ) . The direction selectivity index was computed as [ ( # right J-turns - # left J-turns ) / ( total # J-turns ) ] for each fish , with a value of 1 indicating all J-turns were to the right , −1 indicating all J-turns were to the left , and 0 indicating no overall bias in J-turn direction . Statistical tests were two-tailed t-tests , if not stated otherwise . For the quantification of prey capture behavior , statistics were performed using the scipy library in Python 3 . The proportion of time larvae spent engaged in hunting behavior was compared between treatment groups using a Mann-Whitney U test . Similarly , the direction selectivity index of initial J-turns was compared between treatment groups using a Mann-Whitney U test . | The retina is the thin layer of tissue in the eye that can receive light stimuli and convert them into electric signals to be transmitted to the brain . The cells that sense fine detail cluster at the center of the retina while the motion-sensing cells that keep track of movement lie at the periphery . When zebrafish larvae hunt , their motion-sensing cells are triggered as a prey crosses their peripheral field of view . They then turn and swim towards it . As they approach , the prey image moves to the detail-sensing part of the retina and appears larger , filling more of the field of view at close range . The signals are then processed in defined parts of the brain , in particular in a region called the optic tectum . How this area is organized in response to the organization of the eye and the requirements of the hunt is still unclear . Förster et al . set out to explore how the hunting routine of zebrafish larvae shapes the arrangement of neurons in the optic tectum . The larvae were exposed to different images representing the various aspects of the prey capture process: small moving dots represented passing prey at a distance , while large moving dots stood for prey just before capture . Measuring activity in the neurons of the optic tectum revealed that , like the eye , different areas specialize in different tasks . The back of the tectum was frequently activated by small dots and worked out which direction they were moving in during the first hunting steps . The front of the tectum responded best to large dots , often ignoring their direction , and helped the larvae to track their prey straight ahead . To test these findings , Förster et al . destroyed the large object-responsive cells with a laser and watched the larvae hunting real prey . Without the cells , the fish found it much harder to track and catch their targets . These results shed light on the link between behavior and how neurons are arranged in the brain . Future work could explore how the different neurons in the optic tectum are connected , and the behaviors they trigger in the fish . This could help to reveal general principles about how sensory information guides behavior and how evolution has shaped the layout of the brain . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2020 | Retinotectal circuitry of larval zebrafish is adapted to detection and pursuit of prey |
Respiration , an essential metabolic process , provides cells with chemical energy . In eukaryotes , respiration occurs via the mitochondrial electron transport chain ( mETC ) composed of several large membrane-protein complexes . Complex I ( CI ) is the main entry point for electrons into the mETC . For plants , limited availability of mitochondrial material has curbed detailed biochemical and structural studies of their mETC . Here , we present the cryoEM structure of the known CI assembly intermediate CI* from Vigna radiata at 3 . 9 Å resolution . CI* contains CI’s NADH-binding and CoQ-binding modules , the proximal-pumping module and the plant-specific γ-carbonic-anhydrase domain ( γCA ) . Our structure reveals significant differences in core and accessory subunits of the plant complex compared to yeast , mammals and bacteria , as well as the details of the γCA domain subunit composition and membrane anchoring . The structure sheds light on differences in CI assembly across lineages and suggests potential physiological roles for CI* beyond assembly .
Respiration is an essential metabolic process that provides the energy and intermediate metabolites needed for growth and maintenance of all eukaryotes . In plants , respiratory pathways are not only involved in energy conversion but also play crucial roles in the procurement of biosynthetic precursors and in the balancing of the cellular redox state ( O'Leary et al . , 2019 ) . Plant respiratory processes are also closely intertwined with photosynthetic pathways . Despite the importance of respiratory processes to plants’ biomass accumulation , carbon flux and acclimation ( O'Leary et al . , 2019; Amthor et al . , 2019; Heskel et al . , 2016 ) , the fundamental mechanisms by which the plant mitochondrial electron transport chain ( mETC ) produces proton ( H+ ) gradients that are converted into chemical energy remain poorly understood . Molecular knowledge of the structures and mechanisms of the plant mETC components , which differ significantly in their assembly and composition from better-studied mammalian systems , is essential to understand how plants efficiently convert energy and balance respiration with photosynthesis . Plant mitochondria possess a ‘canonical’ mETC shared with most eukaryotes that is composed of four large membrane protein complexes ( complexes I-IV , CI-IV ) and an associated ATP synthase in the inner mitochondrial membrane ( IMM ) . Complexes I-IV couple oxidoreduction reactions to H+ pumping against the concentration gradient across the IMM to produce a large H+ electrochemical potential ( ‘proton motive force’ ) that is then dissipated through ATP synthase’s rotary mechanism to produce ATP in the mitochondrial matrix . Additionally , plants also possess an ‘alternative’ mETC that dissipates reduction equivalents in a non-H+-pumping , non-energy-conserving fashion ( Millar et al . , 2011; Schertl and Braun , 2014 ) . Complex I ( CI ) is the main energy-conserving entry point for electrons into the mETC . In plants , as in most eukaryotes so far studied , CI is the largest ( ~1 MDa ) and mechanistically least understood component of the mETC ( Sazanov , 2015; Hirst , 2013 ) . CI oxidizes NADH and reduces coenzyme Q ( CoQ , ubiquinone ) , pumping four H+ per two electrons from NADH ( Jones et al . , 2017 ) . CI is an L-shaped multiprotein complex , with a membrane arm and a peripheral arm . In eukaryotes , the peripheral arm of CI extends into the mitochondrial matrix , while the membrane arm is buried within the IMM . Both arms are composed of ‘modules’ with specific functions and distinct evolutionary origins ( Efremov and Sazanov , 2012 ) . The peripheral arm contains the NADH dehydrogenase N-module and the CoQ-reducing Q-module , which provide the binding sites for NADH and quinone , respectively , as well as the chain of FeS clusters needed for electron transfer ( Figure 1A ) . The membrane arm contains four proton pumps , two of which are located in the proximal-pumping module ( PP ) , with the remaining two pumps in the distal-pumping module ( PD; Figure 1A; Dröse et al . , 2011 ) . Through a still poorly understood mechanism , the energy released from NADH-CoQ oxidoreduction in the peripheral arm ( N- and Q-modules ) is coupled to conformational changes along the membrane arm ( PP and PD ) , resulting in proton pumping from the mitochondrial matrix into the mitochondrial intermembrane space ( IMS ) . Across the studied eukaryotes , mitochondrial CI is composed of 14 highly conserved ‘core’ subunits that are responsible for electron transport and H+ pumping , and 30–35 ‘accessory’ subunits that are involved in CI’s assembly , stability and regulation ( Millar et al . , 2011; Meyer , 2012 ) . The exact number of subunits in plant mitochondrial CI is still unclear , with several mass spectrometry measurements revealing differing compositions ( Meyer , 2012 ) . Nonetheless , it is known that several plant CI accessory subunits are not found in fungi and metazoans ( opisthokonts ) . Most notably , five gamma-type carbonic anhydrase ( γCA ) proteins ( CA1 , CA2 , CA3 , CAL1 , and CAL2 ) have been shown to be associated with CI in plants ( Sunderhaus et al . , 2006; Perales et al . , 2004 ) . These proteins are located on the matrix side of CI’s membrane arm , likely as a heterotrimer of CAL1 or CAL2 monomer plus a CA1/CA2 hetero- or homodimer ( Fromm et al . , 2016 ) . Hence , only a subset of the five γCA proteins are expected to be simultaneously associated with CI . Although the exact γCA protein combinations are likely tissue- and development-stage-dependent ( Cï Rdoba et al . , 2019 ) , the role of the γCA domain in plant CI’s function is unknown ( Martin et al . , 2009 ) . Another major difference between plants and metazoans occurs in the CI assembly pathway . In metazoans , the N-module ( which is responsible for NADH oxidation ) is assembled onto the rest of the complex ( Q- , PP- and PD-modules ) as the final step of assembly ( Formosa et al . , 2018; Guerrero-Castillo et al . , 2017; Garcia et al . , 2017; Stroud et al . , 2016; Figure 1—figure supplement 1A ) . In plants , more similar to what occurs in bacterial CI assembly ( Friedrich et al . , 2016 ) , the final assembly step is the attachment of the PD-module onto an intermediate ( termed CI* ) that already contains the N- , Q- and PP-modules ( Ligas et al . , 2019; Figure 1—figure supplement 1B ) . This difference in the order of assembly of CI in plants vs . metazoans is significant: in metazoans , adding the NADH dehydrogenase N-module last ensures that no assembly intermediate is capable of transferring electrons from NADH to CoQ . This is believed to have protective roles , to prevent the formation of reactive oxygen species during the CI assembly process ( Parey et al . , 2019 ) . In contrast , the plant CI* intermediate contains all the subunits and co-factors needed to carry out NADH:CoQ oxidoreduction . In contrast to the large number of recent high-resolution structures of mammalian and yeast respiratory complexes and supercomplexes , the most detailed plant CI structures known were obtained by negative-stain electron microscopy ( EM ) two-dimensional ( 2D ) classifications from Solanum tuberosum ( potato ) and Arabidopsis thaliana or sub-tomogram averaged reconstructions that lack secondary structure details ( Bultema et al . , 2009; Dudkina et al . , 2005; Davies et al . , 2018 ) . The paucity of functional and structural data for plant mETC complexes stems in large part from the limited availability of sufficient protein sample needed for structural analysis ( Dudkina et al . , 2015 ) . Indeed , it has been difficult to obtain intact plant mitochondria in sufficient amounts for preparative biochemical fractionation . A typical reported yield of mitochondria is ~0 . 2–0 . 5 mg mitochondria/g fresh weight of starting plant material ( Luster and Fites , 1987 ) , which contrasts with a yield of ~30 mg mitochondria/g fresh weight from mammalian sources . In light of these challenges , most of the biochemical data on plant mETC have used intact mitochondria ( e . g . oxygen-consumption experiments ) or complexes that have been electro-eluted from electrophoretic gels ( Bultema et al . , 2009; Dudkina et al . , 2005; Dudkina et al . , 2006; Eubel et al . , 2005 ) . Although such electro-eluted protein samples have yielded the low-resolution structures described above and have proven suitable for proteomic studies , the low yields and low activities of these protein samples have so far thwarted detailed functional or structural analyses of the plant mETC complexes . A detailed understanding of the energy-converting mechanisms of plant respiratory mETC complexes and supercomplexes requires improved protocols for their extraction from plant mitochondrial membranes , and their purification in sufficient amounts while maintaining them in a functionally active state . Here , we present a cryoEM structure of an ~800 kDa assembly intermediate of plant mitochondrial CI from etiolated Vigna radiata ( mung bean ) hypocotyls at 3 . 9 Å resolution . This assembly intermediate , CI* ( Ligas et al . , 2019 ) , contains the intact peripheral arm ( N- and Q-modules ) as well as the PP-module and γCA domain , but lacks the PD-module . Our structure allowed us to build the first atomic model for any mitochondrial CI species from the plant kingdom and revealed important differences in the CI core and accessory subunits between plants , mammals , yeast and bacteria . Such subunit differences shed light on the known differences in CI assembly in plants versus opisthokonts . The structure also allowed us to define the interface between the γCA domain and the membrane arm of CI and revealed a key role for lipids in this interaction . We also discuss the implications of our findings on the possibility that CI* may provide additional flexibility to plants’ mETC .
In order to investigate the plant mitochondrial electron transport chain , we identified V . radiata ( mung bean ) as an optimal model system . V . radiata offers several advantages for plant mitochondrial research: i ) it can be easily sprouted and harvested within six days , ii ) it can be grown in the dark ( etiolated ) to minimize development of chloroplasts , which would otherwise contaminate the mitochondrial preparations , iii ) its age and growth conditions can be controlled experimentally , iv ) its genome has been sequenced and v ) its mitochondrial content has been reported to be higher than other plant sources previously used for plant mitochondrial research ( Luster and Fites , 1987 ) . Moreover , we have optimized standard plant mitochondria isolation protocols ( Millar et al . , 2007 ) to routinely obtain ~1 g of wet weight mitochondria per 1 kg of etiolated V . radiata hypocotyls , approximately 3–4 times what has been previously reported ( Luster and Fites , 1987 ) . Isolation of the mitochondrial electron transport complexes of V . radiata was performed by extraction from washed mitochondrial membranes using the gentle detergent digitonin , followed by exchange into the amphipathic polymer A8-35 to further stabilize the complexes . The presence of complex I ( CI ) -containing bands was analyzed using a standard in-gel NADH-dehydrogenase activity assay for CI on a blue-native gel ( BN-PAGE ) ( Schertl and Braun , 2015 ) . As expected from previously reported plant mitochondrial extractions ( Bultema et al . , 2009; Dudkina et al . , 2005; Eubel et al . , 2004a; Eubel et al . , 2004b; Eubel et al . , 2003; Krause et al . , 2004 ) , we observed a number of bands with NADH-dehydrogenase activity , representing CI in different assembly states , such as in mitochondrial supercomplexes ( Bultema et al . , 2009; Dudkina et al . , 2005; Eubel et al . , 2004a; Eubel et al . , 2004b; Eubel et al . , 2003; Krause et al . , 2004; Dudkina et al . , 2010; Figure 1—figure supplement 2A ) . The amphipol-stabilized complexes and supercomplexes were separated on a linear sucrose gradient ( Figure 1—figure supplement 2B–C ) . Two peaks displaying NADH-dehydrogenase activity were of sufficient amount to be further purified by size-exclusion chromatography ( Figure 1—figure supplement 2D ) . These purified fractions retained their NADH-dehydrogenase activity by in-gel activity assays ( Figure 1—figure supplement 2E ) . Moreover , these fractions also showed NADH-decylubiquinone oxidoreductase activity using a standard CI spectroscopic activity assay ( Huang et al . , 2015; Figure 1—figure supplement 2F ) . These fractions were investigated by single-particle cryoEM . Here , we present results from the lower molecular weight fraction ( ‘peak 2’ ) ( Figure 1—figure supplement 2G–H ) . Structural analysis revealed that this fraction contained an ~800 kDa CI subcomplex , previously identified as a plant mitochondrial CI assembly intermediate termed complex I* ( CI* , Figure 1B ) , which we were able to resolve to a nominal resolution of 3 . 9 Å ( Figure 1C , Tables 1–2 , Video 1 ) . The existence of this assembly intermediate has been determined by genetic and mitochondrial proteomics experiments of CI's assembly pathway in etiolated seedlings ( Heazlewood et al . , 2003 ) and non-etiolated seedlings and leaves of Arabidopsis thaliana ( Ligas et al . , 2019; Meyer et al . , 2011; Schertl et al . , 2012; Schimmeyer et al . , 2016; Senkler et al . , 2017 ) , as well in non-etiolated leaves of Nicotiana sylvestris ( Pineau et al . , 2008 ) . Moreover , the A . thaliana and N . sylvestris CI* intermediate shows NADH-dehydrogenase activity by the same in-gel activity assay used in our preparation ( Meyer et al . , 2011; Pineau et al . , 2008; Haïli et al . , 2013 ) . CI* contains CI’s intact peripheral arm ( N- and Q-modules ) , PP-module and γCA domain . However , it is missing the two membrane arm core subunits NU4M and NU5M and their associated accessory subunits that form the PD-module ( Figure 1B ) . As expected from complexome profiling analyses ( Ligas et al . , 2019; Senkler et al . , 2017 ) , our structure of CI* is composed of over 30 subunits of the N-module , Q-module , PP-module and the γCA domain . Throughout this manuscript , we use the plant nomenclature for the subunits ( see Table 3 for subunit name conversions ) . The peripheral and membrane arm core subunits present in the structure of CI* are structurally homologous to the bacterial , yeast and mammalian CI core subunits , with a few notable differences . The N-terminus of core Q-module subunit NDUS2 is shortened in V . radiata compared to NDUS2 from Y . lipolytica and mammals , in which the N-terminus of NDUS2 extends from the interface of the peripheral and membrane arms of the complex along the matrix side of the membrane arm . Whereas in Y . lipolytica the N-terminus of NDUS2 binds to the matrix surface of core H+-pumping subunit NU2M , in mammals the N-terminus of NDUS2 extends further along the membrane arm and binds to the matrix surface of core H+-pumping subunit NU4M , bridging across the PP- and PD-modules . In contrast , V . radiata NDUS2 is ~40 amino acid residues shorter on the N-terminus compared to mammals and does not extend along the membrane arm . Moreover , the equivalent path for the Y . lipolytica or mammalian NDUS2 N-terminus in V . radiata is blocked by the γCA domain to the plant PP-module on the membrane arm . The N-terminus of core peripheral arm subunit NDUS8 is also divergent between plants , fungi and mammals . In V . radiata , the N-terminus possesses an additional α-helix that binds between the Q-module accessory subunit NDUA5 and the PP-module core membrane subunit NU2M , enlarging the interaction interface between the peripheral and membrane arms ( Figure 1C ) . In Y . lipolytica , the N-terminus of NDUS8 forms an extended coil that reaches up along the peripheral arm between the Q-module accessory subunits NDUA5 and NDUA7 , making contact with the core Q-module subunit NDUS3 . In contrast , the N-terminus of mammalian NDUS8 folds back along the surface of the membrane arm and tucks underneath the Q-module accessory subunit NDUA7 . In Y . lipolytica , this binding site , underneath the NDUA7 homologue ( NUZM ) , is occupied by NUZM’s C-terminus , which folds back under itself . However , in V . radiata the binding site underneath NDUA7 is occupied by an unidentified subunit that extends from this pocket under NDUA7 toward the core transmembrane subunits adjacent to the NU3M transmembrane helix ( TMH ) 1–2 loop and the NU6M TMH3 , which undergo conformational changes during CI’s enzymatic turnover in the fungal structures ( Agip et al . , 2018; Letts et al . , 2019; Parey et al . , 2018 ) . Although the identity of this sequence in the V . radiata structure remains unclear , it appears to be unique to plant CI . Core subunit NU2M in V . radiata CI* contains three N-terminal transmembrane helices that are present in yeast and bacterial complexes , but lost in the metazoan lineage ( Birrell and Hirst , 2010 ) . Moreover , V . radiata CI* contains a homologue of Y . lipolytica’s accessory subunit NUXM ( absent in metazoans ) , which binds to the NU2M N-terminal transmembrane helices . Based on the Y . lipolytica subunit name , we coined this subunit of V . radiata CI NDUX1 . The presence of this subunit in both plants and fungi suggests that this subunit was present in the ancestral eukaryotic CI before the unikont/bikont lineage divergence but was lost in metazoans when NU2M became N-terminally truncated . The first transmembrane helix of NU2M in Y . lipolytica is notably short ( only 15 amino acids ) , enters only to the midplane of the membrane and is bound by a membrane-penetrating loop of the accessory subunit NUXM . In contrast , in bacteria ( T . thermophilus and E . coli ) and V . radiata , the first transmembrane helix of NU2M spans the full length of the membrane . Furthermore , the loop connecting V . radiata’s NU2M TMH1-2 in the mitochondrial matrix is longer than in any of the other CI structures and extends into the matrix , where it contacts the N-terminal helix of NDUS8 discussed above . Given the universality of the hinging motion between CI’s peripheral and membrane arms , seen in the structures of several organisms ( Agip et al . , 2018; Letts et al . , 2019; Parey et al . , 2018 ) , the additional interaction surface formed by NDUS8 and NU2M in V . radiata CI is likely functionally relevant . Although the majority of the accessory subunits present in CI* have homologues in fungi and mammals ( opisthokonts ) , there are a number of notable differences . In the plant complex , the peripheral arm accessory subunit NDUS6 lacks an N-terminal domain that is seen in both the Y . lipolytica and mammalian structures ( Figure 2A ) . In Y . lipolytica , mammals and V . radiata , the C-terminal , Zn2+-containing domain of NDUS6 binds mainly to the core subunits NDUS1 , NDUS8 and NDUS2 at the interface of the N- and Q-modules . However , in opisthokonts , the N-terminal domain of NDUS6 binds to the Q-module at an additional site through contacts with the membrane-anchored NDUA9 accessory subunit ( Figure 2A ) . In order to bind across these two locations , NDUS6 in opisthokonts extends above the C-terminus of accessory subunit NDUA12 . This arrangement determines the order of assembly of these subunits in opisthokonts , as NDUA12 must be bound to the peripheral arm before the N-terminal domain of NDUS6 binds . However , due to the lack of the N-terminal domain in V . radiata’s NDUS6 , there is no interaction with NDUA9 nor traversing of the NDUA12 C-terminus . This difference has important implications for the assembly of CI in plants versus opisthokonts . In opisthokonts , the interaction between NDUS6 , NDUA12 and the NDUA12-homologous assembly factor NDUFAF2 establishes an important checkpoint for assembly of the peripheral arm . Thus , the lack of the NDUS6 N-terminus may in part explain observed differences between the assembly pathways of plant and opisthokont CI ( see Discussion ) . Other key differences can be seen on the intermembrane space side of the membrane arm in accessory subunits NDUA8 and NDUC2 . Compared to both Y . lipolytica and mammals , the double-CHCH domain of the PP-module NDUA8 subunit , which binds to the ‘heel’ of the complex on the intermembrane space ( Figure 2B ) , is C-terminally truncated in V . radiata . In the Y . lipolytica structure , the C-terminus of NDUA8 folds back onto itself with an additional α-helix , forming a bulkier subunit and a further interaction interface with the core transmembrane subunit NU1M . More interestingly , in mammals , the C-terminus of NDUA8 extends as a long coil halfway along the membrane arm and binds in a pocket between NU2M and NU4M at the interface of the PP-module and PD-module . The PP-module accessory subunit NDUC2 is also C-terminally truncated in V . radiata and Y . lipolytica relative to NDUC2 in mammals ( Figure 2C ) . In all mitochondrial CI structures to date , this subunit binds to the final transmembrane helix of the core NU2M subunit . However , in mammals , the NDUC2 C-terminus forms an extended coil on the intermembrane space side of the complex that extends along the membrane arm to interact with NDUB10 and NDUB11 , bridging the PP- and PD- modules . This bridging interaction is also present in Y . lipolytica via an extended loop on the PD-module core subunit NU4M . This pattern of truncated core and accessory subunits or missing interactions ( e . g . NDUS2 , NDUA8 and NDUC2; Table 4 ) in V . radiata relative to those in opisthokonts likely diminishes the stability of the attachment of PP-module to the PD-module , which may have consequences for CI’s function and assembly ( see Discussion ) . Compared to the mammalian and Y . lipolytica structures , two accessory subunits are absent from the Q-module in the CI* structure , namely the LYR-protein subunit NDUA6 and its accompanying acyl-carrier protein ( ACPM1 ) . The absence of the NDUA6 and ACPM1 subunits in CI* is notable given that , when the Y . lipolytica NDUA6 homologue is knocked out or mutated , this severely impacts the activity of the complex ( Angerer et al . , 2014 ) . Therefore , although it is not completely understood how NDUA6 modulates the activity of CI , the lack of NDUA6 in CI* may be a way to regulate the activity of the assembly intermediate . Although densities for NDUA6 and ACPM1 are absent in our CI* structure , density can be seen for a short α-helix bound under NDUS1 , where the C-terminus of NDUA6 binds in both the Y . lipolytica and mammalian structures . This suggests that NDUA6 may be bound to CI* via its C-terminus , without fully engaging with the complex . Although this would be surprising , the density for the amino acid sidechains in this region is consistent with the sequence of the NDUA6 C-terminus; thus , this density was modelled as such . If correct , this suggests that NDUA6 may be attached to the Q module but unable to fully bind to its main site on NDUS2 . V . radiata CI* does not have any plant-specific accessory subunits on the peripheral arm . Notwithstanding the unique features of NDUS6 and the absence of NDUA6 and ACPM1 discussed above , all of the V . radiata CI* N- and Q-module subunits have homologues in fungi and metazoans . However , this is not the case for the PP-module . Most notably , a large ( ~90 kDa ) hetero-trimeric γCA domain lies on top of the core membrane arm subunit NU2M ( Figure 1C ) . The identity of the components of the plant γCA has remained elusive , with different three-way combinations of the five plant γCA proteins proposed based on different genetic and biochemical studies ( Sunderhaus et al . , 2006; Perales et al . , 2004; Fromm et al . , 2016; Cï Rdoba et al . , 2019 ) . Our structure allowed us to unambiguously assign the identity of the subunits of the γCA domain despite high sequence identity between the five carbonic anhydrase proteins in plants . Based on unambiguous density for key non-conserved residues , we were able to definitively assign the three different subunits of V . radiata CI* as CA1 , CA2 and CA2L ( Figure 3A ) . The interaction surface between the γCA domain and the PP-module ( subunits NU2M , NDUC2 , P2 and NDUX1 ) is large , covering an approximate surface of 3 , 740 Å2 . As expected ( Sunderhaus et al . , 2006 ) , the γCA interacts with the PP-module tightly , with an approximate gain of solvation free energy of −210 kcal/mol , which is almost twice as large as the solvation energy gain of association of the γCA hetero-trimer itself ( Figure 3A , Table 5 ) . As has been previously demonstrated by proteomic analysis , the N-terminal mitochondrial signal pre-sequences for CA1 and CA2 remain uncleaved ( Klodmann et al . , 2010 ) . We show here that these two N-terminal sequences together form a short α-helical coiled-coil-like structure ( Figure 3C ) . This coiled coil is amphipathic and binds on the matrix surface of the inner mitochondrial membrane , contacting the NDUC2 and P2 subunits ( see below ) adjacent to the NU2M core subunit . In contrast , no density was observed for the N-terminal pre-sequence of CA2L , consistent with it being post-translationally cleaved ( Huang et al . , 2009 ) . The physiological role of the γCA domain on plant CI is unknown . Although recombinant mitochondrial γCA from plants has been shown to bind bicarbonate ( HCO3- ) , it remains unclear whether it exhibits enzymatic activity ( Martin et al . , 2009 ) . The canonical γCA trimer possesses three active sites , one at each interface between two protomers . Each active site is formed by three essential Zn2+-coordinating histidine residues . At each active site , two histidine residues are provided by one subunit and the third is provided by the adjacent subunit . However , in the plant CI γCA heterotrimer , the CA2L subunit is lacking two of the three essential histidine residues ( Ala-147 and Arg-152 in V . radiata ) that would be necessary to form active sites at the interfaces with the CA1 and CA2 subunits . This renders two of the possible three catalytic sites non-functional ( Figure 3A , Figure 3—figure supplement 1 ) . Furthermore , the V . radiata CA1 subunit is also missing one of the three Zn2+-coordinating histidine residues ( Gln-135 ) . Therefore , only one potentially catalytically active interface with all three Zn2+ coordinating residues remains in V . radiata’s γCA—namely , the site between CA1 and CA2 at the "top" ( most matrix-exposed periphery ) of the domain . Clear density for a Zn2+ can only be seen at this site ( Figure 3B ) . In contrast , no Zn2+ is seen at either of the two other sites , whose mutated residues are chemically incompatible with ion coordination . It is also important to note that the plant CA1 , CA2 and CAL2 proteins belong to the CamH subclass of γCAs , which lack the acidic loop containing the catalytically important ‘proton shuttle’ glutamate residue ( Glu89 in the canonical γCA from Methanosarcina thermophila ) ( Zimmerman et al . , 2010 ) . While some members of the CamH subclass are catalytically active , some are not ( Soto et al . , 2006; Jeyakanthan et al . , 2008 ) . Therefore , carbonic anhydrase activity of the γCA domain of CI must be confirmed experimentally ( Ferry , 2010 ) . The other plant-specific subunit we were able to assign in CI* was the single-transmembrane subunit P2 . This subunit binds on top of NDUX1 , adjacent to NU2M and directly underneath the γCA domain . The N-terminus of P2 interacts directly with the γCA domain in the matrix . Together , P2 , NDUX1 , NU2M and NDUC2 form a lipid-filled cavity positioned directly below the γCA domain ( Figure 3C and D ) . Several positively charged residues from the γCA domain subunits can be seen interacting with these lipids , demonstrating that this lipid pocket also forms an important part of the γCA domain/membrane arm interface . We were unable to assign four small regions of density in the CI* structure . One is the region near the N-terminus of NDUS8 discussed above ( Figure 4A ) . Another is the likely C-terminal helix of NDUA6 also discussed above ( Figure 4B ) . The third is on the intermembrane space side of the membrane arm ( Figure 4C ) . In both Y . lipolytica and mammalian CI , this binding site is occupied by the C-terminus of the PP- and PD-module-spanning subunit NDUB5 . In Y . lipolytica and mammals , NDUB5 spans nearly the entire length of the membrane arm . In V . radiata CI* , the density for this subunit follows the equivalent path of NDUB5 in Y . lipolytica and mammals but becomes disordered by the PP-module's core subunit NU2M , which is adjacent to the C-terminus of accessory subunit NDUC2 . The final stretch of unassigned density is for a single-transmembrane accessory subunit bound above NU6M TMH1 that contacts NU6M and NDUS5 on the intermembrane space side of the membrane arm ( Figure 4D ) . This unassigned subunit protrudes away from CI* toward the location where CIII2 binds in the mammalian supercomplex I+III2 ( Letts and Sazanov , 2015 ) , suggesting a possible role for this subunit in supercomplex formation . No equivalent subunit is seen in either Y . lipolytica or mammalian CI , suggesting that this is a plant-specific subunit . However , due to local disorder , the density was too poor to assign the sequence from the reconstruction alone . All the cofactors necessary for the transfer of electrons between NADH and CoQ are present in the CI* intermediate . This includes the flavin mononucleotide ( FMN ) in NDUV1 , all seven FeS-clusters of the main electron transport pathway ( N3[V1] , N1b[S1] , N4[S1] , N5[S1] , N6a[S8] , N6b[S8] , N2[S7] ) , and the off-pathway FeS cluster N1a[V2] ( Figure 5A ) . Moreover , density can be seen in the cryoEM map in the region of the Q-tunnel , in an equivalent position to that of CoQ in the Y . lipolytica structure ( Parey et al . , 2019; Figure 5B ) . This likely represents a CoQ molecule bound at the entry of the CI* Q-tunnel . However , this density is indistinct and thus we have not modeled a CoQ at this position . Analogously to the Y . lipolytica structure , no density for CoQ can be seen deeper in the Q-tunnel where CoQ would need to bind to accept electrons from the terminal FeS cluster ( Figure 5B ) . The loops that cap the Q-tunnel at the interface of the peripheral and membrane arms of the complex , namely the NU3M TMH1-2 and NU1M TMH5-6 loops , are disordered . This is analogous to what is observed in the open or deactive structures of the mammalian and Y . lipolytica complexes ( Agip et al . , 2018; Letts et al . , 2019; Parey et al . , 2018 ) . Conformational changes in these loops are thought to play an important role in CI’s coupling mechanism , which transduces the energy of NADH-quinone oxidoreduction in the Q module to proton pumping along the membrane arm ( Parey et al . , 2018; Cabrera-Orefice et al . , 2018 ) . In particular , a π-bulge in NU6M’s TMH3 in mammals has been seen to undergo a major conformational change , refolding into an α-helix during complex I’s open-to-closed transition ( Agip et al . , 2018; Letts et al . , 2019 ) . This π-bulge in NU6M’s TMH3 is also present in V . radiata CI* . The ‘E-channel’ ( Baradaran et al . , 2013 ) and the hydrophilic axis of polar amino acid residues that are involved in proton translocation and span the membrane arm of CI are also evident in V . radiata CI* ( Figure 5C ) . Given the lack of additional accessory subunits or assembly factors to cap the end of CI*’s shortened membrane arm , hydrophilic-axis residue Lys399 on NU2M’s TMH12 is exposed to the midplane of the membrane . In all other structures of CI , the final transmembrane core subunit NU5M contains a transmembrane helix ( TMH15 ) that caps the hydrophilic axis at the end of the transmembrane arm of full-length CI . The lack of such a cap on V . radiata NU2M in CI* suggests that , although Lys399 of NU2M is mostly surrounded by protein , the core hydrophilic axis may be in contact with lipid .
The structure of V . radiata CI* presented here is the first atomic resolution structure of any plant mitochondrial electron transport chain complex and reveals several key features of mitochondrial CI from vascular plants . CI* is an established assembly intermediate of plant CI , previously identified with genetic and proteomic studies in non-etiolated seedlings and mature leaves of A . thaliana and N . sylvestris ( Ligas et al . , 2019; Meyer et al . , 2011; Schertl et al . , 2012; Schimmeyer et al . , 2016; Senkler et al . , 2017; Pineau et al . , 2008 ) . Furthermore , CI* exhibits NADH-dehydrogenase activity in in-gel activity assays ( Meyer et al . , 2011; Pineau et al . , 2008; Haïli et al . , 2013 ) . Thus , it is unlikely that CI* in our mitochondrial preparations is a peculiarity of our etiolating growth conditions or our choice of model organism . Nevertheless , it may be the case that etiolating conditions promote the accumulation of CI* in V . radiata hypocotyls compared to seedlings grown in the light ( see Appendix ) . Moreover , it is also unlikely that CI* is a degradation product of CI rather than the assembly intermediate . Firstly , our membrane-extraction conditions ( 1% w:v digitonin , 4:1 g:g detergent:protein; see Materials and methods ) are very gentle and were chosen after optimization to preserve protein:protein interactions in protein complexes and supercomplexes . Furthermore , immediately after extraction , we stabilize the detergent-extracted complexes with amphipathic polymers , which wrap around the complexes and further protect them from degradation/dissociation ( Breibeck and Rompel , 2019 ) . A large section of membrane stabilized and co-purified by our gentle digitonin/amphipol treatment is clearly seen around the perimeter of CI* at low contour ( Figure 1—figure supplement 4E ) . Secondly , using digitonin at a higher concentration ( 5% w:v ) , an A . thaliana complexome profiling study ( Senkler et al . , 2017 ) obtained not only full-length CI and CI* , but also full-length CI in a higher order assembly with complex III ( supercomplex SC I+III2 ) [Bultema et al . , 2009; Dudkina et al . , 2005; Eubel et al . , 2004a; Eubel et al . , 2004b] . Protein:protein interactions between complexes in supercomplexes are known to be more labile than intra-complex protein:protein interactions . Given that the more fragile CI:CIII2 interactions are maintained in 5% digitonin ( Senkler et al . , 2017 ) , this argues that the presence of CI* —both in Senkler et al . , 2017 and in this study— is not due to a digitonin-induced dissociation of the PD domain , but rather that it is the true assembly intermediate . Thirdly , controlled-degradation experiments of plant CI in the presence of harsh detergents have shown that , analogous to mammalian CI , plant CI’s detergent-induced dissociation occurs via detachment of the full peripheral arm ( PP-PD ) from the full matrix arm ( N-Q ) ( Klodmann et al . , 2010 ) , not by dissociation between the PP and PD modules . Fourthly , we have reproducibly obtained the CI* fraction , which retains its in-gel and spectroscopic NADH-oxidase activity and chromatographic peak for several days , even after freeze/thaw cycles . For these reasons , it is evident that our structure corresponds to the CI* assembly intermediate , rather than to a degradation product of V . radiata CI . A major unique feature of plant CI compared to the other known structures is the large γCA domain located on the mitochondrial matrix side of the membrane arm of the complex ( Sunderhaus et al . , 2006 ) . Here , we were able to define the interface and anchoring interactions between the γCA domain and the rest of the complex at high resolution ( Figure 3 ) . In line with expectations from the early biochemical experiments on the plant γCA domain ( Sunderhaus et al . , 2006 ) , the structure clearly shows that the interface between the γCA domain and the PP-module is extensive and strong ( Table 5 ) . Additionally , we established that the γCA domain is membrane-targeted via two amphipathic helices that contact the CI membrane arm and through specific interactions with lipids in a lipid-filled pocket formed by core subunit NU2M , accessory subunits NDUX1 , NDUC2 and plant-specific accessory subunit P2 . Furthermore , our structure unambiguously resolves the identities of the hetero-trimeric components of the γCA domain of etiolated V . radiata as CA1 , CA2 and CA2L . Unexpectedly , our structure also reveals that , due to this composition , only one out of the three potential active sites formed at the interfaces between CA1 , CA2 and CA2L is capable of coordinating the Zn2+ ion required for carbonic anhydrase catalysis . Nevertheless , whether the combination of γCA subunits and , consequently , the active site arrangements are different in different species , tissues or developmental stages ( Sunderhaus et al . , 2006; Perales et al . , 2004; Fromm et al . , 2016; Cï Rdoba et al . , 2019 ) remains to be confirmed . Structure alone is not sufficient to demonstrate catalytic ability of the plant CI γCA domain . Indeed , only bicarbonate binding to the plant mitochondrial γCAs has been shown ( Martin et al . , 2009 ) and , despite extensive attempts , no catalytic activity has been measured to date ( Fromm et al . , 2016; Martin et al . , 2009 ) . Further functional and structural studies with purified CI or CI* samples are necessary to determine whether the γCA domain possesses enzymatic activity . Less is known about CI assembly in plants than in fungi or metazoans ( opisthokonts ) . In metazoans , detailed models of CI assembly have been generated and over a dozen CI assembly factors have been identified ( Formosa et al . , 2018; Guerrero-Castillo et al . , 2017; Garcia et al . , 2017 ) . In plants , only three assembly factors have been thus far identified: L-galactono-1 , 4-lactone dehydrogenase ( GLDH ) ( Senkler et al . , 2017 ) , the FeS protein INDH ( Wydro et al . , 2013 ) and an LYR protein termed CIAF1 ( Ivanova et al . , 2019 ) . One possibility is that some of the unassigned densities observed in our reconstruction correspond to assembly factors that are bound to CI* . Current models of plant CI biogenesis predict that , of these three , only GLDH should be bound to the CI* intermediate ( Ligas et al . , 2019 ) . However , GLDH is a large ( ~60 kDa ) globular enzyme ( Leferink et al . , 2008 ) , for which we do not see any consistent density in our structure . Nonetheless , it is possible that GLDH is bound via a flexible loop and thus averaged out in our reconstructions . Further assembly factors have been predicted to bind and cap NU2M in the membrane ( Ligas et al . , 2019 ) . However , as noted above , we do not observe any additional transmembrane subunits capping the end of the shortened transmembrane arm . There are major differences in CI assembly between plants and metazoans ( Figure 1—figure supplement 1 ) . In metazoans , the N-module ( responsible for NADH oxidation ) is assembled onto the Q- , PP- and PD-modules last ( Formosa et al . , 2018; Guerrero-Castillo et al . , 2017; Garcia et al . , 2017 ) . This ensures that no assembly intermediate is capable of transferring electrons from NADH to CoQ . In contrast , in plants the final assembly step is the attachment of the PD-module onto the CI* intermediate ( Ligas et al . , 2019 ) . As noted above , the V . radiata CI* intermediate contains all of the subunits and co-factors needed to carry out NADH:CoQ oxidoreduction: CI* is , in principle , catalytically competent . Indeed , we were able to measure NADH-DQ oxidoreductase activity in the isolated CI* fraction ( Figure 1—figure supplement 2 ) . The V . radiata CI* structure presented here reveals that this difference in assembly may in part stem from a significant difference in the structure of the peripheral-arm accessory subunit NDUS6 . The plant NDUS6 subunit lacks an N-terminal domain relative to the NDUS6 homologues of opisthokonts . In opisthokonts , the N-terminal domain of NDUS6 binds over top of NDUA12 to interact with the Q-module accessory subunit NDUA9 ( Figure 2A ) . Moreover , the assembly factor NDUFAF2 –a paralogue of NDUA12 that occupies the same binding site—sterically prevents the binding of NDUS6 ( Parey et al . , 2019 ) . Thus , in opisthokonts , NDUFAF2 must be removed and replaced with NDUA12 before NDUS6 can bind on the peripheral arm to complete the assembly of CI . In plants , a NDUFAF2 homologue on CI has yet to be observed experimentally ( Meyer et al . , 2019 ) . Additionally , due to the lack of the N-terminal domain on NDUS6 , plant NDUS6 does not cross over NDUA12 but binds next to it on the surface of the peripheral arm . Thus , in plants , NDUS6 may assemble on CI independent of the status of NDUFAF2/NDUA12 . Furthermore , attaching the N-module before the PD-module in plants may provide additional flexibility to their mitochondrial ETC ( see discussion below and Appendix ) . It is clear from the currently available structures that the interface between the PP-module and PD-module is more extensively stabilized by accessory subunit interactions in mammals than in Y . lipolytica or V . radiata ( Table 4 ) . Although we currently only have the structure of the CI* intermediate for V . radiata ( which only contains the PP-module ) , key truncations in core subunit NDUS2 and accessory subunits NDUA8 and NDUC2 , discussed above ( Figure 2B and C ) , already make this distinction clear . The lack of the NDUA8 and NDUC2 bridging interactions suggest that the interface between the PP- and PD-modules in plants may be weaker , which may also help explain the differences in the CI assembly pathway in plants versus opisthokonts . Identification of other possible bridging interactions across the PP- and PD-modules in plants will have to await the structure of full-length plant CI . The bioenergetic regulation of plants , which generate their energy through respiration and photosynthesis , is more intricate and dynamic than that of heterotrophs , whose main bioenergetic process is respiration . Mitochondrial respiration is the major source of ATP in plants’ non-photosynthetic tissues such as roots . In photosynthetic tissue in the light , the role of mitochondrial respiration in ATP production is debated ( Shameer et al . , 2019; Gardeström and Igamberdiev , 2016 ) ( see Appendix ) . Moreover , in photosynthetic tissue , conditions of intense light may lead to an over-production of reducing equivalents ( NAD ( P ) H ) , which could be detrimental to the cells via the production of reactive oxygen species ( ROS ) . To mitigate this , the plant mitochondrial electron transport chain ( mETC ) contains several ‘alternative’ oxidoreductases and oxidases that shunt electrons to molecular oxygen without pumping H+ , thus preventing the over-reduction of the NADH pool ( Millar et al . , 2011; Schertl and Braun , 2014 ) . However , given that alternative complexes do not pump any H+ , energy is instead dissipated as heat . Based on the fact that CI* is missing two of its four standard H+ pumps ( those in the PD module ) , and on our finding that CI* shows NADH-DQ oxidoreduction activity ( Figure 1—figure supplement 2 ) , we hypothesize that CI* may be an NADH-CoQ oxidoreductase with a lower H+-pumping-to-electron-transfer ratio than full-length CI . Namely , we hypothesize that CI* could pump protons at a 2H+:2e- ratio rather than the 4H+:2e- of full-length CI ( Jones et al . , 2017 ) . Decreased H+:e- ratios have previously been reported in functional yeast and bacterial CI mutants ( Dröse et al . , 2011; Steimle et al . , 2011 ) . A mutant of Y . lipolytica CI in which the PD-module accessory subunit NB8M ( homologue of plant NDUB7 ) is deleted ( nb8mΔ ) fails to assemble the PD-module ( Dröse et al . , 2011 ) . The resulting CI subcomplex is analogous to CI* , as it lacks only the PD-module . The nb8mΔ mutant CI is a functional H+-pumping NADH-CoQ oxidoreductase . However , its H+:e- ratio , which is normally 4H+:2e- in fully assembled CI , is reduced to 2H+:2e- ( Dröse et al . , 2011 ) . This is consistent with two of the four H+-pumping subunits ( NU4M and NU5M ) being absent in the nb8mΔ mutant subcomplex . Similar results are seen in E . coli mutants with mutations in its distal H+-pumping subunit NuoL ( homologue of plant NU5M ) . Deletion of NuoL or truncation of its transmembrane helices 15–16 , which bridge the PP and PD modules , result in a functional CI mutant whose H+:e- coupling is 2H+:2e- ( Steimle et al . , 2011 ) . We hypothesize that a lower-H+-pumping CI* could provide additional flexibility to plants’ bioenergetic regulation , beyond the interplay between the canonical and alternative pathways of the mETC . For instance , having a 2H+:2e- ratio would allow CI* to contribute to ATP generation in situations where the mitochondrial [NAD+]/[NADH] ratio would not support H+ pumping by CI ( see Appendix for an in-depth discussion ) . Thus , CI* may provide additional energy-converting flexibility to plants’ electron flow and energy conservation . This would be analogous to the flexibility seen for the electron transport chain of chloroplasts , which employ several dynamic mechanisms at different levels of regulation to adjust the H+:e- coupling and the energetic and redox outputs to changing environmental conditions ( Heber and Kirk , 1975; Scheibe et al . , 2005; Rochaix , 2011; Murchie and Ruban , 2020 ) . Here , we present the structure of a mitochondria CI assembly intermediate , CI* , isolated from etiolated hypocotyls of V . radiata . CI* showed NADH-dehydrogenase activity in native in-gel and spectroscopic activity assays . Although we did not introduce experimental manipulations to prevent the assembly of mitochondrial CI , we were nonetheless able to isolate sufficient amounts of the CI* assembly intermediate for structure determination . This suggests that there are significant steady-state amounts of CI*in V . radiata mitochondria under these etiolating conditions and that CI* may be playing an independent physiological function beyond its role in CI assembly . The structure of V . radiata CI* presented here provides a wealth of information on mitochondrial CI composition , assembly and evolution and raises several questions on the dynamics and regulation of plant respiration . In order to address these questions , further research is needed into the structures of the fully assembled plant mitochondrial CI , as well as of its supercomplex with CIII2 . In addition , biochemical , cell biological and genetic approaches are paramount to test hypotheses on the potential functions of CI* .
V . radiata seeds were purchased from Todd’s Tactical Group ( Las Vegas , NV ) . Seeds were incubated in 1% ( v:v ) bleach for 20 min and rinsed until the water achieved neutral pH . Seeds were subsequently imbibed in a 6 mM CaCl2 solution for 20 hr in the dark . The following day , the imbibed seeds were sown in plastic trays on damp cheesecloth layers , at a density of 0 . 1 g/cm2 and incubated in the dark at 20°C for 6 days . The resulting etiolated mung beans were manually picked , and the hypocotyls were separated from the roots and cotyledons . The hypocotyls were further processed for mitochondria purification based on established protocols ( Millar et al . , 2007 ) . Briefly , hypocotyls were homogenized in a Waring blender with homogenization buffer ( 0 . 4 M sucrose , 1 mM EDTA , 25 mM MOPS-KOH , 10 mM tricine , 1% w:v PVP-40 , freshly added 8 mM cysteine and 0 . 1% w:v BSA , pH 7 . 8 ) before a centrifugation of 10 min at 1000 x g ( 4°C ) . The supernatant was collected and centrifuged for 30 min at 12 , 000 x g ( 4°C ) . The resulting pellet was resuspended with wash buffer ( 0 . 4 M sucrose , 1 mM EDTA , 25 mM MOPS-KOH , freshly added 0 . 1% w:v BSA , pH 7 . 2 ) and gently centrifuged at 1000 x g for 5 min ( 4°C ) . This supernatant was then centrifuged for 45 min at 12 , 000 x g . The resulting pellet was resuspended in wash buffer , loaded on to sucrose step gradients ( 35% w:v , 55% w:v , 75% w:v ) and centrifuged for 60 min at 52 , 900 x g . The sucrose gradients were fractionated with a BioComp Piston Gradient Fractionator ( Fredericton , Canada ) connected to a Gilson F203B fraction collector , following absorbance at 280 nm . The fractions containing mitochondria were pooled , diluted 1:5 in 10 mM MOPS-KOH , 1 mM EDTA , pH 7 . 2 and centrifuged for 20 min at 12 , 000 x g ( 4°C ) . The pellet was resuspended in final resuspension buffer ( 20 mM HEPES , 50 mM NaCl , 1 mM EDTA , 10% glycerol , pH 7 . 5 ) and centrifuged for 20 min at 16 , 000 x g ( 4°C ) . The supernatant was removed , and the pellets were frozen and stored in a −80°C freezer . The yield of these mitochondrial pellets was 0 . 8–1 mg per gram of hypocotyl . Frozen V . radiata mitochondrial pellets were thawed at 4°C , resuspended in 10 ml of chilled ( 4°C ) double-distilled water per gram of pellet and homogenized with a cold Dounce glass homogenizer . Chilled KCl was added to the homogenate to a final concentration of 0 . 15 M and further homogenized . The homogenate was centrifuged for 45 min at 32 , 000 x g ( 4°C ) . The pellets were resuspended in cold Buffer M ( 20 mM Tris , 50 mM NaCl , 1 mM EDTA , 2 mM DTT , 0 . 002% PMSF , 10% glycerol , pH 7 . 4 ) and further homogenized before centrifugation at 32 , 000 x g for 45 min ( 4°C ) . The pellets were resuspended in 3 ml of Buffer M per gram of starting material and further homogenized . The protein concentration of the homogenate was determined using a Pierce BCA assay kit ( Thermo Fisher , Waltham , MA ) , and the concentration was adjusted to a final concentration of 10 mg/ml and 30% glycerol . Washed membranes were thawed at 4°C . Digitonin ( EMD Millipore , Burlington , MA ) was added to the membranes at a final concentration of 1% ( w:v ) and a digitonin:protein ratio of 4:1 . Membranes complexes were extracted by tumbling this mixture for 60 min at 4°C . The extract was centrifuged at 16 , 000 x g for 45 min ( 4°C ) . Amphipol A8-35 ( Anatrace , Maumee , OH ) was added to the supernatant at a final concentration of 0 . 2% w:v and tumbled for 30 min at 4°C , after which gamma-cyclodextrin ( EMD Millipore , Burlington , MA ) was added to a final amount of 1 . 2x gamma-cyclodextrain:digitonin ( mole:mole ) . The mixture was centrifuged at 137 , 000 x g for 60 min ( 4°C ) . The supernatant was concentrated with centrifugal protein concentrators ( Pall Corporation , NY , NY ) of 100 , 000 MW cut-off , loaded onto 10–45% ( w:v ) or 15–45% ( w:v ) linear sucrose gradients in 15 mM HEPES , 20 mM KCl , pH 7 . 8 produced using factory settings of a BioComp Instruments ( Fredericton , Canada ) gradient maker and centrifuged for 16 hr at 37 , 000 x g ( 4°C ) . The gradients were subsequently fractionated with BioComp Piston Fractionatr connected to a Gilson F203B fraction collector , following absorbance at 280 nm . Select fractions were pooled , concentrated with protein concentrators ( Pall Corporation , NY , NY ) of 100 , 000 MW cut-off and purified on a Superose6 10/300 chromatography column ( GE Healthcare , Chicago , IL ) using an NGC 10 Medium-Pressure chromatography system ( Biorad , Hercules , CA ) . For grid preparation , the relevant fractions were buffer-exchanged into 20 mM HEPES , 150 mM NaCl , 1 mM EDTA , pH 7 . 8 ( no sucrose ) and concentrated to a final protein concentration of 6 mg/ml and mixed one-to-one with the same buffer containing 0 . 2% digitonin ( w:v ) , for a final concentration of 0 . 1% digitonin ( w:v ) . Mitochondrial membrane extractions were diluted in 2X BN-loading buffer ( 250 mM aminocaproic acid , 100 mM Tris-HCl , pH 7 . 4 , 50% glycerol , 2 . 5% ( w:v ) Coomassie G-250 ) , loaded on pre-cast 3–12% NativePAGE Bis-Tris gels ( Invitrogen , Carlsbad , CA ) and run at 4°C . The cathode buffer was 50 mM Tricine , 50 mM BisTris-HCl , pH 6 . 8 plus 1X NativePAGE Cathode Buffer Additive ( 0 . 02% Coomassie G-250 ) ( Invitrogen , Carlsbad , CA ) and the anode buffer was 50 mM Tricine , 50 mM BisTris-HCl , pH 6 . 8 . Gels were run at 200 V constant voltage for ∼30 min , after which the cathode buffer was switched for a ‘light blue’ cathode buffer containing 50 mM Tricine , 50 mM BisTris-HCl , pH 6 . 8 plus 0 . 1X NativePAGE Cathode Buffer Additive ( 0 . 002% Coomassie G-250 ) ( Invitrogen , Carlsbad , CA ) . The settings were changed to 7 mA constant amperage and run for another ∼90 min . The CI in-gel NADH dehydrogenase activity assay was performed based on Schertl and Braun , 2015 . The BN-PAGE gel was incubated in 10 ml of freshly prepared reaction buffer ( 1 mg/ml nitrotetrazoleum blue in 10 mM Tris-HCl pH 7 . 4 ) . Freshly thawed NADH was added to the container with the gel , to a final concentration of 150 μM . The gel with the complete reaction buffer was rocked at room temperature for ∼10 min . Once purple bands indicating NADH-dehydrogenase activity appeared , the reaction was quenched with a solution of 50% methanol ( v:v ) and 10% acetic acid ( v:v ) . The spectroscopic NADH dehydrogenase activity assay was performed based on Huang et al . , 2015; Letts et al . , 2019 . CI NADH:decylubiquinone ( DQ ) activity was measured by spectroscopic observation of NADH oxidation at 340 nm wavelength at 30°C using a Molecular Devices ( San Jose , CA ) Spectramax M2 spectrophotometer . Reactions were carried out in 96-well plates . Protein samples were added to 190 μL of reaction buffer ( 100 mM HEPES , pH 7 . 4 , 50 mM NaCl , 10% glycerol , 4 μM KCN , 1 mg/ml BSA , 10 μM cyt c , with or without 100 μM DQ as required ) and mixed by pipetting . The reaction was initiated by addition of NADH to a final concentration of 150 μM and briefly mixed by pipetting and plate stirring for 10 s before recording . Measurements were done in triplicate , averaged and background-corrected . The known extinction co-efficient of NADH ( 6 . 22 mM−1 cm−1 ) was used in the calculations . Statistical significance was determined using a two-tailed t-test . The CI* sample ( 6 mg/ml protein in 20 mM HEPES , 150 mM NaCl , 1 mM EDTA , 0 . 1% digitonin , pH 7 . 8 ) was applied onto glow-discharged holey carbon grids ( Quantifoil , 1 . 2/1 . 3 300 mesh ) followed by a 60 s incubation and blotting for 9 s at 15°C with 100% humidity and flash-freezing in liquid ethane using a FEI Vitrobot Mach III . CryoEM data acquisition was performed on a 300 kV Titan Krios electron microscope equipped with an energy filter and a K3 detector at the UCSF W . M . Keck Foundation Advanced Microscopy Laboratory , accessed through the Bay Area Cryo-EM Consortium . Automated data collection was performed with the SerialEM package ( Schorb et al . , 2019 ) . Micrographs were recorded at a nominal magnification of 60 , 010 X , resulting in a pixel size of 0 . 8332 Å2 . Defocus values varied from 1 . 5 to 3 . 0 µm . The dose rate was 20 electrons per pixel per second . Exposures of 3 s were dose-fractionated into 118 frames , leading to a dose of 0 . 72 electrons per Å2 per frame and a total accumulated dose of 86 . 4 electrons per Å2 . A total of 9816 micrographs were collected , 8541 of which were used for further analysis . Software used in the project was installed and configured by SBGrid ( Morin et al . , 2013 ) . All processing steps were done using RELION 3 . 0 ( Zivanov et al . , 2018 ) unless otherwise stated . Motioncor2 ( Zheng et al . , 2017 ) was used for whole-image drift correction of each micrograph . Contrast transfer function ( CTF ) parameters of the corrected micrographs were estimated using Ctffind4 ( Rohou and Grigorieff , 2015 ) and refined locally for each particle in RELION . Automated particle picking using crYOLO ( Wagner et al . , 2019; Wagner and Raunser , 2020 ) resulted in ~1 . 5 million particles . The particles were extracted using 4002 pixel box binned two-fold and sorted by reference-free 2D classification followed by re-extraction at 5122 pixel box . Reference-free 2D classification resulted in the identification of 190 , 951 CI* particles . An ab initio model was generated in RELION from these particles ( Punjani et al . , 2017 ) . This model , lowpass-filtered at 30 Å , was used for initial 3D classification with a regularization parameter T of 4 . This initial processing resulted in ~34 , 000 particles of good quality , which separated into a single class ( Figure 1—figure supplement 3C ) . The best class was refined to a nominal resolution of 3 . 9 Å according to the gold standard FSC criteria ( Scheres and Chen , 2012 ) . It was clear that the local resolution of this refinement was impacted by hinge-like motions between the membrane and peripheral arms of the complex . Therefore , sub-region refinements were also performed masking around the membrane arm and peripheral arm , respectively ( Figure 1—figure supplement 3C ) . This resulted in significantly , improved map quality , especially for the γCA domain on the membrane arm ( Figure 1—figure supplement 3C ) . These improved maps were used for model building and refinement . The two focused refined maps were then combined into a composite map using Phenix . Starting models for isolated ovine CI ( Letts and Sazanov , 2015 ) and bacterial γCA ( Iverson et al . , 2000 ) , corrected for the V . radiata sequence , were used as templates . Additionally , starting models were generated using the Phyre2 web portal ( Kelley et al . , 2015 ) . These models were split and fit into the highest-resolution focused refinement maps for separate atomic model building of the CI* peripheral arm and CI* membrane arm in Coot ( Emsley and Cowtan , 2004 ) . Real-space refinement of the model was done in PHENIX ( Liebschner et al . , 2019; Goddard et al . , 2018; Pettersen et al . , 2004 ) and group atomic displacement parameters ( ADPs ) were refined in reciprocal space . The single cycle of group ADP refinement was followed by three cycles of global minimization , followed by an additional cycle of group ADP refinement and finally three cycles of global minimization ( Letts et al . , 2019 ) . Molecular graphics and analyses were performed with UCSF Chimera ( Pettersen et al . , 2004 ) , developed by the Resource for Biocomputing , Visualization , and Informatics at the University of California , San Francisco , with support from NIH P41-GM103311 , as well as the PyMOL Molecular Graphics System , Version 2 . 0 Schrödinger , LLC . | Respiration is the process used by all forms of life to turn organic matter from food into energy that cells can use to live and grow . The final stage of this process relies on an intricate chain of protein complexes which produce the molecule that cells use for energy . Complexes in the chain are made up of specific proteins that are carefully assembled , often into discrete modules or intermediate complexes , before coming together to form the full protein complex . Understanding how these complexes are assembled provides important insights into how respiration works . The precise three-dimensional structure of these complexes has been identified for bacteria , yeast and mammals . However , less is known about how these respiration complexes form in plants . For this reason , Maldonado et al . studied the structure of an intermediate complex that is only found in plants , called Cl* . This intermediate structure goes on to form complex I – the largest complex in the respiration chain . A technique called cryo-electron microscopy was used to obtain a structure of Cl* at a near-atomic level of detail . This structure revealed how the proteins that make up Cl* fit together , highlighting differences and similarities in how plants assemble complex I compared to bacteria , yeast and mammals . Maldonado et al . also studied the activity of Cl* , leading to the suggestion that this complex may be more than just a stepping stone towards building the full complex I and could have its own role in the cell . The structure of this complex provides new insights into the respiration mechanism of plants and could help scientists improve crop production . For instance , new compounds may be able to block respiration in pests , while leaving the crop unharmed; or genetic modifications could create plants that respire more efficiently in different environments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2020 | Atomic structure of a mitochondrial complex I intermediate from vascular plants |
Sas-6 and Ana2/STIL proteins are required for centriole duplication and the homo-oligomerisation properties of Sas-6 help establish the ninefold symmetry of the central cartwheel that initiates centriole assembly . Ana2/STIL proteins are poorly conserved , but they all contain a predicted Central Coiled-Coil Domain ( CCCD ) . Here we show that the Drosophila Ana2 CCCD forms a tetramer , and we solve its structure to 0 . 8 Å , revealing that it adopts an unusual parallel-coil topology . We also solve the structure of the Drosophila Sas-6 N-terminal domain to 2 . 9 Å revealing that it forms higher-order oligomers through canonical interactions . Point mutations that perturb Sas-6 or Ana2 homo-oligomerisation in vitro strongly perturb centriole assembly in vivo . Thus , efficient centriole duplication in flies requires the homo-oligomerisation of both Sas-6 and Ana2 , and the Ana2 CCCD tetramer structure provides important information on how these proteins might cooperate to form a cartwheel structure .
Centrioles are complex microtubule ( MT ) based structures that are required for the formation of centrosomes and cilia/flagella . These organelles have many important functions in cells , and their dysfunction has been linked to a plethora of human pathologies , ranging from cancer to microcephaly to obesity ( Nigg and Raff , 2009; Bettencourt-Dias et al . , 2011 ) . Thus , understanding how these organelles assemble and function is an important goal of both basic and biomedical research . Although several hundred proteins are thought to be concentrated at centrioles , only a small number appear to form a conserved ‘core’ pathway that is essential for centriole assembly ( Delattre et al . , 2006; Pelletier et al . , 2006; Gönczy , 2012 ) . During canonical centriole duplication , the protein kinase Plk4/Sak/ZYG-1 is recruited to the mother centriole by SPD-2 in worms ( Delattre et al . , 2006; Pelletier et al . , 2006; Shimanovskaya et al . , 2014 ) , by Asterless ( Asl ) in flies ( Blachon et al . , 2008; Dzhindzhev et al . , 2010 ) , or by a combination of the two ( Cep192 and Cep152 , respectively ) in humans ( Cizmecioglu et al . , 2010; Hatch et al . , 2010; Kim et al . , 2013; Sonnen et al . , 2013 ) . The protein kinase recruits STIL/Ana2/SAS-5 and Sas-6 to a single site on the side of the mother centriole where they assemble with CPAP/Sas-4 into a cartwheel structure that helps to establish the ninefold symmetry of the centriole ( Dammermann et al . , 2004; Delattre et al . , 2004; Leidel et al . , 2005; Nakazawa et al . , 2007; Peel et al . , 2007; Strnad et al . , 2007; Stevens et al . , 2010a; Tang et al . , 2011; Arquint et al . , 2012 ) . CPAP/Sas-4 can interact with tubulin ( Hung et al . , 2004 ) and is required to recruit the centriole MTs to the outer region of the cartwheel ( Pelletier et al . , 2006 ) , possibly working together with Cep135/Bld10 ( Hiraki et al . , 2007; Lin et al . , 2013 ) —although no homologue of this protein has been identified in worms , and it does not appear to be essential for centriole duplication in flies ( Carvalho-Santos et al . , 2012; Mottier-Pavie and Megraw , 2009; Roque et al . , 2012 ) . Great progress has been made recently in understanding how these proteins interact and how these interactions are regulated to ensure that a new centriole is only formed at the right place and at the right time . In particular , the crystal structure of Sas-6 from several species has revealed how this protein forms a dimer through its C-terminal coiled-coil domain ( C–C ) that can then further homo-oligomerise through an N-terminal headgroup interaction ( N–N ) to form a ring structure from which the C–C domains emanate as spokes ( Kitagawa et al . , 2011; van Breugel et al . , 2011 , 2014; Hilbert et al . , 2013 ) . This Sas-6 ring structure can be modelled into EM tomographic reconstructions of the cartwheel from Trichonympha centrioles ( Guichard et al . , 2012 , 2013 ) , strongly suggesting that these Sas-6 rings form the basic building blocks of the cartwheel . In support of this hypothesis , mutant forms of Sas-6 that cannot homo-oligomerise through the N–N interaction are unable to support efficient centriole duplication ( Kitagawa et al . , 2011; van Breugel et al . , 2011 ) , although they can still target to centrioles , a function that seems to rely on the C–C domain ( Fong et al . , 2014; Keller et al . , 2014 ) . A crystal structure of the interface between Ana2/STIL and Sas-4/CPAP has also recently been solved ( Cottee et al . , 2013; Hatzopoulos et al . , 2013 ) , as has the interaction interface between Plk4 and both Cep192/SPD-2 and Cep152/Asl ( Park et al . , 2014 ) ; mutations that perturb these interactions in vitro perturb centriole duplication in vivo , indicating that these interactions are also essential for centriole duplication . More recently , it has been shown that Plk4 can recruit STIL to centrioles in human cells ( Ohta et al . , 2014; Kratz et al . , 2015 ) and that Plk4/Sak can phosphorylate the conserved STIL/Ana2 ( STAN ) domain in STIL/Ana2 proteins in humans and flies , thereby promoting the interaction of the STAN domain with Sas-6 ( Dzhindzhev et al . , 2014; Ohta et al . , 2014; Kratz et al . , 2015 ) . Mutant forms of STIL/Ana2 that could not be phosphorylated strongly perturbed Sas-6 recruitment to centrioles and centriole duplication . Together , these studies have shed important light on the molecular mechanisms of centriole assembly , but many important questions remain . In particular , it has been proposed that the homo-oligomerisation properties of Sas-6 establish the ninefold symmetry of the centriole ( Kitagawa et al . , 2011 ) , and , remarkably , a ninefold symmetric ring structure is formed in crystallo by Leishmania major Sas-6 ( van Breugel et al . , 2014 ) . However , although Sas-6 oligomers appear to have a propensity towards ninefold symmetry , Sas-6 proteins spontaneously assemble into oligomers of varying stoichiometry in vitro ( Kitagawa et al . , 2011; van Breugel et al . , 2011 ) , suggesting that the homo-oligomerisation properties of Sas-6 alone may be insufficient to enforce the rigorous ninefold symmetry that is observed in centrioles from virtually all species ( Cottee et al . , 2011 ) . Additionally , recent Cryo-EM analysis suggests that the basic building block of the cartwheel stack is not a single ring and spoke structure , but rather a pair of rings that sit on top of one another: these rings do not make direct contact with each other , but are joined in the more peripheral regions through their spokes ( Guichard et al . , 2012 , 2013 ) . Our current knowledge of Sas-6 self-association cannot explain this important feature of the cartwheel structure . We previously showed that overexpressed Sas-6 can form higher-order aggregates in Drosophila spermatocytes , but these aggregates only adopt a cartwheel-like structure when Ana2 is also overexpressed ( Stevens et al . , 2010b ) , and the STIL/Ana2 protein family is essential for the proper recruitment of Sas-6 to centrioles ( Dzhindzhev et al . , 2014; Ohta et al . , 2014 ) . We reasoned therefore , that Ana2 was likely to also play an important part in determining the structure of the central cartwheel . We set out to investigate the potential structural features of Ana2 that might be important for centriole assembly .
The Drosophila Ana2 protein contains four regions that have significant homology to Ana2/STIL proteins from other species ( Figures 1A , 2A ) ( Cottee et al . , 2013 ) . Fly Ana2 lacks the conserved region 1 found towards the N-terminus in vertebrate STIL proteins ( Figure 2A ) , but contains a CR2 domain that interacts with Sas-4 ( Cottee et al . , 2013; Hatzopoulos et al . , 2013 ) , a predicted central coiled-coiled domain ( CCCD ) , a STAN domain ( Stevens et al . , 2010a ) that interacts with Sas-6 ( Dzhindzhev et al . , 2014; Ohta et al . , 2014 ) and a short C-terminal CR4 domain ( Figure 1A ) ( Cottee et al . , 2013 ) . To examine the potential function of these conserved regions , we synthesised mRNAs in vitro that contained either wild type ( WT ) or truncated versions of Ana2 fused to either an N- or C-terminal GFP ( Figure 1A ) . These mRNAs were injected into WT early embryos ( that contain unlabelled endogenous WT Ana2 protein ) expressing RFP-Centrosomin ( Cnn ) as a centrosomal marker ( Conduit et al . , 2010 ) . The localisation of the encoded GFP-fusion protein was assessed 90–120 min after mRNA injection ( Figure 1B , C ) . 10 . 7554/eLife . 07236 . 003Figure 1 . A structure/function analysis of Drosophila Ana2 . ( A ) A schematic representation of Drosophila Ana2 highlighting the conserved domains and illustrating the GFP constructs analysed in this study . In vitro transcribed mRNA encoding each of these constructs was injected into Drosophila embryos expressing the PCM marker , RFP-Cnn; the distribution of each fusion protein was analysed in living embryos . ( B ) Micrographs show examples of typical centrosomes in embryos injected with the Ana2 constructs shown in ( A ) . The localisation of the GFP-fusion protein ( green ) is shown on its own ( left panel ) and merged with RFP-Cnn ( right panel ) . ( C ) Bars quantify the localisation behaviour of the various GFP-fusions . Images of 30–80 embryos were analysed for each construct . Images of each embryo were collected and then manually sorted into various categories based on the centrosomal localisation of the GFP-fusion construct ( see colour table at bottom of figure ) . All sorting was performed blind . DOI: http://dx . doi . org/10 . 7554/eLife . 07236 . 00310 . 7554/eLife . 07236 . 004Figure 2 . The Ana2 and STIL central coiled-coil domain ( CCCD ) regions form tetramers in solution . ( A ) A schematic representation of D . melanogaster Ana2 and human STIL highlighting the conserved domains . Note that vertebrate STIL proteins contain a conserved region 1 ( CR1 ) of unknown function that is not present in Ana2 . ( B ) A SEC-MALS analysis of the Drosophila Ana2 CCCD ( aa193–229 ) was performed . Injected protein concentrations are indicated by different shades of blue—solid lines show the relative Rayleigh ratio , dashed lines the observed mass . The black horizontal line indicates the theoretical mass for an Ana2 CCCD tetramer , the grey bar indicates a ±5% tolerance . ( C ) An analysis of the observed mass of human STIL CCCD ( 717–758 ) at various injected protein concentrations obtained from SEC MALS experiments . Error bars represent an estimated ±5% error in the MALS mass measurement , as each data point represents a single injection and mass measurement . The black line and grey bar represent the theoretical tetramer mass ±5% tolerance . The data were fitted to a hyperbolic function in Graphpad Prism 6 . 01 , including a 5% SEM for each mass value , with no extrapolation . This fitting estimated that the STIL CCCD was tending towards a mass of 23 . 4 kDa ( theoretical tetramer mass = 24 . 5 kDa ) with an R2 value of 95% . DOI: http://dx . doi . org/10 . 7554/eLife . 07236 . 00410 . 7554/eLife . 07236 . 005Figure 2—figure supplement 1 . Electrospray-ionisation mass spectrum of the Ana2 CCCD . The masses ( in Da ) for dimeric , trimeric and tetrameric Ana2 CCCD species can still be observed , demonstrating that the Ana2 CCCD self-interaction can partially survive these normally denaturing experimental conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 07236 . 005 Both N- and C-terminal GFP fusions of full length Ana2 ( constructs 1 and 6 ) showed a strong , compact , localisation to centrioles , as did fusions lacking either CR2 or CR4 ( constructs 2 and 7 ) , suggesting that these domains are not involved in Ana2 centriolar targeting . In contrast , fusions retaining the STAN domain , but lacking the CCCD ( constructs 8 , 9 and 11 ) showed a weak and diffuse localisation to the PCM . This PCM localisation appeared to be dependent on the STAN domain , as constructs lacking both the CCCD and the STAN domain were no longer detectable at centrioles or in the PCM ( constructs 5 , 10 , 12 and 13 ) . In contrast , constructs lacking the STAN domain , but retaining the CCCD , localised as a tight dot to centrioles ( although much more weakly than constructs that contained both domains ) and were not detectable in the PCM ( constructs 3 and 4 ) . These observations suggest that the CCCD is required for the centriolar localisation of Ana2 , while the STAN domain increases the efficiency of centriolar localisation and can also weakly target Ana2 to the PCM if the CCCD is absent . These findings are in agreement with recent data showing that STIL , the human homologue of Ana2 , is recruited to centrioles through a direct interaction between regions of STIL containing the CCCD and Plk4 ( Ohta et al . , 2014; Kratz et al . , 2015 ) . Interestingly , the CCCD alone could not target GFP to centrioles ( constructs 14 and 15 ) , demonstrating that , at least in this context , the CCCD was not able to directly target proteins to the centriole . We reasoned that the CCCD might function as an oligomerisation domain for Ana2 . To test this possibility , we bacterially expressed and purified the 37aa CCCD region ( residues 193–229 ) —as predicted by the COILS server ( Lupas et al . , 1991 ) —as a His-tagged diLipoyl peptide ( Figure 2A , B ) ( Cottee et al . , 2013 ) . A SEC-MALS analysis revealed that the purified protein , either with or without the Lipoyl tags , formed a tetramer at a wide range of concentrations ( 36–900 μM ) ( Figure 2B; Figure 4A ) . The CCCD tetramer was very stable and we could not find in-solution conditions under which it was dissociated , so we could not calculate a Kd . Even when examined using the usually denaturing technique , Electrospray-Ionisation Mass Spectrometry , the tetramer did not fully disassemble ( Figure 2—figure supplement 1 ) . We also expressed and purified the 42aa predicted CCCD ( residues 717–758 ) from the human STIL protein as a His-tagged diLipoyl peptide . This also formed a tetramer , although this was less stable than the fly CCCD tetramer and only formed at higher protein concentrations ( Figure 2C ) . The purified Ana2 CCCD protein readily formed protein crystals that diffracted extremely well , enabling us to refine a structure to 0 . 80 Å resolution ( Figure 3 , Figure 3—figure supplement 1 , Table 1 ) . The structure demonstrated that the Ana2 CCCD forms a parallel , symmetrical 4-helix bundle , with a left-handed supercoil ( Figure 3A ) . This structure appears to be unusual as we could find only one other natural soluble protein in the PDB that homo-tetramerises through a parallel four-helical bundle ( NSP4 , a tetrameric enterotoxin secreted by rotaviruses ) . Analysis using the PISA server ( Krissinel and Henrick , 2007 ) showed that residues located at the g , a , d and e positions of the helical heptad repeat were all buried at the tetramer interface ( Figure 3A , yellow residues ) . The tetramer is stabilised by at least three mechanisms: first , the knob-into-holes and van der Waals packing of hydrophobic residues ( Figure 3B , C ) ; second , the packing of internally facing polar residues ( Figure 3D ) ; third , a cross-chain salt bridge formed between R208 and E210 ( Figure 3E ) . 10 . 7554/eLife . 07236 . 006Figure 3 . The Ana2 CCCD forms a parallel four helical tetramer . ( A ) Left , the structure of the Ana2 CCCD tetramer generated around the crystallographic fourfold symmetry axis . The primary amino acid sequence is shown above the structure; residues in the g , a , d and e positions of the helical heptad repeat are indicated below the sequence . All these residues were ≥30% buried ( according to PISA server analysis ) and are coloured in yellow , with side-chains in stick format—other residues are coloured in cyan ( side-chains not shown ) . The TEV cleavage remnant is shown in grey . Right , schematic transverse view of the tetramer indicating how the g , a , d and e residues of the heptad repeat are buried at the tetramer interface . Note that the g and d residues ( coloured red , and highlighted with a red circle underneath the primary amino acid sequence ) form one side of this interface; these 10 residues were mutated to generate forms of the protein that could no longer form tetramers ( see main text ) . ( B–E ) Schematics illustrate the molecular determinants of tetramerisation , with interfacing residues shown as grey sticks . ( B ) A hydrophobic cluster of interface residues . The labelled residues sit at the g , a , d and e positions of the heptad repeat , and pack closely forming a hydrophobic environment . ( C ) A side on view of the same cluster , with one chain shown as a surface . ( D ) A transverse N-C view of a QQQ triad which adopts positions g , a and b of the heptad . These polar side-chains form an inward facing hydrogen-bond network . ( E ) A side-on view showing a salt bridge between adjacent chains of the tetramer . DOI: http://dx . doi . org/10 . 7554/eLife . 07236 . 00610 . 7554/eLife . 07236 . 007Figure 3—figure supplement 1 . Representative electron density for the Ana2 CCCD crystal structure at 0 . 8 Å resolution . The Ana2 CCCD is displayed in stick format , with the 2Fo-Fc map contoured at 1 . 8 σ . DOI: http://dx . doi . org/10 . 7554/eLife . 07236 . 00710 . 7554/eLife . 07236 . 014Table 1 . Ana2 CCCD dataset and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 07236 . 014Dataset statistics BeamlineDiamond I03 Wavelength ( Å ) 0 . 7293 SpacegroupI4 Unit cell dimensions ( Å/° ) 33 . 27 , 33 . 27 , 74 . 49/90 . 00 , 90 . 00 , 90 . 00 Resolution ( Å ) ( overall/inner/outer ) 30 . 36–0 . 80/30 . 36–3 . 58/0 . 82–0 . 80 Completeness ( overall/inner/outer ) 97 . 8/99 . 9/79 . 2 Rmerge ( overall/inner/outer ) 0 . 038/0 . 034/0 . 547 Rpim ( overall/inner/outer ) 0 . 013/0 . 011/0 . 312 CC ( 1/2 ) ( overall/inner/outer ) 1 . 00/0 . 999/0 . 782 I/σI ( overall/inner/outer ) 21 . 5 . 71 . 3/2 . 2 Mulitiplicity ( overall/inner/outer ) 5 . 8/11 . 2/3 . 3Refinement statistics ( parentheses = highest res shell ) Resolution range ( Å ) 30 . 36–0 . 80 ( 0 . 82 . 0 . 80 ) Rwork/Rfree/% test set size10 . 6/11 . 6/5 . 06% ( 21 . 1/20 . 3/4 . 84% ) Number of reflections working set/test set39 , 348 ( 2338 ) /2099 ( 119 ) Number of atoms ( non-H ) 499 Waters53 Rmsd from ideal values: bond length ( Å ) /angles ( ° ) 0 . 025/2 . 230 Average B factor ( Å2 ) 10 . 70 Ramachandran outliers0% Ramachandran favoured100% MolProbity score ( N number , percentile ) 1 . 22 ( 222 , 88% ) Ramachandran and Molprobity scores were calculated using MolProbity ( Chen et al . , 2010 ) . To test the potential importance of tetramerisation of the CCCD in vivo , we created point mutations within the CCCD that our structural studies suggested would disrupt the ability of the CCCD to tetramerise . We replaced all ten residues at the d and g positions of the CCCD with either Ala ( CCCD-A ) , Ser ( CCCD-S ) or Asp ( CCCD-D ) ( Figure 3A , residues circled in red ) . A SEC-MALS analysis revealed that all of these mutant CCCD proteins behaved as monomers rather than tetramers in vitro ( Figure 4A ) . We then made equivalent CCCD mutations within the context of the full length Ana2 protein and tested their localisation in our embryo RNA injection assay . All three mutant proteins were undetectable at centrioles but still localised diffusely to the PCM ( Figure 4B–D ) , indicating that the mutant proteins are not simply misfolded or degraded , as the STAN domain can still target them to the PCM . 10 . 7554/eLife . 07236 . 008Figure 4 . Mutations of the CCCD that perturb tetramer formation in vitro perturb the localisation of Ana2 to centrioles in vivo . ( A ) A SEC-MALS analysis of wild type ( WT ) and mutant forms of the CCCD where the 10 d and g residues important for tetramer formation ( circled in red , Figure 3A ) have been mutated either to Ala ( CCCD-A ) , Ser ( CCCD-S ) or Asp ( CCCD-D ) . Horizontal black lines illustrate the theoretical molecular mass of a tetramer and monomer , grey shading represents ±5% tolerance . Note that , in contrast to the SEC-MALS analysis presented in Figure 1A , the diLipoyl domains of the fusion proteins have not been removed in this experiment , so the masses of the monomer and tetramer are higher . ( B ) A schematic representation of the GFP-Ana2 fusions that contain mutations of the CCCD ( constructs #1 and #11 are the same constructs shown in Figure 1A ) . In vitro transcribed mRNA encoding each of these constructs was injected into Drosophila embryos expressing the PCM marker , RFP-Cnn; the distribution of each fusion protein was analysed in living embryos . ( C ) Micrographs show examples of typical centrosomes in embryos injected with the Ana2 constructs shown in ( A ) . The localisation of the GFP-fusion protein ( green ) is shown on its own ( left panel ) and merged with RFP-Cnn ( right panel ) . ( D ) Bars quantify the localisation behaviour of the various GFP-fusions . Images of 34–40 embryos were analysed for each construct . Images of each embryo were collected and then manually sorted into various categories based on the centrosomal localisation of the GFP-fusion construct ( see colour table at bottom of figure ) . All sorting was performed blind . The data shown here for constructs #1 and #11 is the same as that presented in Figure 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 07236 . 008 We next generated stable Drosophila transgenic lines that express full length Ana2-GFP containing the CCCD-A mutations under the control of the ubiquitin promoter ( Ana2-CCA-GFP ) . This promoter consistently results in the strong overexpression of both WT Ana2-GFP and mutant Ana2-CCA-GFP relative to the endogenous protein ( Figure 5A ) . While Ana2-GFP strongly rescued the centriole duplication defect seen in ana2 mutants , Ana2-CCA-GFP rescued much more weakly , although at least one centrosome-like structure ( CLS ) was detectable in ∼35% of cells expressing one copy of the transgene ( Figure 5B , F′ , F′′ ) . We do not know if these structures contain bona fide centrioles , but they stained for multiple centriole/centrosome markers and were almost invariably located at the spindle poles in mitotic cells , demonstrating that they retain at least some centriole and centrosome function ( Figure 5F–F′′; data not shown ) ; we therefore refer to these structures as CLSs . Interestingly , doubling the dosage of the Ana2-CCA-GFP , which already appeared to be overexpressed even with one gene dose ( Figure 5A ) , increased the efficiency of rescue , and nearly 90% of cells now contained at least one CLS ( Figure 5B ) . Several of these flies were clearly less uncoordinated than the ana2 mutant flies ( data not shown ) , strongly suggesting that flies rescued by a double dose of Ana2-CCA-GFP can form at least some functional cilia , again arguing that the CLSs retain some centriole activity . Taken together , these observations demonstrate that the ability of Ana2 to tetramerise is important for Ana2 function and for centriole assembly , but that Ana2-CCA retains some residual ability to promote the assembly of CLSs in vivo . 10 . 7554/eLife . 07236 . 009Figure 5 . A Mutant form of Ana2 that cannot tetramerise efficiently in vitro cannot support efficient centriole duplication in vivo . ( A ) Two exposures of a western blot illustrating the relative expression levels of endogenous Ana2 ( arrowhead ) and either WT Ana2-GFP or Ana2-CCA-GFP ( arrow ) ( expressed from one ( 1× ) or two ( 2× ) copies of the transgene ) in third instar larval brains in either a WT or ana2 mutant background . Actin is shown as a loading control . ( B ) The bar chart shows the number of centrosomes or centrosome-like structures ( CLSs ) observed in mitotic third instar larval brain cells ( scored by the presence of both the centriole marker Asl and the centrosome marker Cnn ) in WT , ana2 mutant and ana2 mutants expressing one ( 1× ) or two ( 2× ) copies of either WT Ana2-GFP or Ana2-CCA-GFP , as indicated . A total of at least 300 mitotic cells from at least five different brains were scored for each genotype; error bars represent the SD . ( C–F′′ ) Micrographs show the distribution of Asl ( green ) and Cnn ( red ) in representative mitotic third instar larval brain cells of the indicated genotypes . DNA is in blue . The images in F–F′′ show cells rescued with the Ana2-CCA-GFP construct that have either no centrosomes ( F ) or one ( F′ ) or two ( F′′ ) CLSs . Scale bar in C: 2 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07236 . 009 It has previously been shown that Sas-6 proteins also need to homo-oligomerise to function in centriole duplication ( Kitagawa et al . , 2011; van Breugel et al . , 2011 ) , so we wanted to explore the relative importance of Sas-6 and Ana2 oligomerisation for centriole duplication . In all species examined to date Sas-6 forms dimers through an extended C-terminal coiled-coil region ( C–C ) ( Kitagawa et al . , 2011; van Breugel et al . , 2011; Qiao et al . , 2012 ) . In Danio rerio , Chlamydomonas and Leishmania these dimers can further homo-oligomerise through an N-terminal headgroup interaction ( N–N ) to form a flat ninefold symmetric ring from which the C–C domains emanate—thus forming the central hub and spokes of the cartwheel ( Figure 6I ) . In Caenorhabditis elegans , however , the SAS-6 headgroup-CC orientation is altered ( Figure 6H ) , and SAS-6 dimers appear to oligomerise into a spiral , rather than a flat-ring ( Hilbert et al . , 2013 ) , potentially explaining why a classical cartwheel with nine spokes has not been visualised by EM in C . elegans centrioles ( Pelletier et al . , 2006 ) . In Drosophila centrioles , EM images reveal a clear central cartwheel hub from which emanating spokes are often visible—but it is difficult to visualise more than a few spoke structures at any one time ( e . g . , Callaini et al . , 1997; Roque et al . , 2012; Helio Roque , personal communication ) , making it unclear whether Drosophila Sas-6 oligomerises into a canonical ring or into a spiral . To address this issue , we attempted to examine the structure of Drosophila Sas-6 ( Figure 6A ) . 10 . 7554/eLife . 07236 . 010Figure 6 . A biochemical and structural analysis of Drosophila Sas-6 . ( A ) A schematic representation of Drosophila Sas-6 highlighting the position of the N-terminal head domain ( blue ) and C-terminal coiled-coil ( CC ) domain ( green ) . Red lines below represent the constructs used in SEC-MALS and EM studies ( top ) and in X-Ray Crystallography studies ( bottom ) . ( B ) A SEC-MALS analysis of WT ( blue trace ) and F143D mutant ( red trace ) Sas-61–241 proteins , injected at 33 µM . The horizontal black line and grey bar represent the theoretical dimer mass ±5% tolerance . The WT protein could not be analysed by MALS as it eluted in the void volume and appeared to form a range of higher-order oligomers . ( C ) Negative-stain EM analysis of purified WT ( [i]–[iv] ) Sas-61–241 protein , showing the chain-like structures formed ( [iii] and [iv] show magnified views of the red boxed areas in [i] and [ii] ) ; these structures are not detectable in preparations of the mutant Sas-6-F143D1–241 protein ( [v] ) . ( D ) The structure of the Sas-6 dimer , coloured according to Consurf conservation scores ( Glaser et al . , 2003 ) from cyan ( variable ) to burgundy ( conserved ) . The conserved PISA domain and the N-CC interface regions are highlighted with dashed circles . ( E–G ) Superimposed structures from D . melanogaster , D . rerio , Chlamydomonas and Leishmania ( as indicated ) of the Sas-6 N-terminal head-group with a short stretch of the coiled-coil domain . ( H ) Superimposed structures of the N-CC interface in D . melanogaster , D . rerio , Chlamydomonas and C . elegans . Note how the interface is rotated by ∼30° in C . elegans ( purple ) compared to the other structures . ( I ) The DmSas-6 structure modelled into a ninefold symmetric flat ring ( green , single dimer shown in red ) , similar to that observed in crystallo for LmSAS-6 . This ring structure was docked into the EM density of the Triconympha cartwheel structure ( Guichard et al . , 2013 ) ( cyan surface , cut away to reveal the DmSAS-6 ring ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07236 . 010 We were unable to purify constructs containing only the N-terminal head-group , however we could purify constructs that contained the N-terminal headgroup and either 59 ( Sas-61–216 ) or 84 ( Sas-61–241 ) residues of the predicted C–C region . In initial attempts to purify Sas-61–241 the protein invariably formed large aggregates ( blue trace , Figure 6B ) that appeared to be elongated chains of protein by negative-stain EM ( Figure 6Ci–iv ) . It has previously been shown that a large hydrophobic residue in the headgroup is essential for the N–N interaction in several species ( Kitagawa et al . , 2011; van Breugel et al . , 2011 ) , so we mutated the equivalent residue , F143 , to Asp . Purified Sas-61–241-F143D behaved as a dimer by SEC-MALS ( red trace , Figure 6B ) and aggregates were no longer detectable by negative-stain EM ( Figure 6Cv ) ; we conclude that aggregate formation is dependent upon the N–N interaction , and the F143D mutation perturbs this interaction in vitro . To investigate how Drosophila Sas-6 might oligomerise into a cartwheel we solved the crystal structure of Sas-61–216-F143D to 2 . 9 Å ( Figure 6D , Table 2 ) . The asymmetric unit contained a dimer of Sas-6 , associated via the coiled-coil interface . To assess whether this Sas-6 N-CC dimer could be built into a canonical flat ring structure , we compared it to other Sas-6 orthologues for which structures are available . The DmSas-6 N-CC dimer could be superimposed with Sas-6 N-CC dimers from D . rerio , Chlamydomonas and Leishmania Sas-6 ( average pairwise RMSD 1 . 87 ± 0 . 31 Å over 617 ± 47 backbone atom pairs ) ( Figure 6E–G ) . However it could not be superimposed onto C . elegans SAS-6 , which has an alternative head-group-spoke conformation ( Figure 6H ) . Furthermore , we found that the DmSas-6 N-CC dimer could be modelled into a flat ninefold ring ( Figure 6I ) , similar to that observed in crystallo for Leishmania Sas-6 ( van Breugel et al . , 2014 ) . The structure of DmSas-6 is therefore highly similar to Sas-6 orthologues from organisms with canonical cartwheels , suggesting that it also forms such a structure . 10 . 7554/eLife . 07236 . 015Table 2 . Sas-61–216 ( F143D ) dataset and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 07236 . 015Dataset statistics BeamlineESRF ID23-2 Wavelength ( Å ) 0 . 8726 SpacegroupP2 Unit cell dimensions ( Å/° ) 47 . 13 , 64 . 74 , 123 . 73/90 . 00 , 98 . 91 , 90 . 00 Resolution ( Å ) ( overall/inner/outer ) 41 . 43–2 . 92/41 . 43–13 . 06/3 . 00–2 . 92 Completeness ( overall/inner/outer ) 97 . 4/90 . 4/96 . 4 Rmerge ( overall/inner/outer ) 0 . 128/0 . 035/0 . 668 Rpim ( overall/inner/outer ) 0 . 060/0 . 017/0 . 318 CC ( 1/2 ) ( overall/inner/outer ) 0 . 994/0 . 988/0 . 799 I/σI ( overall/inner/outer ) 12 . 1/48 . 4/2 . 6 Mulitiplicity ( overall/inner/outer ) 5 . 2/4 . 8/5 . 1Refinement statistics ( parentheses = highest res shell ) Resolution range ( Å ) 41 . 43–2 . 92 ( 3 . 10–2 . 92 ) Rwork/Rfree/% test set size18 . 3/21 . 5/5 . 00% ( 26 . 2/34 . 5/5 . 90% ) Number of reflections working set/test set14 , 976 ( 2456 ) /788 ( 154 ) Number of atoms ( non-H ) 3405 Waters33 Rmsd from ideal values: bond length ( Å ) /angles ( ° ) 0 . 007/1 . 055 Average B factor ( Å2 ) 76 . 30 Ramachandran outliers0% Ramachandran favoured94 . 9% Molprobity score ( N number , percentile ) 1 . 54 ( 3648 , 100% ) Ramachandran and Molprobity scores were calculated using MolProbity ( Chen et al . , 2010 ) . To test whether the ability of Sas-6 to form higher-order oligomers was important for Sas-6 function , as has been observed in several other systems ( Kitagawa et al . , 2011; van Breugel et al . , 2011 ) , we generated stable transgenic lines expressing either WT GFP-Sas-6 or GFP-Sas-6-F143D under the control of the ubiquitin promoter . This promoter consistently resulted in the overexpression of both WT GFP-Sas-6 and GFP-Sas-6-F143D compared to the endogenous protein ( Figure 7A ) . While WT GFP-Sas-6 strongly rescued the centriole duplication defect seen in Sas-6 mutants , GFP-Sas-6-F143D rescued much more weakly , although , at least one CLS was detectable in ∼60% of cells expressing one copy of the transgene ( Figure 7B–F′′ ) . As was the case with the rescue of the ana2 mutation by Ana2-CCA-GFP , these structures stained for multiple centriole/centrosome markers and were usually located at the spindle poles in mitotic cells , demonstrating that they retain at least some centriole and centrosome function ( Figure 7F–F′′; data not shown ) . From our qualitative analysis , however , the CLSs formed when Sas-6 mutants were rescued by GFP-Sas-6-F143D often appeared smaller and more fragmented than those observed when ana2 mutants were rescued by Ana2-CCA-GFP , suggesting that the CLSs formed in the presence of GFP-Sas-6-F143D may be less well organised than those formed in the presence of Ana2-CCA-GFP . Moreover , as described below , females carrying even one copy of this transgene invariably laid embryos that arrested early in development , so we could not generate flies carrying two copies of the transgene to test if the rescuing activity of the transgene increased with gene dosage—as we observed for Ana2-CCA-GFP ( Figure 5B ) . Nevertheless , these data demonstrate that the ability of Sas-6 to form higher order oligomers is important for Sas-6 function and for centriole assembly , but that GFP-Sas-6-F143D retains some residual ability to promote the assembly of CLSs in vivo ( Figure 6B , C ) . 10 . 7554/eLife . 07236 . 011Figure 7 . A Mutant form of Sas-6 that cannot oligomerise efficiently in vitro cannot support efficient centriole duplication in vivo . ( A ) A western blot illustrating the relative expression levels of endogenous Sas-6 ( arrowhead ) and either WT GFP-Sas-6 or GFP-Sas-6-F143D ( arrow ) in third instar larval brains in either a WT or Sas-6 mutant background . Actin is shown as a loading control , and an ( * ) marks a non-specific band . ( B ) The bar chart shows the number of centrosomes or CLSs observed in mitotic third instar larval brain cells in WT or Sas-6 mutants ( first and second bars ) ; in Sas-6 mutants expressing either WT GFP-Sas-6 or GFP-Sas-6-F143D ( third and fourth bars ) ; or WT brains expressing either WT GFP-Sas-6 or GFP-Sas-6-F143D ( fifth and sixth bars ) . At least 600 mitotic cells from at least five different brains were scored for each genotype; error bars represent the SD . ( C–F′′ ) Micrographs show the distribution of Asl ( green ) and Cnn ( red ) in representative mitotic third instar larval brain cells of the indicated genotypes . DNA is in blue . The images in F , F′ and F′′ show Sas-6 mutant cells rescued by Sas-6-F143D showing examples of the CLSs . ( G , H ) Micrographs show the distribution of Asl ( green ) and Cnn ( red ) in either a WT primary spermatocyte ( G ) or a WT primary spermatocyte overexpressing GFP-Sas-6-F143D ( H ) . ( I ) Graph shows the quantification of centriole length ( as measured by Asl staining ) in WT primary spermatocytes ( blue circles ) or WT primary spermatocytes overexpressing GFP-Sas-6-F143D ( red boxes ) ; at least 1500 centrioles from at least 30 different testes were scored for each genotype , and each circle or box represents the mean from an individual testes . Statistical significance was assessed using an unpaired two-tailed t-test: ( *** ) indicates p-value < 0 . 001 . ( J–K ) Micrographs show the distribution of the centriole marker RFP-PACT ( red ) and either WT GFP-Sas-6 ( green ) ( J–J′′ ) or GFP-Sas-6-F143D ( K–K′′ ) in WT primary spermatocytes . Scale bars: 2 µm in C–F′′ and J–K′′ and 5 µm in H–I . DOI: http://dx . doi . org/10 . 7554/eLife . 07236 . 011 In light of the proposed mechanism of Sas-6-supported cartwheel assembly , the overexpression of mutant forms of the protein that cannot form higher order oligomers through the N–N interaction might be expected to act as dominant-negatives , capable of ‘poisoning’ cartwheel assembly by forming hetero-dimers with the WT protein that can incorporate into the cartwheel through the WT headgroup , but which cannot then interact with another headgroup—thus blocking further cartwheel assembly ( Figure 8A ) . Surprisingly , however , although GFP-Sas-6-F143D was overexpressed in all tissues we examined ( embryos , brains and testes ) , it had very little , if any , negative effect on centriole duplication in WT brain cells ( Figure 7B ) or spermatocytes ( Figure 7G , H ) , although the centrioles were ∼10% shorter in spermatocytes overexpressing GFP-Sas-6-F143D ( Figure 7I ) . Interestingly , small amounts of GFP-Sas-6-F143D could be detected in spermatocyte centrioles , but this was more diffusely localised throughout the centriole length when compared to the WT GFP-Sas-6 , which was strongly concentrated at the proximal and distal ends of the centrioles ( as reported previously ) ( Peel et al . , 2007 ) ( Figure 7J , K ) . Moreover , WT flies overexpressing GFP-Sas-6-F143D were not noticeably uncoordinated , demonstrating that they can form functional cilia . 10 . 7554/eLife . 07236 . 012Figure 8 . GFP-Sas-6-F143D dominantly suppresses centrosome assembly in early embryos . ( A ) A schematic illustration of how GFP-Sas-6-F143D could act as a dominant-negative in cartwheel assembly . WT Sas-6 ( light and dark green ) can form WT–WT homodimers or WT-mutant heterodimers with GFP-Sas-6-F143D ( red ) . The homodimers can support cartwheel assembly while the heterodimers can incorporate into the growing cartwheel ( through the WT headgroup ) , but cannot support further cartwheel assembly . The heterodimer must dissociate before a WT homodimer can incorporate into the cartwheel , so allowing cartwheel assembly to proceed . ( B–H′′ ) Micrographs show images from WT embryos expressing no transgene ( F ) or expressing either WT GFP-Sas-6 ( B , D ) or GFP-Sas-6-F143D ( C , E , G , H ) stained to reveal the distribution of GFP , Asl , Cnn or α-tubulin , as indicated . ( B , D ) Embryos expressing WT GFP-Sas-6 develop normally , and the fusion protein strongly localises to a bright spot in the centre of the centrosomes . ( C , E ) Embryos expressing GFP-Sas-6-F143D arrest during the early syncytial stages; some of these embryos are reasonably well organized and centrioles are observed at the spindle poles , but these contain very little detectable GFP-Sas-6-F143D . ( F–G ) Most embryos are less well organized and contain abnormal microtubule ( MT ) arrays organized by fragmented centrosomes ( G , H ) when compared to WT ( F ) . Scale bars: 10 µm in B , C , 2 µm in D–D′′ and F–H′′ and 3 µm in E–E′′ . DOI: http://dx . doi . org/10 . 7554/eLife . 07236 . 012 In embryos , however , GFP-Sas-6-F143D had a strong dominant-negative affect , and females expressing this transgene laid embryos that invariably arrested early in development after they had gone through only a few rounds of nuclear division ( Figure 8B , C ) . The MTs in these embryos appeared to be organised by centrioles that had incorporated only very small amounts of GFP-Sas-6-F143D ( Figure 8D , E ) , and which often appeared small and fragmented ( Figure 8F–H ) . These observations have important implications for the mechanism of Sas-6-mediated cartwheel assembly as they suggest that GFP-Sas-6-F143D can effectively poison cartwheel assembly in rapidly dividing syncytial embryos that have to assemble centrioles very quickly ( and centrioles are essential for early embryo development in flies [Stevens et al . , 2007; Varmark et al . , 2007] ) , but not in brain cells or spermatocytes that have a slower cell cycle and so can presumably assemble their centrioles over a longer time-frame ( see ‘Discussion’ ) .
Rogala et al . have now shown that C . elegans SAS-5 , the functional homologue of Ana2/STIL , can also form higher-order oligomers and this is essential for SAS-5 function . SAS-5 has two major oligomerisation domains ( a central coiled-coil region and an implico domain ) that allow the protein to form a mix of tetramers and hexamers in vitro . This difference may reflect that SAS-6 forms a spiral , rather than a cartwheel , structure in C . elegans ( Rogala et al . , 2015 ) .
Fragments of Ana2 were PCR amplified from cDNA and subcloned into modified pRNA destination vectors ( Conduit et al . , 2014 ) using the Gateway ( Life Technologies , Carlsbad , CA ) system . These vectors contain a T3 RNA polymerase promoter , a polyA tail and encode GFP in-frame , 5′ or 3′ of the insert . Deletion constructs were generated using a Quikchange II XL mutagenesis kit ( Agilent Technologies , Santa Clara , CA ) . Mutation constructs ( CCCD-A/CCCD-S/CCCD-D ) were synthesised de novo ( Genewiz , South Plainfield , NJ ) using Drosophila-optimised codons for each substituted residue . In vitro mRNA was synthesised from linearised ( AscI ) pRNA vectors using an mMESSAGE mMACHINE T3 Transcription Kit ( Life Technologies ) , and purified using an RNeasy MinElute kit ( Qiagen , Hilden , Germany ) . The molar concentration of RNA was normalised according to overall RNA yield , and the theoretical length of each transcript . Embryos expressing RFP-Cnn ( Conduit et al . , 2010 ) were injected and incubated at 25°C for 90–120 min to allow the mRNA to be translated . Images were acquired using a Perkin Elmer ERS spinning disk system ( Volocity software ) mounted on a Zeiss Axiovert microscope using a 63× 1 . 4 NA oil immersion objective and an Orca ER CCD camera ( Hamamatsu Photonics , Japan ) . Images were processed using either Volocity ( Perkin Elmer , USA ) or Fiji software ( Schindelin et al . , 2012 ) . Images of injected embryos were classified into different catagories based on a qualitative assessment of the ability of the injected fusion-protein to localize to centrosomes . This was performed blind after randomisation of the images . Flies were kept at 25°C , OregonR and w67 ( Bloomington Stock Centre ) served as wild-type controls . The following mutant alleles and stocks were used in this study: ana2169 , ana2719 ( Wang et al . , 2011 ) , Sas-6c02901 ( Peel et al . , 2007 ) , Ubq-Ana2-GFP ( Cottee et al . , 2013 ) , Ubq-Ana2-CCA-GFP ( this study ) , WT GFP-Sas-6 ( this study ) and GFP-Sas-6-F143D ( this study ) . All transgenes were generated by standard P-element mediated transformation ( performed by either the Genetics Department , University of Cambridge , UK or Bestgene Inc . , Chino Hills , CA ) , and all fusion proteins are expressed from the ubiquitin promoter , which drives moderate expression in all cell types ( Lee et al . , 1988 ) . GFP-tagged full length Sas-6 was generated by cloning the full length cDNA into the pUbq-GFP ( NT ) destination vector using the Gateway System ( Life Technologies ) . Point mutations were introduced into full-length ana2 and Sas-6 cDNA using site-directed mutagenesis ( QuickChange II XL , Agilent Technologies ) . Brains were dissected , squashed and stained as previously described ( Stevens et al . , 2009 ) . Adult testes were dissected and fixed as described ( Dix and Raff , 2007 ) . Testes were then incubated with primary antibodies overnight at 4°C followed by washes with PBT and secondary antibody incubation for 4 hr at RT . Slides were washed in PBT and mounted for analysis . Embryos from 0–2 hr egg collections were aged for 1 hr at 25°C and were fixed and stained as previously described ( Stevens et al . , 2009 ) . To preserve the GFP signal in embryos expressing either WT GFP-Sas-6 or GFP-Sas-6-F143D , embryos were fixed in 14 . 4% microfiltered FA solution containing 100 mM PIPES ( pH 7 . 0 ) , 2 mM EGTA and 1 mM MgSO4 for 5 min . The following antibodies were used: sheep anti-Cnn ( 1:1000 ) ( Cottee et al . , 2013 ) , guinea pig anti-Asl ( 1:500 ) ( Cottee et al . , 2013 ) ; GFP-Booster ( ChromoTek , Germany ) was used at 1/500 to enhance the GFP signal . Secondary antibodies conjugated to either Alexa Fluor 488 or Alexa Fluor 568 ( Life Technologies ) were used 1:1000 . Hoechst33258 ( Life Technologies ) was used to visualise DNA . Centrosomes were counted on a Zeiss Axioskop 2 microscope using a 63× 1 . 25 NA objective . Images were acquired in Metamorph ( Molecular Devices , UK ) using a CoolSNAP HQ camera ( Photometrics , Tucson , AZ ) and processed using Fiji ( Schindelin et al . , 2012 ) and Inkscape ( www . inkscape . org/ ) for image assembly . Only brain cells in metaphase were scored ( based on DNA morphology ) , and only centrosomes that clearly stained for both Asl and Cnn were counted . A total of at least 300 cells from at least five brains were analysed for each genotype . Centriole length was measured in fixed meiosis II spermatocytes using the line drawing and measuring tool in Fiji . Length in pixels was converted into µm . At least 30 testes were analysed for each genotype . Centrioles were also examined in living testes dissected in PBS . Testes were transferred to a coverslip with a drop of saline buffer and gently squashed between the coverslip and slide and imaged on the Zeiss Axioskop 2 system described above . The following primary and secondary antibodies were used: rabbit anti-Ana2 ( 3:500 ) , ( Stevens et al . , 2010a ) , rabbit anti-Sas-6 ( 1:500 ) ( Basto et al . , 2006 ) , mouse anti-GFP ( 1:500 , Roche , Switzerland ) , mouse anti-actin ( 1:1000 , Sigma-Aldrich , St . Louis , MO ) , anti-mouse HRP ( 1:3000 , GE Healthcare , UK ) and anti-rabbit HRP ( 1:3000 , GE Healthcare ) . The cDNA sequences encoding Drosophila melanogaster Ana2193–229 ( CCCD ) was inserted into a custom ‘pLip’ vector , which encodes two , TEV protease cleavable , His-tagged lipoyl domains ( from Bacillus stearothermophilus dihydrolipoamide acetyltransferase ) , one fused at either terminus of the insert ( Cottee et al . , 2013 ) . We term the resulting peptide-fusion a ‘diLipoyl fusion protein’ . CCCD-A , CCCD-S , and CCCD-D variants were subcloned from the pRNA plasmids described above . All constructs were expressed in Escherichia coli B834 ( DE3 ) cells in LB broth , and purified using Ni-NTA affinity , and size exclusion chromatography . The Ana2 CCCD was purified from its diLipoyl fusion protein by proteolytic cleavage , size exclusion , and ion exchange chromatography . The construct contains a GGS motif at the N-terminus , and an EFGENLYFQ motif at the C-terminus—remnants of the cloning and protease cleavage sites . E . coli codon-optimised cDNA encoding Human STIL717–758 ( CCCD ) was synthesised ( Genewiz , South Plainfield , NJ ) and inserted into the pLip vector . diLipoyl-STIL717–758 was expressed in E . coli C41 ( DE3 ) and purified as for diLipoyl-Ana2193–229 . STIL717–758 alone was purified by proteolytic cleavage , followed by reverse Ni-NTA chromatography , and size exclusion . The construct contains the same remnants of the cloning and protease cleavage sites as described above . Sas6 fragments were cloned from D . melanogaster cDNA ( AAL68137 ) into a pETM-14 ( EMBL ) vector encoding a cleavable N-terminal His tag . The F143D mutation was inserted using a Quikchange II XL mutagenesis kit ( Agilent Technologies , Santa Clara , CA ) . Sas-61–241 ( WT ) and Sas-61–241 ( F143D ) were expressed in E . coli B834 ( DE3 ) and purified using Ni-NTA and SEC chromatography . Sas-61–216 ( F143D ) was similarly expressed , however the His-tag was removed via proteolytic cleavage and reverse Ni-NTA chromatography prior to SEC . Sas-61–216 ( F143D ) contains a GP at the N-terminus , and a G after the initiator methionine , due to the cloning and protease cleavage sites . Protein samples were diluted to 33 . 3 µg/ml in water . 30 µl of sample was deposited for 2 min onto a 200 mesh , glow discharged carbon coated copper grid . The sample was negatively stained by applying 2% wt/vol uranyl acetate for 10 s before blotting , and air-drying the grid . Samples were viewed using an FEI Tecnai 12 TEM ( FEI , Hillsboro , OR ) , at 120 kV , 43 , 000× magnification . Samples were dialysed into 50 mM Tris pH 7 . 5 , 150 mM NaCl , 1 mM DTT . 100 µl of protein sample was injected onto an S200 10/300 column ( GE Healthcare ) . The light scattering and refractive index were respectively measured in-line by Dawn Heleos-II and Optilab rEX instruments ( Wyatt Technology , Santa Barbara , CA ) , as the samples eluted from the column . Data were analysed using ASTRA software ( Wyatt Technology ) assuming a dn/dc value of 0 . 186 ml/g . Protein samples were desalted with a Chromolith RP-18e column ( Merck , Kenilworth , NJ ) . These samples in Acetonitrile:water + 0 . 1% Formic acid were introduced by electrospray ionisation into a Micromass LCT Premier XE orthogonal acceleration reflecting TOF mass spectrometer in positive ion mode ( Micromass , Milford , MA ) . The resultant m/z spectra were converted to mass spectra by using the maximum entropy analysis MaxEnt in the MassLynx suite of programs . D . melanogaster Ana2 CCCD was dialysed into 20 mM Tris pH 7 . 5 , 150 mM NaCl , 1 mM DTT , and concentrated to 41–43 mg/ml . Initial crystals grew readily at 20°C overnight in sitting drops , using the Stura/Macrosol and Morpheus screens ( Molecular Dimensions , Newmarket , UK ) . The best diffracting crystal grew in an optimisation screen , using 160 nl protein solution and 40 nl of mother liquor ( 100 mM HEPES mix ( 71% pH 7 . 2 , 29% pH 8 . 2 ) , 42% PEG 600 ) . Crystals typically grew to their maximal size within 2–4 days and were fished and flash frozen in liquid nitrogen within 1–14 days . PEG 600 in the mother liquor served as cryoprotectant . D . melanogaster Sas-61–216 ( F143D ) was purified in 50 mM Tris pH 7 . 5 , 150 mM NaCl , and concentrated to 41 . 6 mg/ml . Small rod-like crystals grew after ∼3 weeks at 21°C in drops containing 300 nl protein solution , and 100 nl mother liquor ( 0 . 1 M Bicine/Trizma mix ( pH 8 . 5 ) , 20% wt/vol PEG 550MME , 10% wt/vol PEG 20 K , 30 mM NaNO3 , 30 mM Na2HPO4 , 30 mM ( NH4 ) 2SO4 ) . Crystals were fished and flashed frozen after ∼4 weeks with PEG 550MME in the mother liquor serving as cryoprotectant . Ana2 CCCD data were collected at Diamond beamline I03 . Due to the high resolution of diffraction , a short wavelength ( 0 . 7293 Å ) was used to maximise the number of reflections collected on the detector . A high ( 0 . 8 Å ) and low resolution sweep were processed using the Xia2 pipeline ( Winter , 2009 ) , using XDS ( Kabsch , 2010 ) and AIMLESS ( Evans and Murshudov , 2013 ) . Processing statistics suggest that , given a more optimal beamline setup , useful data could be collected to a higher resolution than 0 . 80 Å . The structure was solved via molecular replacement ( Molrep ) using a helix ( chain A , residues 2–31 ) from PDB entry 1UO4 ( Yadav et al . , 2005 ) —an engineered coiled coil peptide . We retrospectively found that the Ana2 CCCD could be trivially solved via direct methods , using ACORN ( Jia-xing et al . , 2005 ) . Autobuilding was carried out using Buccaneer ( Cowtan , 2006 ) . Refinement was carried out in Phenix . refine ( Afonine et al . , 2012 ) and Refmac ( Murshudov et al . , 2011 ) , using anisotropic B factor refinement and hydrogens modelled in riding positions . Manual rebuilding was performed in Coot ( Emsley and Cowtan , 2004 ) . Sas-61–216 ( F143D ) data were collected at the ESRF beamline ID23-2 . Data were processed using the Xia2 pipeline ( Winter 2009 ) , using XDS ( Kabsch , 2010 ) and AIMLESS ( Evans and Murshudov , 2013 ) . Phasing was carried out by molecular replacement in Phaser ( McCoy et al . , 2007 ) using an ensemble of monomeric SAS-6 structures ( 2Y3V ( A/B/D ) , 2Y3W ( A/B ) 3Q0X ( A/B ) ) ( Kitagawa et al . , 2011; van Breugel et al . , 2011 ) prepared for MR using Chainsaw to trim sidechains to the last common atom ( Stein , 2008 ) . Autobuilding was initially carried out using Buccaneer ( Cowtan , 2006 ) to build into maps that had been solvent flattened using Parrot ( Cowtan , 2010 ) . Density for the C-terminal part of the CC is weak , and only continuous at lower map contours ( ∼0 . 6 σ ) and was initially built with the aid of a solvent mask in autoBUSTER ( Bricogne et al . , 2011 ) . Refinement and model building were carried out using autoBUSTER and Phenix . refine ( Afonine et al . , 2012 ) with model building carried out in Coot ( Emsley and Cowtan , 2004 ) . | Most animal cells contain structures known as centrioles . Typically , a cell that is not dividing contains a pair of centrioles . But when a cell prepares to divide , the centrioles are duplicated . The two pairs of centrioles then organize the scaffolding that shares the genetic material equally between the newly formed cells at cell division . Centriole assembly is tightly regulated and abnormalities in this process can lead to developmental defects and cancer . Centrioles likely contain several hundred proteins , but only a few of these are strictly needed for centriole assembly . New centrioles usually assemble from a cartwheel-like arrangement of proteins , which includes a protein called SAS-6 . Previous work has suggested that in the fruit fly Drosophila melanogaster , Sas-6 can only form this cartwheel when another protein called Ana2 is also present , but the details of this process are unclear . Now , Cottee , Muschalik et al . have investigated potential features in the Ana2 protein that might be important for centriole assembly . These experiments revealed that a region in the Ana2 protein , called the ‘central coiled-coil domain’ , is required to target Ana2 to centrioles . Furthermore , purified coiled-coil domains were found to bind together in groups of four ( called tetramers ) . Cottee , Muschalik et al . then used a technique called X-ray crystallography to work out the three-dimensional structure of one of these tetramers and part of the Sas-6 protein with a high level of detail . These structures confirmed that Sas-6 proteins also associate with each other . When fruit flies were engineered to produce either Ana2 or Sas-6 proteins that cannot self-associate , the flies' cells were unable to efficiently make centrioles . Furthermore , an independent study by Rogala et al . found similar results for a protein that is related to Ana2: a protein called SAS-5 from the microscopic worm Caenorhabditis elegans . Further work is needed to understand how Sas-6 and Ana2 work with each other to form the cartwheel-like arrangement at the core of centrioles . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2015 | The homo-oligomerisation of both Sas-6 and Ana2 is required for efficient centriole assembly in flies |
Plastids are supported by a wide range of proteins encoded within the nucleus and imported from the cytoplasm . These plastid-targeted proteins may originate from the endosymbiont , the host , or other sources entirely . Here , we identify and characterise 770 plastid-targeted proteins that are conserved across the ochrophytes , a major group of algae including diatoms , pelagophytes and kelps , that possess plastids derived from red algae . We show that the ancestral ochrophyte plastid proteome was an evolutionary chimera , with 25% of its phylogenetically tractable nucleus-encoded proteins deriving from green algae . We additionally show that functional mixing of host and plastid proteomes , such as through dual-targeting , is an ancestral feature of plastid evolution . Finally , we detect a clear phylogenetic signal from one ochrophyte subgroup , the lineage containing pelagophytes and dictyochophytes , in plastid-targeted proteins from another major algal lineage , the haptophytes . This may represent a possible serial endosymbiosis event deep in eukaryotic evolutionary history .
Since their origin , the eukaryotes have diversified into an extraordinary array of organisms , with different genome contents , physiological properties , and ecological adaptations ( Dorrell and Smith , 2011; de Vargas et al . , 2015; Dorrell and Howe , 2012a ) . Perhaps the most profound change that has occurred within individual eukaryotic cells is the acquisition of plastids via endosymbiosis , which has happened at least eleven times across the tree of life ( Dorrell and Smith , 2011 ) . All but one characterized group of photosynthetic eukaryotes possess plastids resulting from a single ancient endosymbiosis of a beta-cyanobacterium by an ancestor of the archaeplastid lineage ( consisting of green algae and plants , red algae , and glaucophytes ) ( Dorrell and Smith , 2011 ) . Photosynthesis has subsequently spread outside of the archaeplastids through secondary , tertiary , or more complex endosymbiosis events . By far the most ecologically successful of these lineages are those that possess plastids derived from secondary or more complex endosymbioses of a red alga ( Dorrell and Smith , 2011; Baurain et al . , 2010; Stiller et al . , 2014 ) . These are the ‘CASH lineages’ , consisting of photosynthetic members of the cryptomonads , alveolates ( such as dinoflagellates ) , stramenopiles ( also referred to as heterokonts ) and haptophytes ( Dorrell and Smith , 2011; Baurain et al . , 2010 ) ( see Table 1 and Figure 1—figure supplement 1 for definitions ) . The most prominent of these are the photosynthetic members of the stramenopiles , termed the ochrophytes ( de Vargas et al . , 2015; Aleoshin et al . , 2016; Ševčíková et al . , 2015 ) . The ochrophytes include the diatoms , which are major primary producers in the ocean ( Bowler et al . , 2010; Armbrust et al . , 2004 ) , multicellular kelps , which serve as spawning grounds for marine animals ( Cock et al . , 2010 ) and the pelagophytes , microscopic algae of which some are known to form harmful blooms ( Gobler et al . , 2011 ) ( Figure 1 , panel A; Figure 1—figure supplement 1 ) . The stramenopiles also contain many aplastidic and non-photosynthetic lineages ( e . g . , oomycetes ) , which diverge at the base of the ochrophytes and play important roles as pathogens and in microbial food webs ( Aleoshin et al . , 2016; Derelle et al . , 2016 ) ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 23717 . 003Figure 1 . Procedure for identification of conserved plastid-targeted proteins in ochrophytes . ( Panel A ) shows a schematic unrooted ochrophyte tree , with the three major ochrophyte lineages ( chrysista , hypogyristea , and diatoms ) denoted by different coloured labels . ‘PX’ refers to the combined clade of phaeophytes , xanthophytes and related taxa , and ‘PESC’ to pinguiophytes , eustigmatophytes , synchromophytes , chrysophytes and relatives . A global overview of the eukaryotic tree of life , including the position of ochrophytes relative to other lineages is shown in Figure 1—figure supplement 1 . ( Panel B ) shows the number of inferred positive control HPPGs ( i . e . , HPPGs encoding proteins with experimentally confirmed plastid localisation , or unambiguously plastid function ) and negative control HPPGs ( i . e . , HPPGs encoding proteins with no obvious plastid-targeted orthologues encoded in ochrophyte genomes , but found in haptophyte and cryptomonad genomes ) detected as plastid-targeted in different numbers of ochrophyte lineages using ASAFind ( i ) and HECTAR ( ii ) . The blue bars show the number of positive controls identified to pass a specific conservation threshold , plotted against the left hand vertical axis of the graph , while the red bars show the number of negative controls that pass the same conservation threshold , plotted against the right hand vertical axis of the graph . The number of different sub-categories included in each conservation threshold is shown in a heatmap below the two graphs , with the specific distribution for each bar in the graph shown in the aligned cells directly beneath it . Each shaded cell corresponds to an identified orthologue in one sub-category of a particular ochrophyte lineage: orange cells indicate presence of chrysistan sub-categories; light brown cells the presence of hypogyristean sub-categories; and dark brown cells the presence of diatom sub-categories . In each graph , black arrows label the conservation thresholds inferred to give the strongest separation ( as inferred by chi-squared P-value ) between positive and negative control sequences . The table ( iii ) tabulates the three conservation patterns identified as appropriate for distinguishing probable ancestral HPPGs from false positives . ( Panel C ) shows the complete HPPG assembly , alignment and phylogenetic pathway used to identify conserved plastid-targeted proteins . ( Panel D ) tabulates the number of HPPGs built using ASAFind and HECTAR predictions , and the number of non-redundant HPPGs identified in the final dataset . The final total represents the pooled total of non-redundant HPPGs identified with both ASAFind and HECTAR . DOI: http://dx . doi . org/10 . 7554/eLife . 23717 . 00310 . 7554/eLife . 23717 . 004Figure 1—figure supplement 1 . Overview of eukaryotic diversity . This figure , adapted from a previous review ( Dorrell and Howe , 2012a ) , profiles the diversity of different eukaryotic nuclear lineages . Each grey ellipse corresponds to one major clade , or ‘supergroup’ of eukaryotes . A brown ellipse within the stramenopile clade delineates the ochrophyte lineages . Dashed lines denote uncertain taxonomic relationships . For each taxon , a type species ( defined either by the presence of a complete genome , extensive transcriptome library , or of particular anthropic significance ) is given in brackets . Taxa that lack plastids are labelled in grey , and taxa with plastids are shaded according to the evolutionary origin of that plastid lineage . DOI: http://dx . doi . org/10 . 7554/eLife . 23717 . 00410 . 7554/eLife . 23717 . 005Table 1 . Glossary Box . A schematic figure of eukaryotic taxonomy , showing the evolutionary origins of nuclear and plastid lineages , adapted from previous reviews ( Dorrell and Howe , 2012a ) , is shown in Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 23717 . 005Complex plastidsPlastids acquired through the endosymbiosis of a eukaryotic alga . These include secondary plastids of ultimate red algal origin ( such as those found in ochrophytes , haptophytes and cryptomonads ) , secondary plastids derived from green algae ( such as those found in euglenids or chlorarachniophytes ) , or tertiary plastids such as those found in dinotoms and certain other dinoflagellates ( resulting from the endosymbioses of eukaryotic algae that themselves contain plastids of complex origin ) . CASH lineagesThe four major lineages of algae with plastids of secondary or higher red origin , that is to say Cryptomonads , Alveolates ( dinoflagellates , and apicomplexans ) , Stramenopiles , and Haptophytes . StramenopilesA diverse and ecologically major component of the eukaryotic tree , containing both photosynthetic members ( the ochrophytes ) , which possess complex plastids of red algal origin , and aplastidic and non-photosynthetic members ( e . g . oomycetes , labyrinthulomycetes , and the human pathogen Blastocystis ) , which form the earliest-diverging branches . It is debated when within stramenopile evolution the extant ochrophyte plastid was acquired . OchrophytesPhotosynthetic and plastid-bearing members of the stramenopiles , including many ecologically important lineages ( diatoms , kelps , pelagophytes ) and potential model lineages for biofuels research ( Nannochloropsis ) . Ochrophytes possess plastids of ultimate red origin , and form the most significant component of eukaryotic marine phytoplankton ( Dorrell and Smith , 2011; de Vargas et al . , 2015 ) . HaptophytesSingle-celled , photosynthetic eukaryotes , possessing complex plastids of ultimate red origin . Some haptophytes ( the coccolithophorids ) are renowned for their ability to form large blooms ( visible from space ) , and to form intricate calcareous shells ( Dorrell and Smith , 2011; Bown , 1998 ) , which if deposited on the ocean floor go on to form a major component of limestone and other sedimentary rocks . HPPG‘Homologous plastid protein group’ . Proteins identified in this study to possess plastid-targeting sequences that are homologous to one another , as defined by BLAST-based HPPG assembly and single gene phylogenetic analysis . Following their acquisition , plastids have undergone a number of evolutionary changes that bound them more intricately with the biology of the host . These include the transfer of plastid-derived genes to the host nucleus ( Dorrell and Howe , 2012a; Ruck et al . , 2014; Stegemann et al . , 2003 ) and the targeting of proteins encoded within the nucleus to the plastid ( Nowack and Grossman , 2012; Kleffmann et al . , 2004 ) . Previous studies have shown that many plastid-targeted proteins are not derived from the genomes of the corresponding endosymbiont lineage ( Curtis et al . , 2012 ) . Proteins encoded by genes acquired from other sources , such as laterally acquired genes ( Qiu et al . , 2013; Morse et al . , 1995 ) or previous endosymbiotic organelles historically possessed by the host ( Dorrell and Howe , 2015 , 2012b ) , or proteins that have been repurposed from endogenous host organelles ( Fast et al . , 2001; Harper and Keeling , 2003 ) have important roles in supporting the biology of plastid lineages . Other gene transfer events , e . g . from food sources ( Nowack et al . , 2016 ) , bacterial symbionts ( Dunning Hotopp et al . , 2007 ) , viruses ( Gornik et al . , 2012 ) , or diazotrophic non-plastid cyanobacterial endosymbionts ( Prechtl et al . , 2004; Thompson et al . , 2012 ) might have also played major roles in the evolution of the diverse range of plastid proteins observed today . It nonetheless remains largely unknown which proteins had the most fundamental roles in establishing current plastid lineages ( Dorrell and Howe , 2012a ) , i . e . , which plastid proteins represent the ancestral components of plastid-targeted proteomes . Ochrophytes represent an excellent system in which to reconstruct the origins of plastid proteomes . Firstly , plastid-targeting sequences in different ochrophytes are relatively well conserved , enabling in silico prediction of plastid-targeted proteins from a wide range of different species ( Gruber et al . , 2015; Gschloessl et al . , 2008 ) , in contrast to plastid-targeting sequences within archaeplastid lineages , which are extremely variable ( Fuss et al . , 2013; Suzuki and Miyagishima , 2010 ) . Secondly , compared to other CASH lineages ( haptophytes , cryptomonads , and dinoflagellates ) , ochrophytes represent an extremely well characterised system for experimental and bioinformatic investigation , with ( to date ) eleven complete genomes , and transcriptome libraries available for over 150 species through MMETSP ( the Marine Microeukaryote Transcriptome Sequencing Project ) and through other sources ( Mock et al . , 2017; Keeling et al . , 2014 ) . Reliable transformation and other manipulation strategies are also available for multiple species , such as the model diatom Phaeodactylum tricornutum ( Siaut et al . , 2007; Takahashi et al . , 2007; Radakovits et al . , 2013 ) . Thirdly , the origin of the ochrophyte plastid is an evolutionarily valuable topic to understand . It is currently not known when the ochrophyte plastid was acquired: whether it originated recently , predates the radiation of aplastidic stramenopile relatives ( Stiller et al . , 2014; Aleoshin et al . , 2016; Derelle et al . , 2016 ) , or was acquired prior to the divergence of stramenopiles from their closest relatives , the alveolates ( Janouskovec et al . , 2010 ) . Verifying a late origin for the ochrophyte plastid would thus enable insights into the cellular changes that accompany the transition from a solely heterotrophic to a phototrophic lifestyle ( Aleoshin et al . , 2016; Derelle et al . , 2016 ) , which is currently not possible for archaeplastids ( Burki et al . , 2016; Cavalier-Smith et al . , 2015 ) , and difficult for haptophytes and cryptomonads , in which these relatives respectively remain unknown or understudied at a genomic level ( Burki et al . , 2016; Yabuki et al . , 2014 ) . It has additionally been proposed , based on the presence of large numbers of genes of putative green algal origin in diatom genomes ( Frommolt et al . , 2008; Petersen et al . , 2006 ) , that the ancestor of ochrophytes once possessed a green algal endosymbiont , which was subsequently replaced via the serial endosymbiosis of a red algal-derived plastid ( Dorrell and Smith , 2011; Moustafa et al . , 2009 ) . This hypothesis remains controversial ( Ku et al . , 2015; Woehle et al . , 2011; Deschamps and Moreira , 2012 ) , in particular due to issues associated with the distinction of genes of red and green algal origins in ochrophyte genomes ( Matsuzaki et al . , 2004; Collén et al . , 2013; Qiu et al . , 2015 ) . A final evolutionary suggestion regarding ochrophytes is that they have acted as endosymbiotic donors into other CASH lineages . One recent study proposed that haptophytes possess plastids acquired via the endosymbiosis of an ochrophyte ( Stiller et al . , 2014 ) , although the exact identity of this endosymbiotic acquisition remain unresolved . Characterising the ancestral ochrophyte plastid proteome might therefore help answer major questions about the ways in which plastids become established in the host cell , and provide valuable insights into the origins and diversification of other ecologically important algal lineages . In this study , we present an experimentally verified in silico reconstruction of the proteins targeted to the plastid of the last common ochrophyte ancestor . We show that this ancestral plastid-targeted proteome was an evolutionary mosaic , containing 770 proteins from a range of different sources . Our dataset indicates that the ochrophyte plastid was acquired late in stramenopile evolution , following the divergence of extant aplastidic relatives , that plastid-targeted proteins of green algal origin played a significant role in its origin , and that there has been bidirectional integration of the biology of the ochrophyte host and plastid proteomes , such as the ancient recruitment of proteins from both host and endosymbiont to dually support the biology of the plastid and mitochondria . Finally , we show evidence for an ancient endosymbiosis of a specific ochrophyte lineage , an ancestor of the pelagophytes and dictyochophytes , by a common ancestor of the haptophytes , which we propose- based on discrepancies between the origins of the haptophyte plastid proteome and genome- reveals a possible serial endosymbiosis event early in haptophyte evolution , preceding the origins of the current haptophyte plastid . Our work resolves several long-standing questions of ochrophyte evolution , and provides new insights into the origins and diversification of CASH lineages as a whole .
We developed an in silico pipeline for identifying putatively ancestral plastid-targeted proteins across the ochrophytes ( Figure 1 ) . We screened a large composite library , comprising eleven different ochrophyte genomes , together with transcriptome data from a further 158 ochrophyte species ( Table S1- sheet 1 [Dorrell et al . , 2016] ) using the ochrophyte plastid targeting predictors ASAFind ( Table S2- sheet 1 [Dorrell et al . , 2016] ) ( Gruber et al . , 2015 ) and HECTAR ( Table S3- sheet 1 [Dorrell et al . , 2016] ) ( Gschloessl et al . , 2008 ) . Sequences with predicted plastid localisation were binned into eleven taxonomic sub-categories within three major groups ( chrysista , hypogyrista , and diatoms ) based on recent multigene phylogenies ( Derelle et al . , 2016 ) ( Figure 1 , panel A; Figure 1—figure supplement 1 ) , then assembled by sequence similarity into homologous plastid-targeted protein groups ( HPPGs , Materials and methods ) . We next tested the level of conservation best able to identify truly ancestral HPPGs . We selected three patterns of conservation that identified the largest number of HPPGs from a positive control dataset of proteins with previously identified plastid-associated functions , and minimised the number identified from a negative control dataset of HPPGs generated using seed sequences from three other published CASH lineage genomes , for which no plastid-targeted orthologues were detected in any ochrophyte genome sequence ( Materials and methods; Table S2- sheet 2 , sections 1–2; Table S3- sheet 2 , sections 1–2 [Dorrell et al . , 2016] ) . The selected conservation patterns were: the presence of the protein in a majority ( ≥2/3 ) of chrysistan sub-categories and a majority of either diatom ( ≥3/5 ) or hypogyristean ( ≥2/3 ) sub-categories; or presence in at least one chrysistan sub-category and a majority of both diatoms and hypogyristea ( Figure 1 , panel B ) . We extracted HPPGs matching the conservation patterns defined above and verified their monophyly within ochrophytes via alignment and single-gene trees ( Figure 1 , panel C; Table S4- sheet 1 [Dorrell et al . , 2016] ) . From this , we identified 770 proteins that were probably targeted to the ancestral ochrophyte plastid ( Figure 1 , panel D; Table S4- sheet 2 [Dorrell et al . , 2016] ) . This dataset is significantly enriched in proteins from within the positive control dataset and contains significantly fewer proteins from the negative control dataset than would be expected through random assortment ( chi-squared test , p<1×10−10; Figure 1 ) , confirming its specificity towards probable ancestral plastid-targeted proteins . We wished to verify that the ancestral ochrophyte plastid-targeted proteins inferred from the in silico pipeline are genuinely plastid-targeted . 106 of our inferred ancestral HPPGs include a P . tricornutum protein with prior experimental plastid localization , or unambiguous plastid function ( Figure 1 , panel D ) , but the remainder do not . We selected ten proteins for experimental localisation ( Figure 2 , panel A; Table S5 [Dorrell et al . , 2016] ) . These were chosen on the basis of having only non-plastid annotations on the first 50 BLAST hits against the NCBI nr database excluding ochrophytes , hence lack specific a priori evidence for a plastid localization . In each case , all of the ochrophyte protein sequences within the alignment had a well conserved central domain , and a highly variable N-terminal domain of between 30 and 50 amino acids containing an ASAFAP motif , consistent with a conserved plastid targeting sequence ( Gruber et al . , 2015 ) ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 23717 . 006Figure 2 . Verification of unusual ancestral plastid-targeted proteins . ( Panel A ) lists the ten proteins selected for experimental characterisation and their most probable previous localisation prior to their establishment in the ochrophyte plastid , based on the first 50 nr BLAST hits . Exemplar alignments and single-gene tree topologies for some of these proteins are shown in Figure 2—figure supplements 1–4 . ( Panel B ) shows the localisation of GFP constructs for copies of two proteins with an unambiguous plastid localisation ( a pyrophosphate-dependent PFK , which localises to the pyrenoid , and a novel plastid protein , with cosmopolitan distribution across the plastid ) and one protein with a periplastid localisation ( a predicted peroxisomal membrane protein ) from the diatom Phaeodactylum tricornutum , the diatom endosymbiont of the dinoflagellate Glenodinium foliaceum and the eustigmatophyte Nannochloropsis gaditana , expressed in P . tricornutum . All scale bars = 10 μm . Expression constructs for seven additional P . tricornutum proteins and three additional N . gaditana proteins with multipartite plastid localisations are shown in Figure 2—figure supplements 5 and 6 , and control images ( wild-type cells , and cells expressing untargeted eGFP ) are shown in Figure 2—figure supplement 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 23717 . 00610 . 7554/eLife . 23717 . 007Figure 2—figure supplement 1 . Exemplar ochrophyte plastid protein alignments . This figure shows untrimmed GeneIOUS alignments for two ancestral HPPGs of unusual provenance . In each case the full length of the protein ( labelled i ) and N-terminal region only ( ii ) are shown , demonstrating the broad conservation of the N-terminus position . Sequences for which exemplar targeting constructs ( Phaeodactylum tricornutum , Nannochloropsis gaditana , Glenodinium foliaceum ) were generated are shown at the top of each alignment . DOI: http://dx . doi . org/10 . 7554/eLife . 23717 . 00710 . 7554/eLife . 23717 . 008Figure 2—figure supplement 2 . Tree of ochrophyte glycyl-tRNA synthetase sequences . This tree shows the consensus unrooted Bayesian topology for a 95 taxa x 487 aa alignment of glycyl tRNA synthetase sequences . The font colour of each sequence corresponds to the taxonomic origin ( see legend below for details ) and are labelled with the taxonomic identifiers previously defined in Table S1 . Sequences labelled with chl_ possess apparent plastid targeting sequences recognisable by CASH lineage plastids . The ancestral ochrophyte plastidic isoform , of apparent chlamydiobacterial origin , is labelled with a blue ellipse . Black circles at each node denote posterior probabilities of 1 . 0 in Bayesian inferences with three different substitution matrices ( GTR , Jones , and WAG ) , and grey circles indicate posterior probabilities of 0 . 8 with at least two of these matrices . Support values for all remaining nodes , is provided using both Bayesian analysis ( above line ) and RAxML tree ( below line ) , using three substitution matrices , as defined in the figure legend . DOI: http://dx . doi . org/10 . 7554/eLife . 23717 . 00810 . 7554/eLife . 23717 . 009Figure 2—figure supplement 3 . Tree of ochrophyte pyrophosphate dependent phosphofructo-1- kinase sequences . This tree shows the consensus Bayesian topology inferred for a 94 taxa x 449 aa alignment of pyrophosphate-dependent PFK , with taxa and support values shown as per Figure 2—figure supplement 2 . The ancestral ochrophyte plastid isoform , of probable aplastidic stramenopile origin , is labelled with a cyan ellipse . DOI: http://dx . doi . org/10 . 7554/eLife . 23717 . 00910 . 7554/eLife . 23717 . 010Figure 2—figure supplement 4 . Tree of a novel ochrophyte plastid-targeted protein . This tree shows the consensus Bayesian topology inferred for a 16 taxa x 103 aa alignment of a plastid-targeted protein seemingly restricted to ochrophytes and one dinoflagellate lineage . Taxa are labelled and support values are shown as per Figure 2—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 23717 . 01010 . 7554/eLife . 23717 . 011Figure 2—figure supplement 5 . Multipartite Phaeodactylum plastid-targeted proteins . This figure shows the localisation of GFP overexpression constructs for copies of seven proteins from the diatom Phaeodactylum tricornutum that are of non-plastid origin , but show multipartite localization to the plastid and one other organelle ( the mitochondria , or in the case of the ‘ER heat shock protein’ to the endoplasmic reticulum ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23717 . 01110 . 7554/eLife . 23717 . 012Figure 2—figure supplement 6 . Heterologous expression constructs of multipartite plastid-targeted proteins . This figure shows the localisation of GFP overexpression constructs for copies of two proteins from the dinotom Glenodinium foliaceum ( Panel A ) , and three proteins from the eustigmatophyte Nannochloropsis gaditana ( Panel B ) that are of non-plastid origin , but show multipartite localisation to the plastid and one other organelle , per Figure 2—figure supplement 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 23717 . 01210 . 7554/eLife . 23717 . 013Figure 2—figure supplement 7 . Exemplar control images for confocal microscopy . This figure shows fluorescence patterns for wild-type Phaeodactylum tricornutum cells ( i ) , and transformant Phaeodactylum cells expressing GFP that has not been fused to any N-terminal targeting sequence ( ii ) , both visualised under the same conditions used for all other cultures . DOI: http://dx . doi . org/10 . 7554/eLife . 23717 . 013 The selected proteins included five aminoacyl-tRNA synthetases that yielded BLAST top hits only against enzymes with cytoplasmic annotations , or of probable prokaryotic origin ( Figure 2—figure supplement 2 ) . Also included were a GroES-type chaperonin of inferred mitochondrial origin , an Hsp90-type chaperonin of inferred endoplasmic reticulum origin and a pyrophosphate-dependent phosphofructokinase , which is related to cytosolic enzymes from other lineages ( Figure 2—figure supplement 3 ) , and is distinct from the ATP-dependent phosphofructokinases used by primary plastid lineages ( Smith et al . , 2012 ) . The Mpv17 membrane protein is most closely related to enzymes with peroxisomal functions and localisation ( Wolfe-Simon et al . , 2006; Gillard et al . , 2008 ) , but lacks any identifiable peroxisomal targeting sequence ( PSL , KRR , or a PTS1 motif ) ( Ramirez et al . , 2014 ) in its C-terminus . Finally , a protein ( 'Novel protein one' ) that lacks any conserved domains , and yielded no BLAST matches outside of the ochrophytes below an expect value of 1 × 10−05 ( except for one dinoflagellate sequence ) , was selected for localisation characterisation ( Figure 2—figure supplement 4; Table S5 [Dorrell et al . , 2016] ) . We generated C-terminal GFP-fusion constructs for each of these proteins using P . tricornutum genes and transformed wild-type P . tricornutum ( Figure 2 , panel B; Figure 2—figure supplement 5; Table S5 [Dorrell et al . , 2016] ) . In each case , we identified GFP fluorescence associated with the plastid . In one case ( the peroxisomal membrane protein; Figure 2 , panel B ) , the GFP accumulated in a ring around the plastid equator , consistent with a periplastid compartment ( PPC ) localisation ( Matari and Blair , 2014; Tanaka et al . , 2015a ) . In other cases ( such as the five aminoacyl-tRNA synthetases , Figure 2—figure supplement 5 ) , the GFP signal localised both within and external to the plastid , consistent with a multipartite localisation within the cell . However , in all cases the proteins tested were at least partially targeted to the plastid . We additionally generated heterologous GFP fusion constructs for five of the proteins using sequences from the ‘dinotom’ Glenodinium foliaceum , a dinoflagellate alga that harbours permanent endosymbionts of diatom origin ( Dorrell and Howe , 2015; Imanian et al . , 2010 ) , and the eustigmatophyte Nannochloropsis gaditana , which as a member of the ‘PESC clade’ is distantly related to P . tricornutum on the ochrophyte tree ( Derelle et al . , 2016 ) . We expressed these constructs in P . tricornutum ( Figure 2 , panel B; Figure 2—figure supplement 6 ) , and , in each case , detected plastid-localized GFP fluorescence similar to the patterns observed with the P . tricornutum gene constructs . Overall , our data therefore supports that the ancestral HPPG dataset consists of genuinely conserved plastid-targeted proteins , rather than misidentified proteins of non-plastid function .
In this study , we have reconstructed an experimentally verified dataset of 770 plastid-targeted proteins that were present in the last common ancestor of all ochrophytes ( Figures 1 and 2 ) . Our dataset accordingly provides windows into the evolutionary origins of the ochrophyte plastid lineage . These include evidence for a green algal contribution to ochrophyte plastid evolution and a late acquisition of the ochrophyte plastid following divergence of the ochrophyte lineage from oomycetes ( Figures 3 and 4 ) . Although each of these findings have been previously suggested by studies of whole stramenopile genomes ( Moustafa et al . , 2009; Stiller et al . , 2009 ) our data represent to our knowledge the first large-scale verification from studies of plastid targeted proteins for both of these important events in the origins of the ochrophyte plastid . The relatively late origin of the ochrophyte plastid is particularly interesting as molecular divergence estimates place the ochrophytes as diverging from the oomycetes no more than 90 million years prior to the radiation of ochrophyte lineages ( Brown and Sorhannus , 2010; Matari and Blair , 2014 ) . Assuming that these estimates are reliable , our dataset represents some of the earliest proteins to support the ochrophyte plastid following its endosymbiotic uptake . We also provide evidence for widespread mixing of proteins of different evolutionary origin in the ancestral ochrophyte plastid ( Figure 5 ) , including evidence for the formation of new fusion proteins through the recombination of domains of different evolutionary origins ( Figure 6 ) , and a bidirectional interaction between proteins derived from the endosymbiont with proteins from host organelles via dual-targeting ( Figure 7 ) . A schematic outline of these results is shown in Figure 10 . 10 . 7554/eLife . 23717 . 049Figure 10 . Schematic diagram of events giving rise to the ancestral ochrophyte plastid proteome . Each cell diagram depicts a different stage in the ochrophyte plastid endosymbiosis; each protein depicted represents one or more proteins inferred in this study to have been nucleus-encoded and plastid-targeted in the last common ancestor of all ochrophytes . An ancient ochrophyte ancestor , which had already diverged from oomycetes and other aplastidic stramenopile relatives , and which may have possessed a green algal plastid ( A ) , acquired a red lineage plastid via secondary or higher endosymbiosis ( B ) . Both the host and the endosymbiont are likely to have been evolutionary chimeras , possessing proteins encoded by genes acquired from endosymbiotic and/or lateral gene transfer events . Both host and symbiont are additionally likely to have possessed chimeric proteins , generated through the fusion of genes of different evolutionary origins , and a large number of mitochondrial- , ER- and ( in the case of the red endosymbiont ) potentially dual-targeted proteins . Following genetic integration of the red endosymbiont with its stramenopile host , the first ochrophytes ( C ) thus possessed a wide range of proteins of plastid function acquired from different sources , with no apparent functional bias in the types of proteins that were retained from different sources . Chimeric proteins and dual-targeted proteins , either acquired directly from the endosymbiont , or generated de novo , were also widespread features of this ancestral plastid proteome . Detailed information regarding the relationship between ultimate the evolutionary origins of each HPPG , and its presence or absence in other CASH lineages , is provided in Figure 10—figure supplement 1 . A schematic diagram of possible models through which the haptophyte plastid may have originated is shown in Figure 10—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 23717 . 04910 . 7554/eLife . 23717 . 050Figure 10—figure supplement 1 . Complex origins of different ancestral ochrophyte HPPGs . ( Panel A ) shows the evolutionary positions of lineages with histories of secondary endosymbiosis in trees of ancestral ochrophyte HPPGs verified by combined BLAST top hit and single-gene tree analysis to be either of red algal ( i ) or green algal origin ( ii ) . In both cases , in more than half of the constituent trees , haptophyte and cryptomonad sequences resolve as closer relatives to the ochrophytes than the red or green algal evolutionary outgroup , either due to resolving in the ochrophyte HPPG or forming a specific sister-group to the ochrophyte lineages . ( Panel B ) plots the distribution of cryptomonads ( i ) and haptophytes ( ii ) in trees for different categories of ancestral ochrophyte HPPG of verified evolutionary origin . HPPGs of green algal origin more frequently show internal or sister positions for the cryptomonad sequences than all other categories of HPPG , and in more than 50% of cases resolve internal or sister positions for the haptophyte sequences . This might be consistent with a green algal contribution to the endosymbiotic ancestor of cryptomonad , haptophyte and ochrophyte plastids . DOI: http://dx . doi . org/10 . 7554/eLife . 23717 . 05010 . 7554/eLife . 23717 . 051Figure 10—figure supplement 2 . Different scenarios for the origins of haptophyte plastids . This schematic tree diagram shows different possibilities for the origins of the haptophyte plastid as predicted from the data within this study . No inference is made here regarding the ultimate origin of the ochrophyte plastid , although the ochrophyte , cryptomonad and haptophyte plastids are likely to be closely related to one another within the red plastid lineages . First , a common ancestor of the pelagophytes and dictyochophytes was taken up by a common ancestor of the haptophytes ( point 1 ) , yielding a permanent plastid that contributed genes for a large number of plastid-targeted proteins in extant haptophytes . This plastid was subsequently replaced via serial endosymbiosis ( point 2 ) yielding the current haptophyte plastid and plastid genome . This serial endosymbiosis event either involved a close relative of extant cryptomonads ( 2A ) or a currently unidentified species that forms a sister-group in plastid gene trees to all extant ochrophytes , but is evolutionarily distinct from the pelagophytes ( 2B ) . It is possible that the haptophyte plastid may have been acquired through the secondary endosymbiosis of a different lineage of red algae to the ochrophyte , either via a cryptomonad intermediate ( 2C ) or directly ( 2D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23717 . 051 Many questions nonetheless remain to be answered . It remains to be determined whether the in silico prediction facilitated by programmes such as ASAFind and HECTAR are sufficient to enable the identification of all ochrophyte plastid proteins ( Gruber et al . , 2015; Gschloessl et al . , 2008 ) . This is particularly pertinent in the context of dual-targeted proteins , insofar as the dataset of 34 potentially ancestrally dual-targeted proteins identified in this study may not include proteins that are dual-targeted to the plastid and other cellular organelles , such as the ER ( Porter et al . , 2015 ) , cytoplasm ( Pham et al . , 2014 ) , or nucleus ( Krause et al . , 2012 ) . We note also that , based on the fluorescence patterns observed with the exemplar proteins within this study ( Figures 2 and 7 ) , ASAFind and HECTAR may identify proteins targeted to the periplastid compartment , as well as to the plastid stroma . While these periplastid and multipartite proteins probably form an important part of plastid physiology , it will be interesting to dissect the specific signals associated with the targeting of proteins to individual sub-compartments within CASH lineage plastids ( Tanaka et al . , 2015a; Liu et al . , 2016 ) . Another major question concerns the origins of plastid-targeted proteins of green algal origin in ochrophytes . Overall , our data supports the targeting of a significant complement of proteins of chlorophyte origin to the ochrophyte plastid ( Figure 4 ) . It remains to be determined , however , what the exact chlorophyte donor was , and how these genes may have been acquired . It is possible that the green genes were transferred into the ochrophyte lineage via lateral gene transfer , either from a range of different green algal sources or repeatedly from one lineage ( for example , a semi-permanent intracellular symbiont [Dorrell and Howe , 2012a] ) , although neither scenario would explain the bias in green algal genes in ochrophyte genomes towards encoding proteins of plastid function ( Figure 4 , panel D ) . An alternative possibility might be a cryptic green algal endosymbiosis in the evolutionary history of the host , as has been previously suggested ( Dorrell and Smith , 2011; Moustafa et al . , 2009 ) ( Figure 10 ) , or a more convoluted pattern of acquisition . We note , for example , that the green genes identified in our study are not only plastid-targeted across the ochrophytes , but are apparently shared with haptophytes and cryptomonads ( Figure 10—figure supplement 1 ) , which would be equally consistent with them having been present in a common ancestor of the CASH lineage plastid , and relocated to each host nuclear lineage following endosymbiosis ( Figure 10 ) . Thus , pinpointing the exact nature and timing of the green gene transfer into ochrophytes rests not only on more extensive sequencing of deep-branching chlorophyte lineages , but also on characterising the genome composition of the closest aplastidic relatives of extant ochrophytes ( e . g . , Develorapax , Pirsonia [Aleoshin et al . , 2016] ) , and the closest red algal relative of CASH lineage plastids , which remains unknown ( Dorrell and Smith , 2011; Baurain et al . , 2010 ) . We also provide evidence for a chimeric origin of the haptophyte plastid ( Figures 8 and 9 ) . A schematic outline of these results is shown in Figure 10—figure supplement 2 . We have shown that a significant number of plastid-targeted proteins found in haptophytes originate from an ancestor of the pelagophytes and dictyochophytes ( Figure 8 ) . Although it has previously been suggested from studies of nuclear genomes that ochrophyte and haptophyte plastids share a close evolutionary history , ( Stiller et al . , 2014; Miller and Delwiche , 2015 ) it has not previously been shown robustly that the haptophyte plastid resolves at a specific position internal to the ochrophyte lineage . This data supports findings from other studies ( such as a possible origin for the plastids of dinoflagellates and apicomplexans within a member of the PESC clade; Ševčíková et al . , 2015 ) that many of the plastids found within CASH lineages are of tertiary or higher endosymbiotic origin . The at least partial pelagophyte/dictyochophyte origin of the haptophyte plastid is supported by multiple lines of evidence- i . e . , uniquely shared proteins , single-gene tree topologies , BLAST top hit analysis , and analysis of synapomorphies in multigene alignments ( Figure 8 and supplements ) . Alongside the bias of haptophyte genes of hypogyristean origin encoding proteins of plastid function ( Figure 8- panel E ) , these observations argue against these genes having been acquired through multiple independent lateral gene transfer events , and instead support an endosymbiosis event . We note that other studies have shown strong evidence for gene transfers between haptophytes and individual members of the hypogyristea: for example , Stiller et al . have demonstrated a strong enrichment in BLAST top hits against haptophytes , from the genome of the pelagophyte Aureococcus anophagefferens , compared to other ochrophyte genomes ( Stiller et al . , 2014 ) . We additionally note that an ancestral gene transfer from a pelagophyte/dictyochophyte ancestor into the haptophytes is a chronologically realistic scenario: molecular clock estimates place the pelagophytes and dictyochophytes diverging between 300 and 700 million years before present ( Brown and Sorhannus , 2010; Parfrey et al . , 2011 ) , which broadly overlaps with the molecular dates estimated for the radiation of the haptophytes in the same studies ( Brown and Sorhannus , 2010; Parfrey et al . , 2011 ) , and precedes the first haptophyte microfossils , identified ca . 220 million years before the present ( Bown , 1998 ) . Finally , we verify that the evolutionary links between haptophyte and the pelagophyte/dictyochophyte clade in terms of plastid-targeted proteins are not supported by phylogenies of the haptophyte plastid genome ( Figure 9 ) . Other multigene phylogenies of red lineage plastid genomes have similarly demonstrated that the haptophyte plastid genome instead resolves as a sister-lineage either to cryptomonads or to all ochrophytes ( Stiller et al . , 2014; Janouskovec et al . , 2010; Khan et al . , 2007; Le Corguillé et al . , 2009 ) . Furthermore , the structure and content of haptophyte and hypogyristean plastid genomes are dissimilar: for example , haptophyte plastids possess an rpl36 gene that has been laterally acquired from a bacterial donor and is shared with cryptomonad plastids but absent from ochrophytes ( Rice and Palmer , 2006 ) , and ochrophyte plastids no longer retain genes encoding the plastid division machinery proteins minD and minE , which remain plastid-encoded in haptophytes and cryptomonads ( de Vries and Gould , 2017 ) . Finally , extant haptophyte plastids have comparatively large plastid genomes and possess a conventional quadripartite structure ( Green , 2011 ) , whereas sequenced pelagophyte plastids ( the harmful coastal species Aureococcus anophagefferens and Aureoumbra lagunensis , and an uncultured member of the predominantly open ocean genus Pelagomonas ) all have a reduced coding content compared to other photosynthetic ochrophytes , cryptomonads and haptophytes , and have secondarily lost the plastid inverted repeat ( Worden et al . , 2012; Ong et al . , 2010 ) , although it is not yet known whether the plastid genomes of other pelagophyte genera and of dictyochophytes share this reduced structure . The discrepancy between the pelagophyte/dictyochophyte origin of the haptophyte plastid proteome and the clear non-ochrophyte origin of its plastid genome might be explained by several different evolutionary scenarios . One possibility would be a serial endosymbiosis event deep in haptophyte evolutionary history , in which an ancient plastid derived from a pelagophyte/dictyochophyte ancestor was acquired by the haptophyte common ancestor , then replaced subsequently by a plastid of non-ochrophyte origin ( Figure 10—figure supplement 2 ) . This discrepancy , alongside others such as the presence of green algal genes in ochrophytes , bolsters the possibility that serial plastid endosymbiosis has been a widespread component of the evolution of CASH lineage plastids other than the dinoflagellates , in which it is a well established phenomenon ( Dorrell and Howe , 2015; Yamada et al . , 2017 ) . Verifying this scenario , or its alternatives ( such as lateral gene transfer from pelagophyte or dictyochophyte algae into the algal ancestors of the haptophyte plastid ) rests on identifying the exact origin of the current haptophyte plastid genome , and in particular demonstrating that the haptophyte plastid genome originates from within ( rather than forms a sister-group to ) a major lineage of eukaryotic algae other than ochrophytes ( Figure 10—figure supplement 2 ) . For this , sequence data from early-diverging members of the cryptomonads and haptophytes will be particularly important ( Yabuki et al . , 2014; Choi et al . , 2017; Kim et al . , 2011 ) . It also remains to be determined whether other CASH lineage plastids , such as the peridinin-type plastids found in most photosynthetic alveolates , originate within the ochrophytes ( Ševčíková et al . , 2015; Dorrell and Howe , 2015 ) . Similar plastid proteome reconstructions , using bespoke datasets for these species , will be particularly useful in unravelling their disparate evolutionary origins . Overall , our dataset provides valuable and deep insights into the chimeric origins and complex fates of a major group of eukaryotic algae . Further studies using more sensitive pipelines , or using analogous datasets from other major CASH lineages , may elucidate the evolutionary and physiological diversification of plastids across the eukaryote tree of life .
Ancestral plastid-targeted proteins in ochrophytes were identified via a composite pathway , consisting of in silico prediction , identification of conserved proteins using BLAST , alignment , and single-gene tree building . First , the complete protein libraries annotated from eleven ochrophyte genomes ( the diatoms Phaeodactylum tricornutum ( Bowler et al . , 2008 ) , Thalassiosira pseudonana ( Armbrust et al . , 2004 ) , Thalassiosira oceanica ( Lommer et al . , 2012 ) , Fistulifera solaris ( Tanaka et al . , 2015b ) , Fragilariopsis cylindrus , Synedra acus ( Galachyants et al . , 2015 ) , and Pseudonitzschia multiseries; the pelagophyte Aureococcus anophagefferens ( Gobler et al . , 2011 ) ; the eustigmatophytes Nannochloropsis gaditana and Nannochloropsis salina ( Radakovits et al . , 2013; Wang et al . , 2014 ) ; and the kelp Ectocarpus siliculosus ( Cock et al . , 2010 ) ; Table S1- sheet 1 [Dorrell et al . , 2016] ) , were screened using the ochrophyte plastid-targeting predictors ASAFind ( Gruber et al . , 2015 ) ( used in conjunction with SignalP version 3 . 0 ( Bendtsen et al . , 2004 ) ; Table S2 [Dorrell et al . , 2016] ) and HECTAR ( Gschloessl et al . , 2008 ) ( integrated into a Galaxy ( Afgan et al . , 2016 ) instance available at http://webtools . sb-roscoff . fr; Table S3 [Dorrell et al . , 2016] ) . All proteins that were deemed to possess plastid-targeting sequences ( regardless of the confidence score applied by ASAFind [Gruber et al . , 2015] ) were retained for further inspection . Possible conserved plastid-targeted sequences ( i . e . homologous plastid-targeted protein groups , or HPPGs ) were next identified using a customised BLAST protocol . First , a library of non-redundant proteins was generated to serve as seed sequences for further searches . Each plastid-targeted protein identified from ochrophyte genome sequences was searched by BLASTp against a modified Uniref ( Suzek et al . , 2007 ) library , and the expect values for all top hits were extracted , to yield a floating BLAST threshold below which orthologous proteins were identified . All sequences from lineages with a history of secondary endosymbiosis were first removed from the Uniref library in order to avoid the confounding effects of gene transfer from current and former symbionts ( Stiller et al . , 2014; Ševčíková et al . , 2015; Maruyama et al . , 2011; Archibald et al . , 2003 ) . The removed lineages included cryptomonads , centrohelids , telonemids , haptophytes , alveolates , rhizaria , euglenids , and plastid-bearing stramenopiles . All of the ochrophyte genome-derived plastid-targeted proteins were searched against one another by BLAST , and proteins that matched one another with an expect score lower than the first outgroup hit ( or were retrieved as a stronger match than the outgroup hit if the expected values of both were zero ) , and thus likely correspond to different proteins within the same monophyletic plastid protein cluster , were merged . Only one protein was retained as the seed sequence for subsequent growth of each cluster: this was defined first via organism ( in order of preference: P . tricornutum , T . pseudonana , P . multiseries , F . cylindrus , S . acus , A . anophageferrens , E . siliculosus , N . gaditana , N . salina , T . oceanica , F . solaris ) and , where more than one protein was available for a given organism , the protein with the lowest BLAST expect value against the corresponding uniref top hit . Next , plastid-targeted protein sequences were sought from all available ochrophyte sequence data . A search database was built from all eleven completed ochrophyte genomes , 147 ochrophyte sequence libraries from the Marine Microeukaryote Transcriptome Sequence Project ( MMETSP ) ( Keeling et al . , 2014 ) , eleven further ochrophyte transcriptome sequencing projects ( Matasci et al . , 2014; Mangot et al . , 2017; Kessenich et al . , 2014 ) and uniref . Cross-contamination was removed from MMETSP transcriptomes as previously described ( Marron et al . , 2016 ) . Briefly , this procedure compares the nucleotide sequences of contigs assembled from each MMETSP library by pairwise BLAST , and defines a separate cross-contamination threshold for each pair of MMETSP libraries based on their distribution of BLAST percent identities . These distributions should each contain a peak centered on the average nucleotide percent identity of transcripts between the two species . In addition , in the presence of cross-contamination , there should be a second peak at 100% identity . The procedure defines the cross-contamination threshold as the minimum between these two peaks; above the threshold , contigs ( and the proteins predicted from them ) are considered to be potentially cross-contaminated . In total , 2 . 5% of the MMETSP contigs were discarded through this method . A summary of the number of contigs discarded is provided in Table S1- sheet 2 , section 1 ( Dorrell et al . , 2016 ) . Each decontaminated sequence was trimmed at the N-terminus to the first methionine present , and binned into one of eleven different evolutionary categories , based on recent multigene phylogenetic trees for ochrophytes and diatoms ( Derelle et al . , 2016; Sorhannus and Fox , 2012; Yang et al . , 2012; Theriot et al . , 2015 ) ( Figure 1 , panel A; Table S1- sheet 1 [Dorrell et al . , 2016] ) . These consisted of: three chrysistan lineages ( the ‘PX clade’ of phaeophytes , xanthophytes and related lineages; raphidophytes; and the ‘PESC clade’ of pinguiophytes , eustigmatophytes , synchromophytes , and synurophytes/chrysophytes ) , three hypogyristean lineages ( pelagophytes; dictyochophytes; and bolidophytes ) , and five diatom lineages ( the basally divergent genus Corethron; radial centric lineages such as Coscinodiscophytes and Rhizosoleniaceae; the polar centric Thalassiosirales and Skeletonemataceae , which appear to be relatively distantly related to pennate diatoms ( Sorhannus and Fox , 2012; Theriot et al . , 2015 ) ; polar centric lineages such as Odontellids and Chaetocerotales that appear to be more closely related to pennate diatoms ( Sorhannus and Fox , 2012; Theriot et al . , 2015 ) ; and finally all pennate lineages ) . These binned sequences were then searched for plastid-targeted proteins by ASAFind and HECTAR as before . The seed sequences for the resulting non-redundant HPPGs were searched against the enlarged plastid sequence library using BLASTp . Proteins that matched against seed sequences with a lower expect value than the outgroup best hit ( or were retrieved as a stronger match than the outgroup hit if the expected values of both were zero ) , were added to each HPPG . Next , three custom thresholds were defined that were particularly successful in distinguishing probable proteins of true plastid localisation from false positives ( Figure 1 , panel B ) . For this , conservation patterns were selected that maximised the relative enrichment in proteins with unambiguous plastid functions ( i . e . , were annotated to function in photosynthesis , to constitute integral parts of the plastid thylakoid or inner membranes , or corresponded to the expression products of genes that are plastid-encoded in red algae but have been apparently relocated to the ochrophyte nucleus [Green , 2011] or that corresponded to proteins previously verified experimentally to localise to ochrophyte plastids [Gruber et al . , 2015; Gschloessl et al . , 2008; Huesgen et al . , 2013; Grouneva et al . , 2011[ ) , and thus should contain relatively fewer examples of mispredicted proteins within the dataset . At the same time , conservation patterns were selected that minimised the number of HPPGs identified as conserved from a negative control dataset ( consisting of HPPGs assembled using seed sequences from the published genome sequences of the cryptomonad Guillardia theta ( Curtis et al . , 2012 ) or the haptophytes Emiliania huxleyi ( Read et al . , 2013 ) and Chrysochromulina tobin ( Hovde et al . , 2015 ) , and for which no plastid-targeted orthologues were detected in any of the ochrophyte genome sequences used in this study ) . The thresholds corresponded to: orthologues in a majority ( ≥2/3 ) of chrysistan and a majority ( ≥3/5 ) of diatom lineages; a majority of chrysistan and a majority ( ≥2/3 ) of hypogyristean lineages; and at least one chrysistan , and a majority of both hypogyristean and diatom lineages ( Figure 1 ) . All of the HPPGs that passed at least one threshold were extracted , and homology for each HPPG was confirmed individually ( Table S4- sheet 1 [Dorrell et al . , 2016] ) . First , each HPPG was aligned using 20 iterations of MUSCLE v8 ( Edgar , 2004 ) , followed by the in-built alignment programme integrated into GeneIOUS v 4 . 76 ( Kearse et al . , 2012 ) , under the default criteria . Each HPPG alignment was manually inspected , and proteins that failed to align with the genomic sequences , clearly terminated within the conserved region of the protein , or were truncated at the N-terminus by a length of greater than 50 amino acids ( i . e . the approximate length of an ochrophyte plastid-targeting sequence [Gruber et al . , 2015; Huesgen et al . , 2013] ) were removed , following which HPPGs that no longer passed the taxonomic criteria defined for conservation were eliminated ( Table S4- sheet 1 [Dorrell et al . , 2016] ) . Next , each HPPG was enriched with the sequences for the top 50 hits obtained when the seed sequence was searched against the modified uniref library as detailed above , alongside the single best hit for composite transcriptome and genome libraries constructed for 36 eukaryotic sub-categories ( Table S1- sheet 1 [Dorrell et al . , 2016] ) , and realigned against this reference . The transcriptome components of the reference sequence libraries were cleaned of residual contamination as defined above , and 23 individual MMETSP libraries were additionally excluded due to evidence of further contamination ( Table S1- sheet 2 [Dorrell et al . , 2016] ) . Sequences that failed to align were removed , and HPPGs that failed to meet the criteria for conservation following alignment were eliminated ( Table S4- sheet 1 [Dorrell et al . , 2016] ) . Finally , each HPPG was trimmed at the N- and C-termini to ( respectively ) the first residue and last residue visually identified to be conserved in >70% of the sequences in the alignment , corresponding to the probable conserved domain of the protein . Each HPPG was then trimmed with trimAl using the -gt 0 . 5 option ( Capella-Gutiérrez et al . , 2009 ) . 100 trees were calculated for each trimmed alignment using RAxML , with the JTT substitution model + gamma correction ( Stamatakis , 2014 ) . The consensus tree from the 100 bootstrap replicates was manually inspected for the presence of a clade of ochrophyte proteins , containing sufficient sequences to pass the criteria for conservation defined above , that was either monophyletic , or paraphyletic to the inclusion of only one of five different non-ochrophyte groups ( prokaryotes , red algae , green algae , aplastidic stramenopiles , and all other eukaryotes excluding CASH lineages , rhizaria and euglenids; Table S4- sheet 1 [Dorrell et al . , 2016] ) . HPPGs that passed this final stage of analysis were deemed to correspond to ancestrally plastid-targeted proteins ( Table S4- sheet 2 [Dorrell et al . , 2016] ) . All identified plastid-targeted proteins , HPPGs , full aligned HPPGs , and single-gene trees have been made publically accessible through the University of Cambridge dSpace server ( https://www . repository . cam . ac . uk/handle/1810/261421 [Dorrell et al . , 2016] ) . Phaeodactylum tricornutum 1 . 86 ( CCMP2561 ) , Nannochloropsis gaditana CCMP526 , and Glenodinium foliaceum PCC499 were maintained in liquid cultures of f/2 medium supplemented with vitamins , and 100 μg/ ml each of ampicillin , streptomycin , kanamycin and neomycin , in a constant 19°C environment in a 12 hr: 12 hr cycle of 150 μE m−2 s−1 light: dark . P . tricornutum was maintained on an orbital shaker at 100 rpm , while N . gaditana and G . foliaceum were maintained as stationary cultures . Large volume cultures of P . tricornutum ( e . g . cultures grown for transformation by bombardment ) were grown in artificial seawater , supplemented with vitamins but without antibiotics . Total cellular RNA was extracted from c . 30 ml volumes of late log phase culture from each species using a modified Trizol phase extraction and DNase treatment protocol as described elsewhere ( Dorrell and Howe , 2012b ) . Each RNA sample was tested for integrity by gel electrophoresis and quantified by a nanodrop spectrophotometer , and confirmed to be free of residual DNA contamination by direct PCR using universal eukaryotic 18S rDNA primers ( Gachon et al . , 2013 ) . Approximately 200 ng purified RNA from each species was used as the template for cDNA synthesis , using a Maxima First Strand cDNA Synthesis Kit ( Thermo , France ) , following the manufacturer's instructions . Nucleotide sequences encoding plastid-targeted proteins of unusual provenance were identified using the complete genome sequences of Phaeodactylum tricornutum and Nannochloropsis gaditana ( Radakovits et al . , 2013; Bowler et al . , 2008 ) , and the Glenodinium foliaceum CCAP1116/3 transcriptome library assembled as part of MMETSP ( Keeling et al . , 2014; Hehenberger et al . , 2016 ) ( Table S5 [Dorrell et al . , 2016] ) . Two primers were designed for each sequence: a PCR forward primer corresponding to the 5' end of the ORF , and a translationally in-frame PCR reverse primer positioned a minimum of 45 bp into conserved domain of the protein sequence ( Table S5 [Dorrell et al . , 2016] ) . These primers were respectively fused to 5' fragments complementing the 3' end of the P . tricornutum FcpA promoter , and the 5' end of the GFP CDS . For one gene ( the novel plastid protein ) , PCR reverse primers were designed complementary to the 3’ end of the CDS of each gene due to the lack of a verifiable CDD; a full-length PCR reverse primer was additionally designed against the histidyl-tRNA synthetase sequence from Nannochloropsis gaditana due to failure to obtain functional expression from N-terminal constructs ( data not shown ) . High-fidelity PCR products were amplified with each primer pair from the corresponding cDNA product using Pfu DNA polymerase ( Thermo , France ) , per the manufacturer's instructions . In two cases ( Nannochloropsis gaditana peroxisomal membrane protein , and the novel plastid protein ) inserts were amplified from synthetic , codon-optimised constructs , designed to maximise expression levels in Phaeodactylum tricornutum ( Eurofins , France ) . Each product was separated by DNA gel electrophoresis , cut , purified using a PCR gel extraction column kit ( Macherey-Nagel , France ) , quantified using a nanodrop spectrophotometer , and verified by Sanger sequencing ( GATC Biotech , France ) . The purified products were then used for Gibson ligation reactions ( Gibson et al . , 2009 ) ( NEB , France ) , following the manufacturer's instructions , using linearised and DpnI-treated vector sequence generated from the pPhat-eGFP vector ( Siaut et al . , 2007 ) , and transformed into chemically competent Top10 E . coli cells , prior to selection on LB-1% agar plates containing 100 μg/ ml ampicillin . Individual colonies were picked , verified to contain the insert sequence by PCR , and grown as overnight liquid cultures on LB medium supplemented with 100 μg/ ml ampicillin , prior to purification of the plasmids by alkaline lysis and isopropanol precipitation ( Feliciello and Chinali , 1993 ) . Purified plasmids were integrated into P . tricornutum cells via biolistic transformation , using the Biolistic PDS-1000/He Particle Delivery System ( BioRad , France ) , essentially as previously described ( Siaut et al . , 2007; Falciatore et al . , 1999 ) . Colonies obtained from each transformation were transferred to liquid f/2 supplemented with vitamins and 100 μg/ ml zeocin , and were left to recover under the same growth conditions as used for liquid cultures of untransformed cells . Expression of GFP was visualised using a TCS SP8 confocal microscope ( Leica , France ) , an excitation wavelength of 488 nm and emission wavelength interval of c . 510–540 nm . Chlorophyll fluorescence ( using an emission interval of 650–700 nm ) and bright field images were simultaneously visualised for each cell . Wild-type cells that did not express GFP were used to identify the maximum exposure length possible without false detection of chlorophyll in the GFP channel ( Figure 2—figure supplement 7 ) . Possible mitochondrial localisations of dual-targeted proteins were identified by staining cells with approximately 100 mM Mitotracker orange ( Thermo ) , dissolved in filtered seawater , for 25 min under standard culture conditions ( Tanaka et al . , 2015a ) . Cells were rinsed and resuspended in fresh filtered seawater prior to visualisation , using the same conditions as stated above for GFP , and a 548 nm excitation laser and 575–585 nm absorbance window for the Mitotracker signal . To ensure that there was no possible crosstalk between the two signals , negative controls consisting of an unstained GFP-expressing wild-type line , and stained wild-type cells , were used respectively to determine the maximum exposure length possible without ( respectively ) false detection of GFP in the Mitotracker channel , and false detection of Mitotracker in the GFP channel ( Figure 7—figure supplement 1 ) . The most probable evolutionary origins of individual plastid-targeted proteins were identified via the combined products of BLAST top hit analysis and phylogenetic sister-group inference . First , a composite reference sequence library was generated by appending the uniref outgroup library previously used for BLAST-based assembly of ancestral HPPGs , with twenty-two combined eukaryotic transcriptome and genomic libraries of taxa with no suspected history of serial endosymbiosis , which was previously used to enrich each single-gene tree ( Table S1- sheet 1 [Dorrell et al . , 2016] ) . Each sequence within the library was then assigned a taxonomic affinity consisting of one of six lineages ( green algae , red algae , aplastidic stramenopiles , all other eukaryotes , prokaryotes , and viruses ) and one of 48 sub-categories , ( Table S1- sheet 1 , section 1 [Dorrell et al . , 2016] ) . Next , each seed protein sequence within each ancestral HPPG was searched by BLASTp against the composite library , with a threshold e-value of 1 × 10−05 . Sequences were annotated by the lineage and sub-category of the first hit obtained , and by the number of consecutive top hits obtained within the same lineage ( Table S4- sheet 2 , section 2 [Dorrell et al . , 2016] ) . To minimise misidentification due to any residual contamination in individual sequence libraries , only sequences for which the first three or more BLAST hits resolved within the same lineage were deemed to be unambiguously related to that lineage . Sister-group relationships were additionally inferred for each ancestral HPPG from the previously generated single-gene trees ( Table S4- sheet 2 , section 3 [Dorrell et al . , 2016] ) . To ensure that only true sister-group relationships were recorded , and to avoid potential misidentifications of individual sister-group relationships due to species-specific gene transfer or contaminants that had not previously been excluded by screening individual species libraries , only trees in which ochrophytes were monophyletic , ( i . e . , not paraphyletic with regard to any one of the five outgroups ) , for which a single sister-group could be identified ( using the most phylogenetically complex node as the outgroup ) , and for which the sister-group contained at least two monophyletic or paraphyletic sequences , from different sub-categories of the same lineage , were used for subsequent analysis . To identify the probable relationships between ochrophytes and other CASH lineage plastids , each ancestral HPPG tree was enriched with sequences from six different groups of organisms with histories of serial endosymbiosis ( cryptomonads , haptophytes , dinotoms , other alveolates , euglenids , and chlorarachniophytes ) , subdivided into thirteen sub-categories ( Table S1 [Dorrell et al . , 2016] ) . For the cryptomonad , haptophyte and dinotom sequences , as plastid-targeted proteins from these lineages may be identified using targeting predictors trained on diatoms such as HECTAR ( Aleoshin et al . , 2016 ) and ASAFind ( Gruber et al . , 2015; Gschloessl et al . , 2008 ) , each of the HPPGs initially generated was enriched with plastid-targeted sequences from each cryptomonad , haptophyte and dinotom sub-category identified by in silico prediction with these programmes ( Table S2- sheet 1; Table S3- sheet 1 [Dorrell et al . , 2016] ) . The position of each group of organisms within the tree was then annotated as falling into one of eight different categories , four of which were internal to the ochrophytes ( diatoms; hypogyristea; chrysista; or an ambiguous internal position ) and four of which were external to the ochrophytes ( as an immediate sister-group to all ochrophytes prior to the first outgroup lineage previously identified; within the red algae; within the green algae; and at any other position external to the ochrophytes; Table S4- sheet 2 , sections 5–6 [Dorrell et al . , 2016] ) . To minimise the incorporation of contaminant and non-plastid sequences , tree positions were only recorded if the branch containing sequences from that particular lineage included at least two of the sub-categories considered ( for alveolates , cryptomonads , and haptophytes ) , contained at least one predicted plastid-targeted sequence ( for dinotoms , cryptomonads and haptophytes ) , and for which only one category could be applied ( i . e . , the tree only contained one evolutionarily distinct group for each lineage , which could be unambiguously allocated one category over all others ) . Each tree annotation was repeated three times independently , and only tree annotations that were recorded consistently in each case were retained for further analysis . To identify plastid-targeted proteins that were uniquely shared between haptophytes and other lineages , every HPPG initially generated was screened for the inclusion of only two of five different lineages ( diatoms including dinotoms , hypogyristea , chrysista , haptophytes , and cryptomonads; Table S2- sheet 2 , section 3; Table S3- sheet 2 , section 3 [Dorrell et al . , 2016] ) . The frequencies of these proteins were then compared to the numbers expected in a random distribution of all uniquely shared HPPGs across the entire dataset: for example , if half of all uniquely shared HPPGs were shared with diatoms and one other lineage , and half were shared with haptophytes and one other lineage , then one-quarter of all uniquely shared HPPGs should be shared between haptophytes and diatoms . The expected numbers were corrected to take account of the expected frequencies calculated through this approach to be uniquely shared within one lineage only: for example , in the above case , one-quarter of the expected frequency would be allocated to HPPGs uniquely present in diatoms; to correct for this , all remaining expected frequencies of uniquely shared HPPGs would therefore be multiplied by four-thirds ( i . e . one-third of all uniquely shared HPPGs should be shared between haptophytes and diatoms ) . The specific evolutionary relationships associated with haptophyte plastid-targeted proteins incorporated into ancestral HPPGs were investigated using a modified BLAST top hit technique . Firstly , all of the plastid-targeted proteins assembled into each ancestral HPPG were extracted and separated into each separate sub-category ( Table S13- sheet 1 [Dorrell et al . , 2016] ) . Each sub-category list was then reduced to only leave one , randomly selected sequence per HPPG ( Table S13- sheet 2 [Dorrell et al . , 2016] ) . Finally , each sequence retained in the reduced list was searched by BLAST against a composite library , consisting of the library previously used for outgroup top hit analysis , enriched with all of the plastid-targeted proteins identified for ochrophytes , haptophytes and cryptomonads , except for those that corresponded to the same particular lineage as the query sequence ( Table S13- sheets 1 , 3 [Dorrell et al . , 2016] ) . For example , in the case of haptophytes , plastid-targeted sequences that had been separated into three individual categories ( pavlovophytes , prymnesiales , and isochrysidales [Simon et al . , 2013] ) were searched against a composite library consisting of all outgroup sequences , and plastid-targeted sequences from diatoms , hypogyristea , chrysista , and cryptomonads , but excluding haptophytes . BLAST top hit analysis was then performed as described above ( Table S13- sheets 1 , 3 [Dorrell et al . , 2016] ) . Finally , to enable the identification of genes with consistent results from multiple analyses , the lineage of the BLAST top hit was compared to the lineage of the haptophyte sister-group in the single-gene tree analysis ( Table S4- sheet 2 , section 5; Table S13- sheet 4 [Dorrell et al . , 2016] ) . To identify residues that are uniquely shared between ochrophytes and other lineages , multigene datasets were constructed of a ) ancestral HPPGs of green algal origin , and b ) ancestral HPPGs for which haptophytes show origins within the ochrophytes . To minimise the incorporation of sequences of misidentified origin , in each case only the HPPGs for which the proposed evolutionary origin were identified both by BLAST top hit and single-gene tree analysis were included . To avoid introducing artifacts due to lineage-specific gene transfers , paralogy events , or other phylogenetic incongruencies that could otherwise bias the eventual results ( Qiu et al . , 2012; Leigh et al . , 2008 ) , the single-gene tree generated for each HPPG was manually inspected to exclude any that contain multiple clades ( defined as monophyletic groups containing more than one sequence from a particular lineage , separated from one another by at least two sequences from outside that particular lineage ) for each of the major lineages of interest within the tree: This left datasets consisting of 32 HPPGs for which the ochrophytes were of clear green algal origin , and 37 HPPGs in which the haptophytes were of clear ochrophyte origin , with no conflicting phylogenetic signal . The rationale for inclusion and exclusion of each HPPG in each analysis is presented in Table S6 , sheets 1 and 3 ( Dorrell et al . , 2016 ) . Next , to eliminate individual sequences remaining within each HPPG that might have arisen through species-specific gene transfer or contamination events , each trimmed sequence within each approved alignment was inspected using a composite BLAST approach . First , each sequence was searched against a composite library containing all uniref , jgi and MMETSP sequences from every lineage within the tree of life , and the top ten hits were tabulated for each sequence . In each case , only sequences for which at least the first three hits were of the same lineage as that of the query were retained . For the haptophyte multigene alignment , the ochrophytes were separately analysed as each of the three component lineages ( chrysista , hypogyristea , and diatoms ) , which is to say that a query obtained from a member of the hypogyristea would only be retained if the first three BLAST top hits originated from other hypogyristean sequences , rather than other ochrophytes . Next , each of the component sequences within each cleaned alignment were searched against all other component sequences within the same alignment using BLASTp , and the top ten hits within the alignment were ranked . In each case , sequences were only approved for incorporation into the multigene dataset if the first non-self hit was to a different sub-category within the same lineage , e . g . if a query sequence from a red alga yielded a top hit against a red algal sequence from a different red sub-category . To allow for possible cases of paraphyly and/or absence of sequences within each alignment , the following modifications were applied: Tabulated outputs for each BLAST analysis are provided in Table S6 , sheets 2 and 4 . Finally , each dataset was reduced to leave only one randomly selected sequence for each given sub-category within each HPPG alignment . The number of residues that were uniquely shared between ochrophytes and green algae in the green gene dataset , and haptophytes and ochrophytes in the haptophyte dataset , were then tabulated ( Table S7 [Dorrell et al . , 2016] ) . Briefly , residues were inferred to be uniquely shared between ochrophytes and green algae if they were present in at least 2/3 of the ungapped ochrophyte sequences , one or more green algal sequence , and if none of the red algal or glaucophyte sequences shared the residue in question , but at least one of these sequences had a non-matching ( i . e . non-gapped ) residue at that position ( Table S7- sheet 1 , section 2 [Dorrell et al . , 2016] ) . Similarly , residues were inferred to be uniquely shared between ochrophytes and haptophytes if they were present in at least 2/3 of the ungapped haptophyte sequences , one or more ochrophyte sequence , and if none of the green algal , red algal , glaucophyte or cyanobacterial sequences shared the residue in question , but at least one of these sequences had a non-matching ( i . e . , non-gapped ) residue at that position ( Table S7- sheet 2 , section 2 [Dorrell et al . , 2016] ) . The origin point of each uniquely shared residue was then inferred by comparison to reference topologies respectively of green algae ( Leliaert et al . , 2011 ) and of ochrophytes ( per Figure 1 ) . Residues were assumed to have originated in a common ancestor of a particular clade if that clade contained more lineages with matching than non-matching or gapped residues ( Table S7- sheets 1–2 , section 5 [Dorrell et al . , 2016] ) . A second analysis was additionally performed in which all gapped residues were deemed to be matching , to identify the earliest possible origin point for each uniquely shared residue , taking into account secondary loss ( Ku et al . , 2015; Qiu et al . , 2015 ) and absence of sequences from each alignment ( Woehle et al . , 2011; Deschamps and Moreira , 2012 ) . Two libraries of non-redundant gene families that were broadly conserved across ochrophytes or haptophytes , and thus might represent gene products of the ancestral genomes of these lineages , were generated using a similar BLAST-based assembly pipeline as used to construct HPPGs ( Table S8; Table S14 [Dorrell et al . , 2016] ) . Ochrophyte gene families were deemed to be conserved if orthologues were detected in one of three different patterns of ochrophyte sub-categories previously defined to correspond to ancestral plastid-targeted proteins ( Figure 1 , panel B; Table S8- sheet 1 , section 3 [Dorrell et al . , 2016] ) . Haptophyte gene families , built through a similar pipeline using seed sequences from the Chrysochromulina tobin and Emiliania huxleyi genomes ( Read et al . , 2013; Hovde et al . , 2015 ) , were deemed to be ancestral if orthologues were identified in at least two of the three haptophyte sub-categories considered ( pavlovophytes , prymnesiales , and isochrysidales; Table S14- sheet 1 , section 3 [Dorrell et al . , 2016] ) . The most probable evolutionary origin of each gene family was inferred by BLAST top hit analysis of the seed sequence ( Table S8- sheets 1 , 2; Table S14- sheets 1 , 2 [Dorrell et al . , 2016] ) . Ochrophyte sequences were searched against the composite uniref + MMETSP library used to previously identify the most likely outgroup to each ancestral plastid-targeted protein ( Table S8- sheet 1 , section 6 [Dorrell et al . , 2016] ) , while haptophyte sequences were searched against the enriched library that also contained all ochrophyte and cryptomonad sequences , to enable the distinction of proteins of probable CASH lineage plastid origin from proteins that had evolved through independent gene transfer events between haptophytes and non-CASH lineage organisms ( Table S14- sheet 1 , section 6 [Dorrell et al . , 2016] ) . Targeting preferences for each protein encoded within each gene family were identified using SignalP v 3 . 0 and ASAFind v 2 . 0 ( Dorrell et al . , 2016 ) , and with HECTAR ( Gschloessl et al . , 2008 ) , as previously discussed ( Table S8- sheet 3; Table S14- sheet 3 [Dorrell et al . , 2016] ) . Targeting preferences that were identified in a plurality of sequences and in ≥2/3 of the sequences within each ochrophyte gene family were recorded ( Table S8- sheet 2 , sections 4–5 [Dorrell et al . , 2016] ) . As only three haptophyte sequences were assembled for each ancestral haptophyte gene family , only targeting predictions that were identified in ≥2/3 of the sequences within the HPPG were inferred to be genuine ( Table S14- sheet 2 , sections 4–5 ( Dorrell et al . , 2016] ) . Core plastid metabolism pathways were identified using recent reviews of ochrophyte metabolism , or reviews of homologous plant plastid metabolic pathways where ochrophyte-specific reviews have not yet been published ( Smith et al . , 2012; Green , 2011; Grouneva et al . , 2011; Allen et al . , 2011; Kroth et al . , 2008; Bromke , 2013; Bertrand , 2010; Miret and Munné-Bosch , 2014; Bandyopadhyay et al . , 2008; Shtaida et al . , 2015 ) . The probable function and KOG classification of each HPPG were annotated using the pre-existing annotations associated with seed protein sequence ( if these existed ) , or if not the annotated function of the top uniref hit previously identified by BLAST searches of the seed sequence ( Table S9 [Dorrell et al . , 2016] ) . Expression dynamics for each ancestral HPPG within the genomes of the model diatoms Phaeodactylum tricornutum and Thalassiosira pseudonana were inferred using microarray data integrated into the DiatomPortal server ( Ashworth et al . , 2016 ) ( Table S10- sheets 1 , 2 [Dorrell et al . , 2016] ) . Correlation coefficients were calculated between each pair of P . tricornutum and T . pseudonana genes that were incorporated into an ancestral HPPG , across all microarray libraries within the dataset ( Table S10- sheets 3 , 4 [Dorrell et al . , 2016] ) , with average values being calculated from all pairwise correlations for different evolutionary categories of protein ( Table S10- sheet 5 [Dorrell et al . , 2016] ) . Possible chimeric proteins , resulting from the fusion of proteins of different evolutionary origins , were identified in the dataset using a modified version of a previously published protocol ( Méheust et al . , 2016 ) ( Table S9- sheet 1 , sections 4 , 5; Table S11 [Dorrell et al . , 2016] ) . Each protein within each HPPG was searched using BLASTp against the composite outgroup MMETSP-enriched library , using the same taxonomic classification used for the identification of the evolutionary origin of each seed protein within the dataset , and all hits with an expect value of 1 × 10−05 . Component sequences were then grouped into component families according to the following rule: if two component sequences overlapped by more than 70% of their lengths on the protein composite , they belonged to the same component family . Overlapping and/ or nested component families were additionally merged if one family was included by more than 70% of its length into the other one . Component families were then assigned a broad evolutionary origin corresponding to their taxonomic composition . If the three best component sequences , according to their BLAST bitscore against the composite gene , matched with the same lineage ( e . g . , green algae , red algae , aplastidic stramenopiles , or other eukaryotes ) , the component was considered to have originated from that lineage . Possible dual-targeted proteins were identified within the dataset by screening all possible plastid-targeted proteins with Mitofates , using a cut-off targeting threshold of 0 . 35 ( Fukasawa et al . , 2015 ) , which was inferred to be more effective in identifying experimentally verified ochrophyte mitochondria-targeted proteins ( Figure 7—figure supplement 2 ) ( Gruber et al . , 2015 ) than other threshold values or targeting prediction programmes such as TargetP ( Emanuelsson et al . , 2007 ) or Mitoprot ( Claros , 1995 ) . The default Mitofates positive cutoff value was modified from 0 . 38 to 0 . 35 in order to maximise the capture of experimentally localised mitochondrial proteins , without admitting proteins with unambiguous plastid localisation ( Figure 7—figure supplement 2 ) . As dual-targeting to plastids and mitochondria may be achieved either by distinct protein isoforms resulting from ambiguous targeting peptides or alternative internal translation initiation sites that allow production of mitochondrial targeting sequences ( Xu et al . , 2013; Hirakawa et al . , 2012 ) , each protein was screened with Mitofates using both the full-length N-termini , and N-termini predicted to result from the next downstream methionine within 30 residues . Possible conserved dual-targeted proteins were then identified via the same BLAST-based assembly pipeline and stringency thresholds used to identify probable ancestral HPPGs ( Table S12- sheet 1 [Dorrell et al . , 2016] ) . All putative dual-targeted proteins have been made publically accessible through the University of Cambridge dSpace server ( https://www . repository . cam . ac . uk/handle/1810/261421 ) ( Dorrell et al . , 2016 ) . For the plastid genome phylogenetic analysis , single-gene alignments were constructed by BLAST searches of published red lineage and glaucophyte plastid genomes ( for the gene rich analysis ) or of these genomes plus all MMETSP libraries for the same lineages ( for the taxon rich analysis ) , using the Phaeodactylum tricornutum protein sequence as query and a threshold e-value of 1 × 10−05 , followed by alignment using GeneIOUS v 4 . 76 ( Kearse et al . , 2012 ) , as before . The gene rich analysis included protein sequences from 54 genes that were identified in 22 different non-green lineage plastid genomes while the taxon-rich analysis included 10 different plastid genes that were identified in all 22 plastid genomes and at least 30 different MMETSP libraries ( Keeling et al . , 2014 ) ( Table S15- sheet 1 [Dorrell et al . , 2016] ) . For the taxon-rich analysis , only species that were represented in ≥6/12 of the single-gene alignments were included in the concatenated alignment . Each concatenated alignment was trimmed using trimal ( Capella-Gutiérrez et al . , 2009 ) using the -gt 0 . 8 option . Single-gene alignments for four plastid-targeted proteins predicted to be of polyphyletic origin in ochrophytes ( 3-dehydroquinate synthase , isopropylmalate dehydratase , sedoheptulose bisphosphatase , and shikimate kinase ) were generated using a similar BLAST-based assembly and alignment pipeline as used to verify ancestral plastid-targeted proteins . In this case , all non-redundant ( as inferred by BLAST top hit evalue ) plastid-targeted sequences for each protein identified from ochrophyte genomes were used as independent queries for the identification of plastid-targeted orthologues , 50 uniref top hits , and top hits from the combined MMETSP and genomic libraries from 36 eukaryotic sub-categories , as before . HPPGs were independently generated , aligned and trimmed for each seed sequence; all HPPGs generated for each protein were then merged , realigned and retrimmed using trimAl to generate a single-gene alignment . Single-gene alignments for each of the constituent genes in each concatenated plastid genome tree were generated by splitting the alignment into its component genes . All alignments have been made publically accessible through the University of Cambridge dSpace server ( https://www . repository . cam . ac . uk/handle/1810/261421 ) ( Dorrell et al . , 2016 ) . Trees were inferred for each concatenated and exemplar single-gene alignment ( Table S15- sheet 2 [Dorrell et al . , 2016] ) using the MrBayes and RAxML programmes in-built into the CIPRES web-server ( Stamatakis , 2014; Miller et al . , 2015; Ronquist et al . , 2012 ) . Bayesian trees were inferred using three substitution models ( GTR , Jones , and WAG ) , a minimum of 600000 generations , and an initial burn-in discard value of 0 . 5 . Trees were only utilised if the final convergence statistic between the two chains run was ≤0 . 1 , and tree calculation was automatically stopped if the convergence statistic fell below 0 . 01 . RAxML trees were inferred using three substitution models ( GTR , JTT , and WAG ) with automatic bootstopping , as previously described ( Dorrell et al . , 2017 ) . The best tree topology for each RAxML tree was inferred , and bootstrapping was performed using a burnin value of 0 . 03 . Alternative tree topologies were tested for the RAxML + JTT tree inferred from each concatenated alignment using CONSEL ( Shimodaira and Hasegawa , 2001 ) , under the default conditions . Tree outputs have been made publically accessible through the University of Cambridge dSpace server ( https://www . repository . cam . ac . uk/handle/1810/261421 ) ( Dorrell et al . , 2016 ) . Modified alignments were generated for both of the plastid concatenated multigene datasets from which individual clades of organisms ( diatoms , hypogyristea , chrysista , haptophytes , cryptomonads , red algae , and different combinations of green algae ) had been removed ( Table S15- sheet 2 [Dorrell et al . , 2016] ) . Fast-site removal was performed using TIGER ( Cummins and McInerney , 2011 ) . Site rate evolution characteristics were calculated for each alignment using the -b 100 option , and modified alignments were constructed from which the rate categories corresponding to the fastest evolving 40–50% of sites were serially removed ( Table S15- sheet 2 [Dorrell et al . , 2016] ) . Amino acid composition for each plastid alignment were calculated , and two modified alignments were generated from which glycines ( which in all alignments occur at significantly lower frequencies in ochrophytes than in haptophytes or cryptomonads; chi-squared , p≤0 . 05; Table S16- sheet 3 [Dorrell et al . , 2016] ) , and from which seven amino acids ( alanine , aspartate , glycine , histidine , leucine , asparagine , threonine and valine ) which were found in at least one alignment to occur at significantly different frequencies in ochrophytes compared to haptophytes or to cryptomonads ( p≤0 . 05; Table S16- sheet 3 [Dorrell et al . , 2016] ) had been removed . Trees were inferred for each modified alignment using RAxML with the JTT substitution , and MrBayes with the Jones substitution , and bootstrap calculation as previously described . Modified alignments and tree outputs have been made publically accessible through the University of Cambridge dSpace server ( https://www . repository . cam . ac . uk/handle/1810/261421 ) ( Dorrell et al . , 2016 ) . Uniquely shared residues were manually tabulated for both of the plastid genome multigene alignments ( Table S17 [Dorrell et al . , 2016] ) . For the gene-rich plastid multigene alignment , residues that were present in all haptophyte sequences and only found in a maximum of one other lineage ( red algae , glaucophytes , cryptomonads , diatoms , hypogyristea , or chrysista ) were tabulated ( Table S17- sheet 1 [Dorrell et al . , 2016] ) . For the taxon-rich alignment , to take into account gaps and missing characters , residues were tabulated if they were found in a majority of haptophyte sequences , and one other lineage , as before ( Table S17- sheet 2 [Dorrell et al . , 2016] ) . The total number of residues shared , and uniquely shared , with each non-haptophyte species and lineage are respectively tabulated in Table S17 , sheets 3 and 4 ( Dorrell et al . , 2016 ) . All supporting datasets for this study , including supplementary tables predicted plastid-targeted and dual-targeted protein libraries , single gene and multigene alignments , and tree outputs , have been made publically and freely accessible through the University of Cambridge dSpace server ( https://www . repository . cam . ac . uk/handle/1810/261421 ) ( Dorrell et al . , 2016 ) . | The cells of most plants and algae contain compartments called chloroplasts that enable them to capture energy from sunlight in a process known as photosynthesis . Chloroplasts are the remnants of photosynthetic bacteria that used to live freely in the environment until they were consumed by a larger cell . “Complex” chloroplasts can form if a cell that already has a chloroplast is swallowed by another cell . The most abundant algae in the oceans are known as diatoms . These algae belong to a group called the stramenopiles , which also includes giant seaweeds such as kelp . The stramenopiles have a complex chloroplast that they acquired from a red alga ( a relative of the seaweed used in sushi ) . However , some of the proteins in their chloroplasts are from other sources , such as the green algal relatives of plants , and it was not clear how these chloroplast proteins have contributed to the evolution of this group . Many of the proteins that chloroplasts need to work properly are produced by the host cell and are then transported into the chloroplasts . Dorrell et al . studied the genetic material of many stramenopile species and identified 770 chloroplast-targeted proteins that are predicted to underpin the origins of this group . Experiments in a diatom called Phaeodactylum confirmed these predictions and show that many of these chloroplast-targeted proteins have been recruited from green algae , bacteria , and other compartments within the host cell to support the chloroplast . Further experiments suggest that another major group of algae called the haptophytes once had a stramenopile chloroplast . The current haptophyte chloroplast does not come from the stramenopiles so the haptophytes appear to have replaced their chloroplasts at least once in their evolutionary history . The findings show that algal chloroplasts are mosaics , supported by proteins from many different species . This helps us understand why certain species succeed in the wild and how they may respond to environmental changes in the oceans . In the future , these findings may help researchers to engineer new species of algae and plants for food and fuel production . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"cell",
"biology"
] | 2017 | Chimeric origins of ochrophytes and haptophytes revealed through an ancient plastid proteome |
Calcium ( Ca2+ ) plays an important role in the function and health of neurons . In vertebrate cone photoreceptors , Ca2+ controls photoresponse sensitivity , kinetics , and light adaptation . Despite the critical role of Ca2+ in supporting the function and survival of cones , the mechanism for its extrusion from cone outer segments is not well understood . Here , we show that the Na+/Ca2+ , K+ exchanger NCKX4 is expressed in zebrafish , mouse , and primate cones . Functional analysis of NCKX4-deficient mouse cones revealed that this exchanger is essential for the wide operating range and high temporal resolution of cone-mediated vision . We show that NCKX4 shapes the cone photoresponse together with the cone-specific NCKX2: NCKX4 acts early to limit response amplitude , while NCKX2 acts late to further accelerate response recovery . The regulation of Ca2+ by NCKX4 in cones is a novel mechanism that supports their ability to function as daytime photoreceptors and promotes their survival .
Local and transient changes in cytosolic Ca2+ concentration regulate a wide variety of cellular processes such as synaptic transmission , muscle contraction , and gene expression ( Berridge et al . , 2000 ) . Cytosolic Ca2+ concentration reflects a dynamic balance between influx and efflux pathways , operating through plasma membrane channels or across intracellular stores such as endoplasmic reticulum and mitochondria . One mechanism for Ca2+ efflux from cells uses the electrochemical gradients of both Na+ and K+ to extrude Ca2+ via the SLC24 family of Na+/Ca2+ , K+ exchangers ( NCKX1-5 ) . Roles for SLC24 proteins in physiology and disease are beginning to emerge ( Herzog et al . , 2015; Li and Lytton , 2014; Parry et al . , 2013; Schnetkamp , 2013 ) and in several cases appear to involve Ca2+ regulation in sensory neurons . One Na+/Ca2+ , K+ exchanger with a well-established function is NCKX4 ( encoded by Slc24a4 ) , which is expressed in olfactory sensory neurons . NCKX4-mediated extrusion of Ca2+ from the cilia of olfactory receptor cells shapes the olfactory response and mediates sensory adaptation ( Stephan et al . , 2011 ) . Another family member , NCKX1 , is expressed strongly in rod photoreceptors and is the dominant mechanism for extruding Ca2+ from their outer segments ( Reiländer et al . , 1992; Vinberg et al . , 2015 ) . As a result , NCKX1 plays a critical role for maintaining the dynamic equilibrium of Ca2+ in rod outer segments and is essential for the normal development , function , and survival of rods and for rod-mediated dim light vision ( Reiländer et al . , 1992; Riazuddin et al . , 2010; Vinberg et al . , 2015 ) . The mechanisms that regulate the extrusion of Ca2+ from the cone photoreceptors , remain controversial . As this process mediates light adaptation , it is critical for the ability of cones to function as our daytime photoreceptors . The prevailing view is that cones also express a cell-specific Na+/Ca2+ , K+ exchanger , NCKX2 , that mediates the extrusion of Ca2+ from their outer segments ( Prinsen et al . , 2000 ) , a role analogous to that of NCKX4 in olfactory neurons or NCKX1 in rods . However , in vivo electroretinogram recordings from NCKX2-deficient mice failed to detect any functional deficits in their cones ( Li et al . , 2006 ) , raising doubts about the role of NCKX2 in regulating cone Ca2+ dynamics . A recent more detailed analysis of isolated cone responses in NCKX2-deficient mice revealed delayed cone response recovery but normal light sensitivity and light adaptation ( Sakurai et al . , 2016 ) . These results suggest the existence of additional , as yet unidentified , mechanism ( s ) for extruding Ca2+ from cone outer segments . The Ca2+ influx/efflux balance in cone photoreceptor outer segments is regulated by the activity of the phototransduction cascade . In the dark , part of the continuous current entering the outer segment through transduction cGMP-gated ( CNG ) channels is carried by Ca2+ ( Miller and Korenbrot , 1987; Picones and Korenbrot , 1995 ) , which is then extruded through mechanisms that remain unclear ( Yau and Nakatani , 1984 ) . Following photoactivation , the closure of CNG channels blocks Ca2+ influx but Ca2+ extrusion continues until a new equilibrium is reached . As a result , light activation is accompanied by a decline in the concentration of outer segment Ca2+ ( Sampath et al . , 1999; Yau and Nakatani , 1985 ) . This triggers the Ca2+-mediated negative feedback on phototransduction , a process required for the timely recovery of the light response and for the adaptation of photoreceptors to background light ( Fain et al . , 2001; Nakatani and Yau , 1988; Sakurai et al . , 2011 ) . This strong modulation of cone phototransduction allowed us to use functional analysis of cone photoreceptors to investigate the mechanisms of Ca2+ extrusion from their outer segments .
To identify candidate proteins that might contribute to calcium homeostasis in cones , we examined published microarray data from rod-dominant wild type ( WT ) and cone-dominant NRL-deficient ( Nrl-/- ) mouse retinas ( Corbo et al . , 2007 ) . We noted that the Na+/Ca2+ , K+ exchanger Nckx4 ( Slc24a4 ) was strongly upregulated in the retina of Nrl-/- mice , suggesting that NCKX4 could potentially be present in cone photoreceptors . This observation is also consistent with results of a recent study on differential expression of genes in rods and cones , where NCKX4 and NCKX2 ( Slc24a2 ) were identified as cone-specific genes ( Hughes et al . , 2017 ) . To establish the identity of the retinal cells expressing Nckx4 , we performed in situ hybridization experiments on retinal sections from WT and Nrl-/- mice . We found sparse expression of Nckx4 in cells at the top of the photoreceptor layer ( ONL ) of WT mouse retinas ( Figure 1a ) . The density and location of the expressing cells suggested that they were cones . Consistent with this notion , the Nckx4 transcript was abundant in the cone-like photoreceptors of Nrl-/- mice ( Figure 1b ) . Together these results demonstrate that Nckx4 is expressed in cone but not in rod photoreceptors . 10 . 7554/eLife . 24550 . 003Figure 1 . NCKX4 is expressed in cone photoreceptors . In situ hybridization with Nckx4 ( Slc24a4 ) probe demonstrating sparse expression of NCKX4 in the photoreceptor layer of a WT mouse retina ( a , arrows ) and strong expression of NCKX4 in the photoreceptor layer of an Nrl-/- ( ‘cone-only’ ) mouse retina ( b ) . A robust expression of NCKX4 is also evident in the inner nuclear layers of WT and Nrl-/- retinas ( a , b ) . OS , outer segment; IS , inner segment; ONL , outer nuclear layer; OPL , outer plexiform layer; INL , inner nuclear layer . Scale bar = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 24550 . 003 To determine the subcellular localization of NCKX4 within cone photoreceptors , we developed a polyclonal NCKX4 antibody ( see Materials and methods ) . Co-localization of peanut agglutinin ( PNA ) , known to label cone outer segments , with the NCKX4 antibody demonstrated that NCKX4 expression was confined mainly to the cone outer segments ( Figure 2a , top row ) . Next , we obtained Nckx4-floxed mice ( Nckx4f/f , [Stephan et al . , 2011] ) and crossed them with the cone-specific HGRP Cre mouse line ( Cre+ , [Le et al . , 2004] ) to generate cone-specific Nckx4 conditional knock-out ( Nckx4f/f Cre+ , see Materials and methods and below ) mice . The expression of Cre recombinase alone had no effect on the presence of NCKX4 in cones ( Figure 2a , middle vs . top rows ) . However , in Nckx4f/f Cre+ mice , NCKX4 immunofluorescence was absent from the cones ( Figure 2a , bottom row ) . In contrast , consistent with the cone-specific expression of Cre , the strong NCKX4 staining in the inner nuclear layer was not diminished in Nckx4f/f Cre+ mice . Comparable results ( not shown ) were obtained with another NCKX4 antibody , recently shown to specifically recognize NCKX4 ( Bronckers et al . , 2017 ) . Further validation of the NCKX4 antibody by western blot demonstrated that it reacts with a ~50 kDa protein as was observed by Bronckers et al . ( Figure 2b ) . Thus , the NKCX4 antibody was selective for NCKX4 and did not cross-react with other proteins expressed in rod or cone photoreceptors . Together , these results indicate that the conditional knockout of NCKX4 in cones was successful . Importantly , the density and morphology of PNA-stained NCKX4-deficient cones were indistinguishable from these of wild type cones , suggesting that the absence of NCKX4 did not affect adversely cone survival . 10 . 7554/eLife . 24550 . 004Figure 2 . NCKX4 is expressed in the outer segments of cone photoreceptors . ( a ) Immunostaining for NCKX4 in vertical sections of mouse retinas ( photoreceptors on the top ) . Nuclei ( DNA , cyan ) , cone photoreceptors ( PNA , red ) , and NCKX4 ( green ) staining in Cre-negative Nckx4+/+ ( top row ) , Cre-positive control Nckx4+/+ Cre+ ( middle row ) and Nckx4f/f Cre+ ( bottom row ) mice . Insets show larger magnification immunostaining for cones in the photoreceptor layer . Scale bar = 50 μm . ( b ) Western blot of wild-type mouse retinal homogenate revealing a protein band of ~50 kDa ( * ) consistent with NCKX4 . DOI: http://dx . doi . org/10 . 7554/eLife . 24550 . 004 To further investigate the cone-specific expression of NCKX4 , whole mount retina was prepared from control C57BL/6 mice and co-stained for the shortwave cone pigment ( S-opsin ) and NCKX4 . In the dorsal retina , only a subset of NCKX4-positive cells were also co-labeled with S-opsin ( Figure 3a ) , a result consistent with the low number of cones that express S-opsin in this region of the retina ( Szél et al . , 1992 ) . In the ventral region , where S-opsin is expressed more uniformly , nearly all cones co-expressed S-opsin and NCKX4 ( Figure 3b ) . Thus , NCKX4 is expressed in both M- and S-cones in the mouse retina , indicating that this exchanger could play a role in regulating cone Ca2+ and , hence , cone function . 10 . 7554/eLife . 24550 . 005Figure 3 . NCKX4 is broadly expressed in cones and in rod bipolar cells . Immunostaining for NCKX4 ( green ) and short-wave opsin ( S-opsin , red ) in flat mounted retinas . In the dorsal region , S-opsin expressing cones were only a fraction of the NCKX4-expressing cones ( a ) consistent with the low density of S-cones and high density of M-cones in the dorsal mouse retina . In contrast , nearly all NCKX4-expressing cones in the ventral region expressed S-opsin as well ( b ) A higher magnification of the labeled cones is shown at the bottom of each panel . S opsin staining appeared stronger in the inner segment and tapered off toward the outer segment , whereas NCKX4 labeling appeared uniform in the outer segment . Immunostaining of retinal sections show NCKX4 expression in cells in the inner nuclear layer ( c ) Staining of the same tissue section with PKCα ( d ) a rod bipolar cell marker , shows extensive overlap ( e ) Scale bars = 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 24550 . 005 In situ hybridization revealed that NCKX4-positive cells were not restricted to the photoreceptor layer but they were also present in the distal inner nuclear layer ( INL , Figure 1a ) . The location of their cell bodies suggested that these NCKX4-postive cells might be rod bipolar cells . To test this hypothesis , we co-labeled retinal sections from control C57BL/6 mice for NCKX4 ( Figure 3c ) and the rod bipolar cell marker PKCα ( Figure 3d ) , whereupon strong overlapping immunofluorescence signals were observed ( Figure 3e ) . These results suggest that in addition to its presence in the outer segments of mouse cone photoreceptors , NCKX4 might be expressed in rod bipolar cells , where it could potentially be involved in regulating retinal synaptic transmission and signal processing . However , further experiments will be required to confirm the expression of NCKX4 in rod bipolar cells and to determine its potential role for rod signaling . To determine the role of NCKX4 in cone phototransduction , we first recorded electrical responses from individual dark-adapted mouse cones to flashes of light in Cre+ control and Nckx4f/f Cre+ mouse retinal slices ( Nikonov et al . , 2006 ) ( see Figure 4a and Materials and methods ) . To facilitate the isolation of cone responses , all recordings were done from mice lacking the α-subunit of rod transducin ( Gnat1-/- ) which prevents rods from generating light responses while leaving the structure and function of cones intact ( Calvert et al . , 2000; Nikonov et al . , 2006 ) . The amplitudes of saturated cone responses ( Rmax ) to bright test flashes that closed all transduction CNG channels were not affected by the absence of NCKX4 ( Figure 4b , c and Table 1 ) . Thus , the absence of NCKX4 did not cause substantial change in the dark current of cones , suggesting a normal complement of CNG channels and normal [cGMP] in darkness . However , the overall kinetics of the flash responses appeared slower in NCKX4-deficient cones . Comparing responses to identical dim-flash stimuli , we found that response termination was delayed dramatically in the absence of NCKX4 ( Figure 4d ) . Quantitative analysis revealed an almost two-fold increase of the time-to-peak ( tp ) and the recovery time constant of the tail phase of dim flash response ( τrec ) , as well as a two-fold increase in integration time ( tint ) upon deletion of NCKX4 ( Table 1 ) . 10 . 7554/eLife . 24550 . 006Figure 4 . NCKX4 accelerates light response termination and decreases the sensitivity of mouse cones . ( a ) Slice preparation under infrared illumination . Photoreceptor outer segments are pointing to the right and the recording glass pipette is visible at the bottom right corner . Scale bar = 20 μm . Representative light responses recorded from a single inner segment of a control Nckx4+/+ Cre+ ( b ) and Nckx4f/f Cre+ ( c ) cone . Light flashes ( flash length = 1 ms , λ = 505 nm , flash strength Q = 200–46 , 100 photons μm−2 ) were delivered at t = 0 s ( arrow ) . ( d ) Population averaged ( mean ± SEM ) responses to a dim flash normalized with Q ( in photons μm−2 ) and maximal response amplitude ( Rmax ) recorded from Nckx4+/+ Cre+ control ( black , Q = 860 photons μm−2 , N = 6 cells from two mice ) and Nckx4f/f Cre+ ( red , Q = 393 photons μm−2 , N = 6 cells from three mice ) cones . The tail of the responses is fit by a single exponential function ( Equation 1 ) with τ = 75 ms and 106 ms in control and NCKX4-deficient cones , respectively . The inset shows the rising phase of dim flash responses on a finer time scale . The horizontal bar measures 50 ms and the vertical bar is 0 . 04% . ( e ) Normalized population averaged response amplitudes ( R/Rmax ) are plotted as a function of flash strength in photons μm−2 for control ( black , N = 6 cells from two animals ) and NCKX4-deficient cones ( red , N = 6 cells from three animals ) . Smooth traces plot Equation 3 with Q1/2 = 1390 photons μm−2 and 730 photons μm−2 for control ( black ) and Nckx4f/f Cre+ ( red ) cones , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 24550 . 00610 . 7554/eLife . 24550 . 007Table 1 . Comparison of control and NCKX4-deficient cone flash response and light adaptation parameters . Rmax , maximal flash response amplitude ( in pA ) in single-cell recordings; tp , time from flash to the peak amplitude ( in ms ) of a dim flash response in single-cell recordings; τrec , the recovery time constant ( in ms ) of the tail phase of a dim flash response in single-cell recordings ( see Equation 1 ) ; Q1/2 , flash strength ( in photons μm−2 ) eliciting 50% of the Rmax in single-cell recordings ( see Equation 3 ) ; τ1 and τ2 , time constants in Equation 2 describing the recovery kinetics of ex vivo ERG signal after step onset ( see Figure 4c , d ) ; I0 , background light intensity ( in photons μm−2 s−1 ) reducing the flash response sensitivity of cones to 50% of that in darkness as derived from ex vivo ERG data; Tint , integration time of dim flash responses ( defined as the area between baseline and response divided by the peak amplitude ) Statistics for parameters ( mean ± SEM ) derived from ex vivo ERG data ( τ1 , τ2 and I0 ) were from five control and eight Nckx4f/f Cre+ mice , and the statistics for the other parameters from single-cell recordings were from six control cells ( two mice ) and six Nckx4f/f Cre+ cells ( three mice ) . * indicates significant ( p<0 . 05 ) difference between control and NCKX4-deficient cones . DOI: http://dx . doi . org/10 . 7554/eLife . 24550 . 007Rmax ( pA ) tp ( ms ) τrec ( ms ) Q1/2 ( hv μm−2 ) τ1 ( ms ) τ2 ( s ) I0 ( hv μm−2 s−1 ) Tint ( ms ) Nckx4+/+ Cre+5 . 8 ± 176 ± 352 ± 81 , 150 ± 240109 ± 271 . 3 ± 0 . 249 , 000 ± 11 , 00080 ± 14Nckx4f/f Cre+5 . 6 ± 0 . 4121 ± 6*101 ± 13*670 ± 73*212 ± 37*1 . 5 ± 0 . 311 , 508 ± 990*160 ± 9* Analysis of the early leading edge of the light response ( normalized with Rmax ) can be used to assess the onset and efficiency of the phototransduction activation reactions , i . e . the amplification of the light signal by the transduction machinery ( Nikonov et al . , 2000; Pugh and Lamb , 1993 ) . The leading edge of the NCKX4-deficient cone response was similar to that of control cones ( Figure 4d , inset ) , indicating that the activation of the phototransduction reactions was not affected by the absence of NCKX4 . Consistent with this , analysis of the amplification constant in control and NCKX4-deficient cones revealed comparable ( p>0 . 05 ) values of 0 . 5 ± 0 . 1 s−2 ( n = 6 ) and 0 . 8 ± 0 . 3 s−2 ( n = 6 ) , respectively . The response onset delay td ( see Equation 5 ) appeared to increase slightly ( p=0 . 02 ) in Nckx4f/f Cre+ cones ( td = 35 ± 2 ms ) compared to Cre+ controls ( td = 29 ± 2 ms ) . The reason for this subtle change is not evident . Thus , the efficiency of phototransduction activation was not compromised by deletion of NCKX4 but light responses were slower , particularly in their recovery , in the Nckx4f/f Cre+ cones compared to control mice , resulting in larger responses ( Figure 4d ) and increased sensitivity of cones lacking NCKX4 ( Figure 4e; Table 1 ) . Together these results demonstrate that NCKX4 is important for the timely recovery of cone phototransduction . This conclusion is consistent with the hypothesis that NCKX4 participates in the extrusion of Ca2+ from cone outer segments , and thus allows a faster decline of Ca2+ concentration to accelerate light response termination . Rapid and efficient adaptation of cones to varying ambient illumination is required for proper visual function . Sluggish sensitivity adjustment in varying background light intensity may , for example , compromise our ability to drive a car in twilight or during the night when large changes in average illumination levels occur rapidly . To study the role of NCKX4 in cone light-adaptation , we measured the sensitivity of Cre+ control and Nckx4f/f Cre+ mouse cones under various background light intensities , using ex vivo ERG recordings ( Vinberg et al . , 2014 , 2015 ) ( see Figure 5a and Materials and methods ) . We began by measuring the kinetics of cone light adaptation in response to near half-saturating steps of light ( see legend of Figure 5 and Figure 6c ) . Both Cre+ control and Nckx4f/f Cre+ cones responded to the onset of a light step with initial hyperpolarization followed by an adaptation-mediated relaxation to a plateau ( Figure 5b ) . Thus , despite the absence of NCKX4 , cones were able to undergo light adaptation in response to steady background light . However , the response relaxation appeared slower in NCKX4-deficient cones compared to that in control cones , suggesting a delay in Ca2+ extrusion . The kinetics of light adaptation in response to a step of light could be fit by a sum of two exponential functions in both control ( Figure 5c ) and Nckx4f/f Cre+ cones ( Figure 5d ) . The faster time constant τ1 was about 100 ms in control cones and increased significantly upon deletion of NCKX4 , whereas the slower time constant was ~1 . 5 s , both in control and Nckx4f/f Cre+ cones ( Table 1 ) . The proportion of the faster component in the fittings ( A1/A , see Equation 2 ) was not affected by the absence of NCKX4 ( A1/A = 46 ± 10% in control and A1/A = 51 ± 10% in Nckx4f/f Cre+ mice ) . These results demonstrate that NCKX4 is required for the rapid light adaptation of cones . However , they also indicate the presence of additional , relatively slow and NCKX4-independent , mechanism ( s ) for extruding Ca2+ from cone outer segments . 10 . 7554/eLife . 24550 . 008Figure 5 . NCKX4 accelerates cone light adaptation . ( a ) A schematic of the system to record transretinal voltage ( ex vivo ERG ) from a superfused isolated mouse retina ( pink , see Materials and methods for details ) . ( b ) Population averaged ex vivo cone ERG responses ( mean ± SEM ) to steps of light normalized to the peak amplitude recorded from Nckx4+/+ Cre+ control ( black , background light intensity I = 37 , 800 photons μm−2 s−1 , N = 5 mice ) and Nckx4f/f Cre+ ( red , I = 15 , 500 photons μm−2 s-1 , N = 5 mice ) retinas . A test flash ( arrow ) was delivered 2 . 5 s after the onset of each light step ( bar ) to probe the sensitivity of cones at different background light intensities ( see below ) . A sum of two exponential functions ( Equation 2 ) was fitted to the recovery phase of the averaged step responses ( after the step onset ) shown in ( b ) for Nckx4+/+ Cre+ control ( c ) and Nckx4f/f Cre+ ( d ) mice . The values of the best-fitting time constants are indicated in each panel . DOI: http://dx . doi . org/10 . 7554/eLife . 24550 . 00810 . 7554/eLife . 24550 . 009Figure 6 . NCKX4 extends the function of cones to brighter light . The sensitivity to a flash of light ( SF ) for Nckx4+/f Cre+ control ( a ) and Nckx4f/f Cre+ ( b ) cones in background light was determined 2 . 5 s after the light step onset . Light step and flash timing are indicated on the bottom of each panel . ( c ) SF as normalized to the sensitivity in darkness ( SF , D , mean ± SEM ) is plotted as a function of light step intensity ( IB ) for Nckx4+/+ Cre+ control ( black , N = 5 retinas ) and Nckx4f/f Cre+ ( red , N = 5 retinas ) mice . Smooth traces plot the Weber-Fechner function ( Equation 4 ) with I0 = 43 , 000 and 11 , 300 photons μm−2 s−1 for control ( black ) and NCKX4-deficient ( black ) cones , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 24550 . 009 Next , we probed the sensitivity of Cre+ control ( Figure 6a ) and Nckx4f/f Cre+ ( Figure 6b ) cones to a flash of light ( flash response amplitude/flash energy ) as it approached steady state , 2 . 5 s after the onset of background light . As has been described for many sensory neurons , including photoreceptors , the sensitivity of control cones declined as a function of background light according to the Weber-Fechner law ( Figure 6c ) . Despite the important role for NCKX4 in setting the temporal properties and dark adapted sensitivity of cones and accelerating light adaptation demonstrated above , at steady state cones lacking NCKX4 still adapted according to the Weber-Fechner law ( Figure 6c ) . However , the background light that reduced cone sensitivity two-fold ( I0 ) shifted to about five-fold lower intensity in the absence of NCKX4 ( Figure 6c , Table 1 ) . This shift in the operating range matched well the combined effect of two-fold longer integration time and about 2 . 3-fold higher sensitivity of dim flash responses of dark adapted Nckx4f/f Cre+ cones as compared to the control cones ( see Table 1 and Figure 4d ) . Thus , removing NCKX4 shifted the operating range of cones to dimmer background lights , effectively restricting the ability of these photoreceptors to function in bright light . This finding demonstrates that NCKX4 is important for setting the operating range of the cones for daytime vision . However , the persistent Weber-Fechner adaptation in steady state light indicates that additional mechanism ( s ) for extruding Ca2+ and sustaining adaptation also exist in these cells . As our data above demonstrates , NCKX4 is important for the timely recovery of the light response in cones . Thus , we also sought to determine if this delay in NCKX4-deficient cones compromises the temporal resolution of cone-mediated vision . We assessed the temporal frequency range of the cone retinal signaling pathway in vivo by recording ERG responses to flickering light from anesthetized Cre+ control and Nckx4f/f Cre+ mice . At low frequencies , the cone b-wave ( positive ERG wave dominated by cone-driven bipolar cell activity in the retina [Shirato et al . , 2008] ) of both control and NCKX4-deficient mice could faithfully follow the flicker stimulus ( see 5 Hz flicker stimulation in Figure 7a and b ) . However , consistent with our single-cell recordings , the b-wave responses from Nckx4f/f Cre+ mice recovered more slowly than these from Cre+ control mice . The responses in control mice could easily follow the stimulus for frequencies up to 20 Hz ( Figure 7a ) . In contrast , as a result of their slower termination , the responses of NCKX4-deficient mice began to overlap and saturate at frequencies as low as 10 Hz and could not follow the flicker stimulus at higher frequencies ( Figure 7b ) . Consistent with this observation , analysis of the fundamental response amplitude ( R , measured from negative to positive peak ) revealed significantly steeper decline with increasing flicker frequency in Nckx4f/f Cre+ mice compared to Cre+ controls ( Figure 7c ) . These results demonstrate that NCKX4 plays an important role in enhancing the temporal resolution of cone-mediated signaling in the retina . 10 . 7554/eLife . 24550 . 010Figure 7 . NCKX4 extends the temporal resolution of cone-mediated vision to higher frequencies . In vivo electroretinogram ( ERG ) responses to flickering light recorded from dark-adapted anesthetized Nckx4+/+ Cre+ ( a ) and Nckx4f/f Cre+ ( b ) mice . The frequency of the flicker stimulus ( f ) is indicated on the left for each trace in Hz . Vertical scale bar = 3 μV . The energy of flickering flashes were either 0 or 0 . 5 lg ( Cd m−2 s ) . ( c ) The fundamental response amplitude ( R ) , measured from the most negative peak/plateau to the most positive peak and normalized to the fundamental response amplitude at 5 Hz stimulation , is plotted as a function of the flicker stimulus for Nckx4+/+ Cre+ ( black , N = 5 eyes from three animals ) and Nckx4f/f Cre+ ( red , N = 8 eyes from four animals ) mice . DOI: http://dx . doi . org/10 . 7554/eLife . 24550 . 010 Cones singly deficient in either NCKX2 ( Sakurai et al . , 2016 ) or NCKX4 ( this study ) still exhibited normal steady-state adaptation and by inference , persistent Ca2+ extrusion from the outer segment . Thus , we generated mice lacking both exchangers in their cones in order to examine their functional redundancy and determine whether additional mechanisms of Ca2+ extrusion exist in the cone outer segment . First , we prepared retinal flat mounts to investigate the effect of the simultaneous removal of these two exchangers on cone number and outer segment morphology . The absence of NCKX2 alone has no discernible effect on these parameters , as we have previously shown ( Sakurai et al . , 2016 ) . Cone outer segments from the ventral region of Nckx2-/- retinas from 3-month-old mice labeled with S-opsin ( Figure 8a ) or the α-subunit of the cone cGMP-gated channel ( CNGA3; Figure 8c ) antibodies appeared homogeneous in diameter and morphology . A tangential slice from a CNGA3-labeled flat mount showed largely circular cross-sections of cone outer segments , consistent with the localization of the CNG-gated channel at the plasma membrane ( Figure 8e ) . In contrast , the outer segment diameter of cones from mice lacking both exchangers varied , as seen in S-opsin ( Figure 8b ) and CNGA3 ( Figure 8d ) labeled cells . A cross-section of the CNGA3-labeled flat mount showed numerous cellular profiles of lines instead of circles , as if the circular structures had collapsed ( Figure 8f ) . Despite this change in outer segment structure , the number of PNA-positive cones was not statistically different between Gnat1-/- control , Nckx4 conditional knockout , and Nckx2/4 double knockout ( Nckx2/4 DKO ) cones in 3-month-old mice ( Figure 8i ) . At 9 months of age , however , strikingly fewer number of PNA-positive cones were detected in the Nckx2/4 double knockout cones when compared to Cre-negative Nckx2-/-Nckx4f/f controls ( Figure 8g–i , p<0 . 0001 , two-tailed t-test ) . Thus , the simultaneous deletion of Nckx2 and Nckx4 resulted in abnormal cone outer segment morphology and progressive cone death . 10 . 7554/eLife . 24550 . 011Figure 8 . Loss of both NCKX2 and NCKX4 expressions leads to alteration of outer segment structure and cone cell death . Confocal stacked images from flat mounted retinas from 3-month-old Nckx2-/- mice ( a , c , e ) and their littermate Nckx2/4 double knockout animals ( b , d , f ) stained with antibodies against S-opsin ( a , b ) or CNGA3 ( c , d ) . A slice from the CNG3A stacked image is shown in e and f . Higher magnification of representative cells are shown in inset . Cones were labeled with peanut agglutinin ( PNA ) in representative flat mounted retina from 3-month-old ( g ) and 9-month-old ( h ) Nckx2/4 double knockout mice . Scale bar = 20 μm . The number of cones from retinas of control , Nckx4 knockout , and Nckx2/4 double knockout mice were counted and plotted ( i ) . One-way ANOVA showed no differences in cone numbers in the group of 3-month-old mice ( p=0 . 3 ) , but significant ( p<0 . 0001 ) reduction in cone numbers in the 9-month-old Nckx2/4 double knockouts compared to Nckx2-/- mice . DOI: http://dx . doi . org/10 . 7554/eLife . 24550 . 011 We next investigated how the ablation of NCKX2 and NCKX4 expression affects the function of cones . Consistent with our observation from single-cell suction recordings ( Figure 4 ) , comparison of flash response families obtained with transretinal recordings demonstrated that the response recovery is delayed greatly in NCKX4-deficient cones compared to controls ( Figure 9 , compare a and b ) . The maximal response amplitudes were similar , consistent with the observed normal cone numbers and cone morphology in NCKX4-deficient cones ( Figures 2 , 8 ) . However , the simultaneous deletion of the two exchangers resulted in a dramatic reduction in the maximal response of NCKX2/4 double knockout ( DKO ) cones ( Figure 9c ) . The delay in cone response recovery caused by the absence of NCKX4 ( Figure 9b ) or NCKX2 ( Figure 9d , inset ) was extended even further when both exchangers were removed simultaneously ( Figure 9d ) , demonstrating the complementary roles of NCKX2 and NCKX4 in mediating the extrusion of Ca2+ from cone outer segments . Nckx2/4 DKO cone sensitivity was also significantly reduced compared to both control and NCKX4-deficient cones ( Figure 9e ) . These results are consistent with the observed morphologic alteration of outer segment structure in these cones ( Figure 8 ) and demonstrate that cone function was severely affected by the simultaneous ablation of NCKX2 and NCKX4 . Together , the abnormal morphology and severe functional deficits in cones lacking both exchangers suggest that the combined activity of NCKX2 and NCKX4 represents the dominant mechanism by which calcium is extruded from cone outer segments . 10 . 7554/eLife . 24550 . 012Figure 9 . Loss of both NCKX2 and NCKX4 expressions severely compromises cone function . Representative light responses recorded from isolated retinas of Nckx4+/+ Cre+ ( a ) , Nckx4f/f Cre+ ( b ) and Nckx2-/- Nckx4f/f Cre+ ( c ) mice using ex vivo ERG method . Light flashes ranged from 390 to 460 , 000 photons ( 505 nm ) μm−2 in ( a ) and ( b ) , and from 4000 to 1 . 4 * 106 photons ( 505 nm ) μm−2 . ( d ) Population averaged ( mean ± SEM ) responses to a dim flash normalized with Q ( in photons μm−2 ) and maximal response amplitude ( Rmax ) recorded from Nckx4+/+ Cre+ control ( black , N = 5 retinas from four mice ) , Nckx4f/f Cre+ ( red , N = 8 retinas from four mice ) and Nckx2-/- Nckx4f/f Cre+ ( blue , N = 4 retinas from three mice ) retinas . The response from Nckx2-/- Nckx4f/f Cre+ mice shown in ( d ) is scaled up by 25-fold to facilitate comparison of response kinetics between the genotypes . The inset shows fractional dim flash responses of control ( WT ) and Nckx2-/- cones modified from ( Sakurai et al . , 2016 ) . ( e ) Normalized population averaged response amplitudes ( R/Rmax ) are plotted as a function of flash strength in photons μm−2 for Nckx4+/+ Cre+ control ( black , N = 5 ) , Nckx4f/f Cre+ ( red , N = 8 ) , and Nckx2-/- Nckx4f/f Cre+ ( blue , N = 4 ) mouse retinas . Smooth traces plot Eq . 3 with Q1/2 = 5200 photons μm−2 and 3900 photons μm−2 for control ( black ) and Nckx4f/f Cre+ ( red ) retinas , respectively . ( f ) Light adaptation persists in the cones from Nckx2-/- Nckx4f/f Cre+ mice . Representative response to a bright step of 530 nm light ( I = 38 , 600 , 000 photons μm−2 s−1 ) recorded from Nckx2-/- Nckx4f/f Cre+ mouse retina . Light step timing is indicated on the bottom of the graph . DOI: http://dx . doi . org/10 . 7554/eLife . 24550 . 012 Finally , we also examined the ability of NCKX2/NCKX4-deficient cone photoreceptors to adapt to background light . The small residual signaling and reduced light sensitivity of Nckx2/4 DKO cones resulted in substantial pigment bleaching by the step of light , which made the analysis challenging and prevented us from quantifying the results . However , the response to a step of light revealed substantial relaxation 1–2 s after the onset of the background light ( Figure 9f ) . This surprising result suggests that these cones might still be able to undergo light adaptation despite the block of both NCKX2- and NCKX4-driven extrusion of Ca2+ . Further studies will be needed to determine whether this adaptation is mediated by a slow NCKX-independent extrusion of Ca2+ from cone outer segments or by a Ca2+-independent mechanisms that become unmasked upon the deletion of NCKX2 and NCKX4 . Our molecular and functional analyses demonstrated that NCKX4 is expressed in mouse cones where it plays an important role in regulating their function . To investigate whether the use of NCKX4 by cones is a widespread phenomenon conserved throughout evolution , we examined the retinas of zebrafish and macaque . PNA staining was used to identify the abundant cone photoreceptors in the zebrafish retina ( Figure 10a ) . NCKX4 immunofluorescence was observed in several cell types in the zebrafish retina , consistent with our observation on the mouse retina . In the outer retina , however , the NCKX4 signal was restricted to cones ( Figure 10a , inset ) , demonstrating that similar to the case of mouse , zebrafish cone photoreceptors also express NCKX4 . 10 . 7554/eLife . 24550 . 013Figure 10 . NCKX4 is expressed in the outer segments of zebrafish and non-human primate cones . Immunostaining of NCKX4 in the vertical sections of zebrafish ( top ) and macaque ( bottom ) retinas ( photoreceptors on the top ) . Nuclei ( DNA , cyan ) , cone photoreceptors ( PNA , red ) , and NCKX4 ( green ) staining . Insets show larger magnification immunostaining of cones in the photoreceptor layer . Scale bar = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 24550 . 013 Three spectrally distinct cone photoreceptors mediate high-acuity daytime color vision in humans . To explore whether NCKX4 could potentially also play a role in human daytime vision , we analyzed its expression in macaques , primates with retinas that are very similar to the human retina . Our NCKX4 antibody immunolabeled the cone outer segments ( identified by PNA labeling ) in sections of the mid-peripheral region of the macaque retina ( Figure 10b ) , demonstrating the expression of NCKX4 in primate cones . Together , these results indicate that the expression of NCKX4 in cones is conserved from zebrafish to mice and non-human primates , suggesting that NCKX4 could potentially also play a role in cone-driven primate and human daytime vision .
The main effect of NCKX4 removal is to slow the cone light response recovery and to increase the peak amplitude of subsaturating flash responses ( Figure 4 ) , indicating slower Ca2+ extrusion and phototransduction feedback . Detailed comparison of dim flash response waveforms from control and NCKX4-deficient cones demonstrates that the efficiency of phototransduction activation reactions is not affected by the presumptive slower Ca2+ extrusion ( Figure 4d ) . Interestingly , dark-adapted NCKX4-deficient cone flash responses resemble those of GCAPs knockout cones in which the Ca2+ feedback that accelerates cGMP synthesis is absent ( Sakurai et al . , 2011 ) . Hence , NCKX4 appears to be important for rapidly lowering [Ca2+] in cones following light-induced closure of CNG channels to mediate the Ca2+ feedbacks in the time scale of flash responses ( ~500 ms ) . Ca2+ contributes significantly to both rod and cone light adaptation ( Fain et al . , 2001; Nakatani and Yau , 1988 ) . Without Ca2+ feedback , it has been suggested that rod cells would simply integrate single photon responses and saturate at very dim background light intensities ( Matthews et al . , 1988; Nakatani and Yau , 1988 ) . Thus , disrupting Ca2+ extrusion is expected to compromise light adaptation . Indeed , we found that deletion of NCKX4 delays light adaptation and shifts the operating range of cones to dimmer light ( Figures 5 and 6 ) . However , cones without NCKX4 still adapt according to the Weber-Fechner law , indicating that Ca2+ is still extruded from their outer segments . The lack of detectable cone degeneration ( Figures 2 and 8 ) and the normal cone light response amplitudes of NCKX4-deficient cones ( Figures 4 and 9 ) also support the existence of additional Ca2+ extrusion pathways in cones . Our results from cones lacking both NCKX2 and NCKX4 ( Figure 9 ) suggest that the Ca2+ extrusion from NCKX4-deficient cones is mainly via NCKX2 . This exchanger is expressed in chicken , human and mouse cones , where it has been shown to regulate cone phototransduction ( Prinsen et al . , 2000; Sakurai et al . , 2016 ) . We propose that the combined activity of NCKX2 and NCKX4 is required for the rapid and efficient extrusion of Ca2+ from cones and that this enables these cells to adapt rapidly and remain functional in a wide range of light backgrounds throughout the day . Interestingly , adaptation was apparent even in cones lacking both NCKX2 and NCKX4 ( Figure 9f ) . Our results do not allow us to distinguish between calcium-dependent and calcium-independent mechanisms of light adaptation in NCKX2/NCKX4-deficient cones . One possibility is that residual Ca2+ extrusion somehow persists even in the combined absence of NCKX2 and NCKX4 . Indeed , slow NCKX-independent extrusion of Ca2+ has been recently suggested for rod outer segments ( Vinberg et al . , 2015 ) . Such a mechanism might be common to all photoreceptors , potentially serving as a safety valve for Ca2+ release . Alternatively , it is possible that some residual calcium-independent adaptation exists in cones and becomes unmasked when Ca2+ extrusion is blocked in NCKX-deficient cones . By comparing the functional deficiency of cones lacking NCKX4 with the recent results from cones lacking NCKX2 ( Sakurai et al . , 2016 ) , it is possible to identify the primary mechanism for Ca2+ extrusion in cones . Whereas the deletion of both NCKX4 and NCKX2 causes a notable delay in the cone response recovery , the effect is substantially more pronounced in NCKX4-deficient cones than in NCKX2-deficient cones ( Figure 9d ) . In addition , the amplitude of the single photon response and , consequently , cone sensitivity are also substantially ( >2 fold ) increased in the absence of NCKX4 compared to controls ( Figure 4 ) , whereas a similar analysis of NCKX2-deficient cones revealed only subtle increase in the single photon response amplitude and no change in cone sensitivity ( Sakurai et al . , 2016 ) . Thus , the extrusion of Ca2+ that modulates phototransduction after photoactivation appears to be dominated initially by NCKX4 , with NCKX2 contributing mostly during the late phase of the flash response ( Figure 9d ) . Such temporal separation could be the result of different kinetics or ionic equilibrium of inactivation of the two exchangers , enabling them to differentially modulate Ca2+ extrusion . Regardless of the molecular mechanisms that regulate the activities of NCKX2 and NCKX4 , our findings lead to the conclusion that NCKX4 is the dominant Ca2+ extrusion pathway in cone outer segments . Whereas the number of cones was similar between single knockout of NCKX2 , NCKX4 and control Gnat1-/- mouse retinas , removal of both exchangers had a marked effect on cone structure/function and survival . Results in the DKO retinas are similar to the findings from NCKX1-deficient rods , where the removal of the sole exchanger presumably renders the cells unable to efficiently extrude Ca2+ . Importantly , lowering of Ca2+ stimulates retinal guanylyl cyclase to synthesize cGMP through the action of GCAPs ( Dizhoor et al . , 1995; Gorczyca et al . , 1994; Mendez et al . , 2001; Palczewski et al . , 1994 ) . In support of this mechanism , we measured lowered cGMP levels in NCKX1-deficient retinas ( Vinberg et al . , 2015 ) . Due to the high cooperativity of the cGMP-gated channel , lowered cGMP provides an explanation for the large decrease in light sensitivity in Nckx1-/- retinas . We hypothesize that loss of both NCKX2 and NCKX4 in cones has the same effect as loss of NCKX1 in rods . That is , in the absence of NCKX-driven Ca2+ extrusion , cGMP synthesis is not stimulated , and more cGMP-gated channels remain closed , leading to the observed dramatic decrease in response amplitude and light sensitivity ( Figure 9 ) . Over the long-term , the disrupted Ca2+ homeostasis may gradually lead to cone death . The striking differences in adaptation capacity and kinetics of daytime color vision and dim-light vision can be attributed in part to the distinctive physiological properties of rods and cones , that is cones have faster light response kinetics and are able to remain functional under brighter light than rods . However , the molecular origins of these differences remain incompletely understood ( Ingram et al . , 2016 ) . Biochemical evidence from non-mammalian species have shown faster decay of Meta II ( active form of visual pigment ) ( Shichida et al . , 1994 ) , higher expression of RGS9 and resulting faster inactivation of PDE ( Tachibanaki et al . , 2012 ) , faster synthesis rate of cGMP ( Takemoto et al . , 2009 ) and higher arrestin content ( Tomizuka et al . , 2015 ) in cones compared to rods . Such differences could potentially explain the faster function of cones compared to rods . However , direct physiological evidence of the significance of these molecular differences is still lacking , and it is not known if these differences hold in mammalian phototoreceptors . As light adaptation in both rods and cones is mediated by the light-induced decline in Ca2+ ( Matthews et al . , 1988; Nakatani and Yau , 1988 ) , it is reasonable to assume that the better light adaptation capacity of cones is mediated by more efficient Ca2+ extrusion and Ca2+ feedback . However , recent results indicate that one of the main Ca2+ feedbacks , mediated by GCAP1/2 , contributes similarly to the light adaptation capacity of rods and cones ( Sakurai et al . , 2011 ) . Moreover , although recoverin appears to modulate cone phototransduction somewhat more efficiently than rod phototransduction , its role in light adaptation is quite marginal ( Makino et al . , 2004; Sakurai et al . , 2015 ) . Thus , the remaining possibility is that Ca2+-dependent modulation of cone CNG channel ( Korenbrot et al . , 2013; Rebrik et al . , 2012 ) together with larger and faster changes of Ca2+ concentration in the outer segments of cones between dark and bright light conditions ( Sampath et al . , 1999 , 1998 ) contribute to the faster and wider range of daytime cone-mediated vision compared to rod-mediated nighttime vision . Our data demonstrates that the cone-specific Ca2+ exchanger NCKX4 accelerates the responses of cones ( Figure 4 ) and cone-mediated vision ( Figure 7 ) and shifts the operating range of cones to brighter background lights ( Figure 6 ) . By doing so , NCKX4 contributes to the unique properties of cones and their ability to operate at brighter light as compared to rods . Our recent work and the results presented here demonstrate that NCKX1 is the sole Na+/Ca2+ exchanger expressed in rods , whereas cones express both NCKX2 and NCKX4 ( but not NCKX1; [Sakurai et al . , 2016; Vinberg et al . , 2015] ) . We also found evidence for expression of NCKX4 in inner retina neurons identified tentatively as rod bipolar cells ( Figures 1 and 3 ) but the physiological relevance of its presence there remains to be addressed . The reason for tissue/cell-specific expression patterns of different NCKX isoforms is still unclear , particularly considering the similarities in their biophysical properties ( Jalloul et al . , 2016 ) . However , it is known that NCKX1 dimers form hetero-oligomers with the rod transduction CNG channel ( Bauer and Drechsler , 1992; Poetsch et al . , 2001; Schwarzer et al . , 1997 , Schwarzer et al . , 2000 ) . Consistent with this notion , our recent results suggest that NCKX1 is important for normal rod channel function and that deletion of NCKX1 leads to reduced expression and lower conductance of rod CNG channels in mice ( Vinberg et al . , 2015 ) . On the other hand , deletion of NCKX2 ( Sakurai et al . , 2016 ) or of NCKX4 ( Figure 4 ) does not affect the dark current of cones . The simplest explanation for this result is that , in contrast to the case in rods , the expression of NCKX does not control the expression or conductance of cone CNG channels . Such a scenario is consistent with biochemical evidence that NCKX2 does not interact directly with CNGA3 in cone outer segments ( Matveev et al . , 2008 ) . However , coexpression of NCKX2 and CNGA3 in HEK293 cells was found to result in direct interaction between these two proteins ( Kang et al . , 2003 ) , indicating that this issue will require further investigation . Our data here also explains the striking phenotypic difference between deletion of NCKX1 ( and nonsense mutation in the SLC24A1 gene encoding NCKX1 that cause night blindness in humans ) leading to 100-fold reduction of rod response amplitudes and severely compromised dim light vision ( Vinberg et al . , 2015 ) , as compared to deletion of either NCKX2 or NCKX4 in cones leading only to altered kinetics of light responses and light adaptation without affecting cone dark current . The redundancy of NCKX2 and NCKX4 may further explain why mutations in NCKX2 or NCKX4 have not been linked to eye diseases . Notably , even cones lacking both NCKX2 and NCKX4 respond robustly to light ( Figure 9 ) , further emphasizing the difference between rods and cones . Interestingly , SNPs in SLC24A4 correlate with lighter eye and hair pigmentation in Europeans ( Sulem et al . , 2007 ) . It is possible that the redundancy in cone exchangers has allowed evolution of the gene coding NCKX4 in Europeans to adapt to environment with lower ambient light levels without at the same time compromising color vision . It may be interesting to test whether Europeans with SNPs in SLC24A4 ( i . e . people with blue/green eyes and blond hair ) or those individuals with Amelogenesis Imperfecta , caused by a nonsense SLC24A4 mutation ( Herzog et al . , 2015 ) , have subtle abnormalities in cone vision , such as altered ERG flicker fusion frequency , confirming the functional importance of NCKX4 in humans .
We crossed Nckx4f/f mice in which exon 5 of the NCKX4 encoding gene ( Slc24a4 ) is floxed ( Stephan et al . , 2011 ) , with the HRGP cone Cre-expressing ( Le et al . , 2004 ) Gnat1-/- ( Calvert et al . , 2000 ) line that has been backcrossed to C57BL/6J background for several generations , to produce cone-specific Nckx4f/f Gnat1-/- Cre+ knock-out and Nckx4+/+ Gnat1-/- Cre+ control mice . A set of experiments were also conducted to compare the cone phenotype between Nckx4f/f Gnat1-/- Cre+ and Nckx4f/f Gnat1-/- Cre- as well as Nckx4+/+ Gnat1-/- Cre+ and Nckx4+/+ Gnat1-/- Cre- littermates . Unless otherwise stated , mice used for the experiments were 2- to 3-month-old males and females . Genotyping for the floxed/WT Slc24a4 as well as for the KO/WT Gnat1 alleles and for HRGP Cre was performed by Transnetyx ( Transnetyx , TN ) and the mice were also tested to be free of the Rd8 mutation ( Mattapallil et al . , 2012 ) . Mice were kept under 12/12 hr light/dark cycle with access to water and standard rodent chow . Before the experiments , mice were dark-adapted overnight and euthanized by CO2 asphyxiation . For single-cell and ex vivo electrophysiology experiments , the eyes were removed quickly after euthanasia and retinas were dissected under an infrared light-equipped microscope as described previously ( Vinberg and Kefalov , 2015 ) . All experimental protocols were in accordance with the Guide for the Care and Use of Laboratory Animals and were approved by the institutional Animal Studies Committee at Washington University . Mouse enucleated eye globes were fixed for 24 hr in 4% paraformaldehyde/PBS , embedded in paraffin , and sectioned in the midsagittal plane at 4 μm . In situ hybridization was carried out using the RNAscope technique ( RNAscope 2 . 0; Advanced Cell Diagnostics , CA ) , as per manufacturer instructions . The target probe set was generated against Slc24a4 transcripts . The probe set consisted of 20 pairs of oligonucleotides spanning a ~1 kb contiguous region of the target mRNA transcript ( Slc24a4 NM_172152 . 2 ) . As a negative control , some sections were hybridized with target probe against DapB , a bacterial gene encoding dihydrodipicolinate reductase , a key enzyme in lysine synthesis . A target probe directed against ubiquitously expressed PolR2A ( DNA-directed RNA polymerase II polypeptide A ) served as a positive control . Following proprietary preamplification and amplification steps , target probes were detected using an alkaline-phosphatase-conjugated label probe with Fast Red as substrate and Gill's Hematoxylin as a counterstain . With this technique , individual mRNA molecules were detected as bright red puncta . Fluorescence immunocytochemistry was performed on paraformaldehyde-fixed ( PAF 4% in buffer phosphate saline pH 7 . 4 , 15 min at room temperature ) frozen or vibratome sections of zebrafish , mouse , and monkey retina , respectively , or flat mount mouse retina . Blocking and antibody incubation were carried out in 2% bovine serum albumin , 2% goat serum and 0 . 1% Triton X-100 in buffer phosphate saline ( pH 7 . 4 ) . The antibody against mouse NCKX4 was produced by Primm Biotech , Inc . MA ( RRID: AB_10792951 ) using a recombinant His-Tag Slc24a4 ( from amino acid 229 to 366 ) as antigen . Secondary antibody used was goat anti-rat Alexa 488 ( Life Technologies , CA ) . In all experiments , no-primary antibody control was run in parallel . Similar cone staining was observed with another NCKX4 antibody ( N414/25 , NeuroMab , UC Davis/NIH NeuroMab Facility ) recently shown to specifically recognize NCKX4 ( Bronckers et al . , 2017 ) . Cone photoreceptor outer segments were labeled with peanut agglutinin ( PNA ) conjugated with Alexa 568 ( Life Technologies , CA ) . Methyl Green ( Prieto et al . , 2014 ) ( emission 663–686 nm ) was used as a DNA counterstain . Confocal microscopy was performed on an Olympus Fluoview 1000 microscope or Zeiss LSM5 Pascal microscope . Samples for western blot analysis were prepared as follows: retinas from two month old mice were homogenized in PBS containing protease inhibitor cocktail ( 88666 , Thermo Scientific Pierce , IL ) and the intact cell nuclei were eliminated by centrifugation . Protein content was determined by BCA assay ( Bio-Rad , CA ) using immunoglobulin G as standard . Proteins ( 10 μg ) and Molecular Weight Standards ( Precision Plus Protein Standards , Bio-Rad ) were separated on SDS-PAGE ( 4%–15% Mini-Protean TGX gels , Bio-Rad ) and then transferred to nitrocellulose membranes . Blots were incubated overnight at 4°C with rat anti-NCKX4 diluted 1:500 . Primary antibody was detected using horseradish peroxidase-conjugated secondary antibodies ( Thermo Fisher Scientific , IL ) . Electrical responses of cone photoreceptors to light stimulation were recorded from single cells using a suction electrode method ( Baylor et al . , 1979; Nikonov et al . , 2006 ) and from isolated retinas by ex vivo ERG method ( Arden and Ernst , 1970; Winkler , 1972 ) . To separate the light responses of cones from those generated by rods , all the mice were bred to Gnat1-/- background rendering rod photoreceptors unresponsive to light without affecting the cone physiology ( Calvert et al . , 2000 ) . Suction electrode recordings were performed as described previously ( Wang and Kefalov , 2010 ) with small modifications . Briefly , the dorsal half ( dominated by M-cones ) of the retina was flat-mounted photoreceptor-side upwards on a filter paper ( HABG01300 , Millipore , MA ) . A manual slicer was then used to cut a 200–250 μm vertical slice that was mounted on the bottom of a specimen holder by attaching the filter paper side of the slice on a wall made from vacuum grease ( 111 , Dow Corning , MI ) . The slice was perfused at 3 ml/min with 37°C Locke’s solution containing ( in mM ) : NaCl , 112; KCl , 3 . 6; MgCl2 , 2 . 4; 1 . 2 , CaCl2; HEPES , 10; NaHCO3 , 20; Na2-succinate , 3; Na-glutamate , 0 . 5; glucose , 10 . In addition , the solution was supplemented with 0 . 1% of MEM vitamins and amino acids ( Sigma-Aldrich , MO ) , and equilibrated with 95%O2/5%CO2 at 37°C . The slice was visualized under infrared light with inverted microscope and the proximal outer nuclear layer was targeted with a glass pipette ( tip resistance ~0 . 6 MΩ , see Figure 4A ) containing ( mM ) : NaCl , 140; KCl , 3 . 6; MgCl2 , 2 . 4; CaCl2 , 1 . 2; HEPES , 3; and glucose , 10; pH adjusted to 7 . 4 with NaOH . Small current signals from the cone inner segments were amplified by using a standard Axopatch 200B amplifier , low-pass filtered at 50 Hz ( 8-pole Bessel , model 3382 , Krohn-Hite , MA ) and collected at 10 kHz ( 1440A Digidata , Molecular Devices , CA ) . A 505 nm LED source ( SR-01-E0070 , Luxeon Star LEDs , Ontario , Canada ) provided calibrated light stimulation as described in detail previously ( Vinberg et al . , 2015 ) . Ex vivo ERG recordings were performed as described previously ( Vinberg et al . , 2015 ) . Isolated dark-adapted mouse retinas were mounted on the specimen holder ( Vinberg et al . , 2014 ) photoreceptor side upwards and perfused 2 ml/min . at 37°C with Locke’s medium similar to that used in suction electrode recordings ( see above , [Vinberg and Kefalov , 2015; Vinberg et al . , 2014] ) . The medium was supplemented with 2 mM L-Aspartate , 40 μM DL-AP4 ( Tocris Bioscience , UK ) and 100 μM BaCl2 to isolate the photoreceptor component of the ERG signal . Transretinal voltage between the electrodes of the specimen holder was amplified ( 100X ) by a differential amplifier ( DP-311 , Warner Instruments , CT ) and data were low-pass filtered and collected at 300 Hz and 10 kHz , respectively , by using the same digitizer and analog low-pass filter as in suction electrode recordings ( see above ) . The same custom-build LED light stimulation was used as in the suction electrode recordings . Flicker in vivo ERG recordings were performed by using UTAS Visual Diagnostic System with BigShotTM Ganzfeld stimulation unit modified for mouse ERG experiments ( LKC Technologies , MD ) . Mice were anesthetized by intraperitoneal injection of ketamine ( 80 mg/kg ) and xylazine ( 15 mg/kg ) cocktail and the eyes were dilated with drops of 1% atropine sulfate . During the recordings the mouse body temperature was maintained at ~37°C by using ATC1000 ( World Precision Instruments , FL ) temperature control system with mouse heating pad ( model 502195 , World Precision Instruments , FL ) . DC ERG signal between corneal electrodes ( STelesSR , LKC Technologies , MD ) and a reference electrode inserted under the skin at the skull as a response to flickering light at frequencies between 5 and 20 Hz were acquired at 1000 Hz with low-pass filter set at 300 Hz . The flash energy of flickering flash was set to either 0 or 5 dB ( i . e . 0 or 5 lg ( Cd m−2 s ) ) . A single exponential function was used to describe the time course of the tail phase of dim flash responses: ( 1 ) r ( t ) =Ae−t−tdτ , where ( td , A ) is the starting point of the fitted function and τ is the time constant of the recovery tail , was fitted ( free parameter = τ ) to the data recorded from single cones ( see Figure 4d ) . A sum of two exponential functions was used to describe the recovery of cone responses following the onset of light steps: ( 2 ) r ( t ) =r0+A1 ( 1−e−t−tdτ1 ) + ( A−A1 ) ( 1−e−t−tdτ2 ) , where A is the amplitude measured from the starting point ( td , r0 ) to the plateau when t →∞ and A1 is the fraction of the recovery contributed by the mechanism underlying the faster exponential time constant τ1 . A best fitting function with a set of freely changing parameters τ1 , τ2 and A1 was fitted to the ex vivo ERG data ( see Figure 5c , d ) . A Naka-Rushton function: ( 3 ) rrmax=QQ+Q½ was fitted to response amplitude ( R ) data ( Figure 4e ) to determine the flash strength producing 50% of the maximal response amplitude rmax ( Q1/2 ) . The Weber-Fechner function: ( 4 ) sFsF , D=I0I0+I was fitted to the light adaptation data ( see Figure 6 ) to determine the background light intensity ( I ) at which the sensitivity of cones is 50% of that in darkness ( I0 ) . The sensitivity in darkness ( SF , D ) and during background lights ( SF ) was determined as the response amplitude to a flash producing <0 . 2 rmax divided by the flash strength in photons μm−2 . The kinetics of phototransduction activation reactions were quantified by using a model originally developed by Lamb and Pugh ( Lamb and Pugh , 1992 ) ( 5 ) rrmax=1−exp[−0 . 5QA ( t−td ) 2] where td is a small delay and A is the amplification in s−2 . Both td and A were selected to provide the best fit to the early rising phase of a dim flash response ( Q = 393–859 photons μm−2 ) . All the assumptions of the model may not be valid for fast cone responses ( see [Smith and Lamb , 1997] ) but the model was used here only to compare quantitatively the activation kinetics of phototransduction between control and NCKX4-deficient cones . The same model has been used previously by ( Nikonov et al . , 2006 ) to describe M- and S-cone phototransduction activation . Electrophysiology data analysis , including statistical analysis , fitting and figure preparation , was performed with Origin 9 ( OriginLab , MA ) . Immunohistochemistry images were prepared with Photoshop 11 . All data , where applicable , are presented as mean ± SEM , and the number of mice , retinas and/or cells used in each experiment are indicated in the Figure and Table legends . The statistical significance of difference between parameters calculated from the mutant and control samples was tested with a two-tailed unpaired Student’s t-test or with one-way ANOVA ( p<0 . 05 ) . | Cells known as photoreceptors sense light in the eye . Light activates signaling pathways inside the photoreceptors that relay visual information to nerve cells , which carry the information to the brain . Photoreceptors called cone cells allow us to distinguish different colors of light and therefore play an important role in daytime vision . Over the course of the day , the overall levels of light in the environment can vary widely and so photoreceptors need to be able to adjust their signaling pathways so that they can still respond to light stimuli . Calcium ions modulate the signaling pathways inside cone cells to help them adjust to changing light levels . These ions also play other important roles in the health and activity of photoreceptors , so the cells need to carefully control how many calcium ions they contain . Cone cells contain a structure known as the outer segment , which is responsible for detecting light stimuli . It is believed that cones control the levels of calcium ions in the outer segment by balancing the flow of calcium ions into and out of the segment . The calcium ions enter the outer segment via channels that sit in the membrane surrounding the cell . A transporter protein known as NCKX2 , which is only found in cone cells , was thought to pump the calcium ions out of the cell . However , recent data has challenged this idea by demonstrating that NCKX2 only plays a minor role in this process . Vinberg et al . investigated how calcium ions leave the outer segments of cone cells in several different animals . The experiments show that a transporter protein called NCKX4 – which belongs to the same protein family as NCKX2 – is the main transporter involved in removing calcium ions from the cone cells of mice . Loss of NCKX4 from mouse cones reduced the ability of these cells to respond to fast and rapidly changing light stimuli , and to operate in bright light . Further experiments show that NCKX4 is also found in the outer segments of zebrafish and monkey cone cells . The next challenges will be to find out if NCKX4 is also present in human cones and whether it plays a role in regulating our daytime vision . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | The Na+/Ca2+, K+ exchanger NCKX4 is required for efficient cone-mediated vision |
To follow the dynamics of meiosis in the model plant Arabidopsis , we have established a live cell imaging setup to observe male meiocytes . Our method is based on the concomitant visualization of microtubules ( MTs ) and a meiotic cohesin subunit that allows following five cellular parameters: cell shape , MT array , nucleus position , nucleolus position , and chromatin condensation . We find that the states of these parameters are not randomly associated and identify 11 cellular states , referred to as landmarks , which occur much more frequently than closely related ones , indicating that they are convergence points during meiotic progression . As a first application of our system , we revisited a previously identified mutant in the meiotic A-type cyclin TARDY ASYNCHRONOUS MEIOSIS ( TAM ) . Our imaging system enabled us to reveal both qualitatively and quantitatively altered landmarks in tam , foremost the formation of previously not recognized ectopic spindle- or phragmoplast-like structures that arise without attachment to chromosomes .
Meiosis is essential for sexual reproduction by reducing the chromosome number to eventually generate gametes with half the genomic DNA content of the parental plant . Moreover , meiosis is central to the formation of genetic diversity by generating recombination between the homologous chromosomes ( homologs ) and by randomly selecting either the maternal or paternal homologs to establish a new set of chromosomes in the gametes . Therefore , understanding the molecular mechanisms underlying recombination and chromosome distribution are also of key interest for breeding to modulate meiosis ( Crismani et al . , 2013; Hand and Koltunow , 2014; Lambing and Heckmann , 2018 ) . Entry into meiosis is tightly regulated in all organisms . In plants , it involves the reprogramming of somatic fate since plants , in contrast to animals , do not have a germline that is set aside early during embryo development ( Schmidt et al . , 2015 ) . Designated meiocytes have to repress stem cell activity ( Zhao et al . , 2017 ) , and differentiate by adopting a characteristic shape that radically changes during the course of meiosis ultimately resulting in the formation of spores . These spores then differentiate into gametophytes that produce the gametes , which will fuse during fertilization ( Dresselhaus et al . , 2016 ) . In recent years , our understanding of meiosis in plants has been fostered by genetic approaches , mostly in the model plants Arabidopsis thaliana , Zea mays and Oryza sativa . These studies have identified more than 80 meiotic genes , including those that control entry and progression through meiosis ( Lambing et al . , 2017; Ma , 2006; Mercier et al . , 2015; Wijnker and Schnittger , 2013; Zhou and Pawlowski , 2014 ) . For instance , mutants in the A1;2 type cyclin TARDY AYNCHRONOUS MEIOSIS ( TAM ) were found to be required for entry into meiosis II ( Cromer et al . , 2012; d'Erfurth et al . , 2010; Magnard et al . , 2001 ) . TAM is a bona fide cyclin and builds at least in vitro an active kinase complex with CYCLIN-DEPENDENT KINASEA A;1 ( CDKA;1 ) , which is of key importance for both mitosis and meiosis ( Dissmeyer et al . , 2009; Nowack et al . , 2012; Wang et al . , 2004a ) . However , the molecular targets of TAM have not been identified and a mechanistic understanding of its role in meiosis is missing . Cytological studies of mutants defective in tam and other meiotic genes have so far exclusively relied on the analysis of fixed material by cytochemical methods such as chromosome spreads and the immuno-detection of proteins . While these techniques have been , and continue to be , very informative , they capture the underlying cellular dynamics only to a small degree . Importantly , these methods do not allow individual cells to be followed over time . Thus , conclusions about the course of meiocyte development and progression through meiosis have to be deduced from the analysis of different cell populations at different time points . So far , only two approaches to observe meiosis in real time in plants have been described , revealing details about spindle dynamics and chromosome paring in maize meiocytes . First , the work of Yu et al . and its modification by Nannas et al . used fluorescence microscopy to observe isolated male meiocytes cultured for a maximum of 9 hr in liquid medium ( Nannas et al . , 2016; Yu et al . , 1997 ) . The method of Nannas et al . combined the DNA dye Syto12 with the expression of β-tubulin fused to CFP , thereby allowing the concomitant observation of chromosomes and MTs . This revealed a spatially asymmetric positioning of the spindle at anaphase I and II , and chromosome-dependent phragmoplast deposition ( Nannas et al . , 2016 ) . The second approach involved imaging entire anthers of maize by exploiting the high depth of field of two-photon microscopy , as earlier proposed by Feijó et al . ( Feijó and Cox , 2001; Feijó and Moreno , 2004; Sheehan and Pawlowski , 2012; Sheehan and Pawlowski , 2009 ) . This method , which allowed imaging for periods of 24 hr , led to the characterization of three different movements and trajectories followed by the chromosomes during pairing in prophase I ( Sheehan and Pawlowski , 2009 ) . The studies in maize relied on visualizing DNA by chemical stains such as Syto12 and DAPI and the power of Arabidopsis as a molecular model , which enables the relatively fast generation of fluorescent reporter lines for different meiotic proteins , has largely not been exploited in combination with live cell imaging of meiosis . A first approach was made by Ingouff et al . who observed methylation changes during Arabidopsis sporogenesis and gametogenesis , albeit without resolving specific meiotic stages ( Ingouff et al . , 2017 ) . Here , we set out to develop a live cell imaging system for meiosis in Arabidopsis . To this end , we have generated an easy applicable microscopic set up , a combination of meiotic reporter lines covering central aspects of meiosis , and an evaluation system based on morphological characteristics that allowed the quantification of meiotic phases with high temporal resolution . This work gives insights into the robustness of meiocyte differentiation steps and provides important criteria to judge and/or re-evaluate mutants affecting meiosis . As a test case , we have re-analyzed tam mutants and find new phenotypic aspects that suggest that TAM is a central factor coordinating the cytoskeleton with nuclear events .
Live cell imaging can be performed at three general levels and all three have been applied to the analysis of meiosis in multicellular organisms . First , imaging can be performed on isolated cells as for instance seen in the case of mammalian oocytes where confocal microscopy was applied to analyze chromosome segregation: kinetochores could be tracked for over 8 hr , revealing that the bi-oriented attachment of homologs is established after a lengthy try-and-error process ( Kitajima et al . , 2011 ) ; microtubule organizing centers ( MTOC ) and actin elements of the cytoskeleton have been shown to be relevant for spindle formation and correct segregation ( Holubcová et al . , 2013; Mogessie and Schuh , 2017; Schuh and Ellenberg , 2007 ) , as well as it was confirmed by live cell imaging of fetal mouse oocytes that cohesin establishment is maintained without detectable turnover and that its loss in older oocytes remains uncorrected , leading to formation of aneuploid and non-viable gametes ( Burkhardt et al . , 2016 ) . This approach usually gives very high spatio-temporal resolution since there is no requirement for the laser beam to penetrate surrounding cells and very little laser power can be used reducing photobleaching and phototoxicity . However , since meiocytes in Arabidopsis are very small , that is 20 μm , and difficult to isolate , we did not explore this possibility further . Next , imaging can be carried out in the context of an entire organism , for example in C . elegans ( Mullen and Wignall , 2017; Rosu and Cohen-Fix , 2017 ) with the benefit of perturbing the analyzed cells as little as possible by preparation procedures . However , this set up is limited to small organisms and/or short observation times due to size restrictions and the problem of movement of the sample and thus , cannot be used for Arabidopsis either . Finally , live cell imaging can be performed on isolated organs or tissues that are typically easy to obtain and that provide the appropriate developmental context for analysis of the selected individual cells , for example in mice ( Enguita-Marruedo et al . , 2018 ) , C . elegans ( Mlynarczyk-Evans and Villeneuve , 2017 ) and Drosophila melanogaster ( Głuszek et al . , 2015 ) . As conventional confocal laser scanning microscopes can reach cells up to a depth of 70–100 μm , they are suited to observe the meiocytes in Arabidopsis that are covered by three cell layers in the anthers . Imaging of isolated organs has already been successfully applied to the analysis of organogenesis in the shoot apical meristem ( SAM ) of Arabidopsis ( Hamant et al . , 2014; Reddy et al . , 2004; Reddy and Meyerowitz , 2005 ) . Since shoots could be maintained for several days without obvious perturbations of development , we decided to adapt and optimize this approach for our purposes . First , we harvested inflorescences and removed all but one young flower primordium presumably containing meiotic stages as indicated by its round shape and an approximate diameter of 0 . 4–0 . 6 mm ( Figure 1 ) , corresponding to stage 9 of flower development ( Smyth , 1990 ) . Next , the upper sepal was removed giving access to two of the six anthers since the petals are shorter than the anthers at this floral stage . Finally , the bud along with the pedicel and a few millimeters of the stem was embedded into Arabidopsis Apex Culture Medium ( ACM ) and stabilized with a drop of agarose ( Figure 1A , B ) . In agreement with the previous analysis of the SAM , we found that the flower buds stayed alive on the ACM medium for up to seven days during which all the flower organs expanded , revealing that cells were able to undergo several divisions on the medium ( Figure 1C ) . Imaging was performed with an up-right confocal laser scanning microscope equipped with a water immersion objective . The entire flower bud was submerged in water and the objective was brought into direct contact with the sample ( Figure 1A ) . During image acquisition the temperature was kept constant at 21°C . To test viability of the sample after image acquisition , we transferred flower buds , which were imaged for 24 or 48 hr , onto new ACM medium and allowed them to grow for 3 days . Concomitant with the growth of the entire flower bud , we found that the microspores derived from imaged meiocytes underwent at least one more cell division giving rise to bi-cellular pollen as revealed by DAPI staining ( Figure 1—figure supplement 1 ) , confirming that meiocytes are still alive after imaging . A generic set up for imaging of cell divisions includes a reporter that highlights DNA/chromatin coupled with a marker for cytoskeletal components , usually MTs , so that chromosome and spindle behavior can be visualized ( Nannas et al . , 2016; Peirson et al . , 1997 ) . Since fusions of histones with fluorescent proteins have often been applied for this purpose , we first scanned through previously generated transgenic lines expressing different histone variants fused to fluorescent proteins , such H2B . However , while these labeled histones clearly marked DNA in somatic cells , the signal was often fuzzy in meiosis . Moreover , since all or most cells in an anther produced these fusion proteins , the identification of meiocytes was sometimes difficult , especially at early stages of meiosis when chromosomes are not condensed and meiocytes cannot easily be recognized by their size and shape . Therefore , we aimed for a meiosis-specific gene and generated a GFP fusion to REC8 , the alpha kleisin subunit of the cohesin complex , also known as SYN1 or DIF1 in Arabidopsis ( Bai , 1999 ) . Cohesin is key for chromosome segregation and its step-wise removal allows the segregation of homologous chromosomes in meiosis I , followed by separation of sister chromatids in meiosis II . In addition , cohesin is required for recombination and repair of DNA double-strand breaks resulting in a highly pleiotropic phenotype that leads to almost complete sterility of rec8 mutant plants ( Bai , 1999 ) . Expression of our genomic PROREC8:REC8:GFP reporter in a homozygous rec8 mutant background completely restored fertility of these plants and analysis of chromosome spreads confirmed that chromosome segregation is indistinguishable from the WT ( Figure 2—figure supplement 1 ) . REC8 replaces the mitotic RAD21 in meiosis and is hence highly specific for meiocytes in all species analyzed so far ( Nasmyth , 2001 ) . Consistent with previous immuno-localization studies , we found that the GFP signal of our functional reporter line was only present in meiocytes providing a straightforward way to identify microspore mother cells ( Figure 2 ) . Moreover , the REC8 reporter allowed us to estimate the sensitivity of our imaging procedure . While REC8 is removed from chromosome arms at the end of meiosis I to allow the resolution of cross-overs , a small fraction remains at the centromeres to maintain sister chromatid cohesion . The detection of the centromeric fraction of REC8 has been challenging in immuno-localization studies . When we followed the first meiotic division , we observed the remaining REC8 at centromeres indicating that our live cell imaging system is highly sensitive ( Figure 2B , Video 1 ) . Next , we combined the PROREC8:REC8:GFP with PRORPS5A:TagRFP:TUB4 or PRORPS5A:TagRFP:TUA5 that label MTs and hence permit observation of the cell shape and the formation of the spindles . The resulting double reporter line is referred to as Kleisin IN Green microtuBules In ReD ( KINGBIRD ) in the following . Plants carrying the double constructs , as well as plants expressing only the PRORPS5A:TagRFP:TUB4 or PRORPS5A:TagRFP:TUA5 reporter did not have meiotic defects and seed production was as in the WT ( Figure 2—figure supplement 1 ) . The separated excitation and emission spectra of the two different fluorochromes permitted faithful and concomitant detection of both , the REC8 and the tubulin reporter . An important question was how many frames per time interval should be taken . Due to photo-bleaching as well as potential photo-toxicity , a sampling rate of several frames per minute was not compatible with capturing the entire meiotic program . Based on a several test samples as well as previously published time courses ( Armstrong et al . , 2003; Sanchez-Moran et al . , 2007; Stronghill et al . , 2014; Yu et al . , 1997 ) , we decided to acquire one frame every 3 to 15 min , so that even the shortest phases such a metaphase I and II , could be captured while photo-bleaching was reduced to a minimum . Meiosis is classically apportioned into nine phases: prophase I , metaphase I , anaphase I , telophase I , interkinesis , prophase II , metaphase II , anaphase II and telophase II . Due to the dramatic changes in chromatin structure and the dynamics of chromosomes , and to its prolonged duration , prophase I is divided into the five sub-phases , that is leptotene , zygotene , pachytene , diplotene and diakinesis . These phases have been derived from observation of fixed material and chromosome spreads , leading to definitions mainly based on chromosome configurations , for example pachytene is defined by the presence of fully synapsed chromosomes . Using the KINGBIRD reporter line , we were able to distinguish five parameters of meiocytes: cell shape , MT array , nucleus position , nucleolus position , and chromosome configurations ( condensation and pairing/synapsis ) ( Figure 3A ) . Some of these parameters could be identified as a direct read-out of the reporters , for example cell shape and MT array are visualized by tubulin while chromosome configurations are revealed by REC8 . Other parameters could be indirectly determined as for the nucleus position , which is defined by the absence of MTs , or the nucleolus position , which corresponds to the area of the nucleoplasm where REC8 is present . Each parameter can adopt different states , which have a distinct order . For instance , we observed that cell shape always changed from rectangular to trapezoidal , then to oval , to circular , to triangular and finally giving rise to tetrads composed of four triangular cells , among which typically only three can be identified on the same focal plane ( Figure 3A ) . The MT array is the parameter with the largest number of states . At the onset of meiosis , MTs are homogeneously distributed in the cytoplasm ( state 1 ) and then progressively form an arc structure , hereby named half moon , which develops on one side of the nucleus , which moves at the same time towards a corner of the cell ( states from 2 to 4 ) . The half moon then develops into a full-moon-like structure surrounding the nucleus ( state 5 ) and contracts to form a pre-spindle similar to what is observed in mitosis ( state 6 ) . At the moment of nuclear envelope break down , the pre-spindle is disrupted ( state 7 ) and MTs rearrange to form the first meiotic spindle ( states 8 and 9 ) . States 10 and 11 are present during the transition between meiosis I and meiosis II , MTs reorganize radially around the two new nuclei while the central MTs broaden their disposition forming a phragmoplast-like structure . The second meiotic division resembles the first meiosis , with the formation of two pre-spindles ( state 12 ) followed by two spindles ( state 13 ) and phragmoplast-like structures ( state 14 ) until the cells undergo cytokinesis and form a tetrad ( state 15 ) . Nucleus and nucleolus are characterized by changes of their positions . At the beginning of meiosis , the nucleus is centrally located ( state1 ) . Around the time of the formation of the MT half moon structure , it then moves to one side of the cell ( state 2 ) and at state 3 the nucleus is back to the center of cell . This state is distinguishable from state 1 due to the size of the nucleus , which is now enlarged . During the two metaphases ( states 4 and 6 ) , the nuclear structure disappears , to reappear as two and four smaller nuclei at the following states 5 and 7 , respectively . The nucleolus becomes visible at the onset of meiosis , together with the accumulation of REC8 . It is initially positioned in the center of the nucleus ( state 2 ) and moves to the nuclear periphery during the progression of prophase ( state 3 ) . At late prophase it disappears ( state 4 ) . Since the nucleolus is only visible when REC8 is expressed , its reappearance after chromosome segregation could not be noticed . The last parameter obtained is the localization of REC8 , which correlates with chromatin conformation during the first meiotic division . The states identified resemble the localization pattern of REC8 described by immunolocalization experiments ( Cai et al . , 2003 ) . At first , a diffuse fluorescence signal accumulates in the nucleoplasm ( state 2 ) , which then condenses to form threads that thicken over time ( states 3 and 4 ) indicating the pairing of chromosomes . At state 5 , the REC8 signal becomes fainter consistent with the onset of REC8 removal from chromosomes by the prophase pathway ( Yang et al . , 2019 ) . Soon , the chromosome threads are unrecognizable but a faint diffuse REC8 staining persists ( state 6 ) until small , distinct dots at state 7 are observed representing fully condensed chromosomes . State 7 is followed by the almost complete disappearance of REC8 ( state 8 ) , corresponding to the onset of anaphase I . Importantly , our markers recapitulated previously described changes in nucleolus position , REC8 localization and MT cytoskeleton organization , corroborating that our imaging system does not disturb the overall progression of meiosis ( Peirson et al . , 1997; Stronghill et al . , 2014; Wang et al . , 2004b ) . Analyzing a first set of movies gave rise to the hypothesis that some of the parameter states are connected , for example the nucleolus apparently dissolves only after the nucleus has moved to one side of the meiocyte and returned to a central position . To assess the nature of these associations , we analyzed a subset of cells ( n = 169 from 35 anthers ) assigning a combination of numbers that represents each parameter state at every time point when a frame was taken , for example 1-1-1-2-2 describes a meiocyte that is rectangular in shape , has an evenly distributed MT array , a centrally located nucleus with a centrally located nucleolus , and with uncondensed , yet not paired chromosomes . In the following , we call a combination of all five parameter states a cellular state . A subsequent analysis of 10 , 671 time points allowed us to judge which parameter states occur together and in which frequency ( Figure 3B ) . By this method we could confirm that certain parameter states indeed co-occur with each other in a highly specific manner , for example cell shape state 1 is only found to be associated with nucleus position state 1 ( Figure 3B , WT heatmap ) , while others parameter states are only more loosely associated . This analysis also revealed that out of the more than 20 , 000 possible cellular states only 101 were actually present in our data set ( Figure 4—source data 1 ) . However , their frequencies were distributed in a very broad range ( from 0 . 01% to 21 . 14% of the total number of observations ) . Hence , the importance of a certain cellular state cannot be deduced from the absolute frequency of occurrence since this is highly biased by the duration of the respective state , that is combinations of parameters that depict long phases such as pachytene are present in higher number of time points than combinations depicting short phases , for example metaphase I . To identify biologically distinct cellular states from the observed data , we defined a local or neighboring score , which quantifies the occurrence of a certain cellular state compared to its neighboring states . A neighboring state was defined as a cellular state that is one transition away ( −1 or +1 ) for at least one , but at most two , parameter states compared to the cellular state analyzed . With this , 2-4-2-3-4 , for example , is a neighbor of 2-3-2-3-4 and of 3-3-2-3-4 , but not of the cellular state 2-2-2-3-4 and not 3-3-2-3-3 ( Figure 4—source data 1 ) . Notably , we only took states into account that were actually observed . The neighboring score was then compared with the subset of neighboring states , to find the predominant state among the surrounding states , and is defined as:Score = count ( state ) −mean ( count ( neighboring states ) ) std ( count ( neighboring states ) ) where counts refers to the number of times a certain state is observed in the data , and std refers to the standard deviation . This analysis revealed 11 clearly distinct cellular states that differed from their neighbors with a score higher than one , denoting that they occurred at least one standard deviation more frequent than the mean of the neighboring stages ( Figure 4 ) . These 11 cellular states ( A1-A11 ) are henceforth called meiotic landmark states or landmarks ( Figure 4 and Figure 5 ) . The states between landmarks are defined as transition states , and often represent alternative routes to the next landmark ( Figure 4 ) , for example the cell shape may first change from rectangular to trapezoidal and then the nucleus moves from a center position to a position at the side of the cell , or the nucleus moves first and then the cell shape changes . However , the nucleus is finally always located at the smaller side of the trapezoidal cell defining the new landmark state . The results of the neighboring score analysis were reproduced and confirmed by bootstrapping ( Figure 4—source data 2 ) . Taken together , we conclude that cellular differentiation steps of meiosis can be variable but then converge on distinct cellular states , the landmarks . The qualitative assortment of the landmarks , possibly their order as well as their duration and the degree of variability ( transition state number and duration ) , represent a new system to describe meiosis . The break-down of the nuclear envelope in diplotene is an important hallmark of meiosis ( Wijnker and Schnittger , 2013 ) . We also could clearly observe the breakdown in our live cell imaging system although , due to its rapid progression , it was only captured in 22 out of 10 , 671 analyzed time points with a sampling interval of one frame every 10 min ( Figure 6 and Video 2 ) . Nonetheless , the nuclear envelope break-down is not included in a landmark state since it appeared to be only loosely connected with the other parameter states , for example the cell shape can be oval or round , and the chromatin can be at different condensation levels when the nuclear envelope breaks down ( Figure 6 ) . Thus , although very distinct when looking at MT conformation ( i . e . state 7 , collapse of pre-spindle Figure 3A ) , a clearly defined landmark state corresponding to nuclear envelope breakdown was not reached with the parameters analyzed . We could also clearly observe other short-lived phases such as diakinesis , anaphase I , prophase II and anaphase II . However , due to their unexpected high variation in terms of association with the here analyzed parameter states , these phases , like nuclear envelope breakdown were also not designated as landmarks . Our sample preparation , which keeps anthers intact , also provided the possibility to follow the differentiation of the tissues surrounding the meiocytes , especially the tapetum cells . These are in direct contact with the meiocytes and are thought to nourish and support the meiocytes and spores ( Pacini et al . , 1985 ) . A key feature of tapetum cells in many plant species , including Arabidopsis , is that they become poly-nucleated through endomitosis , that is a cell cycle variant in which cytokinesis is skipped ( Jakoby and Schnittger , 2004 ) . The poly-nuclearization of tapetum cells was clearly visible in our KINGBIRD line ( Video 3 , from minute 980 to minute 1207 ) , possibly representing a sixth cell parameter next to the five meiotic parameters presented above ( Figure 3A ) . Notably , tapetum cell differentiation was previously suggested as a criterion to judge stages of meiosis ( Stronghill et al . , 2014; Wang et al . , 2004b ) . We observed that polynucleated tapetum cells are not found before A4/zygotene and conversely , when all tapetum cells are poly-nucleated , meiosis has progressed into A7/diplotene . However , endomitosis only poorly correlated with any of the meiotic stages between A4 and A7 ( Figure 3B ) and hence , was not incorporated into the landmark system . In turn , we conclude that the meiotic progression and tapetum cell differentiation are not tightly correlated . One obvious future experiment is to combine the here-developed landmark system with additional reporter lines for meiotic regulators , for example ZYP1 and ASY3 ( Yang et al . , 2019 ) . However , green fluorescent proteins ( GFP and possibly mNEONgreen ) are still by far the most powerful reporters due to their high quantum yield and photostability , especially for poorly expressed genes . Hence , we were interested in which landmarks could be revealed using only one color , that is either PROREC8:REC8:GFP or PRORPS5A:TagRFP:TUB4 alone leaving the second color , RFP and GFP , respectively , for labeling another protein of interest . PROREC8:REC8:GFP allows the observation of chromatin condensation levels and of nucleolus position , while PRORPS5A:TagRFP:TUB4 reveals cell shape , nucleus position and MT array ( Figure 3A ) . As expected , excluding some parameters by relying on only one reporter resulted in a reduced number of observed cellular states: 52 for PRORPS5A:TagRFP:TUB4 and 14 for PROREC8:REC8:GFP , compared with the 101 states identified by analyzing the KINGBIRD line; ( Figure 5—source data 1 ) . Ultimately , this led to a lower number of landmarks with a neighboring score higher than 1 in comparison to the analysis with two reporters ( Figure 5 and Figure 5—source data 1 ) . Analysis of PROREC8:REC8:GFP by itself delivered landmarks A1 , A5 and A7 while PRORPS5A:TagRFP:TUB4 revealed landmarks A1 , A3 and A7 to A11 ( Figure 5 ) . Notably , the landmarks revealed by the KINGBIRD line are not simply the addition of the landmarks unraveled by PROREC8:REC8:GFP and PRORPS5A:TagRFP:TUB4 . A2 , A4 , and A6 only appeared as landmarks when both reporters are used indicating the added value of using multiple reporters . Among the two reporters , PRORPS5A: TagRFP:TUB4 turned out to be the most informative when used alone due to the fact that its accumulation pattern covers the complete division from pre-meiotic phases to tetrad stage , and that MT behavior is very distinct in meiosis . Thus , PRORPS5A:TagRFP:TUB4 can be used as a landmark reporter alone and is especially useful for stages after meiosis I . The traditional definition of meiosis , mostly relying on chromosome spreads , and the here-established landmark system by live cell imaging are based on different parameters and aspects of meiosis . Nonetheless , we could , at least roughly , align our landmark-based classification with the traditional definition of meiosis ( Figure 5 ) . We could attribute A1 to an interval between S-phase and early leptotene based on the starting expression of REC8 and its loading onto chromosome arms ( Cai et al . , 2003 ) . A2 could be associated to late leptotene when chromosomes appear as thin threads , as recognized in our case by the REC8 reporter . In addition , the nucleolus , which can be detected by our system by the absence of REC8 , moves to the periphery of the nucleus , which is marked by REC8 , as described for leptotene cells ( Armstrong and Jones , 2003; Ross et al . , 1996; Stronghill et al . , 2014 ) . A3 and A4 fall into zygotene as zygotene cells have previously been found to have the majority of organelles and MTs localized on only one side of the nucleus ( Armstrong and Jones , 2003; Peirson et al . , 1997; Ross et al . , 1996; Stronghill et al . , 2014 ) . Additionally , we observe a thickening of chromosome threads that would be consistent with the formation of the synaptonemal complex , which starts to be formed in zygotene ( Higgins et al . , 2005 ) . Pachytene is characterized by the complete synapsis of homologous chromosomes and the re-positioning of the nucleus into the central area of the cell ( Armstrong et al . , 2003; Armstrong and Jones , 2003; Ross et al . , 1996 ) . Therefore we could link landmarks A5 and A6 to pachytene . A7 is characterized by a diffuse signal of REC8 , which is consistent with the release of synapsis in diplotene ( Cai et al . , 2003; Ross et al . , 1996 ) . A8 can be identified by the formation of a single spindle and five highly condensed chromosomes that align in the metaphase plate , hence representing metaphase I . In A9 two nuclei appear , still connected by MTs , revealing that this stage is telophase I/interkinesis , followed by the formation of two spindles in A10 , which is metaphase II . Finally A11 , where three to four distinct nuclei can be detected without being separated by cell walls , represents telophase II . The assignment of landmarks to the classical stages allowed us then to compare the length of meiotic phases in our live cell imaging approach with the previously performed time course experiments in which the length of meiosis has been estimated by pulse-chase experiments applying either the modified thymine analog 5-bromo-2’-deoxyuridine ( BrdU ) or 5-ethynyl-2’-deoxyuridine ( EdU ) to plants . After a given amount of time , meiotic spreads were prepared and tested for the appearance of these analogs in meiotic chromosome configurations . In these experiments , male meiosis in Arabidopsis was judged to last from G2 onwards approximately 32 to 33 hr with leptotene spanning between 6 and 7 hr , zygotene and pachytene together lasting between 12 and 16 hr . Notably , these previous pulse-chase experiments were not able to resolve stages after diplotene and the rest of meiosis ( from diplotene onwards ) were estimated to approximately persist for 3 hr ( Armstrong et al . , 2003; Sanchez-Moran et al . , 2007; Stronghill et al . , 2014 ) ( Figure 7 ) . A straightforward way to assess the duration of meiosis by our live cell imaging system is by evaluating long movies spanning an entire meiosis . However , long movies with more than 30 hr containing all meiotic stages could only occasionally be obtained and were rarely fully informative due to loss of the focal plane by sample growth . In contrast , 58 movies captured only subsections of meiosis , yet combined provided a complete coverage of meiosis I and II containing each landmark at least 4 times ( Figure 7—source data 1 ) . To faithfully judge the duration of each landmark , the length of one movie had to be long enough to capture at least two transitions of two sequential landmarks in one individual meiocyte ( Video 2 and Figure 7—source data 1 ) . The transition states between two landmarks were added to the observed time of the preceding landmark . Hence , the duration of diakinesis is included in A7 , anaphase I in A8 , prophase II in A9 and anaphase II in A10 ( Figure 5 ) . Since we could not faithfully determined S-phase , the transitions between S- and G2-phase was excluded in our time estimates . We then tracked single meiocytes over time with up to 18 meiocytes per anther . Our measurements of the meiotic phase lengths over all delivered a similar time frame as seen by the previous pulse-chase experiments and we determined the duration of meiosis from late leptotene till telophase II to be 26 hr ( Figure 7 ) . This value excludes the length of landmark A1 ( 8 . 5 hr in total ) , which is marked by the onset of REC8 expression , since this time point is not clearly defined with respect to beginning of S-phase . Prophase I , as expected , resulted to be the longest phase ( minimum 20 hr ) with late leptotene ( A2 ) lasting 1 . 5 hr , zygotene ( A3-A4 ) 6 hr , pachytene ( A5-A6 ) 9 . 5 hr and diplotene and diakinesis ( A7 ) together 3 hr . Importantly , we could also resolve meiotic phases thereafter and determine metaphase I and anaphase I ( A8 ) together with 1 hr , telophase I , interkinesis and prophase II ( A9 ) with 1 hr and meiosis II ( A10-A11 ) all together with 4 hr ( Figure 7 , Figure 7—source data 1 and 2 ) . Summing up , the here-presented landmark system allows a dissection of meiosis with unprecedented temporal resolution . Given that the overall length of meiosis as well as the evaluation of individual sub-phases match previously determined durations , we conclude that our imaging system does not perturb meiosis and hence can be applied to analyze different mutants and to assess environmental conditions in the future . To test whether our imaging system can help to promote our understanding of meiotic mutants , we decided to analyze tam , which is one of the most studied meiotic mutants in Arabidopsis with at least six published reports focusing on its function ( Bulankova et al . , 2013; Bulankova et al . , 2010; Cromer et al . , 2012; d'Erfurth et al . , 2010; Magnard et al . , 2001; Wang et al . , 2004a ) . What has been reported for tam null mutants is that their meiotic progression is delayed in prophase I from pachytene onwards and that they eventually terminate meiosis after the first division . This termination is especially prominent on the male side with nearly 100% of all meiocytes producing dyads instead of tetrads . We introduced the KINGBIRD construct into tam mutants and subjected the resulting plants to our live cell imaging procedure . A total of 31 movies capturing 143 male meiocytes were generated covering the entire meiosis in tam . We first asked which states of the five cellular parameters can be found in mutant plants and annotated the cellular states of 62 meiocytes . The same states for nucleolus position and chromatin condensation were found in tam in comparison to the WT . When looking into cell shape states , we found neither the triangular nor the tetrad configuration that are characteristic for meiosis II , consistent with the finding that tam terminates meiosis after the first meiotic division . Matching a premature termination , we also noted that MT and nucleus position states of meiosis II are absent in tam . Strikingly , we discovered an additional state with aberrant MT configurations ( state 6 ) , which has not been recognized in previous analyses of tam mutants ( Figure 3A ) . This state is characterized by the formation of ectopic spindle-like or phragmoplast-like structures in the cytoplasm of meiocytes during diplotene , that is when chromosomes are already connected by chiasmata , but before nuclear envelope breakdown . The ectopic MT configurations were found to adopt different conformations , that is one large array on one side of the nucleus , two separate entities on one side of the nucleus or different clusters of MTs surrounding the nucleus ( Figure 8 ) . As an additional feature , we also observed , albeit rarely , small dark areas in the cytoplasm occurring already before nuclear envelope breakdown but clearly visible after telophase I , possibly indicating micronuclei ( Figure 8 and Video 4 and Video 5 ) . All together , 102 cellular states were extracted from the movies of tam and we next checked the binary co-occurrence of these states ( Figure 3B ) . The heatmap reveals a higher degree of disorder for some of these states when compared with the WT situation ( highlighted in magenta in Figure 3C ) . Especially prominent was an altered relationship between cell shape and MT array underlining that TAM might regulate MT organization . Likewise , the correlation between chromatin and MT array as well as between nucleus position and cell shape was less stringent in tam when compared to the WT . Possibly , the latter difference was induced by the appearance of the aberrant MT structures , which might push the nucleus back to the center of the cell in some cases whereas in the WT the nucleus is strictly positioned at one side of the cell . In turn , this might also lead to cell shape changes in tam causing an oval shape again at a time point when in the WT the cell shape has reached a round shape ( Figure 3B ) . Next , we performed our neighboring analysis , coupled with a determination of the time course of meiosis in tam , as performed before for the WT ( Figures 4 , 8 and 9 , Video 4 and Figure 4—source data 3 , Figure 7—source data 1 , Figure 9—source data 1 and 2 ) . The analysis revealed a total of 12 landmarks in tam , named t1 to t12 ( Figures 4 and 8 , Figure 4—source data 3 ) . The initial cellular stage ( 1-1-1-1-1 ) was added in the scheme and named START correspondingly to what was done for the analysis of WT meiosis . As expected , landmarks describing meiosis II were never observed in male meiosis of tam mutant plants while new landmarks appeared , that is t5 , t7 , and t8 ( Figure 4 , Figure 8A , Figure 9A , and Video 5 ) . Notably , transition states ( non-landmark states colored from blue to light yellow in Figure 4 ) , reflecting cellular variation , become abundant during the late prophase in tam exactly when the here discovered new aspects of the mutant appear . Prior to the defects in late prophase , we already saw that the wild-type landmark A3 ( 2-2-2-3-3 ) is strongly underrepresented in tam , that is it is recorded only once out of 6092 total analyzed time points ( 0 . 02% ) while in WT , A3 is scored 497 times with a percentage of 4 . 6% and neighboring score pair to 1 . 32 . Consequently , this state obtains a very low neighboring score of −0 . 49 and it is not a landmark in tam ( Figure 4—source data 3 ) . The reason for the disappearance of the landmark corresponding to wild-type A3 in tam is not entirely clear . Likely linked to the disappearance of this landmark is the extension of landmark t2 ( corresponding to A2 in the WT ) suggesting that TAM is required to organize the MT is one corner of the cell . Once this is achieved , tam mutant cells appear to quickly change their cell shape and proceed to state t3 ( corresponding to A4 in the WT ) . Since t2 takes longer than A2 but then A3 is skipped , the total length of early prophase is very similar between the WT and tam ( Figure 9 ) . In contrast , late prophase is extended in tam ( Figure 9 , Video 4 and Video 5 ) . This prolongation is linked to the appearance of the additional landmark t5 between the WT landmarks A5 and A6 , as well as to the presence of the new landmarks t7 and t8 both between the wild-type landmarks A6 and A7 . When looking into the defects of late prophase I in tam in detail , we recognized the co-existence of two different populations ( A and B ) of meiocytes within the same anther . Population A ( 45 . 8% of cells , n = 155 ) , develops the above-described spindle-like and phragmoplast-like structure in the cytoplasm while the nucleus is still intact , and proceeds from landmark t6 via t7 to t9 , skipping landmark t8 . In contrast , population B progresses through landmarks which largely resembled the ones seen in the WT , going from t6 via t8 to t9 , that is skipping landmark t7 , yet progressing through meiosis with a much slower speed than the WT ( Figures 8A and 9A , Video 4 and Video 5 ) . The reason for the appearance for the two populations is as yet not clear and they have probably not been recognized before since population A also is slower that the WT . Thus , tam meiocytes proceed through meiosis at a similar speed ( Figure 9 ) , and do not show strong asynchrony in one anther prior to telophase ( Video 4 and Video 5 ) . Taken together , our analysis assigns TAM the function of a major regulator of MT organization in meiosis . Likely , TAM acts already early in prophase I as seen by the extension of t2/A2 in leptotene . One major function appears to be then the repression of premature spindle and/or phragmoplast formation in diplotene .
Our meiotic description is based on five morphological criteria of male meiocytes that we could distinguish with our reporter genes , that is , cell shape , position of the nucleus , position of the nucleolus , REC8 status and information about chromatin state , and MT array . Importantly , we found that these cellular parameters have two aspects , which make them suitable for a classification system . First , they change in the course of meiosis in a unidirectional manner , for example , cell shape changes from rectangular over trapezoidal and oval to circular . We never found an example where a WT meiocyte skipped one of these cell shape changes or changed back from a later stage to an earlier stage . Second , these parameters are linked with each other and build a matrix . For instance , full-moon MT array was never found to be associated with a rectangular cell shape of the meiocyte ( Figure 3B ) . Our analysis of cellular parameters allowed us to identify 11 prominent morphological states , called landmarks A1-A11 . These differ from each other by at least one parameter state , and always occur in the same order in any cell progressing through meiosis . The pathway taken by an individual meiocyte to reach each landmark could differ slightly , presumably due to biological variation , and is described by a network of the transition states ( Figure 4 ) . It is an interesting question to what degree this developmental plasticity depends on meiotic genes and/or environmental factors such as temperature . The 11 landmarks together with their transitions could be assigned to the classical phases of meiosis ( Figure 5 ) . However , it has to be noted that the alignment of our landmarks with the classically defined stages remains fuzzy for certain phases . For example , leptotene is defined by the beginning of the chromosome pairing process , with the appearance of the first thin threads , a cell feature that we could not clearly resolve in our analysis . However , as more meiotic reporter lines are generated , for instance for the lateral or central elements of the synaptonemal complex , pairing and synapsis will be resolved with enhanced resolution in future . In this regard , the landmark system is highly modular and expandable depending on the resolution needed by the researcher . Already with the current setup , our system allows an accurate and robust determination of meiotic stages . This is important since not all cell characteristics can always be unambiguously resolved , for example when the fluorescent signal diminishes because of photobleaching . Hence , the combined parameter states together with the knowledge about the previous cell stages maximize the information gained . Our landmark system provides a powerful novel platform to study meiocyte differentiation and quantify meiotic progression . The observation that some of the cellular parameters are connected possibly indicates a common regulatory base and/or regulatory dependency . While some associations were expected , for example changes in MT cytoskeleton and cell shape , and are possibly directly linked , other combinations are new and unexpected , for instance the correlation between nucleolus movement and the MT cytoskeleton . These correlations can of course be indirect , yet exploring these combinations in future and identifying which genetic factors underlie them opens a new perspective into meiosis . In turn , their potential uncoupling provides additional , qualitative criteria to describe meiotic mutants . By observing more features in the future through the use of additional reporter lines and the analysis of mutants affecting meiosis , it will be possible to obtain a highly informative network of functional relationships , that is coupled and uncoupled parameters , within a meiocyte thus heading towards a system-biology understanding of meiosis . Importantly , it will be interesting to see to what degree these cellular parameters can be found and are coupled with each other in female meiocytes . Similarities and differences can further be compared with the behavior of meiocytes in other plant species . Applying our technique to tam , a long-known and well-described meiotic mutant in Arabidopsis , revealed surprising new phenotypes and allowed us to quantitatively dissect this mutant . Most strikingly , we observed the formation of spindle or phragmoplast-like structures in the cytoplasm of meiocytes in diplotene , that is at a timepoint when chromosomes are still enclosed in the nucleus . Thus , spindle- and phragmoplast formation can be uncoupled from the presence of chromosomes . However , as soon as the nuclear envelope breaks down and chromosomes are accessible , the ectopic spindle-/phragmoplast-like structures are quickly disassembled and reorganized into a proper meiotic spindle consistent with the finding that chromosomes have a strong MT organizing force . We therefore conclude that one of the major functions of TAM is to prevent self-organization of MTs prior to the presence of highly condensed chromosomes . These results possibly resemble a situation found in mitotic cells where high CDK activity inhibits the function of NACK1 , a kinesin , and NPK1 , a MAP3K . NACK1 together with NPK1 trigger a MAP kinase phosphorylation cascade that results in the activation of MAP65-3 , a central MT organizing force that drives phragmoplast formation ( Sasabe et al . , 2011 ) . It is tempting to speculate that TAM functions in addition to prevent cytokinesis after the separation of chromosomes in anaphase I in a similar manner . However , the targets of a meiotic CDK-TAM complex are not known and it remains to be seen whether a meiotic NACK1 and/or NPK1 homolog is subject to phospho-regulation by a CDK-TAM complex . The second major finding when analyzing tam mutants was that there are two different populations of mutant meiocytes with only one of these populations developing ectopic spindle- and phragmoplastlike structures . The other population progressed through meiosis reaching landmarks that are also found in the WT , yet in a much slower fashion . This could indicate a crucial dose-dependency for kinase activity , that is cells that for unknown , stochastic reasons have very little kinase activity in addition to the loss of TAM develop ectopic spindle- and phragmoplast like structures while intermediate levels of kinase activity due to the loss of TAM are only slowed down . Alternatively , high kinase activity in diplotene might be needed to establish a special state , for example a spatial mark . In absence of TAM , the establishment of the state is less stable and causes the formation of ectopic spindle- and phragmoplast like structures in these cells . Clearly , additional work is required to unravel the complexity of TAM action in meiosis . However , the here unraveled phenotypes give a clear direction for future experiments and underline the power of life cell imaging without which the behavior of different populations of meiocytes is hardly possible to identify .
The Arabidopsis thaliana plants used in this study were all derived from the Columbia ( Col-0 ) ecotype . The REC8 T-DNA insertion line rec8 ( At5g05490 , SAIL_807_B08 ) and the TAM T-DNA insertion line tam ( At1g77390 , SAIL_505_C06 ) were obtained from Syngenta via NASC . All genotypes were determined by polymerase chain reaction ( PCR ) using the primers indicated in Table 1 . All seeds were surface-sterilized with chlorine gas , sown on 1% agar plates ( half-strength Murashige and Skoog ( MS ) salts , 1% sucrose , pH 5 . 8 ) and stored 3 days at 4°C in the dark for stratification . Antibiotics were added for seed selection when required . For germination , plates were transferred to long-day condition ( 16 h day/8 hr night regime at 22°C/18°C ) . After germination , plants were transferred to soil and grown under short-day conditions for 2 weeks ( 12 h day/12 hr night regime at 21°C/18°C ) , and then transferred to long-day conditions until seed production . For all crosses , flowers of the female parent were emasculated 1 day before anthesis and hand-pollinated 1 to 2 days later . To generate the PROREC8:REC8:GFP construct , a 7 , 145 bp genomic fragment of the REC8 gene containing a 1 . 8 kbp fragment upstream of the start codon ( ATG ) and 0 . 5 kbp fragment downstream of the stop codon was amplified with the primers AT5G05490-F and AT5G05490-R ( Table 1 ) and cloned into pENTR/D-TOPO . A SmaI site was inserted in front of the stop codon of the REC8 construct by PCR using the primers REC8 CterSmaI-F and REC8 CterSmaI-R ( Table 1 ) . The ORF for monomeric GFP ( mGFP ) was inserted into the SmaI site to create pENTR/PRORECREC8:REC8:GFP , followed by LR recombination reaction into the destination vector pGWB501 ( Nakagawa et al . , 2007 ) . A REC8 reporter line was established by floral dip transformation of rec8 heterozygous plants with the above-described construct followed by selection of T1 plants on 0 . 5X MS agar plates supplemented with 25 mg/L Hygromycin B and 50 mg/L Carbenicillin . T2 seeds from individual T1 plants were germinated on 0 . 5X MS agar plates supplemented with 25 mg/L Hygromycin . T2 line #3 has been selected as the best performing line in terms of rec8 rescued phenotype and fluorescence intensity . The PRORPS5A:TagRFP:TUB4 line has been provided by Takashi Ishida , ( Kumamoto University ) . A PROREC8:REC8:GFP/PRORPS5A:TagRFP:TUA5 was generated using the MultiSite Gateway . The PRORPS5A:TagRFP:TUA5 part was amplified from pGWB501/PRORPS5A:TagRFP:TUA5 with the primers attB4 TUA5-F and attB1r TUA5-R and cloned into pDONR-P4P1r to create pDONR-P4P1r/PRORPS5A:TagRFP:TUA5 . The pENTR/PROREC8:REC8:GFP and pDONR-P4P1r/PRORPS5A:TagRFP:TUA5 were combined into the destination vector R4pGWB501 by LR recombination reaction . The KINGBIRD reporter line in the WT background has been generated via crossing of plants containing the PROREC8:REC8:GFP and PRORPS5A:TagRFP:TUB4 constructs as described above . The KINGBIRD reporter line in the tam background was generated via Agrobacterium-mediated transformation of the vector PROREC8:REC8:GFP/PRORPSRPS5A:TagRFP:TUA5 into heterozygous tam plants . Rescue of the rec8 phenotype was assessed at pollen level using a Peterson staining protocol as described in Peterson et al . ( 2010 ) and monitoring meiotic progression at a cytological level via cell spreads as described in Ross et al . ( 1996 ) . Flowers of 0 . 4–0 . 6 mm were isolated and prepared as presented in the results section ‘Specimen preparation’ . Up to four samples were positioned on the same petri dish and cultured in Arabidopsis Apex Culture Medium ( ACM ) : half-strength Murashige and Skoog ( MS ) salts , 1% sucrose , 0 . 8% agarose , pH5 . 8 . Supplements were added to a 1X concentration from a 1000X stock solution ( stock solution: 10% Myoinositol , 0 . 1% nicotinic acid , 0 . 1% pyridoxine hydrochloride , 0 . 1% thiamine hydrochloride , 0 . 2% glycine dissolved in Millipore water and subsequently filter sterilized ) ( Hamant et al . , 2014 ) . Time lapses were acquired using a Zeiss LSM 880 confocal microscope and ZEN 2 . 3 SP1 software ( Carl Zeiss AG , Oberkochen , Germany ) . During image acquisition the petri dish was filled with autoclaved water and placed under a W-plan-Apochromat 40X/1 . 0 DIC objective ( Carl Zeiss AG , Oberkochen , Germany ) . GFP was excited at λ 488 nm , and detected at λ between 498–550 nm . RFP was excited at λ 561 nm and detected at λ between 578–650 nm . Autofluorescence from chloroplasts was highlighted in blue using excitation at λ 488 , and detection at λ between 680–750 nm . Time lapses were acquired as series of Z-stacks ( six planes , 50 μm distance ) . Interval time was varying from a max of 15 to a min of 3 min depending on sample conditions . The functions ‘Autofocus’ and ‘Automatized positions’ were used to acquire images . Room temperature and sample temperature were controlled and stabilized at 18°C and 21°C respectively . First , the time lapses were converted into sequential images . The focal plane was then selected at each time point using the function ‘Review Multi Dimensional Data’ of the software Metamorph , version 7 . 8 . 0 . 0 . The files were then exported and saved as . tiff . Image drift was corrected by the Stack Reg plugin ( Rigid Body option ) for Fiji ( Fiji version 1 . 52b , https://imagej . net/Fiji ) . Cell numbers were assigned manually . The landmark system is based on the analysis of a subset of data on male meiocytes from WT and tam plants carrying the KINGBIRD reporter constructs . A subset of the analyzable male meiocytes was described at every time point by assigning manually a value for each of the five parameters assessed . For the WT , a total of 169 meiocytes from 35 anthers were annotated , leading to a total of 18 , 531 data points spanning 3 , 269 hr . For 7860 observations one or more of the parameters could not be annotated by a well-defined state , with 5893 observations not having a single parameter recognizable . The resulting dataset , consisting of 10 , 671 time points , was used to determine the co-occurrences of parameter states and the landmarks . For tam a total of 62 meiocytes from 19 anthers were annotated , leading to a total of 10 , 224 data points spanning 1 , 694 hr . For 4127 observations one or more of the parameters could not be annotated by a well-defined state , with 3109 observations not having a single parameter recognizable . The resulting dataset , consisting of 6097 time points , was used to determine the co-occurrences of parameter states and the landmarks . For the analysis of the single reporters , the same data set derived from the observation of WT meiocytes was used . We subdivided it into two groups: the first group contained the annotations for the two parameters REC8/Chromatin and nucleolus position , while the second group included the annotations of the remaining three parameters: cell shape , nucleus position and MT array . Each group contained the annotations of the same cells at the same time points . Two additional datasets were annotated using the landmark system directly , that is assigning the cellular states A1-A11 and t1-t12 respectively and were used for the calculation of the time course of WT and tam plants . The landmark attribution for each cell at each time point was done manually . Data were recorded in the CSV format and data analysis was done using the Python programming language ( Version 3 . 6 , Python Software Foundation , https://www . python . org; https://gitlab . com/wurssb/arabidopsis-thaliana---landmark-analysis ) ( van Rosmalen et al . , 2019 ) . The manually created data set used for landmark extraction contains a description of the state of each of the five parameters ( two parameters in the case of PROREC8:REC8:GFP alone or three for PRORPS5A:TagRFP:TUB4 alone ) that were recorded for individual cells at 15 min intervals . Missing data points were labeled by ‘n’ , this was done to ensure that unmeasured periods are noted properly and to avoid assigning any unrealistic transitions . The combination of states of each of the cellular parameters makes up the cellular state . Transitions from one cellular state to another occur when one or more parameters change to a new state . To create the co-occurrence heat map in Figure 4 , we counted the number of times a combination of two parameter states occurred . Since some time lapses were measured with different temporal resolution ( e . g . 10 min intervals versus 15 min intervals ) , we first resampled the data points from all time lapses to have the same time between measurements . Co-occurrence counts were normalized by the total number of counts of the columns parameter state , including the counts where the state of the 2nd parameter could not be measured . To assess the robustness of the selected landmarks and thus our theoretical framework , we performed a bootstrapping procedure on our data set . The total set of observations was randomly sampled with replacement to obtain a data set 1 . 5 times the size of the original data set . Scores for each state in this data set were calculated using the procedure described in the previous paragraph and in the section ‘A meiotic landmark system’ of the results . This process was repeated 1 , 000 times to obtain estimates for the mean value , standard deviation and quantiles of the score of each cellular state . Results of the bootstrap can be seen in Figure 4—source data 2 and 4 . The duration of each landmark was automatically extracted from the CSV files ( Material and methods , Quantitative analysis of live cell imaging data , Data set description ) using custom software based on consecutive landmark transitions ( Seifert , 2019; copy archived at https://github . com/elifesciences-publications/LandmarkSummaryGenerator ) . This resulted in a dataset of 327 landmark durations from 136 meiocytes of 17 different anthers for WT plants , and of 245 landmark durations from 76 meiocytes of 15 different anthers for tam plants . | In plant cells , as in other cells , genetic information is stored within structures known as chromosomes . Most of the cells in a plant contain a duplicated set of chromosomes that the plant needs to survive . However , plants also produce some cells known as sex cells that only have a single set of chromosomes . This ensures that , when plants sexually reproduce , a male and female sex cell will fuse together and eventually grow into a new plant that carries a doubled set of chromosomes . Cells known as meiocytes make sex cells by dividing through a process known as meiosis . Previous studies have identified several genes that regulate meiosis in plants . For example , a gene known as TAM is required to make sex cells in a small weed known as Arabidopsis thaliana , which is often used as a model plant in research studies . During meiosis , meiocytes need to copy and move their chromosomes at precisely the right time to ensure that each sex cell they produce has a complete set of chromosomes . Studies of how chromosomes behave during meiosis in plants have so far almost exclusively relied on traditional microscopy techniques that kill the cells in the process of preparing them for imaging . Before being placed under a microscope , the dead cell material is often spread out to make it easier to see the chromosomes . These techniques provide snapshots of meiosis that provide good spatial resolution of chromosome behavior , but information about how chromosomes and other cellular components behave in the course of meiosis is lost . Prusicki et al . developed a new microscopy approach to observe meiosis in living A . thaliana cells . The experiments found that the structure of all the cells changed during meiosis in several distinct stages ( referred to as ‘landmarks’ ) . Some of these landmarks were absent or happened at a different time in mutant plant cells that lacked the TAM gene . As a result , a structure called the spindle that is required to move chromosomes during meiosis formed at the wrong time in the mutant cells . The findings of Prusicki et al . reveal new insights into the role of TAM in meiosis . The next step following on from this work is to use the same approach to study other mutant plants with defects in meiosis and analyze the effects of a changing environment on meiosis . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology",
"cell",
"biology",
"tools",
"and",
"resources"
] | 2019 | Live cell imaging of meiosis in Arabidopsis thaliana |
The integration of most membrane proteins into the cytoplasmic membrane of bacteria occurs co-translationally . The universally conserved YidC protein mediates this process either individually as a membrane protein insertase , or in concert with the SecY complex . Here , we present a structural model of YidC based on evolutionary co-variation analysis , lipid-versus-protein-exposure and molecular dynamics simulations . The model suggests a distinctive arrangement of the conserved five transmembrane domains and a helical hairpin between transmembrane segment 2 ( TM2 ) and TM3 on the cytoplasmic membrane surface . The model was used for docking into a cryo-electron microscopy reconstruction of a translating YidC-ribosome complex carrying the YidC substrate FOc . This structure reveals how a single copy of YidC interacts with the ribosome at the ribosomal tunnel exit and identifies a site for membrane protein insertion at the YidC protein-lipid interface . Together , these data suggest a mechanism for the co-translational mode of YidC-mediated membrane protein insertion .
At present , a mechanistic understanding of the function of YidC , as well as its mitochondrial and chloroplast counterparts Oxa1 and Alb3 , respectively , is limited by a lack of structural information ( Kol et al . , 2008; Dalbey et al . , 2011 ) . High resolution structures are available only for the first periplasmic domain ( P1 ) of Escherichia coli YidC ( Figure 1A; Oliver and Paetzel , 2008; Ravaud et al . , 2008 ) , however , this domain is poorly conserved , only present in Gram-negative bacteria and not essential for functionality ( Jiang et al . , 2003 ) . Furthermore , the region ( s ) of YidC mediating the interaction with the ribosome have not been identified , and the oligomeric state of YidC during co-translational translocation remains controversial ( Kohler et al . , 2009; Herrmann , 2013; Kedrov et al . , 2013 ) . Hence , we set out to determine a molecular model of ribosome-bound YidC during co-translational translocation of the substrate FOc ( van der Laan et al . , 2004 ) , an integral membrane subunit of the ATP synthase complex . 10 . 7554/eLife . 03035 . 003Figure 1 . Evolutionary covariation based structural model of E . coli YidC . ( A ) Membrane topology of YidC , with helix coloring as in all subsequent Figures . ( B ) Matrix of coupling strengths between pairs of YidC residues based on an alignment of 2366 non-redundant sequences . Helix–helix pairs with posterior probabilities higher than 57% are outlined in boxes; the 50 residue–residue pairs with highest coupling coefficients are indicated with red crosses . ( C ) Overall arrangement of TM helices viewed from the cytoplasm based on the prediction of helix–helix pairs ( black lines ) and exposure to lipid ( yellow ) or protein ( green ) . The first residue of each helix is indicated with an asterisk . ( D ) Linear representation of YidC with the seven most probable helix–helix pairs indicated by arches , with thicknesses approximating posterior probabilities . ( E and F ) Side view and cytoplasmic view , respectively , of the E . coli YidC model based on covariation analysis , with predicted residue–residue pairs indicated by yellow pseudobonds . DOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 00310 . 7554/eLife . 03035 . 004Figure 1—figure supplement 1 . Evaluation of possible helix-helix contacts . ( A ) Calibration plots for the prediction of helix–helix interactions . Calibration plots for dataset #2 ( left ) , dataset #3 ( middle ) and combined datasets #2 and #3 ( right ) . The empirical fraction of true positives is plotted depending on the uncalibrated probability ( raw score ) obtained from our method . Points correspond to empirical averages over bins of 60 predictions ( ordered by increasing uncalibrated probability ) . Lines correspond to maximum likelihood fits of the calibration plots using a transformed Bernoulli distribution with 4 parameters . ( B ) Histogram of posterior probabilities for helix–helix interactions . Distribution of predicted calibrated posterior probabilities for YidC ( TM2–TM6 ) which contains seven predicted helices , thus 21 possible helix–helix contacts . The histogram of predicted probabilities shows the specificity of the predictions: there is a large gap between 15% and 55% probability and most possible contacts have probability <15% . DOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 004
In order to build an initial structural model of YidC , we predicted contacts between pairs of residues based on covariation analysis ( Marks et al . , 2011; Hopf et al . , 2012 ) . For that purpose , we constructed a multiple sequence alignment of E . coli YidC excluding the nonconserved first transmembrane helix ( TM1 ) and the P1 domain ( Figure 1A ) and computed direct evolutionary couplings between pairs of YidC residues ( Kamisetty et al . , 2013 ) . The resulting matrix of coupling strengths ( Figure 1B ) contains several diagonal and anti-diagonal patterns of stronger coupling coefficients , which are indicative of parallel or anti-parallel helix–helix pairs , respectively . We computed probabilities for each possible helix–helix contact by aggregating the evidence of stronger coupling coefficients over the expected interaction patterns and calibrating the resulting raw scores on an independent dataset of helix–helix interactions to obtain accurate interaction probabilities . Seven helix–helix contacts attained probabilities above 57% ( Figure 1B–D ) while all other possible contacts scored below 15% , demonstrating the specificity of the method ( Figure 1—figure supplement 1B ) . We roughly positioned the five TM helices of E . coli YidC relative to each other using the predicted helix–helix contacts as constraints , and rotated them according to their predicted lipid or protein exposure ( Lai et al . , 2013; Figure 1C ) . Next , we used MODELLER ( Eswar et al . , 2008 ) to create full length models based on the TM core , secondary structure prediction and the 50 residue–residue contacts with the highest coupling coefficients ( 39 excluding intrahelical contacts , indels and topology violations ) . In the resulting model ( Figure 1E , F ) , the conserved membrane integrated core of YidC forms a helical bundle arranged like the vertices of a pentagon , in the order 4-5-3-2-6 ( clockwise ) when viewed from the cytoplasm ( Figure 1F ) . Notably , all the predicted interactions between TM domains can be explained by monomeric YidC suggesting that dimer or oligomer formation may not be strictly required for YidC activity ( see also below ) . Outside the membrane region , strong helix–helix contacts were predicted within the cytoplasmic loop between TM2 and TM3 , which can be explained the by formation of a helical hairpin ( Figure 1F ) . The base of this ‘helical paddle domain’ ( HPD ) is structurally constrained by predicted contacts with TM3 , its tip on the other hand is more mobile and appears to interact with lipid headgroups ( see below ) . While this manuscript was under review , two crystal structures were published of Bacillus halodurans YidC2 ( BhYidC2 , 34% sequence identity with E . coli YidC ) ( Kumazaki et al . , 2014 ) , providing us with a unique opportunity to directly assess the accuracy of our model . Overall , the root mean square deviation ( RMSD ) between the TM helices of our model and those of BhYidC2 is 7 . 5 Å ( 3WO6 ) and 7 . 3 Å ( 3WO7 ) ( Table 1 ) , which is within the resolution limits of our method . The global arrangement of TM helices is the same as in BhYidC2 , yet , their tilt angle relative to the plane of the membrane is slightly different ( Figure 2 ) . The tilt angle of the HPD also differs , as well as its side that faces the membrane ( Video 1 ) , which may be indicative of a high degree of flexibility of this domain , consistent with its high crystallographic B-factors ( Kumazaki et al . , 2014 ) . Notably , the HPD is not essential for YidC function in E . coli since the deletion of the entire domain is possible without compromising cell viability ( Jiang et al . , 2003 ) . 10 . 7554/eLife . 03035 . 005Table 1 . Deviations among YidC structuresDOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 005RMSD ( Å ) RMSD ( Å ) ( TM core ) 3WO63WO73 . 11 . 83WO6model9 . 47 . 53WO7model9 . 87 . 3Overall root mean square deviations ( RMSD ) between ( the TM helices of ) our model of E . coli YidC and the two BhYidC2 crystal forms . 10 . 7554/eLife . 03035 . 006Figure 2 . Covariation-based model vs homology model . Comparison of the E . coli YidC covariation-based model ( A and B ) to a homology model of E . coli YidC based on the crystal structure of BhYidC2 ( 3WO6 ) ( C and D ) . Predicted residue–residue pairs are indicated by yellow pseudobonds . Note that extracellular helix 1 ( white ) was not present in our multiple sequence alignment and is thus not included in the model . DOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 00610 . 7554/eLife . 03035 . 007Figure 2—figure supplement 1 . Local deviations among YidC structures . Smoothed Cα distances between the two BhYidC2 crystal forms ( 3WO6 vs 3WO7 , red ) , between our model of E . coli YidC and 3WO6 ( green ) and between our model and 3WO7 ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 00710 . 7554/eLife . 03035 . 008Video 1 . Conformational states of YidC . Animation showing conformational differences in YidC starting from BhYidC2 crystal form 1 ( 3WO6 ) , towards crystal form 2 ( 3WO7 ) and ending with our covariation based YidC model . Views are from within the membrane ( left ) and from the cytoplasm ( right ) . Note the movement of the HPD and the closing of the hydrophilic groove between TM3 ( orange ) and TM5 ( green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 008 A qualitative difference between our model and BhYidC2 that may have more mechanistic importance is the relative position of TM3 . In the structure of BhYidC2 a hydrophilic groove is formed on the cytoplasmic side of the TM bundle that has been proposed to form a binding site for YidC substrates ( Kumazaki et al . , 2014 ) . Interestingly , the opening state of this groove differs between the two crystal forms , that is it is more open in 3WO6 than in 3WO7 ( Video 1 ) , largely due to movement of the N-terminal half of TM3 ( Figure 2—figure supplement 1 ) . In our model on the other hand , this hydrophilic groove is even more closed than in 3WO7 because we imposed covariation-based constraints between TM3 and TM5 ( Pro425-Pro499 ) and between TM3 and TM6 ( Cys423-Gln528 & Phe433-Thr524 ) ( Figure 2; Video 1 ) . Strikingly , in BhYidC2 the distances between the Cβ atoms of these three pairs are outliers compared to other residue–residue pairs ( 20 . 5 Å/20 . 9 Å/14 . 9 Å vs an average of 8 . 2 Å , Table 2 ) . Thus , given that ( i ) the position of TM3 differs in the two crystal forms , and ( ii ) that covariation analysis predicts with high accuracy a closer interaction of TM3 with TM6 and one contact with TM5 , we conclude that movement of TM3 is a genuine feature of YidC . This movement and the accompanying dynamics of the hydrophilic groove may represent a crucial step in the functional cycle of the YidC insertase . 10 . 7554/eLife . 03035 . 009Table 2 . Top 50 scoring residue–residue pairs in covariation analysisDOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 009Residue 1# Residue 1RegionResidue 2# Residue 2Regiondmodel ( Å ) d3WO6 ( Å ) Reason for exclusionTRP354TM2–––indelGLY355TM2–––indelPHE356TM2–––indelPHE356TM2ARG533c-termtopology violationILE358TM2<–>GLY512Loop5-69 . 16 . 1ILE359TM2<–>VAL519TM66 . 55 . 2ILE359TM2<–>LEU515TM68 . 57 . 9ILE359TM2–––indelILE361TM2<–>LEU436TM37 . 98 . 2THR362TM2PRO371TM2intrahelicalPHE363TM2<–>VAL523TM65 . 26 . 1GLY367TM2<–>VAL523TM66 . 08 . 2MET369TM2<–>ILE432TM39 . 98 . 4Leu372Loop2-3PRO510Loop5-6topology violationSER379Loop2-3<–>PRO425TM310 . 29 . 9LEU386Loop2-3<–>VAL417Loop2-37 . 57 . 1LEU386Loop2-3<–>LEU411Loop2-36 . 26 . 1PRO388Loop2-3GLN429TM3topology violationLYS389Loop2-3<–>ALA414Loop2-310 . 59 . 8LYS389Loop2-3<–>GLU415Loop2-311 . 210 . 0ILE390Loop2-3<–>MET408Loop2-36 . 86 . 2MET393Loop2-3<–>ILE404Loop2-37 . 97 . 4MET393Loop2-3<–>LEU411Loop2-38 . 27 . 7ARG394Loop2-3<–>ILE404Loop2-38 . 58 . 1ARG396Loop2-3<–>GLU407Loop2-38 . 98 . 4CYS423TM3<–>GLN528TM616 . 220 . 9PRO425TM3<–>PRO499TM510 . 220 . 5PHE433TM3<–>THR524TM611 . 014 . 9LEU436TM3<–>GLY512Loop5-67 . 68 . 3TYR437TM3<–>LEU513Loop5-69 . 86 . 4TRP454Loop3-4<–>ASP462Loop3-46 . 67 . 0TRP454Loop3-4<–>PRO468TM416 . 011 . 5TRP454Loop3-4<–>SER511Loop5-69 . 88 . 3ILE455Loop3-4<–>LEU467TM49 . 810 . 1ILE455Loop3-4<–>ILE466TM411 . 08 . 0ASP462Loop3-4<–>PRO468TM412 . 56 . 8ASP462Loop3-4<–>SER511Loop5-611 . 14 . 2TYR465TM4<–>LEU507TM510 . 48 . 7LEU467TM4<–>LEU515TM611 . 66 . 6PRO468TM4<–>LEU513TM614 . 58 . 8LEU470TM4<–>ILE518TM66 . 35 . 4MET471TM4<–>PHE502TM58 . 84 . 9GLY472TM4<–>THR503TM56 . 75 . 3GLY472TM4GLN479TM4intrahelicalTHR474TM4<–>ASN521TM64 . 73 . 7THR474TM4<–>ILE525TM66 . 77 . 8ILE478TM4<–>ILE525TM69 . 05 . 0THR485Loop4-5–––indelPHE506TM5<–>VAL514TM614 . 44 . 2GLY512Loop5-6GLN532TM6topology violationØ9 . 38 . 1Table showing the 50 residue–residue pairs with the highest covariation scores , and the distances between the Cβ atoms in the final model of the 39 pairs that were used as constraints for model building . For comparison , the corresponding distances in 3WO6 are also given . The 11 residue–residue pairs that were excluded for model building are in italics , with the reason for their exclusion indicated on the right . In summary , the overall structure of our YidC model agrees well with the BhYidC2 crystal structure , and a comparison of both structures reveals dynamic regions in YidC that may be of mechanistic importance . This further illustrates the power of covariation analysis not merely for structure prediction but also for obtaining dynamic insights ( Hopf et al . , 2012 ) . Next , in order to further characterize and validate our obtained YidC model , we assessed its stability and biochemical properties in the bacterial membrane by employing traditional molecular dynamics ( MD ) simulations . Overall , the model was found to be very stable during the simulation . While the five TM helices enable a rigid protein core , the polar loop regions tend to swim on the membrane surface ( Figure 3A ) . An analysis of inter-residue interactions within the TM region ( Figure 3B ) provides a firm basis to the observed stability of YidC: hydrophobic residues on the exterior of the TM bundle stabilize interactions with the apolar lipid tails . The YidC core , in turn , is stabilized both via short and long-range interactions between the five helices . Residues towards the cytoplasmic side of the core are primarily polar or charged and , therefore , engaged in strong electrostatic or charge–dipole interactions . In contrast , residues on the periplasmic side are primarily aromatic and involved in stacking and other nonpolar dispersion interactions . 10 . 7554/eLife . 03035 . 010Figure 3 . Molecular dynamics simulation of the YidC model . ( A ) Side view ( left ) and cytoplasmic view ( right ) of the stable YidC model after a 500 ns MD simulation in a lipid bilayer composed of 3:1 POPE:POPG . ( B ) Ribbon representation of the stable model according to inter-helix energy ( in kcal/mol ) , blue: −7 . 5 to −1; white: −1 to −0 . 002; red: ≥ −0 . 00 . 2 . Residues that inactivate YidC upon mutagenesis are indicated by spheres . ( C ) Ribbon representation of the stable model according to flexibility ( in Å2 ) , blue: 0 . 04 to 0 . 09; white: 0 . 09–1; red: ≥1 . 0 . ( D ) In vivo complementation assay of YidC mutants T362A ( TM2 ) and Y517A ( TM6 ) . ( E ) Thickness of the cytoplasmic and periplasmic leaflet of the simulated bilayer after 500 ns , highlighting the membrane thinning effect in the vicinity of YidC . The membrane surface is defined by positions of polar head groups in the lipids , and thickness at a given point on the surface is taken to be the shortest distance between the head groups from opposite leaflets . The thickness values are averaged over the MD trajectory and presented as a contour plot on the membrane surface with a color-scale from red , indicating thicker region representing bulk bilayer lipids , to blue showing thinned regions close to YidC suggesting hydrophobic mismatch . ( F ) Distribution of hydrophobic ( red ) and hydrophilic residues ( blue ) in YidC at various heights of the membrane , highlighting the hydrophilic environment in the center of YidC on the cytoplasmic side . DOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 01010 . 7554/eLife . 03035 . 011Figure 3—figure supplement 1 . Complementation of MD-based mutants . In vivo complementation assay of YidC mutants identified as structurally important by MD simulations . Positions in YidC that were also identified by covariation analyses are indicated in the right column . DOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 01110 . 7554/eLife . 03035 . 012Figure 3—figure supplement 2 . Expression of MD-based mutants . Western blot of whole FTL10 cells grown on arabinose or glucose , showing the stable expression of inactive YidC mutants that were identifed by MD simulations . DOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 012 In order to verify the functional relevance of residues suggested by the MD simulations , we created alanine mutants and subjected them to an in vivo complementation assay . Some of the most stabilizing residues , T362 in TM2 and Y517 in TM6 , both of which are located at the same height in the membrane , completely inactivated YidC when mutated to alanine ( Figure 3D , Figure 3—figure supplement 1 ) . Both mutants were stably expressed , indicating that the lack of complementation was not caused by instability of YidC ( Figure 3—figure supplement 2 ) . Several residues close to this pair show intermediate activity levels ( F433 , M471 and F505 ) , whereas residues further away do not show an effect ( Figure 3—figure supplement 1 ) . Taken together , we provide a model for the overall arrangement of the conserved domains of YidC that is in good agreement with our covariation analysis , lipid exposure prediction , MD simulation , in vivo complementation analysis as well as the recent crystal structures . Interestingly , we observed that YidC induces thinning of the lipid bilayer during the MD simulation . A significant thinning of 7–10 Å results from the hydrophobic mismatch between the TM helices and the membrane ( Figure 3E ) . The thinning is similar in the upper and lower leaflet , and the thinnest region is in proximity of TM3 and TM5 . Since membrane inserting YidC substrates have been chemically cross-linked to both these helices ( Klenner et al . , 2008; Yu et al . , 2008; Klenner and Kuhn , 2012 ) , we argue that thinning of this region in particular may be relevant for the molecular mechanism of YidC-dependent membrane insertion . In addition , the distribution of hydrophilic and hydrophobic residues within YidC revealed the presence of a hydrophilic environment on the cytoplasmic side of the YidC TM bundle ( Figure 3F ) , which continues into the mentioned hydrophobic cluster of aromatic residues towards the periplasmic side . It is tempting to speculate that this hydrophilic environment may receive the polar termini and loops of YidC substrates during the initiation of translocation , thus facilitating their transfer across the hydrophobic core of the ( thinned ) lipid bilayer ( see below ) . Notably , essentially the same conclusions have been drawn on the basis of the BhYidC2 crystal structures and accompanying cross-linking studies ( Kumazaki et al . , 2014 ) . In order to provide a molecular model of YidC in its active state , we reconstituted purified full length YidC ( extended with the C-terminus of R . baltica YidC [Seitl et al . , 2014] ) with ribosome nascent chains ( RNCs ) exposing the first TM helix of FOc , and subjected the complex to cryo-EM and single particle analysis to a resolution of ∼8 Å ( Figure 4A , B ) . In agreement with previous structural studies ( Kohler et al . , 2009; Seitl et al . , 2014 ) , YidC binds to the ribosomal exit site , however , the improved resolution now allows for a more detailed interpretation . Firstly , we were able to separate the weaker electron density of the detergent micelle from that of YidC ( Figure 4A ) . Secondly , the presence of elongated structural features ( Figure 4D–F ) allowed us to dock our molecular model in a distinct orientation ( cross correlation coefficient 0 . 865 ) . Following placement of the YidC-core model , two prominent densities in the membrane region , one next to TM3 and one next to TM5 , remained unaccounted for . These could be attributed to either TM1 of YidC or to the TM helix of the nascent chain ( NC ) FOc . Given that ( i ) YidC substrates are known to crosslink to TM3 ( Klenner et al . , 2008; Yu et al . , 2008; Klenner and Kuhn , 2012 ) , and ( ii ) that the density neighboring TM3 is aligned with the ribosomal exit tunnel and ( iii ) that at the same relative position nascent chains have been observed inside the SecY channel ( Frauenfeld et al . , 2011 ) ( Figure 4—figure supplement 1 ) , the most plausible assignment to the density near TM3 appeared to be the TM helix of FOc . To verify this , and to exclude that the density neighboring TM5 corresponds to the nascent chain , we reconstituted single cysteine mutants of YidC either in TM3 ( M430C and P431C ) or in TM5 ( V500C and T503C ) with RNCs of a single cysteine mutant of FOc ( G23C ) , and exposed them to disulphide crosslinking . Upon exposure to the oxidator 5 , 5′-dithiobis- ( 2-nitrobenzoicacid ) ( DTNB ) , only in the TM3 mutants a DTT-sensitive ∼90 kDa product appeared that reacted with antibodies against the nascent chain ( NC-tRNA∼30 kDa , Figure 4C ) as well as YidC ( ∼60 kDa , Figure 4C ) . Thus , the adduct represented indeed the inserting FOc TM domain crosslinked to TM3 of YidC . RNCs lacking a cysteine in the nascent chain ( Figure 4—figure supplement 2 ) or YidC mutants with cysteines in TM5 did not yield any crosslinks ( Figure 4C ) . Hence , we conclude that the unaccounted electron density next to TM3 represents the TM of the nascent chain , and that the density neighboring TM5 represents TM1 of YidC ( Figure 4D–F ) . 10 . 7554/eLife . 03035 . 013Figure 4 . Cryo-EM structure of RNC bound YidC and structural model of the active state . ( A ) Side view of the ∼8 Å resolution cryo-EM based electron density of the RNC:YidC complex , with the small subunit depicted in yellow , the large subunit in gray , P-site tRNA and nascent chain in green , YidC in red and the detergent micelle in blue . ( B ) As in A , but sliced through the ribosomal exit tunnel . ( C ) Validation of the active state model by disulphide crosslinking . RNCs carrying the mutant FOc ( G23C ) were reconstituted with the indicated single cysteine YidC mutants , oxidized , applied to a linear sucrose gradient and harvested from the 70S peak before SDS-PAGE and western blotting . Immunodetection was performed with antibodies raised against the HA-tag ( located in the nascent chain inside the ribosomal exit tunnel ) and anti-YidC antibodies . YidC , nascent chain-tRNA ( NC-tRNA ) and the expected crosslink product ( NC-tRNA x YidC ) are indicated . ( D–F ) Structural model of YidC during membrane protein insertion , viewed from two sides within the membrane ( D and E ) and from the cytoplasm ( F ) . The detergent micelle was removed for clarity , the TM helix of FOc is depicted in magenta , and the disulphide crosslink between YidC and FOc with -SS- . DOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 01310 . 7554/eLife . 03035 . 014Figure 4—figure supplement 1 . Comparison of the active states of YidC and SecY . Left: molecular model of YidC during co-translational translocation of FOc , and the contour of active SecY . Middle: composite model of active YidC with FOc replaced by the hydrophilic part of nascent FtsQ as found in active SecY . Right: molecular model of SecY during co-translational translocation of FtsQ . For clarity , the N-terminal signal anchor of FtsQ was omitted . DOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 01410 . 7554/eLife . 03035 . 015Figure 4—figure supplement 2 . Negative control for RNC-YidC crosslinking . Crosslinking was performed with a cysteine-less FOc RNC as described in the legend to Figure 3C . A poorly reproducible unknown product is indicated with an asterisk . DOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 015 We attribute the remaining unaccounted electron density in the periplasmic region to the P1 domain; however , because it is substantially smaller than the crystal structure of P1 , we did not include it in our molecular model . Flexibility relative to the conserved membrane region of YidC is the most likely explanation for this finding . We did not observe density for the HPD , in agreement with its flexibility observed in both , the crystal structures of BhYidC2 and the MD simulations ( Figure 3C ) . In order to validate our molecular model of co-translationally active YidC , we mutated residues that would be in direct contact with the ribosome ( Figure 5A , B ) and analyzed their effect on functionality in the in vivo complementation test . Indeed , mutation of residues Y370A and Y377A ( contacting ribosomal RNA helix 59 ) and D488K ( contacting ribosomal protein uL23 ) severely interfere with YidC activity ( Figure 5C , Figure 5—figure supplement 1 ) thereby emphasizing their functional significance . All these mutants were stably expressed , indicating that the lack of complementation was not caused by instability of YidC ( Figure 5—figure supplement 2 ) . Given that YidC in general is known to be very tolerant to point mutations ( Jiang et al . , 2003 ) , this provides further support for the overall correctness of our model of ribosome-bound YidC during membrane protein insertion . 10 . 7554/eLife . 03035 . 016Figure 5 . Contacts between active YidC and the ribosome . ( A and B ) Close-up views from within the membrane region highlighting the contact between H59 of the ribosome and the 2/3 loop of YidC ( A ) and ribosomal protein uL23 and the 4/5 loop of YidC ( B ) . Residues that inactivate YidC upon mutagenesis or deletion are indicated by magenta spheres . ( C ) In vivo complementation assay of YidC point mutants D488A , D488K , deletion mutant Δ487-489 and the double mutants Y370A/Y377A and Y370F/Y377F . ( D ) Periplasmic view of the active ribosome-bound YidC model , with the YidC contour outlined in red . The polypeptide exit tunnel is indicated with an asterisk . ( E ) Cartoon based comparison of active SecY ( left ) and active YidC ( right ) during membrane insertion of FtsQ and FOc , respectively . The ribosome is depicted in gray , the aqueous channel in SecY as well as the hydrophilic environment within YidC are shaded blue , hydrophobic TM domains of the substrates are depicted magenta , hydrophilic parts in green and the P1 domain by a dashed oval . DOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 01610 . 7554/eLife . 03035 . 017Figure 5—figure supplement 1 . Complementation of ribosome interaction mutants . In vivo complementation assay of YidC mutants involved in ribosome binding . DOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 01710 . 7554/eLife . 03035 . 018Figure 5—figure supplement 2 . Expression of ribosome interaction mutants . Western blot of whole FTL10 cells grown on arabinose or glucose , showing the stable expression of inactive YidC mutants that interact with the ribosome . DOI: http://dx . doi . org/10 . 7554/eLife . 03035 . 018
Finally , it is notable that we observe only a single monomer of YidC bound to the active ribosome . This is in agreement with recent literature showing clearly that both YidC ( Herrmann , 2013; Kedrov et al . , 2013; Seitl et al . , 2014 ) and the SecY complex ( Frauenfeld et al . , 2011; Park and Rapoport , 2012; Taufik et al . , 2013; Park et al . , 2014 ) can be fully active as monomers . However , the comparison of models for active YidC and active SecY ( Figure 5E , Figure 4—figure supplement 1 ) reveals an important difference between the two proteins that has mechanistic implications . While SecY is known to translocate hydrophilic nascent chains through its central aqueous channel ( Cannon et al . , 2005; Rapoport , 2007; Driessen and Nouwen , 2008 ) and insert TM domains through a lateral gate ( Van den Berg et al . , 2004; Gogala et al . , 2014 ) , our model suggests that the YidC substrates are inserted at the protein-lipid interface . Two principal findings of our work suggest how YidC may facilitate this process: ( i ) it provides a hydrophilic environment within the membrane core for receiving the hydrophilic moieties ( termini or loops ) of a substrate , and ( ii ) it reduces the thickness of the lipid bilayer: initial interaction of the hydrophilic moieties of YidC substrates with the hydrophilic environment of YidC would allow for a partial insertion into the membrane , while facilitating exposure of the hydrophobic TM domains to the hydrophobic core of the bilayer . The latter in turn may compensate for the energetic penalty of driving the hydrophilic moieties across the ( already thinned ) bilayer . Further biochemical and structural studies that capture the earlier stages of this translocation process are eagerly awaited to fully elucidate this mechanism .
We constructed a multiple sequence alignment of YidC excluding the unconserved first transmembrane helix ( TM1 ) and the periplasmic P1 domain . We searched for homologous sequences of E . coli YidC starting from the PFAM seed alignment of family PF02096 ( Punta et al . , 2012 ) and using the sensitive homology detection software HHblits ( Remmert et al . , 2012 ) . First , five iterations of HHblits were run against the clustered Uniprot database with no filtering , to retrieve as many homologous sequences as possible . Then , we post-processed the alignment using HHfilter to generate a non-redundant alignment at 90% sequence identity . This resulted in an alignment containing 2366 sequences aligned across YidC helices TM2-TM6 . Using this multiple sequence alignment , we computed direct evolutionary couplings between pairs of YidC residues using the method of Kamisetty et al . ( 2013 ) . To compute probabilities for each possible helix–helix contact , we aggregated the evidence of stronger coupling coefficients over the expected interaction patterns for helix–helix contacts , taking into account the expected periodicity of ∼3 . 5 residues per alpha helix turn . We built three non-redundant datasets of mainly-alpha proteins from the CATH database ( Sillitoe et al . , 2013 ) . For each protein , we slid a square pattern ( of size 17 × 17 residues = 289 cells ) over the matrix of coupling strengths . For each pattern position , we used Bayes theorem to calculate the raw probability for a helix–helix interaction , given the 289 coupling strengths . The distributions of coupling strengths for interacting and non-interacting helix residues were fitted on dataset #1 ( 1118 proteins ) . We assigned different weights to the pattern cells , depending on their position within the pattern and the direction of the helix–helix interaction ( parallel or antiparallel ) ; these weights were optimized on dataset #2 ( 204 proteins ) . Finally , we calibrated the resulting raw scores on dataset #3 ( 85 proteins ) to obtain accurate interaction probabilities . For cross-validation purposes , we also performed optimization on dataset #3 and calibration on dataset #2 . Optimization on either dataset #2 or dataset #3 results in the same choice of weights for the pattern cells . The final posterior probabilities were obtained as the average of the values calibrated on datasets #2 and #3 , weighted by dataset size . The calibration plots for datasets #2 and #3 are shown in Figure 1—figure supplement 1A . The histogram of final posterior probabilities obtained for YidC is shown in Figure 1—figure supplement 1B , which illustrates the specificity of the helix–helix predictions . The conserved TM helices of E . coli YidC were positioned according to the covariation based helix–helix contact prediction , and rotated based on their predicted lipid or protein exposure ( Lai et al . , 2013 ) , resulting in a starting model of the conserved TM core of YidC . Additional information based on direct residue–residue interactions ( covariance analysis ) and secondary structure predictions by Jpred 3 ( Cole et al . , 2008 ) were used as structural restraints in MODELLER ( Eswar et al . , 2008 ) . From a total of 10 output models that differed mainly in the relative orientation of the loop regions , the model that satisfied the imposed constraints best was used for further studies . RNC constructs encoding residues 1–46 of FOc ( preceded by an N-terminal His-tag and 3C rhinoprotease cleavage site , and followed by an HA-tag and TnaC stalling sequence ) were cloned into a pBAD vector ( Invitrogen , Life Technologies , Karlsruhe , Germany ) by standard molecular biology techniques , and expressed and purified as described before ( Bischoff et al . , 2014 ) . Briefly , E . coli KC6 ΔsmpBΔssrA ( Seidelt et al . , 2009 ) carrying the plasmid for FOc was grown in LB with 100 µg/ml ampicilin at 37°C to an OD600 = 0 . 5 and expression was induced for 1 hr by adding 0 . 2% arabinose . Cells were lysed and debris was removed by centrifugation for 20 min at 16 . 000 rpm in a SS34-rotor ( Sorvall ) . The cleared lysate was spun overnight through a sucrose cushion at 45 . 000 rpm in a Ti45 rotor ( Beckmann ) , the ribosomal pellet was resuspended for 1 hr at 4°C and RNCs were purified in batch by affinity purification using Talon ( Clontech ) . After washing the Talon beads with high salt buffer the RNCs were eluted and loaded onto a linear 10%–40% sucrose gradient . The 70S peak was collected , RNCs were concentrated by pelleting , resuspended in an appropriate volume of RNC Buffer ( 20 mM HEPES pH 7 . 2 , 100 mM KOAc , 6 mM MgOAc2 , 0 . 05% ( wt/vol ) dodecyl maltoside ) , flash frozen in liquid N2 and stored at −80°C . The complete sequence of the nascent chain is: MGHHHHHHHHDYDIPTTLEVLFQGPGTMENLNMDLLYMAAAVMMGLAAIGAAIGIGILGGKFLEGAARQPDLIYPYDVPDYAGPNILHISVTSKWFNIDNKIVDHRP . For purification and reconstitution studies , E . coli YidC extended with the C-terminus from R . baltica ( Seitl et al . , 2014 ) was re-cloned into pET-16 ( Novagen ) with an N-terminal His-tag followed by a 3C rhinovirus protease site . Expression and purification was performed essentially as described ( Lotz et al . , 2008 ) . Briefly , E . coli C43 ( DE3 ) cells ( Miroux and Walker , 1996 ) harboring the YidC construct were grown at 37°C to an OD600 = 0 . 6 and expression was induced by adding 0 . 5 mM IPTG . YidC was solubilized with Cymal-6 ( Anatrace ) and purified by affinity chromatography using TALON ( Clontech ) . The N-terminal His-tag of the eluted protein was cleaved off with 3C protease during overnight dialysis at 4°C , followed by gel filtration chromatography ( Superdex 200; GE Healthcare ) . Fractions of the monodisperse peak were pooled , concentrated to ∼1 mg/ml in YidC Buffer ( 20 mM NaPO4 pH 6 . 8 , 100 mM KOAc , 10% glycerol , 0 . 05% Cymal-6 ) and directly used for further structural or biochemical assays . For disulphide crosslink analysis , FOc ( G23C ) -RNCs and single cysteine mutants of YidC were purified separately and reconstituted by incubating 100 pmol of RNCs with 500 pmol of freshly purified YidC for 30 min at 37°C . The endogenous cysteine in YidC at position 423 was replaced by serine . Disulphide crosslinking was induced by adding 1 mM 5 , 5′-dithiobis- ( 2-nitrobenzoicacid ) ( DTNB ) for 10 min at 4°C and quenched by adding 20 mM N-Ethylmaleimide ( NEM ) for 20 min at 4°C . Crosslinked RNC-YidC complexes were separated from non-crosslinked YidC using a 10%–40% linear sucrose gradient , and the 70S peak was harvested and analyzed by SDS-PAGE followed by western blotting . For in vivo complementation studies , wildtype E . coli YidC was recloned into pTrc99a ( Pharmacia ) , and mutants were created by standard molecular cloning techniques . E . coli FTL10 cells ( Hatzixanthis et al . , 2003 ) harboring pTrc99a plasmids encoding the YidC variants were grown overnight at 37°C in LB medium supplemented with 100 µg/ml ampiciline , 50 µg/ml kanamycin and 0 . 2% arabinose . YidC depletion was carried out by transferring the cells to LB medium supplemented with 100 µg/ml ampiciline , 50 µg/ml kanamycin and 0 . 2% glucose , followed by and additional incubation for 3 hr at 37°C . Cell suspensions of all constructs were adjusted to OD600 = 0 . 1 and either loaded onto SDS-PAGE gels for subsequent Western blotting , or further diluted to OD600 = 10−5 . Each dilution was spotted on LB agar plates supplemented 100 µg/ml ampiciline , 50 µg/ml kanamycin and either 0 . 2% arabinose or 0 . 2% glucose , and incubated overnight at 37°C . For cryo-EM analysis , FOc-RNC:YidC complexes were reconstituted by incubating 10 pmol of RNCs with 100 pmol of freshly purified YidC for 30 min at 37°C in a final volume of 50 µl of RNC buffer . Samples were applied to carbon-coated holey grids according to standard methods ( Wagenknecht et al . , 1988 ) . Micrographs were collected under low-dose conditions on a FEI TITAN KRIOS operating at 200 kV using a 4 k × 4 k TemCam-F416 CMOS camera and a final pixel size of 1 . 035 Å on the object scale . Image processing was done using the SPIDER software package ( Shaikh et al . , 2008 ) . The defocus was determined using the TF ED command in SPIDER followed by automated particle picking using Signature ( Chen and Grigorieff , 2007 ) . The machine-learning algorithm MAPPOS ( Norousi et al . , 2013 ) was used to subtract ‘false positive’ particles from the data set and initial alignment was performed using an empty 70S ribosome as reference . The complete data set ( 876376 particles ) was sorted using competitive projection matching in SPIDER followed by focused sorting for ligand density ( Leidig et al . , 2013 ) , and refined to a final resolution of ∼8 . 0 Å ( Fourier shell correlation [FSC] cut-off 0 . 5 ) . The final dataset consisted of 58 , 960 particles showing electron density for P-site tRNA and ligand density at the tunnel exit . We have deposited our cryo-EM map at the EMDB under accession number 2705 , and the model of the transmembrane domains at the PDB under accession number 4utq . | Cells are surrounded by a plasma membrane that acts like a barrier to help to keep the cell intact . Proteins are embedded in this plasma membrane; and some of these membrane proteins act as channels that allow molecules to enter and leave the cell , while others allow the cell to communicate with its surroundings . Like all proteins , membrane proteins are chains of amino acids that are joined together by a molecular machine called a ribosome . Most membrane proteins are inserted into the membrane as they are being built . All bacteria contain a protein called YidC that inserts proteins into the plasma membrane of bacterial cells . However , the mechanism behind this activity and the parts of the YidC protein that interact with the ribosome and plasma membrane are unknown . Wickles et al . have now used data from a range of sources to predict the three-dimensional structure of the YidC protein taken from a bacterium called E . coli . The model shows how the YidC protein is threaded back-and-forth through the membrane , a total of five times . Some of the protein also extends into the inside of the bacterial cell . Wickles et al . then used a technique called cyro-electron microscopy to look at the structure of a YidC protein bound to a ribosome that is building a new protein . Fitting the more detailed model of YidC into this overall structure of the whole complex revealed how a single YidC protein might interact with the ribosome to insert a newly built protein into a membrane . Wickles et al . then used a combination of theoretical modeling and other experiments to identify the amino acids in the YidC protein that bind to the ribosome: as expected , the binding takes place where the newly formed protein chain exits the ribosome . Further experiments also identified the amino acids in the YidC protein that interact with the newly built membrane protein , thus revealing where it might leave the YidC protein and be inserted into the membrane . The next challenge will be to investigate how the YidC protein assists the folding of new membrane proteins into their own highly specific three-dimensional structure . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | A structural model of the active ribosome-bound membrane protein insertase YidC |
Catecholamines modulate the impact of motivational cues on action . Such motivational biases have been proposed to reflect cue-based , ‘Pavlovian’ effects . Here , we assess whether motivational biases may also arise from asymmetrical instrumental learning of active and passive responses following reward and punishment outcomes . We present a novel paradigm , allowing us to disentangle the impact of reward and punishment on instrumental learning from Pavlovian response biasing . Computational analyses showed that motivational biases reflect both Pavlovian and instrumental effects: reward and punishment cues promoted generalized ( in ) action in a Pavlovian manner , whereas outcomes enhanced instrumental ( un ) learning of chosen actions . These cue- and outcome-based biases were altered independently by the catecholamine enhancer melthylphenidate . Methylphenidate’s effect varied across individuals with a putative proxy of baseline dopamine synthesis capacity , working memory span . Our study uncovers two distinct mechanisms by which motivation impacts behaviour , and helps refine current models of catecholaminergic modulation of motivated action .
Catecholamine ( i . e . dopamine and noradrenaline ) transmission has long been implicated in key aspects of adaptive behaviour , including learning , action , and motivation . Deficits in these aspects of adaptive behaviour are observed in a wide range of neuropsychiatric disorders , such as attention deficit hyperactivity disorder , Parkinson’s disease , and addiction ( Dagher and Robbins , 2009; Prince , 2008; Skolnick , 2005 ) , and many of those deficits can be treated with catecholaminergic drugs ( Faraone and Buitelaar , 2010; Wigal et al . , 2011 ) . While overwhelming evidence implicates catecholamines in both motivated activation and motivated learning of behaviour ( Bromberg-Martin et al . , 2010; Robbins and Everitt , 1996; Wise , 2004 ) , their respective contributions are still highly debated . In this study , we dissect the contribution of catecholamines to motivational biases in behavioural activation and learning . The neuromodulator dopamine has been linked particularly strongly to behavioural activation in the context of reward ( Taylor and Robbins , 1986; 1984 ) , putatively by amplifying the perceived benefits of action over their costs ( Collins and Frank , 2014; Niv et al . , 2007 ) . This behavioural activation to reward-predicting cues is likely to be , at least partly , Pavlovian in nature , with the conditioned cues eliciting innately specified responses ( Figure 1A ) . The Pavlovian nature of these motivational biases has been demonstrated using Pavlovian-instrumental transfer ( PIT ) paradigms ( Estes and Skinner , 1941; Estes , 1943 ) . In PIT , conditioned cues elicit innately specified responses that may potentiate ( or interfere with ) instrumental responding , e . g . appetitive cues promote active responding ( appetitive PIT ) , whereas aversive cues increase behavioural inhibition ( aversive PIT; Davis and Wright , 1979; Huys et al . , 2011 ) . Enhanced dopamine increases appetitive PIT ( Wyvell and Berridge , 2000 ) , while appetitive PIT is lowered when striatal dopamine is reduced ( Dickinson et al . , 2000; Hebart and Gläscher , 2015; Lex and Hauber , 2008 ) . Striatal dopamine has also been linked to controlling aversively motivated behaviour ( Faure et al . , 2008; Lloyd and Dayan , 2016 ) . Together , these results show that appetitive cues promote activation and aversive cues promote inhibition in a Pavlovian manner , mediated by the dopamine system . 10 . 7554/eLife . 22169 . 003Figure 1 . Distinct mechanisms by which motivational valence may bias behavioural activation . ( A ) Pavlovian response bias: appetitive cues ( green edge ) elicit generalized behavioural activation ( ‘Go’ ) , whereas aversive cues ( red edge ) elicit behavioural inhibition ( ‘NoGo’ ) . This Pavlovian response bias is introduced in model M3a as the parameter π ( c . f . Figure 3 ) . ( B ) Instrumental learning bias: rewarding outcomes ( upper panel ) facilitate learning of action ( ‘Go’ , thick arrow ) relative to inaction ( ‘NoGo’ , thin arrow ) . Thus , learning effects at the individual trials t will result in a cumulative selective increase of the rewarded action on later trials tn . Punishment outcomes ( lower panel ) hamper the unlearning of inaction ( ‘NoGo’ , dashed arrow ) relative to action ( ‘Go’ , solid arrow ) , resulting in sustained inaction . Neutral outcomes are equally well associated with actions and inactions , and are not illustrated here . The instrumental learning bias is introduced as the parameter κ in model M3b ( c . f . Figure 3 ) . We assess whether these two mechanisms ( i ) act in parallel , and ( ii ) are modulated by the catecholamine system . To test the latter , we administered methylphenidate ( MPH ) , which prolongs the effects of catecholamine release via blockade of the catecholamine receptors . We first assess whether MPH affects the strength of the Pavlovian response bias , introduced as the parameter πMPH in model M5a , and instrumental learning bias , implemented as the parameter κMPH-selective in model M5b ( c . f . Figure 5 ) . ( C ) We hypothesise that prolonged effects of dopamine release following reward outcomes will reduce ( temporal ) specificity , leading to spread of credit: Credit is assigned to other recent actions ( thin arrow ) , in addition to the performed ( and rewarded ) Go response ( thick arrow ) , resulting in additional learning of the alternative ( not-performed ) Go response . This MPH-induced diffuse learning bias is implemented by the parameter κMPH-diffuse in model M5c ( c . f . Figure 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 003 While Pavlovian response biases can be helpful in reducing computational load by shaping our actions in a hardwired manner , they are inherently limited because of their general nature ( Dayan et al . , 2006 ) . In order to select the best action in a specific environment , instrumental systems allow organisms to adaptively learn action-outcome contingencies , by assigning value to actions that in the past have led to good outcomes , while reducing value of actions that led to negative outcomes ( Dickinson and Balleine , 1994; Rescorla and Wagner , 1972; Robbins and Everitt , 2007 ) . Pavlovian and instrumental learning are often presented as a dichotomy , whereby cue-based , Pavlovian effects are solely responsible for motivational biases , while adaptive ‘rational’ choice results from instrumental learning . For example , multiple recent studies showing that reward or punishment cues bias action , eliciting appetitive activation and/or aversive inhibition , have been interpreted specifically in terms of a Pavlovian response bias ( for a review see Guitart-Masip et al . , 2014a ) . We hypothesised that these motivational biases of behavioural activation may also arise from asymmetrical , or biased , instrumental learning ( Figure 1B ) , in addition to Pavlovian response biases . Such biases in learning , like response biases , may reflect predominant statistics of the environment . For example , we might be quicker to believe that an action led to a reward , because actions often cause rewards . However , we may not attribute a punishment to having held back , because holding back usually helps avoid a punishment . Such an instrumental learning bias may arise from a circuitry where reinforcers are more potent at driving learning following active ‘Go’ than inactive ‘NoGo’ actions . This means that Go responses ( relative to NoGo responses ) are easier to learn and unlearn following reward and punishment respectively . This instrumental learning bias would predict that Go responses that elicited a reward are more likely to be repeated ( i . e . better learned ) than NoGo responses that elicited a reward . Similarly , Go responses that elicited a punishment are relatively less likely to be repeated ( i . e . better unlearned ) than NoGo responses that elicited a punishment . These instrumental learning and Pavlovian response biasing accounts of motivated ( in ) action could not be dissociated in earlier studies ( Cavanagh et al . , 2013; Guitart-Masip et al . , 2014b; 2012 ) , because they allowed for only a single Go response: With only one response option , general activation of action cannot be disentangled from facilitated learning of a specific response . In our proposed framework , motivational biases in behavioural ( in ) activation are likely the result of a combination of Pavlovian response biasing plus an asymmetry in instrumental learning of Go and NoGo responses ( Figure 1 ) . At the neurobiological level , this hypothesis arises from current theorizing about the mechanism of action of reinforcement-related changes in dopamine . Specifically , a potential substrate for this proposed learning asymmetry could be provided by the striatal dopamine system , which is notably involved in instrumental learning via modulation of synaptic plasticity ( Collins and Frank , 2014 for review and models ) . In this framework , dopamine bursts elicited by better than expected outcomes reinforce the actions that led to these outcomes ( Montague et al . , 2004; Schultz et al . , 1998; Schultz et al . , 1997 ) via long-term potentiation ( LTP ) in the striatal direct ‘Go’ pathway ( Frank et al . , 2004 ) . The temporal specificity of the phasic dopamine bursts allows for assigning credit to the most recent action , by potentiating the recently active striatal neurons . Due to the LTP in the ‘Go’ pathway , rewards may be more effective in reinforcing neurons coding for active Go responses than NoGo responses . Conversely , dopamine dips elicited by worse-than-expected outcomes ( Matsumoto and Hikosaka , 2009; Tobler et al . , 2005 ) lead to long-term depression ( LTD ) of the ‘Go’ pathway and LTP in the ‘NoGo’ pathway , making it less likely that the same cue would elicit an active than inactive response next time . In short , the striatal system is biased to attribute rewards and punishments to active Go responses , which ecologically may be more commonly the cause of observed outcomes . The implication of this is that is easier to learn to take action based on reward , but easier to withhold making an action based on punishment . A key additional prediction of this model is that prolonging the presence of dopamine , e . g . by blocking dopamine reuptake with methylphenidate , would lead to a spread of credit assignment ( Figure 1C ) . Here , credit is assigned to striatal neurons that were recently active , due to recent actions that did not actually lead to the current reward and phasic dopamine burst ( ‘spread of effect’; Thorndike , 1933 ) . In this framework , the dopamine system can produce biased motivated behaviour due to ( i ) direct Pavlovian biases ( e . g . predicted rewards potentiate the Go pathway during action selection ) , and ( ii ) disproportionate potentiation of instrumental learning of Go actions that ( recently ) led to reward . Put more simply , ( i ) dopamine bursts prompted by reward-predicting cues can potentiate activation of the Go pathway , giving rise to the cue-based , Pavlovian activation , and ( ii ) dopamine bursts prompted by reward outcomes can potentiate plasticity within the Go pathway , making rewards more effective in reinforcing Go responses than NoGo responses . In this study , we aimed to assess whether biases in instrumental learning contribute to the pharmaco-computational mechanisms subserving well-established reward/punishment biases of motivated ( in ) action . To dissociate biased instrumental learning from Pavlovian response biases , we developed a novel experimental paradigm including multiple active response options ( Figure 2 ) , and combined this task with a catecholamine challenge ( catecholamine reuptake blocker methylphenidate - MPH ) . We tested the following hypotheses: ( i ) cue-valence ( appetitive vs . aversive cues ) biases action in a Pavlovian manner , whereas outcome-valence ( reward vs . punishment ) biases instrumental learning of Go vs . NoGo actions; ( ii ) blocking the catecholamine reuptake with MPH enhances the strength of the Pavlovian response bias as a result of prolonged dopamine release to reward cues; ( iii ) MPH reduces the specificity of credit assignment to specific actions that elicited rewards , as the prolonged DA release to reward outcomes would spread credit to non-chosen active actions ( Figure 1 ) . 10 . 7554/eLife . 22169 . 004Figure 2 . Motivational Go/NoGo learning task and performance . ( A ) On each trial , a Win or Avoid cue appears on screen . Subjects can respond during cue presentation . Response-dependent feedback follows . ( B ) In total eight cues are presented for which the correct response needs to be learned . ( C ) Each cue has only one correct response ( Go-left , Go-right , or NoGo ) , which subjects can learn from the feedback . ( D ) Feedback is probabilistic . Correct responses are followed by reward ( Win cues ) or a neutral outcome ( Avoid cues ) in 80% of the time and by a neutral outcome ( Win cues ) or punishment ( Avoid cues ) otherwise . For incorrect responses , these probabilities are reversed . ( E ) Trial-by-trial proportion of Go responses ( ±SEM ) for Go cues ( solid lines ) and NoGo cues ( dashed lines ) , collapsed over Placebo and MPH . Left: All cue types . From the first trial onwards , subjects made more Go responses to Win vs . Avoid cues ( i . e . green lines are above red lines ) , reflecting the motivational bias . Additionally , subjects clearly learn whether to make a Go response or not ( proportion of Go responses increases for Go cues and decreases for NoGo cues ) . Right: Go cues only . For the Go cues , a Go response could be either correct or incorrect . The motivational bias is present in both correct and incorrect Go responses , but incorrect Go responses are unlearnt . Note that the total p ( Go ) in this plot sums up to the solid lines in the left plot . ( F ) Mean ( ±SED ) proportion Go responses . Proportion Go responses is higher for Go vs . NoGo cues , indicative of task learning . Additionally , subjects made more correct and incorrect Go responses to Win vs . Avoid cues . Source data of task performance are available in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 00410 . 7554/eLife . 22169 . 005Figure 2—source data 1 . Source data for task performance under MPH and placebo . This matlab datafile contains choice data ( subject x drug ( MPH/Placebo ) x trial ) for the Go-to-Win , Go-to-Avoid , NoGo-to-Win , and NoGo-to-Avoid cues . Go ( NoGo ) responses are coded as 1 ( 0 ) and choices are collapsed over the two cues of each category . Additionally , accuracy is provided for the Go cues , where correct ( incorrect ) responses are coded as 1 ( 0 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 00510 . 7554/eLife . 22169 . 006Figure 2—figure supplement 1 . Individual traces ( black lines ) and group average ( coloured lines ) of correct and incorrect Go responses using a sliding average of 5 trials . Traces are averaged within cue types and over sessions . Individual traces are semi-transparant , so that darker areas reflect more overlaying subjects . Across trials , subjects increased correct Go responses ( top ) and decreased incorrect Go responses ( bottom ) . Subjects performed at ceiling level more rapidly for the Go-to-Win cues ( top-left ) than Go-to-Avoid cues ( top-right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 006 Finally , MPH prolongs the effects of catecholamine release by blocking the reuptake of catecholamines , without stimulating release or acting as a receptor ( ant ) agonist ( e . g . Volkow et al . , 2002 ) . Accordingly , it is likely that the effect of MPH on catecholamine-dependent function is a function of dopamine synthesis capacity and release . Simply put , if there is no release , there is no reuptake to block . To assess these potential sources of individual variability in MPH effects , we took into account two measures that have been demonstrated with PET to relate to dopamine baseline function: working memory span for its relation to striatal dopamine synthesis capacity ( Cools et al . , 2008; Landau et al . , 2009 ) and trait impulsivity for its relation to dopamine ( auto ) receptor availability ( Buckholtz et al . , 2010; Kim et al . , 2014; Lee et al . , 2009; Reeves et al . , 2012 ) , and collected a large sample ( N = 106 ) to expose individual differences .
Subjects successfully learned this difficult task , in which they needed to identify the correct response out of 3 options ( Go-left/Go-right/NoGo ) for eight different cues , as evidenced by increased Go responding to cues indicating the need to Go vs . NoGo ( Required Action: X2 ( 1 ) =624 . 3; p<0 . 001; Figure 2E , F ) . In other words , subjects were able to adjust Go responding to the required action . As expected , cue valence also influenced Go responding ( Valence: X2 ( 1 ) =40 . 0; p<0 . 001 ) , reflecting a motivational bias in responding . Overall subjects made more Go responses for Win than Avoid cues . The effect of cue valence was highly significant for both Go and NoGo cues ( Go cues: X2 ( 1 ) =37 . 5 , p<0 . 001; NoGo cues: X2 ( 1 ) =13 . 3 , p<0 . 001 ) , though marginally stronger for the Go cues ( Required Action x Valence: X2 ( 1 ) =3 . 1; p=0 . 076 ) . Because each Go cue was associated with only one correct Go response , we confirmed that this motivational bias was present for both correct and incorrect Go responses . Subjects made more Go responses to Win than avoid cues for both correct ( Valence: X2 ( 1 ) =26 . 1 , p<0 . 001 ) and incorrect ( Valence: X2 ( 1 ) =25 . 6 , p<0 . 001 ) Go responses . Next , we tested the hypothesis that this motivational bias arose from a combination of a Pavlovian response bias and biased instrumental learning ( Figure 1A–B ) . We used a computational modelling approach to quantify latent processes that we hypothesised to underlie the behavioural performance . Specifically , our first aim was to disentangle the contribution of Pavlovian response biases and instrumental learning biases to the observed valence effect in behaviour . To this end we extended a simple reinforcement learning model using hierarchical Bayesian parameter estimation . We developed five nested base models ( M1 , M2 , M3a , M3b , M4 ) with increasing complexity to assess whether additional parameters explained the observed data better , while penalizing for increasing complexity . In all models , the probability of each response is estimated based on computed action weights . In the simplest model ( M1 ) the action weights are fully determined by the learned action values ( Q-values ) . Action values are updated with the prediction error , i . e . the deviation of the observed outcome from the expected outcome ( standard ‘delta-rule’ learning; Rescorla and Wagner , 1972 ) . M1 contains two free parameters: a learning rate ( ε ) scaling impact of the prediction-error , and feedback sensitivity ( ρ ) scaling the outcome value . Next , to allow for a non-selective bias in Go responses unrelated to valence , a go bias parameter ( b ) is added to the action weights of Go responses in M2 . This parameter simply captures how likely people are to make a ‘Go’ response overall . In this task , we explicitly instructed the cue valence , by colouring the edge of each cue , where green signalled that subjects could win a reward , while red signalled they had to avoid a punishment ( Figure 2A ) . As a consequence , we observed an effect of the instructed cue valence on Go responses already from the first trial onwards ( Figure 2E ) , implying a motivational bias before learning could occur , which is therefore likely Pavlovian in nature . To assess this Pavlovian response bias , cue values are added to the action weights in M3a . In this model positive ( negative ) Pavlovian values increase ( decrease ) the action weight of Go responses , where π scales the weight of the Pavlovian values ( Cavanagh et al . , 2013; Guitart-Masip et al . , 2014b; 2012 ) . Thus , the Pavlovian bias parameter increases the probability of all Go responses for Win cues and decreases the probability of all Go responses for Avoid cues . In M3b we assessed whether a motivational learning bias affects behaviour . Specifically , we included an instrumental learning bias parameter ( κ ) , to assess whether reward is more effective in reinforcing Go responses than NoGo responses , whereas punishment is less effective in unlearning NoGo responses than Go responses . This biased learning parameter indexes the degree to which the specific Go response that elicited a reward would be relatively more likely to be repeated in subsequent trials , resulting in increased instrumental learning of Go responses for reward . Note that earlier studies used only a single Go response and could thus not dissociate this specific learning vs . Pavlovian bias account . In addition to this effect on learning from rewards , κ indexes the degree to which punishment is biased to potentiate activity in the NoGo versus Go pathway , thus biasing unlearning to be more effective after Go responses than after NoGo responses , ( i . e . , making punishment-based avoidance learning of NoGo responses more difficult than punishment-based avoidance learning of Go responses; Figure 1B ) . Because the Pavlovian and instrumental learning bias might explain similar variance in the data , we tested model M4 , where we included both π and κ to test whether there was evidence for the independent presence of both the instrumental learning bias and the Pavlovian response bias . Stepwise addition of the go bias ( Appendix 5 ) , Pavlovian response bias and instrumental learning bias parameter improved model fit , as quantified by Watanabe-Akaike Information Criteria ( WAIC; Figure 3; Table 1 ) . The Pavlovian bias parameter estimates ( π ) of the winning model M4 were positive across the group ( 96 . 4% of posterior distribution >0 ) . The Pavlovian bias estimates were modest across the group ( Figure 3; Table 1 ) , and showed strong individual variability ( Figure 3—figure supplement 2; Figure 3—figure supplement 3 ) . This strong inter-individual variability is consistent with previous reports , e . g . Cavanagh et al . ( 2013 ) , who show that differences in the strength of the Pavlovian bias is inversely predicted by EEG mid-frontal theta activity during incongruent relative to congruent cues , putatively reflecting the ability to suppress this bias on incongruent trials . The further improvement of model fit due to the instrumental learning bias parameter ( M3a vs . M4 ) provides clear evidence for the contribution of biased action learning on top of the Pavlovian response bias described in previous studies . The biased instrumental learning parameter estimates were also positive across the group ( 100% of posterior distribution >0 ) . In other words , in the winning model , the motivational bias , as reflected by an increase in Go responses to Win relative to Avoid cues , is explained by the presence of both a Pavlovian response bias and biased instrumental learning . Figure 3 and accompanying Figure supplements illustrate the model predictions and parameter estimates . 10 . 7554/eLife . 22169 . 007Figure 3 . Model evidence and parameter inference of base models . ( A ) Model evidence , relative to simplest model M1 , clearly favours M4 . The simplest model M1 contains a feedback sensitivity ( ρ ) and learning rate ( ε ) parameter . Stepwise addition of the go bias ( b ) , Pavlovian bias ( π; Figure 1A ) , and instrumental learning bias ( κ; Figure 1B ) parameter improves model fit , quantified by WAIC ( estimated log model evidence ) . Lower ( i . e . more negative ) WAIC indicates better model fit . ( B ) Temporal dynamics of the correlation between the motivational bias parameters ( M4 ) and the predicted motivational bias , i . e . probability to make a Go response to Win relative to Avoid cues . The impact of the Pavlovian bias ( π ) on choice decreases over time ( although , importantly , the parameter itself remains constant ) . This is because the instrumental values of the actions are learnt and thus will increasingly diverge . As a result , π is less and less 'able' to tip the balance in favour of the responses in direction of the motivational bias ( i . e . it can no longer overcome the difference in instrumental action values ) . In contrast , the impact of κ on choice increases over time , reflecting the cumulative impact of biased learning ( also Figure 3—figure supplement 2 ) . ( C ) Posterior densities of the winning base model M4 . Appendix 5 shows posterior densities for all models . ( D ) One-step-ahead predictions and posterior predictive model simulations of winning base model M4 ( coloured lines ) , to assess whether the winning model captures the behavioural data ( grey lines ) . Both absolute model fit methods use the fitted parameters to compute the choice probabilities according to the model . The one-step-ahead predictions compute probabilities based on the history of each subject's actual choices and outcomes , whereas the simulation method generates new choices and outcomes based on the response probabilities ( see Materials and methods for details ) . Both methods capture the key features of the data , i . e . responses are learnt ( more 'Go' responding for 'Go' cues relative to 'NoGo' cues ) and a motivational bias ( more Go responding for Win relative to Avoid cues ) . We note that the model somewhat underestimates the initial Pavlovian bias ( i . e . difference in Go responding between Win and Avoid trials is , particularly trial 1–2 ) , while it overestimates the Pavlovian bias on later trials . This is likely the result from the fact that while the modelled Pavlovian bias parameter ( π ) is constant over time , the impact of the Pavlovian stimulus values weakens over time , as the subjects’ confidence in the instrumental action values increases . Interestingly , notwithstanding the constancy of the Pavlovian bias parameter , we do capture some of these dynamics as Figure 3B shows that the impact of the Pavlovian bias on choice decreases over time . Source data of M4 simulated task performance are available in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 00710 . 7554/eLife . 22169 . 008Figure 3—source data 1 . Source data for model M4 simulated task performance . This matlab datafile contains the probability of Go responses ( subject x trial ) for the Go-to-Win , Go-to-Avoid , NoGo-to-Win , and NoGo-to-Avoid cues , as simulated by model M4 . Posterior predictive model simulations used the M4 sampled parameter combinations of each subject . Simulations were repeated for each sampled parameter combination ( 4000 times ) , in line with the Bayesian nature of the sampling procedure and to minimize randomness , and choice probabilities were averaged over simulations . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 00810 . 7554/eLife . 22169 . 009Figure 3—figure supplement 1 . Subject traces of model M4 ( green/red ) overlaid on observed behavior ( black ) . M4 one-step-ahead predictions capture the individual variability in task performance . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 00910 . 7554/eLife . 22169 . 010Figure 3—figure supplement 2 . Illustration of the behavioural effects associated with the Pavlovian bias and instrumental learning bias parameters . Model M4 one-step-ahead predictions ( coloured ) overlaid on real data ( grey ) for the subjects with the upper versus lower tertile of parameter estimates . ( A ) Effects of Pavlovian bias ( π ) . A strong Pavlovian bias ( top 33% of π estimates ) predicts higher Go responding for the Win than Avoid cues from the first trial onward , vice versa for a weak Pavlovian bias ( 33% lowest π estimates ) . ( B ) Effects of instrumental learning bias ( κ ) . A strong instrumental learning bias ( 33% highest κ estimates ) predicts steeper Go-to-Win learning and shallower Go-to-Avoid learning , vice versa for a weak instrumental learning bias ( 33% lowest κ estimates ) . See also Figure 3B for the temporal dynamics of the parameter-behaviour correlations . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 01010 . 7554/eLife . 22169 . 011Figure 3—figure supplement 3 . M4 subject-level parameters in model space ( i . e . untransformed ) . The diagonal panels contain the posterior densities for the subject-level parameter means . The off-diagonal panels show the correlation over subjects in mean parameter estimates . Importantly , the two key parameters , Pavlovian bias ( π ) and instrumental learning bias ( κ ) are not correlated to any of the other parameters . We do note that the feedback sensitivity parameter ( ρ ) is anti-correlated with the learning rate ( ε ) , such that the impact of high feedback sensitivity estimates is restricted by low learning rates . This correlation is not problematic , because independent estimation of learning rate and feedback sensitivity is no direct interest to the questions we ask . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 01110 . 7554/eLife . 22169 . 012Table 1 . Base models . Median [25–75 percentile] of subject-level parameter estimates in model space . See Appendix 5 for subject-level / top-level parameters in sampling space ( i . e . untransformed ) . Absolute WAIC is reported at the top as the estimate of model evidence , where a smaller WAIC indicates higher evidence . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 012Base modelsM1M2M3aM3bM4WAIC7101469038676786760266987ρ42 . 7 [19 . 3 79 . 8]41 . 6 [18 . 7 72 . 4]35 . 2 [15 . 8 66 . 4]33 . 4 [13 . 9 59 . 8]32 . 5 [14 . 9 56 . 4]ε00 . 013 [0 . 008 0 . 059]0 . 015 [0 . 008 0 . 054]0 . 017 [0 . 009 0 . 064]0 . 022 [0 . 010 0 . 070]0 . 021 [0 . 010 0 . 063]b−0 . 25 [−0 . 45 0 . 04]−0 . 25 [−0 . 46 0 . 04] . 01 [−0 . 33 0 . 27]−0 . 03 [−0 . 29 0 . 19]π0 . 47 [0 . 02 1 . 00]0 . 12 [−0 . 29 0 . 70]ε rewarded Go ( ε0+κ ) 0 . 037 [0 . 016 0 . 122]0 . 034 [0 . 016 0 . 109]ε punished NoGo ( ε0-κ ) 0 . 006 [0 . 002 0 . 014]0 . 008 [0 . 003 0 . 022] Next , we asked whether acute administration of MPH altered the motivational bias . As noted above , the effects of dopaminergic drugs often depend on baseline dopamine function . We therefore used two neuropsychological measures that have been shown to predict baseline dopamine function using PET: working memory span , predictive of baseline dopamine synthesis capacity ( Cools et al . , 2008; Landau et al . , 2009 ) , and trait impulsivity , predictive of D2 autoreceptor availability ( Buckholtz et al . , 2010; Kim et al . , 2014; Lee et al . , 2009; Reeves et al . , 2012 ) . Importantly , both working memory span and trait impulsivity predict dopaminergic drugs effects on various cognitive functions ( Clatworthy et al . , 2009; Cools et al . , 2009; 2007; Frank and O'Reilly , 2006; Gibbs and D'Esposito , 2005; Kimberg et al . , 1997; Zimet et al . , 1988 ) . MPH enhanced the effect of cue valence on Go responding proportional to working memory span ( Valence x Drug x Listening Span: X2 ( 1 ) =5 . 9; p=0 . 016; Figure 4B ) , in the absence of a Valence x Drug effect across the group ( Valence x Drug: X2 ( 1 ) =1 . 5; p=0 . 221; Figure 4A ) . While high-span subjects showed a drug-induced increase in motivational bias ( MPH versus placebo increased Go responding to Win vs . Avoid cues ) , low-span subjects showed a drug-induced decrease in motivational bias . This span-dependent bias emerged under MPH ( X2 ( 1 ) =4 . 6 , p=0 . 032 ) , and was not significant under placebo ( X2 ( 1 ) =0 . 9 , p=0 . 335; Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 22169 . 013Figure 4 . MPH-induced changes in motivational bias ( i . e . proportion of Go responses to Win relative to Avoid cues ) . ( A ) Mean ( ±SED ) proportion Go responses under MPH relative to Placebo . MPH did not significantly alter the motivational bias across the group ( p=0 . 22; ns indicates p>0 . 05 ) . ( B ) MPH increased the motivational bias in high span subjects , yet decreased it in low span subjects ( R = 0 . 21; p=0 . 016 ) . ( C ) MPH altered the motivational bias particularly for incorrect Go proportional to working memory span ( incorrect Go: p<0 . 001; correct Go: p=0 . 152 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 01310 . 7554/eLife . 22169 . 014Figure 4—figure supplement 1 . Simple effects of MPH-induced changes in motivational bias . ( A ) The span-dependent motivational bias emerged under MPH ( right; p=0 . 032 ) , and was not significant under placebo ( left; p=0 . 34 ) . ( B ) MPH did not significantly alter the motivational bias proportional to working memory span for correct Go responses ( correct Go: p=0 . 15 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 014 A break-down of this effect into correct and incorrect responses revealed that it was driven by incorrect Go responses ( Valence x Drug x Listening Span: X2 ( 1 ) =11 . 9 , p<0 . 001; Figure 4C ) . MPH did not significantly affect the correct Go responses ( Valence x Drug x Listening Span: X2 ( 1 ) =2 . 0 , p=0 . 152 ) . In other words , higher span subjects were more likely to make Go responses to Win cues under MPH , but this Go response was more likely to be incorrect . We reasoned that an enhanced learning bias would manifest primarily in increased correct Go responses to Win cues ( i . e . the correct responses are better learned ) , while an enhanced Pavlovian bias or diffusion of credit assignment would manifest in increased correct and incorrect Go responses to Win cues ( due to overall action invigoration and potentiation respectively ) . Thus , we expected that the altered valence effect on incorrect Go responses under MPH can best be attributed to MPH alteration of Pavlovian response bias or diffusion of credit assignment , which we formally test using computational modelling ( see below ) . In contrast to listening span , trait impulsivity did not significantly predict the effect of MPH on the motivational bias ( all p>0 . 05; see Appendix 3 for an overview of the mixed model effects ) . We confirmed that the MPH effects were not explained by session effects , i . e . whether MPH was received on the first or second testing day ( X2 ( 2 ) =2 . 1 , p=0 . 349 ) , nor did the factor Testing day improve model fit ( X2 ( 1 ) =2 . 0 , p=0 . 162 ) . Finally , we confirmed that including nuisance variables Gender and NLV scores ( measuring verbal intelligence ) , did not improve model fit either ( X2 ( 2 ) =0 . 4 , p=0 . 815 ) . Continuing our modelling approach , we next assessed whether the MPH-induced motivational bias could be attributed to an altered Pavlovian response bias and/or instrumental learning bias . To this end we extended the winning base model M4 into competing models . In M5a we included an MPH-induced Pavlovian bias parameter ( πMPH ) , to assess whether MPH altered the Pavlovian response bias . Here πMPH alters the individual’s Pavlovian bias parameter under MPH . In M5b we included an MPH-induced instrumental learning bias ( κMPH-selective ) . Thus , M5b tests whether MPH affects the strength of the instrumental learning bias in individuals . We further tested whether MPH might make the learning bias more diffuse , because of its mechanisms of action . Because MPH blocks reuptake , it prolongs dopamine release , such that reinforcement and synaptic potentiation might not be attributed only to the temporally coincident neurons that code for the recently selected action , but could be spread to other actions ( diffuse learning ) . To test this hypothesis , M5c contains a MPH-induced diffuse learning bias ( κMPH-diffuse ) , where κMPH-diffuse is a learning rate that alters the value of all Go responses following a reward , under MPH ( Figure 1C ) by scaling the prediction error following all rewarded Go responses . Model fit improved markedly when extending the winning base model M4 with the MPH-induced Pavlovian bias parameter πMPH ( M5a; Figure 5; Table 2 ) . Extending M4 with the MPH-induced selective learning bias parameter κMPH-selective ( M5b ) only slightly increased model fit . Conversely , the MPH-induced diffuse learning bias parameter κMPH-diffuse ( M5c ) also strongly improved model fit relative to base model M4 . This observation is in line with our earlier prediction that the MPH effects are predominantly driven by changes in the proportion of incorrect Go responses . Confirming the model comparison results , the MPH modulation of Pavlovian bias and diffuse learning parameters both covaried with Listening Span ( πMPH: R = 0 . 25 , p=0 . 013; κMPH-diffuse: R = 0 . 28 , p=0 . 006 ) , while the MPH selective learning bias did not ( κMPH-selective: R = −0 . 01 , p=0 . 9 ) . In other words , κMPH-selective did not explain our effect of interest and improved model fit relatively weakly . 10 . 7554/eLife . 22169 . 015Figure 5 . Model evidence and parameter inference of extended MPH models . ( A ) Model evidence ( WAIC ) relative to winning base model M4 . We tested whether MPH alters the strength of the Pavlovian response bias ( πMPH; M5a ) , the instrumental learning bias ( κMPH-selective; M5b ) , or has a diffuse effect on the learning bias ( κMPH-diffuse; M5c; Figure 1C ) . Model selection favoured the composite model M6 , including the πMPH and κMPH-diffuse parameters . ( B ) Posterior densities of the top-level parameters of M6 . ( C ) Subject-level estimates of MPH-induced Pavlovian bias parameter ( upper ) and the MPH-induced diffuse learning bias parameter ( lower; logistic scale ) correlated significantly with Listening Span . ( D ) One-step-ahead model predictions and posterior predictive model simulations of M6 using subject-level parameter estimates . The model predictions and simulations echo the observed data , i . e . that the motivational bias correlates positively with working memory span ( Figure 4B ) , confirming the winning model M6 captures the MPH-induced increase in Go responses to Win vs . Avoid cues . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 01510 . 7554/eLife . 22169 . 016Figure 5—figure supplement 1 . Illustration of the behavioural effects of MPH related to the Pavlovian bias and diffuse learning bias parameters . Model M6 one-step-ahead predictions ( coloured ) overlaid on real data ( grey ) for the subjects with the 33% strongest vs . weakest parameter estimates . The coloured bars at the bottom indicate the trial-by-trial correlation across all subjects , of the parameter estimate with the effect of MPH on Go responding per cue . The R value indicates the average correlation . ( A ) The effect of MPH on Pavlovian bias ( πMPH ) . Strong πMPH estimates predict that MPH increases the motivational bias ( increased Go to Win cues and decreased Go to Avoid cues ) , and vice versa for weak πMPH estimates . The influence of πMPH is present from the first trial onward and decreases over time as indicated by the correlation coefficients . ( B ) Effect of MPH on diffuse learning bias ( κMPH-diffuse ) . Strong κMPH-diffuse estimates predict that MPH increases the motivational bias for Win cues specifically , whereas this effect is diminished for subjects with relatively weak κMPH-diffuse estimates . The effect of κMPH-diffuse is experience-dependent and evolves over time . These one-step-ahead predictions illustrate how each parameter results in an increased motivational bias under MPH , but with unique temporal dynamics , even though the parameter themselves are constant . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 01610 . 7554/eLife . 22169 . 017Figure 5—figure supplement 2 . M6 subject-level parameters in model space ( i . e . untransformed ) . The diagonal panels contain the posterior densities for the subject-level parameter means . The off-diagonal panels contain the parameter correlations over subjects . Importantly , the parameters estimating the effects of MPH on Pavlovian bias ( πMPH ) and diffuse learning bias ( κMPH-diffuse ) are not correlated any of the other parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 01710 . 7554/eLife . 22169 . 018Table 2 . MPH models . Median [25–75 percentile] of subject-level parameter estimates in model space . Absolute WAIC is reported as the estimate of model evidence , where a smaller WAIC indicates higher evidence . Biased instrumental learning rate for rewarded Go and punished NoGo responses as computed by ε0±κ under placebo and by ε0± ( κ+κMPH ) under MPH . ( MPH ) indicates the value of that parameter under MPH . DOI: http://dx . doi . org/10 . 7554/eLife . 22169 . 018Extended MPH modelsM5aM5bM5cM6WAIC66383668836659566069ρ31 . 2 [14 . 7 53 . 6]31 . 6 [15 . 6 57 . 0]55 . 8 [19 . 6 104 . 8]51 . 9 [20 . 6 98 . 7]ε00 . 022 [0 . 010 0 . 067]0 . 021 [0 . 011 0 . 061]0 . 011 [0 . 006 0 . 051]0 . 012 [0 . 006 0 . 055]b−0 . 04 [−0 . 33 0 . 18]−0 . 05 [−0 . 34]−0 . 10 [−0 . 37 0 . 13]−0 . 14 [−0 . 42 0 . 10]π π ( MPH ) 0 . 27 [−0 . 50 . 71] 0 . 20 [−0 . 38 . 71]0 . 15 [−0 . 28 . 70]0 . 05 [−0 . 46 . 61]0 . 27 [−0 . 47 . 74] −0 . 05 [−0 . 70 . 50]ε rewarded Go ε rewarded Go ( MPH ) 0 . 037 [ . 017 . 116]0 . 030 [ . 018 . 103] 0 . 031 [ . 016 . 104]0 . 018 [ . 009 . 082]0 . 019 [ . 009 . 085]ε punished NoGo ε punished NoGo ( MPH ) 0 . 009 [ . 004 . 030]0 . 009 [ . 003 . 021] 0 . 008 [ . 002 . 021]0 . 004 [ . 002 . 013]0 . 005 [ . 002 . 017]ε diffuse ( MPH ) 0 . 002 [ . 002 . 004]0 . 003 [ . 002 . 004] To assess whether πMPH and κMPH-diffuse explained unique Listening Span-dependent effects of MPH ( i . e . whether there was evidence for both of these effects ) , we constructed a composite model ( M6 ) containing both effects . Model comparison showed that indeed this composite model explained the data best ( Figure 5 ) . In this model , both parameters again significantly varied proportional to Listening Span ( πMPH: R = 0 . 24 , p=0 . 020; κMPH-diffuse: R = 0 . 22 , p=0 . 032; Figure 5 ) . Taken together , these modelling results attribute the MPH-induced motivational bias partly to an altered Pavlovian response bias ( πMPH ) , and partly to a reward-driven diffusion of credit during instrumental learning ( κMPH-diffuse ) . In other words , MPH ( i ) alters the impact of cue valence on action , which is present and persists from the first trial onward , and ( ii ) alters the impact of rewarding outcomes on the learning of actions , which fully depends on and evolves with experience . Following a reward , the effect of κMPH-diffuse is to increase the value of incorrect Go responses in addition to the correct Go response . Finally , we tested whether our best fitting model was sufficient to reproduce the key features of the data . This is important because model selection only provides relative , but not absolute evidence for the winning model ( e . g . , Nassar and Frank , 2016 ) . We used two approaches to compute the post hoc absolute model fit , namely data simulation and ‘one-step-ahead’ model predictions . In the simulation method , the first choice is simulated based on the initial values; the corresponding outcome used for learning; the next choice is simulated based on the updated , learned values; and so on . Thus , this simulation method ignores any subject-specific sequential/history effects to determine the current choice probability . Therefore , this can result in choice/outcome sequences that diverge completely from the subjects’ actual experiences . Violating the subject-specific choice and outcome history will change the learning effects , making this method less robust in generating the exact learning effects compared to experience-independent effects . We therefore included a second absolute model fit method that does take into account the subjects’ choice and outcome histories: the post-hoc absolute fit method ( also known as ‘one-step-ahead prediction’; Pedersen et al . , 2016; Steingroever and Wagenmakers , 2014 ) . Here , the initial choice probabilities are determined based on the initial values . For each subsequent trial , the choice probabilities are determined based on the learned values using the actual ( subject’s ) choices and outcomes on all preceding trials . We used both methods as the strongest test providing converging evidence that the models could capture the observed results . Using both absolute model fit methods , we simulated choices for each individual , using model M6 with each individual’s parameter estimates . Both methods confirmed that M6 can capture the observed effects , replicating the Listening Span dependent effect of MPH on choice , where MPH increased Go responses to Win vs . Avoid cues more in higher span subjects ( simulations: R = 0 . 27 , p=0 . 008; one-step-ahead: R = 0 . 20 , p=0 . 050; Figure 5 ) . These simulations echo the results reported above , demonstrating the MPH-induced Pavlovian bias parameter πMPH and diffuse learning bias κMPH-diffuse are sufficient to both explain and predict the span-dependent MPH-induced increase in Go responses to Win vs . Avoid cues . Figure 5 and accompanying Figure supplements illustrate the model predictions and parameter estimates .
Cue valence affected activation versus inhibition of behaviour , consistent with previous reports ( Geurts et al . , 2013; Guitart-Masip et al . , 2012 ) . Even though cue valence was orthogonal to what subjects should be doing , subjects made more Go responses when pursuing reward , and fewer Go responses when trying to avoid punishment . We and others have previously suggested that this motivational asymmetry in behavioural activation entails Pavlovian control over instrumental behaviour ( Cavanagh et al . , 2013; Geurts et al . , 2013; Huys et al . , 2011 ) . Here we challenge this initial idea , and argue that motivational valence may also bias instrumental learning . To disentangle the hypothesised contribution of a Pavlovian response bias from biased instrumental learning , we extended existing paradigms by incorporating multiple Go response options . For the cues requiring active responses , only one response option was considered correct , enabling us to disentangle general activation from specific action learning . For cues where subjects had to activate responding ( ‘Go’ cues ) , they increased both correct and incorrect Go responses when pursuing reward compared with when avoiding punishment . Thus , the increased activation towards reward was in part beneficial , and in part detrimental . We used computational models to formalise our hypothesis regarding a dissociable contribution of Pavlovian activation and biased instrumental learning . We then fitted competing models to the subjects' choices , and compared the performance of all models . We demonstrate that cue valence shapes behavioural activation/inhibition in a Pavlovian manner , and additionally that outcome valence biases instrumental learning of activation/inhibition: reward enhances the learning of specific active actions , and punishment suppresses the unlearning of inactions . In short , we are quicker to believe that an action led to a reward , but reluctant to attribute a punishment to having held back . Current views of striatal dopamine function ( Collins and Frank , 2015; 2014; Frank , 2006; Frank , 2005; Lloyd and Dayan , 2016 ) suggest that the striatal architecture is well suited to implement the Pavlovian asymmetry in behavioural activation . Appetitive ( aversive ) conditioned cues elicit peaks ( dips ) in mesolimbic dopamine release in the striatum ( Cohen et al . , 2012; Day et al . , 2007; Matsumoto and Hikosaka , 2009; Tobler et al . , 2005 ) . Increased striatal dopamine levels activate the direct D1 ( ‘Go’ ) pathway ( Hernández-López et al . , 1997 ) , which promotes behavioural activation ( DeLong and Wichmann , 2007; Mink and Thach , 1991 ) , whereas decreased striatal dopamine levels activate the indirect D2 ( ‘NoGo’ ) pathway ( Hernandez-Lopez et al . , 2000 ) , promoting behavioural inhibition . In striatal dopamine models , increased dopamine biases action selection to be driven more by the potential rewards of alternative actions encoded in D1 neurons and less by the costs encoded in D2 neurons ( Collins and Frank , 2014; see also recent optogenetic experiment supporting this notion; Zalocusky et al . , 2016 ) , but this can also be manifest in terms of Pavlovian biases . Taken together , the striatal ( in ) direct pathways provide a neural mechanism for implementing Pavlovian activation to appetitive vs . aversive cues . In parallel with our behavioural findings , the same striatal pathways may also generate the asymmetry in action learning . Here , dopamine bursts elicited by reward prediction errors ( Montague et al . , 2004; Schultz et al . , 1998; Schultz et al . , 1997 ) during the outcome , enhance long-term potentiation ( LTP ) of the corticostriatal synapses associated with the just-performed response ( Frank et al . , 2004 ) . Importantly , enhancing LTP in the ‘Go’ pathway should promote learning of active responses , relative to learning the inhibition of actions . Recent experiments show temporally and spatially selective enhancement of corticostriatal spines given glutamatergic input ( putatively representing the selected action ) and followed closely in time by dopaminergic bursts ( Yagishita et al . , 2014 ) . Thus , prolonged release of DA ( e . g . after DAT blockade ) might reduce this selectivity , and diffuse the specificity of credit assignment . Conversely , striatal dopamine dips following negative prediction errors can drive avoidance by promoting long-term depression ( LTD ) in the ‘Go’ pathway and LTP in the ‘NoGo’ pathway ( Beeler et al . , 2012; Frank , 2005; Shen et al . , 2008 ) . Indeed , transient optogenetic inhibition of DA induces behavioural avoidance of recently selected actions ( Danjo et al . , 2014; Hamid et al . , 2016 ) , an effect that depends on D2 receptors ( Danjo et al . , 2014 ) . D2 neurons are excited in response to losses ( Zalocusky et al . , 2016 ) ; their activation during losses induces subsequent avoidance learning ( Kravitz et al . , 2012; Zalocusky et al . , 2016 ) , and their disruption prevents avoidance learning ( Hikida et al . , 2010 ) . While LTP in the NoGo pathway would be beneficial for unlearning to perform actions , LTP in the NoGo pathway would be detrimental in case of unlearning to make NoGo responses ( i . e . attributing a punishment to a NoGo response ) . To summarize , the dopamine peaks following positive reinforcement can enhance learning of actions by enhancing LTP in the striatal ‘Go’ pathway . Conversely , the dopamine dips following negative outcomes can disrupt learning to initiate responses by increasing LTD in the ‘Go’ pathway and LTP in the NoGo pathway . Blocking the reuptake of catecholamines with MPH altered the extent to which subjects were influenced by the cue and outcome valence . This effect of MPH was highly variable between individuals , and depended on working memory span . In high relative to low span subjects , MPH enhanced the influence of valence , such that subjects made even more active responses when pursuing reward and displayed more inhibition when avoiding punishment . This effect was driven particularly by changes in the proportion of incorrect Go responses that subjects made . Formal modelling showed that this effect was due to MPH affecting both generalized Pavlovian activation and a diffusion of credit assignment . Specifically , MPH induced a spread of credit assignment following rewarded active responses , rather than magnifying the selective instrumental learning bias . We argue that both of these effects can be understood as reflecting prolonged catecholamine presence in the synaptic cleft with MPH . Blocking catecholamine reuptake with MPH extends the duration of dopamine presence in the synaptic cleft ( Dreyer and Hounsgaard , 2013 ) . This prolonged dopamine presence ( i . e . reduced temporal specificity ) would be less selective in potentiating the actions that were selected immediately prior to rewards ( e . g . Yagishita et al . , 2014 ) . This would reduce credit assignment of specific active actions , but still bias reinforcement of actions more generally ( e . g . Collins and Frank , 2015; Syed et al . , 2016 ) . This account explains why MPH modulates the strength of the Pavlovian activation ( which is inherently global ) but not of the specific instrumental learning bias ( which is inherently selective ) . Our results indeed provided evidence for this diffusing effect of MPH on the instrumental learning bias , such that reward potentiates actions globally . The data were best explained by a combination of this diffuse instrumental learning and Pavlovian response bias modulation . Thus , on the one hand MPH modulated the impact of the cue valence on behavioural activation , which surfaces already before any learning has taken place . On the other hand , MPH spread credit assignment following rewarded responses to all Go responses , which is an experience-dependent effect . Our results are highly consistent with those predicted from current models of dopamine in the basal ganglia , suggesting that the effects of MPH are due to modulation of striatal dopamine . Of course , the present study does not allow us to exclude the possibility that ( part of ) the effects were mediated by extra-striatal , e . g . prefrontal regions ( Spencer et al . , 2015 ) , or by the noradrenaline system ( Arnsten and Dudley , 2005 ) . Future studies are needed to investigate directly the site of the presently observed effects of MPH , e . g . with fMRI , and dopamine dependence and selectivity , e . g . with selective dopamine antagonists . Individuals vary strongly in the extent to which MPH increases extracellular dopamine ( Volkow et al . , 2002 ) . We therefore anticipated that the effect of MPH would covary with measures relating to baseline dopamine function . We assessed whether MPH effects were predicted by ( i ) working memory span , given its known relation to dopamine synthesis capacity ( Cools et al . , 2008; Landau et al . , 2009 ) , and ( ii ) trait impulsivity , for its known relation to D2 ( auto ) receptor availability ( Buckholtz et al . , 2010; Kim et al . , 2014; Lee et al . , 2009; Reeves et al . , 2012 ) . MPH affected choice behaviour proportional to working memory span , but not trait impulsivity . Subjects with higher working memory span , linked to higher striatal synthesis capacity , showed a relative increase in both Pavlovian response bias and spread of credit assignment under MPH . This finding that transporter blockade has stronger effects in those individuals with putatively higher baseline dopamine is in line with the observation that MPH increases dopamine levels more in individuals with higher dopamine cell activity ( van der Schaaf et al . , 2013; Volkow et al . , 2002 ) . Indeed , baseline dopamine cell activity is a better predictor of effects of MPH than either D2 auto-receptor availability or DAT occupancy under MPH ( Volkow et al . , 2002 ) . Together this may explain why the observed MPH effects covary with working memory span but not trait impulsivity . The finding that drug effects depend on working memory is highly consistent with the hypothesis that they reflect modulation of striatal dopamine ( c . f . Frank and Fossella , 2011 ) . However , we need to be cautious in our interpretation . First , both striatal and prefrontal dopamine are known to contribute to working memory performance ( updating and maintenance respectively; e . g . Cools and D'Esposito , 2011 ) . The Listening Span task does not dissociate between working memory updating and maintenance , and thus a contribution of modulation of prefrontal dopamine cannot be excluded . Another possibility raised by the finding that drug effects depend on span , is that they reflect modulation of working memory itself , rather than reflecting dependence on baseline dopamine synthesis capacity . However , we argue that this is unlikely , because there was no significant effect of baseline working memory on motivational bias under placebo conditions . Rather , this relationship was induced by MPH . For future studies , it would be of interest to also include other measures related to baseline dopamine levels , such as eyeblink rates . More broadly , further research is required to identify the optimal combination of the various proxy measures of individual variability in the dopamine system in order to account for the large inter-individual variability in dopaminergic drug response . This is one of the major aims of our ongoing work . Across subjects , MPH increased subjective experiences of positive affect and alertness , and decreased calmness ( Appendix 2 ) . In contrast to the MPH-induced Pavlovian response bias and diffuse learning bias , these non-specific mood changes did not covary with working memory span . In other words , the MPH-induced mood changes are orthogonal to our effect of interest . Therefore , the MPH effect on Pavlovian activation and biased instrumental learning cannot be attributed to MPH-induced changes in mood . This study elucidates two distinct mechanisms by which motivational valence can bias behaviour . Cue valence promotes activation/inhibition in a Pavlovian manner , whereas outcome valence affects action/inhibition learning . Blocking the reuptake of catecholamines with methylphenidate altered the Pavlovian response bias , and had a diffuse , rather than selective , effect on biased learning . The effect of methylphenidate on the Pavlovian bias and biased learning was predicted by working memory span , such that methylphenidate enhanced Pavlovian activation and biased learning proportional to working memory span . These results help bridge the study of motivational biasing of action and instrumental learning , and help refine current models of catecholamines in motivated action . The present observations suggest that we need to add a new dimension to the suggested dichotomy of the role of dopamine in learning versus performance . Our study brings together two literatures that emphasise the role of ( midbrain ) dopamine in reward ( prediction-error ) based learning on the one hand ( Collins and Frank , 2014; Frank et al . , 2004; Schultz et al . , 1997 ) , and motivation-driven performance and behavioural activation on the other ( Beierholm et al . , 2013; Berridge , 2007; Robbins and Everitt , 2007; Shiner et al . , 2012; Smittenaar et al . , 2012 ) . Our results suggest that these two interact , resulting in biased learning of action-reward and inaction-punishment links , putatively via the same striatal mechanism that drive motivational Pavlovian response biases . Like motivational response tendencies , such biased learning would allow us to optimally profit from stable environmental statistics , as this instrumental learning bias supports rapid learning of likely action-outcome associations ( e . g . that an action caused a reward ) , while avoiding learning unlikely , spurious , associations ( e . g . that inhibition caused a punishment ) .
The study consisted of two test sessions with an interval of one week to two months . The first test day started with informed consent , followed by a medical screening . Participation was discontinued if subjects met any of the exclusion criteria ( Appendix 1 ) . On both test days , subjects first completed baseline measures . Next subjects received a capsule containing either 20 mg MPH ( Ritalin , Novartis ) or placebo , in a double-blind , placebo-controlled , cross-over design . MPH blocks the dopamine and noradrenaline transporters , thereby diminishing the reuptake of catecholamines . When administered orally , MPH has a maximal plasma concentration after 2 hr and a plasma half-life of 2–3 hr ( Kimko et al . , 1999 ) . After an interval of 50 min , subjects started with the task battery containing the motivational Go/NoGo learning task . See Appendix 2 for an overview of the task battery . On average the motivational Go/NoGo learning task was performed 2 hr after capsule intake , well within the peak of plasma concentration . Both test days lasted approximately 4 . 5 hr , which subjects started at the same time ( maximum difference of 45 min ) . Blood pressure , mood and potential medical symptoms were monitored three times each day: before capsule intake , upon start of the task battery and after finishing the task battery . Subjects were told to abstain from alcohol and recreational drugs 24 hr prior to testing and from smoking and drinking coffee on the days of testing . Subjects completed self-report questionnaires at home between ( but not on ) test days . Upon completion of the study , subjects received a monetary reimbursement or study credits for participation . The study was in line with the local ethical guidelines approved by the local ethics committee ( CMO / METC Arnhem Nijmegen: protocol NL47166 . 091 . 13 ) , pre-registered ( trial register NTR4653 , http://www . trialregister . nl/trialreg/admin/rctview . asp ? TC=4653 ) , and in accordance with the Helsinki Declaration of 1975 . Baseline measures , self-report questionnaires , mood- and medical symptom-ratings are reported in Appendix 2 . As individual differences were a main focus of the study , we collected a large sample of 106 native Dutch volunteers ( aged 18–28 years , mean ( SD ) = 21 . 5 ( 2 . 3 ) ; 53 women; 84 right-handed; sample size calculation reported in CMO protocol NL47166 . 091 . 13 ) . Four subjects dropped out after the first test day ( due to too much delay between test days , loss of motivation , nausea , and mild arrhythmia ) . Two subjects dissolved the capsules before swallowing and are discarded because of uncertainty in the pharmacodynamics . One subject did not sufficiently engage in the task ( only 13/2% Go responses on day 1/2 ) and was discarded as well . We repeated the analyses with these subjects included to confirm that this did not alter the conclusions ( Appendix 3 ) . Of the resulting 99 subjects , 48 subjects received MPH on the first day . Exclusion criteria comprised a history of psychiatric , neurological or endocrine disorders . Appendix 1 presents a complete overview of the exclusion criteria . Each trial started with the on-screen presentation of a cue ( Figure 2A ) . During cue presentation subjects could decide to press a button ( Go response ) or not ( NoGo response ) . Subjects could either press the left ( Go-left ) or right ( Go-right ) button on a button box . Subjects received feedback based on their response . Each cue had a red or green edge . Cues with a red edge ( Avoid cues ) were followed by neutral feedback or punishment . Cues with a green edge ( Win cues ) were followed by reward or neutral feedback . Subjects were informed about these contingencies . Note that the explicit cue valence is in contrast to previous studies where subjects needed to learn the cue valence during the task ( e . g . Cavanagh et al . , 2013; Guitart-Masip et al . , 2012 ) . The rationale of explicit cue valence was to directly observe effects of cue valence on choice and minimize individual differences in learning the cue valence . Punishment consisted of the display of the red text ‘−100’ , accompanied by a low buzz , reward of the green text ‘+100’ together with a flourish sound , and the neutral feedback of the grey text ‘000’ together with a short beep . All cues had unique shapes and colours well distinguishable from the red and green edge . Cue colour and shape were randomized over cue types . Two separate stimulus sets were used for the two test days to prevent transfer effects , and set order was counterbalanced across subjects . For each cue , there was one correct response ( Go-left , Go-right or NoGo; Figure 2C ) , which subjects had to learn by trial and error . Feedback validity was 80% , that is , correct ( incorrect ) responses were followed by the desirable outcome 80% ( 20% ) of the time ( Figure 2D ) . There were eight cues in total ( Figure 2B ) . The number of Go and NoGo cues was kept equal to prevent reinforcing an overall Go bias . The order of cue presentation was pseudorandom , as cues could be repeated once at most . Each cue was presented 40 times . The task lasted approximately 30 min , including instructions and a self-paced break halfway . The instructions were presented on screen . Subjects were informed about the probabilistic nature of the feedback and that each cue had one optimal response . At the end of the task the total number of points won or lost was displayed on screen and subjects were informed beforehand that these points would be converted to a monetary bonus at the end of the study ( mean = EUR2 . 90 , SD = 1 . 49 ) . Working memory span was assessed with the Listening Span Test ( Daneman and Carpenter , 1980; Salthouse and Babcock , 1991 ) , which was also used in two FMT PET studies showing positive correlations with striatal dopamine synthesis capacity ( Cools et al . , 2008; Landau et al . , 2009 ) . Subjects completed the Listening Span Test on day two prior to capsule intake . The Listening Span Test consists of sets of pre-recorded sentences , increasing from 2 to 7 sentences . Subjects are presented with the sentences , and required to simultaneously answer written verification questions regarding the content of each sentence . At the end of each set , subjects recalled the final word of each sentence in the order of presentation . The Listening Span reflects the set size of which the subject correctly recalled the final words on at least two out of three trials . Listening span increased with half a point , when only one trial of the next level was correct . Trait impulsivity was assessed with the Barratt Impulsiveness Scale ( BIS-11 ) ( Patton et al . , 1995 ) . The BIS-11 is a self-report questionnaire , consisting of 30 questions tapping in common ( non ) impulsive behaviours and preferences . The BIS-11 total impulsivity scores reflect the tendency towards impulsivity . Subjects completed the questionnaire at home between test days . To assess the influence of motivational valence on behavioural activation , we first analysed Go vs . NoGo responses ( irrespective of Go-left vs . Go-right ) . Second we tested whether effects on Go responses were explained by correct or incorrect Go responses . We were specifically interested how MPH altered Go/NoGo responding to Win vs . Avoid cues as a function of Listening Span and Impulsivity . To account for both between and within subject variability , choice data were analysed with logistic mixed-level models using the lme4 package in R ( Bates et al . , 2014; R Developement Core Team , 2015 ) . Reflecting our objectives , the mixed models included the within subject factors Drug ( MPH vs . placebo ) , Valence ( Win vs . Avoid cue ) , and Required Action ( Go vs . NoGo ) , and the between subject factors Listening Span and Impulsivity . The analysis of correct and incorrect Go responses included only the Go cues; hence this analysis did not include the factor Required Action . Models included all main effects and interactions , except for the interactions between Listening Span and Impulsivity . All models contained a full random effects structure ( Barr , 2013; Barr et al . , 2013 ) . We performed control analyses using a model comparison approach , where we tested whether the following factors improved model fit: Drug Order , Testing Day , Gender , and NLV ( a measure for verbal intelligence ) . For completeness , we analysed reaction times ( RTs ) as a measure of behavioural vigour ( Appendix 4 ) . In all models , action weights ( w ) are estimated for each response option ( a ) for all trials ( t ) per cue ( s ) . Based on these action weights choice probabilities are computed using a softmax function , as follows: ( 1 ) p ( at|st ) = [ exp ( w ( at , st ) ) ∑a'exp ( w ( a' , st ) ) ] In the simplest model ( M1 ) the action weights are fully determined by the learned action values ( Q-values ) . To compute the action values , we used standard delta-rule learning with two free parameters; a learning rate ( ε ) scaling the update term , and feedback sensitivity ( ρ ) scaling the outcome value ( comparable to the softmax temperature ) . ( 2 ) Qt ( at , st ) = Qt−1 ( at , st ) + ε ( ρrt− Qt−1 ( at , st ) ) Here outcomes are reflected by r , where r∈{−1 , 0 , 1} . In the current paradigm cue valence is instructed , by means of the green and red cue edges . Therefore , the initial expected outcome is 0 . 5 for Win cues and −0 . 5 for Avoid cues . Initial Q-values ( Q0 ) are set accordingly to ρ*0 . 5 for Win cues and ρ*−0 . 5 for Avoid cues . In M2 a go bias parameter ( b ) is added to the action weights of Go responses . We then explored the influence of Pavlovian biases that modulate Go responding according to predicted reward value . Pavlovian values ( V ) contribute to the action weights in M3a , increasing ( decreasing ) the weight of Go responses for positive ( negative ) Pavlovian values respectively . ( 3 ) w ( at , st ) = {Q ( at , st ) + πV ( s ) + bif a=GoQ ( at , st ) else Here the weight of the Pavlovian values is determined by the parameter π . Pavlovian values are fixed at 0 . 5 for Win cues and at −0 . 5 for Avoid cues , again because cue valence is instructed . In M3b we included the instrumental learning bias parameter ( κ ) instead of the Pavlovian bias , to assess whether the motivational bias can be explained in terms of enhanced learning of Go following a reward , and disrupted learning from punishment following NoGo . ( 4 ) ϵ={ϵ0+κif rt=1 & a=goϵ0−κif rt=−1 & a=nogoϵ0else In model M4 , we included both the Pavlovian bias parameter and the instrumental learning bias parameter . We used a sampling method for hierarchical Bayesian estimation of group-level and subject-level parameters . The group-level parameters ( X ) serve as priors for the individual-level parameters ( x ) , such that x ~ 𝒩 ( X , σ ) . The hyperpriors for σ are specified by a half-Cauchy ( Gelman , 2006 ) with a scale of 2 . The hyperpriors for X are centered around 0 ( with the exception of Xρ ) and weakly informative: Xρ ~ 𝒩 ( 2 , 3 ) , Xε , κ ~ 𝒩 ( 0 , 2 ) , Xb , π ~ 𝒩 ( 0 , 3 ) . All parameters are unconstrained , with the exception of ρ ( positivity constraint; exponential transform ) and ε ( [0 1] constraint; inverse logit transform ) . To ensure that the effect of κ on ε ( Equation 4 ) was symmetrical in model space ( i . e . after sigmoid transformation to ensure [0 1] constraint ) , ε was computed as: ( 5 ) ε= {ε0=inv . logit ( ε ) εpunished NoGo=inv . logit ( ε−κ ) εrewarded Go=ε0+ ( ε0−εpunished NoGo ) Model estimations were performed using Stan software in R ( RStan ) ( Stan Development Team , 2016 ) . Stan provides full Bayesian inference with Markov chain Monte Carlo ( MCMC ) sampling methods ( Metropolis et al . , 1953 ) . The number of Markov chains was set at 4 , with 200 burn-in iterations and 1000 post burn-in iterations per chains ( 4000 total ) . Model convergence was considered when the potential scale reduction factor R^ < 1 . 1 for all parameters ( Gelman and Rubin , 1992 ) . In case model convergence was not reached , both ( post ) burn-in samples were increased to 1500 . Not all models reached convergence at this point . Therefore , we repeated model estimation while excluding the subjects ( N = 5 ) for whom initially R^ > 1 . 1 in any one of the models , resulting in model convergence for all models . We report model evidence including all subjects in Appendix 5 , showing that model selection and parameter inference remains the same when excluding these subjects . Model comparison was evaluated using the Watanabe-Akaike Information Criteria ( WAIC ) ( Watanabe , 2010 ) . WAIC is an estimate of the likelihood of the data given the model parameters , penalized for the effective number of parameters to adjust for overfitting . Lower ( i . e . more negative ) WAIC values indicate better model fit . As WAIC is reported on the deviance scale ( Gelman et al . , 2014 ) , a difference in WAIC value of 2–6 is considered positive evidence , 6–10 strong evidence , and >10 very strong evidence ( Kass and Raftery , 1995 ) . Having established the mechanisms by which motivational valence may affect instrumental learning and activation , we extended the winning model to test which of these mechanisms are affected by MPH , putatively driven by a prolonged striatal presence of catecholamines ( dopamine ) following reward , due to reuptake inhibition by MPH . In M5 we tested whether MPH altered the Pavlovian response bias . This model includes a parameter allowing for an MPH-induced change in the Pavlovian weight ( πMPH ) : ( 6 ) π= {π0if placeboπ0+πMPHif MPH Next , we tested two mechanisms by which MPH might alter the bias in instrumental learning ( κ ) . In M5b we tested whether MPH simply enhanced or reduced the learning bias parameter , estimating an additive effect of κMPH-selective: ( 7 ) κ= {κ0if placeboκ0+κMPH−selectiveif MPH Alternatively , the prolonged presence of catecholamines following reward under MPH could induce a more diffuse credit assignment , rather than a selective learning bias effect . To test this hypothesis , in M5c we included a MPH-induced learning bias parameter ( κMPH-diffuse ) , which was used to update both Go responses , on all trials where any active Go response was followed by reward , in addition to the regular learning update for the chosen Go response: ( 8 ) if MPH , rt=1 , & achosen=Go:Qt ( achosenGo , t , st ) =Qt−1 ( achosenGo , t , st ) + ( ε+κ0+κMPH−diffuse ) ⋅PEQt ( aunchosenGo , t , st ) =Qt−1 ( aunchosenGo , t , st ) +κMPH−diffuse⋅PE Where PE is the prediction error following the rewarded Go response: PE=ρrt− Qt−1 ( at , st ) . Thus where κMPH-selective enhances the learning of the selected Go response after reward , κMPH-diffuse induces learning of all Go responses when a Go response elicited reward . To test whether MPH affected both the Pavlovian response bias and instrumental learning bias , M6 include πMPH parameter as well as the winning model of the two learning bias mechanisms ( M5c - κMPH-diffuse ) . For completeness , we report the composite model including the parameters πMPH and κMPH-selective in Appendix 5 . The hyperpriors are again centered around 0 and weakly informative: Xκmph ~ 𝒩 ( 0 , 2 ) and Xπmph ~ 𝒩 ( 0 , 3 ) , where only Xκmph-diffuse is constrained ( [0 1] constraint; inverse logit transform ) . Having established the winning model , we used two absolute model fit approaches to confirm that the winning model captures the effects of interest; the post-hoc absolute-fit approach ( also called one-step-ahead prediction ) and posterior predictive model simulation approach ( Steingroever and Wagenmakers , 2014 ) . The posterior predictive model simulations simply 'play' the task , using the estimated parameters . This approach , however , ignores sequential/history effects of actually observed choices and outcomes . The 'one-step-ahead' prediction fits parameters to trials t1 - tn-1 , and then predicts the choice on trial tn . Taking these sequential effects into account is particularly important to assess effects of the parameters that estimate the effect of previous choice/outcome combinations , i . e . the learning rate parameters , relative to the constant parameters like the Pavlovian and go biases . For both the one-step-ahead predictions and model simulations , we computed action probabilities for all subjects on all trials using the sampled combinations of all individual-level parameter estimates . For the one-step-ahead predictions the observed choices and outcomes were used to update the action probabilities . For the model simulations choices were simulated depending on the computed action probabilities , and outcomes were determined according to the ground-truth outcome probabilities ( i . e . a correct response would lead to the desired outcome 80% of the time ) . Subsequently , outcomes corresponding to the simulated choices were used to update the action probabilities . The one-step-ahead prediction and simulations were repeated for all sampled parameter combinations ( 4000 times ) , and action probabilities were averaged over repetitions . Averaging over repetitions also minimizes effects of randomness due to the stochastic nature of the choice simulation . | When we see a threat , we tend to hold back . When we see a reward , we have a strong urge to approach . Most of the time , these hardwired tendencies – or biases – are the right thing to do . However , our behaviour is not all hardwired; we can also learn from our previous experiences . But might this learning be biased too ? For example , we might be quicker to believe that an action led to a reward , because actions often do bring rewards . Conversely , we might be less likely to attribute a punishment to having held back , because holding back usually helps us to avoid punishments . Swart et al . have now tested whether rewards and punishments influence our actions solely via hardwired behavioural tendencies , or whether they also bias our learning . That is , are we biased to learn that taking action earns us rewards , while holding back spares us punishments ? Previous work has shown that chemical messengers in the brain called catecholamines help us to take action when we anticipate a reward . Swart et al . therefore also examined whether catecholamine levels contribute to any bias in learning . One hundred young healthy adults twice performed a task in which they could earn rewards and avoid losses by taking or withholding action . By using a mathematical model to work out what influenced the choices made by the volunteers , Swart et al . found that rewards and punishments did indeed bias learning . Moreover , this learning bias became stronger when the volunteers took methylphenidate ( also known as Ritalin ) , a drug that increases catecholamine levels and which is used to treat ADHD and narcolepsy . The volunteers varied markedly in how strongly methylphenidate affected their choices . This emphasises how important it is to account for differences between people when evaluating the effects of medication . Motivations are what get us going and keep us going . The findings of Swart et al . mean that we now have a better understanding of how motivations , such as desired rewards or unwanted punishments , influence our behaviour . A future challenge is to understand how we can overcome these motivations when they work against us , such as in addiction or obesity . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | Catecholaminergic challenge uncovers distinct Pavlovian and instrumental mechanisms of motivated (in)action |
The membrane-bound transcription factor ATF6α plays a cytoprotective role in the unfolded protein response ( UPR ) , required for cells to survive ER stress . Activation of ATF6α promotes cell survival in cancer models . We used cell-based screens to discover and develop Ceapins , a class of pyrazole amides , that block ATF6α signaling in response to ER stress . Ceapins sensitize cells to ER stress without impacting viability of unstressed cells . Ceapins are highly specific inhibitors of ATF6α signaling , not affecting signaling through the other branches of the UPR , or proteolytic processing of its close homolog ATF6β or SREBP ( a cholesterol-regulated transcription factor ) , both activated by the same proteases . Ceapins are first-in-class inhibitors that can be used to explore both the mechanism of activation of ATF6α and its role in pathological settings . The discovery of Ceapins now enables pharmacological modulation all three UPR branches either singly or in combination .
Most secreted and transmembrane proteins utilize the endoplasmic reticulum ( ER ) as a dedicated folding compartment . It is estimated that about one third of all newly synthesized proteins pass through the ER , where they fold and assemble into multi-subunit complexes , and where post-translational modifications such as disulfide bridge formation and glycosylation occur ( Braakman and Hebert , 2013 ) . Dedicated quality control mechanisms ensure that only properly folded proteins exit the ER . These mechanisms are part of the cell’s 'proteostasis network' and include chaperone systems to aid in protein folding and ER associated degradation ( ERAD ) to remove terminally misfolded proteins ( Ruggiano et al . , 2014 ) . In addition , the ER has the ability to adjust its folding capacity upon demand through activation of a homeostatic signaling network , called the unfolded protein response ( UPR ) . The UPR directs cell fate - cells that cannot restore homeostasis then initiate apoptosis to prevent secretion or cell surface presentation of misfolded , non-functional proteins ( Lin et al . , 2007; Lu et al . , 2014 ) . Three principal transmembrane sensors of the UPR independently monitor folding stress in the ER – IRE1 , PERK and ATF6 ( Gardner et al . , 2013 ) . These UPR branches function cooperatively to decrease the load of incoming polypeptides and to increase both the protein folding and degradative capacity of the ER through regulation of transcription and translation . ATF6 and IRE1 increase the folding capacity of the ER by upregulating transcription of UPR target genes . ATF6 is a membrane-tethered transcription factor activated by regulated trafficking and proteolysis producing ATF6-N , the ATF6 fragment that constitutes the functional transcription factor ( Haze et al . , 1999; Ye et al . , 2000; Chen et al . , 2002 ) . In contrast , the highly conserved kinase-endoribonuclease IRE1 removes an intron from the mRNA encoding the UPR effector XBP1 allowing translation of XBP1s ( 's' for spliced ) , the functional transcription factor variant of this protein ( Yoshida et al . , 2001 ) . ATF6-N and XBP1s bind to ER stress response ( ERSE ) ( Yoshida et al . , 1998; 2000; Roy and Lee , 1999 ) and unfolded protein response ( UPRE ) elements ( Yamamoto et al . , 2004 ) , respectively in the promoters of UPR target genes . ATF6 upregulates transcription of chaperones , foldases and lipid synthesis genes ( Wu et al . , 2007; Yamamoto et al . , 2007; Adachi et al . , 2008 ) , while XBP1 upregulates ER chaperones and the ERAD machinery ( Lee et al . , 2003; Acosta-Alvear et al . , 2007 ) . Decreasing the load of proteins entering the ER is coordinated by IRE1 and PERK . Regulated IRE1-dependent mRNA decay ( RIDD ) cleaves ER-targeted mRNAs leading to their degradation ( Hollien et al . , 2009 ) . PERK , a second transmembrane kinase , phosphorylates itself and the α-subunit of the initiation factor ( eIF2-α ) leading to transient inhibition of cap-dependent translation and an increase in translation of UPR effectors with upstream open reading frames ( Harding et al . , 1999; Sidrauski et al . , 2015 ) , including the transcription factor ATF4 and the apoptotic effector CHOP . The balance between survival and death is controlled temporally . Both IRE1 and ATF6 signaling attenuate after prolonged ER stress , removing most of the cytoprotective functions of the UPR ( Lin et al . , 2007; Lu et al . , 2014; Haze et al . , 2001 ) . PERK signaling is maintained and , through CHOP , commits the cell to apoptosis ( Zinszner et al . , 1998; Palam et al . , 2011 ) . This dual capacity of the UPR to boost the protein folding capacity or drive cell death has been implicated in many disease models ( Ryno et al . , 2013 ) . Many small molecule inhibitors and activators of the PERK and IRE1 enzymes have been isolated ( Maly and Papa , 2014; Maurel et al . , 2015; Mendez et al . , 2015 ) . In contrast , no pharmacological agents promoting the selective modulation of ATF6 have been developed , in part due to the fact that , unlike IRE1 and PERK , ATF6 is not an enzyme . Two closely related homologs define the ATF6 family of ER stress sensors; ATF6α and ATF6β , which respond to the same stress inducers and are activated with similar kinetics . ATF6α and ATF6β act redundantly during development as single knockout mice are viable and fertile while double knockout animals are pre-implantation lethal ( Yamamoto et al . , 2007 ) . Conversely , ATF6α and ATF6β do not appear to act redundantly during ER stress as ATF6α knockout cells or animals die when challenged with ER stressors ( Wu et al . , 2007; Yamamoto et al . , 2007 ) . The transcriptional targets of ATF6β and its role during ER stress remain poorly defined . Regulation of ATF6 signaling is by spatial separation of the substrate , ATF6 , in the ER and the proteases , site-1 and site-2 proteases ( S1P and S2P , respectively ) , in the Golgi apparatus: upon ER stress ATF6 moves from the ER to the Golgi apparatus . The mechanism by which ATF6 trafficking is regulated is poorly understood but the transport requires the COPII coat ( Nadanaka et al . , 2004; Schindler and Schekman , 2009 ) , which is not unique to ATF6 . Inhibition of ATF6 was achieved by inhibiting the proteases that release it from the membrane – S1P and S2P , respectively ( Ye et al . , 2000; Okada et al . , 2003 ) . These proteases do not uniquely process ATF6 but also play essential roles in regulating cholesterol homeostasis via processing of SREBP ( Brown and Goldstein , 1997 ) and lysosome biogenesis via processing of the α/β-subunit precursor of the N-acetylglucosamine-1-phosphotransferase complex ( Marschner et al . , 2011 ) . Their pleiotropic engagement limits the usefulness of S1P and S2P inhibitors for studies of the UPR . To date , the ATF6 signaling pathway was considered 'undruggable' . Here we developed cell-based screens to identify a series of pyrazole amides as the first selective inhibitors of the ATF6α branch of the UPR . We show in the accompanying manuscript that these compounds trap ATF6α in the ER in discrete foci , which inspired us to name the compounds 'Ceapins' from the Irish verb 'ceap' meaning 'to trap' ( Gallagher and Walter , 2016 ) . Ceapins do not inhibit activation of either IRE1 or PERK in response to ER stress , nor do they inhibit trafficking and cleavage of SREBP in response to low sterols or ATF6β in response to ER stress . Through structure-activity studies we increased the potency of the series ten-fold to an IC50 of 600 nM . Inhibition of ATF6α with Ceapin analogs has no toxicity in unstressed cells but increases the sensitivity of cells to ER stress inducers , closely mimicking the genetic ablation of ATF6α in mice . This makes Ceapins the most selective class of ATF6α inhibitors identified to date and the first to act through a mechanism distinct from protease inhibition or general trafficking between the ER and the Golgi .
To isolate small molecule modulators of ATF6 signaling , we used an assay based on the activation of transcription of ATF6 target genes . To this end , we cloned two copies of the ER stress response element ( ERSE ) upstream of a minimal promoter driving expression of luciferase ( Figure 1A ) into a retroviral vector , which we used to generate a HEK293T-based ERSE-luciferase reporter stable cell line . To induce ER stress , we treated the reporter cells with thapsigargin ( Tg ) , an inhibitor of the ER calcium pump . ER stress causes a 3 . 8 ± 0 . 2-fold induction of luciferase activity ( Figure 1B ) . The ER stress-induced luciferase was not affected by inhibition of IRE1 ( Figure 1—figure supplement 1 ) ( Patterson et al . , 2011 ) , indicating that the cell line reported selectively on the ATF6 branch . We screened 106 , 281 compounds at a single concentration for their ability to inhibit Tg-induced luciferase activity in the reporter cells ( Figure 1C , see also Figure 3—figure supplement 2 for overview of screen workflow ) . About 1% of the compounds ( 1142 ) showed >69% inhibition ( amounting to three standard deviations from the mean of stressed controls ) . To focus on ATF6 pathway specific modulators , we next removed from consideration compounds that showed inhibitory activity in analogous assays based on luciferase reporters induced by either IRE1-dependent mRNA splicing ( Figure 1—figure supplement 1 ) ( Mendez et al . , 2015 ) or PERK-dependent translational control ( Sidrauski et al . , 2013 ) . When fresh stocks were retested in dose response assays , all 598 remaining compounds showed inhibitory activity with an IC50 < 13 . 5 μM against Tg-induced luciferase activity in the reporter cells . 10 . 7554/eLife . 11878 . 003Figure 1 . Isolation of small molecule inhibitors of ATF6 mediated transcriptional response induced by ER stress . ( A ) Schematic representation of ERSE-luciferase construct used to make screening cell line . Two copies of the ER Stress Response Element ( ERSE ) were cloned in front of a minimal CMV promoter ( MCP ) driving expression of luciferase . ( B ) Luciferase activity is induced 3 . 84 ± 0 . 16 fold upon ER stress ( 100 nM Tg ) in ERSE-Luciferase 293T reporter cell line . Mean of three independent experiments with at least duplicate wells is plotted; error bars are standard error of the mean . ( C ) Primary screen data from ERSE-luciferase transcriptional reporter cell line . Each plate was internally normalized from 0–100% inhibition using stressed and unstressed controls respectively . 106 , 281 compounds were added in combination with ER stressor and assayed for their ability to inhibit stress-induced production of ERSE-luciferase . Plot shows % inhibition for each control and compound tested - blue lines denote mean and standard deviation of each population , black dots indicate those wells more than two standard deviations away from the mean of the population . 1142 compounds scoring more than three standard deviations from the mean ( >69% inhibition , orange line ) were classified as hits . ( D–J ) . 293 cells expressing doxycycline inducible MPZ-GFP were uninduced ( D ) or induced with 50 nM doxycycline without ( E ) or with inhibitors ( F–I ) for seven hours and then fixed and stained for GFP ( green ) , actin ( red ) and DNA ( blue ) . Inhibitors tested were the protein synthesis inhibitor cycloheximide at either 0 . 01 μg/mL ( F ) or 0 . 1 μg/mL ( G ) , the ER stressor thapsigargin ( 100 nM , H ) or the S1P inhibitor ( 50 μM Pf-429242 , I ) . ( J ) Mean induction of GFP per cell per image was quantified and plotted as fold induction relative to uninduced controls . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 00310 . 7554/eLife . 11878 . 004Figure 1—figure supplement 1 . IRE1 inhibitor blocks induction of luciferase activity through XBP-luciferase but not ERSE-luciferase . 293T reporter cell lines expressing either ERSE-luciferase or XBP-luciferase were treated with either vehicle ( DMSO ) or ER stressor ( 100 nM Tg ) without or with 10 μM IRE1 inhibitor for nine hours . Mean of duplicate experiments each with duplicate wells is plotted , error bars are standard error of the mean . For each cell line , ER stress ± IRE1 inhibitor were compared using two-tailed t-test – n . s non-significant , *p = 0 . 013 . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 004 We further triaged the reconfirmed compounds to remove non-specific inhibitors of transcription or translation . To this end , we developed a high-throughput assay to determine if the compounds blocked expression of green fluorescent protein-tagged myelin protein zero ( MPZ-GFP ) under the control of a doxycycline-inducible promoter ( Figure 1D–I ) . Addition of doxycycline induced the MPZ-GFP reporter 2 . 6 ± 0 . 06-fold ( Figure 1E ) compared to uninduced controls ( Figure 1D ) . As expected , the translation inhibitor cycloheximide ( 'CHX' , Figure 1F , G ) prevented MPZ-GFP expression , and treatment of the cells with an ER stressor ( Figure 1H ) or with an S1P inhibitor ( Figure 1I ) ( Hay et al . , 2007; Hawkins et al . , 2008 ) did not block doxycycline-induced MPZ-GFP expression . We quantified the fold-induction of MPZ-GFP by doxycycline ( Figure 1J ) . Sixty of the 598 compounds isolated in the primary screen inhibited doxycycline-induced MPZ-GFP expression and were removed from further consideration , yielding a collection of 538 compounds that inhibited ER stress-induced ATF6 signaling without inhibiting either transcription or translation or inhibiting signaling through other UPR branches . Each step of ATF6α activation occurs in a different organelle – stress-sensing in the ER , proteolytic processing in the Golgi and activation of transcription in the nucleus . To begin mapping the action of the inhibitors to the steps of ATF6 activation , we next determined the subcellular localization of ATF6α using a cell line that stably expresses GFP-ATF6α ( Figure 2A–D ) . Using an antibody against GRP94 to mark the ER and a DNA stain to mark the nucleus , we examined the ratio of GFP intensity between the nucleus and the ER . In unstressed cells , GFP-ATF6α colocalized with GRP94 , indicating its predominant localization in the ER ( Figure 2A ) . Upon ER stress , GFP-ATF6α translocated to the nucleus and colocalized with DNA ( Figure 2B , yellow in merged image ) . As a positive control , when we inhibited ATF6 cleavage by S1P , GFP-ATF6α no longer translocated to the nucleus but accumulated in perinuclear punctae , corresponding to the Golgi apparatus ( Figure 2C , see also Figure 3 in Gallagher and Walter ( 2016 ) ) . We classified compounds that decreased or inhibited nuclear translocation as 'Class 1 inhibitors' . Figure 2D shows Ceapin-A1 as an example in this class showing decreased nuclear GFP signal and perinuclear GFP-ATF6α punctae . We classified compounds that allowed nuclear translocation but inhibited reporter transcription as 'Class 2 inhibitors' . By our definition , Class 2 inhibitors act downstream of ER-Golgi trafficking , proteolysis , and nuclear import of ATF6 . These inhibitors may act by preventing DNA binding or interaction with transcriptional co-activators , such as NF-Y ( Yoshida et al . , 2000; Li et al . , 2000 ) . 10 . 7554/eLife . 11878 . 005Figure 2 . Isolation of small molecule inhibitors of ER stress induced nuclear translocation of ATF6 . ( A–D ) Nuclear translocation assay in U2-OS GFP-ATF6α cells . U2-OS cells expressing GFP-ATF6α were treated with either vehicle ( unstressed , DMSO , A ) or ER stressor in the absence ( ER stress , 100 nM Tg , B ) or presence of S1P inhibitor ( 20 μM Pf-429242 , C ) or screen hit ( 6 . 6 μM Ceapin-A1 , D ) . After five hours , cells were fixed and stained for GFP ( green ) , GRP94 ( ER marker , blue ) and DNA ( nuclear marker , DAPI , red ) . ( E–F ) Quantification of nuclear translocation assay . ( E ) Single cell ratios of nuclear: ER GFP intensity were calculated for four images per well for each treatment ( unstressed and stressed control wells are present seven times per plate ) and plotted as histograms . For each plate , the minimum nuclear: ER ratio where the percentage of stressed cells is greater than the percentage unstressed cells is calculated and annotated as the threshold for activation by ER stress ( light grey vertical dashed line ) . ( F ) For each plate , the percent activation by ER stress is calculated for the control wells ( unstressed n = 1904 , ER stress n = 2095 , unstressed + S1P inhibitor n = 366 , stressed + S1P inhibitor n = 330 cells ) and used to generate a Z’ score for the plate . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 00510 . 7554/eLife . 11878 . 006Figure 2—figure supplement 1 . Annotation of nuclear and ER regions used for calculation of the ratio of nuclear to ER GFP-ATF6α signal for each cell . ( A–E ) Calculation of nuclear: ER ratio of GFP intensity for nuclear translocation assay using CellProfiler . For each individual cell , nuclear DNA stain ( DAPI , A ) and ER staining ( anti-GRP94 , B ) were used to generate outlines of the nucleus and ER respectively . These outlines were overlaid on the corresponding GFP-ATF6α image ( C ) and used to quantify the intensity of GFP signal in the nucleus ( D ) and ER ( E ) of each single cell in each image . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 00610 . 7554/eLife . 11878 . 007Figure 2—figure supplement 2 . Example of a heat map for a plate from the nuclear translocation assay secondary screen showing percent inhibition of test compounds compared to controls . Example of the heat map generated for each plate showing the percent inhibition of each compound tested . In this presentation , wells in column 1 and 12 represent stressed and unstressed cells , respectively , with H1 and H12 containing the S1P inhibitor . Columns 2–11 contain stressed cells treated with test compounds . 0% and 100% inhibition are set by the controls - unstressed wells ( 100% , vehicle , A12-G12 ) and stressed wells ( 0% , A1-G1 ) respectively . S1P inhibitor alone ( H12 ) or in combination with ER stressor ( H1 ) was used as the positive control in each plate . Wells G8 , F6 , and A4 represent the top three hits on this plate . Wells containing less than 50% the number of cells of ER stressed controls were annotated as toxic with white X signs – wells E5 , G5 and G10 . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 00710 . 7554/eLife . 11878 . 008Figure 2—figure supplement 3 . Ceapin-A1 inhibits nuclear translocation of GFP-ATF6 . ( A ) Single cell ratios of nuclear: ER GFP intensity from four images per well for Ceapin-A1 , unstressed and stressed controls from a single plate are plotted as histograms . ( B ) The percent activation by ER stress is for the control wells ( unstressed n = 1904 , ER stress n = 2095 , unstressed + S1P inhibitor n = 366 , stressed + S1P inhibitor n = 330 cells ) and for Ceapin-A1 ( stressed + Ceapin-A1 n = 256 cells ) from a single plate . From three independent plates , nuclear translocation of GFP-ATF6α by Ceapin-A1 was inhibited by 35 . 07 ± 18 . 15% . Note that in this assay , compounds were screened at 6 . 6 μM , which is below the IC50 for the Ceapin-A1 stock in the library plate ( 8 . 49 μM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 00810 . 7554/eLife . 11878 . 009Figure 2—figure supplement 4 . Combining data from high content assays identified non-specific inhibitors of trafficking and toxic compounds . ( A–E ) Nuclear translocation assay showing unstressed ( A ) and ER stressed ( B ) GFP-ATF6α U2-OS cells . GFP-ATF6α is green , GRP94 is blue and DNA is in red . Two compounds with similar scaffold ( 47304 , C and 50243 , D ) scored as inhibitors of ATF6 processing . ( E ) 153970 is an example of a toxic compound . ( F–J ) Inducible GFP assay showing uninduced ( F ) and induced ( G ) MPZ-GFP T-Rex cells . MPZ-GFP is in green , actin is in red , DNA is in blue . Both compounds identified as inhibitors of ATF6 processing also show defects in trafficking of MPZ-GFP from the ER to the plasma membrane ( 47304 , H and 50243 , I ) and are unlikely to be selective inhibitors of ATF6 . ( J ) 153970 is also toxic to this cell line . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 00910 . 7554/eLife . 11878 . 010Figure 3 . Isolation of Ceapin-A1 , a small molecule inhibitor of ATF6 but not SREBP processing . ( A ) ERSE-luciferase assay in HEK293T cells . Cells were treated without ( DMSO ) or with ER stressor ( 100 nM Tg ) in the presence or absence of inhibitors for nine hours . Increasing concentrations of either S1P inhibitor ( Pf-429242 , red ) or Ceapin-A1 ( green ) but not IRE1 inhibitor ( 4 μ8C , grey ) block ER stress-induced luciferase activity . Plotted is one representative experiment showing mean and standard deviation for each inhibitor concentration ( triplicate wells per point ) . Dashed grey lines indicate the relative luciferase activity of unstressed and stressed controls . ( B ) ER stress induced upregulation of the endogenous ATF6α target gene GRP78 in U2-OS cells . Cells were treated without ( DMSO , open circles ) or with ER stress ( 100 nM Tg , black squares ) in the absence or presence of inhibitors for four hours prior to isolation of mRNA . Upregulation of GRP78 mRNA was measured using qPCR . mRNA levels for GRP78 were normalized to GAPDH for each well and then compared to unstressed and stressed controls . ER stress induced GRP78 mRNA induction is inhibited by co-incubation with either S1P inhibitor ( 2 . 3 μM Pf-429424 , red ) or Ceapin-A1 ( 10 μM , green ) but not the inactive Ceapin analog A5 ( 10 μM , blue ) . Inhibition of the ISR ( 200 nM or 400 nM ISRIB , orange ) partially inhibits GRP78 induction while inhibition of IRE1 ( 10 μM 4 μ8C , grey ) has only minor effects . Data plotted is the mean percent activation of GRP78 transcription relative to unstressed ( 0% ) and stressed ( 100% ) controls from two or three independent experiments , each with duplicate reactions carried out on duplicate wells . ( C ) Induction of SREPB processing by lipoprotein depletion in HeLa cells . HeLa cells were grown in lipoprotein deficient media for 16 . 5 hr prior to addition of either sterols or inhibitors for five hours . One hour prior to lysis proteasome inhibitor ( 25 μg/mL ALLN ) was added to prevent the degradation of the cleaved SREBP-N fragment . Whole cell lysates were analyzed by Western blotting for SREPB1 and GAPDH . Arrowheads denote positions of full-length ( SREBP ) and cleaved ( SREBP-N ) variants of SREBP1 . Lipoprotein depletion induces cleavage of SREBP ( lanes 1 , 11 ) that is inhibited by addition of sterols ( 10 μg/mL cholesterol , 1 μg/mL 25-hydroxycholesterol , lanes 2 , 12 ) or increasing concentrations of a S1P inhibitor ( Pf-429242 , lanes 7–10 ) but not increasing concentrations of Ceapin-A1 ( lanes 3–6 ) . Data shown is representative of three independent experiments . ( D ) Induction of ATF6α processing by ER stress in T-Rex cells expressing FLAG-tagged ATF6α . Arrowheads denote positions of full-length ( ATF6α ) , cleaved membrane-bound ( ATF6α-M ) and cleaved nuclear ( ATF6α-N ) variants of ATF6 . Cells were treated without ( lanes 1 , 6 , 11 ) or with ( lanes 2 , 7 , 12 ) ER stressor ( 100 nM Tg ) alone or in combination with either S1P inhibitor ( 0 . 75 μM Pf-429242 , lanes 3 , 4 , 8 , 9 , 13 , 14 ) or Ceapin-A1 ( 14 . 95 μM , lanes 5 , 10 , 15 ) for two hours prior to harvesting . One hour prior to lysis proteasome inhibitor ( MG132 , 10 μM ) was added . Cells were harvested and separated by centrifugation into total , membrane and nuclear fractions and analyzed by Western blot for ATF6α ( anti-FLAG ) , PERK ( membrane control ) , ATF4 ( nuclear control ) and GAPDH ( loading control ) . Note that totals were run on 10% gels while membrane and nuclear fractions were run on gradient gels to visualize the migration differences between ATF6α-N and ATF6α-M and between PERK and phosphorylated PERK respectively . Data shown is representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 01010 . 7554/eLife . 11878 . 011Figure 3—figure supplement 1 . Identification of Ceapin-A1 , a small molecule that inhibits ATF6 processing in response to ER stress . ( A–B ) qPCR analysis of endogenous ATF6α target genes GRP78 ( A ) and HERPUD1 ( B ) . U2-OS cells were treated without ( DMSO ) or with ER stressor ( 100 nM Tg ) alone or in combination with ATF6 processing hits ( either 10 x IC50 or 10 μM , depending on solubility ) for four hours prior to harvesting and mRNA extraction . Fold induction of each target gene is calculated relative to unstressed controls . Dashed orange lines denote 50% inhibition . Data plotted are means and standard deviations from duplicate reactions from duplicate wells . It is unclear why so many compounds failed at this point having passed through earlier filters . These results validate the effects of the Ceapin-A1 on endogenous gene expression and so based on these results we focused exclusively on the Ceapin series . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 01110 . 7554/eLife . 11878 . 012Figure 3—figure supplement 2 . Screening workflow Summary of screening workflow that lead to the identification of Ceapin-A1 consisting of primary ( yellow ) , secondary ( orange ) and tertiary ( green ) screens . The primary screen was run on the entire library , the secondary screens on cherry-picked compounds from the primary screen and the tertiary screens on repurchased material . At each stage compounds were triaged and filtered according to the criteria described . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 01210 . 7554/eLife . 11878 . 013Figure 3—figure supplement 3 . Ceapin-A1 inhibits ER stress induced ERSE-luciferase activity ERSE-luciferase assay in HEK293T cells . Cells were treated without ( DMSO ) or with ER stressor ( 2 μg/mL Tm ) in the presence or absence of inhibitors for ten hours . Increasing concentrations of either S1P inhibitor ( Pf-429242 , red ) or Ceapin-A1 ( green ) but not IRE1 inhibitor ( 4μ8C , grey ) block ER stress-induced luciferase activity . Plotted is a representative experiment showing mean and standard deviation for each inhibitor concentration ( triplicate wells per point ) . Dashed grey lines indicate the relative luciferase activity in unstressed and stressed controls . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 01310 . 7554/eLife . 11878 . 014Figure 3—figure supplement 4 . Mutation of S1P cleavage site in ATF6α leads to production of ATF6α-M upon ER stress . Induction of ATF6α processing by ER stress in T-Rex cells expressing FLAG-tagged ATF6α either wild-type or with S1P cleavage site mutated ( R416A ) . Arrowheads denote positions of full-length ( ATF6α-FL ) , cleaved membrane bound ( ATF6α-M ) and cleaved nuclear ( ATF6α-N ) variants of ATF6α . Cells were treated without ( lanes 1 , 7 ) or with ( lanes 2 , 8 ) ER stressor ( 100 nM Tg ) alone or in combination with either S1P inhibitor ( 0 . 75 μM Pf-429242 , lanes 3 , 4 ) or Ceapin-A1 ( 15 μM , lanes 5 , 6 ) for two hours prior to harvesting . Cells were harvested and analyzed by Western blot for ATF6 ( anti-FLAG ) . White line indicates where intervening lanes have been removed . Data shown is representative of duplicate experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 014 To quantify GFP-ATF6α localization , we defined a threshold for activated cells – i . e . cells that responded to ER stress and show nuclear translocation of GFP-ATF6 . Using CellProfiler ( Carpenter et al . , 2006 ) , we used the GRP94 and DNA images to generate masks for the ER and nuclei , respectively ( Figure 2—figure supplement 1 ) . We next calculated the ratio of nuclear to ER GFP signal ( nuc:ER ratio ) for each cell and plotted the nuc:ER ratios for the unstressed and stressed cells as histograms ( Figure 2E ) . The distributions showed a wide range of responses within the population of cells in each well . To compare wells treated with different inhibitors , we extracted from these distributions a single metric representing the percentage of stressed cells for each well , as described in the Methods . This allowed us to convert single cell measurements of GFP-ATF6α from the images to compare across plates . We performed these analyses in biological triplicates . Statistical evaluation of unstressed and stressed controls yielded a mean Z’ of 0 . 7 +/- 0 . 18 , which is an exceptionally robust read-out for a cell- and image-based high-throughput assay . We further validated the analyses using S1P inhibitor as a positive control ( Figure 2E and F ) . An example of a heat map for one plate assessing the inhibitory activity of different compounds is shown in Figure 2—figure supplement 2 . An example of a hit from this assay , Ceapin-A1 is shown in Figure 2—figure supplement 3 . Analysis of the data using t-tests instead of the threshold method gave the same results , indicating that thresholding did not introduce a bias into the data analysis . Of 598 compounds tested in this way , 85 showed robust inhibition of ER stress-induced nuclear translocation of GFP- ATF6α above three standard deviations from the mean of the negative ( Tg-alone ) control . We further triaged top-scoring compounds to remove false positives . Toxic compounds scored as hits in the ERSE-luciferase assay but identified in image-based assays as compounds that reduced the cell number per well below 50% of stressed controls ( e . g . , wells E5 , G5 and G10 in Figure 2—figure supplement 2; sample images in Figure 2—figure supplement 4E and J ) . They were removed from further consideration . Likewise , we removed 18 compounds that inhibited nuclear translocation of GFP- ATF6α by globally inhibiting protein trafficking from the ER by examining the image data from our MPZ-GFP assay ( Figure 2—figure supplement 4C , D , H and I ) . MPZ-GFP is targeted to the ER where it is folded prior to export to the plasma membrane ( Figure 2—figure supplement 4G ) ( Pennuto et al . , 2008 ) . Furthermore , we applied a potency cut-off of IC50 < 5 μM to Class 2 ( transcriptional ) inhibitors . We next analyzed the chemical structures of the compounds to remove pan-assay interference compounds ( PAINS ) ( Baell and Holloway , 2010; Dahlin et al . , 2015 ) . After completion of these assays , we were left with 38 Class 1 and 128 Class 2 inhibitors , of which 29 and 70 were repurchased . We performed dose-response ERSE-luciferase assays on the repurchased compounds using two different ER stressors – Tg or tunicamycin ( Tm ) . Tm inhibits N-linked glycosylation and induces the UPR . Of 99 repurchased compounds , 98 showed inhibitory activity with Tg , of which 82 also inhibited tunicamycin induced ERSE-luciferase . We next assessed compounds displaying well-shaped dose-response curves and IC50’s < 19 μM in both Tg and Tm ERSE-luciferase assays for their ability to inhibit induction of endogenous ATF6α target genes , GRP78 ( encoding the ER HSP70 BiP ) and HERPUD1 ( encoding a ubiquitin-domain-containing protein involved in ERAD ) ( Figure 3A , and Figure 3—figure supplement 1 ) ( Wu et al . , 2007; Yamamoto et al . , 2007; Adachi et al . , 2008 ) . This qPCR assay did not rely on the presence of reporters and proved the most stringent ( a workflow for the screen is shown in Figure 3—figure supplement 2 ) . Of the Class 1 inhibitors , only one consistently blocked ATF6α target gene upregulation: Ceapin-A1 ( 'A1' standing for Analog 1 the founding compound in the series described below ) . Ceapin-A1 inhibited both ERSE-luciferase induction by both Tg ( Figure 3A ) and Tm ( Figure 3—figure supplement 3 ) and induction of GRP78 mRNA ( Figure 3B ) with an IC50 = 4 . 7 ± 1 . 1 μM and to the same extent as the S1P inhibitor . Inhibition of IRE1 ( Cross et al . , 2012 ) had little effect in the same assays ( Figure 3A and B ) . In contrast , inhibition of the integrated stress response ( ISR ) ( Sidrauski et al . , 2013 ) showed a decrease in ATF6α target gene induction , as previously reported for PERK knockout cells ( Wu et al . , 2007 ) ( Figure 3B ) . By these criteria , Ceapin-A1 behaved like a selective inhibitor of ATF6α . Since Ceapin-A1 treatment of stressed cells resulted in the same punctate localization of GFP- ATF6α as the S1P inhibitor in the nuclear translocation assay ( Figure 2C and D ) , we next investigated if Ceapin-A1 was in fact an S1P inhibitor . To this end , we monitored the processing of endogenous SREBP1 in HeLa cells ( Figure 3C ) . Cells grown in lipoprotein-deficient media activate SREBP processing , indicated by the presence of faster migrating , cleaved SREBP-N ( Figure 3C , lanes 1 and 11 ) . This cleavage was blocked by the addition of either sterols ( Figure 3C , lanes 2 and 12 ) or increasing concentrations of the S1P inhibitor ( Figure 3C , lanes 7–10 ) to the cell culture media . Addition of Ceapin-A1 to 25 μM ( > five times its IC50 ) had no effect on lipoprotein depletion-mediated SREBP processing ( Figure 3C , lanes 3–6 ) . Therefore , Ceapin-A1 neither inhibits S1P nor S2P and emerges as the first small molecule inhibitor of ATF6 that does not inhibit SREBP or other pathways known to use these proteases . We then analyzed the cleavage of ATF6α directly , using a stable HEK293 based cell line that expressed FLAG-tagged ATF6α from an inducible promoter . Induction of ER stress led to cleavage of full-length ATF6α to produce the faster migrating , active transcription factor ATF6α-N ( Figure 3D , lanes 1 and 2 ) . Surprisingly , in the presence of the S1P inhibitor , ER stress-induced cleavage of ATF6α was not blocked ( Figure 3D , lanes 3 and 4 ) . This was unexpected , given that inhibition of S1P prevented both nuclear translocation of GFP- ATF6α ( Figure 2C and F ) and upregulation of ATF6α target genes ( Figure 3A and B , and Figure 3—figure supplement 1 and 3 ) . As a first clue on how to resolve the paradox , we observed that the cleavage product , henceforth referred to as ATF6α-M , produced in the presence of the S1P inhibitor migrated more slowly on SDS polyacrylamide gels than the product ATF6α-N produced by ER stress alone . We next analyzed the subcellular localization of ATF6α-M using differential centrifugation . In contrast to ER stressed cells , where ATF6α was recovered in the membrane fraction ( Figure 3D , lane 12 ) and ATF6α-N is in the nuclear fraction ( Figure 3D , lane 7 ) , both ATF6α and ATF6α-M were recovered in in the membrane fraction in cells treated with ER stress inducer and the S1P inhibitor ( Figure 3D , lane 14 ) . We therefore denoted this fragment as ATF6α-M , since it remains membrane-bound and as such is incapable of entering the nucleus and activating ATF6α target genes . The transmembrane protein PERK and the transcription factor ATF4 served as controls for membrane and nuclear fractions respectively ( Figure 3D ) . We observed the same fragment in cells expressing a variant of ATF6α , in which the S1P cleavage site was mutated ( Figure 3—figure supplement 4 ) , indicating that an alternate protease generated ATF6α-M and that ATF6α-M was not a substrate for subsequent membrane-release by S2P cleavage . Importantly , cells treated with both ER stress inducer and Ceapin-A1 did not show any ATF6α cleavage product ( Figure 3D , lanes 5 , 10 , 15 ) . This difference strongly suggested that the mechanism of inhibition of ATF6α processing by Ceapin-A1 is not via protease inhibition , but that ATF6α is trapped in a place or state where it cannot be cleaved . For this reason , we chose to name the chemical scaffold of the inhibitor 'Ceapin' , after the Irish verb ‘ceap’ meaning to trap . The original hit Ceapin-A1 is an N-benzyl pyrazol-4-yl amide ( Figure 4A ) . We used the ERSE-luciferase assay to guide structure-activity relationship ( SAR ) studies aimed at defining the essential pharmacophore within the Ceapin scaffold . We found that all four rings present in Ceapin-A1 are necessary for activity ( Figure 4B ) . Thus , deletion of the furan ring ( as in Ceapin-A2 ) led to a loss of detectable activity in the ERSE-luciferase assay . Among several aryl and heteroaryl rings examined in place of the furan , only a simple phenyl ring as in Ceapin-A3 ( IC50 6 . 9 ± 0 . 7 ) afforded activity comparable to Ceapin-A1 ( IC50 = 4 . 9 ± 1 . 2 μM ) . The benzyl substituent on the pyrazole ring was similarly sensitive to modification . Thus , analogs lacking the ortho and para chloro substituents ( Ceapin-A4 ) or bearing methyl in place of the benzyl group ( Ceapin-A5 ) were inactive ( Figure 4B and D ) . Further SAR of ring substitution in the benzyl side chain revealed that substitution at both the ortho and para position with spheroid hydrophobes was optimal for activity ( Figure 4C ) . Thus , the 2 , 4-dibromobenzyl analog Ceapin-A6 was about two-fold more potent than Ceapin-A1 , while the corresponding bis-trifluoromethyl congener Ceapin-A7 ( IC50 = 0 . 59 ± 0 . 17 μM ) was approximately ten-fold more potent than Ceapin-A1 ( Figure 4C and D ) . In contrast , analogs bearing a single trifluoromethyl group at either the para ( Ceapin-A8 ) or ortho position ( Ceapin-A9 ) were notably less potent than Ceapin-A7 . Additional studies were performed with Ceapin-A7 , representing an improved lead from the Ceapin series and with the inactive Ceapin analog A5 as a negative control . 10 . 7554/eLife . 11878 . 015Figure 4 . SAR studies of Ceapin-A1 improve potency by an order of magnitude . ( A–C ) Summary of structure activity relationship for Ceapin analogs . ( A ) Chemical structure of the initial screen hit , Ceapin-A1 . SAR of rings ( B ) or substituents on bis-substituted phenyl ring ( C ) of Ceapin scaffold . IC50 values were obtained using ERSE-luciferase assay in 293T cells where compounds were tested in dose-response in combination with ER stressor ( 100 nM Tg ) . IC50 values are from at least four independent experiments for each compound . ( D ) ERSE-luciferase assay showing improved potency of Ceapin-A7 ( purple ) and lack of activity of Ceapin-A5 ( blue ) compared to Ceapin-A1 ( green ) . 293T cells with stably integrated ERSE-luciferase reporter were treated with ER stressor ( 100 nM Tg ) and increasing concentrations of Ceapin analogs for nine hours prior to reading luciferase activity . Data plotted are mean values from a representative experiment with each point done in triplicate , error bars are standard deviation . ( E ) qPCR analysis of endogenous ATF6α target genes . U2-OS cells were treated without or with ER stressor in the presence of increasing concentrations of Ceapin analogs for four hours prior to harvesting of mRNA for qPCR analysis . mRNA levels for GRP78 , HERPUD1 and ERO1B were normalized to GAPDH for each well and then compared to unstressed controls . Data plotted are mean IC50 values calculated from duplicate experiments , each with duplicate qPCR reactions from duplicate wells for S1P inhibitor ( Pf-429242 , red ) , Ceapin-A1 ( green ) and Ceapin-A7 ( purple ) . Error bars are standard deviation . ( F ) Calculated mean IC50 values and standard deviations from qPCR analysis described above . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 015 We next confirmed the potency of Ceapin analogs on endogenous ATF6α target induction using qPCR analysis of GRP78 , HERPUD1 , and ERO1B , an ER oxidoreductase ( Wu et al . , 2007; Yamamoto et al . , 2007; Adachi et al . , 2008 ) ( Figure 4E ) . The mean IC50 values for the S1P inhibitor , Ceapin-A1 and Ceapin-A7 calculated from dose-response qPCR assays correlated well with the values obtained using the ERSE-luciferase reporter ( Figure 4F ) , validating both the analogs and the use of the ERSE-luciferase assay for SAR studies . As expected , the inactive Ceapin analog A5 does not inhibit induction of these targets ( Figure 3B ) . The increased IC50 values for HERPUD1 reflect the contribution from the IRE1/XBP1 and PERK/ATF4 branches to transcriptional upregulation of this target ( Lee et al . , 2003; Ma and Hendershot , 2004 ) , further underscoring the selectivity of Ceapin for the ATF6 branch of the UPR . Inhibition of IRE1 did not significantly alter ATF6 signaling as seen using ERSE-luciferase or GRP78 mRNA induction upon ER stress ( Figure 3A and B ) . We next validated that Ceapins selectively inhibit the ATF6 over the IRE1 and PERK branches of the UPR monitoring the activation of each UPR branch directly . To this end , we used a polyclonal antibody against ATF6α developed by the Mori lab ( Haze et al . , 1999 ) , which allowed us to look at endogenous ATF6α in U2-OS cells ( Figure 5A ) . As expected , treatment of cells with Tm produced both a faster migrating unglycosylated form of ATF6α ( ATF6α* ) and cleaved ATF6α-N ( Figure 5A , compare lanes 1 and 2 ) . Cells treated with both Tm and Ceapin-A7 contained the unglycosylated form of ATF6α but not ATF6α-N ( Figure 5A , lane 3 ) , indicating that despite the accumulation of unglycosylated proteins , ATF6α was not cleaved . ATF6α–derived bands from cells treated with Tm and the inactive Ceapin analog A5 were identical to Tm alone ( Figure 5A , lane 4 ) . ATF6α–derived bands from cells treated with Tg alone or in combination with the inactive Ceapin analog A5 ( inactive analog ) showed both ATF6α and ATF6α-N ( Figure 5A , lanes 6 and 8 ) while cells treated with both Tg and active Ceapin-A7 were indistinguishable from those derived from unstressed cells ( Figure 5A , lanes 5 , 7 ) . Induction of ER stress using either Tg or Tm led to upregulation of BiP protein levels ( Figure 5B , compare lanes 2 and 6 to lanes 1 and 5 ) . Consistent with our qPCR analysis of its mRNA levels ( GRP78 mRNA ) , Ceapin-A7 but not the inactive Ceapin analog A5 inhibited ER stress-induced upregulation of BiP ( Figure 5B , compare lanes 3 and 7 to lanes 4 and 8 ) . Thus consistent with our analyses above , Ceapins inhibit cleavage and functional activation of endogenous ATF6α in response to ER stress . 10 . 7554/eLife . 11878 . 016Figure 5 . Ceapins are selective inhibitors of the ATF6α branch of the UPR . ( A ) U2-OS cells were treated without ( DMSO ) or with ER stressor ( 100 nM Tg or 2 . 5 μg/mL Tm ) in the absence or presence of Ceapin analogs ( 6 μM each ) for two hours . One hour prior to lysis , proteasomal inhibitor ( 10 μM MG132 ) was added to prevent the degradation of the cleaved ATF6α-N fragment . Cells were harvested and analyzed by Western Blot for ATF6α , PERK , XBP1 and GAPDH ( loading control ) . Arrowheads denote the positions of full-length ( ATF6α ) , unglycosylated full-length ( ATF6α* ) and cleaved ( ATF6α-N ) variants of ATF6α and also PERK and phospho-PERK . Data shown is representative of three independent experiments . ( B ) U2-OS cells were treated without ( DMSO ) or with ER stressor ( 100 nM Tg or 2 . 5 μg/mL Tm ) in the absence or presence of Ceapin analogs ( 6 μM each ) . After eight hours , cells were harvested and analyzed by Western Blot for BiP and GAPDH ( loading control ) . ( C ) U2-OS cells were treated without ( DMSO ) or with ER stressor ( 100 nM Tg , black ) in the absence or presence of ten-fold the IC50 of either S1P inhibitor ( Pf-429242 , 3 . 2 μM , red ) or Ceapin-A1 ( 35 . 7 μM , green ) . Four hours later cells were harvested and mRNA extracted . mRNA levels for XBP1s or DDIT3 were normalized to GAPDH for each well and then compared to unstressed controls . Data plotted are from duplicate qPCR reactions from duplicate wells , error bars are standard deviation . ( D ) U2-OS cells were treated without ( DMSO ) or with ER stressor ( 100 nM Tg or 2 . 5 μg/mL Tm ) in the absence or presence of increasing concentration of Ceapin-A7 ( 0 . 6 , 1 . 89 , 6 , 18 . 9 μM ) or S1P inhibitor ( 5 μM Pf-429242 ) for four and a half hours . One hour prior to lysis , proteasomal inhibitor ( 10 μM MG132 ) was added to prevent the degradation of the cleaved ATF6α-N and ATF6β-N fragments . Cells were harvested and analyzed by Western Blot for ATF6α , ATF6β and GAPDH ( loading control ) . Arrowheads denote the positions of full-length ( ATF6α , ATF6β ) , unglycosylated full-length ( ATF6α* , ATF6β* ) , cleaved membrane bound ( ATF6α-M ) and cleaved ( ATF6α-N , ATF6β-N ) variants of ATF6α and ATF6β . Data shown is representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 01610 . 7554/eLife . 11878 . 017Figure 5—figure supplement 1 . Induction of XBP1s and DDIT3 mRNA is only partially inhibited by either the S1P inhibitor or Ceapin-A1 U2-OS cells were treated without ( DMSO ) or with ER stressor ( 2 . 0 μg/mL Tm , black ) in the absence or presence of ten-fold the IC50 of either S1P inhibitor ( Pf-429242 , 3 . 2 μM , red ) or Ceapin-A1 ( 35 . 7 μM , green ) . Four hours later cells were harvested and mRNA extracted . mRNA levels for XBP1s or DDIT3 were normalized to GAPDH for each well and then compared to unstressed controls . Data plotted are from duplicate qPCR reactions from duplicate wells , error bars are standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 01710 . 7554/eLife . 11878 . 018Figure 5—figure supplement 2 . ATF6β-N is generated in ER stressed cells treated with Ceapin-A7 but not ER stressed cells treated with the S1P inhibitor . U2-OS cells were treated with ER stressor ( 100 nM Tg ) in the absence ( lanes 1 , 2 , 7 and 8 ) or presence of S1P inhibitor ( 5 μM Pf-429242 , lanes 3 , 4 , 9 and 10 ) or Ceapin-A7 ( 6 μM , lanes 5 , 6 , 11 and 12 ) for four hours . One hour prior to lysis , proteasomal inhibitor ( 10 μM MG132 ) was added to prevent the degradation of the cleaved ATF6α-N and ATF6β-N fragments . Cells were harvested and separated by centrifugation into total ( lanes 1 , 3 , 5 , 7 , 9 and 11 ) and nuclear fractions ( lanes 2 , 4 , 6 , 8 , 10 and 12 ) and analyzed by Western blot for ATF6α , ATF6β , PERK and ATF4 . Arrowheads denote the positions of full-length ( ATF6α , ATF6β ) , cleaved membrane bound ( ATF6α-M ) and cleaved ( ATF6α-N , ATF6β-N ) variants of ATF6α and ATF6β . The transcription factor ATF4 is present in both total and nuclear extracts while membrane proteins PERK , ATF6α and ATF6β are present only in total extracts . In ER stressed cells faster migrating bands corresponding to ATF6α-N ( lanes 1 and 2 ) and ATF6β-N ( lanes 7 and 8 ) are found in both total and nuclear fractions . In ER stressed cells treated with the S1P inhibitor the faster migrating ATF6α and ATF6β bands found in the total extract are not present in the nuclear extract ( compare lanes 3 and 4 for ATF6α and lanes 9 and 10 for ATF6β ) indicating that neither ATF6α-N nor ATF6β-N were produced and these bands are likely ATF6α-M and ATF6β-M . In ER stressed cells treated with Ceapin-A7 ATF6α is not cleaved and no ATF6α-N is found in the nuclear extract ( lanes 5 and 6 ) . In contrast , ATF6β is cleaved and this faster migrating band is found in the nuclear extract indicating it is ATF6β-N ( lanes 11 and 12 ) . Data shown is representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 018 The same lysates were also analyzed for activation of the other branches of the UPR ( Figure 5A , bottom panels ) . Ceapin-A7 did not inhibit activation of either the PERK ( shown by a slower migrating band representing the phosphorylated form of PERK [Harding et al . , 1999] ) or the IRE1 ( shown by production of XBP1s protein ) branches of the UPR . These results validate and extend our analyses of the UPR in the 293T-based FLAG-ATF6α reporter cell line . Ceapin-A1 inhibited cleavage of FLAG-ATF6α without inhibiting induction of ATF4 ( Figure 3D , compare lanes 6 , 7 and 10 ) or autophosphorylation of PERK ( indicated by the shift in mobility; Figure 3D , compare lanes 11 , 12 and 15 ) . There is cross-talk between UPR branches . Specifically , effectors of both the IRE1 and PERK branches , XBP1 and CHOP respectively , are non-exclusive transcriptional targets of ATF6 ( Wu et al . , 2007; Adachi et al . , 2008 ) . Treatment of U2-OS cells with ER stressors showed upregulation of both spliced XBP1 mRNA ( XBP1s , Figure 5C ) and DDIT3 mRNA ( encodes CHOP , Figure 5C ) . Consistent with Ceapin inhibiting the ATF6-branch selectively , co-incubation of cells with ER stressor and either S1P inhibitor or Ceapin-A1 showed only a partial decrease in upregulation of XBP1s and DDIT3 mRNA ( Figure 5C , Figure 5—figure supplement 1 ) , in agreement with data from ATF6α knockout mouse embryonic fibroblasts ( MEF ) ( Wu et al . , 2007; Adachi et al . , 2008 ) . Taken together , these data show that Ceapins are selective inhibitors of the ATF6 branch of the UPR and do not inhibit either IRE1 or PERK signaling . The ER contains two related bZip proteins – ATF6α and ATF6β ( Haze et al . , 2001 ) . ATF6α and ATF6β show 41% identity in their amino acid sequences , with both S1P and S2P cleavage sites conserved . ATF6β is activated with similar kinetics and in response to the same stressors as ATF6α and , as ATF6α , moves from the ER to the Golgi apparatus where it is processed by S1P and S2P to release ATF6β-N from the membrane allowing its nuclear translocation . Given the similarity between these related proteins , we next tested if Ceapins inhibit processing of ATF6β in response to ER stress . To this end , we treated U2-OS cells either with Tg or Tm in the absence or presence of increasing concentrations of Ceapin-A7 or with S1P inhibitor and analyzed the migration pattern of both endogenous ATF6α and ATF6β-derived bands ( Figure 5D ) . Consistent with our previous results , we observed no ATF6α-N in lysates from Ceapin-A7 treated cells ( Figure 5D , lanes 3–6 ( Tg ) and lanes 10–13 ( Tm ) ) but observed ATF6α-M in lysates from S1P inhibitor-treated cells ( Figure 5D , lane 7 ( Tg ) and lane 14 ( Tm ) ) . In contrast , Ceapin-A7 had no effect on the production of ATF6β-N , even at the highest concentration ( 18 . 9 μM; corresponding to 33x its IC50 for ATF6α ) ( Figure 5D , lanes 17–20 and 24–27 ) . We further confirmed that while ATF6β-N is not present in nuclear extracts of ER stressed cells treated with S1P inhibitor , cells subjected to ER stress in the absence or presence of Ceapin-A7 both have nuclear localized ATF6β-N ( Figure 5—figure supplement 2 ) . ATF6α knockout mice and MEF cells show impaired survival in the face of acute ( Yamamoto et al . , 2007 ) or chronic ( Wu et al . , 2007 ) ER stress . We tested if Ceapins would similarly impair survival of human cells treated with an ER stressor . To this end , we treated U2-OS cells with increasing concentrations of Tg and monitored cell viability over a seventy-two hour time course ( Figure 6A ) . From the survival curve , we measured an IC50 of 7 . 1 nM for Tg alone in U2-OS cells ( Figure 6B ) . Cells co-incubated with both Tg and Ceapin-A7 showed an almost two-fold increase in sensitivity to ER stress ( IC50 = 4 . 5 nM ) , whereas the inactive Ceapin analog A5 showed no difference ( Figure 6B ) . This two-fold difference is consistent with the data from genetic ablation of ATF6α in mice ( Yamamoto et al . , 2007 ) . 10 . 7554/eLife . 11878 . 019Figure 6 . Ceapin-A7 sensitizes cells to ER stress . ( A–B ) U2-OS cells were treated with increasing concentrations of ER stressor ( Tg ) in the absence ( black ) or presence of six micromolar Ceapin analogs - Ceapin-A5 ( inactive , blue ) or Ceapin-A7 ( purple ) . ( A ) After seventy-two hours reducing potential of living cells was assayed to determine cell viability . Data plotted are the means of four independent experiments performed in triplicate , error bars represent the standard error of the mean . ( B ) EC50 values calculated for ER stressor in the absence or presence of Ceapin analogs showing mean and 95% confidence limits . ( C ) . U2-OS cells were treated with increasing concentrations of ER stressor ( Tg ) in the absence ( black ) or presence of 6 μM Ceapin-A7 ( purple ) . To analyze cell death , cells were stained with Annexin V and 7AAD and analyzed by flow cytometry . Data plotted are the means from three independent experiments performed in duplicate; error bars represent standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 11878 . 019 Since changes in cell viability may be due to cytostatic and/or cytotoxic effects , we next determined whether Ceapin-A7 displayed increased apoptosis in response to ER stress . To this end , we stained cells with annexin V , which measures phosphoserine flipping as a marker for apoptotic cells , and 7-aminoactinomycin D ( 7AAD ) , a membrane impermeable dye taken up only by cells with compromised plasma membranes as a marker for late apoptotic / necrotic cell death . We treated cells with or without ER stressor at different concentrations in the absence or presence of Ceapin-A7 and analyzed the cells by flow cytometry ( Figure 6C ) . Cells treated with Ceapin-A7 alone showed no difference in cell death compared to vehicle alone , consistent with previous work demonstrating that homozygous ATF6α knockout mice are viable and fertile ( Wu et al . , 2007; Yamamoto et al . , 2007 ) . At low concentrations of ER stress ( 10 nM Tg ) , inhibition of ATF6α did not enhance cytotoxicity , however as the concentration of ER stressor was increased ( 30 nM and 90 nM ) , ATF6α inhibition resulted in a two-fold increase in apoptotic cells compared to cells treated with ER stressor alone . Thus human cells treated with ER stress and Ceapin-A7 phenocopy the results obtained using genetic ablation of ATF6α in mouse models . Ceapins therefore define a first-in-class series of ATF6α inhibitors that selectively blocks ATF6α and not ATF6β , SREBP or other UPR branches without relying on inhibition of the proteases that are also used by other critical signaling pathways .
In this work we describe the isolation , specificity and chemical refinement of the Ceapin scaffold , the first selective and highly potent pharmacological inhibitors of the ATF6α branch of the UPR . The ATF6α pathway is the least well-understood UPR branch – its mechanism of activation in response to ER stress remains unknown , and the lack of unique enzymes in the pathway precluded development of biochemical assays for screening purposes . Prior to this work , the few compounds shown to inhibit ATF6α signaling ( PDI and S1P inhibitors ) target important housekeeping enzymes that are shared among multiple pathways and hence elicit pleiotropic effects ( Maurel et al . , 2015 ) . Indeed , ATF6 was considered 'undruggable' ( Maly and Papa , 2014 ) . Our study highlights the value of combining chemical biology with cell-based screening to identify precise tools for ‘intractable’ pathways , as amply demonstrated in other pioneering work ( Cassidy-Stone et al . , 2008; Schreiber et al . , 2015 ) . To identify Ceapins , we combined cell-based screens to isolate a highly selective inhibitor of ATF6α . Using a broad range of secondary assays for different steps of pathway activation , we built a toolbox of small molecules with which to interrogate ATF6α signaling at different points . We successfully isolated compounds that act both upstream of proteolytic cleavage , corresponding to the early stages of activation of ATF6α , and downstream of nuclear import , corresponding to transcriptional activation . From over 100 , 000 compounds screened , we found one , Ceapin-A1 , that is not only selective for ATF6α but acts precisely at its initial activation stage that currently is the least understood step . As we detail in the accompanying manuscript , Ceapins induce rapid , reversible clustering of ATF6α , preventing exit of ATF6α from the ER ( Gallagher and Walter , 2016 ) . Ceapins inhibit activation of ATF6α in response to ER stress without inhibiting ATF6β or SREBP activation , two other ER-bound transcription factors that are similarly trafficked and processed by S1P and S2P in the Golgi apparatus . Ceapins also do not inhibit the IRE1 or PERK branches of the UPR , indicating that the compounds do not generally interfere with sensing ER stress . Further underscoring their high selectivity , Ceapins show no toxicity to unstressed cells , consistent with the fact that ATF6α knockout mice are viable and fertile . Loss of ATF6α becomes detrimental when cells or animals are treated with ER stressors , and Ceapin action mimics this effect . Thus Ceapins promise to be invaluable , first-in-class tools to investigate the role of ATF6α independently or in combination with inhibition of other UPR branches in models of human disease . Efforts to study the role of ATF6α in disease models have been limited to knockdown or overexpression experiments . ATF6α was shown to be required for the survival of dormant tumor cells ( Schewe and Aguirre-Ghiso , 2008 ) . In a cystic fibrosis model , knockdown of ATF6α was shown to increase delivery of the poorly folded △F508 variant of the cystic fibrosis transmembrane conductance regulator to the plasma membrane where it could function ( Kerbiriou et al . , 2007 ) . Thus ATF6α inhibitors could be beneficial in diseases where decreased ER quality control would ameliorate cell function . In contrast , enhancing ATF6α activity using inducible activation of ATF6α-N increased quality control of protein folding in the ER and decreased secretion and extracellular aggregation of amyloidogenic proteins involved in light chain amyloidosis ( Cooley et al . , 2014 ) . Ceapins offer a new strategy to investigate the role of ATF6α in existing disease models without requiring the introduction of knockdown or overexpression constructs . While the pharmacokinetic properties of current Ceapin analogs limit their usefulness in animal studies , continued SAR studies to improve metabolic stability offer promise for the future . Much is made in reviews about the therapeutic potential of modulating the UPR and proteostasis networks ( Ryno et al . , 2013; Lindquist and Kelly , 2011; Brandvold and Morimoto , 2015 ) . While there has been an explosion in small molecule modulators of the UPR , it is still unclear which kinds of modulations open therapeutic windows for individual disease states . Some cell types exclusively activate one branch , others all three , and often it is not even clear whether down- or up-regulation of one branch or the other would lead to the desired outcome . In the UPR network , extensive cross-talk between the three branches further complicates the issue , with both compensation and interdependence between the UPR branches having implications for how to manage UPR modulation for therapeutic benefit ( Wu et al . , 2007 ) . For example , both PERK and ATF6 act downstream of vascular endothelial growth factor ( VEGF ) mediated cell survival and angiogenesis ( Karali et al . , 2014 ) . Additionally , inhibition of IRE1 alone in models of diabetes led to hyper-activation of ATF6 that led to severe nephropathy ( Madhusudhan et al . , 2015 ) . As these examples illustrate , appreciating the level of cross talk and compensation between UPR branches is essential to the design of successful UPR-based therapeutic strategies . To date , this was only possible for IRE1 and PERK . With the identification of Ceapins , we finally have tools to modulate all three branches and to unleash the full potential of UPR modulation that has remained theoretical up to this point .
Growth media was DMEM with high glucose ( Sigma D5796 ) supplemented with 10% FBS ( Life technologies # 10082147 ) , 2 mM L-glutamine ( Sigma G2150 ) , 100 U penicillin 100 μg/mL streptomycin ( Sigma P0781 ) . Additional cell line specific supplements are detailed below . Cells were incubated at 37°C , 5% CO2 unless stated otherwise . Human bone osteosarcoma ( U2-OS ) cells ( ATCC HTB-96 ) and human embryonic kidney ( HEK ) 293T cells ( ATCC CRL-3216 ) were obtained from the American Type Culture Collection . U2-OS cells stably expressing GFP-ATF6α were purchased from Thermo Scientific ( 084_01 ) . Growth media was supplemented with 500 μg/mL G418 ( Roche 04 727 878 001 ) to maintain expression of GFP-ATF6 . HeLa-NF cells were a generous gift from Paul Wade ( NIH ) ( Fujita et al . , 2003 ) . The XBP1 reporter cell line ( HEK293T XBP1-Luciferase ) was derived from the HEK 293T cell line ( ATCC CRL-3216 ) and was described previously ( Mendez et al . , 2015 ) . The ERSE-luciferase reporter cell line was also derived from the HEK 293T cell line ( ATCC CRL-3216 ) and is described below . 293 T-REx cells expressing doxycycline-inducible 6xHis-3xFLAG-HsATF6α ( wild type ( Sidrauski et al . , 2013 ) or mutant ) or MPZ-GFP are derived from ( Tet ) -ON 293 human embryonic kidney ( HEK ) cells ( Clontech ) containing a ferritin-like protein ( Flp ) recombination target ( FRT ) site ( Cohen and Panning , 2007 ) and are described below . Commercially available cell lines were authenticated by DNA fingerprint STR analysis by the suppliers . All cell lines were visually inspected using DAPI DNA staining and tested negative for mycoplasma contamination . Data from CellProfiler analysis were exported to MATLAB ( R2009a , MathWorks ) for further computation . At 18–24 hr prior to drug treatment , 12 , 000 U2-OS cells were plated per well of a 96 well plate ( Costar 3595 ) , covered with breathable seals ( Aeraseal ) and incubated in a humidified chamber 37°C , 5% CO2 . Each compound was tested at in duplicate wells using either 100 nM Tg or 2 μg/mL Tm as ER stress inducer . Eight unstressed and stressed wells ( four per inducer ) per plate were included as controls . Quarter-log serial dilutions of inhibitors in DMSO at 500x assay concentration were prepared in 96w plates = 'compound plate' ( Applied Biosystems MicroAmp Optical 96-well reaction plate N8010560 ) . Media without or with inducer was prepared to 6 . 073x and added to 96w plates to which 500x inhibitor stocks were added to 6x final = 'inducer plate' . Media without inducer contained DMSO as vehicle for unstressed control . Media containing either vehicle , ER stress inducer or ER stress inducer and inhibitor was added to cells to 1x final , covered with breathable seals and incubated for four hours at 37°C , 5% CO2 . The final volume of DMSO was equal between all wells ( 0 . 2% ) . Cells were lysed , RNA was prepared and PCR reactions were assembled using the Power SYBR Green Cells-to-CT kit ( Life Technologies #4402955 ) according to manufacturers instructions . Oligos used for qRT-PCR were as follows: GRP78 ( BiP ) : 5’-CATGGTTCTCACTAAAATGAAAG-3’ and 5’-GCTGGTACAGTAACAACTG-3’ . Herpud1: 5’-CAGAAATCAACGCCAAGGTG-3’ and 5’-GAACTTCCCTTTGCCTTAAACC-3’ Ero1LB: 5’-AATCTGAAGCGACCTTGTCC-3’ and 5’-GCCCAGCTTTTATTCCAACC-3’ XBP1s: 5’-GGAGTTAAGACAGCGCTTGG-3’ and 5’-CCTGCACCTGCTGCG-3’ DDIT3 ( CHOP ) : 5’-AGCCAAAATCAGAGCTGGAA-3’ and 5’-TGGATCAGTCTGGAAAAGCA-3’ GAPDH: 5’-TGGAAGATGGTGATGGGATT-3’ and 5’-AGCCACATCGCTCAGACAC-3’ For each experiment , duplicate reactions were performed on duplicate wells giving four values for each dose . qRT-PCR reactions were run using a CFX96 Real Time System ( Bio-Rad ) . Expression was normalized first to GAPDH internal control and then compared to stressed controls using CFX Manager 3 . 0 software ( Bio-Rad ) . For each inhibitor data was log transformed , dose-response curves ( log ( inhibitor ) versus response , variable slope , four parameter ) were plotted and IC50 were calculated using Prism 5 . Of ( GraphPad Software , Inc ) . Cleavage of SREBP in HeLa cells was analyzed essentially as described ( Hua et al . , 1995; Espenshade et al . , 1999; Sakai et al . , 1996 ) . Briefly , two days prior to drug treatment HeLa cells were plated in growth media at a density of 4 × 104 cells per well of a six-well plate . The following day , 16 . 5 hr prior to drug treatment , growth media was replaced with lipoprotein deficient media – DMEM with high glucose ( Sigma D5796 ) , 10% lipoprotein deficient serum from fetal calf ( LPDS , Sigma S5394 ) , 50 μM compactin ( aka mevastatin , a HMG-CoA reductase inhibitor , Santa Cruz Biotechnology sc-200853 ) , 50 μM mevalonolactone to facilitate non-sterol isoprenoids ( Sigma 68519 ) ( Goldstein and Brown , 1990 ) , 2 mM L-glutamine ( Sigma G2150 ) , 100 U penicillin 100 μg /mL streptomycin ( Sigma P0781 ) . After 16 . 5 hr in lipoprotein deficient media , sterols or inhibitors were added to cells . Sterols added were cholesterol ( 10 μg/mL , Sigma C3045 ) and 25-hydroxycholesterol ( 1 μg/mL , Sigma H1015 ) . Serial dilutions of inhibitors in DMSO at 1000x were prepared for both Ceapin and the S1P inhibitor , Pf-429242 . Final concentrations of inhibitors on cells were 0 . 5 , 5 , 15 and 25 μM respectively . For each well not receiving either sterols or inhibitors the corresponding vehicle was added such that the final concentration of ethanol and DMSO was equal for all wells . Four hours after addition of inhibitors , a proteasome inhibitor ( 25 μg/mL ALLN , Sigma A6185 ) was added to prevent degradation of cleaved SREBP-N . One hour later cells were harvested and protein lysates were prepared . For lysis , 0 . 5 mL of scraping buffer was added per well . Cells from each well were scrapped into eppendorf tubes and centrifuged at 3000 g for five minutes at four degrees . Each cell pellet was resuspended in 10 μL of lysis buffer and incubated on ice for twenty minutes . Tubes were then vortexed for five minutes at four degrees to shear genomic DNA , incubated on ice for five minutes , centrifuged at 1000 g for two minutes at four degrees . Total protein concentration per sample was determined from 2 μL of each sample using the BCA assay according to manufacturers instructions ( Thermo Scientific 23225 ) . For each sample , 8 . 92 μg total protein was loaded per lane of a fifteen well SDS-PAGE gel . After western blotting , membranes were probed with anti-SREBP1 ( abcam ab3259 ) or anti-GAPDH ( abcam ab9485 ) . Scraping buffer is 10 μM MG132 ( Sigma C2211 ) , 1x complete protease inhibitor ( Roche Diagnostics 05056489001 ) in phosphate buffered saline ( Sigma D8537 ) . Lysis buffer is 200 mM Tris pH 8 . 0 , 1% SDS , 100 mM NaCl , 10 μM MG132 , 1x complete protease inhibitor . Loading buffer was added to each sample from a 5x stock to 1x final . 1x loading buffer is 40 mM Tris pH 8 . 0 , 0 . 2% SDS , 8 mM DTT , 6% glycerol , 10 μM MG132 , 1x complete protease inhibitor , bromophenol blue . N-{1-[ ( 2 , 4-Dichlorophenyl ) methyl]-1H-pyrazol-4-yl}-5- ( furan-2-yl ) -1 , 2-oxazole-3-carboxamide ( Ceapin-A1 ) was purchased from Chemdiv . N-{1-[ ( 2 , 4-Dichlorophenyl ) methyl]-1H-pyrazol-4-yl}-5-methyl-1 , 2-oxazole-3-carboxamide ( Ceapin-A2 ) andN-{1-[ ( 2 , 4-dichlorophenyl ) methyl]-1H-pyrazol-4-yl}-5-phenyl-1 , 2-oxazole-3-carboxamide ( Ceapin-A3 ) were purchased from ChemBridge . N- ( 1-Benzyl-1H-pyrazol-4-yl ) -5- ( furan-2-yl ) -1 , 2-oxazole-3-carboxamide ( Ceapin-A4 ) and 5- ( furan-2-yl ) -N- ( 1-methyl-1H-pyrazol-4-yl ) -1 , 2-oxazole-3-carboxamide ( Ceapin-A5 ) were purchased from Enamine . Reagents and solvents were purchased from Sigma- Aldrich , Acros , Combi-Blocks , AK Scientific , ChemBridge , Enamine or TCI America and used as received unless otherwise indicated . Flash column chromatography was carried out using a Biotage Isolera Four system and SiliaSep silica gel cartridges from Silicycle . Hydrogenation reactions were carried out in ThalesNano H-Cube reactor using 30 mm 10% Pt/C catalyst cartridges . 1H NMR spectra were recorded on a Varian INOVA-400 400MHz spectrometer . Chemical shifts are reported in δ units ( ppm ) relative to residual solvent peak . Coupling constants ( J ) are reported in hertz ( Hz ) . Characterization data are reported as follows: chemical shift , multiplicity ( s=singlet , d=doublet , t=triplet , q=quartet , br=broad , m=multiplet ) , coupling constants , number of protons , mass to charge ratio . LC/MS analyses were performed on a Waters Micromass ZQ/Waters 2795 Separation Module/Waters 2996 Photodiode Array Detector/Waters 2424 Evaporative Light Scattering Detector system . Separations were carried out on an XTerra MS C18 5 µm 4 . 6 × 50 mm column at ambient temperature using a mobile phase of water-methanol containing 0 . 1% formic acid . To a solution of 4-nitro-1H-pyrazole ( 1 equiv ) in N , N-dimethylformamide , were added potassium carbonate ( 2 equiv ) and the benzyl bromide ( 1 equiv ) . The mixture was stirred at ambient temperature until judged complete by LC/MS . The reaction mixture was then diluted with ethyl acetate ( 10 mL ) , washed with saturated ammonium chloride solution ( 10 mL ) , water ( 10 mL ) and brine ( 10 mL ) . The organic layer was dried over magnesium sulfate , concentrated under reduced pressure and purified by flash column chromatography ( ethyl acetate/hexanes ) to obtain the product . A methanolic solution of the nitro compound was passed through 10% Pt/C catalyst at a rate of 1 mL/min in the H-Cube reactor under atmospheric pressure and ambient temperature until the reaction was judged complete by LC/MS . The reaction mixture was concentrated under reduced pressure to obtain the crude amine that was used without further purification . To a solution of the carboxylic acid ( 1 equiv ) in N , N-dimethylformamide , were added HATU ( 1 . 1 equiv ) , the amine ( 1 equiv ) , and N , N-diisopropylethylamine ( 2 equiv ) . The mixture was stirred at ambient temperature until the reaction was judged complete by LC/MS . The reaction mixture was then diluted with ethyl acetate ( 10 mL ) , washed with saturated ammonium chloride solution ( 10 mL ) , water ( 10 mL ) and brine ( 10 mL ) . The organic layer was dried over magnesium sulfate , concentrated under reduced pressure and purified by flash column chromatography ( ethyl acetate/hexanes ) to obtain the product . To a cooled ( 0°C ) solution of 2 , 4-dibromobenzyl alcohol ( 0 . 1 g , 0 . 37 mmol ) in dichloromethane ( 1 . 0 mL ) and N , N-diisopropylethylamine ( 0 . 13 mL , 0 . 75 mmol ) was added dropwise methanesulfonyl chloride ( 0 . 032 mL , 0 . 41 mmol ) . The mixture was stirred at 0°C for 15 min followed by addition of potassium carbonate ( 0 . 1 g , 0 . 75 mmol ) and a solution of 4-nitro-1H-pyrazole ( 0 . 042 g , 0 . 37 mmol ) in N , N-dimethylformamide ( 0 . 5 mL ) . The mixture was stirred at 50°C for 18 hr . The reaction mixture was then diluted with ethyl acetate ( 10 mL ) , washed with saturated ammonium chloride solution ( 10 mL ) , water ( 10 mL ) and brine ( 10 mL ) . The organic layer was dried over magnesium sulfate , concentrated under reduced pressure and purified by flash column chromatography ( 20% ethyl acetate/hexanes ) to obtain 0 . 43 g ( 82% ) of the title compound as a cream colored solid . 1H NMR ( 400 MHz , CDCl3 ) δ 8 . 15 ( s , 1H ) , 8 . 08 ( s , 1H ) , 7 . 79 ( S , 1H ) , 7 . 48 ( d , J = 8 . 2 Hz , 1H ) , 7 . 12 ( d , J = 8 . 2 Hz , 1H ) , 5 . 37 ( s , 2H ) . LCMS m/z 361 ( MH+ ) . To a refluxing mixture of 1-{[2 , 4-dibromophenyl]methyl}-4-nitro-1H-pyrazole ( 0 . 06 g , 0 . 17 mmol ) and ammonium chloride ( 0 . 09 g , 1 . 7 mmol ) in a 2:1 mixture of ethanol/water ( 6 . 0 mL ) , was added in portions , iron ( 0 . 028 g , 0 . 5 mmol ) over a period of 30 min . After refluxing for an additional 5 hr and cooling to ambient temperature , dichloromethane ( 20 mL ) was added to the reaction mixture . The organic layer was washed with brine , dried over magnesium sulfate and concentrated to obtain 1-{[2 , 4-dibromophenyl]methyl}-1H-pyrazol-4-amine which was used without further purification . To a solution of 5- ( furan-2-yl ) -1 , 2-oxazole-3-carboxylic acid ( 0 . 016 g , 0 . 09 mmol ) in N , N-dimethylformamide ( 0 . 5 mL ) , were added HATU ( 0 . 038 g , 0 . 099 mmol ) , 1-{[2 , 4-dibromophenyl]methyl}-1H-pyrazol-4-amine ( 0 . 03 g , 0 . 09 mmol ) , and N , N-diisopropylethylamine ( 0 . 031 mL , 0 . 18 mmol ) . The mixture was subjected to conditions described in method C and purified by flash column chromatography ( 20% ethyl acetate/hexanes ) to obtain 0 . 016 g ( 36% ) of the title compound as a pink colored solid . 1H NMR ( 400 MHz , CDCl3 ) δ 8 . 52 ( s , 1H ) , 8 . 09 ( s , 1H ) , 7 . 74 ( S , 1H ) , 7 . 62 ( s , 1H ) , 7 . 58 ( s , 1H ) , 7 . 39 ( dd , J = 8 . 3 , 1 . 9 Hz , 1H ) , 6 . 98 ( d , J = 3 . 4 Hz , 1H ) , 6 . 90 ( s , 1H ) , 6 . 87 ( d , J = 8 . 0 Hz , 1H ) , 6 . 56–6 . 68 ( m , 1H ) , 5 . 34 ( s , 2H ) ; LCMS m/z 492 ( MH+ ) . To a solution of 4-nitro-1H-pyrazole ( 0 . 2 g , 1 . 8 mmol ) in N , N-dimethylformamide ( 2 . 0 mL ) , were added potassium carbonate ( 0 . 489 g , 3 . 5 mmol ) and 2 , 4-bis ( trifluoromethyl ) benzyl bromide ( 0 . 332 mL , 1 . 8 mmol ) . The mixture was subjected to conditions described in method A and purified by flash column chromatography ( 15% ethyl acetate/hexanes ) to obtain 0 . 43 g ( 72% ) of the title compound as a white solid . 1H NMR ( 400 MHz , CDCl3 ) δ 8 . 15 ( d , J = 8 . 1 Hz , 2H ) , 7 . 98 ( s , 1H ) , 7 . 82 ( d , J = 8 . 1 Hz , 1H ) , 7 . 39 ( d , J = 8 . 1 Hz , 1H ) , 5 . 57 ( s , 2H ) . A solution of 1-{[2 , 4-bis ( trifluoromethyl ) phenyl]methyl}-4-nitro-1H-pyrazole ( 0 . 04 g , 0 . 1 mmol ) in methanol ( 20 mL ) was subjected to hydrogenation conditions described in general method B to obtain about 37 mg of crude 1-{[2 , 4-bis ( trifluoromethyl ) phenyl]methyl}-1H-pyrazol-4-amine as a colorless oil which was used without further purification . LCMS m/z 310 ( MH+ ) . To a solution of 5- ( furan-2-yl ) -1 , 2-oxazole-3-carboxylic acid ( 0 . 018 g , 0 . 1 mmol ) in N , N-dimethylformamide ( 0 . 5 mL ) , were added HATU ( 0 . 042 g , 0 . 11 mmol ) , 1-{[2 , 4-bis ( trifluoromethyl ) phenyl]methyl}-1H-pyrazol-4-amine ( 0 . 031 g , 0 . 1 mmol ) , and N , N-diisopropylethylamine ( 0 . 035 mL , 0 . 2 mmol ) . The mixture was subjected to conditions described in general method C to obtain 0 . 045 g ( 95% ) of the title compound as a white powder . 1H NMR ( 400 MHz , CDCl3 ) δ 8 . 52 ( s , 1H ) , 8 . 11 ( s , 1H ) , 7 . 94 ( s , 1H ) , 7 . 72 ( d , J = 8 Hz , 1H ) , 7 . 67 ( s , 1H ) , 7 . 58 ( s , 1H ) , 7 . 09 ( d , J = 8 . 1 Hz , 1H ) , 6 . 98 ( d , J = 3 . 5 Hz , 1H ) , 6 . 90 ( s , 1H ) , 6 . 56–6 . 57 ( m , 1H ) , 5 . 58 ( s , 2H ) ; LCMS m/z 471 ( MH+ ) . To a solution of 4-nitro-1H-pyrazole ( 0 . 25 g , 2 . 21 mmol ) in N , N-dimethylformamide ( 2 . 5 mL ) , were added potassium carbonate ( 0 . 61 g , 4 . 42 mmol ) and 4- ( trifluoromethyl ) benzyl bromide ( 0 . 34 mL , 2 . 21 mmol ) . The mixture was stirred at ambient temperature for 18 hr and then subjected to conditions described in method A to obtain 0 . 61 g of the crude product as a white solid which was used without further purification . 1H NMR ( 400 MHz , CDCl3 ) δ 8 . 11 ( d , J = 2 . 7 Hz , 2H ) , 7 . 66 ( d , J = 8 . 0 Hz , 2H ) , 7 . 39 ( d , J = 8 . 0 Hz , 2H ) , 5 . 37 ( s , 2H ) . A solution of 1-{[4- ( trifluoromethyl ) phenyl]methyl}-4-nitro-1H-pyrazole ( 0 . 029 g , 0 . 1 mmol ) in methanol ( 15 mL ) was subjected to hydrogenation conditions described in method B to obtain about 0 . 026 g of crude 1-{[4- ( trifluoromethyl ) phenyl]methyl}-1H-pyrazol-4-amine as a colorless oil which was used without further purification To a solution of 5- ( furan-2-yl ) -1 , 2-oxazole-3-carboxylic acid ( 0 . 019 g , 0 . 1 mmol ) in N , N-dimethylformamide ( 0 . 5 mL ) , were added HATU ( 0 . 042 g , 0 . 11 mmol ) , 1-{[4- ( trifluoromethyl ) phenyl]methyl}-1H-pyrazol-4-amine ( 0 . 026 g , 0 . 1 mmol ) , and N , N-diisopropylethylamine ( 0 . 035 mL , 0 . 2 mmol ) . The mixture was subjected to conditions described in method C and purified by flash column chromatography ( 30% ethyl acetate/hexanes ) to obtain 0 . 017 g ( 43% ) of the title compound as a white powder . 1H NMR ( 400 MHz , CDCl3 ) δ 8 . 50 ( s , 1H ) , 8 . 07 ( s , 1H ) , 7 . 58–7 . 60 ( m , 3H ) , 7 . 33 ( d , J = 7 . 9 Hz , 2H ) , 6 . 97 ( d , J = 3 . 5 Hz , 1H ) , 6 . 90 ( s , 1H ) , 6 . 56 ( dd , J = 3 . 5 , 1 . 7 Hz , 1H ) , 5 . 34 ( s , 2H ) ; LCMS m/z 403 ( MH+ ) . To a solution of 4-nitro-1H-pyrazole ( 0 . 25 g , 2 . 21 mmol ) in N , N-dimethylformamide ( 2 . 5 mL ) , were added potassium carbonate ( 0 . 61 g , 4 . 42 mmol ) and 2- ( trifluoromethyl ) benzyl bromide ( 0 . 53 g , 2 . 21 mmol ) . The mixture was stirred at ambient temperature for 18 hr and then subjected to conditions described in method A to obtain 0 . 55 g of the crude 1-{[2- ( trifluoromethyl ) phenyl]methyl}-4-nitro-1H-pyrazole as a white solid which was used without further purification . 1H NMR ( 400 MHz , CDCl3 ) δ 8 . 07–8 . 10 ( m , 2H ) , 7 . 74 ( d , J = 7 . 7 Hz , 1H ) , 7 . 48–7 . 59 ( m , 2H ) , 7 . 28 ( d , J = 8 . 0 Hz , 1H ) , 5 . 52 ( s , 2H ) . LCMS m/z 272 ( MH+ ) . A solution of 1-{[2- ( trifluoromethyl ) phenyl]methyl}-4-nitro-1H-pyrazole ( 0 . 032 g , 0 . 1 mmol ) in methanol ( 15 mL ) was subjected to hydrogenation conditions described in method B to obtain about 0 . 029 g of crude 1-{[2- ( trifluoromethyl ) phenyl]methyl}-1H-pyrazol-4-amine as a colorless oil which was used without further purification . LCMS m/z 242 ( MH+ ) . To a solution of 5- ( furan-2-yl ) -1 , 2-oxazole-3-carboxylic acid ( 0 . 019 g , 0 . 1 mmol ) in N , N-dimethylformamide ( 0 . 5 mL ) , were added HATU ( 0 . 042 g , 0 . 11 mmol ) , 1-{[2- ( trifluoromethyl ) phenyl]methyl}-1H-pyrazol-4-amine ( 0 . 026 g , 0 . 1 mmol ) , and N , N-diisopropylethylamine ( 0 . 035 mL , 0 . 2 mmol ) . The mixture was subjected to conditions described in method C and purified by flash column chromatography ( 25% ethyl acetate/hexanes ) to obtain 0 . 028 g ( 70% ) of the title compound as a white powder . 1H NMR ( 400 MHz , CDCl3 ) δ 8 . 50 ( s , 1H ) , 8 . 04 ( s , 1H ) , 7 . 68 ( d , J = 8 . 0 Hz , 1H ) , 7 . 66 ( s , 1H ) , 7 . 58 ( s , 1H ) , 7 . 45–7 . 49 ( m , 1H ) , 7 . 40 ( d , J = 7 . 6 Hz , 1H ) , 7 . 01 ( d , J = 8 . 0 Hz , 1H ) , 6 . 97–6 . 99 ( m , 1H ) , 6 . 90 ( s , 1H ) , 6 . 55–6 . 57 ( m , 1H ) , 5 . 52 ( s , 2H ) ; LCMS m/z 403 ( MH+ ) . U2-OS cells were detached using Accutase ( Innovative Cell Technologies #AT-104-500 ) and the cell number was determined with the Scepter Cell Counter as described by the manufacturer ( Millipore ) . 2000 cells were seeded into each well of a 96 well black walled optical bottom plate ( Corning 3882 ) in normal growth medium . After 16–18 hr , the media was removed from the cells and 120 μL of fresh media with increasing concentrations of thapsigargin plus or minus 6 uM Ceapin 2 or 3 . The highest concentration of thapsigargin was 90 nM and six 1:3 serial dilutions were performed . The final DMSO concentration for all samples including the DMSO only control was 0 . 034% . Breathable seals were used to seal the plate and placed in the incubator in a humidified chamber . At 72 hr , PrestoBlue Cell Viability Reagent ( Life Technologies #A13262 ) was added to each well and incubated at 37C for 10 min and read on a plate reader as described by the manufacturer . PrestoBlue Cell Viability Reagent is a Resazurin based cell viability indicator that is reduced to a highly fluorescent molecule by viable cells . For data analysis the background ( wells with media without cells ) was subtracted from the experimental wells and viability relative to the DMSO treated wells was calculated in Microsoft Excel . The means of four independent experiments performed in triplicate were graphed using GraphPad Prism using the non-linear regression Sigmoidal , 4PL , X is log ( Concentration ) equation . The absolute EC50 was calculated in GraphPad Prism to interpolate X at 50% with 95% confidence intervals . US-OS cells were detached with Accutase ( Innovative Cell Technologies #AT-104-500 ) and cells counted using the Sceptor Cell Counter as described by the manufacture ( Millipore ) . 20 , 000 cells in 0 . 5 mL of growth medium were added to each well of a 24 well plate ( Corning 3526 ) . After 16–18 hr the media was carefully removed from the wells and fresh medium with DMSO , 10 , 30 or 90 nM Thapsigargin plus or minus 6 μM Ceapin 3 was added . Final DMSO concentration for all wells was 0 . 034% . After 72 hr , media was removed from each wells and placed in 1 . 5 mL Eppendorf tubes . 250 μL of Accutase was used to detach cells in the well and the entire volume was added to Eppendorf tube containing the kept media . 100 μL of the media and cell mixture in the Eppendorf tubes were transferred to 2 mL microtubes ( VWR #16466–030 ) containing 100 μL of room temperature Muse Annexin V and Dead Cell Reagent ( EMD Millipore #MCH100105 ) . The kit contains Annexin V to detect phosphatidylserine for use as an early apoptotic marker and the membrane impermeable DNA dye , 7-Aminoactinomycin D ( 7AAD ) , to detect late apoptotic and necrotic cells . Tubes were gently vortexed for 5 s and incubated at room temperature for 20 min in the dark . Flow cytometry was performed using the Muse Cell Analyzer ( EMD Millipore ) and gated as directed by the manufacturer . To minimize reading times between samples , 1000 events were read for each sample and the percentage of live and apoptotic/dead cells was calculated as described by the manufacturer . The means of three independent experiments performed in duplicate were graphed in GraphPad Prism . 7AAD: 7-aminoactinomycin D ATF: Activating Transcription Factor CFTR: Cystic Fibrosis transmembrane conductance regulator CHOP: CCAAT-enhancer binding proteins ( C/EBP ) Homologous Protein CMV: Cytomegalovirus ERSE: ER Stress response Element GFP: Green Fluorescent Protein GRP: Glucose Regulated Protein HEK293T: Human embryonic Kidney 293T cells . IRE1: Inositol Requiring Enzyme One MCP: Minimal cytomegaloviral promoter PERK: Protein Kinase R ( PKR ) -like Endoplasmic Reticulum Kinase RIDD: Regulated IRE1-Dependent Decay Tg: Thapsigargin Tm: Tunicamycin U2-OS: U2 Osteosarcoma cells UPRE: Unfolded Protein Response Element VEGF: Vascular Endothelial Growth Factor XBP1: X-box Binding Protein One | Newly made proteins must be folded into specific three-dimensional shapes before they can perform their roles in cells . Many proteins are folded in a cell compartment called the endoplasmic reticulum . The cell closely monitors the quality of the work done by this compartment . If the endoplasmic reticulum has more proteins to fold than it can handle , unfolded or misfolded proteins accumulate and trigger a stress response called the unfolded protein response . This increases the capacity of the endoplasmic reticulum to fold proteins to match the demand . However , if the stress persists , then the unfolded protein response instructs the cell to die to protect the rest of the body . A protein called ATF6α is one of three branches of the unfolded protein response . This protein is found in the endoplasmic reticulum where it is inactive . Endoplasmic stress causes ATF6α to move from the endoplasmic reticulum to another compartment called the Golgi apparatus . There , two enzymes cut ATF6α to release a fragment of the protein that then moves to the nucleus to increase the production of the machinery needed to fold proteins in the endoplasmic reticulum . Errors in protein folding can cause serious diseases in humans and other animals . Drugs that target ATF6α might be able to regulate part of the unfolded protein response to treat these diseases . However , no drugs that act on ATF6α had been identified . Now , two groups of researchers have independently identified small molecules that specifically target ATF6α . Gallagher et al . screened over 100 , 000 compounds for their ability to reduce the activity of ATF6α-regulated genes . The experiments reveal that a class of small molecules termed Ceapins can selectively block the activity of ATF6α during endoplasmic reticulum stress , but had no effect on other proteins involved in the unfolded protein response . Furthermore , when human cells experiencing stress were treated with Ceapins , a greater number of cells died in comparison to cells that had not received Ceapins . An accompanying study by Gallagher and Walter reports on the mechanism by which Ceapins act on ATF6α . Independently , Plate et al . identified a type of small molecule that can activate ATF6 . Together , the findings of Gallagher et al . and Plate et al . may lead to the development of new drugs for treating diseases associated with incorrect protein folding in the endoplasmic reticulum . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2016 | Ceapins are a new class of unfolded protein response inhibitors, selectively targeting the ATF6α branch |
Mutations in the human telomerase reverse transcriptase ( TERT ) promoter are the most frequent non-coding mutations in cancer , but their molecular mechanism in tumorigenesis has not been established . We used genome editing of human pluripotent stem cells with physiological telomerase expression to elucidate the mechanism by which these mutations contribute to human disease . Surprisingly , telomerase-expressing embryonic stem cells engineered to carry any of the three most frequent TERT promoter mutations showed only a modest increase in TERT transcription with no impact on telomerase activity . However , upon differentiation into somatic cells , which normally silence telomerase , cells with TERT promoter mutations failed to silence TERT expression , resulting in increased telomerase activity and aberrantly long telomeres . Thus , TERT promoter mutations are sufficient to overcome the proliferative barrier imposed by telomere shortening without additional tumor-selected mutations . These data establish that TERT promoter mutations can promote immortalization and tumorigenesis of incipient cancer cells .
Activation of telomerase is the critical step for the immortalization of more than 90% of all human tumors ( Greider and Blackburn , 1985; Counter , 1992; Kim et al . , 1994 ) . Non-coding mutations in the promoter of the catalytic subunit of telomerase ( TERT ) emerged recently as one of the most prevalent mutations in human cancer ( Bojesen et al . , 2013; Horn et al . , 2013; Huang et al . , 2013; Killela et al . , 2013; Fredriksson et al . , 2014; Weinhold et al . , 2014 ) . Interestingly , all TERT promoter mutations associated with cancer formation thus far generate novel binding sites for the ETS ( E26 transformation-specific ) family of transcription factors and are located close to the translational start site of TERT ( e . g . , −57A/C , −124C/T , and −146C/T ) ( Horn et al . , 2013; Huang et al . , 2013 ) . Transient transfection experiments using ectopic TERT Luciferase-reporter constructs suggest that TERT promoter mutations can increase TERT transcription by 1 . 5–2 fold when assayed in tumor cells ( Horn et al . , 2013; Huang et al . , 2013 ) . To date , the physiological events that select for these specific mutations are still unclear , as they have been mostly investigated for their impact in tumor cell lines that are already immortal , maintain telomere length , and have aberrant karyotypes . These tumor cell lines have sufficient telomerase activity to maintain an immortal phenotype , but so do tumor cells without these TERT promoter mutations . Thus , changes in telomerase levels and telomere length provide incomplete information regarding the functional differences between cells that do or do not carry TERT promoter mutations . In untransformed human tissues , telomerase activity is restricted to embryonic cells and some adult stem cell or progenitor compartments due to transcriptional silencing of TERT upon differentiation ( Gunes and Rudolph , 2013; Aubert , 2014 ) . As a consequence , differentiated somatic cells undergo progressive telomere shortening with cell division , which limits their proliferative capacity and has thus been proposed as a tumor suppressor mechanism ( Wright et al . , 1996 ) . Critically short telomeres are detected as sites of DNA damage leading to cell death or replicative senescence ( Palm and de Lange , 2008 ) . Long-term inhibition of TERT ( Herbert et al . , 1999 ) or interference with telomerase recruitment to telomeres ( Nakashima et al . , 2013; Sexton et al . , 2014 ) lead to cell death in telomerase-positive cancer and stem cells . Inversely , ectopic telomerase expression is sufficient to immortalize normal human fibroblasts by allowing them to bypass senescence ( Bodnar et al . , 1998; Morales et al . , 1999 ) . Since the discovery of telomerase reactivation in cancer , many cis-regulatory elements and corresponding transcription factors have been suggested to contribute to the regulation of TERT in healthy cells and its aberrant expression in tumor cells ( Greenberg et al . , 1999; Ducrest et al . , 2002; Lin and Elledge , 2003; Kyo et al . , 2008 ) . GWAS analysis identified a specific set of TERT promoter mutations in melanomas that all occur in a very small region close to the transcriptional start site and each results in novel putative TTCCGG- ETS binding sites ( Horn et al . , 2013; Huang et al . , 2013 ) . While ETS-factors are a large family of transcription factors that can recognize this binding site , recent data suggest that TERT promoter mutations are bound predominantly by GABP ( Bell et al . , 2015 ) . This specificity does not appear restricted to melanomas as the same TERT promoter mutations have emerged as a major driver in a multitude of human solid tumors ( Heidenreich et al . , 2014 ) , including glioblastomas , medulloblastomas , carcinomas of the bladder , urothelial cancer ( Borah et al . , 2015 ) , thyroid and squamous cell carcinomas of the tongue , as well as in liposarcomas and hepatocellular carcinomas ( Heidenreich et al . , 2014 ) . Based on this tumor spectrum , TERT promoter mutations have been hypothesized to preferentially promote tumor progression in tissues with relatively low rates of self-renewal ( Killela et al . , 2013 ) . Several studies have suggested that TERT promoter mutations can provide a biomarker to stratify human cancer subtypes ( Heidenreich et al . , 2014; Borah et al . , 2015 ) . However , the mechanism by which these mutations promote tumor formation is unknown . The key outstanding questions are: ( 1 ) whether TERT promoter mutations are sufficient to immortalize cells and ( 2 ) why TERT promoter mutations occur in specific tumors subtypes . Here we address these questions by genetically engineering human embryonic stem cells ( hESCs ) to carry the three most prevalent cancer-associated TERT promoter mutations in an isogenic background . The impact of these mutations was studied by measuring their effect on TERT expression , telomerase activity , and telomere length in stem cells as well as in differentiated cell types . We demonstrate that two out of three cancer-associated TERT mutations caused no effect and only the most prevalent promoter mutations mildly increased TERT levels in hESCs , which did not result in significantly increased telomerase activity . We find that increased TERT expression is not functionally linked to an increase in active telomerase , as TR , the telomerase RNA component , but not TERT is limiting in hESCs . However , the importance of these mutations in tumorigenesis becomes clear when hESCs are differentiated into normally telomerase-negative cells . Under these conditions all cancer-associated TERT mutations prevent repression of TERT , resulting in a retention of telomerase activity relative to wild-type differentiated cells . Ultimately , the resulting TERT expression led to aberrant telomerase enzymatic activity in terminally differentiated cells and abnormally long telomeres , thereby bypassing the telomere shortening tumor suppressor pathway .
We aimed to understand the molecular basis by which the cancer-associated TERT mutations impact telomerase biology . To address this question , we employed CAS9- ( clustered regularly interspaced short palindromic repeats ( CRISPR ) /CRISPR-associated systems 9 ) ( Jinek et al . , 2012 ) mediated genome editing to derive human pluripotent stem cells ( WIBR#3 ) that carry TERT promoter mutations at the endogenous TERT locus . Initially we attempted conventional donor-based genome editing strategies with sgRNAs targeting sequences proximal to the targeting site . These attempts were however unsuccessful , likely due to the TERT promoter mutations being in a genomic region with ∼80% GC content . This non-random base composition does not allow for the design of specific sgRNAs without a large number of potential off-targets . We tested several sgRNAs in proximity to TERT promoter mutations and found them to be toxic to cells shortly after transfection into cancer cells and human primary fibroblasts . We overcame this challenge by employing a two-step targeting approach ( Figure 1A ) . In a first editing step we homozygously deleted a 1 . 5 kb region in the TERT gene using two sgRNAs that cut at positions −1462 and +67 relative to the first ATG ( Figure 1A ) . In a second editing step we reintroduced the deleted region either with or without the promoter mutations into the endogenous TERT locus ( Figure 1A , B ) . 10 . 7554/eLife . 07918 . 003Figure 1 . Generation of isogenic TERT promoter mutation-containing hESCs reveals a modest increase of TERT expression only for the −124C/T mutation . ( A ) Schematic overview of the two-step approach used to genome-edit TERT promoter mutations in hESCs . First , TERT knock-out cell line ( TERTΔ/Δ ) that lacks 1 . 5 kb upstream and 66 bp downstream of the first ATG was established using two CAS9/sgRNAs ( sg-1 and sg-15 ) . Second , an sgRNA against the newly synthesized NHEJ-derived junction ( −1462 and +67: sg1+15; see Figure 1—figure supplement 1B ) were co-electroporated with donor plasmids containing the deleted regions with or without the cancer-associated TERT promoter mutations . ( B ) Sequence analysis of targeted cells confirmed successful restoration and introduction of the TERT promoter mutations . ( C ) Telomeric repeat amplification protocol ( TRAP ) assay of whole cell extracts from TERTΔ/Δ hESC lines ( n = 2 ) using 200 ng protein . TERTΔ/Δ #1 and #2 cells were collected at day 89 and day 146 after the first editing respectively . IC: internal control . ( D ) Telomere restriction fragment assay of wild-type ( WT ) , TERTΔ/Δ , and the targeted hESCs over a time course after targeting ( day 0: first editing step , day 73: second editing ) . TERTΔ/Δ #1 cells are telomerase-deficient , undergo telomere shortening and die around day 120 unless they regain telomerase activity through the second targeting step . At the first time point ( day 101 ) , the majority of the cells are untargeted TERTΔ/Δ cells , therefore telomere length is heterogeneous and short . This short telomere length results in reduced hybridization intensity with the TTAGGG radioactive probe . In contrast at the second time point ( day 129 ) , uncomplemented TERTΔ/Δ #1 died due to progressive telomere shortening and the targeted populations are enriched . In this targeted population the restoration of telomerase resulted in substantial telomere elongation and an overall increase in telomere signal intensity . 2 µg of genomic DNA after digestion with MboI and AluI were loaded in each lane . Quantification of the average telomere length signal is indicated at the bottom of the gel . Throughout all figures we refer to non-targeted wild-type WIBR#3 hESCs as WT . We refer to wild-type cells generated by reintroducing the wild-type promoter into TERTΔ/Δ as wt . ( E ) Relative expression levels of TERT mRNA by mutant and wt promoter-containing hESCs over a time course after targeting measured by quantitative RT-PCR . Expression is relative to WT hESCs ( black line ) . Expression of TERT was normalized to GAPDH . Also shown is TERTΔ/Δ cells ( green line ) until day 123 . This is the last time point in which RNA could be isolated before TERTΔ/Δ cultures died . ( F ) TRAP assay of whole cell extracts from WT and promoter mutation-containing hESCs ( day 147 ) using decreasing amount of protein ( 200 , 40 , 8 ng ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07918 . 00310 . 7554/eLife . 07918 . 004Figure 1—figure supplement 1 . Genotyping of TERTΔ/Δ hESCs prior to the second targeting that introduced the mutated promoter sequences . ( A ) Southern blot analysis for TERTΔ/Δ hESCs . Genomic DNA isolated form individual clones was digested with BamHI and hybridized with the 3′ probe ( the top panel ) . The correctly targeted allele appears as a 9 . 5 kb band and the untargeted wild type allele is 11 kb . Homozygous targeted hESC clones are shown in blue , heterozygous targeted clones are in red , and untargeted clones are in black . The parental cell lines used for the second targeting , TERTΔ/Δ and TERTΔ/Δ#2 , are clone #26 and #17 respectively . The correct deletion events were also confirmed by PCR using external primers ( bottom panel ) . ( B ) The newly formed NHEJ-derived junction of the deleted region in the homozygous targeted lines TERTΔ/Δ and TERTΔ/Δ #2 was determined by sequencing . The sgRNA for the second targeting was designed across the junction ( sg-1+15 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07918 . 00410 . 7554/eLife . 07918 . 005Figure 1—figure supplement 2 . Independent confirmation of promoter mutation experiments ( shown in Figure 1C , D ) using an independent TERTΔ/Δ#2 cell line . ( A ) Relative expression levels of TERT mRNA of mutant and wt promoter-containing hESCs ( TERTΔ/Δ #2 ) over a time course after targeting measured by quantitative RT-PCR . Expression of TERT was normalized to GAPDH . ( B ) Telomere restriction fragment assay of WT , TERTΔ/Δ#2 and the targeted hESCs over a post-targeting time course ( day 0: first editing; day 86: second editing ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07918 . 00510 . 7554/eLife . 07918 . 006Figure 1—figure supplement 3 . The clonal analysis of TERT promoter mutation containing hESCs confirmed the results of the bulk analysis . ( A ) Southern blot analysis for TERTΔ/Δ hESCs clones . The correctly targeted allele appeared as a 11 kb band at the size of WT hESCs and the untargeted allele as 9 . 5 kb at the size of parental TERTΔ/Δ hESCs . ( B ) Quantitative RT-PCR of TERT and OCT4 in individual clones of the targeted hESCs . Expression levels are shown relative to WT hESCs and normalized to GAPDH . ( C ) Average expression of the data shown in ( B ) . Expression levels are compared to TERTΔ/Δ wt hESC clones . Error bars represent the SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 07918 . 006 As the first deletion step removed the translational start site as well as some coding sequence of TERT , this targeting strategy resulted in telomerase-negative hESCs ( TERTΔ/Δ ) ( Figure 1C ) . Correct targeting was confirmed by Southern blot and PCR-sequencing of the genomic deletion ( Figure 1—figure supplement 1A , B ) . As expected from our previous characterization of TERT−/− hESCs ( Sexton et al . , 2014 ) , these cells proliferated normally for 3 to 4 months , followed by cell death due to progressive telomere shortening with no survivors after 140 days ( Sexton et al . , 2014 ) . In a second targeting step we edited the newly formed genomic site with a specific sgRNA spanning the new junction ( −1462 to +67 ) to reinsert and restore the deleted region with either the wild-type promoter or an altered region containing the most frequent cancer-associated TERT promoter mutations: −57A/C , −124C/T , or −146C/T ( Figure 1B ) . This complementation approach restored the TERT gene and cellular viability of targeted cells . Accordingly , cells with a restored TERT gene gradually outcompeted none-rescued parental TERTΔ/Δ cells and lead to substantial telomere elongation by the time untargeted TERTΔ/Δ hESCs had died ( Figure 1D ) . This complementation strategy therefore successfully generated hESCs that differed exclusively at the TERT locus by expressing TERT either from its wild-type promoter or from a promoter that contained one of the cancer-associated point mutations . We first analyzed the impact of the TERT promoter mutations on TERT mRNA levels by qRT-PCR until cultures established stable TERT expression levels and all TERTΔ/Δ had died ( Figure 1E ) . This analysis revealed that the most frequent TERT mutation ( −124C/T ) resulted in a 2–3-fold increase in TERT expression when compared to the isogenic wild-type control . This increase is in agreement with previous reports that evaluated these mutations using Luciferase reporter constructs ( Horn et al . , 2013; Huang et al . , 2013 ) . Noticeably , the two other mutations did not result in similarly increased TERT expression in hESCs . We confirmed this finding for an independent TERTΔ/Δ cell line ( Figure 1—figure supplement 2 ) and individual single cell-derived hemizygously targeted clones ( Figure 1—figure supplement 3 ) . Interestingly , the promoter mutation cell lines carrying the −124C/T mutation had elevated levels of TERT mRNA expression , without a equivalent increase in telomerase activity ( Figure 1F ) . A lack of a significant change in telomerase activity despite increased TERT levels in hESCs that carry the −124C/T mutations suggested that TERT mRNA levels are not rate-limiting for telomerase activity in hESCs . Similar observations were made previously for some tumor cell lines in which telomerase activity is limited by levels of TR ( Cristofari and Lingner , 2006; Xu and Blackburn , 2007 ) , and might explain the tissue-specific impact of TERT and TR mutations in patients with dyskeratosis congenita ( Batista et al . , 2011; Strong et al . , 2011; Armanios and Blackburn , 2012 ) . Telomerase biogenesis is a complex biological process that has been shown in human pluripotent stem cells to depend on several activities that , when depleted , can become limiting ( Yang et al . , 2008; Batista et al . , 2011; Batista and Artandi , 2013 ) . To test the hypothesis that in wild type human pluripotent stem cells TR is the limiting factor for telomerase activity , we ectopically expressed TERT , TR , or both from the AAVS1 safe harbor locus ( Hockemeyer et al . , 2009 , 2011 ) ( Figure 2A ) . The introduction of such transgenes into this locus in isogenic settings overcomes concerns of random integration of the transgene . Overexpression levels were verified by western and northern blotting and qRT-PCR ( Figure 2B , C; Figure 2—figure supplement 1A ) and quantitative analysis showed that TERT mRNA was overexpressed >40 fold and TR levels by approximately 20 fold . TERT protein was detectable by immunoblotting when overexpressed , contrasted to the lack of detectable endogenous TERT protein . In addition , we determined telomerase activity levels ( Figure 2D and Figure 2—figure supplement 1B , C ) and telomere length changes in hESCs 36 days after targeting ( Figure 2E ) . TR overexpression strongly increased telomerase activity and led to rapid telomere elongation in hESCs , while overexpression of TERT alone did neither . However , when differentiated into fibroblasts or neural precursor cells ( NPCs ) , we observed the inverse behavior . In this setting , telomerase activity levels were significantly increased when TERT was overexpressed while increased levels of TR did not affect telomerase activity ( Figure 2D ) . This finding showed that in hESCs TERT levels were not limiting , and that increased TERT expression did not result in a significant increase of telomerase activity or telomere length . Hence , hESCs are unlikely to reveal the impact of the TERT promoter mutations . Therefore , observation of the effect of TERT promoter mutations requires the analysis of differentiated cells in which TERT down-regulation results in it becoming limiting for telomerase activity . 10 . 7554/eLife . 07918 . 007Figure 2 . Telomerase activity is restricted by levels of TERT in differentiated cells while TR is limiting in wild-type hESCs . ( A ) Targeting schematic of GFP , 3XFLAG-TERT ( F-TERT ) , TR , and F-TERT+TR overexpression from the AAVS1 locus in wild-type hESCs . ( B ) Northern blot detection of total TR and 7SL in targeted hESC lines . TR runs as a doublet in UREA PAGE . ( C ) SDS-PAGE immunoblot of total TERT and tubulin proteins in editing hESC lines from whole cell extract . ( D ) TRAP assay of whole cell extracts from NPCs and fibroblast-like cells differentiated from GFP ( G ) , F-TERT ( T ) , TR ( R ) , or F-TERT+TR ( T&R ) overexpressing hESCs using 200 ng protein . ( E ) Telomere restriction fragment assay of GFP , F-TERT , TR , and F-TERT+TR overexpressing hESCs . DOI: http://dx . doi . org/10 . 7554/eLife . 07918 . 00710 . 7554/eLife . 07918 . 008Figure 2—figure supplement 1 . Quantification of TERT and TR expression levels and telomerase activity in the overexpression hESCs . ( A ) Quantitative RT-PCR of TERT , TR , and OCT4 in GFP , F-TERT , TR , or F-TERT+TR overexpressing hESCs . Expression level is relative to GFP hESCs and normalized to GAPDH . Error bars represent the SEM of three biological replicates taken 1 week apart . ( B ) TRAP assay of whole cell extracts from GFP , F-TERT , TR , or F-TERT+TR overexpressing hESCs using decreasing amount of protein ( 250 , 50 , 10 , 2 ng ) . ( C ) Relative telomerase activity of the hESC lines was assayed by QTRAP . Values were set as fold activity relative to the GFP control line ( n = 3 , ANOVA with Tukey's test , SEM bars ) . R: TR , T: F-TERT and T&R: F-TERT+TR . DOI: http://dx . doi . org/10 . 7554/eLife . 07918 . 008 We differentiated edited hESCs into embryonic bodies ( EBs ) and eventually into fibroblasts and determined TERT mRNA levels over a 15 day differentiation period ( Figure 3A ) . All cell lines differentiated with equal efficiencies , as evidenced by up-regulation of the differentiation marker COL1A1 and repression of OCT4 transcription ( Figure 3B , C ) . Although TERT expression was successfully down-regulated in cells with wild-type TERT promoter , all three promoter mutation lines retained significant levels of TERT expression ( Figure 3A ) . This failure of TERT transcriptional silencing became apparent as early as 3 days after the induction of differentiation and accumulated into a fourfold increase in TERT expression in cells that carried the −57A/C or the −146C/T mutation and an 8–12-fold increase in cells in which transcription depended on the endogenous TERT promoter with the −124C/T mutation ( Figure 3A , B ) . This failure to appropriately repress TERT transcription during EB differentiation became even more apparent when the cells were differentiated into fibroblast-like cells . As expected , TERT transcription was undetectable in differentiated wild-type fibroblasts . In contrast , fibroblasts with the cancer-associated promoter mutations showed high levels of TERT expression ( Figure 3D ) . This difference was not due to impaired differentiation of cells with the TERT promoter mutations , as these cells had silenced OCT4 and appropriately induced COL1A1 expression ( Figure 3D ) . Importantly , while telomerase activity is not detectable in wild-type fibroblasts , the aberrant TERT expression resulted in robust telomerase activity in fibroblasts that contained the promoter mutations ( Figure 3E ) . As before , we confirmed these findings in an independent TERTΔ/Δ cell line ( Figures 3—figure supplement 1A–D ) and in individual single cell-derived targeted clones ( Figures 3—figure supplement 1E , F ) . Furthermore we confirmed that this failure to repress TERT expression persists in fibroblasts as late as 45 days after differentiation ( Figures 3—figure supplement 2 ) . 10 . 7554/eLife . 07918 . 009Figure 3 . Fibroblasts carrying cancer-associated TERT promoter point mutations failed to silence TERT expression upon differentiation and have telomerase activity . ( A ) , ( B ) and ( C ) Relative expression level of TERT , OCT4 or COL1A1 in the promoter-mutated hESC-derived fibroblasts compared to WT hESCs over a time course of differentiation ( left panel ) . Relative expression level of TERT , OCT4 or COL1A1 compared to TERTΔ/Δ and WT fibroblasts over a time course of differentiation . The right panel shows the same data as in the left panels , normalized to TERTΔ/Δ wt fibroblasts . Expression of TERT , OCT4 , or COL1A1 was normalized to GAPDH . ( D ) Quantitative RT-PCR of TERT , OCT4 , COL1A1 , and GAPDH in the fibroblasts carrying the promoter mutations 24 days after differentiation . Expression level is relative to WT hESCs . ( E ) TRAP assay of whole-cell extracts from WT hESCs , and the fibroblasts carrying the TERT promoter mutations ( 24 days after differentiation ) using 2 µg of protein . DOI: http://dx . doi . org/10 . 7554/eLife . 07918 . 00910 . 7554/eLife . 07918 . 010Figure 3—figure supplement 1 . Independent confirmation of the failure of TERT repression and telomerase activity upon fibroblast differentiation shown in Figure 3 using an independent TERTΔ/Δ#2 cell line . The results obtained by the bulk analysis were also confirmed by clonal analysis of fibroblasts carrying the TERT promoter mutations . ( A ) and ( B ) Relative expression levels of TERT and OCT4 in the promoter-mutated hESC-derived fibroblasts compared to WT hESCs over a time course of differentiation ( left panels ) . The right panels show the same data normalized to TERTΔ/Δ #2 wt cells . ( C ) Relative expression level of COL1A1 in the promoter-mutated hESC-derived fibroblasts compared to WT fibroblasts ( day 15 ) over a time course of differentiation ( left panels ) . The right panels show the same data normalized to TERTΔ/Δ #2 wt cells . ( D ) Relative expression level of BRACHYURY ( T ) in the promoter-mutated hESC-derived fibroblasts compared to WT embryonic bodies ( day 3 ) over a time course of differentiation ( left panels ) . The right panels show the same data normalized to TERTΔ/Δ #2 wt cells . ( E ) Quantitative RT-PCR of TERT , OCT4 , and GAPDH in individual clones of fibroblasts carrying the mutations . Expression levels are shown relative to WT hESCs . ( F ) TRAP assay of whole cell extracts from fibroblasts differentiated from clonal hESCs carrying the TERT promoter mutations using 2 µg of protein . DOI: http://dx . doi . org/10 . 7554/eLife . 07918 . 01010 . 7554/eLife . 07918 . 011Figure 3—figure supplement 2 . The failure of TERT repression in fibroblasts was retained throughout long-term culture . ( A ) , ( C ) and ( D ) Relative expression level of TERT , OCT4 , COL1A1 , and GAPDH in individual hESC clones differentiated to fibroblasts carrying the indicated mutations at days 28 , 40 , and 45 after differentiation . Expression levels are shown relative to WT hESCs . ( B ) Average expression of data shown in ( A ) . Expression levels are compared to WT hESC clones . Error bars represent the SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 07918 . 011 The failure to silence telomerase could be specific to the fibroblast differentiation paradigm or a more general defect during differentiation . To address this issue , we first generated NPCs using the highly robust dual SMAD inhibition protocol ( Chambers et al . , 2009 ) , establishing NPCs that can be maintained in culture for extended periods of time with low levels of telomerase expression . These NPCs can be further differentiated towards terminally differentiated non-proliferating post mitotic neurons that are characterized by the expression of the pan-neural marker proteins TUJ1 and NEUN . Independent of their genotype , all cells were able to differentiate into NPCs and neurons showing equal down-regulation of OCT4 expression and induction of neuronal marker genes ( Figure 4A–C ) . However , a striking difference became apparent in TERT levels as both NPCs and neurons that carried the TERT promoter mutations failed to repress TERT transcription ( Figure 4A , C and Figures 4—figure supplement 1 ) and showed robust telomerase activity ( Figure 4D ) . Even when neurons were maintained in the presence of a mitotic inhibitor , the promoter mutations led to elevated TERT mRNA and telomerase activity levels , suggesting that telomerase activity can accumulate in slowly and non-dividing cells as late as 1 month after induction of terminal differentiation ( Figure 4C , D ) . 10 . 7554/eLife . 07918 . 012Figure 4 . Neural precursors and neurons differentiated from the promoter mutation hESCs failed to repress TERT and telomerase activity . ( A ) Quantitative RT-PCR of GAPDH , TERT , NESTIN , and OCT4 in the neural precursors carrying the promoter mutations 20–25 days after differentiation from hESCs . Expression levels are relative to the WT hESCs . ( B ) Phase-contrast and immunofluorescence images of neurons differentiated from wild-type hESCs or the TERT promoter mutation-containing hESCs . Shown are cells 28 days after neural induction from NPCs and treated with mitotic inhibitor for 16 days . The left panel shows IF staining against NeuN ( red ) , Tuj1 ( green ) , and DAPI staining ( blue ) . ( C ) Quantitative RT-PCR of GAPDH , TERT , TUJ1 , and OCT4 in the neurons carrying the promoter mutations . The top panel shows expression levels of neurons 7 days after neuronal differentiation from NPCs . The bottom panel shows expression levels of neurons 28 days after induction of neuronal differentiation from NPCs and treated with mitotic inhibitor for 16 days . Expression level is relative to the WT hESCs . ( D ) TRAP assay of whole cell extracts from NPCs ( 35 days after differentiation from hESCs ) and neurons ( 28 days after neuronal differentiation from NPCs and treated with mitotic inhibitor for 16 days ) using 1 µg protein . DOI: http://dx . doi . org/10 . 7554/eLife . 07918 . 01210 . 7554/eLife . 07918 . 013Figure 4—figure supplement 1 . Clonal analysis of TERT promoter mutation NPCs confirmed results from bulk analysis . ( A ) Quantitative RT-PCR of TERT , OCT4 , NESTIN , and GAPDH in NPCs differentiated from individual clones of the targeted hESCs ( 28 days after differentiation from hESCs ) . Expression levels are shown relative to WT hESCs . ( B ) Average expression of data shown in ( A ) . Expression levels are compared to WT hESCs . Error bars represent the SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 07918 . 013 Next , we assessed the impact of the aberrant telomerase activity of TERT promoter-mutation-containing cells by directly comparing the telomerase levels in fibroblasts and NPCs that carried the promoter mutations to the telomerase activity found in three established tumor cell lines ( Figure 5A , B ) . Remarkably , telomerase activity levels in −124C/T NPCs were equivalent to the activity found in immortal Hela S3 cells and about 50% of the activity found in hESCs , HCT116 , and 293T cells . Importantly , telomerase activity in these cells was greatly increased relative to wild type cells . Moreover , −124C/T fibroblasts had about 50% of the telomerase activity measured in HeLa S3 cells , 30% of that found in 293T cells , and 25% of that of HCT116 colon carcinoma cells . These findings suggested that TERT promoter mutations induce telomerase levels that are sufficient to enable immortalization or at least significantly delay telomere length-induced senescence . Finally , we analyzed the functional consequences of increased TERT expression by evaluating telomere length changes in hESCs , NPCs , and fibroblasts derived from the isogenic set of TERT promoter-edited cell lines ( Figure 5C , D , and Figure 5—figure supplement 1A ) . All differentiated cell lines with cancer-associated promoter mutations showed an increase in telomere length compared to the wild-type controls . 10 . 7554/eLife . 07918 . 014Figure 5 . Fibroblasts and neural precursors carrying cancer-associated TERT promoter point mutations showed comparable telomerase activity to cancer cell lines , and telomere length was maintained over long-term culture and tumor development . ( A ) and ( B ) TRAP assay of whole-cell extracts from cancer cell lines ( HeLaS3 , 293T , and HCT116 ) , WT hESCs , and the NPCs or fibroblasts carrying the TERT promoter mutations ( 24 days after differentiation for fibroblasts and 20 days for NPCs ) using decreasing amount of protein ( 200 , 40 , 8 ng ) . For comparison , TRAP samples in ( A ) and ( B ) were prepared simultaneously and samples from cancer cell lines are identical in ( A ) and ( B ) . TRAP signals relative to HeLa S3 were quantified and are shown at the bottom of the gels . ( C ) Telomere restriction fragment assay of the hESCs and NPCs ( 65 days after differentiation from hESCs ) . Median telomere length signals were quantified and shown at the bottom . It is important to note the telomere shortening in NPCs and fibroblasts in wild type cells exceeds the initial telomere length difference found in the hESCs . ( D ) Telomere restriction fragment assay of the fibroblasts ( 30 days after differentiation from hESCs ) . Median telomere length signals were quantified and shown at the bottom . ( E ) Telomere restriction fragment assay of teratoma tumor tissue generated from wt and promoter-mutation containing hESCs ( 75 days after injection ) . Median telomere length signals were quantified and shown at the bottom . DOI: http://dx . doi . org/10 . 7554/eLife . 07918 . 01410 . 7554/eLife . 07918 . 015Figure 5—figure supplement 1 . Fibroblasts and teratoma tissue carrying cancer-associated TERT promoter point mutations maintained telomere length over long-term culture and tumor development . ( A ) Telomere restriction fragment assay of the fibroblasts ( 36 days after differentiation from hESCs ) . ( B ) Telomere restriction fragment assay of teratoma tumor tissue generated from wt and promoter-mutation containing hESCs . Teratoma tissues in the group #1 were explanted 84 days after injection except −57 mutant samples ( 64 days after injection ) . Teratoma tissues in the group #2 were explanted 56 days after injection . DOI: http://dx . doi . org/10 . 7554/eLife . 07918 . 015 To further explore the in vivo relevance of these findings in the context of long-term differentiation as well as tumor progression , cells with TERT promoter mutations were assayed for teratoma tumor formation in immune-compromised mice ( Figure 5E and Figure 5—figure supplement 1B ) . For this assay all cell lines were injected subcutaneously into NOD/SCID mice , allowing pluripotent cells to differentiate and form a teratoma comprised of cells derived from all three germ layers . All cell lines injected formed teratomas of approximately equal size , were explanted simultaneously ( 75 days after injection ) , and analyzed for their telomere length . Using this unbiased approach , we again found that cells with TERT promoter mutations carried aberrantly long telomeres , with the −124C/T mutation having the strongest defect in silencing telomerase activity and retaining almost identical telomere length as undifferentiated hESCs ( Figure 5C , E and Figure 5—figure supplement 1B ) . These data demonstrated the causal relationship between the TERT promoter mutations and telomere maintenance and showed that the TERT promoter mutations can up-regulate TERT levels sufficiently to suppress telomere erosion without additional tumor-selected changes .
A key challenge in cancer research is to understand how mutations that sequentially occur in normal cells eventually produce a tumor . For noncoding mutations identified by GWAS , this is a particular challenge . Here we were able to dissect how the most frequent noncoding mutations in human cancer exert their tumorigenic effect . We are able to do so because cancer genomics have identified candidate mutations , genome editing is facile and robust , and we tested the effects in an otherwise wild-type background , thus being able to attribute a phenotypic effect specifically to a single genetic change . Using genome editing of the endogenous TERT locus we generated a panel of three hESC lines that differed exclusively at a single position in the TERT promoter associated with cancer . Analyzing the impact of these mutations in hESCs , we showed that the most frequent mutation , −124C/T , increased TERT mRNA levels in hESCs by about 2–3 fold . However , neither the −57A/C nor the −146C/T mutation led to an increase in TERT transcription , despite the fact that these mutations generate the same putative ETS-binding motif of TTCCGG . This suggests a strong positional effect between the location of the ETS mutation and the core transcription initiation machinery . The possibility of such context-dependent positional constraints between the TERT promoter mutations and the core transcriptional machinery is further supported by the fact that a single point mutation at position −89 ( C/G , −31 bp upstream of the TSS ) would result in the same TTCCGG sequence . The fact that this mutation has as of now not been reported to be associated with cancer despite it being between the −57 and −124 sites is likely due to the need of the core transcription factors to bind to this site . It is intriguing to speculate why all promoter mutations are located in close proximity to the TSS; it seems possible that TERT transcriptional regulation is closely linked with the core transcriptional machinery rather than regulated through the canonical positioning of transcription factors along an extended promoter . The experimental approach established in this study of genome editing the TERT promoter provides an experimental system to uncover cis-regulatory elements that are necessary for telomerase expression in stem cells and its transcriptional regulation upon differentiation . Increased expression of TERT mRNA in −124C/T containing hESCs did not lead to a significant increase in telomerase activity or pronounced telomere lengthening , establishing that in hESCs TR levels , but not TERT levels , are limiting for telomerase assembly and telomere lengthening . Therefore , immortal hESCs are as uninformative with regard to cancer-associated TERT mutations as immortal tumor tissue or cell lines . However , whereas upon differentiation wild-type hESCs efficiently silence TERT transcription , resulting in loss of telomerase activity and telomere shortening , the cancer-associated TERT promoter mutations were sufficient to maintain expression of TERT and resulted in telomerase activity levels comparable to immortal cancer cell lines . These experiments uncover that the underlying cancer-causing mechanism is likely a failure to repress telomerase upon differentiation into somatic cells . It is remarkable that TERT promoter mutations are sufficient to up-regulate TERT expression without additional cancer-selected changes in the genome such as increased levels of ETS factors . TERT promoter mutations are not frequently found in leukemias and colorectal cancers ( Heidenreich et al . , 2014 ) . Direct evidence ( Chiu et al . , 1996; Schepers et al . , 2011 ) as well as the pathology of the telomerase-related disease dyskeratosis congenita in which patients with mutations in the telomere maintenance pathway present with bone marrow failure as well as lung , intestinal , and skin pathologies show that TERT is expressed in these highly proliferative tissues and is required for their long-term self-renewal capacity and ability to maintain tissue homeostasis ( Armanios and Price , 2012; Aubert et al . , 2012 ) . Tumor-initiating events in these cancers predominantly drive proliferation pathways that spur formation of hyperplasia and niche-independent proliferation that allow incipient cancer cells to outcompete their neighbors ( Barker et al . , 2009; Zhou et al . , 2009; Merlos-Suarez et al . , 2011; Magee et al . , 2012 ) . In this setting , mutations in the TERT promoter or alterations in the telomerase biogenesis pathway might be at first neutral , not providing a direct proliferative advantage as telomeres are still long or telomerase is active ( Figure 6A ) . During the genesis of these tumors , telomere shortening might present a challenge at a later stage when cells have already outcompeted their neighbors . 10 . 7554/eLife . 07918 . 016Figure 6 . Model explaining the tumor spectrum associated with TERT promoter mutations . Shown are the differential outcomes of a cell acquiring a cancer-associated TERT promoter mutation or a proliferation-inducing mutation dependent on telomere length of the cell . ( A ) In a cell with long telomeres and telomerase activity , a proliferation-promoting mutation will result in a strong proliferative advantage and can act as the tumor-initiating event . Cells with long telomeres arise from tissues that have a telomerase positive stem cell compartment such as the hematopoietic or intestinal system . In contrast , mutations in the TERT promoter do not provide a proliferative advantage , they are neutral and do not promote tumor formation . Cell states are depicted on the left; cells that acquire mutations are shown in red . A schematic depicting telomere length changes as a function of the number of cell divisions is shown on the right . The dashed line indicates the critical telomere length at which cells are subjected to the Hayflick limit and stop proliferating or die . The red line indicates the telomere length changes predicted for cells that acquire either a proliferation-promoting mutation ( top ) or a TERT promoter mutation ( bottom ) . The blue line indicates the telomere length changes predicted for wild-type cells . Indicated is a case where telomere shortening is suppressed by the TERT promoter mutations . However , since these cell already have long telomeres and/or naturally express telomerase , telomeres in neither wild-type cells or cells acquiring a proliferation inducing mutation will shorten to the point that the cells are subjected to the Hayflick limit . ( B ) Schematic as shown in ( A ) but for a telomerase negative cells with short telomeres . A proliferation-promoting mutation will also provide a growth advantage in telomerase-negative differentiated cells with short telomeres , however , these cells will enter replicative senescence or die . In contrast , a cell with short telomeres acquiring a TERT promoter mutation can bypass the Hayflick limit ( dashed lines ) , immortalize , and outcompete its neighboring cells . Cell states are depicted on the left; cells that acquire mutations are shown in red . Schematic depicting telomere length changes as a function of the number of cell divisions is shown on the right . The orange line indicates the telomere length changes predicted for wild-type cells . The table to the right shows the frequency of TERT promoter mutations found in different types of tumors ( adapted form Heidenreich et al . , 2014 ) . The table includes references that report the specific tumor subtypes and frequencies used to generate this table . DOI: http://dx . doi . org/10 . 7554/eLife . 07918 . 016 In contrast , tumor-initiating cells that are thought to not directly arise from a canonical telomerase-positive stem cell compartment ( e . g . , liposarcomas ) , that undergo high numbers of divisions after differentiating ( e . g . , neural-crest derived melanocytes ) , or that have to reenter a proliferation cycle in response to chronic injury ( e . g . , urothelial cells and hepatocytes ) could be challenged by the telomere-dependent proliferative barrier comparatively early in their progression . In these cell types TERT promoter mutations will provide an immediate and strong proliferative advantage over neighboring cells . In this case telomerase activation occurs in cells in which telomerase is absent or low and which have an otherwise mostly intact genome . In these cells activating TERT promoter mutations will be present in most tumor cells and detected as a frequent and thus early event ( Figure 6B ) . It is important to note that this model depicts the very extreme cases of a TERT-positive adult stem cell with long telomeres contrasted to the expected outcome of a TERT promoter mutation in a telomerase-negative cell with short telomeres . Likely this sharp distinction between canonical telomerase-positive stem cell compartments and telomerase-negative compartments is rather continuous in vivo . Telomerase expression , telomere length , and the number of cell divisions will differ between tissues and with age and therefore the benefit of the TERT promoter mutation will be complexly graded . Given this , it will be critical to determine exactly which cells of the human body are telomerase-positive , when and how telomerase is silenced upon differentiation , and how many divisions cells undergo in human tissue after becoming telomerase-negative . Telomerase inhibition has been proposed as a target for cancer therapies . We demonstrate that TERT promoter mutations are sufficient to de-repress TERT , providing a potential target to inhibit TERT expression and telomerase activity . In order to identify therapeutic approaches specific to these promoter mutations , a model system in which TERT is dysregulated solely by these mutations is necessary . Our model system fulfills this requirement and allows for a direct assessment of any potential inhibition by measuring TERT expression following differentiation . In contrast , this approach will be challenging in cancer cells , as TERT mRNA levels , telomerase levels , and telomere length vary dramatically regardless of whether they carry any of the TERT promoter mutations . Further mechanistic studies in such tumor cells are also challenged by the high frequency of concurrent TERT copy number variations , promoter polymorphisms , and cancer-associated dysregulation of factors implicated in TERT regulation such as MYC . As such , it will be challenging to evaluate the effectiveness of such an inhibitor due to these potentially compensatory effects arising from these misregulations . As such , it is imperative to test any potential therapeutic approach directed at these promoter mutations in a model system that only carries these mutations in an otherwise wild-type background , such as the model system described here . Specifically targeting the TERT promoter mutations is an attractive approach , as TERT promoter mutations are exclusive to the tumor cells and are not present in surrounding normal tissue . Therefore , any intervention that is targeted specifically against their mode of operation is expected to affect tumor cell survival , but not the telomerase-positive adult stem cells of the patient .
Genome-editing experiments were performed in WIBR#3 hESCs ( Lengner et al . , 2010 ) , NIH stem cell registry # 0079 . Cell culture was carried out as described previously ( Soldner et al . , 2009 ) . Briefly , all hESC lines were maintained on a layer of inactivated mouse embryonic fibroblasts ( MEFs ) in hESC medium ( DMEM/F12 [Lifetech] ) supplemented with 15% fetal bovine serum [Lifetech] , 5% KnockOutTM Serum Replacement [Lifetech] , 1 mM glutamine [Lifetech] , 1% non-essential amino acids [Lifetech] , 0 . 1 mM β-mercaptoethanol [Sigma] , 1000 U/ml penicillin/streptomycin [Lifetech] , and 4 ng/ml FGF2 [Lifetech] . Cultures were passaged every 5–7 days either manually or enzymatically with collagenase type IV [Lifetech] ( 1 . 5 mg/ml ) and gravitational sedimentation by washing 3 times in wash media ( DMEM/F12 [Lifetech] supplemented with 5% fetal bovine serum [Lifetech] , and 1000 U/ml penicillin/streptomycin [Lifetech] ) . For the formation of EBs hESC colonies were grown on petri dishes in fibroblast medium ( DMEM/F12 [Lifetech] ) supplemented with 15% fetal bovine serum [Lifetech] , 1 mM glutamine [Lifetech] , 1% non-essential amino acids [Lifetech] , and penicillin/streptomycin [Lifetech , Carlsbad , CA] . After 9 days EBs were transferred to tissue culture dishes to attach . Fibroblast-like cells were passaged with Trypsin EDTA ( [Lifetech] , 0 . 25% ) , triturated into a single-cell suspension and plated on tissue culture dishes . Cultures were maintained in fibroblast media and passed every 6 days . Before differentiation to NPCs , hESCs were cultured under feeder-free conditions on matrigel [Corning]-coated plates in E8 media ( DMEM/F12 [Lifetech] ) supplemented with 64 µg/ml L-ascorbic acid [Sigma] , 19 . 4 µg/ml insulin [Sigma , St . Louis , MO] , 14 µg/l sodium selenite [Sigma] , 543 ng/l sodium bicarbonate [Sigma] , 1000 U/ml penicillin/streptomycin [Lifetech] , 100 ng/ml FGF2 [Lifetech] , and 10 . 7 µg/ml Transferrin [Sigma] . hESCs were passaged with accutase [Invitrogen] and triturated to a single-cell solution and plated on matrigel-coated plates at 50 , 000 cell/cm2 . The dual SMAD inhibition protocol for the differentiation of hESCs to NPCs was adapted from Chambers et al . ( 2009 ) . Differentiation was induced when cells reached 90–100% confluency . NPCs were maintained in N2 media ( 50% DMEM/F12 [Lifetech] , 50% Neurobasal Media [Lifetech] supplemented with 0 . 75% BSA ( wt/vol ) [Sigma] , N2 Supplement [Lifetech] , 20 ng/ml insulin [Sigma] , 1 mM glutamine [Lifetech] , 1000 U/ml penicillin/streptomycin [Lifetech] , 25 ng/ml FGF2 [Lifetech] and 40 ng/ml EGF [R&D systems] ) and passaged every 5 days . For the terminal differentiation to neurons NPCs were plated at 50 , 000 cells/cm2 on matrigel-coated plates in N2B27 media ( 50% DMEM/F12 [Lifetech] , 50% Neurobasal Media [Lifetech] supplemented with 0 . 75% BSA ( wt/vol ) [Sigma] , N2 Supplement [Lifetech] , B27 Supplement [Lifetech] , 1 mM glutamine [Lifetech] , 1000 U/ml penicillin/streptomycin [Lifetech] ) . Neurons were treated with 250 nM mitotic inhibitor ( Cytosine-β-D-arabinofuranoside [Sigma] ) . All targeting experiments were preformed as previously described ( Hockemeyer et al . , 2009 , 2011 ) . CAS9 and all sgRNAs were expressed using the px330 plasmid ( Cong et al . , 2013 ) . Cancer associated TERT promoter mutation containing cell lines were generated by two targeting steps . First , 1–2 × 107 hESCs were co-electroporated with 15 µg of two CAS9 plasmids targeting −1418 to −1399 bp ( aaccgcccctttgccctag ) and +110 to +129 bp ( taccgcgaggtgctgccgc ) from the TSS and 7 . 5 µg of a GFP-expression plasmid was electroporated along with px330 . Cells were sorted for GFP fluorescence 72 hr after electroporation . Single-cell derived hES colonies were isolated and their targeting was confirmed by Southern blotting and PCR followed by sequencing . 120 clones were analyzed and three homozygous targeted hESC lines ( TERTΔ/Δ ) were obtained . For the second targeting , px330 plasmids were designed with sgRNAs against the newly formed NHEJ-derived junction site in TERTΔ/Δ cells and electroporated with 35 µg of a repair plasmid that carried either the wild type TERT promoter element ( wt ) or the respective TERT promoter mutations ( 57A/C , 124 C/T , 146C/T ) . After the second targeting , cells were continuously passaged . Over a period of 120 days all TERTΔ/Δ lines that did not undergo the second targeting step died due to critically short telomeres . However , cells that were correctly targeted in the second targeting step regained TERT expression and outgrew untargeted cells . These cells were analyzed in bulk or as single cell derived clones , after the parental TERTΔ/Δ control culture , that did not undergo the second targeting step , had completely died . Targeting of individual clones was confirmed by Southern blot analysis . RNA was extracted with TRIzol ( Lifetech ) and treated with DNaseI ( NEB ) . 600 ng RNA were converted to cDNA with the iScript Reverse Transcriptase ( BioRad ) and random and poly A priming . TR cDNA was prepared by gene specific reverse transcription . qRT-PCR was performed with KAPA SYBR fast [KAPA Biosystems] or SYBR Select Master Mix ( ABI ) in 96-well or 384-well format with a total reaction volume of 20 µl or 10 µl respectively . 2 µl cDNA from the iScript reaction mixture was used for the detection of TERT mRNA . For measuring the expression levels of all other genes , cDNA was diluted 1:10 and 2 µl were used for qPCR . Due to different expression levels of GAPDH between hESCs and differentiated cells , GAPDH data are shown in the figures that required comparison of expression in different cell types . Relative expression levels were calculated based on Δ/Δ Ct and/or ΔCt analysis . qRT-PCR primers used in this study are summarized in Supplementary file 1 . For analysis by IF , cells were briefly rinsed with PBS , and fixed with 4% formaldehyde in PBS . Cells were blocked with PBS 0 . 3% Triton X-100 with 5% horse serum . Fixed cells were incubated with antibodies against NEUN ( mouse , monoclonal , [Millipore] , MAB377; 1:1500 ) and TUJ1 ( B-III-Tubulin , chicken , polyclonal , [Millipore] , AB9354 , 1:500 ) , in PBS 0 . 3% Triton X-100 with 1% BSA over night . After washing with PBS the cells were stained with secondary antibodies ( Alexa Fluor 546 goat α mouse , Alexa Fluor 488 goat α chicken [Lifetech]; 1:500 ) , for 1 hr in PBS 0 . 3% Triton X-100 with 1% BSA . Cells were then washed with PBS and stained with 1 ng/µl DAPI ( Sigma ) in PBS . RNA for northern blot was purified using TRIzol according to the manufacturer's protocol ( Lifetech ) . Northern blot detection of TR was performed as previously described ( Fu and Collins , 2003 ) . 7SL RNA was detected using 32P end-labeled probe ( TGAACTCAAGGGATCCTCCAG ) under similar conditions as TR , except hybridization took place at 37°C . Southern blots analysis was performed as previously described ( Hockemeyer et al . , 2009 , 2011 ) using a 3′- probe for TERT ( 6280 bp −6846 bp downstream of the TERT first ATG ) and probe T1 ( amplified from hES genomic DNA with primers Fw: GTGACTCAGGACCCCATACC and Rev: ACAACAGCGGCTGAACAGTC ) . PCR-based telomeric repeat amplification protocol ( TRAP ) was performed as previously described using TS ( AATCCGTCGAGCAGAGTT ) and ACX ( GCGCGGCTTACCCTTACCCTTACCCTAACC ) for amplification of telomereic repeats and TSNT ( AATCCGTCGAGCAGAGTTAAAAGGCCGAGAAGCGAT ) and NT ( ATCGCTTCTCGGCCTTTT ) as an internal control ( Kim et al . , 1994 ) . Real-time quantitative telomeric repeat amplification ( QTRAP ) was performed similar to previously published protocols ( Wege et al . , 2003 ) . Cell extract was generated from CHAPS lysis and samples were normalized using the BCA Protein Assay Kit ( Pierce ) . 200 ng of total protein was used per 20 µl QTRAP reaction , which was composed of iTaq Universal SYBR Green Supermix ( Bio-Rad ) and 0 . 1 µg TS and 0 . 02 µg ACX primers . Samples were incubated at 30°C for 30 min before a 2 min 95°C hot-start and 35 cycles of 95°C for 15 s and 61°C for 90 s . Relative telomerase activity was calculated by ΔCt to the reference sample . After heating to 80°C for 5 min , protein samples were cooled to room temperature and resolved by SDS-PAGE . Protein was then transferred to nitrocellulose membrane and subsequently incubated with mouse α-tubulin ( 1:500 , DM1A , [Calbiochem] ) and mouse anti-TERT ( 1:3000 [Geron] ) in 4% nonfat milk ( Carnation ) in TBS buffer ( 150 mM NaCl , 50 mM Tris pH 7 . 5 ) overnight at 4°C . The membrane was washed in TBS and incubated with goat α-mouse Alexa Fluor 680 ( 1:2 , 000 , [Life Technologies] ) in 4% nonfat milk in TBS for 1 hr at room temperature . After extensive washing with TBS , the membrane was visualized on a LI-COR Odyssey imager ( Fu and Collins , 2003 ) . For preparation of genomic DNA , hESC lines were washed with PBS , released from the feeder cell layer by treatment with 1 . 5 mg/ml collagenase type IV and washed 3× in wash media by gravitational sedimentation to minimize contaminating MEF cells . Genomic DNA was then prepared as described previously ( Hockemeyer et al . , 2005 ) . While this method removes the vast majority of MEFs , the signal from mouse telomeres is disproportionate to human telomeres due to amplified relative length and concentration into a smaller area ( Kipling and Cooke , 1990 ) . Because MEF telomeres are size-resolved from human telomeres they do not interfere with analysis of hESC telomere length . Genomic DNA was digested with MboI and AluI overnight at 37°C . The resulting DNA was normalized and run on 0 . 75% agarose ( Seakem ME Agarose , Lonza ) , dried under vacuum for 2 hr at 50°C , denatured in 0 . 5 M NaOH , 1 . 5 M NaCl for 30 min , shaking at 25°C , neutralized with 1 M Tris pH 6 . 0 , 2 . 5 M NaCl shaking at 25°C , 2× for 15 min . Then the gel was pre-hybridized in Church's buffer ( 1% BSA , 1 mM EDTA , 0 . 5M NaP04 pH 7 . 2 , 7% SDS ) for 1 hr at 55°C before adding a 32P-end-labeled ( T2AG3 ) 3 telomeric probe . The gel was washed 3× 30 min in 4× SSC at 50°C and 1× 30 min in 4× SSC + 0 . 1% SDS at 25°C before exposing on a phosphorimager screen . Teratoma formation assays where performed as previously described in ( Hockemeyer et al . , 2008 ) . | The bulk of the DNA in the human genome is divided between 23 pairs of chromosomes . The ends of these chromosomes contain a repetitive stretch of DNA known as a telomere . Every time a cell divides , a portion of the telomere is lost and can be restored by an enzyme called telomerase . If the telomeres shorten below a critical length , the cell can no longer divide and eventually dies . Thus , long telomeres increase the number of times a cell can divide . In the majority of human cells—with the exception of stem cells—telomerase activity is absent due to the down regulation of the active protein component ( called TERT ) after birth . Therefore , the telomeres in these cells shorten after each cell division . However , 90% of human cancers have very high TERT activity , which enables them to divide continuously to drive tumor growth . Genes are sections of DNA that code for proteins and other molecules . The start of a gene contains a region known as the promoter , which controls when and where in the body the gene is active . Cancer cells often contain mutations in the promoter of the gene that encodes TERT . However , it remains poorly understood how these mutations lead to the formation of tumors . Chiba et al . have now used a technique called genome editing to introduce mutations that are commonly found in cancer cells into the promoter of the gene for TERT in human embryonic stem cells . Unexpectedly , these changes did not increase the activity of the telomerase enzyme in these cells , nor did they increase the length of the telomeres . Chiba et al . next caused these genetically engineered stem cells to develop into more specialized cell types—such as nerve cells . These ‘differentiated’ cells normally silence the gene that encodes TERT , but the mutations prevented the gene from being silenced . This led to abnormally high levels of telomerase activity and long telomeres . The experiments also showed that TERT activity in these cells was similar to that found in cancer cells that can divide indefinitely . Cells containing the promoter mutations were then injected into mice . The cells formed a mass of tumors that contained very long telomeres . These results together suggest that cancer-causing mutations in the gene for TERT stop this gene from being properly silenced in more specialized cells , and that this , on its own , can promote the formation of tumors . These findings are likely to underpin future efforts to treat cancers by targeting the expression and activity of the telomerase enzyme . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"developmental",
"biology"
] | 2015 | Cancer-associated TERT promoter mutations abrogate telomerase silencing |
An opioid epidemic is spreading in North America with millions of opioid overdoses annually . Opioid drugs , like fentanyl , target the mu opioid receptor system and induce potentially lethal respiratory depression . The challenge in opioid research is to find a safe pain therapy with analgesic properties but no respiratory depression . Current discoveries are limited by lack of amenable animal models to screen candidate drugs . Zebrafish ( Danio rerio ) is an emerging animal model with high reproduction and fast development , which shares remarkable similarity in their physiology and genome to mammals . However , it is unknown whether zebrafish possesses similar opioid system , respiratory and analgesic responses to opioids than mammals . In freely-behaving larval zebrafish , fentanyl depresses the rate of respiratory mandible movements and induces analgesia , effects reversed by μ-opioid receptor antagonists . Zebrafish presents evolutionary conserved mechanisms of action of opioid drugs , also found in mammals , and constitute amenable models for phenotype-based drug discovery .
The current North American opioid epidemic is staggering , with millions of emergency visits and over 45 , 000 deaths annually ( Florence et al . , 2016 ) . Synthetic opioids like fentanyl and oxycodone , or natural opioids like heroin and morphine are highly addictive ( Wilkerson et al . , 2016 ) and can lead to respiratory depression ( Dahan et al . , 2010; Montandon et al . , 2016a; Nagappa et al . , 2017 ) , that can be lethal with overdose ( Gomes and Juurlink , 2016; Jones et al . , 2013 ) . Opioids are critical pain therapies and there are currently no other medications that offer effective relief of pain ( IASP , 2018 ) . However , the side-effects of opioids limit their effective use and may lead to sub-optimal pain treatments . Opioid medications act on µ- , δ- , and κ- opioid receptors ( Gutstein , 2001 ) , which are expressed in discrete brain circuits . Although a wide range of opioids targeting these opioid receptors exist , the most potent opioids , such as fentanyl and oxycodone , are largely selective for the µ-opioid receptors ( MORs ) ( Sora et al . , 1997 ) . Because MORs are expressed in neural circuits involved in breathing and nociception , opioids present effects beyond their intended therapeutic analgesic properties . MOR drugs induce side-effects such as addiction ( Le Merrer et al . , 2009 ) , sedation ( Montandon et al . , 2016a ) , and hypoventilation ( Dahan et al . , 2010; Montandon and Slutsky , 2019; Nagappa et al . , 2017 ) . Hypoventilation is characterized by reduced respiratory rate ( Macintyre et al . , 2011 ) , reduced airflow ( Ferguson and Drummond , 2006 ) , prolonged apneas ( Mogri et al . , 2008; Nagappa et al . , 2017 ) , and severe hypoxemia ( Brown et al . , 2006 ) . Complete respiratory arrest and bradycardia are the main causes of death by opioid overdose ( Pattinson , 2008 ) . Importantly , respiratory depression and cessation of breathing with opioid overdose are due to the action of opioid drugs on brainstem respiratory network ( Montandon and Slutsky , 2019 ) . The only available treatment to reverse respiratory depression during an opioid overdose is the MOR antagonist naloxone which directly blocks the binding of opioid ligands to MORs . However , naloxone also blocks the analgesic properties of opioid drugs , so it cannot be used as an adjunct treatment . The discovery of novel opioid drugs or combinations of drugs with potent analgesic properties without the side-effects of respiratory depression has been hindered due to the lack of animal models allowing behavioral assessments of pain and breathing , while allowing for large-scale drug discovery . Identification of new molecular targets and drugs is critical to the development of safe opioid therapies . Due to the complexity of the machinery regulating MOR inhibition , wide scale screening of drugs acting on the known mechanisms regulating MOR inhibition in rodent models is not feasible . It is therefore critical to identify amenable model systems where large-scale screening can be developed , while preserving the complexity of vertebrate central nervous system and behaviours . The challenge is to quantify complex phenotypes like respiratory neural activity and nociception on a large scale , which is normally limited by the complexity of the animal models . Rodents are not optimal because their large size , generation time , and drug availability limit screen scale . An alternative is the larval zebrafish , which shares anatomy , physiology , and large parts of the genome with humans . The small size of larval zebrafish , ease of production , and their external fertilization which allows easy gene knockdown , makes the zebrafish an attractive model organism to screen genes and drugs . Here , we propose to use the larval zebrafish as simple , yet powerful , animal model , so that the molecular mechanisms of MOR respiratory and nociceptive inhibition can be identified . Importantly , the zebrafish larvae can also be used for large-scale drug screening to identify new drug candidates . The larval zebrafish has emerged as an ideal model system for drug and gene discovery since it combines the biological complexity of in vivo models with a nervous system homolog to humans including a respiratory neural network . Although fish use a different strategy to absorb oxygen and eliminate carbon dioxide than mammals , they rhythmically produce mandibular movements to move water through their gills . In lampreys , the respiratory neural network includes the paratrigeminal respiratory group ( pTRG ) and the vagal nucleus and this network generates rhythmic mandible movements ( Bongianni et al . , 2016 ) . Because of their close evolutionary origins ( Missaghi et al . , 2016 ) , the respiratory network shares similarities with the mammalian respiratory network ( Cinelli et al . , 2013 ) which generates breathing and regulates respiratory depression by opioids ( Montandon et al . , 2011 ) . Interestingly , the pTRG presents similar functional properties than the mammalian respiratory network such as sensitivity to substance P ( Mutolo et al . , 2010 ) and to opioid ligands ( Mutolo et al . , 2007 ) . Here , we propose that respiratory mandible movements can be used as an index of respiratory network activity , similar to respiratory activity of the trigeminal muscle in mammals ( Jacquin et al . , 1999 ) . The zebrafish also possesses a nociceptive system encompassing spinal cord , brainstem and sub-cortical circuits ( Taylor et al . , 2017 ) , with homology to mammals . Although pain circuit activity and its response to opioids cannot be directly assessed , the swimming escape response to nociceptive stimuli can be easily assessed in larval zebrafish . Nociceptive stimuli such as formalin ( Magalhães et al . , 2017 ) or ( allyl ) -isothiocyanate ( AITC , also known as ‘mustard oil’ ) can be administered to larval zebrafish . AITC acts on transient receptor potential ankyrin 1 channels and transient receptor potential cation channel vanilloid receptor 1 ( Oda et al . , 2016 ) which are involved in nociception to chemical compounds ( Bamps et al . , 2021 ) . Here , we propose to determine whether respiratory depression and analgesia can be measured in larval zebrafish . Most opioid analgesics and drugs of abuse elicit their effects by binding to MORs ( Gutstein , 2001 ) . In zebrafish , the MOR has high homology to the mammalian MOR and shares 74% of its amino acid sequence with its mammalian counterpart ( Sanchez-Simon and Rodriguez , 2008 ) . The MOR also possesses similar binding properties to the mammalian MOR for morphine and DAMGO ( Marron Fdez de Velasco et al . , 2009; Mutolo et al . , 2007 ) . Because of the homologies of the respiratory , pain , and opioid systems in zebrafish and mammals , we propose that the larval zebrafish may mimic respiratory depression and analgesia by opioids and other phenotypes associated with opioid drugs in mammals . The objectives of this study are to demonstrate that larval zebrafish have evolutionary conserved opioid properties mimicking mammals and that they can be used to investigate opioid-induced respiratory depression and analgesia . We aim to demonstrate that larval zebrafish replicates the opioid pharmacology observed in mammals and humans and that current pharmacotherapies to block respiratory depression can be tested in larval zebrafish .
Zebrafish use gills to promote gas exchange with water . Flow of water through the gills is generated by a complex respiratory network in the brainstem which presents similar anatomy and properties to the mammalian respiratory network ( Figure 1A ) . The respiratory network generates rhythmic movements and rhythmically activate muscles which promotes water flow through the gills . One of these muscles , the mandible muscle controls opening and closing of the mandible and is activated by the mandible nerve which originate from the trigeminal nerve . Mandible movements can be easily recorded using a high-definition camera positioned on top of the swimming zebrafish ( Figure 1B , C ) , and is a robust index of respiratory network activity . Mandible movements can be quantified by counting the changes in pixel intensity around the mandible area ( Figure 1B ) and can be displayed over time ( Figure 1D ) . The rate of mandible movements , identified as the number of mandible movement peaks ( red circles , Figure 1D ) , indicates respiratory network rhythm , and is considered here as an index of respiratory rate . Using the convenience of multi-well plates , we administered combinations of drugs to embryo medium as described in Figure 1E . Combinations of drugs were administered to separate groups of fish . Each animal only received one combination of drugs and did not receive consecutive combinations of drugs because it would not be feasible to remove drugs from the water . To determine whether zebrafish can be used as a model of opioid-induced respiratory rate depression , we administered embryo water ( control ) or fentanyl - a commonly used opioid analgesic - to larval zebrafish while quantifying the rate of mandible movements ( Figure 2A ) . Larval zebrafish under control conditions ( embryo water ) presented rates of mandible movements ranging from 15 to 115 movements per minute . Larvae were exposed to a concentration of fentanyl of 1 µM and respiratory rate was recorded over a 30 min time-period ( Figure 2B ) . In the control group , respiratory rate initially increased compared to baseline but did not significantly change during the 30 min following baseline ( Figure 2B ) . In the fentanyl group , respiratory rate was strongly reduced 4 min after fentanyl administration . Since the strongest respiratory rate depression was observed 5 min after exposure , we compared respiratory rates ( measured over a one-minute time-period between minutes 5 and 6 ) of control and fentanyl groups ( Figure 2C ) . Fentanyl ( 0 . 2 , 0 . 4 µM ) did not significantly decrease rate of mandible movements compared to control ( p=0 . 473 and p=0 . 797 , Figure 2C ) . Fentanyl ( 1 and 3 µM ) significantly decreased rate compared to control groups ( p=0 . 006 and p=0 . 038 , Figure 2C ) . When data were normalized according to baseline rate , variability within groups was significantly reduced ( Figure 2D ) . Fentanyl significantly decreased normalized rate at 1 µM ( p<0 . 001 ) and 3 µM ( p<0 . 001 ) compared to the control group ( Figure 2E ) , but not at 0 . 2 µM ( p=0 . 275 ) and 0 . 4 µM ( p=0 . 200 ) . In summary , fentanyl-induced respiratory rate depression was observed at 1 and 3 µM of fentanyl tested . We determined that such normalization best represented respiratory rate depression by fentanyl and we only presented normalized data for the subsequent figures with respiratory depression . To determine the pharmacology of opioid receptors in zebrafish , we compared groups of fish exposed to fentanyl with fish exposed to fentanyl and the MOR antagonists naloxone and CTAP ( Figure 3 ) . Exposure to fentanyl reduced the rate of mandible movements ( Figure 3A ) . The MOR antagonist naloxone at 10 µM did not significantly reverse respiratory rate depression induced by fentanyl ( p=0 . 178 , Figure 3B ) . However , naloxone ( 20 µM ) significantly reversed respiratory rate depression by fentanyl ( p<0 . 001 , Figure 3B ) . The selective MOR antagonist CTAP ( 4 µM ) did not significantly reverse respiratory rate depression by fentanyl ( p=0 . 077 , Figure 3B ) , although a trend toward significance was observed . To determine whether naloxone or CTAP may increase respiratory rate when administered alone , we compared groups of fish exposed to CTAP ( 4 µM ) or naloxone ( 5 and 20 µM ) to control fish ( Figure 3C ) . Neither CTAP nor naloxone showed effects on respiratory rate when administered alone ( one-way ANOVA , p=0 . 122 ) . In summary , fentanyl induced a significant respiratory rate depression which was blocked by the MOR antagonist naloxone but not CTAP . In the previous experiments , naloxone was administered concomitantly with fentanyl and prevented respiratory depression ( Figure 3A , B ) . We then tested whether naloxone can completely reverse respiratory depression by fentanyl ( 1 µM ) when administered after respiratory depression ( Figure 3D ) . As expected , fentanyl significantly depressed respiratory rate ( p=0 . 009 ) , an effect not reversed by naloxone at 9 µM ( p=0 . 111 ) . Although a sequential approach may be of interest due to its repeated measure design , it presented two challenges . First , drugs were pipetted twice in the multi-well plate and affected the fish’s behavior , which may explain the large variability observed with naloxone . Second , due to the low volume of fish water in wells and volumes of stock solution administered , it was challenging to increase naloxone concentration in these experiments without diluting the concentration of fentanyl . We therefore concluded that sequential administration of drugs was not the most consistent and compared drug combinations using separate groups of fish . Because zebrafish strains may present genetic variability and different sensitivities to opioid drugs ( Marron Fdez de Velasco et al . , 2009; Sanchez-Simon and Rodriguez , 2008 ) , we compared respiratory depression by fentanyl in various zebrafish lines ( Figure 3E ) . Fentanyl ( 1 µM ) significantly reduced respiratory rate in AB fish as previously determined in previous experiments ( p<0 . 001 ) , but not in Tübingen ( TU ) ( p<0 . 001 ) or AB x TU zebrafish crosses ( p<0 . 001 ) . To conclude , only the AB zebrafish strain presented sensitivity to fentanyl and was used for subsequent experiments . In summary , respiratory rate was depressed by fentanyl through activation of opioid receptors , an effect reversed by naloxone . Morphine is a widely-used opioid drug that induces less severe respiratory depression than the highly potent opioids such as fentanyl ( Gutstein , 2001 ) . We tested whether morphine induces respiratory rate depression in zebrafish . There was no dose-dependent decrease in respiratory rate with increasing morphine dosages ( 1 , 10 , 20 , 50 and 200 µM , p=0 . 088 , Figure 3F ) . However , multiple comparison tests showed that morphine at 1 µM significantly depressed respiratory rate ( p=0 . 021 ) when compared to controls , but not at the other concentrations tested ( p>0 . 488 ) . These results suggest that morphine depressed respiratory rate at low concentration , but that at higher concentrations , morphine may induce an excitatory effect . To determine whether breathing in zebrafish larvae can be stimulated by specific excitatory drugs as observed in mammals , we administered two respiratory stimulants previously shown to stimulate breathing when depressed by opioids . We first administered the 5-HT4A serotoninergic ligand BIMU8 ( Manzke et al . , 2003 ) with or without fentanyl ( Figure 4A ) . BIMU8 ( 10 µM ) combined with fentanyl did not show reversal of respiratory rate when compared to the fentanyl group ( p=0 . 193 ) . Fish exposed to BIMU8 alone showed significant higher respiratory rate compared to fentanyl alone ( p=0 . 008 ) but was not different from the control group ( p=0 . 294 ) . BIMU8/fentanyl showed significantly lower respiratory rates compared to BIMU8 alone and control ( p=0 . 018 and p=0 . 007 ) . In summary , the addition of BIMU8 to fentanyl did not significantly reversed respiratory depression by fentanyl . The AMPA positive allosteric modulator CX614 significantly reversed respiratory rate depression by fentanyl ( p<0 . 001 , Figure 4B ) . CX614/fentanyl showed higher respiratory rate than fentanyl alone , as well as higher respiratory rate than control ( p=0 . 021 ) , suggesting that CX614 at this dosage reversed respiratory depression by fentanyl . CX614 alone did not present higher respiratory rate when compared to control group ( p=0 . 058 ) but showed a trend toward significance . Higher dosages of CX614 induced seizure and were not tested . To determine whether the respiratory depression observed with fentanyl was due to a direct effect of fentanyl on respiratory rate rather than an indirect effect on sedation ( Montandon and Horner , 2019 ) or pain circuits ( Jiang et al . , 2004 ) , we administered lidocaine , an analgesic not acting on MORs ( Lopez-Luna et al . , 2017 ) . Lidocaine did not significantly change respiratory rate compared to control ( p=0 . 267 , Figure 4C ) . To dissolve drugs , we used DMSO as a solvent which may affect fish respiratory rate . DMSO administered at 0 . 0016% did not change respiratory rate ( p=0 . 342 ) . To determine the analgesic properties of opioid drugs in larval zebrafish , we established a simple model of nociception combining exposure to formalin as a nociceptive stimulus and measurements of the subsequent swimming escape response . Using video recordings of larvae positioned in a multi-well plate , we tracked the individual movements of larvae exposed to embryo medium , formalin and a combination of formalin and fentanyl ( Figure 5 ) . We quantified two swimming behaviors ( Figure 5A ) : swimming velocity ( swimming distance in mm per second ) and angular velocity ( turn angle in degrees per second , Figure 5—figure supplement 1 ) . In larvae exposed to formalin , swimming velocity was substantially increased by formalin ( Figure 5B ) . Fentanyl combined with formalin showed considerable reduction in velocity compared to formalin alone ( Figure 5B ) . Following exposure to formalin , swimming velocity significantly increased during the initial 3 min ( Figure 5C , shaded blue ) , a response not observed in control larvae or formalin/fentanyl larvae ( Figure 5C ) . We then compared the averaged swimming velocity during the first 3 min following drug exposure for the different groups of larvae . The formalin group presented significantly faster velocity compared to the control group ( p<0 . 001 , Figure 5D ) . Velocity in the formalin/fentanyl group was significantly slower than in the formalin group ( p=0 . 037 ) , suggesting that fentanyl reduced the escape response to formalin , which may indicate an analgesic effect . However , this effect may be due to the effects of fentanyl directly on swimming . Fentanyl administered alone did not decrease velocity when compared to control ( p=0 . 221 ) , showing that it did not affect swimming by itself . We then determined the effects of the µ-opioid receptor antagonists naloxone and CTAP ( Figure 5D ) . In naloxone groups , velocities were not significantly higher than formalin/fentanyl ( all p>0 . 122 ) , suggesting that naloxone did not block the analgesic effects of fentanyl . In CTAP/formalin/fentanyl group , there was no difference compared to formalin/fentanyl or formalin alone ( p=1 . 000 and p=1 . 000 ) . Naloxone/formalin/fentanyl showed higher velocity than fentanyl or control groups ( Figure 5D ) . These effects may be due to the impact of naloxone alone at 50 µM ( Figure 5E ) . We then looked at the effect of formalin on angular velocity . The formalin group showed reduced angular velocity compared to control ( p<0 . 001 , Figure 5—figure supplement 1 ) , an effect not reversed by fentanyl ( p=0 . 151 ) , which suggest angular velocity cannot be used to assess the escape response to formalin . Although naloxone did not prevent or reverse to analgesic effects of fentanyl , it may be due to the large variability in swimming response induced by formalin . In fact , swimming velocity in formalin/fentanyl/naloxone ranged from 0 . 3 to 4 mm/sec , which is substantially larger than the variability observed in control or fentanyl groups ( Figure 5D ) . To induce more consistent nociceptive response , we administered ( allyl ) -isothiocyanate ( AITC ) , a chemical acting AITC on transient receptor potential ankyrin 1 channels and transient receptor potential cation channel vanilloid receptor 1 ( Oda et al . , 2016 ) . We used a similar experimental approach than we used with formalin by combining AITC , fentanyl and/or naloxone . AITC ( 100 µM ) induced a strong increase in velocity compared to controls , an effect reduced by fentanyl ( Figure 6A ) . AITC produced an initial increase in swimming velocity during the first 3 min following drug exposure , which is consistent with the formalin response . Conversely , in fish with a combination of fentanyl ( 6 µM ) and AITC ( 100 µM ) , no swimming response was observed compared to AITC alone ( Figure 6B ) , showing that fentanyl reduced the nociceptive response to AITC . Compared to the concentration of fentanyl used in the formalin assay ( 3 µM ) , a higher concentration ( 6 µM ) was necessary to eliminate the effects of AITC . This concentration of fentanyl ( 6 µM ) did not decrease velocity when administered alone compared to controls ( Figure 6B ) . Averaged velocities in AITC fish were significantly higher than in controls ( p<0 . 001 , Figure 6C ) , and this increase was significantly lower in fish exposed to fentanyl and AITC ( p<0 . 001 ) . Angular velocity was increased by AITC , but not reversed by fentanyl ( Figure 6—figure supplement 1 ) . Interestingly , groups with naloxone at 20 and 50 µM combined with AITC and formalin presented higher swimming velocities than groups exposed to AITC/fentanyl ( p=0 . 003 and p<0 . 001 , respectively ) , whereas CTAP did not ( p=0 . 342 ) . In summary , AITC produced a fast increase in swimming velocity , an effect significantly reduced by fentanyl , which was reversed by naloxone . To determine whether larval zebrafish are sensitive to other types of analgesics , we exposed larvae to lidocaine , a non-opioid anesthetic ( Figure 7A ) . As previously shown , formalin increased velocity compared to control ( p<0 . 001 , Figure 7B ) . In fish exposed to lidocaine and formalin , swimming velocity was significantly reduced compared to formalin group ( p=0 . 002 ) . Swimming velocity in lidocaine group was not different from control ( p=0 . 693 ) . Since we used a relatively high concentration of DMSO ( 1% ) to dissolve lidocaine , we compared whether DMSO ( 1% ) was different from control fish and we did not observed differences between the two groups ( p=0 . 879 , Figure 7A ) . Lidocaine did not affect angular velocity compared to formalin alone ( Figure 7—figure supplement 1 ) . In conclusion , two different types of analgesics ( fentanyl and lidocaine ) reduced the swimming response to formalin , therefore suggesting that these assays may properly assess analgesia . The two stimulants BIMU8 and CX614 were administered to zebrafish to determine whether they can reverse respiratory depression by fentanyl . Only CX614 was able to significantly reverse respiratory depression . We therefore determine whether BIMU8 and CX614 were also reversing the analgesic properties of fentanyl . The combination of BIMU8 ( 20 µM ) /formalin/fentanyl did not significantly change swimming velocity compared to the control ( p=1 . 000 ) or formalin/fentanyl group ( p=1 . 000 , Figure 7B ) . However , CX614 ( 5 µM ) /formalin/fentanyl group presented higher swimming velocities than the control group ( p<0 . 001 ) and formalin/fentanyl ( p=0 . 007 ) , therefore showing that the ampakine CX614 reversed both respiratory depression and analgesia by fentanyl . Finally , Tübingen fish did not show respiratory depression by fentanyl . We therefore tested whether this strain is similarly insensitive to fentanyl in our analgesia assay . No difference in swimming velocity was observed between formalin and formalin/fentanyl group in TU fish ( p=0 . 132 ) , ( Figure 7C ) , demonstrating that fentanyl did not reduce nociception elicited by formalin .
Opioid drugs are widely used as analgesics but present the severe side-effect of respiratory depression that can be lethal with overdose . No safe opioid therapy currently exists to treat severe and chronic pain . To identify new potent pain treatments without side-effects , opioid drug discovery needs to explore new research avenues . Novel approaches to drug discovery have been recently used including structure-based drug discovery using computer simulations ( Manglik et al . , 2016 ) or cell assays ( Winpenny et al . , 2016 ) . Most drug discovery approaches use target-based approaches which exclude many drug candidates that do not specifically target proteins of interest . Here we propose phenotype-based drug discovery approaches in larval zebrafish to identify potent opioid pain therapies without the side-effects of respiratory depression . The larval zebrafish has emerged as a powerful model system for drug and gene discovery since it combines the biological complexity of in vivo models with a nervous system homolog to humans including the respiratory neural network . Such approach has the potential to validate many drug candidates without making any assumptions on their mechanisms of action or targets . Our drug discovery approach proposes to test whether drugs or combinations of drugs can be used as analgesics without the side-effects of respiratory depression and such strategy is of considerable interest ( Dahan et al . , 2018; Montandon and Slutsky , 2019 ) . We combined several assays to test the respiratory depressant effects of pain killers and their analgesics properties . We demonstrated that larval zebrafish ( days post-fertilization 12–14 ) can be used to test the analgesic properties of opioid drugs and to test the severity of respiratory depression . Our assays provide simple , yet effective , ways to quickly test potential drug candidates in a complex animal model while preserving its central nervous system and brain blood barrier . Due to the high production of fish embryos , we can use these phenotypic assays to perform high-throughput drug screening protocols that are otherwise not feasible in other animal models . Importantly , drug screening in rodents is not feasible as it would require extensive resources and time . Here , we propose new behavioral assays in larval zebrafish where live animals with intact central nervous system can be leveraged for phenotype-based drug discovery combining respiratory depression and analgesia by opioid drugs . Respiratory depression by opioids has been widely studied in mice , rats , dogs , and sheep ( Krause et al . , 2009; Montandon et al . , 2011; Montandon et al . , 2016b ) . The use of complex animal models to assess respiratory variables comes with many challenges . To accurately assess respiratory depression by opioid drugs , it is critical to quantify respiratory variables without the use of anesthetics which interact with the respiratory system , drug metabolism , and mechanisms of action of opioid drugs . Respiratory assessments in non-anesthetized rodents is the gold standard but it raises concerns due to the stress associated with animal handling , opioid injection , and changes due to altered arousal states and behaviours ( Montandon et al . , 2016a; Montandon and Slutsky , 2019 ) , which may ultimately affect research outcomes . In larval zebrafish , we demonstrated , for the first time , that respiratory network activity can be assessed , that larval zebrafish present respiratory rate depression by opioid drugs in a similar fashion than humans , and that opioid pharmacology observed in humans is preserved in zebrafish . Most fishes have a complex respiratory system to move water through their gills . Although fishes use a different strategy to absorb oxygen and eliminate carbon dioxide than mammals , they rhythmically produce mandibular movements to move water through their gills . In lampreys , the paratrigeminal respiratory group ( pTRG ) generates rhythmic mandible movements ( Bongianni et al . , 2016 ) . Because of their close evolutionary origins ( Missaghi et al . , 2016 ) , the pTRG shares similarities with the mammalian respiratory network ( Cinelli et al . , 2013 ) which generates breathing and regulates respiratory depression by opioids ( Montandon et al . , 2011 ) . The pTRG also presents similar functional properties than mammals ( Gray et al . , 1999 ) such as sensitivity to substance P ( Mutolo et al . , 2010 ) and to opioid drugs ( Mutolo et al . , 2007 ) . Here , we propose that respiratory mandible movements controlled , at least in part , by the pTRG can be used as an index of respiratory network activity , similarly to respiratory activity of the trigeminal muscle in mammals ( Jacquin et al . , 1999 ) . We showed that the rate of respiratory movements presented dose-dependent decreases in response to increasing dosages of the MOR ligand fentanyl . Although it is difficult to compare administration of fentanyl through fish water in zebrafish and systemic injection in mammals , the dosages used in zebrafish corresponds to those used in rodents ( Yassen et al . , 2008 ) . Respiratory depression by fentanyl was substantially blocked by the MOR antagonist naloxone , which is consistent with the fact that fentanyl is acting mostly through MORs ( James and Williams , 2020 ) . However , the highly selective MOR antagonist CTAP did not prevent respiratory depression by fentanyl , which is similar to the partial antagonism observed in some rodent studies ( Zhang et al . , 2007 ) . These results suggest that either other opioid receptors are involved in the effect of fentanyl or that CTAP did not cross the blood brain barriers in zebrafish due to its large molecular weight . Interestingly , morphine reduced significantly respiratory rate in larval zebrafish at the low concentration of 1 µM , which is consistent with clinical studies ( Montandon et al . , 2016a ) , but not at higher concentrations . The lack of respiratory depression at high morphine concentrations may be due to disinhibition of excitatory networks regulating breathing in the zebrafish . In rodents , morphine disinhibits the ventral tegmental area ( Chen et al . , 2015 ) , a brain area involved in sleep-wake states and arousal ( Venner et al . , 2019 ) . Since increased arousal state reduces the severity of respiratory depression by opioids ( Montandon et al . , 2016a; Montandon and Horner , 2019 ) , it may lead to a reduced respiratory depression at high concentrations of morphine . The two major strains of wild-type zebrafish , AB and Tübinger ( TU ) , showed different sensitivities to fentanyl . AB fish present pronounced respiratory depression , an effect not observed in TU fish , regardless of the concentration used ( data not shown ) . This strain-specificity suggests that the mechanisms of action of opioid drugs differ between zebrafish strains which could be due to polymorphisms of the MOR gene as it can be found in humans ( Oertel et al . , 2006 ) or genetic differences in various genes involved in MOR inhibition ( Bian et al . , 2012 ) . To determine whether respiratory stimulants can reverse respiratory depression by opioids in larval zebrafish , we tested two types of agonists targeting excitatory receptors: ampakines which comprise drugs acting on AMPA receptors ( Ren et al . , 2006 ) and serotoninergic agents ( Manzke et al . , 2003 ) . CX-614 , an ampakine allosteric modulator of AMPA receptors , reversed respiratory depression in our larval zebrafish models , which is consistent with its effects in rodents ( Ren et al . , 2006 ) or humans ( Dahan et al . , 2018 ) . BIMU8 , a 5-HT4A agonist , was administered in combination with fentanyl and compared with fentanyl alone . BIMU8 did not significantly reverse respiratory depression at the concentrations tested , which is consistent with the low efficacy of this treatment in humans ( Lötsch et al . , 2005 ) . Other serotonin agents such as 5-HT1 and 5-HT3 agonists ( van der Schier et al . , 2014 ) can be easily tested using our zebrafish models . Zebrafish larvae were previously used as animal models to study pain and analgesia . Acetic or citric acids were added to fish water to induce pain ( Lopez-Luna et al . , 2017; Steenbergen and Bardine , 2014 ) . Our models did not use acids as nociceptive stimuli as they change water pH , to which zebrafish are sensitive ( Avdesh et al . , 2012; Steenbergen and Bardine , 2014 ) . Instead , we used formalin‚ a nociceptive stimulus widely used to induce pain in rodent models ( Yoon et al . , 2015 ) , and showed that formalin increased swimming velocity . Here , we propose that increased swimming velocity in response to formalin represents the fish escape response to nociceptive stimuli . In fact , lidocaine , a non-opioid analgesic , applied with formalin showed a significant reduction compared to formalin alone . Lidocaine induces analgesia by inactivating voltage-gated sodium channels in neurons , without activating opioid receptors ( Tetzlaff , 2000 ) , and these results support the concept that formalin induces pain in larval zebrafish . To promote analgesia in larval zebrafish , fentanyl was combined with formalin and it reduced the swimming response to formalin , therefore suggesting that it reduced pain . It could be suggested that the escape response observed in our assays is due to the effect of formalin on locomotor activity ( not nociception ) and that the reduced response with fentanyl may be due to inhibition of motor activity . However , fentanyl alone did not reduce swimming velocity compared to control larvae . To complement the nociceptive assay with formalin , we used AITC , also known as ‘mustard oil’ , which acts on cation channel transient receptors potential ankyrin 1one and transient receptor potential vanilloid 1 ( Oda et al . , 2016 ) , two receptors involved in nociception to chemical compounds ( Bamps et al . , 2021 ) . Like formalin , the swimming response to AITC was reduced by fentanyl . To determine the role of µ-opioid receptors in the analgesic responses to fentanyl , we administered the antagonist naloxone . In the formalin assay , naloxone did not block the effects of formalin , but blocked it in the AITC assay , which is consistent with the effects of naloxone in humans ( Gutstein , 2001 ) . Although naloxone did not significantly block analgesia when formalin was used , it is clear in Figure 5d that naloxone increased substantially velocity in some animals , but also reduced it in other animals . In summary , the fact that fentanyl reduced the nociceptive responses of AITC and formalin , two chemicals promoting nociception through different mechanisms , suggests that our approaches correctly assess opioid analgesia in larval zebrafish , while opioid pharmacology remains unclear in the formalin/fentanyl assays . As observed with the respiratory the CX-614 reversed analgesia which is consistent with their effects in rodents ( Dahan et al . , 2018 ) . BIMU-8 did not reverse analgesia which is consistent with its lack of effects in clinical trials ( Dahan et al . , 2018 ) . In conclusion , we established opioid analgesia assays that can be used to quantify pain and the analgesic properties of new opioid therapies that can easily be used for large scale drug discovery using larval zebrafish . Opioid analgesics constitute essential pain therapies that present the lethal side-effect of respiratory depression therefore limiting their effective use in the clinical and at-home settings . There is currently no effective safe pain therapy due to the difficulty at identifying new drug combinations with potent analgesia but reduced respiratory side-effects ( Dahan et al . , 2018 ) . Using larval zebrafish , we propose models allowing phenotype-based drug discovery approaches , i . e . permitting drug testing without assumptions related to mechanisms of action and targets . Our novel drug discovery models allow high-throughput drug screening in a simple and amenable animal model presenting similar pharmacological and genetic profiles than humans ( MacRae and Peterson , 2015 ) . Although zebrafish has been used to identify new anesthetics and their mechanisms of actions ( McGrath et al . , 2020; Yang et al . , 2019 ) , our study is the first to assess respiratory depression by opioid drugs in larval zebrafish combined with analgesia . In addition to high-throughput screening approaches , our models can also be used in combination of transgenic or knockout zebrafish to better understand the mechanisms opioid inhibition , analgesia and respiratory depression , as well as live microscopy of neural circuits of pain and respiration ( Ahrens et al . , 2013 ) .
Animal practices and experiments in adult fish , larvae , and breeding pairs followed laboratory standards ( Avdesh et al . , 2012 ) , and were carried out according to the procedures outlined by the Canadian Council on Animal Care and were approved by St . Michael’s Hospital animal care committee . Only wildtype AB strain zebrafish larvae at 12–14 days post-fertilization ( dpf ) were used for experiments , except when comparisons between strains were made . Tübingen ( TU ) and crosses between TU and AB were used to compare strains . All experiments were performed in AB fish except when TU was specifically mentioned . All fish were housed on a 14/10 hr light/dark cycle and kept at a constant water temperature of 28°C ± 0 . 5° . Larvae and adults were originally obtained from the Hospital for Sick Children and the University of Toronto Mississauga ( Toronto , ON ) . For breeding , male and female adult fish at 4 months of age were placed in a breeding tank and separated by a divider . The next day the divider was removed , and fish mated within the first half hour of the lights-on period and were returned to the rack ( Aquaneering , CA , United States ) . The eggs were collected and placed in petri-dish filled with E2 embryo medium ( NaCl , 15 . 0 mM; KCl , 0 . 5 mM; MgSO4 , 1 . 0 mM; CaCl2 , 1 . 0 mM; Na2HPO4 , 0 . 05 mM; KH2PO4 , 0 . 15 mM; NaHCO3 , 0 . 7 mM ) . Unfertilized eggs were removed . Beginning at 5 days post-fertilization , larvae were fed with Ziegler AP100 ( artificial plankton ) dry larval diet ( 100 microns ) and were moved to 0 . 8 litre tanks filled with system water , with a density of 20 fish/100 mL and were kept there until the experiment . Water quality was kept at a pH of 7 . 5 ± 0 . 5 and with a conductivity of 500-1000ppm . Dissolved oxygen was maintained at 6-7ppm . Nitrites and ammonia were kept at <150 ppm . All experiments were undertaken during daylight hours and fish were placed in an incubator at 28 . 5 ± 0 . 5°C . All opioid drugs were used with Health Canada approval . Fentanyl citrate and morphine sulfate were obtained from Sandoz ( QC , Canada ) . Naloxone hydrochloride was obtained from Omega ( QC , Canada ) . The MOR antagonist CTAP ( D-Phe-Cys-Tyr-D-Trp-Arg-Thr-Pen-Thr-NH2 ) , lidocaine , the Ampakine CX614 , and the 5-HT4A receptor agonist BIMU8 were obtained from Tocris ( ON , Canada ) . AITC was obtained from Sigma-Aldrich ( ON , Canada ) . Respiratory mandible movements were quantified in live zebrafish larvae using a custom-made system including a 4K high-definition camera ( Basler 4K , 4096pi x 3000pi , model acA4112-30um , Edmunds Optics ) and partially telecentric 7x zoom lens ( Edmund Optics ) . The water was kept at a constant 28 . 5 ± 0 . 5°C temperature using a heating pad placed under the multi-well plate . With this system , simultaneous recordings of 12 zebrafish larvae can be performed . As an index of respiratory network activity , mandible movements were measured , and respiratory rate was quantified ( Figure 1A ) . In a custom-made 12-well clear plate , larval zebrafish were placed in wells ( diameter 10 mm , depth 4 mm ) containing 54 µL of embryo medium . Using our video recording system , a dorsal video was performed for each drug combination . We used two approaches to determine the rate of mandible movement . Rate of mandible movements was visually counted by two independent researchers , blind to the drug combinations . Using this approach , the number of mandible movements per minute was calculated for each condition and each animal . In a subset of zebrafish larvae , mandible movements were also quantified by measuring pixel intensity changes in a defined area around the fish head . Tracings of pixel intensity over time were plotted and the rate of intensity changes was quantified using a custom-made software in Matlab ( Mathworks , US ) . The validity of visual quantification was then compared to software quantification . Zebrafish larvae were positioned in wells and were left for 10 min to acclimatize to the environment . A video of baseline respiratory movement was then taken for 1 min . Following the baseline recordings , combinations of drugs were applied to each well in a volume of 6 µL per well . VA video was recorded for 30 min following drug administration . To determine whether various drugs induce respiratory rate depression , we compared the rates of mandible movements between several groups of larvae: control group ( embryo medium ) , fentanyl group , fentanyl/naloxone group ( combination of fentanyl and naloxone ) , fentanyl/other drugs ( fentanyl and potential drug candidates ) . When possible , animals from different groups were tested simultaneously in 12-well plates ( for example: four larvae per group with three groups were tested ) . However , several separate recordings of different 12-well plates were necessary to provide the correct number of animals per group . Respiratory rate values collected in larval zebrafish were normalized according to baseline mandible rates acquired before drugs were administered . Normalization considerably reduced the high variability encountered in larval zebrafish . Larval zebrafish may present different growth and behaviors at 12-14dpf due to variable development rates and access to food and nutrients within the tank . To consistently select larvae at similar development stages , we established selection criteria . For respiratory assays , larvae with rates of mandible movements below 60 and above 160 movements/min at baseline ( before drugs were administered ) were not considered . Such exclusion criteria considerably reduce the inter-group variability observed in zebrafish larvae . Nociception was assessed by measuring the swimming escape response to nociceptive stimuli . Zebrafish were placed individually into wells of a custom-made 24-well plate containing 90 µL of embryo water . The well plate was made of white acrylic with a back-light to allow video recording . The plate was heated at 28 . 5 ± 0 . 5°C . The same custom-made video recording apparatus , as mentioned above , was used to record swimming in larval zebrafish . Swimming experiments were analyzed using a commercial software ( EthoVision XT v15 , Noldus Information Technology , Netherlands ) , and swimming velocity ( mm/sec ) and angular velocity ( degrees/sec ) were quantified . Before applying drugs to the plate wells containing larvae , we measured a 10 min baseline . Only larvae with velocity between 0 . 1 and 2 mm/sec during baseline ( before drugs were administered ) were considered for analysis . Formalin or ( allyl ) -isothiocyanate ( AITC ) were then used as nociceptive stimuli . Such stimuli were previously validated in mammalian and adult zebrafish ( Magalhães et al . , 2017; Oda et al . , 2016 ) studies . AITC directly acts on transient receptor potential A1 which is involved in pain response . We applied formalin or AITC in combination with opioid analgesics to determine opioid analgesia in larval zebrafish . Zebrafish larvae ( 12–14 days post-fertilization ) were positioned in wells of a 24-well plate and left 10 min to acclimatize . A video of the baseline movements was taken for 10 min . Following baseline recordings , combinations of drugs were applied to each well in a volume of 10 µL and video was recorded for 15 min . To determine whether various drug combinations induce analgesia , we compared the swimming velocity between several groups of larvae: control group ( embryo medium was administered ) , formalin or AITC group , fentanyl/formalin or fentanyl/AITC group ( combination of fentanyl and formalin ) , formalin/fentanyl/naloxone or AITC/fentanyl/naloxone . CTAP was also used instead of naloxone . When possible , these groups were tested simultaneously in 24-well plates ( six larvae per group if four groups were tested ) . Multiple recordings were needed to provide the correct number of animals per group . The swimming and angular velocities were calculated for the first 3 min following treatment ( minutes 0–3 ) and were used for analysis . Data in larval zebrafish do not always follow a normal distribution . Therefore , normality was tested using a Shapiro-Wilk method and equality of variances using the Brown-Forsythe test . All respiratory data presented in the Results section followed normal distribution when respiratory rate was normalized according to baseline respiratory rate ( before any drugs were administered ) . When data followed a normal distribution , we compared multiple groups using a one-way ANOVA followed by Holm-Sidak post-hoc tests to compare individual groups . We presented individual data points , group means and standard deviation as error bars . However , data in larval zebrafish are often not normally distributed . For instance , swimming velocity and angular velocity did not follow normal distribution , even when normalized according to baseline . When data distribution was not normal , we compared experimental groups using Wilcoxon rank tests and All Pairwise Multiple Comparison Procedures ( Dunn’s method ) as post-hoc tests . We presented data as individual data points and medians with error bars showing 25th and 75th percentiles ( interquartile range ) . In some cases , data presented clear outliers . To objectively identify outliers , we selected all data points below and above 1 . 5 x the interquartile range and eliminated these data points from the data set ( Michel et al . , 2020 ) . All statistical tests and graphs were done using Sigmaplot ( version 14 , SAS ) and figures were prepared with Adobe Illustrator ( Creative Suite 5 , Adobe ) . | When it comes to treating severe pain , a doctor’s arsenal is somewhat limited: synthetic or natural opioids such as morphine , fentanyl or oxycodone are often one of the only options available to relieve patients . Yet these compounds can make breathing slower and shallower , quickly depriving the body of oxygen and causing death . This lethal side-effect is particularly devastating as opioids misuse has reached dangerously high levels in the United States , creating an ‘opioid epidemic’ which has claimed the lives of over 80 , 000 Americans in 2020 . It is therefore crucial to find safer drugs that do not have this effect on breathing , but this research has been slowed down by the lack of animal models in which to study the effect of new compounds . Zebrafish are small freshwater fish that reproduce and develop fast , yet they are also remarkably genetically similar to mammals and feature a complex nervous system . However , it is not known whether the effect of opioids on zebrafish is comparable to mammals , and therefore whether these animals can be used to test new drugs for pain relief . To investigate this question , Zaig et al . exposed zebrafish larvae to fentanyl , showing that the fish then exhibited slower lower jaw movements – a sign of decreased breathing . The fish also could also tolerate a painful stimulus for longer , suggesting that this opioid does reduce pain in the animals . Together , these results point towards zebrafish and mammals sharing similar opioid responses , demonstrating that the fish could be used to test potential pain medications . The methods Zaig et al . have developed to establish these results could be harnessed to quickly assess large numbers of drug compounds , as well as decipher how pain emerges and can be stopped . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"medicine",
"neuroscience"
] | 2021 | Respiratory depression and analgesia by opioid drugs in freely behaving larval zebrafish |
Distinctions between cell types underpin organizational principles for nervous system function . Functional variation also exists between neurons of the same type . This is exemplified by correspondence between grid cell spatial scales and the synaptic integrative properties of stellate cells ( SCs ) in the medial entorhinal cortex . However , we know little about how functional variability is structured either within or between individuals . Using ex-vivo patch-clamp recordings from up to 55 SCs per mouse , we found that integrative properties vary between mice and , in contrast to the modularity of grid cell spatial scales , have a continuous dorsoventral organization . Our results constrain mechanisms for modular grid firing and provide evidence for inter-animal phenotypic variability among neurons of the same type . We suggest that neuron type properties are tuned to circuit-level set points that vary within and between animals .
The concept of cell types provides a general organizing principle for understanding biological structures including the brain ( Regev et al . , 2017; Zeng and Sanes , 2017 ) . The simplest conceptualization of a neuronal cell type , as a population of phenotypically similar neurons with features that cluster around a single set point ( Wang et al . , 2011b ) , is extended by observations of variability in cell type features , suggesting that some neuronal cell types may be conceived as clustering along a line rather than around a point in a feature space ( Cembrowski and Menon , 2018; O'Donnell and Nolan , 2011; Figure 1A ) . Correlations between the functional organization of sensory , motor and cognitive circuits and the electrophysiological properties of individual neuronal cell types suggest that this feature variability underlies key neural computations ( Adamson et al . , 2002; Angelo et al . , 2012; Fletcher and Williams , 2019; Garden et al . , 2008; Giocomo et al . , 2007; Kuba et al . , 2005; O'Donnell and Nolan , 2011 ) . However , within-cell type variability has typically been deduced by combining data obtained from multiple animals . By contrast , the structure of variation within individual animals or between different animals has received little attention . For example , apparent clustering of properties along lines in feature space could reflect a continuum of set points , or could result from a small number of discrete set points that are obscured by inter-animal variation ( Figure 1B ) . Moreover , although investigations of invertebrate nervous systems show that set points may differ between animals ( Goaillard et al . , 2009 ) , it is not clear whether mammalian neurons exhibit similar phenotypic diversity ( Figure 1B ) . Distinguishing these possibilities requires many more electrophysiological observations for each animal than are obtained in typical studies . Stellate cells in layer 2 ( SCs ) of the medial entorhinal cortex ( MEC ) provide a striking example of correspondence between functional organization of neural circuits and variability of electrophysiological features within a single cell type . The MEC contains neurons that encode an animal’s location through grid-like firing fields ( Fyhn et al . , 2004 ) . The spatial scale of grid fields follows a dorsoventral organization ( Hafting et al . , 2005 ) , which is mirrored by a dorsoventral organization in key electrophysiological features of SCs ( Boehlen et al . , 2010; Dodson et al . , 2011; Garden et al . , 2008; Giocomo et al . , 2007; Giocomo and Hasselmo , 2008a; Pastoll et al . , 2012a ) . Grid cells are further organized into discrete modules ( Stensola et al . , 2012 ) , with the cells within a module having a similar grid scale and orientation ( Barry et al . , 2007; Gu et al . , 2018; Stensola et al . , 2012; Yoon et al . , 2013 ) ; progressively more ventral modules are composed of cells with wider grid spacing ( Stensola et al . , 2012 ) . Studies that demonstrate dorsoventral organization of integrative properties of SCs have so far relied on the pooling of relatively few measurements per animal . Hence , it is unclear whether the organization of these cellular properties is modular , as one might expect if they directly set the scale of grid firing fields in individual grid cells ( Giocomo et al . , 2007 ) . The possibility that set points for electrophysiological properties of SCs differ between animals has also not been considered previously . Evaluation of variability between and within animals requires statistical approaches that are not typically used in single-cell electrophysiological investigations . Given appropriate assumptions , inter-animal differences can be assessed using mixed effect models that are well established in other fields ( Baayen et al . , 2008; Geiler-Samerotte et al . , 2013 ) . Because tests of whether data arise from modular as opposed to continuous distributions have received less general attention , to facilitate detection of modularity using relatively few observations , we introduce a modification of the gap statistic algorithm ( Tibshirani et al . , 2001 ) that estimates the number of modes in a dataset while controlling for observations expected by chance ( see 'Materials and methods' and Figure 1—figure supplements 1–5 ) . This algorithm performs well compared with discreteness metrics that are based on the standard deviation of binned data ( Giocomo et al . , 2014; Stensola et al . , 2012 ) , which we find are prone to high false-positive rates ( Figure 1—figure supplement 4A ) . We find that recordings from approximately 30 SCs per animal should be sufficient to detect modularity using the modified gap statistic algorithm and given the experimentally observed separation between grid modules ( see 'Materials and methods' and Figure 1—figure supplements 2–3 ) . Although methods for high-quality recording from SCs in ex-vivo brain slices are well established ( Pastoll et al . , 2012b ) , typically fewer than five recordings per animal were made in previous studies , which is many fewer than our estimate of the minimum number of observations required to test for modularity . We set out to establish the nature of the set points that establish the integrative properties of SCs by measuring intra- and inter-animal variation in key electrophysiological features using experiments that maximize the number of SCs recorded per animal . Our results suggest that set points for individual features of a neuronal cell type are established at the level of neuronal cell populations , differ between animals and follow a continuous organization .
Before addressing intra- and inter-animal variability , we first describe the data set used for the analyses that follow . We established procedures to facilitate the recording of integrative properties of many SCs from a single animal ( see 'Materials and methods' ) . With these procedures , we measured and analyzed electrophysiological features of 836 SCs ( n/mouse: range 11–55; median = 35 ) from 27 mice ( median age = 37 days , age range = 18–57 days ) . The mice were housed either in a standard home cage ( dimensions: 0 . 2 × 0 . 37 m , N = 18 mice , n = 583 neurons ) or from postnatal day 16 in a 2 . 4 × 1 . 2 m cage , which provided a large environment that could be freely explored ( N = 9 , n = 253 , median age = 38 days ) ( Figure 2—figure supplement 1 ) . For each neuron , we measured six sub-threshold integrative properties ( Figure 2A–B ) and six supra-threshold integrative properties ( Figure 2C ) . Unless indicated otherwise , we report the analysis of datasets that combine the groups of mice housed in standard and large home cages and that span the full range of ages . Because SCs are found intermingled with pyramidal cells in layer 2 ( L2PCs ) , and as misclassification of L2PCs as SCs would probably confound investigation of intra-SC variation , we validated our criteria for distinguishing each cell type . To establish characteristic electrophysiological properties of L2PCs , we recorded from neurons in layer 2 that were identified by Cre-dependent marker expression in a Wfs1Cre mouse line ( Sürmeli et al . , 2015 ) . Expression of Cre in this line , and in a similar line ( Kitamura et al . , 2014 ) , labels L2PCs that project to the CA1 region of the hippocampus , but does not label SCs ( Kitamura et al . , 2014; Sürmeli et al . , 2015 ) . We identified two populations of neurons in layer 2 of MEC that were labelled in Wfs1Cre mice ( Figure 3A–C ) . The more numerous population had properties consistent with L2PCs ( Figure 3A , G ) and could be separated from the unidentified population on the basis of a lower rheobase ( Figure 3C ) . The unidentified population had firing properties that were typical of layer 2 interneurons ( Gonzalez-Sulser et al . , 2014 ) . A principal component analysis ( PCA ) ( Figure 3D–F ) clearly separated the L2PC population from the SC population , but did not identify subpopulations of SCs . The properties of the less numerous population were also clearly distinct from those of SCs ( Figure 3A , C ) . These data demonstrate that the SC population used for our analyses is distinct from other cell types also found in layer 2 of the MEC . To further validate the large SC dataset , we assessed the location-dependence of individual electrophysiological features , several of which have previously been found to depend on the dorso-ventral location of the recorded neuron ( Boehlen et al . , 2010; Booth et al . , 2016; Garden et al . , 2008; Giocomo et al . , 2007; Pastoll et al . , 2012a; Yoshida et al . , 2013 ) . We initially fit the dependence of each feature on dorsoventral position using a standard linear regression model . We found substantial ( adjusted R2 >0 . 1 ) dorsoventral gradients in input resistance , sag , membrane time constant , resonant frequency , rheobase and the current-frequency ( I-F ) relationship ( Figure 3G ) . In contrast to the situation in SCs , we did not find evidence for dorsoventral organization of these features in L2PCs ( Figure 3G ) . Thus , our large dataset replicates the previously observed dependence of integrative properties of SCs on their dorsoventral position , and shows that this location dependence further distinguishes SCs from L2PCs . To what extent does variability between the integrative properties of SCs at a given dorsoventral location arise from differences between animals ? Comparing specific features between individual animals suggested that their distributions could be almost completely non-overlapping , despite consistent and strong dorsoventral tuning ( Figure 4A ) . If this apparent inter-animal variability results from the random sampling of a distribution determined by a common underlying set point , then fitting the complete data set with a mixed model in which animal identity is included as a random effect should reconcile the apparent differences between animals ( Figure 4B ) . In this scenario , the conditional R2 estimated from the mixed model , in other words , the estimate of variance explained by animal identity and location , should be similar to the marginal R2 value , which indicates the variance explained by location only . By contrast , if differences between animals contribute to experimental variability , the mixed model should predict different fitting parameters for each animal , and the estimated conditional R2 should be greater than the corresponding marginal R2 ( Figure 4C ) . Fitting the experimental measures for each feature with mixed models suggests that differences between animals contribute substantially to the variability in properties of SCs . In contrast to simulated data in which inter-animal differences are absent ( Figure 4B ) , differences in fits between animals remained after fitting with the mixed model ( Figure 4D ) . This corresponds with expectations from fits to simulated data containing inter-animal variability ( Figure 4C ) . To visualize inter-animal variability for all measured features , we plot for each animal the intercept of the model fit ( I ) , the predicted value at a location 1 mm ventral from the intercept ( I+S ) , and the slope ( lines ) ( Figure 4E ) . Strikingly , even for features such as rheobase and input resistance ( IR ) that are highly tuned to a neurons’ dorsoventral position , the extent of variability between animals is similar to the extent to which the property changes between dorsal and mid-levels of the MEC . If set points that determine integrative properties of SCs do indeed differ between animals , then mixed models should provide a better account of the data than linear models that are generated by pooling data across all animals . Consistent with this , we found that mixed models for all electrophysiological features gave a substantially better fit to the data than linear models that considered all neurons as independent ( adjusted p<2×10−17 for all models , χ2 test , Table 1 ) . Furthermore , even for properties with substantial ( R2 value >0 . 1 ) dorsoventral tuning , the conditional R2 value for the mixed effect model was substantially larger than the marginal R2 value ( Figure 4D and Table 1 ) . Together , these analyses demonstrate inter-animal variability in key electrophysiological features of SCs , suggesting that the set points that establish the underlying integrative properties differ between animals . Because neuronal integrative properties may be modified by changes in neural activity ( Zhang and Linden , 2003 ) , we asked whether experience influences the measured electrophysiological features of SCs . We reasoned that modifying the space through which animals can navigate may drive experience-dependent plasticity in the MEC . As standard mouse housing has dimensions less than the distance between the firing fields of more ventrally located grid cells ( Brun et al . , 2008; Hafting et al . , 2005 ) , in a standard home cage , only a relatively small fraction of ventral grid cells is likely to be activated , whereas larger housing should lead to the activation of a greater proportion of ventral grid cells . We therefore tested whether the electrophysiological features of SCs differ between mice housed in larger environments ( 28 , 800 cm2 ) and those with standard home cages ( 740 cm2 ) . We compared the mixed models described above to models in which housing was also included as a fixed effect . To minimize the effects of age on SCs ( Boehlen et al . , 2010; Burton et al . , 2008; Supplementary file 2 ) , we focused these and subsequent analyses on mice between P33 and P44 ( N = 25 , n = 779 ) . We found that larger housing was associated with a smaller sag coefficient , indicating an increased sag response , a lower resonant frequency and a larger spike half-width ( adjusted p<0 . 05; Figure 4E , Supplementary file 3 ) . These differences were primarily from changes to the magnitude rather than the location-dependence of each feature . Other electrophysiological features appeared to be unaffected by housing . To determine whether inter-animal differences remain after accounting for housing , we compared mixed models that include dorsoventral location and housing as fixed effects with equivalent linear regression models in which individual animals were not accounted for . Mixed models incorporating animal identity continued to provide a better account of the data , both for features that were dependent on housing ( adjusted p<2 . 8×10−21 ) and for features that were not ( adjusted p<1 . 4×10−7 ) ( Supplementary file 4 ) . Together , these data suggest that specific electrophysiological features of SCs may be modified by experience of large environments . After accounting for housing , significant inter-animal variation remains , suggesting that additional mechanisms acting at the level of animals rather than individual neurons also determine differences between SCs . To address the possibility that other experimental or biological variables could contribute to inter-animal differences , we evaluated the effects of home cage size ( Supplementary files 3–4 ) , brain hemisphere ( Supplementary file 5 ) , mediolateral position ( Figure 4—figure supplement 1 and Supplementary file 6 ) , the identity of the experimenter ( Supplementary file 7 ) and time since slice preparation ( Supplementary files 8 and 9 ) . Several of the variables influenced some measured electrophysiological features , for example properties primarily related to the action potential waveform depended on the mediolateral position of the recorded neuron ( Supplementary file 6; Canto and Witter , 2012; Yoshida et al . , 2013 ) , but significant inter-animal differences remained after accounting for each variable . We carried out further analyses using models that included housing , mediolateral position , experimenter identity and the direction in which sequential recordings were obtained as fixed effects ( Supplementary file 10 ) , and using models fit to minimal datasets in which housing , mediolateral position and the recording direction were identical ( Supplementary file 11 ) . These analyses again found evidence for significant inter-animal differences . Inter-animal differences could arise if the health of the recorded neurons differed between brain slices . To minimize this possibility , we standardized our procedures for tissue preparation ( see 'Materials and methods' ) , such that slices were of consistent high quality as assessed by low numbers of unhealthy cells and by visualization of soma and dendrites of neurons in the slice . Several further observations are consistent with comparable quality of slices between experiments . First , if the condition of the slices had differed substantially between animals , then in better quality slices , it should be easier to record from more neurons , in which case features that depend on tissue quality would correlate with the number of recorded neurons . However , the majority ( 10/12 ) of the electrophysiological features were not significantly ( p>0 . 2 ) associated with the number of recorded neurons ( Supplementary file 12 ) . Second , analyses of inter-animal differences that focus only on data from animals for which >35 recordings were made , which should only be feasible with uniformly high-quality brain slices , are consistent with conclusions from analysis of the larger dataset ( Supplementary file 13 ) . Third , the conditional R2 values of electrophysiological features of L2PCs are much lower than those for SCs recorded under the same experimental conditions ( Table 1 and Supplementary file 1 ) , suggesting that inter-animal variation may be specific to SCs and cannot be explained by slice conditions . Together , these analyses indicate that differences between animals remain after accounting for experimental and technical factors that might contribute to variation in the measured features of SCs . The dorsoventral organization of SC integrative properties is well established , but whether this results from within animal variation consistent with a small number of discrete set points that underlie a modular organization ( Figure 1B ) is unclear . To evaluate modularity , we used datasets with n ≥ 34 SCs ( N = 15 mice , median age = 37 days , age range = 18–43 days ) . We focus initially on rheobase , which is the property with the strongest correlation to dorsoventral location , and resonant frequency , which is related to the oscillatory dynamics underlying dorsoventral tuning in some models of grid firing ( e . g . Burgess et al . , 2007; Giocomo et al . , 2007 ) . For n ≥ 34 SCs , we expect that if properties are modular , then this would be detected by the modified gap statistic in at least 50% of animals ( Figure 1—figure supplements 2C and 3 ) . By contrast , we find that for datasets from the majority of animals , the modified gap statistic identifies only a single mode in the distribution of rheobase values ( Figure 5A and Figure 6 ) ( N = 13/15 ) and of resonant frequencies ( Figure 5B and Figure 6 ) ( N = 14/15 ) , indicating that these properties have a continuous rather than a modular distribution . Consistent with this , smoothed distributions did not show clearly separated peaks for either property ( Figure 5 ) . The mean and 95% confidence interval for the probability of evaluating a dataset as clustered ( pdetect ) was 0 . 133 and 0 . 02–0 . 4 for rheobase and 0 . 067 and 0 . 002–0 . 32 for resonant frequency . These values of pdetect were not significantly different from the proportions expected given the false positive rate of 0 . 1 in the complete absence of clustering ( p=0 . 28 and 0 . 66 , binomial test ) . Thus , the rheobase and resonant frequency of SCs , although depending strongly on a neuron’s dorsoventral position , do not have a detectable modular organization . When we investigated the other measured integrative properties , we also failed to find evidence for modularity . Across all properties , for any given property , at most 3 out of 15 mice were evaluated as having a clustered organization using the modified gap statistic ( Figure 6 ) . This does not differ significantly from the proportion expected by chance when no modularity is present ( p>0 . 05 , binomial test ) . Consistent with this , the average proportion of datasets evaluated as modular across all measured features was 0 . 072 ± 0 . 02 ( ± SEM ) , which is similar to the expected false-positive rate . By contrast , the properties of grid firing fields previously recorded with tetrodes in behaving animals ( Stensola et al . , 2012 ) were detected as having a modular organization using the modified gap statistic ( Figure 1—figure supplement 5 ) . For seven grid-cell datasets with n ≥ 20 , the mean for pdetect is 0 . 86 , with 95% confidence intervals of 0 . 42 to 0 . 996 . We note here that discontinuity algorithms that were previously used to assess the modularity of grid field properties ( Giocomo et al . , 2014; Stensola et al . , 2012 ) did indicate significant modularity in the majority of the intrinsic properties measured in our dataset ( N = 13/15 and N = 12/15 , respectively ) , but this was attributable to false positives resulting from the relatively even sampling of recording locations ( see Figure 1—figure supplement 4A ) . Therefore , we conclude that it is unlikely that any of the intrinsic integrative properties of SCs that we examined have organization within individual animals resembling the modular organization of grid cells in behaving animals . Finally , because many of the measured electrophysiological features of SCs emerge from shared ionic mechanisms ( Dodson et al . , 2011; Garden et al . , 2008; Pastoll et al . , 2012a ) , we asked whether dorsoventral tuning reflects a single core mechanism and whether inter-animal differences are specific to this mechanism or manifest more generally . Estimation of conditional independence for measurements at the level of individual neurons ( Figure 7A ) or individual animals ( Figure 7B ) was consistent with the expectation that particular classes of membrane ion channels influence multiple electrophysiologically measured features . The first five dimensions of a principal components analysis ( PCA ) of all measured electrophysiological features accounted for almost 80% of the variance ( Figure 7C ) . Examination of the rotations used to generate the principal components suggested relationships between individual features that are consistent with our evaluation of the conditional independence structure of the measured features ( Figure 7D and A ) . When we fit the principal components using mixed models with location as a fixed effect and animal identity as a random effect , we found that the first two components depended significantly on dorsoventral location ( Figure 7E and Supplementary file 14 ) ( marginal R2 = 0 . 50 and 0 . 09 and adjusted p=1 . 09×10−15 and 1 . 05 × 10−4 , respectively ) . Thus , the dependence of multiple electrophysiological features on dorsoventral position may be reducible to two core mechanisms that together account for much of the variability between SCs in their intrinsic electrophysiology . Is inter-animal variation present in PCA dimensions that account for dorsoventral variation ? The intercept , but not the slope of the dependence of the first two principal components on dorsoventral position depended on housing ( adjusted p=0 . 039 and 0 . 027 ) ( Figure 7E , F and Supplementary file 15 ) . After accounting for housing , the first two principal components were still better fit by models that include animal identity as a random effect ( adjusted p=3 . 3×10−9 and 4 . 1 × 10−86 ) , indicating remaining inter-animal differences in these components ( Supplementary file 16 ) . A further nine of the next ten higher-order principal components did not depend on housing ( adjusted p>0 . 1 ) ( Supplementary file 15 ) , while eight differed significantly between animals ( adjusted p<0 . 05 ) ( Supplementary file 16 ) . Together , these analyses indicate that the dorsoventral organization of multiple electrophysiological features of SCs is captured by two principal components , suggesting two main sources of variance , both of which are dependent on experience . Significant inter-animal variation in the major sources of variance remains after accounting for experience and experimental parameters .
Theoretical models suggest how different cell types can be generated by varying target concentrations of intracellular Ca2+ or rates of ion channel expression ( O'Leary et al . , 2014 ) . The within cell type variability predicted by these models arises from different initial conditions and may explain the variability in our data between neurons from the same animal at the same dorsoventral location ( Figure 8A ) . By contrast , the dependence of integrative properties on position and their variation between animals implies additional mechanisms that operate at the circuit level ( Figure 8B ) . In principle , this variation could be accounted for by inter-animal differences in dorsoventrally tuned or spatially uniform factors that influence ion channel expression or target points for intracellular Ca2+ ( Figure 8C ) . The mechanisms for within cell type variability that are suggested by our results may differ from inter-animal variation described in invertebrate nervous systems . In invertebrates , inter-animal variability is between properties of individual identified neurons ( Goaillard et al . , 2009 ) , whereas in mammalian nervous systems , neurons are not individually identifiable and the variation that we describe here is at the level of cell populations . From a developmental perspective in which cell identity is considered as a trough in a state-landscape through which each cell moves ( Wang et al . , 2011b ) , variation in the population of neurons of the same type could be conceived as cell autonomous deviations from set points corresponding to the trough ( Figure 8A ) . Our finding that variability among neurons of the same type manifests between as well as within animals , could be explained by differences between animals in the position of the trough or set point in the developmental landscape ( Figure 8B ) . Our comparison of neurons from animals in standard and large cages provides evidence for the idea that within cell-type excitable properties are modified by experience ( Zhang and Linden , 2003 ) . For example , granule cells in the dentate gyrus that receive input from SCs increase their excitability when animals are housed in enriched environments ( Green and Greenough , 1986; Ohline and Abraham , 2019 ) . Our experiments differ in that we increased the size of the environment with the goal of activating more ventral grid cells , whereas previous enrichment experiments have focused on increasing the environmental complexity and availability of objects for exploration . Further investigation will be required to dissociate the influence of each factor on excitability . Dorsoventral gradients in the electrophysiological features of SCs have stimulated cellular models for the organization of grid firing ( Burgess , 2008; Giocomo and Hasselmo , 2008b; Grossberg and Pilly , 2012; O'Donnell and Nolan , 2011; Widloski and Fiete , 2014 ) . Increases in spatial scale following deletion of HCN1 channels ( Giocomo et al . , 2011 ) , which in part determine the dorsoventral organization of SC integrative properties ( Garden et al . , 2008; Giocomo and Hasselmo , 2009 ) , support a relationship between the electrophysiological properties of SCs and grid cell spatial scales . Our data argue against models that explain this relationship through single cell computations ( Burgess , 2008; Burgess et al . , 2007; Giocomo et al . , 2007 ) , as in this case , the modularity of integrative properties of SCs is required to generate modularity of grid firing . A continuous dorsoventral organization of the electrophysiological properties of SCs could support the modular grid firing generated by self-organizing maps ( Grossberg and Pilly , 2012 ) or by synaptic learning mechanisms ( Kropff and Treves , 2008; Urdapilleta et al . , 2017 ) . It is less clear how a continuous gradient would affect the organization of grid firing predicted by continuous attractor network models , which can instead account for modularity by limiting synaptic interactions between modules ( Burak and Fiete , 2009; Bush and Burgess , 2014; Fuhs and Touretzky , 2006; Guanella et al . , 2007; Shipston-Sharman et al . , 2016; Widloski and Fiete , 2014; Yoon et al . , 2013 ) . Modularity of grid cell firing could also arise through the anatomical clustering of calbindin-positive L2PCs ( Ray et al . , 2014; Ray and Brecht , 2016 ) . Because many SCs do not appear to generate grid codes and as the most abundant functional cell type in the MEC appears to be non-grid spatial neurons ( Diehl et al . , 2017; Hardcastle et al . , 2017 ) , the continuous dorsoventral organization of SC integrative properties may also impact grid firing indirectly through modulation of these codes . Our results add to previous comparisons of medially and laterally located SCs ( Canto and Witter , 2012; Yoshida et al . , 2013 ) . The similar dorsoventral organization of subthreshold integrative properties of SCs from medial and lateral parts of the MEC appears consistent with the organization of grid cell modules recorded in behaving animals ( Stensola et al . , 2012 ) . How mediolateral differences in firing properties ( Figure 4—figure supplement 1; Canto and Witter , 2012; Yoshida et al . , 2013 ) might contribute to spatial computations within the MEC is unclear . The continuous dorsoventral variation of the electrophysiological features of SCs suggested by our analysis is consistent with continuous dorsoventral gradients in gene expression along layer 2 of the MEC ( Ramsden et al . , 2015 ) . For example , labelling of the mRNA and protein for the HCN1 ion channel suggests a continuous dorsoventral gradient in its expression ( Nolan et al . , 2007; Ramsden et al . , 2015 ) . It is also consistent with single-cell RNA sequencing analysis of other brain areas , which indicates that although the expression profiles for some cell types cluster around a point in feature space , others lie along a continuum ( Cembrowski and Menon , 2018 ) . It will be interesting in future to determine whether gene expression continua establish corresponding continua of electrophysiological features ( Liss et al . , 2001 ) . What are the functional roles of inter-animal variability ? In the crab stomatogastric ganglion , inter-animal variation correlates with circuit performance ( Goaillard et al . , 2009 ) . Accordingly , variation in intrinsic properties of SCs might correlate with differences in grid firing ( Domnisoru et al . , 2013; Gu et al . , 2018; Rowland et al . , 2018; Schmidt-Hieber and Häusser , 2013 ) or behaviors that rely on SCs ( Kitamura et al . , 2014; Qin et al . , 2018; Tennant et al . , 2018 ) . It is interesting in this respect that there appear to be inter-animal differences in the spatial scale of grid modules ( Figure 5 of Stensola et al . , 2012 ) . Modification of grid field scaling following deletion of HCN1 channels is also consistent with this possibility ( Giocomo et al . , 2011; Mallory et al . , 2018 ) . Alternatively , inter-animal differences may reflect multiple ways to achieve a common higher-order phenotype . According to this view , coding of spatial location by SCs would not differ between animals despite lower level variation in their intrinsic electrophysiological features . This is related to the idea of degeneracy at the level of single-cell electrophysiological properties ( Marder and Goaillard , 2006; Mittal and Narayanan , 2018; O'Leary et al . , 2014; Swensen and Bean , 2005 ) , except that here the electrophysiological features differ between animals whereas the higher-order circuit computations may nevertheless be similar . In conclusion , our results identify substantial within cell type variation in neuronal integrative properties that manifests between as well as within animals . This has implications for experimental design and model building as the distribution of replicates from the same animal will differ from those obtained from different animals ( Marder and Taylor , 2011 ) . An important future goal will be to distinguish causes of inter-animal variation . Many behaviors are characterized by substantial inter-animal variation ( e . g . Villette et al . , 2017 ) , which could result from variation in neuronal integrative properties , or could drive this variation . For example , it is possible that external factors such as social interactions may affect brain circuitry ( Wang et al . , 2011a; Wang et al . , 2014 ) , although these effects appear to be focused on frontal cortical structures rather than circuits for spatial computations ( Wang et al . , 2014 ) . Alternatively , stochastic mechanisms operating at the population level may drive the emergence of inter-animal differences during the development of SCs ( Donato et al . , 2017; Ray and Brecht , 2016 ) . Addressing these questions may turn out to be critical to understanding the relationship between cellular biophysics and circuit-level computations in cognitive circuits ( Schmidt-Hieber and Nolan , 2017 ) .
All experimental procedures were performed under a United Kingdom Home Office license and with approval of the University of Edinburgh’s animal welfare committee . Recordings of many SCs per animal used C57Bl/6J mice ( Charles River ) . Recordings targeting calbindin cells used a Wfs1Cre line ( Wfs1-Tg3-CreERT2 ) obtained from Jackson Labs ( Strain name: B6;C3-Tg ( Wfs1-cre/ERT2 ) 3Aibs/J; stock number:009103 ) crossed to RCE:loxP ( R26R CAG-boosted EGFP ) reporter mice ( described in Miyoshi et al . , 2010 ) . To promote expression of Cre in the mice , tamoxifen ( Sigma , 20 mg/ml in corn oil ) was administered on three consecutive days by intraperitoneal injections , approximately 1 week before experiments . Mice were group housed in a 12 hr light/dark cycle with unrestricted access to food and water ( light on 07 . 30–19 . 30 hr ) . Mice were usually housed in standard 0 . 2 × 0 . 37 m cages that contained a cardboard roll for enrichment . A subset of the mice was instead housed from pre-weaning ages in a larger 2 . 4 × 1 . 2 m cage that was enriched with up to 15 bright plastic objects and eight cardboard rolls ( Figure 2—figure supplement 1 ) . Thus , the large cages had more items but at a slightly lower density ( large cages — up to 1 item per 0 . 125 m2; standard cages — 1 item per 0 . 074 m2 ) . All experiments in the standard cage used male mice . For experiments in the large cage , two mice were female , six mice were male and one was not identified . Because we did not find significant effects of sex on individual electrophysiologically properties , all mice were included in the analyses reported in the text . When only male mice were included , the effects of housing on the first principal component remained significant , whereas the effects of housing on individual electrophysiologically properties no longer reach statistical significance after correcting for multiple comparisons . Additional analyses that consider only male mice are provided in the code associated with the manuscript . Methods for preparation of parasagittal brain slices containing medial entorhinal cortex were based on procedures described previously ( Pastoll et al . , 2012b ) . Briefly , mice were sacrificed by cervical dislocation and their brains carefully removed and placed in cold ( 2–4°C ) modified ACSF , with composition ( in mM ) : NaCl 86 , NaH2PO4 1 . 2 , KCl 2 . 5 , NaHCO3 25 , glucose 25 , sucrose 75 , CaCl2 0 . 5 , and MgCl2 7 . Brains were then hemisected and sectioned , also in modified ACSF at 4–8°C , using a vibratome ( Leica VT1200S ) . To minimize variation in the slicing angle , the hemi-section was cut along the midline of the brain and the cut surface of the brain was glued to the cutting block . After cutting , brains were placed at 36°C for 30 min in standard ACSF , with composition ( in mM ) : NaCl 124 , NaH2PO4 1 . 2 , KCl 2 . 5 , NaHCO3 25 , glucose 20 , CaCl2 2 , and MgCl2 1 . They were then allowed to cool passively to room temperature . All slices were prepared by the same experimenter ( HP ) , who followed the same procedure on each day . Whole-cell patch-clamp recordings were made between 1 to 10 hr after slice preparation using procedures described previously ( Pastoll et al . , 2013; Pastoll et al . , 2012a; Pastoll et al . , 2012b; Sürmeli et al . , 2015 ) . Recordings were made from slice perfused in standard ACSF maintained at 34–36°C . In these conditions , we observe spontaneous fast inhibitory and excitatory postsynaptic potentials , but do not find evidence for tonic GABAergic or glutamatergic currents . Patch electrodes were filled with the following intracellular solution ( in mM ) : K gluconate 130; KCl 10 , HEPES 10 , MgCl2 2 , EGTA 0 . 1 , Na2ATP 2 , Na2GTP 0 . 3 and NaPhosphocreatine 10 . The open tip resistance was 4–5 MΩ , all seal resistances were >2 GΩ and series resistances were <30 MΩ . Recordings were made in current clamp mode using Multiclamp 700B amplifiers ( Molecular Devices , Sunnyvale , CA , USA ) connected to PCs via Instrutech ITC-18 interfaces ( HEKA Elektronik , Lambrecht , Germany ) and using Axograph X acquisition software ( http://axographx . com/ ) . Pipette capacitance and series resistances were compensated using the capacitance neutralization and bridge-balance amplifier controls . An experimentally measured liquid junction potential of 12 . 9 mV was not corrected for . Stellate cells were identified by their large sag response and the characteristic waveform of their action potential after hyperpolarization ( see Alonso and Klink , 1993; Gonzalez-Sulser et al . , 2014; Nolan et al . , 2007; Pastoll et al . , 2012a ) . To maximize the number of cells recorded per animal , we adopted two strategies . First , to minimize the time required to obtain data from each recorded cell , we measured electrophysiological features using a series of three short protocols following initiation of stable whole-cell recordings . We used responses to sub-threshold current steps to estimate passive membrane properties ( Figure 2A ) , a frequency modulated sinusoidal current waveform ( ZAP waveform ) to estimate impedance amplitude profiles ( Figure 2B ) , and a linear current ramp to estimate the action potential threshold and firing properties ( Figure 2C ) . From analysis of data obtained with these protocols , we obtained 12 quantitative measures that describe the sub- and supra-threshold integrative properties of each recorded cell ( Figure 2A–C ) . Second , for the majority of mice , two experimenters made recordings in parallel from neurons in two sagittal brain sections from the same hemisphere . The median dorsal-ventral extent of the recorded SCs was 1644 µm ( range 0–2464 µm ) . Each experimenter aimed to sample recording locations evenly across the dorsoventral extent of the MEC , and for most animals , each experimenter recorded sequentially from opposite extremes of the dorsoventral axis . Each experimenter varied the starting location for recording between animals . For several features , the direction of recording affected their measured dependence on dorsoventral location , but the overall dependence of these features on dorsoventral location was robust to this effect ( Supplementary file 9 ) . Electrophysiological recordings were analyzed in Matlab ( Mathworks ) using a custom-written semi-automated pipeline . We defined each feature as follows ( see also Nolan et al . , 2007; Pastoll et al . , 2012a ) : resting membrane potential was the mean of the membrane potential during the 1 s prior to injection of the current steps used to assess subthreshold properties; input resistance was the steady-state voltage response to the negative current steps divided by their amplitude; membrane time constant was the time constant of an exponential function fit to the initial phase of membrane potential responses to the negative current steps; the sag coefficient was the steady-state divided by the peak membrane potential response to the negative current steps; resonance frequency was the frequency at which the peak membrane potential impedance was found to occur; resonance magnitude was the ratio between the peak impedance and the impedance at a frequency of 1 Hz; action potential threshold was calculated from responses to positive current ramps as the membrane potential at which the first derivative of the membrane potential crossed 20 mv ms−1 averaged across the first five spikes , or fewer if fewer spikes were triggered; rheobase was the ramp current at the point when the threshold was crossed on the first spike; spike maximum was the mean peak amplitude of the action potentials triggered by the positive current ramp; spike width was the duration of the action potentials measured at the voltage corresponding to the midpoint between the spike threshold and spike maximum; the AHP minimum was the negative peak membrane potential of the after hyperpolarization following the first action potential when a second action potential also occurred; and the F-I slope was the linear slope of the relationship between the spike rate and the injected ramp current over a 500 ms window . The location of each recorded neuron was estimated as described previously ( Garden et al . , 2008; Pastoll et al . , 2012b ) . Following each recording , a low magnification image was taken of the slice with the patch-clamp electrode at the recording location . The image was calibrated and then the distance measured from the dorsal border of the MEC along the border of layers 1 and 2 to the location of the recorded cell . Analyses of location-dependence and inter-animal differences used R 3 . 4 . 3 ( R Core Team , Vienna , Austria ) and R Studio 1 . 1 . 383 ( R Studio Inc , Boston , MA ) . To fit linear mixed effect models , we used the lme4 package ( Bates et al . , 2014 ) . Features of interest were included as fixed effects and animal identity was included as a random effect . All reported analyses are for models with the minimal a priori random effect structure , in other words the random effect was specified with uncorrelated slope and intercept . We also evaluated models in which only the intercept , or correlated intercept and slope were specified as the random effect . Model assessment was performed using Akaike Information Criterion ( AIC ) scores . In general , models with either random slope and intercept , or correlated random slope and intercept , had lower AIC scores than random intercept only models , indicating a better fit to the data . We used the former set of models for all analyses of all properties for consistency and because a maximal effect structure may be preferable on theoretical grounds ( Barr et al . , 2013 ) . We evaluated assumptions including linearity , normality , homoscedasticity and influential data points . For some features , we found modest deviations from these assumptions that could be remedied by handling non-linearity in the data using a copula transformation . Model fits were similar following transformation of the data . However , we focus here on analyses of the untransformed data because the physical interpretation of the resulting values for data points is clearer . To evaluate the location-dependence of a given feature , p-values were calculated using a χ2 test comparing models to null models with no location information but identical random effect specification . To calculate marginal and conditional R2 of mixed effect models , we used the MuMin package ( Bartoń , 2014 ) . To evaluate additional fixed effects , we used Type II Wald χ2 test tests provided by the car package ( Fox and Weisberg , 2018 ) . To compare mixed effect with equivalent linear models , we used a χ2 test to compare the calculated deviance for each model . For clarity , we have reported key statistics in the main text and provide full test statistics in the Supplemental Tables . In addition , the code from which the analyses can be fully reproduced is available at https://github . com/MattNolanLab/Inter_Intra_Variation ( Nolan , 2020; copy archived at https://github . com/elifesciences-publications/Inter_Intra_Variation ) . To evaluate partial correlations between features , we used the function cor2pcor from the R package corpcor ( Schafer et al . , 2017 ) . Principal components analysis used core R functions . To establish statistical tests to distinguish ‘modular’ from ‘continuous’ distributions given relatively few observations , we classified datasets as continuous or modular by modifying the gap statistic algorithm ( Tibshirani et al . , 2001 ) . The gap statistic estimates the number of clusters ( kest ) that best account for the data in any given dataset ( Figure 1—figure supplement 1A-C ) . However , this estimate may be prone to false positives , particularly where the numbers of observations are low . We therefore introduced a thresholding mechanism for tuning the sensitivity of the algorithm so that the false-positive rate , which is the rate of misclassifying datasets drawn from continuous ( uniform ) distributions as ‘modular’ , is low , constant across different numbers of cluster modes and insensitive to dataset size ( Figure 1—figure supplement 1D-G ) . With this approach , we are able to estimate whether a dataset is best described as lacking modularity ( kest = 1 ) , or having a given number of modes ( kest > 1 ) . Below , we describe tests carried out to validate the approach . To illustrate the sensitivity and specificity of the modified gap statistic algorithm , we applied it to simulated datasets drawn either from a uniform distribution ( k = 1 , n = 40 ) or from a bimodal distribution with separation between the modes of five standard deviations ( k = 2 , n = 40 , sigma = 5 ) ( Figure 1—figure supplement 2A ) . We set the thresholding mechanism so that kest for each distinct k ( where kest ≥2 ) has a false-positive rate of 0 . 01 . In line with this , testing for 2 ≤ kest ≤ 8 ( the maximum k expected to occur in grid spacing in the MEC ) , across multiple ( N = 1000 ) simulated datasets drawn from the uniform distribution , produced a low false-positive rate ( P ( kest ) ≥2 = ~0 . 07 ) , whereas when the data were drawn from the bimodal distribution , the ability to detect modular organization ( pdetect ) was good ( P[kest]≥2 = ~0 . 8 ) ( Figure 1—figure supplement 2B ) . The performance of the statistic improved with larger separation between clusters and with greater numbers of data points per dataset ( Figure 1—figure supplement 2C ) and is relatively insensitive to the numbers of clusters ( Figure 1—figure supplement 2D ) . The algorithm maintains high rates of pdetect when modes have varying densities and when sigma between modes varies in a manner similar to grid spacing data ( Figure 1—figure supplement 3 ) . Recently described algorithms ( Giocomo et al . , 2014; Stensola et al . , 2012 ) address the problem of identifying modularity when data are sampled from multiple locations and data values vary as a function of location , as is the case for the mean spacing of grid fields for cells at different dorsoventral locations recorded in behaving animals using tetrodes ( Hafting et al . , 2005 ) . They generate log-normalized discontinuity ( which we refer to here as lnDS ) or discreteness scores , which are the log of the ratio of discontinuity or discreteness scores for the data points of interest and for the sampling locations , with positive values interpreted as evidence for clustering ( Giocomo et al . , 2014; Stensola et al . , 2012 ) . However , in simulations of datasets generated from a uniform distribution with evenly spaced recording locations , we find that the lnDS is always greater than zero ( Figure 1—figure supplement 4A ) . This is because evenly spaced locations result in a discontinuity score that approaches zero , and therefore the log ratio of the discontinuity of the data to this score will be positive . Thus , for evenly spaced data , the lnDS is guaranteed to produce false-positive results . When locations are instead sampled from a uniform distribution , approximately half of the simulated datasets have a log discontinuity ratio greater than 0 ( Figure 1—figure supplement 4A ) , which in previous studies would be interpreted as evidence of modularity ( Giocomo et al . , 2014 ) . Similar discrepancies arise for the discreteness measure ( Stensola et al . , 2012 ) . To address these issues , we introduced a log discontinuity ratio threshold , so that the discontinuity method could be matched to produce a similar false-positive rate to the adapted gap statistic algorithm used in the example above . After including this modification , we found that for a given false-positive rate , the adapted gap statistic is more sensitive at detecting modularity in the simulated datasets ( Figure 4—figure supplement 1B ) . To establish whether the modified gap statistic detects clustering in experimental data , we applied it to previously published grid cell data recorded with tetrodes from awake behaving animals ( Stensola et al . , 2012 ) . We find that the modified gap statistic identified clustered grid spacing for 6 of 7 animals previously identified as having grid modules and with n ≥ 20 . For these animals , the number of modules was similar ( but not always identical ) to the number of previously identified modules ( Figure 1—figure supplement 5 ) . By contrast , the modified gap statistic does not identify clustering in five of six sets of recording locations , confirming that the grid clustering is likely not a result of uneven sampling of locations ( we could not test the seventh as location data were not available ) . The thresholded discontinuity score also detects clustering in the same five of the six tested sets of grid data . From the six grid datasets detected as clustered with the modified gap statistic , we estimated the separation between clusters by fitting the data with a mixture of Gaussians , with the number of modes set by the value of k obtained with the modified gap statistic . This analysis suggested that the largest spacing between contiguous modules in each mouse is always >5 . 6 standard deviations ( mean = 20 . 5 ± 5 . 0 standard deviations ) . Thus , the modified gap statistic detects modularity within the grid system and indicates that previous descriptions of grid modularity are , in general , robust to the possibility of false positives associated with the discreteness and discontinuity methods . | The brain consists of many types of cells that are specialised to perform different tasks . This is similar to how different groups of people will have different responsibilities in a large company . But within each group with the same role , individual employees will also do their jobs in different ways . Does the same apply to the brain ? In other words , do individual neurons of the same type – with the same role – process information differently ? To find out , Pastoll et al . studied stellate cells in the mouse brain: these neurons take their name from their distinctive star-shaped arrays of projections , and they work together in groups known as modules to help animals navigate their environment . To determine whether stellate cells differ between mice , and how they might differ within a single animal , Pastoll et al . measured the activity of more than 800 stellate cells in more than two dozen individuals . The results revealed that stellate cells process the same information differently between mice , which may contribute to variations in behaviour across the species . But even within an individual , stellate cells also showed differences in information processing . In fact , the properties of the stellate cells within each mouse varied along a continuum . This discovery rules out several previous theories on how stellate cells form the modules that support navigation . The work by Pastoll et al . helps to understand how the brain supports thinking and memory . In the long term , these findings could also have implications for treating brain disorders , as they suggest that variations between people in the properties of their neurons could lead to variations in drug response . Researchers may need to take inter-individual differences into account when planning experiments , and ultimately when designing drugs . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2020 | Inter- and intra-animal variation in the integrative properties of stellate cells in the medial entorhinal cortex |
In all animals , sleep pressure is under continuous tight regulation . It is universally accepted that this regulation arises from a two-process model , integrating both a circadian and a homeostatic controller . Here we explore the role of environmental social signals as a third , parallel controller of sleep homeostasis and sleep pressure . We show that , in Drosophila melanogaster males , sleep pressure after sleep deprivation can be counteracted by raising their sexual arousal , either by engaging the flies with prolonged courtship activity or merely by exposing them to female pheromones .
The two-process model for regulation of sleep , first postulated by Borbély in 1982 , is still considered the most accurate conceptual framework to describe how sleep pressure builds and dissipates along the day ( Borbély , 1982 ) . According to the model , sleep propensity at any given time integrates two independent biological mechanisms: a circadian regulator ( process C ) and a homeostatic regulator ( process S ) . The circadian regulator is under control of the circadian clock and independent of external factors . The homeostatic regulator , on the other hand , is a tracker of past sleep and is responsible for the accumulation of sleep pressure upon sleep deprivation , or its release after a nap ( Borbély and Achermann , 1999 ) . The idea of a ‘process S’ is historically based upon electrophysiological recordings obtained in mammals and , in particular , on the observation that low-frequency electrophysiological correlates of neuronal activity — Slow Wave Sleep — increase with sleep deprivation , dissipate with sleep , and thus can ultimately act as biological markers for sleep pressure . The basic separation between a circadian and a homeostatic regulator , however , is a fundamental concept that lives beyond electrophysiology and can be adopted to model sleep pressure also in animals where the electroencephalographic ( EEG ) correlates of sleep are very different or unknown ( Campbell and Tobler , 1984 ) . In virtually all animals tested so far , sleep deprivation leads to a subsequent increase in sleep rebound ( Cirelli and Tononi , 2008 ) . Understanding the biological underpinnings of process C and process S is an important investigative task , not only to uncover the mechanisms regulating sleep , but ultimately its function too . Discovering how and why sleep pressure increases upon sleep deprivation may be critical to ultimately unravel what sleep is for . Besides a homeostatic and a circadian controller , we do know that other factors can modulate sleep . Most people in western society will lament poor sleep habits and this is generally not due to malfunctioning of process S or process C but , instead , to societal or generally environmental and emotional causes ( e . g . stress , anxiety , excitement , hunger , love ) ( Ohayon , 2002; Adolescent Sleep Working Group et al . , 2014 ) . From the experimental perspective , changes in environmental temperature and food restriction constitute two important examples of sleep modulation by environmental conditions . In flies , an increase in temperature during the night has been shown to have profound effects on sleep pattern , but not necessarily on total sleep amounts ( Lamaze et al . , 2017; Parisky et al . , 2016 ) . In rats ( Danguir and Nicolaidis , 1979 ) , humans ( MacFadyen et al . , 1973 ) , and flies ( Keene et al . , 2010 ) starvation has been shown to lead to a rapid decrease in sleep amount . In mammals , this also correlates with qualitative differences in the EEG pattern ( Danguir and Nicolaidis , 1979 ) . Besides a strong evolutionary conservation at the behavioural level , caloric intake and sleep are also genetically linked , as the same proteins and neuromodulators have been shown to control both ( Willie et al . , 2001 ) . However , the relationship between the two is also complicated by the fact that caloric restriction has profound consequences on metabolism . Here we describe a new paradigm to study the behavioural , neuronal , and genetic connection between environment and sleep: sex drive . We find that , in male flies , sexual arousal has profound effects on sleep , and that sexual experience or even exposure to pheromones alone are sufficient stimuli to counteract sleep pressure after sleep deprivation .
After being forcefully deprived of sleep , Drosophila melanogaster consistently show an increase in sleep pressure , in the form of a concomitant increase in sleep amount ( Huber et al . , 2004 ) and in arousal threshold ( Faville et al . , 2015 ) . In other words , flies , like mammals , appear to sleep longer and deeper after sleep deprivation and both are clear signs of what is normally referred as ‘sleep rebound’ , a hallmark of sleep homoeostasis . To deprive flies of sleep , most researchers would use mechanical machines , such as laboratory shakers , that subject animals to frequent , if not continuous , vibratory stimuli ( Faville et al . , 2015; Huber et al . , 2004 ) . We previously showed that a spatially restricted , forced interaction between two males also leads to a robust sleep deprivation that has the same behavioural and cellular characteristics of mechanical deprivation , including a similar extent of detectable rebound and comparable biochemical correlates ( Gilestro et al . , 2009 ) . To further investigate how social interaction affects sleep , we devised an experimental paradigm based on computer-assisted video analysis of behaviour . Using ethoscopes , video tracking machines recently developed in our laboratory ( Geissmann et al . , 2017 ) , we monitored and annotated the behaviour of flies either in isolation ( baseline and rebound days ) or in groups of two ( interaction day ) . The advantages of using video tracking over the infrared beam split system when measuring sleep have been discussed at length elsewhere ( Donelson et al . , 2012; Gilestro , 2012; Zimmerman et al . , 2008 ) and , arguably , video tracking becomes even more compelling when monitoring multiple flies interacting in the same space . In our archetypical experiment , wild-type male flies ( CantonS ) were kept in social isolation in small glass tubes for five days , in order to acclimatise to their environment and to record their baseline activity ( see Materials and methods ) . Then , at the beginning of interaction day , we introduced a second individual in the restricted recording space: the intruder . For MM interactions , the intruder was another male of different eye colours ( white1118 , cyan in figures ) . For MF interactions , a wild-type virgin female ( peach in figures ) . Mock control male animals underwent the same experimental manipulation but were kept in isolation also during interaction day ( mock , grey in figures ) . In all cases , interaction lasted no more than 24 hr . Confirming previous results ( Gilestro et al . , 2009 ) , we found that MM interactions consistently led to a sleep deprivation during the interaction period , and to a noticeable rebound immediately after ( Figure 1A–D and Figure 1—figure supplement 1A , B ) . The MF interaction led to an even greater deprivation of sleep ( Figure 1C ) but , surprisingly , did not show any subsequent rebound ( peach in Figure 1B ) . Why ? One first explanation could be that our tracking system overestimates the extent of sleep deprivation experienced in the MF interaction . To explore this possibility , we video-recorded interacting animals and manually scored their behaviour ( Figure 1D and interactive video currently available at https://lab . gilest . ro/projects/raw-data/regulation-of-sleep-homeostasis-by-sex-pheromones-supplementary-videos/ – MF: 1082 bins scored per day; MM: 1247 bins scored per day ) as well as their euclidean coordinates ( scored 347 times in a day for Mock , MM and MF . Nmock = 12 , NMM = 11 , NMF = 11 ) . Human scoring confirmed machine scoring , as well as previous results ( Fujii et al . , 2007 ) , and showed that the MF interaction led indeed to a sustained increase in activity ( Figure 1D and Figure 1—figure supplement 2 ) . In particular , even though all couples copulated within minutes from the start of the interaction ( 16 . 6 ± 15 . 7 min; mean ±SD ) , male flies still spent on average 47% of their time actively courting the female ( 47 ± 16% over 24 hr; 61 ± 25% during the day and 33 ± 12% during the night; mean ±SD ) . Flies engaged in MM interaction , on the other hand , were not as physically active as flies in MF ( Figure 1D and Figure 1—figure supplement 2 ) , thus not explaining but instead reinforcing the apparent paradox of absence of sleep rebound after interaction with a female . To further characterise the consequences of social interaction , we also used a recently established CaLexA assay ( Liu et al . , 2016 ) to compare , a posteriori , the neuronal activity in the R2 neurons of the ellipsoid body after 24 hr of social interaction ( MM or MF ) or 24 hr of mechanical sleep deprivation ( Figure 1E , F ) . The CaLexA system uses a calcium-responsive transcription factor to drive a green fluorescent protein ( GFP ) in neurons that undergo prolonged firing activity ( Masuyama et al . , 2012 ) . Firing rate of R2 neurons was shown to correlate with sleep drive , therefore an increase of CaLexA fluorescence in those neurons can be interpreted as a bona fide proxy for neuronal firing , and ultimately , for sleep pressure ( Liu et al . , 2016 ) . In all three experimental conditions , R2 neurons labelled by the R30G03-GAL4 driver showed a sustained and similar increase in detectable CaLexA-GFP levels compared to mock ( Figure 1E , F ) , suggesting that all three conditions elicit a comparably efficient sleep deprivation . Together , these results show that males who engaged in sexual interaction ( a ) experience a highly efficient sleep deprivation ( Figure 1C and Figure 1—figure supplement 2 ) and ( b ) exhibit increases in neuronal markers which typically appear after prolonged wakefulness ( Figure 1E and F ) . Why do they show no rebound sleep , then ? One possibility is that the memory of their recent sexual encounter could motivate them to keep searching for a mating partner . To test this hypothesis , we subjected two canonical memory mutants to the same experimental paradigm: dunce and rutabaga ( Figure 2A , B and Figure 2—figure supplement 1A , B ) . Both mutants are amongst the first and the best-characterised memory mutants discovered in Drosophila ( Davis and Dauwalder , 1991; Levin et al . , 1992 ) and have been shown to be unable to consolidate memory in many paradigmatic conditions , including courtship conditioning ( Griffith and Ejima , 2009 ) . While the role of dunce and rutabaga in the context of courtship conditioning is well described ( Joiner and Griffith , 1999 ) , it is not known whether flies possess any memory of past sexual experience . We speculated that if recollection of past experience is responsible for the suppression of rebound , one may expect to see a regular rebound in forgetful flies . This was not the case ( Figure 2A , B and Figure 2—figure supplement 1A , B ) . As observed in wild-type flies , learning mutants also experienced a strong degree of sleep deprivation when forced to interact with females ( Figure 2—figure supplement 1A , B ) , but they also lacked sleep rebound the day after . Regular rebound was once again observed after MM interaction . If it is not a memory of the past experience that is responsible for the suppression of sleep rebound , could it be due to a physical trace left in the environment ? Could either a volatile or non-volatile sex pheromone be left in the tube after the MF interaction , thus contributing to prolonging a signal of sexual arousal ? We reckoned one way to approach this hypothesis would be to force an inter-species sexual interaction: a large number of olfactory and gustatory stimuli contributes to the complex courtship ritual between males and females ( Dweck et al . , 2015 ) and a convenient way to rule many at once is to force interaction between D . melanogaster males and a close evolutionary relative , such as D . simulans ( Manning , 1959; Sturtevant , 1919 ) . We therefore placed D . melanogaster wild-type males with D . simulans females on interaction day and video-recorded , then scored , their behaviour . In accordance with the classical literature ( Schilcher and Dow , 1977 ) , the inter-species MF interaction resulted in limited copulation ( only 2 flies out of the 11 that were visually monitored , Figure 2—figure supplement 2A ) , but with some degree of courting mainly during the day ( 10 . 8 ± 2 . 3% over 24 hr but only 2 . 3 ± 1 . 4% during ZT 12–24; mean ±SD ) , followed by sleep deprivation throughout the night ( Figure 2—figure supplement 2B ) . However , after inter-species MF interaction , male flies finally did show a sleep rebound that was even greater than the rebound observed after MM interactions ( Figure 2C and Figure 2—figure supplement 1C ) . Interestingly , even though we never observed fighting behaviour between D . melanogaster males and D . simulans females , the activity profile , the limited courtship , and the rebound were more reminiscent of MM interaction than MF interaction . The sleep rebound observed after inter-species interaction suggests that a possibly arousing chemical signal left by the female may be responsible for the suppression of rebound after D . melanogaster specific MF interaction . In Drosophila , some pheromones have a certain degree of volatility ( Farine et al . , 2012 ) and , to test whether an olfactory signal was involved with this process , we measured rebound after social interaction in the anosmic orco mutants ( Larsson et al . , 2004 ) but found no difference between wild-type and orco mutant flies: anosmic males also lacked rebound after MF-induced sleep deprivation ( Figure 2D and Figure 2—figure supplement 1D ) . If there is an arousing signal that anosmic flies can still perceive , could this be a non-volatile pheromone ? Females of D . melanogaster and D . simulans have different cuticular hydrocarbons acting as sex pheromones , with the former bearing predominantly 7 , 11-Heptacosadiene ( 7 , 11-ND ) and the latter 7-Tricosene ( 7-HD ) ( Jallon , 1984; Marcillac et al . , 2005 ) . If olfactory signals are not involved , we reasoned that a D . melanogaster specific cuticle pheromone could be responsible for the puzzling phenotype . We , therefore , subjected wild-type male flies to MM interaction and then , at the dawn of rebound day , we removed the intruder and inserted in the recording tube a fragment of paper on which we had previously diluted a mix of the species-specific sex pheromones 7 , 11-ND and 7 , 11-HD or the solvent alone as control ( Figure 2E and Figure 2—figure supplement 1E ) . At last , we found that the mere presence of D . melanogaster female cuticular pheromones could indeed inhibit sleep rebound after MM interaction , suggesting the pheromones left by the female were sufficient to counteract sleep pressure accumulated at rebound day . Male flies sense female non-volatile pheromones through neurons located on the distal tip of their forelegs ( Lu et al . , 2012; Starostina et al . , 2012; Thistle et al . , 2012; Toda et al . , 2012; Vijayan et al . , 2014 ) . At the beginning of the sexual courtship ritual , male flies tap the female to presumably taste and recognise sex-specific signals ( Spieth , 1974 ) that are important for courting to continue . In particular , 7 , 11-ND and 7 , 11-HD are sensed by neurons expressing members of the degenerin/epithelial sodium channel ( DEG/ENaC ) family — Ppk25 , Ppk23 , and Ppk29 — and male flies mutants in either of these receptors show a decreased level of courtship ( Liu et al . , 2012; Starostina et al . , 2012; Toda et al . , 2012; Vijayan et al . , 2014 ) . Behavioural and electrophysiological data showed that sex-pheromones detection is almost completely lost in ppk23 mutant males ( Lu et al . , 2012; Thistle et al . , 2012; Toda et al . , 2012 ) and therefore we reasoned that ppk23 mutants could serve as a good model to test , once more , the hypothesis that suppression of sleep rebound is due to pheromone signalling . We then subjected male ppk23 mutant flies to four experimental conditions: MF interaction , MM interaction , and MM interaction with or without the addition of exogenous pheromones ( Figure 3 ) . As predicted , flies underwent the expected level of sleep deprivations ( Figure 3B–E ) , but MF condition did not show an abnormal sleep rebound ( Figure 3A ) , indicating that ppk23 signalling during the interaction plays a role in suppressing sleep rebound . The data collected until this point show that male flies , exposed to female pheromone , will downregulate their natural need for sleep rebound after sleep deprivation . However , the attentive reader will have realised that all experiments performed so far rely on a social paradigm for sleep deprivation , therefore introducing a confounding condition: is the mere presence of pheromones truly sufficient to suppress sleep deprivation , or is this somehow connected to the social nature of our behavioural paradigm ? After all , we do know that social interaction in flies can have profound effects on their sleep ( Ganguly-Fitzgerald et al . , 2006 ) . To test sufficiency of pheromones effect on sleep rebound , we devised two sets of experiments , in which we replaced social-driven sleep deprivation with mechanical sleep deprivation , using the sleep deprivation module of our ethoscopes ( Figures 4 and 5 ) . Ethoscopes can interact with single flies in a context dependant manner , triggering events upon a real-time analysis of behaviour ( Geissmann et al . , 2017 ) . We programmed the ethoscopes to rotate a tube whenever the animal inside was detected completely inactive for 60 s . We call this paradigm: dynamic sleep deprivation ( Geissmann et al . , 2017 ) . In the first set of experiments ( Figure 4 ) , CantonS flies were subjected to dynamic SD for 24 hr , then transferred into their same tube ( mock manipulation – Figure 4A , B ) , to a clean and fresh tube ( Figure 4C , D ) , or transferred into a tube where a virgin female was previously housed for five days ( Figure 4E , F ) . Sleep rebound after SD was observed in the first two cases , but not in the last , suggesting that pheromones left by the previously hosted female are indeed sufficient to suppress rebound . To ultimately test the sufficiency of sex pheromones , we conducted a second set of experiments in which flies were engineered to express the thermo-activated channel TrpA1 in the pheromone sensing cells expressing the ppk23-GAL4 driver ( Figure 5 ) . These flies were also subjected to dynamic SD for 24 hr and then the temperature was raised from the inactivating ( 22°C ) to the activating ( 29°C ) condition to synaesthetically stimulate pheromone sensation . Of all experimental conditions ( Figure 5 and Figure 5—figure supplement 1 ) , lack of rebound after SD was observed only when pheromone sensing cells were thermogenetically stimulated ( Figure 5E , F ) . The results collected so far indicate that pheromone signalling has the ability to suppress sleep rebound after sleep deprivation . However , is pheromone the only cue able to do so ? After all , suppression of sleep observed in the archetypical experiment ( Figure 1A ) appears to be even stronger than the suppression observed after exposing flies to pheromone alone . To address this question , we performed two sets of experiments . In a first set , we subjected flies to the usual social interaction paradigm but , at the end of the interaction day , we removed the intruders and transferred the focal flies not in their own tube as done previously , but instead in a clean , fresh tube ( Figure 6A , B ) . Indeed , we found that also when transferred to a clean tube — and therefore in the absence of residual female pheromones in their environment — male flies did show a suppressed rebound ( Figure 6A , B - compare with Figure 1A , B ) . In a second set of experiments , we used the thermosensitive form of the neuronal inhibitor shibire ( shiTS ) to selectively silence ppk23 neurons at the dawn of rebound day , after social interaction ( Figure 6C–H ) . As expected , silencing of the pheromone signalling cells after MF interaction did not rescue the sleep rebound phenotype . These findings strongly suggests that suppression of sleep rebound may be due to a general state of sexual arousal , which could be elicited either by the recent sexual experience or by the presence of sex pheromones: the combination of both factors may then act in synergy to manifest the stronger effect shown in Figure 1A , B . Sexual arousal in D . melanogaster males is known to be largely under control of the sexually dimorphic P1 cluster of fruitless expressing neurons ( Yamamoto and Koganezawa , 2013 ) . P1 neurons are activated by contact with females ( Kohatsu and Yamamoto , 2015 ) and , conversely , experimental activation of P1 neurons is sufficient to trigger or enhance courtship behaviour ( Kohatsu et al . , 2011 ) , possibly generating an internal state of sexual arousal . Therefore , to test the ultimate shape of our hypothesis , we expressed the thermoactivated channel TrpA1 in the P1 neurons and raised the temperature to the activating condition ( 29°C ) for 24 hr ( Figure 7 ) . Sustained and prolonged activation of P1 neurons led to a phenotype of prolonged activity and almost total suppression of sleep ( Figure 7D ) , largely similar to the one observed when pairing a male fly with a female partner . Most importantly , the sleep deprivation induced by activating P1 neurons also did not lead to sleep rebound but , on the contrary , to a noticeable reduction of sleep on rebound day ( Figure 7D , E ) . To further confirm that suppression of sleep rebound is indeed controlled by sexual arousal , we performed two final genetic manipulations , both involving the octopamine and tyramine synthesis pathway ( Figure 8A ) . The octopaminergic system is a key regulator of Drosophila behaviour , involved among others , with modulation of sexual activity ( Huang et al . , 2016; Zhou et al . , 2012 ) , male-male aggression ( Zhou et al . , 2008 ) and sleep ( Crocker and Sehgal , 2008; Crocker et al . , 2010; Yang et al . , 2015 ) . In the first set of experiments , we subjected flies mutant for the TβH enzyme to the social interaction paradigm . In TβH mutant flies , octopamine synthesis is impaired and this has been linked to a deficit in the ability to form courtship conditioning ( Zhou et al . , 2012 ) a paradigm in which male flies learn to suppress their sexual instincts , after having been repeatedly rejected ( Griffith and Ejima , 2009 ) . In an educated guess , we reasoned that TβH mutant flies lacking courting conditioning may , therefore , show an abnormal increase of sexual arousal after prolonged social interaction and , possibly , show an increased effect in suppression of rebound . This was indeed the case ( Figure 8B , C ) . TβH mutant flies showed a clear rebound after MM interaction and a strong suppression of rebound after MF interaction . Finally , to investigate whether mere sexual arousal is responsible for this effect , we used flies mutant in the TDC2 gene , that possess lower levels of tyramine and octopamine ( Crocker and Sehgal , 2008 ) and were previously shown to court male as well as female flies ( Huang et al . , 2016 ) . We hypothesised that if these flies are sexually aroused by both male and female partners , they should then respond with a suppression of sleep rebound to both conditions of social interaction . This was what we observed indeed ( Figure 8D , E ) . In flies with a bi-sexual orientation , both MF and MM interaction lead to a strong suppression of sleep rebound .
The main finding of this work is that sexual arousal has the ability to modulate sleep pressure . We use different behavioural paradigms to promote a state of sexual arousal in male flies and show that , in all cases , this results in a suppression of sleep rebound following sleep deprivation . Why is this important ? In the past 15 years , Drosophila has emerged as one of the most promising animal models to study the biological underpinnings of sleep . Many genes that affect sleep in Drosophila have been identified so far , and many neuronal circuits that can alter sleep when manipulated have been described ( Potdar and Sheeba , 2013; Tomita et al . , 2017 ) . Given that the framework for sleep regulation is stably centred around the two-process model , newly identified neurons modulating sleep are normally classified either as involved with circadian regulation — and thus belonging to process C — or as involved with homeostatic regulation — and thus belonging to process S . Here , we identified an internal state that has the ability to modulate sleep and sleep pressure but arguably does not belong to either process . Historically , accessory regulation of sleep has been attributed to neuromodulators and , again , Drosophila has proven instrumental in understanding how neuromodulators influence sleep ( Griffith , 2013 ) . However , environmental control of sleep is likely to extend beyond neuromodulators and indeed likely to encompass specific sensory and central circuits . Using optogenetics and thermogenetics , it is now possible to activate and silence single neurons or entire circuits looking for functional correlates of behaviour . A proper characterisation of possible outcomes is a necessary step: how can we distinguish if a neuron’s main job is to directly regulate sleep pressure or , for instance , to create a state of anxiety , hunger or sexual arousal , that indirectly modulates sleep pressure ? Paraphrasing a famous assay by Thomas Nagel , we cannot know what is like to be a fly ( Nagel , 1974 ) : does exposure to sex pheromones create an inner status of sexual arousal that then counteracts sleep , or does it directly interfere with sleep regulation without any further sexual implication ? Manipulation of ppk23 neurons , either thermogenetically or by the use of chemicals , does not elicit any clear sign of courtship ( data not shown ) and this is in accordance with previous literature , where it was also shown that activation of ppk23 neurons alone is not sufficient to induce any sign of sexual behaviour in isolated flies ( Starostina et al . , 2012; Toda et al . , 2012 ) and that the right pheromones can act , instead , to potentiate other concomitant sexual stimuli . In our paradigm , activation of P1 neurons also does not show any clear sign of courtship , such as singing through wing extension ( data not shown ) . For the sleep field , this work offers a novel experimental paradigm that could be used to dissect , in an ecologically meaningful way , how internal drives or environmental stimuli affect sleep regulation and sleep homeostasis . The interaction between sex and sleep in Drosophila , and more specifically the hierarchy of those two concurrent biological drives , was initially described in the frame of circadian interaction ( Fujii et al . , 2007 ) and very recently the neuronal underpinnings were investigated by two independent groups ( Chen et al . , 2017; Machado et al . , 2017 ) . In particular , Machado et al . ( 2017 ) and Chen et al . ( 2017 ) also find a role for the P1 neurons in diverting an animal’s interest from sleep to sex . Our work , however , does not focus on the binary choice between sleep and courtship , but rather uncovers a new role for sexual arousal on modulation of sleep homeostasis , also in absence of a female partner . The concept that sleep homeostasis is not inviolable and can actually be modulated is not a novel one: migratory birds and cetaceans were reported to have the ability to suppress sleep at certain important periods of their lives , namely during migration or immediately after giving birth ( Fuchs et al . , 2009; Lyamin et al . , 2005; Rattenborg et al . , 2004 ) ; flies , similarly , were shown to lack sleep rebound after starvation-induced sleep deprivation ( Thimgan et al . , 2010 ) or after induction of sleep deprivation through specific neuronal clusters ( Seidner et al . , 2015 ) . Perhaps even more fitting with our findings is the observation that male pectoral sandpipers , a type of Arctic bird , can forego sleep in favour of courtship during the three weeks time window of female fertility ( Lesku et al . , 2012 ) . It appears , therefore , that animals are able to balance sleep needs with other , various , biological drives . It would be interesting to see whether these drives act to suppress sleep through a common regulatory circuit . Rebound sleep has always been considered one of the most important features of sleep itself . Together with the reported death by sleep deprivation , it is frequently used in support of the hypothesis that sleep is not an accessory phenomenon but a basic need of the organism ( Cirelli and Tononi , 2008 ) . Understanding the regulation of rebound sleep , therefore , may be crucial to understanding the very function of sleep . Interestingly , in our paradigm rebound sleep is not postponed , but rather eliminated . Moreover , on rebound day , the sleep architecture of sexually aroused male flies does not seem to be affected: the sleep bout numbers appear to be similar to their mock control counterparts , while the length of sleep bouts is , if anything , slightly reduced ( Figure 1—figure supplement 1 ) . The last remark that arises from our finding concerns the use of Drosophila melanogaster as a model for complex brain functions , such as emotions . Drosophila neurobiology is experiencing a period of renaissance , driven by a Cambrian explosion of genomics , ethomics and connectomics . The field may soon be able to use fruit flies for behavioural models that were once considered to be an exclusive of mammals - or even humans . Past examples of these behaviours are aggression or sleep itself . Studying emotions or internal states in animals is not an easy task , given their subjective nature . However , studying the effects of emotions on sleep may open a window of opportunity , by providing an easily quantifiable output .
Flies were raised under a 12 hr light:12 hr dark ( LD ) regimen at 25 on standard corn and yeast media . Following lines were used in the study: CantonS from Ralf Stanewsky ( UCL , UK ) ; D . simulans from Virginie Orgogozo ( IJM , France ) ; ppk23-GAL4 and ppk23Δ mutants ( Toda et al . , 2012 ) from Barry Dickson ( HHMI , USA ) ; CaLexA ( Masuyama et al . , 2012 ) from Marc Dionne ( ICL , UK ) ; UAS-shiTS from James Jepson ( UCL ) ; R30G03-GAL4 ( #49646 ) ( Liu et al . , 2016 ) , dunce1 ( #6020 ) , rutabaga1 ( #9404 ) , and orco1 ( #23129 ) mutants , UAS-TrpA1 ( #26263 ) from Bloomington Drosophila Stock Centre ( Indiana , USA ) . The P1-split-GAL4 driver was created and provided by Eric Hoopfer ( Hoopfer et al . , 2015 ) . TβHnM18 and TDC2R054 are from Stephen Goodwin ( CNCB , Oxford ) . Animals were grown and treated in the same conditions as in behavioural experiments . After a day of social interaction or mechanical sleep deprivation , animals were anaesthetised and their brains were dissected and fixed as previously described ( Beckwith et al . , 2013 ) . For CaLexA measurements , fly brains were immunostained with anti-GFP ( 1:400 , ab290 Abcam ) . Images were taken under 40X magnification and analysed in Fiji/ImageJ ( Schindelin et al . , 2012 ) . To measure signal intensities , a maximal intensity projection of all the stack comprising the R2 ring was generated . A doughnut shaped region of interest was superimposed to measure mean grey value for each R2 ring . Intensity on an adjacent non-labelled region was measured and subtracted . To allow comparison with previously published data ( Liu et al . , 2016 ) , mechanical sleep deprivation was conducted by placing the flies on top of a laboratory shaker controlled by an Arduino timer activated in pulses of 5 to 30 s at pseudo-random intervals of 1 to 7 min ( Arduino code and instructions on https://github . com/gilestrolab/fly-sleepdeprivator ) . Sleep recordings were performed using ethoscopes ( Geissmann et al . , 2017 ) under 12:12 LD condition , 50–70% humidity , in incubators set at 25 . In all experiments , environmental values of temperature , humidity , and light were recorded and monitored every 5 min . For social interactions , zero to one-day old flies were removed from a shared vial and placed in 70 mm x 5 mm glass tubes containing standard food . Twenty tubes were placed in each ethoscope arena . Flies were acclimated in behavioural glass tubes for 5 days of which the last 2 days were recorded as a baseline . On the interaction day , intruders ( CantonS females or white1118 males ) were added at ZT0 . Intruders were then removed from ZT23 to ZT24 , finishing 10 min before the dark to light transition . Rebound period was then recorded for two consecutive days . All figures show the last baseline day and the first rebound day . Manipulation of flies and recording of interaction was performed as in all other experiments with the only difference that experiments were video-recorded using the recording function of ethoscopes . Videos were recorded with a resolution of 1920 × 1080 pixels and a frame rate of 25 FPS . The degree of interaction was then scored using a web-based graphical interface , available upon request . Behavioural labelling was done at a frequency of approximately once every 60 s , while positional scoring with a frequency of once every 240 s . For the pheromone experiments , a small fragment of 3 MM filter paper containing the pheromones mix ( 70 ng in 10 µl of 7 ( Z ) , 11 ( Z ) -Nonacosadiene and 70 ng in 10 µl of 7 ( Z ) , 11 ( Z ) -Heptacosadiene; Cayman Chemicals , Ann Arbor , Michigan 48108 USA ) or the vehicle ( hexane ) was added to the tube just after removal of the intruder male . Sleep deprivation was conducted using the servo motors module of the ethoscope platform ( Geissmann et al . , 2017 ) . All bouts of immobility lasting at least 60 s were automatically interrupted by the machine rotating individual experimental tubes , thus awakening the flies only when they were quiescent . For each stimulation , motors rotate three times: −85° 200 ms , +170° 300 ms , −85° 200 ms . For experiments employing TrpA1 and shiTS , animals were raised in incubators set at 22°C . Baseline recordings and sleep deprivation were performed also in incubators set at the same temperature but the actual recorded temperature oscillated between 22°C and 24°C due to heat produced by ethoscopes themselves . Thermo-manipulation was conducted at 29°C . In all experiments , environmental conditions of light , temperature , and humidity were recorded with a frequency of once every 5 min . For the shiTS experiments shown in Figure 6C–H , the temperature was raised to 29°C at ZT23:30 . All data analysis was performed in R ( Core Team , 2014 ) or in Python ( Team , 2015 ) . Behavioural data were analysed with the R package Rethomics ( https://github . com/gilestrolab/rethomics ) and statistical analysis consisted of pairwise Wilcoxon rank sum test ( i . e . Mann–Whitney U test ) with P value adjustment for multiple comparisons ( Benjamini and Hochberg , 1995 ) . For ethograms , bootstrap re-sampling with 5000 replicates , was performed in order to generate 95% confidence interval ( Carpenter and Bithell , 2000 ) ( shadowed ribbons around the mean in the figures ) . All experiments were replicated two to five times . In all figures , Ns represent the total number of flies over all experiments . Statistics were done on aggregated data . Outliers were never excluded . Flies that died during the course of the experiment were excluded from all analysis . All figures were generated in R , using ggplot2 ( Wickam , 2009 ) . For all boxplots , the bottom and top of the box ( hinges ) show the first and third quartiles , respectively . The horizontal line inside the box is the second quartile ( median ) . Tuckey's rule ( the default ) , was used to draw the ‘whiskers’ ( vertical lines ) : the whiskers extend to last extreme values within ±1 . 5 IQR , from the hinges , where IQR is Q3-Q1 . A detail summary of all statistical comparisons is provided as Supplementary file 1 . Supplementary Videos S1 , S2 , S3 , S4 ( available at https://lab . gilest . ro/projects/raw-data/regulation-of-sleep-homeostasis-by-sex-pheromones-supplementary-videos/ given the interactive nature of this figure ) . ( S1 ) Interaction between wild-type D . melanogaster male and female flies ( MF ) . ( S2 ) Interaction between wild-type D . melanogaster male and white eyed males ( MM ) . ( S3 ) Interaction between ppk23Δ mutant D . melanogaster male and wild-type D . melanogaster female flies . ( S4 ) Interaction between wild-type males and D . simulans female . S1 and S2 show the same dataset as in Figure 1D . S3 has the same dataset as in Figure 3D . S4 shows the same dataset as in Figure 2—figure supplement 2 . In all videos , hover the mouse cursor on the ethogram to highlight the corresponding region of interest in the video . Click on the ethogram or on the day/night bar to seek to the relative video position . Legend for behavioural classification as in Figure 1C . Videos are compressed to facilitate access . Full , uncompressed dataset accessible on Zenodo via doi: 10 . 5281/zenodo . 167551 or https://zenodo . org/record/167551# . Wbo5OkqGMsk ( Gilestro et al . , 2016 ) . When clicking on ethograms , allow an error on time axis of ±10 min on the relative video . | Humans spend one-third of their lifetime sleeping , but why we ( and other animals ) need to sleep remains an unresolved mystery of biology . Our desire to sleep changes depending on how much sleep we’ve already had . If we’ve had a long nap during the day , we may find it harder to fall asleep at night; conversely , if we stay up all night partying , we’ll have a difficult time staying alert the next day . This change in the pressure to sleep is known as “sleep homeostasis” . Can sleep homeostasis be suppressed ? We know that some migratory birds are able to resist sleep while flying over the ocean . In addition , males of an Arctic bird species forgo sleep for courtship during the three-week window every year when females of its species are fertile . These examples suggest that some behavioral or environmental factors may influence sleep homeostasis . Beckwith et al . now show that sexual arousal can disrupt sleep homeostasis in fruit flies . In “blind date” experiments , young male fruit flies were kept in a small tube with female fruit flies , prompting a 24-hour period of courtship and mating . The males went without sleep during that period , and they did not make up for the lost sleep afterward . In other experiments , male fruit flies were kept awake by a robot that disturbed them every time they tried to sleep . After such treatment , the flies normally attempted to nap . But if the sleep-deprived flies were exposed to a chemical emitted by female flies that increased their sexual arousal , they no longer needed to sleep . Overall , the results presented by Beckwith et al . show that sleep is a biological drive that can be overcome under certain conditions . This will be important for sleep researchers to remember , because it means that it’s possible to affect sleep regulation ( perhaps by making the animal stressed or aroused ) without activating the brain circuits directly involved in regulating sleep . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"neuroscience"
] | 2017 | Regulation of sleep homeostasis by sexual arousal |
In diverse bacterial species , the global regulator Hfq contributes to post-transcriptional networks that control expression of numerous genes . Hfq of the opportunistic pathogen Pseudomonas aeruginosa inhibits translation of target transcripts by forming a regulatory complex with the catabolite repression protein Crc . This repressive complex acts as part of an intricate mechanism of preferred nutrient utilisation . We describe high-resolution cryo-EM structures of the assembly of Hfq and Crc bound to the translation initiation site of a target mRNA . The core of the assembly is formed through interactions of two cognate RNAs , two Hfq hexamers and a Crc pair . Additional Crc protomers are recruited to the core to generate higher-order assemblies with demonstrated regulatory activity in vivo . This study reveals how Hfq cooperates with a partner protein to regulate translation , and provides a structural basis for an RNA code that guides global regulators to interact cooperatively and regulate different RNA targets .
The RNA chaperone Hfq contributes to the control of mRNA translation through different modes of action . In one mode , Hfq acts indirectly by facilitating base-pairing interactions of cognate mRNA targets with small regulatory RNAs ( Vogel and Luisi , 2011; Wagner and Romby , 2015 ) . In a second mode , the RNA chaperone can directly repress translation independently of small RNAs , by binding A-rich sequences at or in the vicinity of translation initiation sites ( Vecerek et al . , 2005; Salvail et al . , 2013; Sonnleitner and Bläsi , 2014 ) . As a member of the extensive Lsm/Sm protein family , Hfq shares an ancient structural core that oligomerizes to form toroidal architectures exposing several RNA-binding surfaces . Crystallographic and biophysical data showed that RNA recognition is mediated by distinct interactions with distal , proximal and rim faces of the hexameric ring ( Schumacher et al . , 2002; Link et al . , 2009; Sauer et al . , 2012; Horstmann et al . , 2012; Panja et al . , 2013 ) , as well as revealing the role of the unstructured C-terminal tail in auto-regulating RNA-binding activities ( Santiago-Frangos et al . , 2016; Santiago-Frangos et al . , 2017 ) . In the opportunistic , Gram-negative pathogen Pseudomonas aeruginosa , Hfq acts as a pleiotropic regulator of metabolism ( Sonnleitner and Bläsi , 2014 ) , virulence ( Sonnleitner et al . , 2003; Fernández et al . , 2016; Pusic et al . , 2016 ) , quorum sensing ( Sonnleitner et al . , 2006; Yang et al . , 2015 ) and stress responses ( Lu et al . , 2016 ) . Many of these roles are likely facilitated through partner molecules , and numerous putative protein interactors of P . aeruginosa Hfq have been identified with functions in transcription , translation and mRNA decay ( Van den Bossche et al . , 2014 ) . In the case of Hfq from E . coli , the functionally important partners include RNA polymerase , ribosomal protein S1 ( Sukhodolets and Garges , 2003 ) , the endoribonuclease RNase E ( Ikeda et al . , 2011 ) , polyA-polymerase , and the exoribonuclease polynucleotide phosphorylase ( Mohanty et al . , 2004; Bandyra et al . , 2016 ) . Most likely , these complexes are RNA mediated and affect the co-localisation of the machineries of transcription , translation and RNA decay ( Worrall et al . , 2008; Resch et al . , 2010; Vecerek et al . , 2010 ) . One P . aeruginosa protein that was found to co-purify with tagged Hfq is the Catabolite repression control protein , Crc ( Van den Bossche et al . , 2014; Moreno et al . , 2015; Sonnleitner et al . , 2018 ) . Crc is involved in carbon catabolite repression ( CCR ) in Pseudomonas , a process that channels metabolism to use preferred carbon sources ( such as succinate ) until they are exhausted , whereupon alternative sources are used ( Rojo , 2010 ) . In addition to carbon catabolite repression , Hfq and Crc link key metabolic and virulence processes in Pseudomonas species . The two proteins affect biofilm formation , motility ( O'Toole et al . , 2000 , Huang et al . , 2012; Zhang et al . , 2012; Pusic et al . , 2016 ) , biosynthesis of the virulence factor pyocyanin ( Sonnleitner et al . , 2003; Huang et al . , 2012 ) , and antibiotic susceptibility ( Linares et al . , 2010; Heitzinger , 2016 ) . Recent ChiP-seq studies indicate that Hfq and Crc have an even broader regulatory impact in Pseudomonas and can work in concert to bind many nascent transcripts co-translationally , uncovering a large number of potential regulatory targets ( Kambara et al . , 2018 ) . Crc is a specialized protein that is found in a limited subset of bacteria including Pseudomonads but excluding Enterobacteriaceae such as E . coli . The Crc protein is a member of the EEP family ( Exonuclease/Endonuclease/Phosphatase ) , but the absence of key catalytic residues in the active site cleft indicate that Crc has lost enzymatic function ( Milojevic et al . , 2013 ) . Crc and Hfq act together to translationally repress mRNA target genes . The targets are then subjected to degradation , which might trigger disassembly of the Hfq/Crc/RNA complex ( Sonnleitner and Bläsi , 2014 ) . The impact of Hfq/Crc on translational repression is countered by the non-coding RNA CrcZ , which increases in levels when preferred carbon sources are exhausted ( Sonnleitner et al . , 2009; Sonnleitner and Bläsi , 2014 ) . CrcZ expression is under control of the alternative sigma factor RpoN and the two-component system CbrA/B ( Sonnleitner et al . , 2009 ) . Although the signal responsible for CbrA/B activation remains unknown , it is thought to be related to the cellular energy status ( Valentini et al . , 2014 ) . The mechanism by which Hfq and Crc act in translational repression of target mRNAs involves binding of both proteins to ribosome-binding sequences . In the case of the translationally repressed amiE mRNA , encoding aliphatic amidase , the distal face of Hfq binds to an A-rich segment termed amiE6ARN that comprises the ribosome binding site ( Sonnleitner and Bläsi , 2014 ) . In contrast , Crc has no intrinsic RNA-binding activity ( Milojevic et al . , 2013 ) but strengthens binding of A-rich target transcripts to the distal side of Hfq ( Sonnleitner et al . , 2018 ) . How Hfq binds to A-rich RNAs is structurally understood ( Link et al . , 2009 ) , but the cooperative role of Crc in this context has not been elucidated . To gain insight into how P . aeruginosa Hfq cooperates with Crc in translational repression of amiE , we determined the structure of the complex they form on the Hfq binding motif of the CCR-controlled amiE mRNA using cryo-electron microscopy ( cryoEM ) . Our analyses revealed that the components form higher order assemblies and explain for the first time how a widely occurring structural motif can support the association of Hfq and RNA into cooperative ribonucleoprotein complexes that have regulatory roles . We observe that the interactions supporting the quaternary structure are required for in vivo translational regulation . These findings expand the paradigm for in vivo action of Hfq through cooperation with the Crc helper protein and RNA to form effector assemblies .
For cryo-EM structural studies of the Hfq/Crc/RNA complex , purified recombinant Hfq and Crc proteins were mixed with an 18 nucleotide Hfq binding motif from the translation initiation region of the CCR-controlled amiE mRNA ( hereafter termed amiE6ARN ) . The defined sequence ( 5'-AAA-AAU-AAC-AAC-AAG-AAG-3' ) has a binding motif comprised of 6 repeats of an A-R-N pattern preferred by the distal face of Hfq ( Sonnleitner and Bläsi , 2014 ) . The 18 nucleotides of amiE6ARN could be distinguished in the cryoEM maps and will be referred to with numbers 1–18 from the 5´ to the 3´ end . The purified sample of Hfq/Crc/amiE6ARN , after mild chemical crosslinking , yielded well defined single particles on graphene oxide in thin , vitreous ice . Analysis of the reference free 2D class averages and subsequent 3D classification indicated three principal types of complexes corresponding to different stoichiometries of Hfq ( hexamer ) :Crc:amiE6ARN with compositions 2:2:2 , 2:3:2 and 2:4:2 ( Figure 1 ) . These higher order assemblies are in agreement with recently observed SEC-MALS and mass spectrometry results which excluded a simple 1:1:1 assembly ( Sonnleitner et al . , 2018 ) . The maps for the complexes are estimated to be 3 . 1 Å , 3 . 4 Å and 3 . 2 Å in resolution , respectively , based on gold-standard Fourier shell correlations ( Figure 1—figure supplement 1 ) . The distribution of the complexes corresponds to roughly 49% , 29% and 23% for the 2:2:2 , 2:3:2 and 2:4:2 complexes ( Figure 1 ) . The individual crystal structures of Hfq and Crc dock well into the cryoEM densities ( Figure 1 and Figure 1—figure supplement 2 ) , and aside from side chain rotations there are few other structural changes of the components ( Figure 1—figure supplement 2 ) . To probe for artefacts introduced by the crosslinking experiments , CryoEM analyses were subsequently performed with samples that had not been treated by crosslinking using grid preparation conditions optimised for the crosslinked specimens . Although limited to lower resolution , data from these specimens revealed that the quaternary structure remained unchanged within the resolution limits of the data for the model being compared ( Supplementary file 1A ) . Link et al . ( 2009 ) described the crystal structure of E . coli Hfq bound to a polyriboadenylate 18-mer and observed that the RNA encircled the distal face of the Hfq hexamer via a repetitive tripartite binding scheme ( Figure 2—figure supplement 1A ) . Each base triplet is partially embedded between adjacent Hfq monomers and is mostly surface exposed , folding into a’ crown-like’ conformation . We observe striking similarities with the fold of the authentic amiE6ARN species on the distal side of the P . aeruginosa Hfq hexamer ( Figure 2 and Figure 2—figure supplement 1A ) . Notably , the cryoEM maps were calculated without any reference to Link et al . ( 2009 ) . The agreement between the co-crystal structure of the analogous complex and the entirely independently derived cryoEM based model provides thus a strong validation . Like its E . coli homologue , Pseudomonas Hfq contains six tripartite binding pockets on the distal side , capable of binding a total of 18 nucleotides . Each of the six RNA triplets of the amiE6ARN RNA fits into an inter-subunit cleft in Hfq ( Figure 2 ) . The specific , star-shaped RNA fold is guided by six positively charged protuberances on the distal face of Hfq , with the phosphate backbone circularly weaving in between these , seemingly to minimise steric hindrance while maximizing surface interactions ( Figure 2 ) . As described by Link et al . ( 2009 ) , each pocket consists of an adenosine specificity site ( A ) , a purine nucleotide specificity site ( R ) , and a presumed RNA entrance/exit site ( N ) which is non-discriminatory ( Figure 2—figure supplement 1A ) . Hfq thus has a structural preference for ( ARN ) n RNA stretches on its distal side , where N is any nucleotide . The adenosine specificity ( A ) sites are organised identically to the corresponding A sites in E . coli Hfq , forming hydrogen bonds between the peptide backbone and carboxyl-groups of Gln33 and the N6 , 7 atoms of the adenosine base , and a polar interaction between Gln52 ( Nε ) and the N1 atom of the adenosine base ( Figure 2 , Figure 2—figure supplement 1A ) . The peptide backbone amide of residue Lys31 interacts with the 5’ phosphate group of adenine . Finally , the adenine base is stacked against the side chain of Leu32 ( Figure 2 ) . The interactions with N1 , by the carbonyl of Q33 , and N7 , by the amide of Q33 , confer pocket-specificity for A as they are not compatible with a G , which would form a repulsive contact via its O6 and peptide carbonyl ( Figure 2 ) . The purine ( R ) specificity site is defined by two neighbouring monomers , where the side chains from Tyr25 and from Leu26’ , Ile30’ and Leu32’ ( where the prime denotes residues from a neighbouring subunit ) contact the nucleotide aromatic base . In amiE6ARN , one R-site is populated by a guanine , forming a hydrogen bond between the Nε of Gln52’ and the guanine exocyclic O6 ( Figure 2 ) . Just like in the E . coli Hfq/polyA18 structure ( Link et al . , 2009 ) , Gln52’ forms a physical link between the A and R sites . Previous structures were obtained from polyA RNA , whereas the structures presented here were solved with the authentic amiE Hfq recognition site . Interestingly , Thr61 Oγ forms a double hydrogen bond with the N1 and the exocyclic N2 from the guanine base , which was not seen previously ( Link et al . , 2009 ) as all R-sites were occupied by adenine residues ( Figure 2 ) . As we will describe further below , the N site base interacts with Crc , but without apparent sequence preference . Strikingly , the A-R-N motif is not supported by Hfq of the Gram-positive bacteria Staphylococcus aureus and Bacillus subtilis , which instead use a R-L ( purine , linker ) motif ( Horstmann et al . , 2012 ) ( Figure 2—figure supplement 1B ) . Phe 30 hinders the RNA backbone , and thus the A-site base , from entering the A-site , while stabilising the R-site together with M32 via stacking interactions . In the core complex ( with 2:2:2 stoichiometric composition of Hfq:Crc:RNA ) , two Hfq hexamers sandwich the RNA and Crc ( Figure 3A ) , with each Hfq interacting with one amiE6ARN RNA and two Crc components . The assembly has C2 symmetry , with the molecular twofold axis passing through the interface of the two Crc molecules . The Crc´s in the assembly self-interact in the same way as observed in the crystal structure of the isolated Crc dimer , where the interface is generated through crystallographic symmetry ( Wei et al . , 2013; Milojevic et al . , 2013 ) . As anticipated from mutational analyses ( Sonnleitner et al . , 2018 ) , the dominating protein/RNA interaction is made by the distal face of Hfq , forming an interface area of roughly 2270 Å2 . The two Crc molecules interact with RNA residues exposed on the surface of Hfq , and both Crc molecules contact the Hfq-rim on the distal side ( Figure 3A ) . The Crc protomers form contacts mainly with the backbone phosphate groups and exposed ribose rings of the RNA ( Figure 3A and C ) . Because the Crc forms antiparallel dimers , there are two modes of interaction of the Crc with amiE6ARN RNA . In one mode of interaction , Arg140 η1-NH2 and Arg141 ε-NH and η1-NH2 interact with the phosphodiester backbone of amiE6ARN . Arg140 and Arg196 form a sandwich with the purine-base of the A3 nucleotide at an Entry/Exit ‘N’ site of amiE6ARN RNA . Interestingly , Arg140 participates in the Crc1-Crc2 dimer interface in the Crc-dimer as solved by crystallography via contacts with E193 ( pdb-ID 4jg3 , data not shown ) , but our maps clearly show a rotamer shift of the R140 side chain away from E193 , forming contacts with amiE as described above . The Arg196 ε-NH and η1-NH2 groups form hydrogen bonds with the U6-amiE6ARN backbone and the U6 O2 group forms a hydrogen bond with the Met156 amide . In the second mode of interaction , Lys155 ζ-NH2 makes a hydrogen bond with the OP2-group of C9 and the ribose hydroxyl group . Additional hydrogen bonds are formed between Trp161 ε1-NH and Arg162 η1/2-NH2 and the phosphate backbone of amiE6ARN ( Figure 3A ) . The highly organised interactions in the core complex ( Figure 3A ) illustrate how the bases of amiE6ARN as presented by Hfq constitute a molecular interface for the RNA-mediated interactions between Hfq and Crc . Crc forms small contact surfaces to the RNA , to Hfq , and to itself as a homodimer; these small areas work together to give an assembly that is most likely stabilised through chelate cooperativity . Notably , there is a striking absence of any lower order assemblies in the cryo EM micrographs . The 2:2:2 complex is therefore likely to be the minimal complex formed when all components are present and must be constructed in an ‘all or nothing’ manner , somewhat like a binary switch . The dimer interface of the Crc pair is the largest protein-protein interface in the 2:2:2-complex and has a buried area of 766 Å2 , which typically corresponds to a moderate intermolecular affinity . The key dimerization interface is maintained by salt bridges between Arg229-Arg230 of one Crc monomer and Glu142 of the second Crc monomer , which is further stabilised by pi-stacking of the Phe231-Phe231 rings across the twofold axis ( Figure 3A ) . The phenylalanine residues are in turn stabilised by stacking interactions with Trp255 of the same Crc monomer ( not shown ) . Two additional polar contacts are formed between Arg137 and the Asn184 carbonyl group of two pairs of helices in the Crc dimer , forming a smaller secondary interface ( Figure 3A ) . To verify selected interactions between the Crc protomers and Crc and RNA , we explored the effects of Crc variants on Hfq/Crc-mediated repression of an amiE::lacZ translational reporter gene encoded by plasmid pME9655 ( Sonnleitner et al . , 2018 ) . In addition , an in vitro co-immunoprecipitation assay was employed to assess the capacity of Hfq and Crc variants to form a complex in the presence of amiE6ARN RNA ( Sonnleitner et al . , 2018 ) . First , we asked whether R140 ( CrcR140 ) is required for the interaction of the protein with the RNA ( Figure 3A , bottom left inset ) . As shown in Figure 3B , the CrcR140E mutant was deficient in translational repression of the amiE::lacZ reporter gene , similarly as observed for the crc deletion strain . Moreover , CrcR140E did not co-immunoprecipitate with Hfq in the presence of amiE6ARN RNA ( Figure 3—figure supplement 1A ) , indicating that the interaction between CrcR140 and RNA is pivotal for Hfq/Crc/RNA complex formation . Next , we focused on the interactions in the E142 and R229/R230 ‘triangle’ ( Figure 3A , top left inset ) for the Crc-Crc interaction . The single mutant proteins CrcR229E and CrcR230E did not affect translational repression of amiE::lacZ , whereas the function of the CrcE142R variant was diminished ( Figure 3B ) , indicating that E142 can interact with either R229 or R230 . The de-repression of amiE:lacZ observed with the CrcE142R variant was partially compensated by the double mutant proteins CrcE142R , R229E and CrcE142R , R230E ( Figure 3B ) . In addition , the CrcE142R and CrcR230E variants were impaired in Hfq/Crc/RNA complex formation as shown by the co-immunoprecipitation assay ( Figure 3—figure supplement 1A ) . Strikingly , the compensatory changes present in the triple mutant protein CrcE142R , R229E , R230E almost fully restored translational repression of the amiE::lacZ reporter gene ( Figure 3B ) . As the respective Crc variant proteins were produced at comparable levels ( Figure 3—figure supplement 1B ) , these mutational studies support the in vivo role for the interactions of the Crc protomers observed in the cryo-EM models . The exchange of Glu142 with an Ala in the triple mutant protein CrcE142A , R229E , R230E restored as well translational repression of the amiE::lacZ reporter gene ( Figure 3B ) . We interpret this as showing that the rescue of the deleterious mutations does not require charge compensation , but removing any potential charge repulsion . The quaternary organisation of the 2:2:2 complex forms a core unit that is also present in the 2:3:2 and 2:4:2 complexes . In that common core , the interaction of Crc with RNA leaves approximately half of the accessible surface of the nucleic acid exposed . For the 2:3:2 and 2:4:2 complexes , additional Crc units are recruited through interactions with the exposed portion of the RNA . As such , the C2 symmetry is broken by the third Crc molecule in the 2:3:2 complex ( Figure 1 ) . Interestingly , recruitment of a fourth Crc monomer to the complex restores the C2 symmetry , preserving the symmetry axis from the core complex , but with a conformationally different Crc dimer interface between Crc molecules 3 and 4 ( Figure 4A ) . The two additional Crc monomers have small surface-area contacts with the rest of the complex and are likely to be comparatively mobile , which may account for the stronger variation in resolution for the 2:3:2 and 2:4:2 maps compared to the rather rigid 2:2:2 core assembly ( Figure 1—figure supplement 1 ) . The protomer interactions of the 2:2:2 assembly are highly interdependent , and once the core complex is generated it can apparently recruit additional Crc molecules , forming the 2:3:2 and 2:4:2 complexes . In the 2:4:2 complex , a second type of Crc dimer seems to assemble with a smaller buried surface ( Figure 4A ) . Such additional dimers can only form when an intact 2:2:2 core complex is present , as they are not observed in the core complex itself . Notably , the additional Crc dimer is in a more ‘open’ conformation than the dimer in the core ( Figure 4C ) . The same key Crc dimer interface is occupied but seems to serve as a dynamic hinge , whereas the secondary , smaller , dimer interface between the Crc helices is absent to allow the new Crc dimer to adopt an ‘open’ conformation . Arg230 is reorganised by Glu193 in the same protomer to self-interact with the corresponding Arg230 in the partner Crc , rather than with Glu142 ( Video 1 ) . Additional hydrogen bonds are formed between Arg233 and Glu193 , whereas Arg229 is no longer part of the dimer interface ( Figure 4A ) . Both Arg230 and Glu193 seem to play important roles in providing the structural freedom to form a dynamic hinge ( Figure 4C ) . Glu193 is part of the Crc1-Crc2 interface in the crystal structure but does not participate in this interface when assembled in the 2:2:2 core complex . Instead it aids in coordinating the Crc3-Crc4 interface in the 2:4:2 assembly . Interestingly , Crc3 and Crc4 do not interact with Crc1 or Crc2 , and seem to be recruited solely via Hfq/amiE6ARN . By reorganising the extra Crc molecules 3 and 4 that bind the 2:2:2 core ( Figure 4A ) , the alternative Crc dimer is able to utilise one of two basic patches on its surface when engaging amiE6ARN without causing steric hindrance to the already bound Crc dimer . The Arg233-Glu193 interaction is unique for the 2:4:2 assembly and was therefore assessed in vivo . Strikingly , the CrcE193R mutation fully abrogated translational repression of the amiE:lacZ reporter gene ( Figure 4B ) . The model predicts that the deleterious CrcE193R mutation can be compensated by the substitution of CrcR230E to re-establish the interaction . This pair does indeed behave as predicted , as the CrcE193R , R230E variant restored translational repression of the amiE::lacZ reporter gene , further confirming the in vivo importance of the 2:4:2 assembly during CCR ( Figure 4B ) . The exchange of Glu193 with Ala in the CrcE193A , R230E mutant protein also restored translational repression of the reporter gene , albeit to a somewhat reduced extent when compared with the CrcE193R , R230E variant ( Figure 4B ) . Again , these results suggest that the rescue of the deleterious mutations does not require charge compensation , but act by removing any potential charge repulsion . In addition to Crc Arg140 and Arg141 , Crc K139 ζ-NH2 makes a hydrogen bond with the OP2-group of A12 , Arg138 η1-NH2 interacts with the ribose hydroxyl group of C9 and K135 ζ-NH2 forms a hydrogen bond with the A11 OP2 . Finally , the O2 of cytosine C12 engages in a hydrogen bond with the backbone amino group of Arg140 . Direct interactions between the reorganised Crc dimer and Hfq are limited to the same Crc β-strand and exposed loop of a sole Hfq monomer , as in the core complex . Due to the open conformation of the alternative Crc dimer , the Hfq Thr49 hydroxyl group now forms a hydrogen bond with the Ala 78 amide group ( Figure 4A ) . Interestingly , a basic half-channel is formed over the core dimer interface , with additional basic patches spread over the RNA binding surface of the Crc dimer ( Figure 4D ) . Speculatively , longer RNA species could travel though the surface exposed half-channel and interconnect all components of the core complex into a highly organised assembly on this target RNA ( Figure 5 ) .
Many functional studies have highlighted the cooperation of global posttranscriptional regulators in controlling the fate of targeted transcripts . A major gap in our current understanding has been the lack of high-resolution structural data of these highly coordinated cellular processes . Here , we report for the first time structural information on how a partner protein can assist the role of Hfq in direct translational repression when bound to a translational initiation region ( amiE6ARN ) by forming a multi-component assembly . The added value of the Crc in the complex as compared to recognition of the A-rich sequence by Hfq is the potential for highly cooperative switches for activation/repression and greater specificity . The amiE RNA translation control site is an A-rich fragment that occupies almost entirely the distal surface of the P . aeruginosa Hfq , weaving in between basic , surface exposed islands ( Figure 2 ) . There are striking similarities to the structure of the polyA18 complex with E . coli Hfq ( Link et al . , 2009 ) , which greatly added to the understanding of RNA binding and chaperone mechanisms , and hinted at how the distinct polyA RNA interaction might enable Hfq-mediated regulation . The polyA18/Hfq structure revealed rules for recognition of motifs of the type A-R-N , where R is purine and N is any base ( Figure 2—figure supplement 1 ) . The P . aeruginosa Hfq interaction with amiE6ARN follows the same rules . The A-R-N repeat occurs in many RNAs , and it is a recurring motif in the nascent transcripts that associate with Hfq and Crc in Pseudomonas ( Kambara et al . , 2018 ) . It has been proposed that the exposed bases ( at the Entry/Exit site , corresponding to ‘N’ in the A-R-N motif ) could mediate RNA to RNA interactions ( Schulz et al . , 2017 ) , but we observe that the exposed bases are presented for protein recognition . The exposed bases at the ‘N’ position and RNA backbone in the Hfq/amiE6ARN complex are available for interactions with Crc to form a cooperative assembly ( Figures 3 and 4 ) that mediates translational repression on target mRNAs , that is carbon catabolite repression in vivo when the preferred carbon source is available . Strikingly , the A-R-N motif is not supported by the Hfq of the Gram-positive bacteria Staphylococcus aureus and Bacillus subtilis ( Horstmann et al . , 2012 ) ( Figure 2—figure supplement 1B ) . In view of this divergence , it seems unlikely that Hfq in the Gram-positive species form assemblies that resemble the P . aeruginosa Hfq/Crc assembly . Previous studies have shown that both Hfq and Crc are required for tight translational repression of mRNAs , which are subject to carbon catabolite repression ( CCR ) ( Sonnleitner and Bläsi , 2014; Moreno et al . , 2015 ) . The presence of Crc did not significantly enhance the affinity of Hfq for amiE6ARN RNA ( Sonnleitner et al . , 2018 ) . However , the simultaneous interactions of Crc with both binding partners resulted in an Hfq/Crc/RNA assembly with increased stability when compared with the Hfq/RNA complex alone ( Sonnleitner et al . , 2018 ) . In light of our structural studies , the enhancing effect of Crc in Hfq-mediated translational repression of target mRNAs during CCR ( Sonnleitner and Bläsi , 2014; Moreno et al . , 2015 ) can be explained by sandwiching the Hfq binding site on mRNA between both binding partners . Thus , the structural model can rationalize the observed increased stability of the Hfq/Crc/amiE6ARN assembly when compared to the sole Hfq/amiE6ARN complex ( Sonnleitner et al . , 2018 ) . It is conceivable that full repression is only achieved when amiE6ARN is masked entirely in the 2:4:2 complex , which is supported by our in vivo studies . The question arises why a higher order assembly such as the 2:2:2 core is formed and not a simpler complex . The structural data indicate that the dimerization of Crc provides the key step for formation of the 2:2:2 complex , because it will pre-organise a copy of the surface that interacts with a binary Hfq/RNA complex so that a second Hfq/RNA complex can be recruited . Thus , all components seem to be necessary to form the complex so that there is no formation of lower order ‘sub assemblies’ . The structural data are consistent with Crc having no capacity for RNA binding by itself ( Milojevic et al . , 2013 ) . The Hfq/Crc/RNA complex may thus be assembled in a checklist-like manner through numerous small contacting surfaces and when the RNA target is presented by Hfq in a specific , well-defined configuration . In this way , the components interact mutually through chelate cooperative effects . It seem likely that the 2:2:2 core is formed first followed by recruitment of the other Crc components . We envisage that the 2:2:2 core and higher order assemblies might interact with other mRNAs . The higher order assembly could capture two of such mRNA substrates as shown in Figure 5i , but chelate effects might instead induce formation of the complex on a single mRNA target . In that scenario , a portion of the mRNA could thread through the central basic half channel ( Figure 4D ) as depicted in Figure 5ii and potentially recruit the second Hfq hexamer . In this context it is worth noting that several mRNAs , including amiE , which are directly repressed by Hfq/Crc comprise another putative Hfq binding motif downstream of the start codon ( Sonnleitner et al . , 2018 ) . Therefore , we are currently exploring the possibility whether this second Hfq binding motif contributes to Hfq/Crc assembly on longer mRNA substrates and whether it likewise may confer specificity for these Hfq/Crc repressed substrates . Under conditions of catabolite repression regulation , pull-down assays showed that Hfq and Crc form a co-complex in the presence of the 426nt long CrcZ RNA ( Moreno et al . , 2015; Sonnleitner et al . , 2018 ) . In the presence of less preferred carbon sources , the expression levels of CrcZ RNA increase ( Sonnleitner et al . , 2009 ) and CrcZ functions as an antagonist in Hfq/Crc mediated translational repression of catabolic genes . The CrcZ RNA has multiple A-R-N triplets that could be sites for Hfq/Crc interaction ( Sonnleitner and Bläsi , 2014 ) and sequester multiple Hfq/Crc proteins ( Figure 5 ) . Thus , under conditions where CCR is relieved , CrcZ RNA would serve as a sponge for Hfq/Crc to prevent repression of genes encoding proteins required for the utilization of less preferred carbon sources ( Figure 5 ) . How the CrcZ RNA is displaced from Hfq/Crc remains unknown . However , the assemblies are likely to be dynamic and the displacement process might resemble that proposed for the step-wise exchange of sRNAs on Hfq ( Fender et al . , 2010 ) . Recent findings show that the regulatory spectrum of Hfq and Crc is much broader than initially expected . Hfq was found to bind more than 600 nascent transcripts co-transcriptionally often in concert with Crc ( Kambara et al . , 2018 ) . These findings indicate that Hfq and Crc together regulate gene expression post-transcriptionally beyond just catabolite repression . Understanding how gene expression is regulated post-transcriptionally in pathogens such as P . aeruginosa may provide potential targets for novel drug design . Hfq and Crc are involved in key metabolic and virulence processes in Pseudomonas species ( O'Toole et al . , 2000; Sonnleitner et al . , 2003; Sonnleitner et al . , 2006; Linares et al . , 2010; Huang et al . , 2012; Zhang et al . , 2012; Zhang et al . , 2013; Sonnleitner and Bläsi , 2014; Pusic et al . , 2016 ) . Disrupting the interface of the core assembly of the Hfq/Crc complex might be one strategy to counter , among other , metabolic regulation and consequently its downstream processes that impact on virulence during infection . A recent study showed how overproduction of the aliphatic amidase AmiE strongly reduced biofilm formation and almost fully attenuated virulence in , amongst others , a mouse model of acute lung infection ( Clamens et al . , 2017 ) . Novel drugs that specifically counteract Hfq/Crc/amiE assembly formation and prevent repression of AmiE production could induce the phenotype described by Clamens et al . ( 2017 ) . The high resolution structures presented here provide a starting point for novel strategies to interfere with carbon regulation in a pathogenic bacterium for therapeutic intervention of threatening infections .
P . aeruginosa Hfq and Crc were produced in E . coli and purified as described by Sonnleitner et al . ( 2018 ) . The synthetic 18-mer amiE6ARN RNA ( 5´-AAAAAUAACAACAAGAGG-3´ ) used in these studies consists of six tripartite binding motifs ( Sonnleitner and Bläsi , 2014 ) . The Hfq/Crc/RNA complex was prepared by first heating the amiE6ARN RNA at 95°C for 5 min followed by 50°C for 10 min and 37°C for 10 min . The RNA was then incubated with the Hfq hexamer at a 1:1 molar ratio on ice for 20 min to form a binary complex , then an equal molar ratio of Crc was added as recently described ( Sonnleitner et al . , 2018 ) . The mixture was incubated on ice for 30 min prior to fractionation by size exclusion chromatography using a Superdex 200 column equilibrated in running buffer composed of 20 mM HEPES , pH 7 . 9 , 10 mM KCl , 40 mM NaCl , 1 mM MgCl2 , and 2 mM TCEP ( tris ( 2-carboxyethyl ) phosphine ) . The peak fractions were buffer exchanged into 20 mM HEPES , pH 7 . 9 , 10 mM KCl , 40 mM NaCl , 5 mM MgCl2 . Samples used for cross-linking were incubated with bis ( sulfosuccinimidyl ) suberate ( BS3 ) at 150 µM for 30 min on ice , followed by quenching at 37 . 5 mM Tris-HCl pH 8 . 0 . Graphene oxide grids are prepared as described by Pantelic et al . ( 2010 ) . Briefly , 2 mg/ml of graphene oxide solution in water ( Aldrich ) was diluted ten times in water . After removing aggregation by spinning for 30 s at 300 rcf , 2 µl of graphene oxide solution was loaded on freshly glow discharged quantifoil Au-grids ( R1 . 2/1 . 3 , 300 mesh ) . Glow discharge was performed prior to graphene oxide coating at 45 mA for 60 s with an Edward Sputter Coater S150B at 0 . 2 m Bar at 0 . 75 KV . After the graphene oxide had been adsorbed for 1 min , the grids were washed 3 times with 20 μl water , then air-dried for 1 hr at room temperature prior to sample application . Specimens for cryoEM analysis were prepared by applying 2 μl of a 0 . 65 µM solution of the Hfq/Crc/RNA complex to the Quantifoil Au grids freshly coated with graphene oxide . After an adsorption time of 60 s , the grids were blotted for 10 s at a blot force of 5 , then plunge frozen into liquid ethane using a Vitrobot ( FEI ) . Images were recorded on a Krios G2 , Falcon III direct electron detector at 300 kV operating in counting mode ( Supplementary file 1C ) . Whole frame motion correction was performed on movies with motioncorr2 with dose weighting followed by CTF estimation using gctf ( Zhang , 2016; Zheng et al . , 2017 ) . RELION-3 . 0 was used for data processing ( Scheres , 2012 ) . Final resolution estimates were calculated after the application of a soft binary mask and phase randomisation and determined based on the gold standard FSC = 0 . 143 criterion ( Scheres and Chen , 2012; Chen et al . , 2013 ) . For the BS3 treated complex , after manually picking 3159 particles and using suitable references for autopicking , 482426 particles were used for early classifications . After three rounds of rejecting particles by 2D classification , 215774 particles were used for initial model generation and 3D classification . An initial model was generated using an SGD algorithm based on a small subset of particles with diverse orientations ( Punjani et al . , 2017 ) . During 3D classification , three different complexes were resolved after 25 iterations with an angular sampling of 7 . 5°: 2Hfq:2Crc:2amiE6ARN ( 2:2:2 ) , 2Hfq:3Crc:2amiE6ARN ( 2:3:2 ) and 2Hfq:4Crc:2amiE6ARN ( 2:4:2 ) . To properly separate , validate and refine the three classes , the same 3D classification was rerun with the new 2:3:2 model as reference model , lowpass filtered to 20 Å resolution . C2 symmetry was observed and imposed for the 2:2:2 and 2:4:2 complexes . Each of the classes was then refined to sub-3 . 5 Å resolution , followed by per-particle frame alignment for movement correction and per-frame damage weighting . The resulting ‘polished’ particles were subjected to a final refinement round with solvent flattening . All reference models were lowpass filtered to 60 Å prior to refinement . The dominant class ( 2:2:2 ) had a resolution of 3 . 13 Å . Local resolution calculations were done with the RELION local resolution estimator , Supplementary file 1A ( Figure 1—figure supplement 1 ) . The 2:4:2 and 2:3:2 maps were sharpened locally with LocScale and LocalDeblur to better resolve the additional Crc components ( Jakobi et al . , 2017; Ramírez-aportela et al . , 2018 ) . Crystal structures for P . aeruginosa Crc ( PDB code 1U1S ) and Hfq ( PDB code 4JG3 ) were manually docked into the EM density map as rigid bodies in Chimera ( Pettersen et al . , 2004 ) . The RNA 18-mers were manually built into the density using Coot ( Emsley et al . , 2010 ) . Refmac5 and Phenix real-space refinement with global energy minimization , NCS-restraints , group B-factor and geometry restraints were used to iteratively refine the multi-subunit complexes at high resolution , followed by manual corrections for Ramachandran and geometric outliers in Coot and ISOLDE ( Supplementary file 1A ) ( Emsley et al . , 2010; Murshudov et al . , 2011; Afonine et al . , 2012; Croll , 2018 ) . Model quality was evaluated with Procheck in CCP4 and MolProbity ( Williams et al . , 2018 ) . In silico 2 Å maps were generated from the atomic models and FSC validation against the experimental maps was performed with the EMDB Fourier shell correlation server ( EMBL-EBI ) ( Figure 1—figure supplement 2B ) . The strains , plasmids and oligonucleotides used in this study are listed in Supplementary file 1B and 1C . To test the proficiency of Crc mutant proteins to co-repress translation of a translational amiE:lacZ reporter gene , derivatives of plasmid pME4510crcFlag ( Supplementary file 1B ) were constructed by means of Quick change site directed mutagenesis ( Agilent Technologies ) . Plasmid pME4510crcFlag was used together with the corresponding mutagenic oligonucleotide pairs ( Supplementary file 1C ) . The parental plasmid templates were digested with DpnI and the mutated nicked circular strands were transformed into E . coli XL1-Blue , generating plasmids pME4510crc ( R140E ) Flag , pME4510crc ( E142R ) Flag , pME4510crc ( R229E ) Flag , pME4510crc ( E193R ) Flag , pME4510crc ( R230E ) Flag , pME4510crc ( E142R , R229E ) Flag , pME4510crc ( E193R , R230E ) Flag , pME4510crc ( E193A , R230E ) Flag , pME4510crc ( E142R , R230E ) Flag and pME4510crc ( E142R , R229E , R230E ) Flag and pME4510crc ( E142A , R229E , R230E ) Flag . The ability of the Crc mutant proteins to repress translation of an amiE::lacZ reporter gene was tested in a PAO1 crc deletion strain bearing plasmids encoding the wt protein or the respective Crc variants ( Supplementary file 1B ) as described by Sonnleitner et al . ( 2018 ) . The β-galactosidase activities were determined as described ( Miller , 1972 ) . The β-galactosidase units in the different experiments were derived from two independent experiments . The R140E , E142R , R230E single aa exchanges in Crc were obtained by using the QuickChange site-directed mutagenesis protocol ( Agilent Technologies ) . The plasmid pETM14lic-His6Crc ( Supplementary file 1B ) was used together with the corresponding mutagenic oligonucleotide pairs ( Supplementary file 1C ) . The entire plasmids were amplified with Pfu DNA polymerase ( Thermo Scientific ) . The parental plasmid templates were digested with DpnI and the mutated nicked circular strands were transformed into E . coli XL1-Blue , generating plasmids pETM14lic-His6CrcR140E , pETM14lic-His6CrcE142R and pETM14lic-His6CrcR230E . The Crc protein and the Crc variants CrcR140E , CrcE142R and CrcR230E were purified from E . coli strain BL21 ( DE3 ) harboring either plasmid pETM14lic-His Crc or the respective derivatives using Ni-affinity chromatography , followed by removal of the His6-tag with GST-HRV14-3C ‘‘PreScission’’ protease as described by Milojevic et al . , 2013 . The co-IP studies in the presence of 40 pmol of Hfq-hexamer , 120 pmol of Crc protein or of the respective Crc mutant proteins and 40 pmol amiE6ARN RNA were performed as described ( Sonnleitner et al . , 2018 ) . Equal amounts of proteins were separated on 12% SDS-polyacrylamide gels , and then electro-blotted onto a nitrocellulose membrane . The blots were blocked with 5% dry milk in TBS buffer , and probed with rabbit anti-Hfq ( Pineda ) and rabbit anti-Crc ( Pineda ) antibodies , respectively . Immuno-detection of ribosomal protein S1 served as a loading control . The antibody-antigen complexes were visualized with alkaline-phosphatase conjugated secondary antibodies ( Sigma ) using the chromogenic substrates nitro blue tetrazolium chloride ( NBT ) and 5-Bromo-4-chloro-3-indolyl phosphate ( BCIP ) . | Living things can adapt rapidly to changes in their surroundings by switching whole groups of genes on and off . These responses must be controlled carefully , and they are often coordinated by regulatory proteins working together . Within a cell , the coded information in genes is copied to create molecules called mRNAs , which are then translated to produce proteins . Stopping the cell from reading the information in mRNAs is one way of shutting down specific genes . Pseudomonas aeruginosa is a species of bacteria that can infect humans and can cause cases of sepsis and pneumonia . It changes the activity of its genes in response to its environment . For example , certain genes are only active inside a human host . This allows the bacteria to make the best use of available nutrients and energy . In these cells , two proteins named Hfq and Crc cooperate to silence groups of genes . They do this by stopping the cell from reading specific mRNA molecules , but how they do this is not fully understood . By using a technique called cryo-electron microscopy , Pei , Dendooven et al . studied Hfq and Crc attached to mRNAs . The results show that two groups of six Hfq molecules and two Crcs attach themselves to two mRNA sections to create a structure that stops an mRNA from being translated into a protein . Since the structure only forms with certain mRNAs , the effect is specific to certain genes . The structure needs both Hfq and Crc to work together , which means it only forms in specific situations – when the affected genes are not needed . P . aeruginosa is highly antibiotic resistant and new drugs are urgently needed to control infections . Understanding how this disease-causing bacterium controls its genes could lead to new treatments . The mechanisms of gene regulation are also common to many other forms of life so may also aid the wider understanding of how cells adapt to rapidly changing environments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2019 | Architectural principles for Hfq/Crc-mediated regulation of gene expression |
Vertebrate photoreceptors are among the most metabolically active cells , exhibiting a high rate of ATP consumption . This is coupled with a high anabolic demand , necessitated by the diurnal turnover of a specialized membrane-rich organelle , the outer segment , which is the primary site of phototransduction . How photoreceptors balance their catabolic and anabolic demands is poorly understood . Here , we show that rod photoreceptors in mice rely on glycolysis for their outer segment biogenesis . Genetic perturbations targeting allostery or key regulatory nodes in the glycolytic pathway impacted the size of the outer segments . Fibroblast growth factor signaling was found to regulate glycolysis , with antagonism of this pathway resulting in anabolic deficits . These data demonstrate the cell autonomous role of the glycolytic pathway in outer segment maintenance and provide evidence that aerobic glycolysis is part of a metabolic program that supports the biosynthetic needs of a normal neuronal cell type .
Sensory neurons capture information from the environment and convert it into signals that can greatly impact the survival of an organism . These systems are thus under heavy selective pressure , including for the most efficient use of energy to support their sensitivity and efficiency ( Niven and Laughlin , 2008 ) . In this regard , the primary photoreceptor cells face a dual challenge . They not only need to preserve their membrane excitability via ion pumps by ATP hydrolysis ( Okawa et al . , 2008 ) but also maintain elaborate membranous organelles ( rhabdomeres in invertebrates and outer segments in vertebrates ) that maximize light capture . The maintenance of such structures requires considerable metabolic resources . Invertebrate photoreceptors exhibit light-dependent endocytosis of their photosensitive membranes ( Satoh et al . , 2005 ) enabling the recycling of these resources . In contrast , vertebrate photoreceptors shed a fraction of their outer segments ( OS ) daily , to be phagocytosed by the juxtaposed retinal pigmented epithelium ( RPE ) ( Basinger et al . , 1976; LaVail , 1976 ) ( Figure 1—figure supplement 1 ) . To sustain a constant volume of the OS , a cell must channel metabolites toward biosynthesis , against the backdrop of very high ATP consumption , which is required to maintain membrane potential . Photoreceptors thus must balance the use of their intracellular carbon pool between oxidative catabolism , to generate the required ATP , and anabolism , to continually renew the OS . The mammalian retina depends upon glucose and glycolysis for survival and function ( Chertov et al . , 2011; Noell , 1951 ) . The majority ( ~80% ) of glucose is converted to lactate via glycolysis ( Cohen and Noell , 1960; Warburg , 1925; Winkler , 1981 ) . The adult retina can produce lactate aerobically ( aerobic glycolysis/Warburg effect ) with an ~50% increase during anaerobic conditions ( Pasteur effect ) ( Cohen and Noell , 1960 ) . The cell types that carry out aerobic glycolysis in the normal adult retina have not been determined . The photoreceptors have been assumed to rely on aerobic glycolysis . This assumption is based on the adverse effects on photoreceptor function after en masse inhibition of whole retinal glycolysis by pharmacological treatments e . g . with iodoacetate ( Winkler , 1981 ) . The Warburg effect exemplifies an elaborate set of metabolic strategies adopted by a cell to preferentially promote glycolysis ( Gatenby and Gillies , 2004; Liberti and Locasale , 2016 ) . One drawback of inhibiting glycolytic enzyme activity in the retina is that such a manipulation does not differentiate between aerobic glycolysis and housekeeping glycolysis- a pathway critical for most cell types . Studies conducted on retinal tissue in vitro indicate that isolated mammalian photoreceptors can consume lactate , which can be produced by glycolysis in retinal Mueller glia ( Poitry-Yamate et al . , 1995 ) . Thus , the decreased photoreceptor function after whole retinal glycolytic enzyme inhibition could be a non-cell-autonomous effect on Muller glia . Although many features of the ‘lactate shuttle’ and its in vivo relevance have recently been questioned ( Hurley et al . , 2015 ) , it is important to devise an experimental strategy that would be able to discern the cell-autonomous versus non-autonomous requirement of glycolysis for the photoreceptors . The cellular origins and purpose of aerobic glycolysis in the retina , its relevance to photoreceptor physiology , and its regulation , are not understood . In this study , we explored the propensity of photoreceptors to produce or consume lactate and utilized genetic manipulations to reveal the regulatory mechanisms of glycolysis . We show that rod photoreceptors rely on glycolysis for their OS biogenesis . Genetic perturbations targeting allostery or key regulatory nodes in the glycolytic pathway impacted the OS size . Fibroblast growth factor ( FGF ) signaling was found to regulate glycolysis , with antagonism of this pathway resulting in anabolic deficits . These data demonstrate the cell autonomous role of the glycolytic pathway in OS maintenance and provide evidence that aerobic glycolysis is part of a metabolic program that supports the biosynthetic needs of a normal neuronal cell type .
We first examined lactate production from the retina and assayed the metabolic consequences of inhibiting aerobic glycolysis . Lactate is produced by reduction of pyruvate , a reaction catalyzed by lactate dehydrogenase ( LDH ) ( Figure 1—figure supplement 2A ) . Freshly isolated retinae were cultured in the presence or absence of sodium oxamate- an LDH inhibitor . These were subsequently transferred to buffered Krebs’-Ringer’s medium that has glucose as the sole source of carbon ( see - an LDH inhibitor . These were subsequently transferred to buffered Krebs’-Ringer’s medium that has glucose as the sole source of carbon ( see Materials and methods ) , and lactate secretion was quantified ( Figure 1A ) . The extracellular secreted lactate was measured because it represents the pyruvate-derived carbons that are diverted away from other intracellular metabolic processes or the mitochondria . Oxamate treatment led to a significant drop in the secreted lactate production rate compared to control . In addition , the ATP levels were monitored and , surprisingly , the steady-state levels of ATP in oxamate-treated retinae did not differ from the control retinae ( Figure 1B ) . This could be due to a relatively minor glycolytic contribution to the total ATP pool , a compensatory metabolic realignment toward mitochondria-dependent ATP production or existence of phosphotransfer enzyme systems such as adenylate kinase or creatine kinase . To differentiate among these possibilities , explants were cultured in oxamate or control conditions followed by a short treatment with NaN3 to inhibit mitochondrial ATP synthesis ( Figure 1B ) . Control retinae displayed ~50% reduction in ATP levels after incubation in NaN3 . Interestingly , oxamate-treated retinae displayed a further 20% decrease in ATP after exposure to NaN3 . Thus , inhibiting lactate synthesis resulted in a greater fraction of the ATP pool that was sensitive to mitochondrial function . 10 . 7554/eLife . 25946 . 003Figure 1 . Ldha-dependent aerobic glycolysis and outer segment maintenance in photoreceptors . ( A ) Freshly explanted retinas were treated with the LDH inhibitor , sodium oxamate , for 8 hr in explant culture medium , transferred to Krebs’-Ringer's for 30 min , and lactate was measured in the supernatant . Control ( n = 5 ) , Oxamate ( n = 4 ) . ( B ) Freshly explanted retinas were treated with oxamate or NaCl ( control ) in explant culture medium for 8 hr , followed by treatment with NaN3 or NaCl ( untreated group ) in Krebs’-Ringer's medium for 30 min . ATP per retina was then measured . n = 7 , Control untreated; n = 8 , Oxamate untreated n = 8 , Control NaN3; n = 8 , Oxamate NaN3 . ( C ) Expression of Ldha and Ldhb as determined by IHC . Glutamine synthetase ( GS ) , a Mueller glia-specific marker , colocalized with LDHB in the cell bodies ( arrowheads ) , processes ensheathing the photoreceptors ( arrows ) and the outer limiting membrane ( OLM , * ) . Scale bar , 50 μm . ( D ) ISH for Ldha and Ldhb . Ldha RNA displayed photoreceptor-enriched expression while Ldhb RNA was not observed in photoreceptors . Scale bar , 100 μm . ( E , F ) Freshly explanted retinas were treated with FX11 or DMSO for 8 hr and transferred to Krebs’-Ringer's for 30 min and secreted lactate was measured ( E ) n = 5 , DMSO; n = 6 , FX11 , or they were transferred to Krebs’-Ringer's buffer with NaN3 or NaCl ( untreated group ) for 30 min for ATP quantitation ( F ) . ATP per retina was measured at the end of the assay . n = 8 , DMSO untreated; n = 8 , FX11 untreated; n = 9 , DMSO NaN3; n = 7 , FX11 NaN3 . ( G ) Freshly explanted retinae were transferred to Krebs’-Ringer's for 30 min and secreted lactate was measured . n = 8 , Bl6/J; n = 8 , Ldhafl/fl; n = 8 , Rod-cre; n = 16 , Rod-cre> Ldhafl/fl; n = 8 , Rod-cre> Ldhafl/+ . ( H ) Photoreceptor outer segment phenotype 42–45 days following in vivo electroporation of a knock-down construct ( shRNA ) for Ldha . CAG-mGFP was used for coelectroporation . Plasmid combinations listed on the left . Magnification of areas outlined in yellow is displayed on right with threshold-adjusted rendering to highlight inner and outer segments . Scale bar , 25 μm . ( I ) Quantification of inner+outer segment ( IS+OS ) lengths . n = 53–74 photoreceptors , 4–5 retinae . ( J ) Photoreceptor outer segment phenotype of dark-reared animals . Electroporated pups were transferred to dark on the day of eye opening ( P11 ) and reared with their mothers for 3 weeks . ( K ) Quantification of inner+outer segment lengths of ( J ) . n = 53–83 photoreceptors , 4–5 retinae . ( L ) Colored end products of redox reactions catalyzed by COX and SDH enzymes in retinal tissue . Scale bar , 200 μm . ( M ) IHC for SDH-A subunit in adult retina . Scale bar , 200 μm . ONL , outer nuclear layer . INL , inner nuclear layer . Data , Mean±SD . Statistics , unpaired , two-tailed t-test with Kolmogorov-Smirnov correction for panels A , E; two-way ANOVA with Tukey’s correction for panels B , F and K; one-way ANOVA with Tukey’s multiple comparison test for panels G , I . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 00310 . 7554/eLife . 25946 . 004Figure 1—source data 1 . Source data for Figure 1A , B , E , F and G . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 00410 . 7554/eLife . 25946 . 005Figure 1—figure supplement 1 . Metabolic challenges of photoreceptor cell . Schematic of the rod photoreceptor and RPE-Outer segment proximity shown on the left . photoreceptors are also ensheathed by Mueller glia that span the thickness of the retina . A meshwork of blood capillaries , the choroidal plexus , supplies nutrients and oxygen to the photoreceptors . These cells shed a fraction of their outer segment to be phagocytosed by the RPE . We estimated , based on published findings , that on diurnal basis , the shed discs account for ~70X the lipid present in the cell outside the outer segment ( LaVail , 1976 ) and necessitate ~2X the rate of protein synthesis if shedding does not occur ( Kwok et al . , 2008 ) . Thus , outer segment shedding poses a considerable biosynthetic demand on the photoreceptors . Intense metabolic activity compels judicious allocation of metabolites to competing pathways ( Right ) . Each photoreceptor consumes ~108 ATP s−1 in darkness primarily via the action of Na+/K+ ATPase . Glucose oxidation can generate the ATP , though this would necessitate regulated channeling of glucose to biosynthetic vs catabolic pathways . Similarly , each photon absorption results in formation of all-trans Retinal , which needs to be reduced to complete the visual cycle using NADPH in stoichiometric amounts . NADPH also plays an important role in lipid biosynthesis and countering oxidative stress , which is a byproduct of mitochondria-based oxidative phosphorylation . The central question in understanding photoreceptor physiology thus is , how carbons are allocated toward biosynthetic vs catabolic processes ? ONL , outer nuclear layer . INL , inner nuclear layer . GCL , ganglion cell layer . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 00510 . 7554/eLife . 25946 . 006Figure 1—figure supplement 2 . Characterization of Ldha knockdown and mitochondrial function . ( A ) Lactate dehydrogenase ( LDH ) catalyzes equilibrium between pyruvate and lactate . Of importance is to note the 1:1:1:1 molar stoichiometry between NAD+ , pyruvate , lactate and NADH underscoring the concept that formation of lactate results in molar equivalent contribution to the cytosolic NAD+ pool which in turn serves as a cofactor to generate molar equivalents of pyruvate via the glycolytic pathway . Secreted lactate represents pyruvate-derived carbons that were unavailable to that cell for other metabolic pathways . ( B ) LDH is composed of four subunits with the two most common encoded by the Ldha and Ldhb genes . The five tetrameric compositions are considered to differ in the ability to produce or consume lactate , although the net reaction direction would be dictated by thermodynamics and flux considerations . ( C ) Retinal cross section from the 8 week old F1 progeny of Rod-cre and mT/mG parents . mtdTomato is constitutively expressed while mGFP is expressed in a cre-dependent manner . ( D ) Retina in ( c ) stained for cone arrestin , a cone photoreceptor marker . ( E ) Immunoblot probing for LDHA expression in 3-week-old retinal lysates of Rod-Cre> Ldhafl/fl and age-matched Cre− ( Ldhafl/fl ) siblings . Six retinae from 3 mice were pooled for lysate preparation in each group . ( F ) IHC for Ldhb on a retinal cross section of a 6-week-old Rod-Cre> Ldhafl/fl mouse . ( G ) Representative immunoblots of 293 T cells transfected with either full length ( FL ) or the coding region ( CDS ) of AU1-tagged murine LDHA driven by the CAG promoter . Cells were cotransfected with constructs encoding short hairpins targeting the murine Ldha transcripts . The short hairpin sh1 targets the 3’ untranslated region ( UTR ) of the mouse Ldha transcript while sh3 and sh4 target the coding region . Cox IV was used as a loading control . UT , untransfected . ( H ) Photoreceptor outer segment phenotype 40 days following in vivo electroporation of LDHAsh , CAG-rLDHB . CAG-mGFP was used for coelectroporation . For Ldhb staining , the gain during acquisition was adjusted to prevent oversaturation of signal intensity in overexpressing photoreceptors so as to preserve detail . Live histogram of pixel-intensity distribution was used in order to prevent clipping at the far-right end of intensities . Thus , Ldhb staining intensity in the IPL seems much lower compared to those observed in other figure panels in this study . Right panel , quantification of inner+outer segment lengths . Data , Mean±SD ( n = 75 photoreceptors , 3 retinae for LDHAsh+ CAG-rLDHB group ) . Statistics , One-way ANOVA with Tukey’s multiple comparison test . ( I ) Control histochemical reactions for SDH and COX activity . SDH reaction on unfixed retinal tissue without the substrate or in presence of Malonate , a competitive inhibitor . The light blue precipitate in –Succinate reaction was observed in outer segments . It differed from the intense purple precipitate when the substrate was included and does not match SDH localization in the retina . COX was inhibited by sodium azide ( +NaN3 ) and failed to form the end product of the histochemical reaction . ( J ) Confocal image of retinal cross section stained with anti-PDHE1 antibody that recognizes a subunit of pyruvate dehydrogenase ( PDH ) . Highest signal is seen in the photoreceptor inner segments , as well as the OPL and IPL synaptic layers . Scale bar , 25 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 00610 . 7554/eLife . 25946 . 007Figure 1—figure supplement 3 . Cell-autonomous effect of Ldha knockdown . Rhodopsin staining of a 42-day-old mouse retina electroporated with a shRNA construct targeting Ldha . Magnification of a retinal cross section focusing at the edge of the electroporation boundary is shown to depict the outer segments within the electroporated patch ( left side of the dotted line ) and outside ( right-hand-side of the dotted line ) . CAG-mGFP was used as a coelectroporation plasmid . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 00710 . 7554/eLife . 25946 . 008Figure 1—figure supplement 4 . Mitochondrial activity after Ldha loss-of-function . Cytochrome oxidase ( COX ) activity after deletion of Ldha in the rods ( Rod-cre> Ldhafl/fl , bottom right ) of a 7-week animal . Age-matched heterozygous sibling ( Rod-cre> Ldhafl/+ , bottom left ) is included as control . For comparison , a retina from wild-type mouse ( previously presented in Figure 1 ) is included ( top panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 008 Next , we wanted to ascertain if photoreceptors produce or consume lactate . As a first step , the expression of the LDH subtypes was examined . LDH is a tetrameric enzyme composed of LDHA and LDHB subunits encoded by the Ldha and Ldhb genes respectively . The subunits can assemble in five different combinations with differing kinetic properties ( Dawson et al . , 1964; Doherty and Cleveland , 2013 ) ( Figure 1—figure supplement 2B ) . A tetramer of all LDHA subunits has high affinity for pyruvate and a higher Vmax for pyruvate conversion to lactate than does an all-LDHB isoenzyme . In addition , many glycolytic cancers have elevated Ldha expression ( Balinsky et al . , 1983; Behringer et al . , 2003 ) . On the contrary , an all-LDHB tetramer is maximally active at low pyruvate concentrations , is strongly inhibited by pyruvate , and is expressed in tissues using lactate for oxidative metabolism or gluconeogenesis ( Dawson et al . , 1964 ) . We examined the expression of LDHA and LDHB subunits in the retina by immunohistochemistry ( IHC ) ( Figure 1C ) . Photoreceptors showed strong expression of LDHA , particularly with respect to the other retinal cell types . Similar results were obtained by others using a different set of commercially available antibodies ( Casson et al . , 2016; Rueda et al . , 2016 ) . Immunohistochemical localization also indicated that LDHB was abundantly expressed in the cells of the inner nuclear layer ( INL ) , which includes interneurons and Mueller glia . To validate the staining pattern obtained by IHC , expression analysis of transcripts of Ldha and Ldhb genes by in situ hybridization ( ISH ) was performed ( Figure 1D ) . This confirmed that Ldha expression is enriched in the photoreceptors , whereas Ldhb is excluded . This was also confirmed by qRT-PCR analysis of Ldha and Ldhb transcripts in isolated rod photoreceptor cDNA ( Supplementary file 1 ) . We conclude that photoreceptors have predominantly LDHA-type subunits . We also assessed the ability of the retina to secrete lactate after treatment with the LDHA-specific inhibitor , FX-11 ( Le et al . , 2010 ) . FX-11 significantly reduced lactate secretion ( Figure 1E ) . Similar to oxamate , FX-11 also resulted in an increased percentage of ATP that was sensitive to azide inhibition ( Figure 1F ) . To investigate if photoreceptors produce lactate in an Ldha-dependent manner , mice with a conditional allele of Ldha ( Wang et al . , 2014b ) , ( Ldhafl/fl ) , were used . The specificity and efficiency of Cre recombinase under control of the rhodopsin regulatory elements ( Le et al . , 2006 ) were first tested , which showed that only rod photoreceptors had a history of cre expression ( Figure 1—figure supplement 2C , D ) ( the mouse line henceforth called Rod-cre ) . The recombination efficiency varied between ~50–90% of photoreceptors among different retinae . The Rod-cre; Ldhafl/fl retinae were examined for LDHA protein , which showed a significant reduction ( Figure 1—figure supplement 2E ) . A compensatory expression of Ldhb in photoreceptors was not detected ( Figure 1—figure supplement 2F ) . Lactate production in these retinae was examined and was found to be significantly reduced ( Figure 1G ) . Thus photoreceptors produce lactate in an Ldha-dependent manner . In order to assess if reduction in Ldha expression created a cellular phenotype in photoreceptors , and if so , whether it was required autonomously , electroporation of a short hairpin RNA ( shRNA ) specifically targeting the 3’ untranslated region ( UTR ) of the Ldha transcript was used ( Figure 1—figure supplement 2G ) . This strategy was taken , vs . examination of the rods in the Rod-cre; Ldhafl/fl retinae , due to the concern that a reduction in lactate by rods might affect closely associated cell types , such as Mueller glia and/or RPE cells , creating non-autonomous effects on rods . A plasmid encoding this shRNA was delivered to the retina in vivo by electroporation . Electroporation occurs in patches comprising 15–30% of the retina , and in a given patch , only ~20% cells are electroporated ( Sui Wang and C . Cepko , unpublished ) . Thus , plasmid transfection via electroporation allowed us to determine if Ldha has a cell-autonomous role in photoreceptors . The electroporated photoreceptors had markedly reduced OS length when compared to control ( Figure 1H , I ) . Genetic complementation by coelectroporation of a sh-resistant Ldha cDNA that lacks the 3’UTR demonstrated that the defect was attributable to Ldha loss-of-function ( Figure 1H , I ) and the phenotype observed with the shRNA was not due to off-target effects . To determine if the catalytic activity of LDHA was required for rescue , an allele of Ldha with a point mutation in the catalytic center ( LDHAH193>A ) was introduced . It failed to rescue the shRNA phenotype . Finally , expression of Ldhb was not sufficient to compensate for Ldha loss-of-function ( Figure 1—figure supplement 2H ) . To confirm that the Ldha knockdown via electroporation conferred a cell-autonomous phenotype , we examined rhodopsin localization in mGFP-negative rods within the electroporated patch ( Figure 1—figure supplement 3 ) . A non-autonomous deleterious effect on rods that did not receive the plasmid ( mGFP- ) was not observed . The rhodopsin localization and the length of the OSs in GFP-negative rods within the patch did not vary from that of the rods lying outside of the electroporated patch ( Figure 1—figure supplement 3 ) . The cyclical process of OS shedding and renewal is regulated by light ( LaVail , 1976 ) . Since LDHA function is necessary to maintain OS length , we assessed the effect of Ldha knockdown in dark-reared mice and compared with mice raised in normal room light . Electroporated mice were raised with their mothers in normal room light until eyes were open ( P11 ) , and then shifted to the dark for 3 weeks . In mice with no Ldha knockdown , there was ~25% increase in IS+OS length after dark rearing compared to the light:dark condition ( Figure 1J , K ) , presumably as a part of an adaptive mechanism that might include less OS shedding ( Penn and Williams , 1986 ) . Interestingly , in mice with the Ldha knockdown , dark rearing resulted in a partial rescue of the Ldha knockdown phenotype ( Figure 1J , K ) . The average IS+OS length after Ldha knockdown was similar to that of light-treated control animals . These data indicate that reducing the need for OS biogenesis , as occurs in the dark , led to a reduced reliance on Ldha function . Cells with immature or dysfunctional mitochondria become reliant on glycolysis by increasing Ldha expression at the expense of Ldhb ( Facucho-Oliveira et al . , 2007; Ross et al . , 2010; Trifunovic et al . , 2004 ) . Although photoreceptors have abundant mitochondria , a reason for their high Ldha and low Ldhb expression could be subpar mitochondrial function , especially when compared to other retinal cell types . Thus , we assessed whether there was a mitochondrial activity difference between the photoreceptors and INL cells by examining succinate dehydrogenase ( SDH ) and cytochrome oxidase ( COX ) activity in fresh , unfixed , adult retinal sections ( Figure 1L ) . SDH/complex II plays a role in the citric acid cycle , as well as in the electron transport chain , and its subunits are encoded by the nucleus . COX or complex IV plays a role in the electron transport chain and has catalytic subunits that are encoded by the mitochondrial genome ( mtDNA ) . SDH activity was not lower in the photoreceptors relative to INL cells . COX activity was high in the photoreceptor layer , even higher than that seen in the other retinal layers . The specificity controls for the histochemical reaction are presented in Figure 1—figure supplement 2I . Finally , IHC for SDH was carried out . The highest IHC signal was observed in the photoreceptor inner segments ( IS ) , as well as the OPL and IPL synaptic layers ( Figure 1M ) , in good agreement with the observed SDH activity . IHC for another mitochondria-specific enzyme , pyruvate dehydrogenase , showed a similar pattern ( Figure 1—figure supplement 2J ) indicating that these are the sites of maximal mitochondrial densities in the retina . These data align with other studies that assessed mitochondrial activity in the retina ( Hansson , 1970; Rueda et al . , 2016 ) . Thus , lactate production by the photoreceptors cannot be attributed to lack of mitochondrial activity . Similarly , a decrease in mitochondrial COX activity was not detectable in the Rod-cre; Ldhafl/fl retinae ( Figure 1—figure supplement 4 ) by the histochemical assay . LDHA supports glycolysis by providing a ready supply of cytosolic NAD+ that is independent of O2 availability and/or mitochondrial function . The phenotype observed following Ldha knockdown might be indicative of a reliance on glycolysis where cells might exhibit a preference for unabated and rapid flux through glycolysis . Alternatively , it could be due to an unidentified role of Ldha in OS maintenance . To understand the extent of photoreceptors’ dependence on glycolysis , we designed an experimental strategy that satisfied the following criteria: ( 1 ) Does not ablate core glycolytic enzymes in order to avoid pleiotropic effects due to their possible non-glycolytic roles , ( 2 ) Targets a glycolytic node such that impact on other biosynthetic pathways , such as Pentose Phosphate Pathway ( PPP ) , would be minimal and ( 3 ) Uncovers glycolytic reliance and differentiates it from ‘housekeeping’ glycolysis . Glucose-derived metabolites are committed towards glycolytic flux by the enzyme 6-phosphofructo-1-kinase ( PFK1 ) , which catalyzes conversion of fructose-6-phosphate ( F6P ) to fructose-1 , 6-bisphosphate ( F-1 , 6-BP ) ( Figure 2—figure supplement 1 . ) . The most potent allosteric activator of PFK1 is fructose-2 , 6-bisphosphate ( F-2 , 6-BP ) ( Hers and Van Schaftingen , 1982 ) . F-2 , 6-BP is synthesized from F6P by the kinase activity of the bifunctional enzyme , 6-phosphofructo-2-kinase/fructose-2 , 6-bisphosphatase ( PFK2 ) ( Figure 2—figure supplement 1A , B ) . To examine the glycolytic dependence of photoreceptors , we targeted the steady-state levels of the metabolite , F-2 , 6-BP as it would satisfy the above criteria . First , we examined expression of PFK2 isoenzymes encoded by Pfkfb1-4 genes ( Figure 2—figure supplement 1C ) . Pfkfb3 expression could not be detected . Pfkfb1 , 2 and 4 were expressed in either a photoreceptor-enriched or photoreceptor-specific pattern , suggesting a propensity of these cell types to regulate glycolysis via a PFK2-dependent mechanism . With the exception of Pfkfb3 , all other PFK2 isoenzymes have kinase and phosphatase domains on the same polypeptide ( Mor et al . , 2011 ) ( Figure 2—figure supplement 1B ) . In addition to potential problems posed by functional redundancy ( i . e . knockdown of one isoenzyme might not be sufficient ) , genetically ablating the PFK2 isoforms would not uncover the preference for directionality ( i . e . an observed phenotype could be attributed to absence of either the kinase or phosphatase function ) . In addition , the structure-function relationships of their kinase and phosphatase domains are not known , thus making kinase- or phosphatase-dead versions is not straightforward . To overcome these problems , we overexpressed Tigar ( TP53-induced glycolysis and apoptosis regulator ) as it is functionally similar to the phosphatase domain of PFK2 with well-characterized F-2 , 6-BPase activity ( Bensaad et al . , 2006 ) ( Figure 2—figure supplement 1B ) and hence reduces the steady state levels F-2 , 6-BP . This is the intended effect and bypasses the aforementioned concerns with the conventional genetic loss-of-function approach associated with PFK2 isoenzymes . In addition to the predicted function of reducing glycolysis , overexpression of Tigar would not negatively affect the PPP ( Bensaad et al . , 2006 ) . We utilized an experimental strategy that addressed the following concerns: ( 1 ) The effect should be autonomous to photoreceptors , ( 2 ) the phenotype should be induced in fully mature photoreceptors , and ( 3 ) the phenotype should discernably be due to perturbations specifically of fructose-2 , 6-bisphosphate . Our experimental scheme utilized a construct that expressed tamoxifen-inducible Cre only in rods ( Figure 2A , B and Figure 2—figure supplement 1D ) . Expression of Tigar specifically in adult photoreceptors resulted in a significant reduction of OS length ( Figure 2C , D ) . This phenotype was specifically attributable to the phosphatase activity because expression of a catalytic dead version of Tigar , Tigar-TM ( triple mutant , H11>A , E102>A , H198>A ) ( Bensaad et al . , 2006 ) , did not cause a change in the photoreceptor OS length ( Figure 2—figure supplement 1E , F ) . To ascertain if the phenotype is specifically attributable to Tigar’s phosphatase activity on F-2 , 6-BP , we decided to coexpress Pfkfb3- a PFK2 isoform that has the kinase activity ~700 fold higher than the phosphatase ( Sakakibara et al . , 1997 ) ( Figure 2A and Figure 2—figure supplement 1B ) . Interestingly , overexpression of Pfkfb3 alone did not result in an overt phenotype- the OS length and morphology were indistinguishable from those of the control electroporated retina ( Figure 2C , D ) . Overexpression of Pfkfb3 was able to rescue the reduction in OS length caused by Tigar expression ( Figure 2C , D ) . Together , these data suggest that adult photoreceptors are sensitive to perturbations targeting F-2 , 6-BP . 10 . 7554/eLife . 25946 . 009Figure 2 . Targeting allostery reveals glycolytic reliance for outer segment maintenance . ( A ) Constructs for spatio-temporal control of expression of Tigar and Pfkfb3 . DsRed used as the cre reporter , mGFP as a coelectroporation marker . ( B ) Scheme for electroporation and tamoxifen induction . i . p , Intraperitoneal . ( C , D ) IS+OS length were measured following introduction of Tigar ( n = 72 cells ) , PFKB3 ( n = 72 cells ) , and Tigar and Pfkfb3 constructs ( n = 78 ) , shown in ( A ) . Controls were -tamoxifen ( n = 62 ) and +tamoxifen ( n = 74 ) . Scale bar , 50 μm . Data are Mean±SD . One-way ANOVA with Tukey’s correction . Outlined areas magnified to show IS and OS morphology . ( E ) AAV genomes for expression of mGFP ( AAV-mGFP ) or Tigar ( AAV-TIGAR ) . ( F ) Cross-sections of AAV-mGFP alone or AAV-TIGAR ( coinjected with AAV-mGFP ) infected retinae harvested at P28 imaged for mGFP expression . ( G ) Intracellular lactate normalized for total protein was quantified for retinae infected with AAV-mGFP ( Control ) or AAV-mGFP + AAV-TIGAR . Data represented as percentages relative to age-matched , freshly isolated retinae . Data , Mean±SD . Unpaired , two-tailed t-test with Kolmogorov-Smirnov correction for non-Gaussian distribution . ONL , outer nuclear layer . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 00910 . 7554/eLife . 25946 . 010Figure 2—source data 1 . Source data for Figure 2G . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 01010 . 7554/eLife . 25946 . 011Figure 2—figure supplement 1 . Allosteric regulation in photoreceptor glycolysis . ( A ) Schematic of the glycolytic pathway . ( B ) PFK2 isoenzyme polypeptide structure depicting the kinase and phosphatase domains . PFKFB3 , has ~700 times kinase to phosphatase activity . Conversely , TIGAR ( TP53-induced glycolysis and apoptosis regulator ) is functionally similar to the phosphatase domain of PFK2 . ( C ) Expression of PFK2 isoenzymes in the retina . In situ hybridization ( left panels ) for Pfkfb1 , Pfkfb2 , and Pfkfb4 on retinal sections and representative immune-stained sections ( right panels ) . ( D ) Confocal images of retinal cross-sections after in vivo electroporation of CAG-mGFP , RHO-ERT2CreERT2 and DsRed cre reporter . In absence of tamoxifen , some leaky expression of the DsRed reporter was seen ( top panels ) , but it remained restricted to the photoreceptors . In tamoxifen injected mice ( bottom ) induction of dsRed was seen that remained confined to the photoreceptor layer . mGFP serves as electroporation control . ( E ) Cross sections of CAG-Tigar-TM electroporated retinae at postnatal day 45 . CAG-mGFP was used as coelectroporation marker . Magnification of outlined area to reveal outer segment morphology . OS , outer segment . ( F ) Expression Tigar-TM ( n = 85 cells , 5 retinae ) did not result in significant reduction in IS+OS length . Unpaired , t-test with Kolmogorov-Smirnov correction . ( G ) Cross-section of a 27-day -ld , AAV-mGFP infected retina stained for cone arrestin . The mGFP expression remained confined to the photoreceptor layer and cone arrestin colocalization was not observed . ( H ) Representative immunoblot of retinal lysates from 28 day old mice infected with AAV-mGFP alone or AAV-mGFP + AAV-TIGAR . GAPDH served as loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 011 Next , the effects of Tigar expression on glycolysis were assayed . Although electroporation answers the question of the cell autonomous effect of a perturbation , the total number of affected cells and their percentage in the electroporated area are too minor to determine biochemical contributions . Adeno-associated virus ( AAV ) -mediated transduction of Tigar into photoreceptors was thus used , as it transduces a greater percentage of cells than electroporation . An AAV construct that drives expression of Tigar and/or mGFP from the bovine rhodopsin ( RHO ) promoter specifically in rods was constructed ( Figure 2E , F and Figure 2—figure supplement 1G ) . The AAVs ( expressing mGFP alone or mGFP and TIGAR ) were injected at postnatal day 6 ( P6 ) , after the end of cell proliferation , and nearly full retinal infection was confirmed by indirect ophthalmoscopy at P24-P27 by assessing the GFP fluorescence . Retinae were harvested at P28 and examined for expression ( Figure 2—figure supplement 1H ) and lactate levels ( Figure 2G ) . Consistent with the idea that Tigar would interfere in allosteric regulation of glycolysis , a significant reduction in retinal lactate was observed in the AAV-TIGAR infected retinae compared to the control AAV-mGFP infected retinae ( Figure 2G ) . Given the essential role of Ldha in postmitotic photoreceptors and proliferating cancer cells , other aspects of metabolism that have been discovered in cancer cells , such as the expression of pyruvate kinase isoforms , were investigated . Pyruvate kinase catalyzes the final irreversible reaction of glycolysis and distinct isoenzymes are encoded by two genomic loci , Pkm ( muscle ) and Pklr ( liver and red blood cell ) . Pklr transcripts were not detected in the retina , ( Figure 3—figure supplement 1 ) , but M1 and M2 splice isoforms of the PKM gene were detected ( Figure 3—figure supplement 2A ) in line with protein expression data reported earlier ( Lindsay et al . , 2014 ) . The M2 isoform is known to regulate aerobic glycolysis , promotes lactate production , and is upregulated in many tumors ( Christofk et al . , 2008a , 2008b ) . This isoform was previously reported to be expressed in photoreceptors ( Casson et al . , 2016; Lindsay et al . , 2014; Morohoshi et al . , 2012; Rajala et al . , 2016; Rueda et al . , 2016 ) . We confirmed that there is photoreceptor-enriched expression of PKM2 by IHC ( Figure 3A ) . PKM1 , known to be expressed in most differentiated cell types in adults ( Jurica et al . , 1998 ) , was expressed in the cells of the INL and ganglion cell layer , as shown by IHC ( Figure 3A ) , but was not detectable in photoreceptor cells . In this regard , our data differed from some published findings ( Casson et al . , 2016; Lindsay et al . , 2014 ) but matched those of others ( Rajala et al . , 2016 ) . To address this discrepancy and validate commercially available antibodies , we performed isoform-specific ISH ( Figure 3A ) and confirmed the expression pattern that we observed using IHC . We also examined transcript abundance by qPCR in mRNA purified from isolated rod photoreceptor cells ( Supplementary file 1 ) and found the M1 isoform to be much less abundant than M2 in the photoreceptors . Postnatally , PKM1 protein expression gradually increased , in correlation with increased differentiation and decreased proliferation in the developing retina ( Figure 3B ) . On the other hand , PKM2 protein expression was detectable during the period of proliferation and its expression did not decrease with increased differentiation , likely due to retention of expression in differentiated photoreceptors . Previous studies on pyruvate kinase in the context of proliferation have suggested that loss-of-function of Pkm2 reduces proliferation attributable to the glycolytic reliance of mitotic cells for growth ( Christofk et al . , 2008b; Israelsen et al . , 2013 ) . To assess if PKM2 plays an essential role in rod photoreceptors , an shRNA construct that specifically targeted mouse PKM2 ( PKM2sh ) , but spared PKM1 ( Figure 3—figure supplement 2B , C , D , E ) , was generated . In vivo electroporation of a plasmid encoding PKM2-specific shRNA resulted in photoreceptors with significantly shorter OS than control ( Figure 3D , E ) . This phenotype could be rescued by coelectroporation of a construct encoding human PKM2 cDNA ( Figure 3C , D ) , which was not targetable by the shRNA ( Figure 3—figure supplement 2F ) . Coelectroporation of plasmid encoding mouse PKM1 with PKM2sh did not rescue the OS length defect ( Figure 3C , D ) . These data demonstrate that PKM1 and PKM2 play nonequivalent roles in rod photoreceptors . In order to further investigate whether PKM2 was needed for an autonomous role in rods , we electroporated a low concentration of plasmid encoding the shRNA ( Figure 3—figure supplement 3 ) . In addition to few electroporated rods , there were very few electroporated INL cells . In this condition , electroporated rods displayed a similar OS phenotype to that observed with higher concentrations of PKM2sh ( Figure 3C ) , that is , reduced OSs . We also generated an shRNA construct that targeted exon 4 , which is shared between mouse PKM1 and PKM2 ( PKM1 +2 sh ) ( Figure 3—figure supplement 2C , D ) . Electroporation of this construct resulted in a significant decrease in the OS length ( Figure 3—figure supplement 2G ) . The photoreceptor morphology and OS length were the same as that observed following electroporation with PKM2sh . While complementation with human PKM2 was sufficient to rescue the IS+OS length defect , we noted some abnormalities with the morphology of some of the photoreceptor ISs and OSs ( Figure 3—figure supplement 2G ) . In 4/6 retinae , many photoreceptors lacked clear borders of IS and OS , though in 2/6 retinae , the morphology closely resembled that of control retinae ( Figure 3—figure supplement 2G ) . The contribution of PKM2 to OS maintenance was further investigated in the retinae of dark-reared mice electroporated with PKM2sh ( Figure 3E ) . Dark rearing significantly increased OS length in these animals ( Figure 3F ) . Taken together , the results from dark-reared animals , in which Ldha or Pkm2 was knocked down , indicate the requirement for the glycolytic pathway in OS maintenance . Since two different genes that promote aerobic glycolysis are necessary for the light-dependent maintenance of OS , the short OS phenotype is likely due to a reduced supply of the building blocks normally supplied by aerobic glycolysis . In order to probe the biochemical effects of PKM2 reduction , lactate production was examined . Since electroporated retinae are not ideal for these experiments , mice that had a conditional deletion of Pkm2 in rods were used . The Pkm2fl/fl mouse strain , in which the M2-specific exon 10 was floxed ( Israelsen et al . , 2013 ) , was crossed with the Rod-cre strain . The retinae with deficiency of PKM2 had a small but significant decrease in lactate production , as compared to the controls ( Figure 3G ) . We also noted upregulation of PKM1 in these retinae ( Figure 3—figure supplement 2H ) similar to what has been reported before ( Israelsen et al . , 2013 ) . However , as noted above , in rods electroporated with PHM2sh , PKM1 expression was not observed ( Figure 3—figure supplement 2I ) . One possibility for the difference in the presence of the M1-specific exon in the mRNA in the knockout vs . the knockdown manipulation might reflect a choice made by the splicing machinery . After the deletion of the ‘preferred’ M2-specific exon in the genome in the knockout , the splicing machinery might include the M1 exon as a default choice . However , when the shRNA was used to knockdown the PKM2 isoform , the splicing event that chose the M2-specific exon would have already happened . 10 . 7554/eLife . 25946 . 012Figure 3 . PKM1 and PKM2 isoforms have nonequivalent roles . ( A ) Biased expression of M1 and M2 isoforms in retinal layers detected by IHC and ISH . ( B ) Immunoblot of retinal lysates from postnatal retina at different developmental stages . HEK293T cell lysates that were from untransfected ( UT ) cells , or those transfected with CAG-FLAGmuPKM1 ( M1 ) or CAG-FLAGmuPKM2 ( M2 ) as controls . Postnatal age in days . A , mature retina ( P25–P30 ) . ( C ) Outer segment phenotype of P45 mice after electroporation with constructs encoding mouse PKM2-specific shRNA ( PKM2sh ) and adding either mouse PKM1 ( muPKM1 ) or human PKM2 ( huPKM2 ) . Selected areas in yellow boxes are magnified on the right . ( D ) Quantification of IS+OS lengths obtained in ( C ) . n = 32–53 cells from 3 to 4 retinae . ( E ) Outer segment phenotype of dark-reared P31 mice electroporated with PKM2sh-encoding plasmid . The yellow-boxed region is magnified and presented on the right . ( F ) Quantification of IS+OS lengths obtained in ( e ) . n = 75 cells from three retinae . ( G ) Secreted lactate from freshly isolated retinae from Pkm2fl/fl ( fl/fl ) ( n = 12 ) or Rod-cre> Pkm2fl/fl ( m2-/- ) ( n = 16 ) mice . ( H ) Outer segment phenotype after CAG promoter-driven overexpression of Flag-tagged mouse PKM1 or PKM2 . Inset , higher magnification of IS and OS . ( I ) Quantification of IS+OS lengths obtained in ( H ) . n = 35 cells from three retinae in PKM1 and PKM2 groups . ONL , outer nuclear layer . Data , Mean±SD . Statistics , one-way ANOVA with Tukey’s correction for panels D , I; two-way ANOVA with Tukey’s multiple comparison test for panel F; unpaired , two-tailed t-test with Kolmogorov-Smirnov correction for panel G . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 01210 . 7554/eLife . 25946 . 013Figure 3—source data 1 . Source data for Figure 3G . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 01310 . 7554/eLife . 25946 . 014Figure 3—figure supplement 1 . Assessment of Pklr expression in retina and liver . PCR analysis to detect Pyruvate kinase Liver RBC ( Pklr ) , Rhodopsin ( Rho ) , Actin ( Act ) and Hepatocyte nuclear factor 4α ( Hnf4α ) . RNA was extracted from the retina and liver of an adult mouse ( >P28 ) . PCR reactions with retinal cDNA ( lanes with reverse transcriptase , +RT ) as a template were compared with reactions where liver cDNA was used . The –RT lanes served as control for genomic DNA contamination . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 01410 . 7554/eLife . 25946 . 015Figure 3—figure supplement 2 . Characterization of PKM1 and PKM2 function in the retina . ( A ) Schematic of the PKM genomic locus and depiction of generation of M1 and M2 isoforms by alternative splicing . ( B ) Plasmids encoding cDNA of either mouse PKM1 or PKM2 and destabilized GFP ( dGFP ) were transfected in 293T cells . CAG-mCherry served as transfection control . Plasmids encoding candidate shRNAs driven by the U6 promoter were cotransfected . ( C ) Live imaging of dGFP or mCherry in transfected cells . PKM1 +2 sh encodes for shRNA that targets both M1 and M2 , whereas the PKM2sh codes for shRNA that is M2-specific . Scale bar , 50 μm . ( D ) Effect of shRNAs on protein steady state . Representative immunoblot of 293T cells transfected with Flag-tagged muPKM1 or muPKM2 driven by the CAG promoter and co-transfected with a plasmid encoding for mCherry and indicated shRNA-encoding constructs . Forty-eight hour post-transfection , cells were lysed and lystaed used for immunoblots using indicated antibodies on the left . This panel also represents shRNAs that had a knockdown effect in the screen in ( b ) , but were not considered in favor of PKM1 +2 owing to its strong knockdown effect . UT , untransfected 293T ( endogenously express PKM2 but not PKM1 ) . ( E ) , Representative immunoblot of 293 T cells transfected with Flag-tagged muPKM1 or muPKM2 driven by the CAG promoter and cotransfected with PKM2sh from ( c ) ( ‘+’ lanes ) or empty sh vector ( ‘- ‘lanes ) and harvested 24 hr later for lysate preparation . ( F ) Human and mouse M2 exon alignment . The region targeted by PKM2sh is highlighted . This shRNA did not knockdown the human PKM2 . ( G ) In vivo electroporation of a plasmid encoding PKM1 +2 shRNA resulted in photoreceptors with significantly shorter inner plus outer segments ( top left ) . This phenotype could be rescued by coelectroporation of a construct encoding human PKM2 cDNA ( top right and bottom left ) . In 4/6 retinae ( top right ) , many photoreceptors lacked clear borders distinguishing inner and outer segments ( arrows ) while some photoreceptors looked normal ( arrowheads ) . In 2/6 retinae ( bottom left ) , the morphology resembled that of control retinae . ( H ) Retinal cross section of a 6 week old Rod-cre; PKM2fl/fl mouse stained for PKM1 . ( I ) Retinal cross section of a P40 mouse electroporated with PKM2sh and CAG-mGFP . Arrows mark inner segments of electroporated photoreceptors . ( J ) Representative immunoblot of 293T cells transfected with Flag-tagged muPKM1 or muPKM2 driven by the CAG promoter . GAPDH served as loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 01510 . 7554/eLife . 25946 . 016Figure 3—figure supplement 3 . Cell-autonomous effect of PKM2 knockdown . Photoreceptor outer segment phenotype 41 days after sparse in vivo electroporation of a plasmid encoding PKM2sh . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 01610 . 7554/eLife . 25946 . 017Figure 3—figure supplement 4 . PKM1 and PKM2 splicing factors . In situ hybridization for Srsf3 mRNA ( left ) and Ptbp1 mRNA ( right ) on retinal sections . SRSf3 mRNA is abundant in photoreceptors while Ptbp1 is more enriched in the inner nuclear layer ( INL ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 01710 . 7554/eLife . 25946 . 018Figure 3—figure supplement 5 . Outer segments in young Rod-cre; Pkm2fl/fl mice . Retinal cross section of a 6-week-old Rod-cre; Pkm2fl/fl ( depicted previously in Figure 3—figure supplement 2H ) stained for PKM1 , Rhodopsin ( RHO ) and counterstained with DAPI . The region of slightly longer outer segments is demarcated with arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 01810 . 7554/eLife . 25946 . 019Figure 3—figure supplement 6 . Age-dependent retinal changes in Rod-cre; Pkm2fl/fl mice . Retinal cross-section of a 37-week-old Rod-cre; PKM2fl/fl mouse stained for PKM2 , Rhodopsin ( RHO ) and counterstained with DAPI . Slight nuclear disorganization in the retinal mosaic corresponding to PKM2 loss is demarcated in the panel corresponding to the DAPI channel ( lower left ) . Aberrant mislocalized RHO+ cell is marked by arrow . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 019 The differential expression of the M1 and M2 isoforms in the retinal layers could be attributable to the differential expression of splicing factors that promote inclusion or exclusion of the M1- or M2-specific exon . To evaluate this possibility , we examined the expression of Srsf3 , a splicing factor known to promote inclusion of the M2 exon ( Wang et al . , 2012 ) , and Ptbp1 , known to repress the M1 exon inclusion ( Chen et al . , 2012 ) ( Figure 3—figure supplement 4 ) . While Srsf3 was expressed at higher levels in photoreceptors , Ptbp1 was more enriched in the INL . Thus the regulation of PKM isoform preferences in retina is more complex than that predicted by canonical splicing models . We also noted slightly reduced rod OS length in the region that had PKM1 expression in the young ( postnatal 6 week ) Rod-cre; Pkm2fl/fl mice ( Figure 3—figure supplement 5 ) . Although the recombination in this line has been reported to be complete by 6 weeks , we cannot exclude the possibility of some recently recombined rods ( that express PKM1 ) in this region . These rods might not have had enough time to have a discernable impact on rhodopsin abundance and OS length . In addition , there might be some non-recombined rods interspersed in the broad region where PKM1 expression was apparent , and contributed to longer OSs . Due to the packing density of rods and abundance of rhodopsin , an immunohistochemical approach might not be a suitable way to reliably assess the OS length and its relation to PKM2 function . We also examined older mice ( 37 weeks , ~8 months old ) of this background hypothesizing that aging might uncover a subtle phenotype ( Figure 3—figure supplement 6 ) . The nuclei in the ONL region that had lost PKM2 protein expression were disorganized and lost their typical columnar arrangement , perhaps due to cell death . We also noted disorganization of OS in some regions where PKM2 was lost . Overall the OS length was reduced compared to aged Bl6/J mice , but this reduction was also noted for photoreceptors that had PKM2 protein expression . We cannot exclude the possibility of a non-autonomous effect on these rods as a response to tissue reorganization or alterations in their metabolic environment . Similarly , maintenance of some rhodopsin expression after PKM2 loss in the surviving rods might be an indication of an adaptive response on part of these cells . The electroporation approach , where only a few cells are transfected , circumvents these concerns and illustrates the critical cell-autonomous requirement of PKM2 for rods . PKM1 is constitutively active while PKM2 is regulatable ( Anastasiou et al . , 2012 ) . Biased expression of PKM2 in photoreceptors suggests that these cells may need to dynamically regulate glycolysis . The inability of PKM1 to rescue PKM2 loss-of-function indicates that merely replacing pyruvate kinase ( PK ) function after Pkm2 knockdown is not sufficient to restore the OS . In addition , it indicates the importance of glycolytic regulation at the PK step in photoreceptors . We examined the effect of forced expression of PKM1 in the presence of endogenous PKM2 , with the hypothesis that the constitutively active isoform might interfere at the regulatory step . We delivered plasmids encoding FLAG-tagged mouse PKM1 and PKM2 via in vivo electroporation ( Figure 3H ) . Photoreceptors electroporated with PKM1-expressing constructs , but not PKM2 expressing constructs , had a reduction in the length of the OS ( Figure 3H , I ) with the majority of the photoreceptors in the PKM1 electroporated retinae lacking discernable OS . The two proteins were expressed at equivalent levels , as assessed by Western blotting for the FLAG epitope in HEK293T cells ( Figure 3—figure supplement 2J ) . PKM2 has been shown to interact with tyrosine phosphorylated proteins ( Christofk et al . , 2008a ) and is tyrosine phosphorylated at position 105 ( pY105 ) in tumor cells ( Hitosugi et al . , 2009 ) leading to promotion of aerobic glycolysis . The pY105 is a shared epitope in PKM1 and PKM2 ( Figure 4A ) . To assess the phosphorylation status of PKM2 at this site , PKM2 was specifically immunoprecipitated from retinal lysates followed by immunoblotting using a phospho-Y105-specific antibody ( Figure 4B ) . We observed that PKM2 was phosphorylated at Y105 . In order to ascertain if phosphorylation of PKM2 at this site might have any physiological significance , its regulation by light was examined . PKM2 was immunoprecipitated from the retinae of mice at 3 hr intervals during a 24-hr time course , and phosphorylation at Y105 was probed ( Figure 4C ) . A light-dependent increase in phosphorylation at Y105 of PKM2 was observed . Thus , this phosphorylation site might be one of the target sites for physiologically relevant signaling events regulating aerobic glycolysis . Light-dependent phosphorylation of this epitope in the retina has also been reported recently using IHC and immunoblot approaches ( Rajala et al . , 2016 ) . The authors observed changes in signal intensity on IHC , that were dependent on light and activation status of the phototransduction pathway . Similarly , the Y105 epitope showed light-dependent phosphorylation as assessed by immunoblot analysis of total retinal lysates . Our results on immunoprecipitated PKM2 confirm that this protein is among the targets of a light-dependent signaling pathway . Thus , phosphorylation of this site was then used as a proxy for the tyrosine kinase signaling pathways that could phosphorylate PKM2 in the retina . Freshly explanted retinae were cultured with antagonists targeting specific pathways: Afatinib ( EGFR ) , Dasatinib ( Src ) , BMS536924 ( Insulin/IGF ) , PD173074 ( FGFR1 ) and Dovitinib/TKI258 ( FGFR1 and FGFR3 ) . PKM2 was immunoprecipitated and its phosphorylation at Y105 probed ( Figure 4D ) . FGF inhibitors , PD173074 and TKI258 , reduced PKM2 phosphorylation . Tyrosine kinase signaling can also target multiple nodes , including pyruvate dehydrogenase kinase and LDHA ( Fan et al . , 2011 ) , and regulate aerobic glycolysis in cancer . We observed that treatment with either PD173074 or TK1258 also resulted in a dose-dependent decrease in LDHA phosphorylation at the Y10 residue ( Figure 4E ) . Thus , FGF signaling potentially targets multiple nodes in order to regulate aerobic glycolysis in the retina . 10 . 7554/eLife . 25946 . 020Figure 4 . FGF signaling regulates aerobic glycolysis and anabolism . ( A ) Schematic of PKM1 and PKM2 polypeptide showing Y105 is a shared epitope between PKM1 and PKM2 . ( B ) Immunoprecipitation ( IP ) of PKM2 from adult retina followed by immunoblot ( IB ) for either PKM1 , PKM2 or pY105 PKM . IP using isotype-matched antibody ( IgG ) is used alongside to control for nonspecific binding . Lysates from skeletal muscle ( expresses PKM1 ) and 293T ( expresses only PKM2 ) included as controls . Molecular weight marker positions are depicted on the right-hand-side ( C ) Retinal lysates were prepared from eyes harvested at 3-hr interval during the 12 hr light 12 hr dark cycle . T0 is the time point of light on in the room . The lysates were subjected to immunoprecipitation with anti-PKM2 . Immunoprecipitates were probed for phosphorylation at Y105 by immunoblotting with the phospho-specific antibody . ( D ) Lysates from explants treated with candidate tyrosine kinase pathway inhibitors or vehicle control ( DMSO ) were subjected to immunoprecipitation with anti-PKM2 . Immunoprecipitates were probed for phosphorylation at Y105 by immunoblotting with the phospho-specific antibody . ( E ) FGF inhibitors also reduce phosphorylation of LDHA at the Y10 residue . Phosphorylation of FRS2 , an FGFR-interacting protein was included as a control . SDHA served as loading control . ( F ) Rate of lactate production from explants treated with DMSO ( n = 5 ) or FGF inhibitors PD173074 ( 5 mM ) ( n = 6 ) , PD173074 ( 20 mM ) ( n = 5 ) , TKI258 ( n = 6 ) . ( G ) Steady-state ATP levels per retina in explants after culture with TKI258 or DMSO . The retinae were transferred to Krebs’-Ringer's with NaN3 or NaCl ( untreated group ) for 30 min followed by harvest for ATP extraction . n = 7 , DMSO+NaCl; n = 9 , TKI258+NaCl; n = 9 , DMSO+NaN3; n = 9 , TKI258+NaN3 . Data are Mean±SD . Statistics , Two-way ANOVA with Tukey’s correction . ( H ) NADPH steady state levels in explants as a percentage of those measured in freshly isolated retina . Explants were treated with DMSO , oxamate , PD173074 , TKI258 or left untreated in culture medium . n = 4 groups . Unpaired t-test with Kolmogorov-Smirnov correction for indicated pairs . ( I ) NADP steady-state levels in explants as a percentage of those measured in freshly isolated retina . Explants were treated with DMSO , oxamate , PD173074 , TKI258 or left untreated in culture medium . Oxamate , n = 5; rest , n = 6 groups . Unpaired t-test with Kolmogorov-Smirnov correction for indicated pairs . ( J ) Blocking glycolysis or FGF signaling reduced EU incorporation in nascent RNA . Explants were treated with DMSO , oxamate , TKI258 or Actinomycin D ( RNA Pol II inhibitor ) followed by incubation with EU . ( K ) Quantitative PCR analysis of transcripts to ascertain relative expression of FGF or non-FGF targets ( Arr3 , Rs1 ) in explants cultured with or without RPE/Sclera complex ( +RPE or –RPE respectively ) . ( L ) Ability to produce lactate from neural retina increased when cultured in the presence of RPE/Sclera complex ( +RPE ) ( n = 11 ) as compared to those that were cultured without the complex ( -RPE ) ( n = 9 ) . Addition of FGF2 in –RPE cultures restored the ability ( -RPE+FGF2 ) ( n = 8 ) . Retinal explants were cultured with RPE attached in the explant culture medium . Before transferring them to Krebs’s-Ringer's for lactate estimation , the RPE/Sclera complex was removed and intact neural retina was used . For –RPE conditions , neural retina was cultured in explant medium followed by transfer to Krebs’-Ringer's . FGF2 was added to the explant culture medium but was absent in the Krebs’-Ringer's for -RPE+FGF2 condition . Data depict median in 1–99 percentile box and whiskers plot . Hinges extend between 25th to 75th percentiles . Statistics , Ordinary one-way ANOVA with Tukey’s correction . ONL , outer nuclear layer . INL , inner nuclear layer . GCL , Ganglion cell layer . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 02010 . 7554/eLife . 25946 . 021Figure 4—source data 1 . Source data for Figure 4F–I , K and L . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 02110 . 7554/eLife . 25946 . 022Figure 4—figure supplement 1 . Model summarizing regulation of glycolysis and its contribution to photoreceptor physiology . Aerobic glycolysis could serve as a metabolic adaptation to promote anabolism and visual cycle in the photoreceptors . Lactate represents carbon that is unavailable for oxidation and ATP generation . We favor a model where allosteric and FGF-mediated promotion of Warburg effect would funnel glucose towards glycolytic pathway enabling generation of NADPH- a key cofactor in lipid biosynthesis and visual cycle , and nucleotides- required RNA biosynthesis to replenish proteins lost in disc shedding and phototransduction . Since methods to quantify selective glucose uptake by photoreceptors in undissociated state are not available , its fate specifically in the photoreceptors and relative contribution of glycolysis to the ATP pool in these cells is currently not decipherable . Other fuels such as fatty acids can be used in the mitochondria for ATP requirements ( Joyal et al . , 2016 ) . A byproduct of mitochondrial activity is ROS generation . NADPH plays a critical role in ROS detoxification as well . Thus , in this way aerobic glycolysis might support sustainable energy generation by the photoreceptors . It is worth noting that the blood pressure in choroidal vasculature ( that lies behind the photoreceptors ) is not regulated and is generally considered to provide unabated supply of nutrients and oxygen to the neural retina . Given this anatomical feature , metabolic regulation in photoreceptors might not be geared toward economizing carbon allocation . Thus , aerobic glycolysis might not be a wasteful process , but a metabolic adaptation to meet multiple physiological needs . DOI: http://dx . doi . org/10 . 7554/eLife . 25946 . 022 To determine if FGF signaling might regulate lactate production , freshly explanted retinae were cultured with TKI258 or PD173074 , and lactate secretion was measured . Significantly reduced lactate secretion ( Figure 4F ) was seen to result from inclusion of either drug . In addition , inhibition of the FGF pathway resulted in increased mitochondrial dependence on ATP steady state maintenance ( Figure 4G ) . Thus , one role for FGF signaling in the adult retina is to promote glycolytic reliance . FGF signaling is required for the maintenance of adult photoreceptors in mice and zebrafish ( Campochiaro et al . , 1996; Hochmann et al . , 2012; Qin et al . , 2011 ) . Since it is possible that some of the effects are via regulation of aerobic glycolysis , we examined whether aerobic glycolysis promotes anabolism in the retina . Inhibition of aerobic glycolysis by oxamate treatment or by FGF inhibition resulted in significantly lower steady state NADPH levels- a key cofactor in biosynthetic pathways for lipids , antioxidant responses , and the visual cycle ( Figure 4H ) . We also observed that interference with aerobic glycolysis did not result in an equivalent reduction in NADP+ steady state levels ( Figure 4I ) . The lowering of NADPH level could be attributable to attenuation of the PPP-shunt as a result of decreased glycolytic flux and/or increased usage of NADPH to quench the reactive oxygen species- an unavoidable consequence of increased mitochondrial dependence . We also assessed other effects on cellular anabolism . Nucleotide availability for nascent RNA synthesis was visualized using ethynyl uridine ( EU ) incorporation after treatment with oxamate and TKI258 . Marked reduction in nascent RNA synthesis was evident following inhibition of LDH or FGF signaling ( Figure 4I ) . Among the large family of FGFs , basic FGF ( bFGF/FGF2 ) has been the most studied in the adult retina . In adult mice and primates , FGF2 is localized to a matrix surrounding photoreceptors and/or is found on their OS ( Gao and Hollyfield , 1992 , 1995; Hageman et al . , 1991 ) . The RPE might contribute to a high FGF2 concentration near photoreceptors via biosynthesis , and/or create a barrier to its diffusion from a retinal source . We first examined the role of the RPE in FGF-signaling . Adult retinal explants were cultured with the RPE/choroid/sclera complex , and expression of FGF target genes in the neural retina was compared with that of explants cultured without the attached complex . In the absence of this complex , the transcripts of known FGF signaling targets displayed reduced steady state levels ( Figure 4J ) . To assess if the reduction in FGF targets was part of a general transcription downregulation or specifically due to dampened FGF signaling , we examined expression of retinoschisin ( RS1 ) or cone arrestin ( Arr3 ) , genes expressed at moderate levels in the retina ( Blackshaw et al . , 2001 ) ( Figure 4J ) . These genes were not downregulated in the absence of the RPE complex . The effect of the RPE complex on aerobic glycolysis was analyzed by quantifying lactate production ( Figure 4K ) . Culturing retinae in the presence of the RPE complex resulted in a small , but significant , increase in the ability to produce lactate . Addition of FGF2 in the culture medium was sufficient to increase lactate production from explants cultured without the RPE . Together these data suggest that the RPE/choroid/sclera complex contributes to FGF signaling in the neural retina and that this signaling pathway plays a role in regulating the Warburg effect .
Several reports suggest that aerobic glycolysis is a feature of some normal proliferating somatic cells ( Agathocleous et al . , 2012; Brand and Hermfisse , 1997; Wang et al . , 2014b; Zheng et al . , 2016 ) , and not just of cancer cells . Our work expands the cell types where aerobic glycolysis can occur to include a mature cell type , the differentiated photoreceptor cell . Like proliferating cells , rod photoreceptors utilize aerobic glycolysis to meet their anabolic needs . A critical aspect of aerobic glycolysis is its ability to be regulated . The data presented here suggest that allostery and FGF signaling are the regulatory mechanisms in the retina . We favor a model where aerobic glycolysis appears to be relevant to photoreceptors not only for organelle maintenance , but likely also helps photoreceptors meet their multiple metabolic demands ( Figure 4—figure supplement 1 ) . In light of this model , it is important to assess the genetic tools that we employed to probe this pathway . Since we drove shRNA expression for Ldha and Pkm2 knockdown from a constitutively active promoter ( U6 ) , we speculate that there could be an effect during retinogenesis . We reproducibly observed retinal thinning , indicated by a reduced number of nuclear rows ( Figure 1H ) , especially in very well electroporated retinae . The thinning could be due to perturbation in the cell cycle of retinal progenitor cells , increased cell death , or both . As it is known that there is a role for LDHA and PKM2 in cell proliferation , it is quite likely that such an effect occurred here . The reduced retinal thickness was also apparent in some retinae from dark-reared animals that received the shRNA-encoding constructs against Ldha ( Figure 1J ) . Many photoreceptors that received knockdown constructs against Ldha or Pkm2 showed a significant increase in their OS length after dark rearing . This result argues for a physiological effect due to light exposure having an effect on the OS length , rather than a developmental defect . In addition , our experiments with Tigar gain-of-function , where expression is achieved in a spatiotemporal manner in order to have a minimal effect on retinal development , suggest that the effects of glycolytic perturbation on photoreceptor OSs can be parsed from the confounding effects on retinogenesis . Aerobic glycolysis in the retina may have implications for blinding disorders . Studies on retinal degenerative disorders indicate that there are metabolic underpinnings to photoreceptor dysfunction , especially those centering around glucose uptake and metabolism ( Aït-Ali et al . , 2015; Punzo et al . , 2009 ) . Furthermore , reducing metabolic stress prolongs survival and improves the function of photoreceptors ( Venkatesh et al . , 2015; Xiong et al . , 2015 ) . In such treated retinae , there is a trend toward upregulation of glycolytic genes ( Venkatesh et al . , 2015 ) or metabolites ( Zhang et al . , 2016 ) . However , a direct cause-and-effect relationship between cell survival and glycolysis has not been established . Our results highlight the metabolic strategies employed by healthy photoreceptors and provide a rational basis for the identification of candidate factors that would further clarify the role of glycolysis in retinal degeneration .
The synthetic promoter , CAG , consisting of cytomegalovirus ( CMV ) enhancer , chicken β-actin and rabbit β-globin gene splice acceptor was used for expression and genetic complementation . The expression pattern from this promoter when delivered by electroporation has been described previously ( Matsuda and Cepko , 2007 ) . Co-electroporation of a plasmid encoding myristoylated/membrane green fluorescent protein ( mGFP ) allowed visualization of cells that received the plasmid and marked the inner and outer segments . Co-electroporation rate of plasmids to the retina is close to 100% ( Matsuda and Cepko , 2007 ) . Full-length rat Ldhb ( rLdhb ) , human TIGAR , mouse PFKFB3 and human Pkm2 cDNA were obtained from Open Biosystems/GE Dharmacon . Subcloning , epitope tagging and site-directed mutagenesis were carried out by routine molecular biology procedures . For short hairpin ( sh ) design targeting PKM , and Ldha following resources/software were used: The RNAi consortium , CSHL RNAi central , iRNAi , Invitrogen Block-iT RNAi designer . Designed sh oligos were subcloned in pLKO . 1 TRC backbone to be driven by the U6 promoter ( Addgene , Cambridge , MA , #10878 ) and the sequences used in this manuscript are listed in Supplementary file 2 . Four hairpin constructs were screened for Ldha and 72 were screened for PKM1/PKM2 . Those hairpins that targeted specific mouse sequences but did not target human Pkm2 were chosen . The murine FLAG-tagged PKM1 and PKM2 cDNAs were obtained from Addgene ( #44240 and #42512 ) and subcloned in pCAG-EN . The pyruvate kinase activity from these ORFs has been already reported ( Anastasiou et al . , 2011 ) . The plasmids were mixed in equal molar ratios by accounting for their lengths and subjected to Phenol:Chloroform extraction followed by ethanol precipitation and resuspended to a final concentration of 1 mg/mL in Phosphate Buffered Saline . Subretinal in vivo injections and electroporation were carried out as described earlier ( Wang et al . , 2014a ) . When possible , the control and experimental constructs were injected in the pups of the same litter and the tail termini were snipped ( or left uncut ) to identify them later . For knockdown assays or testing expression from plasmids , transfection in HEK293T cells was carried out as using polyethylenimine ( PEI ) . These cells were maintained as a lab stock and were subjected to periodic in-house testing for mycoplasma . Since , these cells were used for protein overexpression and knockdown analyses , concerns of misidentification do not apply to the current work and hence were not checked by third-party testing services . For making the AAV-mGFP and AAV-TIGAR constructs , the CMV promoter in the empty AAV-MCS8 vector ( Harvard Medical School DF/HCC DNA Resource Core ) was replaced with the bovine rhodopsin promoter ( Matsuda and Cepko , 2007 ) . Woodchuck hepatitis virus posttranscriptional response element ( WPRE ) was added to enhance expression . Capsid type 8 AAVs were produced and titered as described previously ( Xiong et al . , 2015 ) . For subretinal injections of AAV , ~3 . 5 × 106–5 × 106 particles ( based on genome copies ) per eye were used . P6 pups were injected in order to transduce cells after the proliferative phase of retinogenesis so as to minimize any detrimental effects on cell division and dilution of replication-incompetent viruses . The extent of infection was assessed with a Keeler indirect ophthalmoscope using the cobalt blue filter and Volk 78 diopter lens on non-anesthetisized animals . Mice with edge-to-edge infection were tagged and used subsequently for lactate assays and immunoblotting . Timed pregnant , wild-type CD1 female mice were obtained from Charles River Laboratories , Boston , MA , and P0-P1 pups thereof were used in electroporations . C57BL/6J and the two-color Cre reporter mouse Gt ( ROSA ) 26Sortm4 ( ActB-tdTomato , -EGFP ) Luo/J ( referred to as mT/mG and described previously [Muzumdar et al . , 2007] ) were obtained from the Jackson Laboratories ( JAX ) , Bar Harbor , ME . Ldhafl/fl ( Wang et al . , 2014b ) , Pkm2fl/fl ( Israelsen et al . , 2013 ) , Rod-cre ( Le et al . , 2006 ) mice have been described before . Rod-Cre; Ldhafl/fl and Rod-cre; Pkm2fl/fl mouse lines were established . For experimentation , these mice were backcrossed with Ldhafl/fl or Pkm2fl/fl parents and Cre+ and Cre− F1 progeny were used to ensure equivalent allelic copies of the Cre transgene , minimum genetic difference and ease of age-matching by using the siblings . Animals were housed at room temperature with 12 hr light and 12 hr dark cycle . Light inside the cages in the room varied from 0 to 3 lx in the cage farthest from the light source to 300 lx in the cage closest to it . As a practice , the electroporated mice inhabited rack spaces where light intensity in the cages varied from ~175 to~235 lx . At weaning , the mice were segregated according to their sexes , thus a cage usually had the control and electroporated pups from the same litter . Tamoxifen injections were carried out as described previously ( Matsuda and Cepko , 2007 ) . For dark rearing , electroporated animals were raised with their mothers until P11 , when the eyes started to open . Following this , they were transferred to animal housing maintained in darkness until weaning age , when they were weaned and group housed in dark until indicated times for harvest . In order to minimize effects due to circadian regulation of OS growth , the light and dark-reared animals from all the groups ( control , LDHAsh and PKM2sh ) were harvested on the same day and within 3 hr of each other . Water and chow were available ad libitum . Animal care was following institutional IACUC guidelines . Wild-type , pigmented C57BL/6J mice ( JAX ) were used for explant cultures since presence or absence of RPE was easily discernable . For adult retinal cultures , P23-P28 animals were euthanized by CO2 asphyxiation and freshly enucleated eyes were dissected rapidly in Hanks buffered saline solution ( HBSS ) ( Invitrogen , Carlsbad , CA ) . Extraocular tissue was trimmed off and the cornea and iris were carefully removed . Sclera along with the RPE was gently removed . This was done primarily for two reasons: ( 1 ) In our assays the presence of Sclera/RPE complex significantly reduced the efficacy of drug treatments and , ( 2 ) secreted lactate was not detected from freshly isolated eyecup with intact sclera . Lens was retained to keep the sphericity of the retina for uniform access to the medium . Explant medium consisted of Neurobasal-A , 0 . 2% B27 supplement , 0 . 1% N2 supplement , 0 . 1% Glutamax and penicillin/streptomycin ( all Invitrogen ) . Retinae were incubated in freshly prepared explant medium constantly supplied with 95% O2 + 5% CO2 ( Medical Technical Gases ) at 37°C in a roller culture system ( B . T . C Engineering , Cambridge , UK ) for indicated times . At the end of incubation period , the lens was removed and the retinae were quickly rinsed with prewarmed Krebs’ Ringers medium ( 98 . 5 mM NaCl , 4 . 9 mM KCl , 2 . 6 mM CaCl2 , 1 . 2 mM MgSO4 , 1 . 2 mM KH2PO4 , 26 mM NaHCO3 , 20 mM HEPES , 5 mM Dextrose ) saturated with 95% O2 . Retinae were again incubated in 0 . 5 mL Krebs’ Ringers medium for 30 min in roller culture with 95% O2 supplied . The supernatant and retinae were rapidly frozen separately at the end of the experiment . DMSO was used as vehicle control for water-insoluble solutes . Sodium oxamate or sodium azide was dissolved in the medium . Equimolar amount of sodium chloride was used as control for osmotic pressure , a colligative property . For +RPE experiments , the extraocular tissue was trimmed off , cornea and iris removed and the eyecups were incubated in the explant culture medium . At the end of the incubation , the RPE/sclera complex was removed along with the lens and the neural retina was incubated in the oxygenated Kebs’s Ringers medium for 30 min as described earlier to assay secreted lactate . Thus , our experiments assess the effect of RPE/sclera complex on the ability to produce lactate by neural retina . Sodium Azide ( 20 mM , Sigma-Aldrich ) , Sodium Oxamate ( 50 mM , Sigma-Aldrich ) , FX11 ( 10 μM , Calbiochem , San Diego , CA ) , BMS 536924 ( 5 μM , Tocris , Minneapolis , MN ) , Afatinib ( 5 μM , Selleckchem , Houston , TX ) , Dovitinib/TKI258 ( 5 μM , Selleckchem ) , Dasatinib ( 5 μM , Selleckchem ) , PD173074 ( 5 μM or 20 μM , Selleckchem ) , Actinomycin D ( 5 μM , Sigma-Aldrich , St . Louis , MO ) , FGF2 ( 2 μg/mL , Cell Signaling , Danvers , MA ) . BL/6J retinae without RPE were homogenized in Lysis buffer ( 5 mM HEPES , 1 mM DTT , 1 mM ATP , 5 mM MgCl2 , 1% glycerol , Complete Protease Inhibitor ( Roche ) and PhosStop phosphatase inhibitor ( Roche , Basel , Switzerland ) . Immunoprecipitation was carried out using rabbit anti-PKM2 and rabbit IgG isotype control followed by sheep anti-rabbit-conjugated Dynabeads ( Life Technologies ) . Immunoprecipitates were boiled and loaded on 10% SDS-PAGE gels followed by transfer on Hybond nitrocellulose membranes ( GE Amersham , Amersham , UK ) . Membranes were blocked with 5% non-fat milk in 1X Tris Buffered Saline +0 . 1% Tween-20 . A conformation-specific mouse-anti rabbit secondary and HRP-conjugated goat-anti-mouse ( Jackson Immunoresrearch , 1:10 , 000 ) tertiary antibodies were used followed by Enhanced Chemiluminescent ( ECL ) detection using substrate from GE Amersham . Enucleated eyes were fixed overnight at 4°C in 4% formaldehyde . The eyes were passed through an increasing concentration of sucrose ( 5% , 15% , 30% ) followed by equilibration in 1:1 30% sucrose: OCT ( Sakura Finetek , Torrance , CA ) and frozen on dry ice . Eighteen micron cryosections were cut using a Leica CM3050S cryostat . Antibodies used are listed in Supplementary file 3 . Heat-mediated antigen retrieval at pH 8 was carried out . For HRP staining , Cell and Tissue staining kit ( R and D systems ) was used . Confocal images were acquired on Zeiss LSM710 or LSM780 inverted microscope . The intensity and pixel saturation were calibrated for inner and outer segment label ( mGFP ) so that details in these cellular features were retained . Thus , due to intense signal of mGFP in the outer segments , the labeling in other cells of the inner retina seems variable and less bright despite electroporation known to target these cells . Images were processed on ImageJ . Maximum intensity projections are depicted . Colocalization was confirmed by individual merges of coplanar sections along the z-axis . For IS/OS length measurements , the orthogonal projections of sections were used . The projections spanning the entire IS/OS volume ensure changes due sectioning angle have a minimal effect . Multiple quantifications across the electroporated field were done for at least three retinae . Expression by IHC was confirmed in both CD1 ( albino ) and BL/6J ( pigmented ) mice . Sclera and RPE were preserved in electroporated eyes to ensure that outer segments were not ripped during the dissections . For all procedures involving antibodies , multiple antibodies were sourced and tested whenever possible ( Supplementary file 3 ) . Previously published antibodies were included and cross-verified with other commercially available antibodies . IHC data were always cross-verified with RNA ISH . In situ hybridization was carried out as described earlier ( Blackshaw et al . , 2001 ) . Probe sequences are presented in Supplementary file 4 . For Pfkfb1 , Pfkfb2 , Pfkfb4 , Srsf3 and Ptbp1 , tyramide amplification ( Perkin , Waltham , MA ) was used . Bright-field images were acquired on Nikon Eclipse E1000 microscope . For ATP estimation , individual retinae were rapidly frozen in liquid nitrogen at the end of the assay . ATP was measured using ATP bioluminescence kit CLS II ( Roche/Sigma-Aldrich ) . For secreted lactate estimation the retinae were incubated in Krebs’ Ringers medium after indicated treatments . The supernatant from above was used with Lactate assay kit ( Eton Bioscience , San Diego , CA ) . Amount of lactate produced in 30 min was assayed . Intracellular lactate was estimated for AAV-transduced retinae because a large number of mice had to be injected and screened for complete , edge-to-edge infection . Thus , infected retinae at specific age were harvested and frozen as they became available . All these retinae were harvested 4–5 hr after lights were turned on in the facility to minimize variability due to possible cyclical diurnal changes in metabolism . Two to three retinae were pooled into a group and frozen together . Five such groups ( n = 5 ) were used for assaying lactate after AAV-mGFP and AAV-TIGAR infection . The retinae were homogenized with the Lactate Assay buffer ( Fluorometric Lactate Assay kit , abcam , Cambridge , MA ) . A small aliquot was removed for protein estimation and subsequent immunoblotting and the remainder was passed through 10 kDa protein filtration column ( abcam ) to remove proteins and thus minimize interference due to endogenous lactate dehydrogenase in the lactate assay . For protein estimation , Qubit protein assay ( Invitrogen ) was used since it is not affected by the presence of detergents in the Lactate Assay buffer . NADP and NADPH was assayed using Fluoro NADP/NADPH kit ( Cell Technology , Fremont , CA ) following manufacturer’s instructions . The quantifications for NADP and NADPH were made separately and thus represent different retinae and treatments . Histochemistry on fresh and unfixed retinal tissue was carried as described earlier for brain tissue ( Ross et al . , 2010 ) . The assay relies on the ability of functional cytochrome oxidase to catalyze oxidative polymerization of 3 , 3'-diaminobenzidine ( DAB ) ( an electron donor ) to brown indamine product . Succinate dehydrogenase assay is based on the ability of this enzyme to oxidize supplied succinate and in turn reduce a ditetrazole ( NBT ) to dark blue diformazan using phenazine methosulfate ( PMS ) , an intermediate electron carrier . Explants were cultured with indicated drug or DMSO for 5 hr followed by 1 mM EU ( Life Technologies ) with the drug for additional 2 . 5 hr . The retinae were fixed , cryosectioned and processed for label detection using Click chemistry reagents ( Life Technologies ) . RNA was isolated using TRIzol reagent ( Life Technologies , Carlsbad , CA ) from 3 to 4 retinae . Two μg RNA was subjected to cDNA synthesis using SuperScript III reverse transcriptase and random hexamers . QPCR was performed using power SYBR Green PCR Master mix ( Applied Biosystems , Foster City , CA ) on a 7500 Fast Real-Time PCR System ( Applied Biosystems ) . Primer sequences are provided in Supplementary file 5 . Rpl13a was used as internal reference and freshly isolated retinal tissue was used as calibrator sample . Expression ratio was calculated using 2-ΔΔCt method . For each target gene , three technical replicates were simultaneously assayed to arrive at the average value for a biological replicate . Mean of three biological replicates was used to derive the Ct value of each target . P0 CD1 mice were electroporated with Rho-dsRed plasmid which encodes for dsRed , driven by bovine rhodopsin promoter , which results in retinas with patches of dsRed expression only in rod photoreceptors ( Matsuda and Cepko , 2004 ) . Once they reached adulthood , mice were then euthanasized via CO2 asphyxiation and the retinas were rapidly removed . The retinas were incubated for 5 min at 37°C in Hank’s Balanced Salt Solution ( HBSS ) supplemented with 10 mM HEPES and 5 mM EDTA and then gently triturated with a P1000 . The dissociated retina was allowed to settle on sylgard-coated petri dishes . Rods expressing the dsRed reporter were identified by their red fluorescence using an inverted microscope and hand-pipeted directly into lysis buffer , and their cDNA amplified using the previously described protocol that utilizes oligo dT priming ( Goetz and Trimarchi , 2012 ) . Data collection was from non-randomized experiments . The primary experimenters were not blinded to treatments . No statistical methods to predetermine sample size were employed . No assumptions for potential outliers were made and hence all data points were included in analyses and depicted . Normality of data distribution was tested using D’Agostino-Pearson omnibus test . Non-parametric statistics were used when Gaussian distribution of data points could not be obtained . p-value denoted as: Not significant ( NS ) , p>0 . 05; *p≤0 . 05; **p≤0 . 01; ***p≤0 . 001; ****p≤0 . 0001 . | Living cells need building materials and energy to grow and carry out their activities . Most cells in the body use sugars like glucose for these purposes . In a process known as glycolysis , cells break down glucose into molecules that are eventually converted to carbon dioxide and water to form the chemical ATP – the cellular currency for energy . Developing cells that have not yet fully specialized , and rapidly dividing cells , like cancer cells , consume large amounts of glucose via aerobic glycolysis ( also known as the Warburg effect ) as they require high levels of energy and building materials . As cells become more specialized and divide less often , they have a reduced need for building blocks , and adjust their consumption and breakdown of glucose accordingly . One exception is the photoreceptor cells , found in the light-sensitive part of our eyes . Although these specialized cells do not divide , they still need a lot of energy and building blocks to constantly renew their light-sensing and processing structures , and to capture and convert the information from the environment into signals . Previous research has shown that the eye also uses the Warburg effect . However , until now , it was not known whether the photoreceptors or other cells in the eye carry out this form of glycolysis . Using genetic tools , Chinchore et al . analysed how the photoreceptor cells in mice used glucose . The experiments demonstrated that the photoreceptors do indeed carry out the Warburg effect . Chinchore et al . further discovered that the Warburg effect is regulated by the same key enzymes and signalling molecules that cancer cells use . This indicates that specialized cells like photoreceptors might choose to retain certain metabolic features of their precursor cells , if they need to . These findings provide new insight into how photoreceptors use glucose . The next step will be to understand how aerobic glycolysis is regulated in healthy eyes as well as in eyes that are affected by degenerating diseases , which may ultimately lead to new ways of treating blindness . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2017 | Glycolytic reliance promotes anabolism in photoreceptors |
Kinesin-12 motors are a little studied branch of the kinesin superfamily with the human protein ( Kif15 ) implicated in spindle mechanics and chromosome movement . In this study , we reconstitute full-length hKif15 and its microtubule-targeting factor hTpx2 in vitro to gain insight into the motors mode of operation . We reveal that hKif15 is a plus-end-directed processive homotetramer that can step against loads of up to 3 . 5 pN . We further show that hKif15 is the first kinesin that effectively switches microtubule tracks at intersections , enabling it to navigate microtubule networks , such as the spindle . hKif15 tetramers are also capable of cross-linking microtubules , but unexpectedly , this does not depend on hTpx2 . Instead , we find that hTpx2 inhibits hKif15 stepping when microtubule-bound . Our data reveal that hKif15 is a second tetrameric spindle motor in addition to the kinesin-5 Eg5 and provides insight into the mechanisms by which hKif15 and its inhibitor hTpx2 modulate spindle microtubule architecture .
The human mitotic spindle is a microtubule-based protein machine with bipolar geometry that mediates chromosome segregation ( Dumont and Mitchison , 2009 ) . Initial assembly of the spindle requires the separation of centrosomes by the homotetrameric plus-end-directed molecular motor Kif11 ( Eg5 ) —a member of the kinesin-5 family ( Tanenbaum and Medema , 2010 ) . In vitro reconstitution experiments reveal that Kif11 can drive the outward sliding of anti-parallel microtubules thus providing a mechanistic basis for centrosome separation ( Kapitein et al . , 2005 ) . Although essential for centrosome separation , Kif11 is not required to maintain subsequent spindle bipolarity due to the compensatory activity of the Kinesin-12 Kif15 ( hKlp2 ) ( Tanenbaum et al . , 2009; Vanneste et al . , 2009 ) . Furthermore , overexpression of hKif15 can overcome the absolute requirement for hKif11 in centrosome separation ( Tanenbaum et al . , 2009; Sturgill and Ohi , 2013 ) . This functional redundancy has led to a model in which hKif15 , like Kif11 , can slide apart anti-parallel microtubules . Kinesin-12 proteins from Xenopus ( xKlp2 ) ( Wittmann et al . , 1998 ) , sea urchin ( Rogers et al . , 2000 ) , and rat ( Liu et al . , 2010 ) are dimeric plus-end-directed motors that are targeted to the spindle by the microtubule-associated protein ( MAP ) Tpx2 ( targeting protein for xKlp2 ) ( Wittmann et al . , 1998 ) . hKif15 is also recruited to the spindle microtubules by hTpx2 in human cells ( Tanenbaum et al . , 2009; Vanneste et al . , 2009 ) with the interaction between dimeric hKif15 and hTpx2 hypothesised to enable the motor to cross-link and slide anti-parallel microtubules ( Tanenbaum et al . , 2009 ) . However , this model has been challenged by recent cell-biological studies showing that hKif15 localizes to kinetochore ( k ) -fibres ( parallel microtubule bundles ) and contributes to the generation of forces that counter those generated by hKif11 ( Sturgill and Ohi , 2013; Vladimirou et al . , 2013 ) . Understanding how hKif11 and hKif15 cooperate during mitosis has important implications for cancer therapy because the clinical efficacy of hKif11 inhibitors , currently used as monotherapy , has proven largely disappointing ( Rath and Kozielski , 2012 ) . hKif15 is therefore emerging as a potentially important therapeutic target ( Groen , 2013 ) . To understand how hKif15/hKif11 operate within the spindle , we will require a detailed mechanistic understanding of how each motor interacts with microtubules and generates force . Such information is already available for Kif11 . In this study , we provide the first insight into the properties of the Kinesin-12 family motor hKif15 .
Full-length human His6-Kif15 ( hKif15 ) , His6-Kif15-eGFP , and His6-Tpx2 ( hTpx2 ) were expressed in insect SF9-cells and the recombinant proteins purified to near homogeneity by sequential affinity chromatography using a cation-exchange and a Co-NTA matrix ( Figure 1A ) . Analysis of His6-hKif15 on native PAGE revealed that the protein migrates at ∼730 KDa ( Figure 1B ) . Given the predicted molecular weight of His6-hKif15 ( 164 . 8 kDa ) our data suggest the presence of hKif15 tetramers , although hydrodynamic analysis of the frog ( xKlp2 ) and Sea urchin ( KRP180 ) Kif15 orthologues indicated that the motor is dimeric with a sedimentation coefficient of 8 . 1S and 8 . 3S respectively ( Wittmann et al . , 1998; Rogers et al . , 2000 ) . To further characterise human Kif15 , we subjected His6-hKif15 to glycerol gradient ultracentrifugation on 5–40% glycerol gradients . At physiological ionic strengths ( 35 mM sodium phosphate buffer , 0-150 mM NaCl ) both His6-hKif15 and His6-hKif15-eGFP appear to be monodispersed and have apparent sedimentation coefficients of ∼12S ( Figure 1C ) . However , increasing the salt concentration to 300 mM NaCl converts hKif15 into a species that runs at ∼8S ( Figure 1C ) . Taken together , our data show that hKif15 exists as a tetramer at physiological ionic strength that can be forced to dissociate into a dimer at high ionic strength . To further confirm that hKif15 can form tetramers we performed a size-exclusion chromatography combined with multi angle light scattering ( SEC-MALS ) , which allows determination of the absolute molecular weight . This analysis confirmed the presence of tetrameric hKif15 motors ( molecular weight of 744 . 4 ± 14 . 1 kDa ) ( Figure 1D , 1st peak , blue line ) . We could also detect a dimeric population of motors ( molecular weight of 360 . 1 ± 6 . 5 kDa ) indicating that there is some degree of complex dissociation in solution under these conditions ( Figure 1D , 2nd peak , orange line ) . 10 . 7554/eLife . 01724 . 003Figure 1 . Kif15 is a tetramer . ( A ) Coomassie stained SDS-PAGE gel of purified His6-hTpx2 , His6-hKif15 , and His6-hKif15-GFP . ( B ) Tetrameric His6-hKif15 on a 4–16% Native-PAGE gel stained with coomassie . Calculated molecular weight: His6-hKif15: 165 kDa . ( C ) Coomassie stained SDS-PAGE gels of fractions 1–23 out of 25 from 5–40% glycerol gradients loaded with either ∼5 µg His6-hKif15 or His6-hKif15-eGFP at different salt concentration ( see table ‘below’ that summarises the apparent S values of hKif15 ( eGFP ) at different salt concentrations ) . ( D ) Elution profile ( grey line , A280 , left y-axis ) from a size-exclusion chromatography ( SEC ) run with subsequent multi angle light scattering ( MALS ) analysis . Outcome of the MALS analysis for the peaks is presented in coloured lines ( blue-hKif15 tetramer , orange hKif15 dimer , MW , right y-axis ) . Molecular weight ± uncertainty is given above each peak . Please note that the presence of dimeric hKif15 species is specific to the protein preparation used in this experiment only ( ‘Materials and methods’ ) . Standard preparations do not contain any significant proportion of dimeric hKif15 ( see panels B and C this Figure ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01724 . 003 Next , we used total internal reflection fluorescence ( TIRF ) -microscopy , to observe the movement of single hKif15-eGFP motors on polarity-marked microtubules that were stabilised with the non-hydrolysable GTP-analogue GMP-CPP . These experiments were done in buffer conditions ( BRB20 , 50 mM KCl ) that would be expected to preserve hKif15 in a tetrameric state ( from here on hKif15-eGFP refers to single tetrameric motors ) . The majority hKif15 motors move either by directional movement towards the plus-end of individual microtubules ( Figure 2A , ‘P’ ) or by bidirectional 1D-diffusion ( Figure 2A , ‘D’ , Figure 2B , right , Figure 2—figure supplement 1D , right ) along the microtubule lattice . These types of movement are equally likely at all salt conditions tested ( Figure 2—figure supplement 1A ) and single motors can switch between both modes ( Figure 2—figure supplement 1B ) . Processively moving , plus-end-directed motors move at a speed of 137 . 8 ± 4 . 1 nm•s−1 median ± SEM , though we observed velocities up to 550 nm•s−1 ( Figure 2B , left ) . The run length of hKif15 motors is 1 . 9 ± 0 . 09 µm median ± SEM ( max . 9 . 6 µm ) ( Figure 2B , middle ) and their residency time 26 . 3 ± 2 . 8 s median ± SEM ( max . >180 s ) ( Figure 2B , right ) . Processive runs can include pauses up to 20 . 5 s ( median ± SEM: 5 . 0 ± 0 . 4 s ) at a frequency of 0 . 30 ± 0 . 04 pauses•µm−1 mean ± SD ( n = 162 , three different experiments ) without motor-dissociation from the microtubule lattice ( Figure 2—figure supplement 1B , D , middle ) . Additionally , when motors reach the microtubule end , more than 90% remain attached ( Figure 2—figure supplement 1D , left note orange column ) . Motors dwell at the tip for 18 . 8 ± 3 . 5 s median ± SEM ( max . >152 s ) ( Figure 2A , ( asterix ) , Figure 2—figure supplement 1D , left ) and may switch to diffusional movement on the end-proximal microtubule lattice ( Figure 2—figure supplement 1C ) . While the majority of hKif15 motors either moves directionally to the plus-end or diffuses along the lattice , we also observed a small proportion of motors that move processively towards the minus-end of polarity-marked microtubules ( 87 . 5% plus-end-directed motors vs 12 . 5% minus-end-directed motors , n = 504 ) ( Figure 2A , right kymograph , ‘MP’ ) . Minus-end-directed motors move at a speed similar to plus-end-directed motors , but have a significantly reduced run length/residency time and do not undergo directional reversals ( Figure 2C ) . 10 . 7554/eLife . 01724 . 004Figure 2 . hKif15 is a processive tetramer . ( A ) Kymographs showing typical behaviour of eGFP-labelled hKif15 motors on GMP-CPP stabilised microtubules ( processive movement–P , diffusion–D , plus-end dwelling–asterix , processive minus-end directed movement–MP ) . Pictures on the left of each kymograph show orientation of the polarity labelled microtubule . ( B ) Frequency distributions showing kinetic properties of processively moving plus-end-directed eGFP-labelled hKif15 motors . Coloured lines within the column plots are Gaussian or exponential decay fits . Insets show the respective median ± standard error of mean and maximum value of the distribution as well as its sample size . All values are derived from kymographs as shown in ( A ) . ( C ) Frequency distributions showing kinetic properties of eGFP-labelled hKif15 motors that move processively to the minus-end of a microtubule . Coloured lines within the column plots are Gaussian or exponential decay fits . Insets show the respective median ± standard error of mean and maximum value of the distribution as well as its sample size . All values are derived from kymographs as shown in ( A ) . ( D ) Kymographs showing a processively moving hKif15-eGFP motor ( left ) and a diffusive hKif15-eGFP motor ( right ) that photo-bleach in three and four steps respectively , indicating presence of four eGFP molecules in each motor . Plot below the kymograph shows the intensity along motor traces in arbitrary units . Intensity had been locally corrected for background intensity . Horizontal green lines within the graphs indicate the fitted average intensity ( arbitrary units ± standard deviation ) in the respective section of the trace . Photo-bleaching steps were manually defined . Please note that for the last bleaching step in the right kymograph we cannot formally exclude a dissociation event . ( E ) Example traces showing the movement of a single molecule ( hKif15 tetramer ) as a function of time ( 1 ms boxcar filtered ) . The motor steps out of the trap centre ( black dotted line ) until it detaches at variable loads between 1 . 5 and 3 . 5 pN and reattaches ( in the trap centre ) for subsequent movements . Movements can also be bidirectional ( green arrowhead , compare with Figure 2A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01724 . 00410 . 7554/eLife . 01724 . 005Figure 2—figure supplement 1 . TIRF-based analysis of hKif15 motility . ( A ) Kymographs showing the behaviour of hKif15-eGFP motors on GMP-CPP stabilised microtubules in dependency indicated salt concentrations apart from the standard condition used throughout the experiments ( BRB20 , 50 mM KCl , Figure 2A ) ( processive movement—P , diffusion—D ) . ( B ) Kymographs showing moving hKif15-eGFP motors that convert from diffusive ( ‘D’ ) into processive movement ( ‘P’ ) or vice versa and show extensive pauses during processive movement ( horizontal sections of the trace ) . ( C ) Kymographs showing end dwelling hKif15 motors that switch to diffusive movements and roam the end proximal region of the microtubule lattice . ( D ) Additional frequency distributions showing kinetic properties of processively and diffusively moving eGFP-labelled hKif15 motors . Coloured lines within the column plots are Gaussian or exponential decay fits . Insets show the respective median ± standard error of mean and maximum value of the distribution as well as its sample size . ( E ) Kymograph showing additional processive moving hKif15-eGFP motors at different velocities that photo-bleach in two steps , indicating the presence of ( at least ) three eGFP molecules per motor . The speed for each slope-section is given ( please compare with velocity distribution in Figure 2B , left panel ) . Graph below the kymographs show the intensity along motor trace in arbitrary units . Intensity had been locally corrected for background intensity . Horizontal green lines within the graphs indicate the fitted average intensity ( arbitrary units ± standard deviation ) in the respective section of the trace . Photo-bleaching steps were manually defined . Compare with Figure 2D . DOI: http://dx . doi . org/10 . 7554/eLife . 01724 . 00510 . 7554/eLife . 01724 . 006Figure 2—figure supplement 2 . Analysis of bead motility and microtubule-binding in laser trap experiments . ( A ) Poisson statistics for the bead-binding probability of n ≥ 1 ( blue ) and n ≥ 2 ( orange ) motors fitted to data from a hKif15 dilution series ( brown ) . Number of analysed beads per concentration >50 except for the highest where n = 30 ( B ) Velocity frequency distributions of hKif15-linked polystyrene beads at zero load followed by DIC microscopy . Orange line within the column plot is a Gaussian fit . Inset shows the respective median ± standard error of mean . Dotted blue line is the TIRF-derived velocity distribution fit from Figure 2B copied for the purpose of comparison . ( C ) Poisson statistics for the bead-binding probability of n ≥ 1 ( blue ) and n ≥ 2 ( orange ) motors fitted to data from a kinesin-1 dilution series ( black ) . Number of analysed beads per concentration n = 25 except for the three lowest where n = 50 . DOI: http://dx . doi . org/10 . 7554/eLife . 01724 . 006 To confirm that motile hKif15 motors are tetramers , we imaged hKif15-eGFP motors under conditions that caused photobleaching and measured the change in fluorescence intensity with time ( ‘Materials and methods’ ) . This analysis showed that both processive and diffusive motors bleach in up to four steps , indicating the presence of four eGFP molecules per motor ( Figure 2D ) . Furthermore , processive tetramers are evident throughout the velocity distribution ( Figure 2—figure supplement 1E ) . These data are in line with our biochemical analysis ( Figure 1 ) and confirm that the observed hKif15-eGFP motors are indeed tetramers . Within the mitotic spindle , motors will experience varying loads and thus it is essential to understand how a motor responds to force . We therefore investigated the mechanical properties of hKif15 motors using our single-bead laser trap set-up ( Carter and Cross , 2005 ) . Single hKif15 motors were adsorbed onto 550-nm diameter polystyrene beads and steered to the microtubule surface with the laser trap at which point the motor can bind to the lattice ( see ‘Materials and methods’ for details and Figure 2—figure supplement 2A ) . With the trap turned off ( zero-load ) the mean velocity of beads moving on the microtubule was consistent with our measurements of hKif15-eGFP by TIRF microscopy ( Figure 2—figure supplement 2B ) . Next , we held the bead in the trap and allowed the motor to walk along the microtubule , away from the centre of the trap . The motion of the bead reports the stepping actions of the motor . Figure 2E shows two position traces in which hKif15 motors make a number of processive steps along the microtubule before detaching at relatively low loads between 1 . 5 and 3 . 5 pN . We also observed that beads move out the trap before moving back in the opposite direction without detaching ( green arrowhead ) . These events may reflect the minus- and plus-end-directed diffusive motion observed in our TIRF experiments ( Figure 2A ) . Together with our TIRF microscopy , we conclude that hKif15 is a homotetramer that undergoes either diffusive movement along microtubules , processive plus-end-directed stepping over long-distances or short duration movements to the minus-end . Our single molecule TIRF experiments also revealed a remarkable property of hKif15 motors as they approach microtubule intersections . The well-studied kinesin-1 will normally pass intersections and only in rare cases the motor will pause or switch microtubule tracks ( Ross et al . , 2008 ) . In contrast , hKif15 motors show significantly elevated pause and switch events at intersections . Every fifth motor that approaches an intersection switches its microtubule track ( Figure 3A , B; Videos 1 and 2 ) . In general all events—pass , switch , pause , and dissociation at intersections—are equally likely . In this way , hKif15 is able to roam microtubule networks and Figure 3C demonstrates that one tetramer ( see bleaching steps , lower panel ) can indeed navigate along several different microtubules: the motor starts processive movement on microtubule #1 ( trace A ) , eventually dissociates and rebinds to microtubule #2 . On this , the motor diffuses a short distance ( trace B ) and dissociates again , just to rebind to microtubule #1 . It processively follows microtubule #1 ( trace C ) and sequentially switches to microtubules #3 ( trace D ) and #4 ( trace E ) , ending at the plus-end of microtubule #4 without detaching from the microtubule lattice ( s ) once again ( see also Video 3 ) . Thus , Kif15 is a processive motor that is able to navigate microtubule networks by effectively switching microtubule tracks . 10 . 7554/eLife . 01724 . 007Figure 3 . hKif15 effectively switches microtubules at intersections . ( A ) Kymographs showing the processive switch of an eGFP-labelled hKif15 motor at a microtubule–microtubule intersection . The image in the upper left gives an overview of the microtubule positons , bar = 5 µm . The schematic below depicts the microtubules whose line scans are depicted as kymographs on the right . The numbering of the microtubules corresponds to that of the respective kymograph and the assigned letters to the traces within the kymograph . White arrows depict the direction of the line-scan; the dotted black arrow depicts the path of the hKif15 motor . Velocities of single trace-sections are given . Please note that the motor never leaves the microtubule lattice during the switch event ( yellow dotted line between the kymographs ) . ( B ) Quantification of motor behaviour at microtubule intersections . Definitions as following: Pass–motor passes on the same microtubule and continues movement ( may include a pause at the intersection ) ; Switch–motor switches the microtubule and continues movement ( may include a pause at the intersection ) ; Pause–motor pauses for more than 5 s and does not continue movement after intersection ( may dissociate or bleach at some point ) ; Dissociate–motor immediately dissociates at intersection . ( C ) Essentially as in ( A ) , now following a motor along four microtubules . Movement involves two dissociation/re-association ( Kymographs: trace A/trace B , trace B/trace C ) and two microtubule switch events ( trace C/trace D , trace D/trace E ) during switch events the motor never leaves the microtubule lattice ( yellow dotted line between the kymographs ) . Plot below the kymograph shows the intensity along motor traces in arbitrary units . Intensity had been locally corrected for background intensity . Horizontal green lines within the graphs indicate the fitted average intensity ( arbitrary units ± standard deviation ) in the respective section of the trace . Photo-bleaching steps were manually defined and show that the observed motor bleaches in three steps indicating presence of four eGFP molecules . DOI: http://dx . doi . org/10 . 7554/eLife . 01724 . 00710 . 7554/eLife . 01724 . 008Video 1 . hKif15 effectively switches microtubules at intersections . Video of the events summarised in Figure 3A ( hKif15-eGFP—cyan , microtubules—yellow ) . The switching hKif15-eGFP motor is marked by the cyan arrowhead . DOI: http://dx . doi . org/10 . 7554/eLife . 01724 . 00810 . 7554/eLife . 01724 . 009Video 2 . hKif15 effectively switches microtubules at intersections . Video of another hKif15-eGFP motor switching polarity-marked microtubules ( hKif15-eGFP—cyan , seed—red , microtubule extension—yellow ) . The switching motor is marked by the cyan arrowhead . Please note that the motor moves for some time towards the minus-end of the microtubule it switched onto and subsequently changes direction towards the plus-end . DOI: http://dx . doi . org/10 . 7554/eLife . 01724 . 00910 . 7554/eLife . 01724 . 010Video 3 . hKif15 effectively switches microtubules at intersections . Video of the events summarised in Figure 3C ( hKif15-eGFP—cyan , seed—red , microtubule extension—yellow ) . The switching hKif15-eGFP motor is marked by the cyan arrowhead . DOI: http://dx . doi . org/10 . 7554/eLife . 01724 . 010 The tetrameric organisation of hKif15 would support the hypothesis that the motor drives extensile sliding of anti-parallel microtubule overlaps like the tetrameric Kif11 motor ( Tanenbaum et al . , 2009 ) . To establish , whether tetrameric hKif15 indeed is able to cross-link microtubules on its own , we carried out experiments , in which biotinylated HiLyte-647-labelled GMP-CPP-microtubules were attached to a cover slip followed by addition of short XRhodamin-labelled GMP-CPP-stabilized- ‘transport’ microtubules and 1 . 4 nM tetrameric hKif15-eGFP . By drawing kymographs , we found that hKif15 could mediate the cross-linking of microtubules ( Figure 4 ) on its own , but that there was very little microtubule–microtubule sliding activity . Besides pivoting microtubules , we did observe episodes of bidirectional short duration sliding indicating a tug-of-war ( Figure 4 , right ) and rarely the transport of short microtubules ( Figure 4 , left ) : behaviours that were recently reported for the yeast Kinesin-8 Kip3 ( Su et al . , 2013 ) . However , we have no clear evidence to support the idea that hKif15 can drive continuous extensile sliding of anti-parallel microtubule bundles like hKif11 . This is presumably because the motors frequently pause during processive excursions on the lattice and dwell at the ends for a long period ( Figure 2—figure supplement 1B–D ) . Consistent with this , we only observed discontinuous short distance movement and pivoting of microtubules ( around their ends ) on surface-bound hKif15 motors , but not continuous microtubule sliding in microtubule gliding assays ( Videos 4 and 5 ) . 10 . 7554/eLife . 01724 . 011Figure 4 . hKif15-eGFP can transport short microtubules as a cargo . Kymographs showing examples of processive ( left ) and tug-of-war-type transport ( right ) , with an overview image of the crosslinked microtubules on top . Left: hKif15-eGFP motors drive slow ( 26 nm•s−1 ) processive movement of a cargo microtubule . The moving microtubule ( magenta ) is attached via Kif15 motors ( cyan ) to a substrate microtubule ( yellow ) for which the plus-end is orientated left in the overview image on top and at the bottom of the kymograph . hKif15 motors can be seen stably associated with one end of the cargo microtubule , which must be the plus-end because the motors move in a plus-end direction and pause at the end ( see Figure 2A , Figure 2—figure supplement 1C , see also schematics to the left ) . Thus , the cargo microtubule moves plus-end leading towards the plus-end of the substrate microtubule . That is , the microtubules are parallel . Right: hKif15-eGFP motors drive tug-of-war-type movement , characterised by frequent and rapid direction changes during movement . DOI: http://dx . doi . org/10 . 7554/eLife . 01724 . 01110 . 7554/eLife . 01724 . 012Video 4 . Surface-bound hKif15 does not support continuous sliding of GMP-CPP stabilised microtubules ( white ) over long distances . DOI: http://dx . doi . org/10 . 7554/eLife . 01724 . 01210 . 7554/eLife . 01724 . 013Video 5 . Surface bound hKif15 does not support continuous sliding of GMP-CPP stabilised microtubules ( white ) over long distances . DOI: http://dx . doi . org/10 . 7554/eLife . 01724 . 013 So far our experiments show that hTpx2 is not required for hKif15 to bind microtubules or for the motors processivity and ability to crosslink microtubules ( as a tetramer ) . However , cell-biological experiments showed that hTpx2 is essential for hKif15 to be recruited to the mitotic spindle ( Tanenbaum et al . , 2009; Vanneste et al . , 2009 ) . To resolve this contradiction , we first analysed possible complex formation of recombinant hKif15 and hTpx2 in solution by ultracentrifugation in glycerol gradients . With nanomolar concentrations of each protein , we could not observe any complex formation over a range of different salt conditions ( 50 mM HEPES pH7 . 5 , 0/150/300/450 mM NaCl [data not shown] and 35 mM sodium phosphate buffer [Figure 5A] ) . However , a 30-fold excess of hTpx2 dimers ( 2 . 5 µM hTpx2 [dimer] + 80 nM hKif15 [tetramer] ) allowed the formation of a hKif15-hTpx2 complex of ∼14 . 0S , which included approximately one third of the provided hKif15 ( Figure 5A , lower gels ) . The dissociation constant of the hKif15–hTpx2 complex is likely to be in the µM range arguing that its formation in vivo is less likely . In contrast , using a microtubule sedimentation assay , we observed that at low nM concentrations , hTpx2 ( 30 nM , dimer ) does enhance the affinity of hKif15 ( 15 nM , tetramer ) binding to GMP-CPP stabilised microtubules when in the presence of the non-hydrolysable ATP analogue AMP-PNP ( the fitted Kd increasing ∼twofold from 993 nM to 490 nM; Figure 5B ) . This observation could explain why cell-biological experiments show that association of hKif15 with the mitotic spindle depends on hTpx2 ( Tanenbaum et al . , 2009; Vanneste et al . , 2009 ) . It is , however , unclear why hTpx2 further increases the microtubule binding affinity of a motor that is already highly processive . 10 . 7554/eLife . 01724 . 014Figure 5 . hTpx2 inhibits hKif15 motility by increasing its microtubule affinity . ( A ) Formation of a stable hKif15/hTpx2 complex occurs only at low µM concentration of Tpx2 . Coomassie stained SDS-PAGE gels showing the first 23 of 25 fractions of a 5–40% glycerol gradient in 35 mM sodium phosphate buffer loaded with the indicated purified proteins alone and in mixture at the indicated concentrations . While hKif15 and hTpx2 do not form a stable complex at nanomolar concentrations , a vast excess of 2 . 5 µM dimeric hTpx2 drives formation of a hKif15/hTpx2 complex of 14 . 0S . ( B ) Above: coomassie stained SDS-PAGE gel of a typical microtubule co-sedimentation experiment of 15 nM tetrameric hKif15 in the presence or absence of 30 nM dimeric hTpx2 at different concentrations of taxol-stabilised microtubules ( SN–supernatant , P–pellet ) . Below: Quantification of pelleted tubulin and bound hKif15 shown above . Crosses indicate the SD of the average from three independent experiments . Deviation in tubulin concentration is due to partial microtubule instability in phosphate buffer at low tubulin concentration and the microtubule stabilising effects of hTpx2 ( compare pelleted tubulin at 0 . 25 µM with and without hTxp2 ) . ( C ) Kymographs show the effect of 5 nM dimeric hTpx2 on the motility of 5 nM tetrameric hKif15 on GMP-CPP stabilised microtubules . Graph below left shows the fraction of motile and static motors normalised to overall microtubule length and subtracted by the fraction of static motors in the control , which sets motile motors in control to 1 . Error bars show SD of three independent experiments . Graph below right shows the velocity distribution of motile motors in the presence of hTpx2 , coloured lines are Gaussian fits revealing a bimodal distribution , compare with Figure 2B . ( D ) Side-by-side comparison of hKif15 ( blue ) and Drosophila kinesin-1 ( green ) stepping in the absence and presence of hTpx2 in the laser trap . Stepping traces for each kinesin are from the same bead before flow-in of 36 nM hTpx2 ( left trace ) , after flow-in of hTpx2 ( right trace ) . The asterix in the above right trace marks the deliberate movement of the stage , showing motor maintained attachment to the microtubule . DOI: http://dx . doi . org/10 . 7554/eLife . 01724 . 014 To address the impact of hTpx2 on moving hKif15 tetramers , we added hTpx2 in equimolar ( i . e . , 5 nM ) concentration to our hKif15-eGFP TIRF setup . Addition of hTpx2 decreased the proportion of motile motors on microtubules by 70% . The remaining motors still showed both directional and diffusive movement , indicating that inhibition by hTpx2 is not selective for either type of movement ( Figure 5C , kymographs and left graph below ) . Further , the velocity distribution of remaining processive motors ( compared to Figure 2A ) now shows a bimodal distribution with a high proportion of slow motors ( 42 . 1 ± 1 . 7 nm•s−1 median ± SEM ) and fewer motors at the median speed of hKif15 motors in the absence of hTpx2 ( 114 . 9 ± 3 . 2 nm•s−1 median ± SEM ) ( Figure 5C , right graph below ) . This may reflect the higher probability of a fast moving motor from engaging hTpx2 molecules that are bound on the microtubule lattice . To further confirm the inhibition of hKif15 by hTpx2 and to estimate the functional impact of hTpx2 inhibition on hKif15 functions beside movement , we included hTpx2 in our optical trap experiment . To do this , we captured a hKif15-linked bead and confirmed that it underwent stepping on the microtubule ( Figures 5D , 1 . blue trace ) . We then exchanged the buffer to allow 18 nM dimeric hTpx2 access to the hKif15-microtubule complex and again recorded the bead position . The hKif15 motor could no-longer step processively , although it remained bound to the microtubule lattice ( 4/5 cases; Figures 5D , 2 . blue trace ) . To rule out that the bead had not simply detached , we moved the microscope stage and observed an increase in force indicating an intact microtubule-motor connection ( see silver asterisk in Figure 5D ) . During this period , the bead can maintain microtubule attachment at forces that—in the absence of hTpx2—would have forced dissociation of the motor ( compare also with Figure 2E ) . Thus , hTpx2 increases the force-holding capability of microtubule-bound hKif15 . We can rule out a generic ‘roadblock’ inhibition-mechanism: Firstly , kinesin-1 motility is unaffected by the presence of hTpx2 ( n = 5; Figure 5D , green traces ) . Secondly , hTpx2 already inhibits hKif15 motility at low nanomolar concentrations when only single molecules are present on the lattice . Kinesin-1 motility in the presence of hTpx2 also shows that inhibition of bead motility in case of hKif15 is not simply mediated by hTpx2-bead-microtubule interactions . Thus , hTpx2 is a selective inhibitor of hKif15 motor stepping presumably by forming stable trimeric hKif15-hTpx2-microtubule complexes .
In this study , we reveal that hKif15 is a second tetrameric spindle motor in addition to the tetrameric Kinesin-5 hKif11 ( Eg5 ) . This finding is in line with initial cell-biological experiments that concluded that these two motors are redundant with Kif15 becoming essential for spindle assembly when hKif11 is inhibited ( Tanenbaum et al . , 2009; Vanneste et al . , 2009 ) . Indeed , hKif15 , like Kif11 , is only able to step under low loads in the 1–3 pN range ( Korneev et al . , 2007 ) . However , our data show that Kif15 motors have a number of biophysical properties that distinguish it from Kif11: while both motors are able to undergo plus-end-directed movement or 1-D diffusion ( Kwok et al . , 2006; Kapitein et al . , 2008; Figure 6 , ‘A’ ) , the median velocity of hKif15 is 10-fold higher than that of Kif11 and processive runs have a fourfold longer duration ( Korneev et al . , 2007 ) . In the case of hKif15 , both these movement types occur at the same ionic strength . In contrast , processive and diffusive movement by Kif11 is exclusive and depends on the ionic strength ( Kapitein et al . , 2008 ) . Additionally , to our knowledge , pauses during processive runs have not been reported for Kif11 . 10 . 7554/eLife . 01724 . 015Figure 6 . Schematic model summarising the biophysical and biochemical properties of hKif15 and illustrating how the motor may operate within a k-fibre ( parallel microtubule bundle ) . A–hKif15 can move uni-directionally towards the plus-end ( minus-end-directed motion can also occur albeit with lower frequency ) by processive stepping or by bi-directionally diffusion along the microtubule lattice . B–hKif15 can switch between microtubule tracks . ( box ) Model for the sequence of events during switch or pass movements at intersections via a bridge structure of the hKif15 tetramer that resolves into a pass ( 1 ) or switch event ( 2 ) depending on which motor domain pair detaches first from the microtubule lattice . C–hKif15 has a significant plus-end dwell time and therefore might modulate plus-end dynamics like other k-fibre motors ( Stumpff et al . , 2012 ) . D–Being a tetramer , hKif15 can crosslink two microtubules and potentially resist sliding of microtubules within the fibre . Note that this link is dynamic as the motor still can step or diffuse within a bundle . ( box ) Once hKif15 and hTpx2 have formed a complex on the microtubule lattice , this crosslink could become static , thereby forming a fixed structural link in the fibre , which can sustain higher loads than hKif15-only crosslinks . E–hKif15 powers transport of ( small ) microtubule fragments . DOI: http://dx . doi . org/10 . 7554/eLife . 01724 . 015 The uniqueness of hKif15 is also reflected in the motors ability to track-switch during processive movement—a property shared with Dynein ( Ross et al . , 2008 ) . It is easy to comprehend how dynein switches between microtubule tracks , as this motor is capable of making variable step sizes , significantly larger than 8 nm ( Reck-Peterson et al . , 2006 ) . However , kinesins step with a step size of ∼8 nm ( Svoboda et al . , 1993; Valentine et al . , 2006; Yardimci et al . , 2008; Huckaba et al . , 2011; Jannasch et al . , 2013 ) suggesting that it is unlikely for a single hKif15 motor domain-pair to mediate switching without leaving the lattice . In this regard , we show that within our 500 ms time resolution single motors do not leave the lattice during switching ( Figure 3 ) . Furthermore , switching events often include a pause at the intersection ( Figure 3C , traces ‘D’ and ‘E’ ) . As hKif15 is a motile tetramer , we propose that a switch starts with attachment of the second , free head-pair to the lattice of the intersecting microtubule , while the first , engaged head-pair of the same molecule stays attached to the old track . This hKif15 bridge can now pause at the intersection or resolve by detachment of the second domain-pair into a pass or by detachment of the firstdomain-pair into a switch event ( Figure 6 , ‘B’+box ) . Microtubule switching is a feature that is very helpful to navigate through the complex microtubule arrangement within the k-fibre . Thereby each switch event is equivalent to a new independent run on a new microtubule , so that a single molecule can travel a distance of ‘median run-length × ( nswitches+1 ) ’ as we have shown in Figure 3C: traces C-E sum up to 7 . 4 µm run length in total , with single traces at 1 . 6 , 3 . 7 and 2 . 1 µm . Thus , even with a moderate median run length of 2 microns , a single molecule would have a high probability of reaching the plus-end ( kinetochores ) of the k-fibre , regardless of any discontinuity within the fibre or roadblocks owing to structural microtubule-associated elements ( Figure 6 , ‘B’+box ) . In contrast , short running , non-switching hKif11 , that additionally is constantly gathered at the spindle poles by a dynein-dependent minus-end-directed flux ( Ma et al . , 2010 ) , cannot travel far enough into the k-fibre network to reach the kinetochore/plus-end . In vivo data backs up this hypothesis , as hKif11 localises mainly to the spindle pole and fades towards the spindle midzone , while hKif15 is localised uniformly along the k-fibres ( Vanneste et al . , 2009; Sturgill and Ohi , 2013 ) . Moreover , Kif15 motors also modulate the capacity of k-fibre microtubule bundles to generate an inward-directed force within the spindle ( in opposition to hKif11 ) ( Sturgill and Ohi , 2013 ) . One possibility is that , due to their prolonged end dwell time , hKif15 motors may influence plus-end dynamics at kinetochores ( Figure 6 , ‘C’ ) . Such is the proposed role of the Kinesin-8 Kif18a , which is reported to dampen microtubule dynamics in vitro ( Du et al . , 2010; Stumpff et al . , 2012 ) . Tetrameric hKif15 motors , like Kif11 , can form cross-links between microtubules , we show that hKif15 crosslinks are dynamic as we observed limited tug-of-war movements of crosslinked cargo microtubules . Such cross-links may be reinforced by inhibitory hTpx2 ( see below ) turning dynamic hKif15 crosslinks into static structural hKif15/hTpx2 crosslinks that can resist higher forces than hKif15 only crosslinks ( Figure 6 , ‘D’+box ) . Our data does not , however , provide any evidence that hKif15 drives a Kif11-like continuous extensile sliding of anti-parallel microtubules in vitro , though we cannot rule out that in vivo other additional factors are involved in such an activity . This factor is unlikely to be hTpx2 since our data shows this protein to inhibit the stepping of hKif15 . How this mechanism is regulated within the spindle will be an important future work , but the fact that hTpx2 and hKif11 are constantly transported to the spindle pole ( Ma et al . , 2010 ) might hint to a partition of the spindle into a hKif11 active zone around the poles and a hKif15 active zone towards the spindle equator . Overall , our data provide the first insight into the biochemical and biophysical properties of the full-length human Kinesin-12 hKif15 . We reveal that hKif15 is a distinct class of kinesin that assembles into stable tetramers , which are highly processive , can navigate microtubule networks by switching track and form high load-bearing microtubule–microtubule crosslinks when bound to the regulatory factor hTpx2 . These data shed new light on the mechanism by which hKif15 motors control spindle and chromosome mechanics .
The hKIF15 ORF was amplified in five 800- to 1000-bp sized fragments from a cDNA library ( derived from hTERT immortalised retinal pigment epithelial ( hTERT-RPE1 cells ) ) and subsequently joined by overlap extension PCRs using Pfu Ultra AD polymerase ( Agilent , Stockport , UK ) . The complete hKIF15 ORF was cloned via AseI/NotI into a modified ( bases 2391–2426 of the pIEx/Bac1-vector coding the Strep-Tag II were replaced by the His6/TEV-cleavage site element of p11 [DNASU Plasmid ID: EvNO00085126 , DNASU Plasmid Repository at Arizona State University] , flanked by a 5′ NcoI and a 3′ NdeI site . [gift form Miho Katsuki , Riken , Japan] ) pIEx/Bac1-vector ( MERCK , Darmstadt , Germany ) opened with NdeI/NotI . For hKIF15-eGFP an AscI-site was introduced in front of the NotI-site and the eGFP inserted via AscI/NotI before the hKIF15 ORF was inserted as above . We observed that the hKIF15 ORF is toxic to Escherichia coli cells when antibiotics other than ampicillin ( e . g . , kanamycin , gentamycin ) are used , so viral genome generation by the Invitrogen Bac-to-Bac system ( Life Technologies , Paisley , UK ) using DH10BAC cells is not possible . The hTPX2 ORF was amplified from I . M . A . G . E . 3509275 ( ATCC clone MG1537 ) and cloned into the pFastBacM13 vector ( MPI-CBG , Dresden , Germany ) via SpeI/SalI . Assembly of viral genomes was carried out according to manufacturer protocols and transfection competent baculovirus particles were generated and used for transfection of 500 ml–1 L SF9-cells expression cultures according to ( Wasilko et al . , 2009 ) . Cells were harvested at 250-g in a SLA-3000 rotor ( Thermo Scientific , Waltham , MA , USA ) and resuspended in lysis buffer ( 50 mM HEPES pH 7 . 5 , 150 mM NaCl , 1 . 5 mM MgCl2 , 3 mM EGTA , 5% glycerol , 0 . 1% Tween-20 , 0 . 1 mM ATP , complete protease inhibitor [Roche , Burgess Hill , UK] ) . For hKif15 purification , lysates were cleared at 48k-g in a SS-34 rotor ( Thermo Scientific ) , diluted 1:3 ( to 50 mM NaCl final ) with 50 mM sodium phosphate buffer and adjusted to pH 7 . 0 . Protein was allowed to bind to SP-sepharose ( GE Healthcare , Little Chalfont , UK ) in batch for 2 hr at 4°C . hKif15 was directly eluted by applying a threefold bed volume of 50 mM sodium phosphate buffer pH 7 . 5 , 50 mM NaCl , 1 mM MgCl2 , 5% glycerol , 0 . 05% Tween-20 , 0 . 1 mM ATP onto Talon-beads ( Takara Biotech , Saint-Germain-en-Laye , France ) . Again , protein was allowed to bind 2 hr at 4°C in batch . Talon-beads where washed with 10-fold bed volume of 50 mM sodium phosphate buffer pH 7 . 5 , 300 mM NaCl , 1 mM MgCl2 , 5% glycerol , 0 . 05% Tween-20 , 0 . 1 mM ATP and a 15-fold bed volume of 50 mM sodium phosphate buffer pH 7 . 5 , 100 mM NaCl , 1 mM MgCl2 , 5% glycerol , 0 . 05% Tween-20 , 10 mM imidazole , 0 . 1 mM ATP . Protein was eluted with 50 mM sodium phosphate buffer pH 7 . 5 , 150 mM NaCl , 1 mM MgCl2 , 10% glycerol , 50 mM imidazole , 0 . 1 mM ATP and purity as well as concentration determined by SDS-PAGE against a BSA standard using ImageJ . For each protein preparation , the oligomerisation state and its dispersity have been checked by glycerol gradients or native PAGE . Note that for the SEC-MALS experiment , protein was eluted with 35 mM sodium phosphate buffer pH 7 . 0 , 1 mM MgCl2 , 50 mM imidazole , 0 . 05 mM ATP and snap frozen in the absence of glycerol in 500 µl aliquots and stored in liquid nitrogen . For purification of hTpx2 lysates were cleared and diluted 1:3 ( to 50 mM NaCl ) with 50 mM sodium phosphate buffer pH 7 . 5 as done for hKif15 . Protein was allowed to bind to SP-sepharose in batch for 2 hr at 4°C . Beads were washed with a fivefold bed volume of 50 mM sodium phosphate buffer pH 7 . 5 , 100 mM NaCl , 1 mM MgCl2 , 5% glycerol , 0 . 05% Tween-20 and protein was eluted with a threefold bed volume of 50 mM sodium phosphate buffer pH 7 . 5 , 300 mM NaCl , 1 mM MgCl2 , 5% glycerol , 0 . 05% Tween-20 , 10 mM imidazole onto Talon-beads . Protein was allowed to bind 2 hr at 4°C in batch and beads were washed with a 25-fold bed volume of 50 mM sodium phosphate buffer pH 7 . 5 , 150 mM NaCl , 10 mM imidazole , 1 mM MgCl2 , 5% glycerol , 0 . 05% Tween-20 . Protein was eluted with 50 mM sodium phosphate buffer pH 7 . 5 , 75 mM NaCl , 150 mM imidazole , 10% glycerol . Purity and concentration was determined as above . All protein was snap frozen in 5–20 µl aliquots and stored in liquid nitrogen . Glycerol gradients were essentially carried out as described in McClelland and McAinsh ( 2009 ) in the presence of 1 mM DTT and complete protease protein inhibitor at 4°C for 15 hr over night . We checked for hKif15/hTpx2 interactions in 50 mM HEPES pH 7 . 5 , 1 mM MgCl2 , 1 mM EGTA , 0 . 1 mM ATP , 0/150/300/450 mM NaCl ( 150 mM NaCl also ± free tubulin in ATP or AMP-PNP ) and 35 mM sodium phosphate buffer pH7 . 0 , 1 mM MgCl2 , 1 mM EGTA , 0/150/300/500 mM NaCl , 0 . 1 mM ATP or AMP-PNP . The 5-mlgradient was fractionated by hand in fractions of 200 µl each . The protein was TCA-precipitated and fractions 1–23 , together with 1/2 input , analysed on the same SDS-PAGE gel stained with coomassie ( SimplyBlue , Invitrogen ) . Protein bands were quantified with ImageJ and peak values were determined by fitting a Gaussian distribution to the data using Origin 8 . 5 . Selected proteins from the High/low molecular weight gel filtration calibration kit ( GE Healthcare ) were used in standard calibration runs . Protein samples were concentrated to ∼2 mg/ml in 35 mM sodium phosphate buffer pH 7 . 0 , 1 mM MgCl2 and loaded onto a size exclusion column ( Wyatt , Santa Barbara , CA , USA ) connected to a Dawn Heleos II MALS detector and Optilab T-rEX refractometer ( Wyatt ) . A dn/dc value of 0 . 185 was used for all calculations . Please note that for the SEC-MALS experiment only , protein-purification was modified in that hKi15 was finally eluted in 35 mM sodium phosphate buffer pH 7 . 0 , 1 mM MgCl2 , 50 mM imidazole , 0 . 05 mM ATP and snap frozen in the absence of glycerol , which is likely to be the reason for the appearance of dimeric hKif15 in the SEC-MALS analysis that otherwise are absent in the preparation used in all the other experiments ( as shown by native PAGE and glycerol gradients , see Figure 1B , C ) . Microtubules for the sedimentation assay were grown in presence of 1 mM GTP in BRB80 ( 80 mM PIPES pH 6 . 8 , 1 mM MgCl2 , 1 mM EGTA ) by stepwise increasing the taxol concentration in solution . Polymerised microtubules were washed once with BRB80+Taxol using a Beckman Airfuge and resuspended in 35 mM sodium phosphate buffer pH 7 . 0 , 1 mM MgCl2 , 1 mM EGTA and 1 mM DTT . Samples were mixed in the same buffer at the indicated concentrations ( 16 . 5 nM tetrameric hKif15 , 33 nM dimeric hTpx2 ) and incubated for 20 min at room temperature in the presence of 2 mM ATP or AMP-PNP . Samples were spun for 20 min at 45k rpm in a TLA100 rotor ( Beckman Coulter , High Wycombe , UK ) using thick-wall polyallomer tubes ( Beckman Coulter ) . The supernatants were TCA precipitated and analysed together with the resuspended pellets by SDS-PAGE . Protein bands were quantified using ImageJ . Since taxol stabilised microtubules are less stable in 35 mM sodium phosphate over long time periods and that hTpx2 has a stabilising effect on microtubules , we also quantified the actual amount of precipitated tubulin in order to compare both setups with and without hTpx2 . This gives rise to the horizontal error bars in Figure 5B . Preparation of sialylated coverslips and flow chamber setup ( with double sided tape ) was essentially carried out as described in Bechstedt et al . ( 2011 ) , except for the initial cleaning step with piranha solution , which had been substituted by incubation with 7 . 4% hydrochloric acid at 60°C over night . For single molecule assays , GMP-CPP ( Jena Bioscience , Jena , Germany ) stabilised , polarity-marked microtubules labelled with XRhodamin- and HiLyte-647-linked tubulin ( Cytoskeleton , Denver , CO , USA ) were adsorbed to the glass surface via anti beta tubulin antibodies ( TUB 2 . 1 , Sigma-Aldrich , St . Louis , MO , USA ) after the flow chamber has been blocked with 1% Pluronic F-127 ( Sigma-Aldrich ) and 1 mg/ml kappa casein . The flow chamber was loaded with 1–10 nM His6-hKIF15-GFP ( tetrameric ) in BRB20 ( 20 mM PIPES pH6 . 8 , 1 mM MgCl2 , 1 mM EGTA ) , 50 mM KCl , 2 mM ATP , 0 . 1 mg/ml kappa casein , 80 µg/ml glucose , 40 µg/ml glucose-oxidase , 16 µg/ml catalase , 1 mM DTT , sealed with VALAP ( vaseline , lanolin , paraffin mixed 1:1:1 ) and imaged at 25°C on a Olympus CELLR/TIRF microscope ( Olympus , Southend-on-Sea , UK ) equipped with a ImagEM emCCD camera ( Hamamatsu Photonics , Welwyn Garden City , UK ) , an environmental chamber and a stage-top-incubator ( Okolab , Ottaviano , Italy ) using a 100x NA 1 . 49 objective with 1 . 6x auxiliary magnification . To visualise hKif15–eGFP movement , 3-min time-lapse videos were recorded at 2 frames per second ( fps ) with a 100 ms exposure using a 488 nm laser line . For the bleaching experiments , videos were recorded at 4 fps . The position of polarised microtubules before and after the time-lapse video was determined by capturing a single image with the 561 nm ( 100 ms exposure ) and 640 nm ( 100 ms exposure ) laser lines . In General , kymographs were generated and analysed with ImageJ and the MultipleKymograph plugin ( http://www . embl . de/eamnet/html/body_kymograph . html ) . For the hTpx2 inhibition experiments , equimolar dimeric hTpx2 was directly loaded together with hKif15 into the flow chamber . Imaging was started after a short incubation period of 5 min at room temperature . To prevent a strong dilution of both proteins on the microtubule lattice , a microtubule density of 150 µm tubulin per field of view ( 51 . 2 × 51 . 2 µm ) was not exceeded . For microtubule sliding experiments , biotinylated ( Cytoskeleton ) and HiLyte-647-labelled GMP-CPP-microtubules were adsorbed via streptavidin to a PEG-biotin-coated ( Surface Solutions Switzerland , Dübendorf , Switzerland ) cover slip that has been blocked with 1 mg/ml kappa casein . Short XRhodamin-labelled GMP-CPP-microtubules were added as cargo to the imaging-mix ( as above ) . Again , 3 min time-lapse videos at 2 frames per second ( fps ) were recorded to track hKif15-GFP and cargo microtubule movement . Exposure times were 100 ms with the 488 nm laser line and 150/200 ms with the 561 nm laser . Positions of the HiLyte-647-labelled substrate microtubules were determined by capturing a single image ( 100 ms exposure ) with a 640 nm laser line both before and after the time-lapse video . For the most part , instrumentation and sample preparation was the same as described in Carter and Cross ( 2005 ) with the following changes: motors were non-specifically bound to polystyrene beads ( 550 nm , NIST calibration size standards , Polysciences , Warrington , PA , USA ) by incubation in a solution of 80 mM Pipes , 2 mM MgSO4 , 1 mM EGTA , 1 mM DTT , 3 mg/ml D-Glucose , 0 . 2 mg/ml casein and 1 μM ATP at pH 7 . 0 . Before addition to the flow cell , the incubated bead–motor solutions were diluted in an assay buffer: 20 mM Pipes , 2 mM MgSO4 , 1 mM ATP , 1 mM EGTA , 1 mM DTT , 3 mg/ml D-Glucose , 0 . 4 mg/ml casein , 4 μM taxol , and a glucose oxidase/catalase-based oxygen scavenging system . For improved stability , recordings were made in flow-cells sealed with Dow Corning high vacuum grease ( Dow Corning , Barry , UK ) . Motor concentration was reduced until no more than one third of beads showed any binding . A bead-binding event was defined as a bead that remained attached to the microtubule for at least 2 s after the laser trap has turned off . Each bead was tested for binding on two or more different microtubules . Bead-binding data was fitted with two poisson probability curves using a linear least-squares method ( Svoboda and Block , 1994 ) . This analysis showed that our experimental hKif15 and Kinesin-1 data fit a ≥1 poisson distribution ( the probability that a bead binds by one or more motors ) , rather than the ≥2 distribution ( the probability that a bead binds by two or more motors ) . These data confirm that we are observing single motor ( i . e . , tetramer ) motility at the low concentrations we used in our trapping experiments ( 3 . 3 and 6 . 5 nM hKif15 , see Figure 2—figure supplement 2A ) . Some hKif15-linked beads bound in a diffusive non force-producing state to the microtubule , a behaviour that was rarely observed with kinesin-1-linked beads and fits to our observation of diffusive motion in our TIRF assays . In the case of the hTpx2 flow-through recordings , the samples were not sealed at the sides to allow wash-through . A bead with active motor was recorded and a further 3–4 cell volumes of assay buffer with 18 nM dimeric hTpx2 was carefully passed through the cell whilst retaining the same bead in the trap . The trapped bead was focused close to the lower cover-glass surface during these solution changes , this offered some boundary layer protection during the solution flow and also greatly reduced the chances of a second bead entering the trap . Following recordings in the presence of hTpx2 , a further 3–4 cell volumes of assay buffer were passed through the flow cell to make the final recordings following hTpx2 wash out . Bead position data were recorded at 88 kHz and averaged down to 22 kHz for analysis and storage . | Before a cell can divide , it produces an extra copy of all its chromosomes , and it must then ensure that each daughter cell ends up with one copy of each chromosome . During the division process , a structure called the spindle forms in the cell . This spindle is made of thread-like extensions called microtubules that grow from two poles at opposite ends of the cell . These microtubules are responsible for getting the chromosomes to line up in the middle of the cell , and then pulling half of the chromosomes to one end of the cell , and half to the other end . The cell then divides into two daughter cells . Two motor proteins—so-called because they consume chemical energy to ‘walk’ along the microtubules—have important roles in this process: Kif11 motor proteins mainly drive the formation of the spindle and thus division of the chromosomes . A cell that does not contain Kif11 can only divide if it contains extra copies of a second motor protein called Kif15: this suggests that Kif15 can serve as some sort of back up for Kif11 . Normal cells only divide when new cells are needed for growth or to replace old cells that have died . Cancer cells , on the other hand , divide in a way that is not controlled . Drugs that interfere with Kif11 have been developed in the hope that they will stop cancer cells dividing , but these drugs have not been very effective in clinical tests , possibly due to the Kif15 back up . Scientists hope , therefore , that a better understanding of the role of Kif15 may lead to improved cancer treatments . Drechsler et al . have isolated individual Kif15 motor proteins and used advanced microscopy techniques to study them in action . These experiments showed that Kif15 motor proteins can travel long distances along a single microtubule , and can also switch to a different microtubule at intersections . This movement of Kif15 is stopped when they bump into Tpx2 proteins , which are sitting on the microtubules . Together , these proteins can also form links between microtubules that can withstand high forces . These properties provide a starting point to understand how Kif15 can act as a back up for Kif11 in cells . In the future , it will be important to work out how Kif11 and Kif15 motor proteins work together in teams to build the spindle . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2014 | The Kinesin-12 Kif15 is a processive track-switching tetramer |
Various TRP channels act as polymodal sensors of thermal and chemical stimuli , but the mechanisms whereby chemical ligands impact on TRP channel gating are poorly understood . Here we show that AITC ( allyl isothiocyanate; mustard oil ) and menthol represent two distinct types of ligands at the mammalian cold sensor TRPM8 . Kinetic analysis of channel gating revealed that AITC acts by destabilizing the closed channel , whereas menthol stabilizes the open channel , relative to the transition state . Based on these differences , we classify agonists as either type I ( menthol-like ) or type II ( AITC-like ) , and provide a kinetic model that faithfully reproduces their differential effects . We further demonstrate that type I and type II agonists have a distinct impact on TRPM8 currents and TRPM8-mediated calcium signals in excitable cells . These findings provide a theoretical framework for understanding the differential actions of TRP channel ligands , with important ramifications for TRP channel structure-function analysis and pharmacology .
Neurons of the somatosensory system act as individually tuned sensory cells that convert specific thermal , mechanical and/or chemical stimuli into electrical signals , which are then conveyed to the central nervous system ( Vriens et al . , 2014 ) . Within the somatosensory system , several members of the TRP superfamily of cation channels act as polymodal molecular sensors of both temperature , and a variety of endogenous and exogenous chemicals , including a plethora of plant-derived compounds ( Clapham , 2003; Tominaga et al . , 1998; Voets et al . , 2005; Vriens et al . , 2014 ) . Chemical activation of TRP channels in nerve endings of trigeminal or dorsal root ganglion neurons is generally believed to underlie typical chemesthetic sensations evoked by such plant-derived substances ( Bandell et al . , 2007 ) , such as the burning heat evoked by capsaicin ( the pungent substance in hot peppers ) , which acts as a selective agonist of the heat-activated TRPV1 ( Caterina et al . , 1997 ) , and the cool sensation evoked by menthol ( the cooling compound in mint plants ) , due to activation of the cold sensor TRPM8 ( McKemy et al . , 2002; Peier et al . , 2002 ) . Such TRP channel ligands are present in widely used foodstuffs and drugs ( Nilius and Appendino , 2013 ) , and are extensively used as pharmacological tools to study somatosensation and/or TRP channel function in vitro and in vivo ( Julius , 2013 ) . Yet , very little is known about the molecular and biophysical mechanisms of action of the various TRP channel ligands . We studied the agonist effects of AITC , also known as mustard oil , a pungent organosulphur compound derived from Brassica plants . AITC is responsible for the characteristic oral sensations that one experiences upon eating Dijon mustard or wasabi , which contain between 5–30 mM of AITC ( Uematsu et al . , 2002 ) . Whereas earlier work has firmly established that AITC activates TRPA1 and TRPV1 in nociceptor neurons , approximately 10% of dorsal root ganglion neurons remained AITC-responsive after combined genetic deletion these two TRP channels ( Bandell et al . , 2004; Bautista et al . , 2006; Everaerts et al . , 2011; Jordt et al . , 2004 ) . In this work we show that AITC excites this subset of somatosensory neurons via direct activation of TRPM8 . Interestingly , a detailed biophysical analysis revealed that AITC activates TRPM8 by inducing a relative destabilization of the closed conformation relative to the transition state . This mode of action is fundamentally different from that of other known TRPM8 agonists such as menthol , which stabilize the open conformation relative to the transition state . Based on these results , we propose to classify TRPM8 agonists as either type I ( menthol-like ) or type II ( AITC-like ) , and provide a kinetic model that accurately describes the differential actions of the two agonist types on channel gating kinetics . Finally , we illustrate that the two agonist types have a distinct impact on TRPM8-mediated currents and calcium signals in excitable cells .
To investigate the origin of TRPV1- and TRPA1-independent AITC responses , we performed Ca2+ imaging experiments on dorsal root ganglion ( DRG ) neurons isolated from TRPV1/TRPA1 double knockout mice . In line with previous work ( Everaerts et al . , 2011 ) , we found that a small fraction of these TRPV1/TRPA1-deficient neurons ( 55 out of 578; 9% ) showed a rapid and reversible increase in intracellular Ca2+ in response to 3 mM AITC ( Figure 1A ) . These AITC-responsive cells consistently responded to menthol ( 54 out of 55; 98% ) ( Figure 1A , B ) . In these cells , the responses to both AITC and menthol were fully inhibited by the TRPM8 antagonist AMTB , and recovered partially upon AMTB washout ( Figure 1C ) . Taken together , these results indicate that TRPV1- and TRPA1-independent AITC responses in DRG neurons depend on the cold- and menthol-sensitive channel TRPM8 . 10 . 7554/eLife . 17240 . 003Figure 1 . AITC excites trigeminal neurons in a TRPM8-dependent manner . ( A ) Examples of fura-2-based intracellular calcium measurements in trigeminal neurons from TRPV1/TRPA1 double knockout mice . The red trace represents a neuron that shows responses to AITC ( 3 mM ) and menthol ( 50 µM ) , which can be reversible inhibited by AMTB ( 2 µM ) . The black trace represents a non-responder . A high K+-solution ( 50 mM K+ ) was used at the end of the experiments to identify neurons from non-neuronal cells . In total , 578 neurons from 6 different mice were analyzed . ( B ) Percentage of AITC-responsive neurons in menthol-sensitive ( n = 55 ) and menthol-insensitive ( n = 523 ) neurons . ( C ) Quantification of the reversible inhibition by AMTB of responses to AITC and menthol ( n = 54 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17240 . 003 To investigate the mechanisms underlying TRPM8-dependent AITC sensitivity in sensory neurons , we tested the effect of acute application of AITC on whole-cell currents in HEK293 cells heterologously expressing human TRPM8 . At room temperature , TRPM8 exhibits substantial activity , which can be recorded as an outwardly rectifying current ( Figure 2A , B ) . Application of AITC at concentrations ≥300 μM caused a rapid and reversible increase in TRPM8 current ( Figure 2A ) . The amplitude of the response increased with AITC concentration , with relatively stronger effects at negative voltages , but did not saturate at the highest concentration tested ( 10 mM; Figure 2C ) . At 3 and 10 mM AITC , activation was followed by a gradual decay of TRPM8 current , reducing current amplitude to levels below the basal level ( Figure 2A , B ) . Following washout of 3 or 10 mM AITC after prolonged exposure , we observed a rapid initial decrease in current followed by a gradual restoration of the current to the basal level ( Figure 2A ) , suggesting that the agonistic effect of AITC reverses more rapidly than the inhibitory effect . Rapid and reversible current responses to AITC were also observed in cell-free inside-out patches from human TRPM8-expressing HEK293 cells , indicating that the effect of AITC on TRPM8 is membrane-delimited ( Figure 2D , E ) . 10 . 7554/eLife . 17240 . 004Figure 2 . AITC activates human TRPM8 . ( A ) Time course of whole-cell currents at +100 and −80 mV in HEK293 cells expressing human TRPM8 , upon stimulation with the indicated concentrations of AITC . ( B ) Current-voltage relations recorded at the time points indicated in ( A ) . ( C ) Relative AITC-induced current increase at +100 and −80 mV ( n = 9 ) . ( D ) Menthol ( 50 µM ) and AITC ( 3 mM ) activate TRPM8 in cell-free inside-out patches during repetitive 100-ms voltage steps to +100 mV . Comparable current activation was measured in 5 out of 5 inside-out patches . ( E ) Current traces recorded at the time points indicated in ( D ) . ( F ) TIRF images showing mCherry-tagged human TRPM8 in the perimembrane region before and during stimulation with 3 mM AITC . Micrographs are 20 × 20 μm . ( G ) Lack of change in perimembrane mCherry-fluorescence during stimulation with AITC ( n = 6 ) . Fluorescence was normalized to the total fluorescence before adding AITC to the bath solution . DOI: http://dx . doi . org/10 . 7554/eLife . 17240 . 004 It has been put forward that AITC induces trafficking of TRPA1 to the plasma membrane ( Schmidt et al . , 2009 ) . To test whether AITC-induced activation of TRPM8 also involves rapid translocation of the channel towards the plasma membrane , we expressed human TRPM8 coupled with mCherry at its C terminus ( TRPM8-mCherry ) , and performed total internal reflection fluorescence ( TIRF ) microscopy to monitor potential AITC-induced transport of TRPM8 towards the plasma membrane . We have recently shown that TRPM8-mCherry is fully functional , and can be used to track cellular TRPM8 transport ( Ghosh et al . , 2016 ) . As shown in Figure 2F , G , application of 3 mM AITC had no detectable effect on the TRPM8-mCherry fluorescence in the close vicinity of the plasma membrane . mCherry fluorescence amounted to 99 ± 1% and 102 ± 2% of the pre-AITC level after 5 and 50 s of AITC application , respectively . Since the onset of current activation by AITC was very rapid , with maximal current achieved within ~2 s ( Figure 2A , D ) , we can exclude a significant contribution of trafficking to the acute agonistic effect of AITC on TRPM8 . To investigate the mechanism of the agonistic effect of AITC in more detail , we recorded TRPM8 currents during voltage steps ranging from −140 to +220 mV , both in control conditions and immediately upon application of AITC ( Figure 3A ) . Analysis of the steady-state conductances revealed that AITC has little or no effect on the maximal conductance at strongly depolarizing potentials , but shifts the voltage-dependent activation curves towards more negative voltages in a concentration-dependent manner ( Figure 3B , C ) . Such ligand-induced shifts in the voltage-dependent activation curve have been shown earlier to describe the effects of agonists on TRP channels , including the effect of menthol on TRPM8 ( Voets et al . , 2004; Voets et al . , 2007 ) ( Janssens and Voets , 2011 ) . 10 . 7554/eLife . 17240 . 005Figure 3 . Voltage dependence of the activating effect of AITC on human TRPM8 . ( A ) TRPM8 currents in response to the indicated voltage step protocol in the absence and presence of AITC ( 1 mM ) . ( B ) Voltage-dependent activation curves in control and in the presence of the indicated AITC concentrations , for the cell shown in ( A ) . Steady-state conductance ( G ) was determined as steady-state current divided by test voltage , and normalized to the estimated maximal conductance ( Gmax ) , which was obtained by fitting a Boltzmann function to the curve in the presence of 10 mM AITC . ( C ) Concentration dependence of the shift of voltage-dependent activation curves ( n = 7 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17240 . 005 However , when analyzing the kinetics of TRPM8 current activation/deactivation during voltage steps in more detail , we observed a remarkable difference between the effects of menthol and AITC . This is illustrated in Figure 4A , which provides a comparison of currents in the absence of ligands and in the presence of either 3 mM AITC or 30 μM menthol , concentrations that provoke similar steady-state TRPM8 current amplitudes at the end of the voltage steps . In the presence of AITC we observed a clear acceleration of the gating kinetics upon depolarization to +120 mV , whereas the current relaxation kinetics upon repolarization to −80 mV were not markedly altered . In stark contrast , in the presence of menthol we found a pronounced slowing of the kinetics of current relaxation , most noticeable upon repolarization to −80 mV ( Figure 4A , B ) . 10 . 7554/eLife . 17240 . 006Figure 4 . Differential effects of AITC and menthol on gating kinetics of human TRPM8 . ( A ) Current traces in response to the indicate voltage protocol in control condition and in the presence of menthol ( 30 μM ) and AITC ( 3 mM ) . The dashed lines overlaying the control and AITC traces represent single exponential fits , the dotted line overlaying the menthol trace represents a double exponential fit . ( B ) Scaled and expanded currents corresponding to the boxed areas in ( A ) . ( C , D ) Current traces in response to the voltage protocol from ( A ) in control condition and the indicated concentrations ( in μM ) of AITC and menthol . ( E ) Mono-exponential time constants for current relaxation at +120 and −80 mV in the presence of indicated concentrations of AITC ( n = 8 ) . Solid lines represent model predictions , obtained by fitting a mono-exponential function to simulated currents like those shown in Figure 5D . ( F ) Fast and slow exponential time constants for current relaxation at +120 mV and −80 mV in the presence of indicated concentrations of menthol ( n = 5 ) . Solid lines represent model predictions , obtained by fitting a double exponential function to simulated currents like those shown in Figure 5E . See Figure 4—figure supplement 1 for more details on the curve fitting . DOI: http://dx . doi . org/10 . 7554/eLife . 17240 . 00610 . 7554/eLife . 17240 . 007Figure 4—figure supplement 1 . Mono- and bi-exponential fits of experimental and modeled current relaxation time courses of human TRPM8 . ( A , B ) Examples of mono-exponential ( A ) and bi-exponential ( B ) fits to experimental relaxation time courses in the presence of different concentrations of menthol ( left ) , along with the corresponding residual plots ( right ) . The data are from Figure 4D , relaxation time course at +120 mV . ( C ) Mono-exponential and bi-exponential fits to modeled relaxation time courses at +120 mV as in Figure 5E ( left ) . Bi-exponential fits virtually overlap with the modeled data , as can be appreciated from the corresponding residual plots ( right ) . ( D ) Comparison of mono-exponential time constants at −80 mV ( red ) and =120 mV ( black ) obtained from fits to experimental ( symbols ) and modeled ( lines ) relaxation time courses . DOI: http://dx . doi . org/10 . 7554/eLife . 17240 . 00710 . 7554/eLife . 17240 . 008Figure 4—figure supplement 2 . Effects of thymol , icilin and linalool on gating kinetics of human TRPM8 . ( A–C ) Same approach as in Figure 4 , showing the effects of thymol ( A; 500 μM ) , icilin ( B; 10 μM ) and linalool ( C; 500 μM ) . ( D ) Effect of thymol , icilin and linalool on the time constant of current relaxation at +120 and −80 mV; n = 5 for each ligand , obtained by fitting a monoexponential function to the data . DOI: http://dx . doi . org/10 . 7554/eLife . 17240 . 008 To quantify the differences in gating kinetics in more detail , we fitted exponential functions to the current time courses during voltage steps . In line with earlier work , we found that in the absence of ligands the time courses at +120 and −80 mV were generally well fitted by a mono-exponential function ( Figure 4A , B; Figure 4—figure supplement 1 ) , yielding the time constants of current relaxation at both potentials ( τ+120 mV and τ−80 mV ) . In the presence of 3 mM AITC , the time courses remained well fitted by a mono-exponential function , and the accelerated kinetics were reflected in a reduction of τ+120 mV compared to control , whereas τ−80 mV was unaltered ( Figure 4A , B , E ) . In contrast , the relaxation kinetics in the presence of 30 μM menthol were consistently slower than in control and were no longer mono-exponential: at least two exponential terms were required to accurately describe the current time course at +120 ( τ+120 mV , fast and τ+120 mV , slow ) and −80 mV ( τ−80 mV , fast and τ−80 mV , slow ) ( Figure 4A , B , F ) . These distinct effects of AITC and menthol on the current relaxation kinetics of TRPM8 were observed over a broad concentration range ( Figure 4C–F ) . Other known TRPM8 agonists , including thymol , icilin , and linalool , act in a similar manner as menthol , slowing down the kinetics of activation and deactivation , albeit less pronounced ( Figure 4—figure supplement 2 ) . Based on these results , we propose that TRPM8 agonists be classified into two types based on their effect on the gating kinetics: Type I ( menthol-like ) agonists induce a slowing of the gating kinetics , which is most prominently observed as slowly deactivating tail currents following repolarization , whereas Type II ( AITC-like ) agonists cause an acceleration of the kinetics of channel activation upon depolarization , with little or no effect on the kinetics of deactivating tail currents ( Figure 4 ) . The differential effects of the two ligand types on the gating kinetics suggest that they act on different conformational states of the channel during the gating process . In particular , the characteristic slowly decaying tail currents upon repolarization in the presence of menthol indicate that menthol impedes voltage-dependent channel deactivation , which points at a stabilization of the channel in an open conformation . Oppositely , the faster current relaxation upon depolarization in the presence of AITC indicates that AITC accelerates voltage-dependent channel activation , which points at a destabilization of the channel in a closed conformation . To further pinpoint the mechanistic basis of different effects of Type I and Type II agonists on channel gating kinetics , we built on a previously described voltage-dependent Monod-Wyman-Changeux ( MWC ) model that was initially developed to describe the concerted actions of Ca2+ and voltage on the gating of large conductance Ca2+-activated potassium ( BK ) channels ( Cox et al . , 1997 ) . We have shown earlier that this model can accurately describe the effect of menthol , voltage and temperature on steady-state TRPM8 currents ( Janssens and Voets , 2011 ) . Moreover , based on analysis of channel chimeras with different combinations of wild type and mutated menthol binding sites , it was found that a single TRPM8 channel can bind up to four ligand molecules , each subunit having and ligand binding site with an affinity Kd ( in the closed state ) , and that every bound ligand shifts the equilibrium between the closed and open channel by a similar extent ( Janssens and Voets , 2011 ) . The energetic effect of ligand binding can be quantified as ΔΔGligand , which represents the change of the difference in Gibbs free energy between the closed and open state of the channel ( ΔG ) upon binding of one ligand molecule to one of the four subunits . In the case of an agonist , ΔΔGligand < 0 , which implies that the open state becomes more stable relative to the closed state . As illustrated by the energy diagrams in Figure 5A , a negative ΔΔGligand can be the result of a ligand-induced relative stabilization of the open state , destabilization of the closed state , or a combination of both , taking the transition state as the reference . In the case of a relative stabilization of the open state , the energy barrier for the transition from open to closed will become higher , which would lead to slower closing rates , as seen with the Type I agonists ( Figure 5A ) . Oppositely , relative destabilization of the closed state will reduce the energy barrier for the transition from closed to open , which would be reflected in faster opening rates , as seen with the Type II agonist AITC ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 17240 . 009Figure 5 . Type I ( menthol-like ) versus Type II ( AITC-like ) TRPM8 agonists . ( A ) ( left ) Energy diagram for the transition between the closed and open channel conformation in a non-liganded channel . Steady-state equilibrium is determined by ΔG0 , whereas Eopen and Eclose determine the opening and closing rates , respectively . ( right ) Alteration in the energy profile upon binding of Type I and Type II ligands . The black line represents the non-liganded channel , whereas the green lines represent channels with 1–4 bound ligands . The corresponding kinetic schemes are provided in Supplementary Figure 2 . ( B , C ) Activation ( B ) and deactivation ( C ) time courses in the absence and presence of the indicated concentrations of menthol or AITC . Overlaid dashed lines represent global fits to the control and ligand-activated current traces . ( D , E ) Model predictions corresponding to the experimental data shown in Figure 4C , D . DOI: http://dx . doi . org/10 . 7554/eLife . 17240 . 00910 . 7554/eLife . 17240 . 010Figure 5—figure supplement 1 . Kinetic schemes of the MWC model , depicting the differential effects of Type I and Type II ligands . DOI: http://dx . doi . org/10 . 7554/eLife . 17240 . 01010 . 7554/eLife . 17240 . 011Figure 5—figure supplement 2 . Combining Type I and Type II agonists . ( A ) Combined effect of menthol and AITC on TRPM8 gating kinetics , using the voltage protocol shown in Figure 4A . ( B ) Model simulation of the combined effect menthol and AITC . To obtain these traces , the effect of AITC was modeled as a fixed decrease of ΔG0mVα† . ( C ) Energy profiles upon simultaneous binding of Type I ( top ) and Type II ( bottom ) ligands . DOI: http://dx . doi . org/10 . 7554/eLife . 17240 . 011 We performed global fits of the MWC model to the experimental current time courses obtained in individual cells during voltage steps in both the absence and presence of different concentrations of AITC or menthol . Values for Kd , menthol , ΔΔGmenthol , Kd , AITC and ΔΔGAITC were obtained from shifts in the steady-state voltage-dependent activation curves ( Janssens and Voets , 2011 ) . Note that values for the on rates for ligand binding ( kon ) were determined from the fits , in contrast to earlier work in BK channels were Ca2+ binding rates were assumed to be diffusion-limited ( Cox et al . , 1997 ) . Off rates ( koff ) were constrained by the Kd and on rates . Importantly , we obtained excellent fits to the experimental data when we set fixed that menthol binding acts exclusively by stabilization the open state , while AITC acts by destabilization of the closed state ( Figure 5B , C ) . Model parameters obtained from the fits are listed in Table 1 . Gratifyingly , the model accurately predicts the concentration-dependent effects of AITC and menthol on TRPM8 , including the mono-exponential time constants in the presence of different AITC concentrations , as well as the bi-exponential relaxation kinetics in the presence of menthol , respectively ( Figure 5D , E; Figure 4E , F; Figure 4—figure supplement 1 ) . Based on these results , we propose that AITC represents the first example of a type II TRPM8 agonist , acting primarily by destabilizing the closed channel , which contrasts to the Type I agonists , such as menthol , icilin , thymol and linalool , which primarily stabilize the open channel . 10 . 7554/eLife . 17240 . 012Table 1 . Experimentally derived model parameters describing the action of menthol and AITC on TRPM8 gating . DOI: http://dx . doi . org/10 . 7554/eLife . 17240 . 012ParameterValueSourcez0 . 82 ( Voets et al . , 2007 ) ΔΔGAITC−2 . 7 ± 0 . 4 kJ mol−1Steady-state activation curves ( n = 7 ) ΔΔGmenthol−4 . 5 ± 0 . 4 kJ mol−1Steady-state activation curves ( n = 6 ) Kd , AITC2 . 9 ± 0 . 6 mMSteady-state activation curves ( n = 7 ) Kd , menthol21 ± 4 μMSteady-state activation curves ( n = 6 ) α0 ( 0 ) 10 . 4 ± 1 . 2 s−1Global kinetic fit ( n = 14 ) β0 ( 0 ) 1 . 11 ± 0 . 15 ×103 s−1Global kinetic fit ( n = 14 ) kon , AITC95 ± 35 ×103 M−1 s−1Global kinetic fit ( n = 7 ) kon , menthol551 ± 210 ×103 M−1 s−1Global kinetic fit ( n = 7 ) Displayed are values for the different parameters that determine the MWC model . For the global kinetic fits , cells were included for which current traces were fit at minimally tree ligand concentrations and two voltages . More details are provided in the text . We also tested the combined effect of AITC and menthol on the kinetics of TRPM8 activation and deactivation . In line with the above , application of 50 μM menthol results in slower activation and deactivation kinetics , due to the stabilization of the open state ( Figure 5—figure supplement 2 ) . Addition of 3 mM AITC in the continued presence of menthol resulted in faster activation kinetics , without affecting the time course of deactivation ( Figure 5—figure supplement 2 ) . These results are in line with the predictions of the MWC model , assuming that Type I and Type II agonists can act simultaneously and independently , resulting in both stabilization of the open and destabilization of the closed state ( Figure 5—figure supplement 2 ) . In the context of a sensory neuron , activation of ion channels such as TRPM8 causes influx of Na+ and Ca2+ , which depolarizes the membrane and , when the threshold is reached , causes action potential generation ( Vriens et al . , 2014 ) . The differential effects of Type I and Type II agonists on the gating kinetics of TRPM8 suggest that they may have distinct effects on TRPM8-mediated currents and calcium signals during rapid neuronal action potentials . To investigate this possibility , we measured TRPM8 currents evoked by voltage waveforms mimicking action potentials in sensory neurons in the presence of AITC ( 3 mM ) or menthol ( 30 µM ) . Note that , at a physiological holding potential of −60 mV , these concentrations resulted in comparable steady-state inward current amplitudes ( Figure 6A ) . In response to the action potential waveforms , the current in the presence of AITC mainly manifested during the upstroke phase , and rapidly deactivates upon action potential repolarization . In comparison , in the presence of menthol , the peak outward current is smaller but a more prominent inward TRPM8 current is observed during the repolarization phase of the action potential ( Figure 6A–C ) . These differential effects of AITC and menthol on TRPM8 currents during an action potential are fully in line with the predictions of the MWC model for type I versus Type II agonists ( Figure 6A ) . We also compared the cumulative influx and efflux of charge during a 1-s train of action potential waveforms at 8 Hz , a typical firing rate of cold-sensitive neurons ( Orio et al . , 2012 ) . As illustrated in Figure 6D , E , net charge influx is larger in the presence of menthol , whereas net charge efflux is larger in the presence of AITC . 10 . 7554/eLife . 17240 . 013Figure 6 . TRPM8 gating during an action potential – Type I versus Type II ligands . ( A ) Voltage protocol simulating a sensory neuron action potential ( left ) ; TRPM8 currents in HEK293 cells in response to the action potential waveform in control condition and during application of menthol ( 30 μM ) and AITC ( 3 mM ) ( middle ) ; and corresponding model simulation ( right ) . Boxed areas are expanded in the inset . ( B ) Peak outward and inward currents during the action potential waveform in the presence of menthol ( cyan ) and AITC ( magenta; n = 6 ) . ( C ) Ratio between peak inward and peak outward current in the presence of menthol or AITC . ***p<0 . 001 . ( D ) TRPM8 current responses during a train of action potentials ( 1 s; 8Hz ) in control condition and during application of menthol and AITC ( top ) ; outward ( middle ) and inward ( bottom ) charge displacement during the action potential train , determined as the integrated current after subtraction of the holding current . ( E ) Mean inward and outward charge displacement for the two ligands ( n = 5 ) . *p<0 . 05; **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 17240 . 013 Under normal physiological conditions , inward TRPM8 current is partially carried by Ca2+ ions ( McKemy et al . , 2002; Peier et al . , 2002 ) . Since our results indicated substantial differences in charge influx during action potentials between Type I and Type II agonists , we expected that menthol and AITC may show differential efficacy in evoking Ca2+ transients in excitable versus non-excitable cells . To test this , we compared the relative responses to menthol and AITC in TRPM8-expressing mouse sensory neurons versus ( non-excitable ) HEK293 cells heterologously expressing TRPM8 . For these latter experiments , we used HEK293 cells transiently expressing the mouse TRPM8 orthologue , and first tested current responses to AITC . Like its human orthologue , mouse TRPM8 was rapidly activated by AITC , and the difference in gating kinetics in the presence of AITC versus menthol was also observed ( Figure 7—figure supplement 1 ) . However , interestingly , AITC-induced current inhibition was much less pronounced in mouse TRPM8 compared to the human orthologue ( Figure 7—figure supplement 2 ) : at the end of a 60-s application of 3 mM AITC , mouse TRPM8 current amounted to 88 ± 6% ( n = 8 ) of the peak current , compared to 21 ± 5% in the case of human TRPM8 ( n = 7; p=0 . 00004 ) . A further analysis of this species difference in AITC-induced inhibition is provided in Figure 7—figure supplement 2 . To specifically analyze TRPM8-mediated responses to AITC and menthol in mouse sensory neurons , we used TRPA1/TRPV1 double knockout mice , only examined cells that showed robust responses to both agonists , and controlled that these responses were fully inhibited in the presence of AMTB , as outlined in Figure 1 . In these cells , we found that the amplitudes of Ca2+ transients evoked by a 60-s-long applications of 3 mM AITC were on average ~30% smaller than those evoked by 30 µM menthol ( Figure 7A , D ) . Likewise , the peak rate of calcium rise , which represents a measure for the maximal inward calcium current , was consistently smaller in response to AITC than to menthol in neurons ( Figure 7A , D ) . Interestingly , we observed an opposite potency of the same concentrations of menthol and AITC in HEK cells expressing mouse TRPM8: AITC evoked larger calcium increases and with a higher peak rate of calcium rise than did menthol ( Figure 7B , D ) . Moreover , if action potential firing in the TRPA1/TRPV1-deficient sensory neurons was blocked using tetrodotoxin ( TTX; 1 μM ) , we found a similar ratio of AITC versus menthol responses as in HEK cells ( Figure 7C , D ) , with AITC being slightly more potent than menthol . Taken together , these data provide further support for the notion that , compared to Type II agonists ( e . g . AITC ) , Type I agonists ( e . g . menthol ) are more potent in evoking calcium influx in excitable cells , due to enhanced calcium influx during the prolonged inward tail currents following action potentials . In cells that do not fire action potentials , rapid changes in membrane potential are not expected , and hence the kinetic differences between the two types of agonists will not affect calcium signals . 10 . 7554/eLife . 17240 . 014Figure 7 . Differential effectiveness of Type I and Type II agonists in excitable versus non-excitable cells . ( A ) Fura-2-based intracellular calcium measurements in mouse trigeminal neurons from TRPV1/TRPA1 double knockout mice showing increases in intracellular calcium in response to AITC ( 3 mM ) and menthol ( 30 µM ) . The upper trace shows the time differential of the intracellular calcium concentration , which represents a measure of net calcium influx/extrusion mechanisms . The TRPM8-dependence of the responses was ensured based on full block by AMTB ( as in Figure 1; not shown ) . ( B ) Same as ( A ) , but in the presence of TTX ( 1 μM ) to block neuronal action potentials . ( C ) Same as ( A ) , but in a HEK293 cell expressing mouse TRPM8 . Non-transfected cells did not show any detectable response to AITC or menthol . ( C ) Relative stimulatory effect of menthol and AITC in control trigeminal neurons ( n = 81 from 9 different mice ) , trigeminal neurons treated with 1 μM TTX ( n = 3 from 3 different mice ) and HEK293 cells ( n = 448 ) . * , ** , ***: p<0 . 05 , 0 . 01 and 0 . 001 , respectively , in paired t-test comparing the response to AITC and menthol within individual cells . ###: P<0 . 001 in unpaired t-tests comparing TG neurons and HEK293 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 17240 . 01410 . 7554/eLife . 17240 . 015Figure 7—figure supplement 1 . Activation of mouse TRPM8 by AITC . ( A ) TRPM8 currents in response to the voltage step protocol shown in Figure 3A in the absence and presence of AITC ( 3 mM ) . ( B ) Voltage-dependent activation curves corresponding to the currents shown in ( A ) . ( C ) Current traces in response to the voltage protocol shown in Figure 4A , in control condition and in the presence of menthol ( 30 μM ) and AITC ( 3 mM ) . ( right ) Scaled and expanded currents corresponding to the boxed areas . DOI: http://dx . doi . org/10 . 7554/eLife . 17240 . 01510 . 7554/eLife . 17240 . 016Figure 7—figure supplement 2 . AITC-induced current inhibition in human versus mouse TRPM8 , as well as in chimeric channels . ( A , B ) Time courses of whole-cell currents at +100 and −80 mV in HEK293 cells expressing human ( left ) or mouse ( right ) TRPM8 , upon sequential ( A ) or simultaneous ( B ) stimulation with menthol ( 50 µM ) and AITC ( 3 mM ) . ( C , D ) Quantification of the AITC-induced current inhibition in mouse and human TRPM8 , as well as in the depicted chimeric channels containing all possible combinations of transmembrane region and N and C termini of both orthologues . In ( C ) inhibition was quantified as I60s/Ipeak , which represents the remaining current after a 60-s application of 3 mM AITC divided by the peak AITC-induced current . In ( D ) inhibition was quantified as I60s/Imenthol , which represents the remaining current after a 60-s application of 3 mM AITC divided by the peak current induced by menthol ( 50 µM ) . Mean ± SEM for 5–8 cells for each chimeric channel . DOI: http://dx . doi . org/10 . 7554/eLife . 17240 . 016
While there are already numerous natural and synthetic agonists known for TRPM8 ( Almaraz et al . , 2014 ) , our results demonstrate that AITC is an atypical agonist , with a mode of action that is fundamentally different from that of all other known TRPM8-activating stimuli . Activation of TRPM8 by cooling or by known agonists such as the natural compounds menthol , thymol and linalool , and the synthetic agonists such as icilin and halothane , is associated with a slowing of the kinetics of voltage-dependent channel gating ( Vanden Abeele et al . , 2013; Voets et al . , 2004; Voets et al . , 2007 ) . This slowing of the gating kinetics can be directly explained by a stabilization of the open channel relative to the transition state , as we illustrated in this work for menthol and elsewhere for cooling ( Voets et al . , 2004 ) . In clear contrast , activation of TRPM8 by AITC resulted in an acceleration of the kinetics of voltage-dependent gating , and we show here that this can be fully explained by a mechanism where AITC leads to a relative destabilization of the closed conformation relative to the transition state . Based thereon , we propose that TRPM8 agonists can be classified as either Type I , causing a relative stabilization the open state , or Type II , causing a relative destabilization the closed state , and we provide a kinetic voltage-dependent Monod-Wyman-Changeux-type model that faithfully reproduces their differential agonist effects . Such classification may also be extended to activating ligands of other voltage- and ligand-sensitive TRP channels . For instance , published current traces suggest that activation of TRPV1 by capsaicin or low pH is associated with faster activation time courses upon depolarization ( Aneiros et al . , 2011; Voets et al . , 2004 ) , classifying them as Type II ligands , whereas the activating effects of phosphatidylinositol-4 , 5-bisphosphate ( PIP2 ) or lysophosphatidic acid are associated with slower activation and longer deactivating tails upon repolarization , classifying them a Type I agonists ( Nieto-Posadas et al . , 2012; Ufret-Vincenty et al . , 2015 ) . The classification of activating ligands as either Type I or Type II is useful for several purposes . First , this information can provide important insights into ligand-induced structural rearrangements during channel gating , and may help interpreting ligand-bound channel structures . Indeed , our results indicate that Type II ligands such as AITC reduce the difference Gibbs free energy between the closed state and the transition state ( ΔGα† ) , without affecting the difference Gibbs free energy between the open state and the transition state ( ΔGβ† ) . This suggests that the AITC-induced conformational change at its binding site occurs early in the gating process , prior to the main close-open transition . Oppositely , Type I ligands such as menthol cause an increase in ΔGβ† , without affecting ΔGα† . This indicates that the menthol-induced conformational change at its binding sites occurs later than the main close-open transition . This analysis and interpretation is reminiscent of the rate-equilibrium free-energy relationship ( REFER ) approach , which has used to evaluate the effects of perturbations ( e . g . ligands or mutations ) on the equilibrium of reactions , including the gating of ion channels such as the nicotinic acetylcholine receptor and CFTR ( Grosman et al . , 2000; Sorum et al . , 2015 ) . In REFER , the effect of a family of perturbations on channel gating is quantified using the φ-value , which is the slope of a plot of the logarithm of the opening rate ( log α ) versus the log of the gating equilibrium constant ( log Keq ) , where Keq is the ratio of the opening ( α ) and closing ( β ) rate ( Keq = α/β ) ( Auerbach , 2007 ) . In the case of type I ligands , ligand binding affects the equilibrium solely by decreasing β , yielding φ = 0 . Following the REFER theorem , this indicates late movement ( Auerbach , 2007 ) . Type II ligands affect the equilibrium entirely by affecting α , yielding φ = 1 , indicating early movement . We also found that the effects of simultaneously applied menthol and AITC on channel gating are well described assuming independent binding and additive effects on the gating equilibria , which further supports the notion that type I and type II ligands act at distant binding sites with different timing for conformational changes . Second , based on our kinetic fits , we obtained estimates for the on rates for ligand binding to TRPM8 . For example , for menthol binding we obtained a kon of 0 . 55 μM−1s−1 , which is well below the diffusion-limited rate ( >100 μM−1s−1 ) , and also one order of magnitude or more lower than binding rates for ligands to synaptic ligand-gated channels , such as the ionotropic receptors for glutamate ( kon≈ 5 μM−1s−1 ) ( Clements and Westbrook , 1991 ) , ATP ( kon≈ 12 μM−1s−1 ) ( Bean , 1990 ) or acetylcholine ( kon≈ 60 μM−1s−1 ) ( Sine et al . , 1990 ) . The relatively slow ligand equilibration kinetics for menthol are in line with the distinct structural properties of ligand binding sites in TRP channels compared to these classical ionotropic receptors . Indeed , whereas binding sites for glutamate , ATP and acetylcholine are located extracellularly ( Hille , 2001 ) , directly accessible from the aqueous phase , the binding site for menthol is located in a hydrophobic domain in between the transmembrane helices ( Bandell et al . , 2006; Voets et al . , 2007 ) . The observation that current relaxation time courses of TRPM8 in the presence of menthol become multi-exponential is then a direct consequence of the slow equilibration rate of menthol with its binding site in comparison to the transition rates between closed and open channel conformations . Finally , we showed that differential effect on voltage-dependent gating of Type I and II agonists is reflected in distinctive TRP channel-mediated currents during rapid changes in membrane voltage , for instance during an action potential in a sensory neuron . Indeed , as predicted by our model , the AITC-induced TRPM8 current during a typical action potential waveform mainly manifest during the upstroke phase , and rapidly deactivates upon action potential repolarization . In contrast , an equipotent concentration of menthol ( i . e . a concentration of menthol provoking a similar steady-state current ) results in a less outward current but more prominent activation of inward TRPM8 current during the repolarization phase following an action potential , and thus leads to more Ca2+ influx via TRPM8 . In line herewith , we found that inhibiting action potential firing using TTX has a more profound effect on menthol-induced responses than on AITC-induced responses in sensory neurons . These findings illustrate the importance of evaluating the mode of action of ligands on voltage-dependent TRP channels , especially when extrapolating results from non-excitable heterologous expression systems to physiological effects in excitable cells such as neurons , cardiomyocytes or pancreatic beta cells . Using the voltage-dependent Monod-Wyman-Changeux-type model , we assumed that for any number of bound ligands the transition between closed and open channel conformation is a one-step process , determined by forward and backward rates αi and βi . Whereas this assumption is in line with the mono-exponential kinetics we generally observed in our experiments in the absence of ligands ( Voets et al . , 2004 ) , it is probably a simplification of the full gating intricacies of TRPM8 , and models with one or more closed-closed transitions preceding channel opening have been proposed ( Fernandez et al . , 2011; Raddatz et al . , 2014 ) . Nevertheless , even when using such more complex models , the ligands’ effects on TRPM8 gating kinetics can only be explained assuming that Type II ligands cause acceleration of the gating transition ( s ) towards the open state , whereas Type I ligands slow down the backward rate ( s ) from the open state . Our results demonstrate that TRPM8 underlies the residual TRPA1- and TRPV1-independent responses to AITC in mouse sensory neurons . Under our experimental conditions , activation of TRPM8 by AITC only occurred in the high micromolar to millimolar concentration range . As such , TRPM8 is about two orders of magnitude less sensitive to AITC than TRPA1 , for which concentrations for half-maximal activation of 5–50 µM have been reported ( Bandell et al . , 2004; Everaerts et al . , 2011; Jordt et al . , 2004 ) , but comparable to TRPV1 , for which a concentration for half-maximal activation of 3 mM was found at room temperature ( Everaerts et al . , 2011 ) . These findings are in line with in vitro experiments in sensory neurons , showing that AITC concentrations ≤100 µM evoke responses that are strictly TRPA1-dependent ( Bautista et al . , 2006; Everaerts et al . , 2011 ) , whereas higher concentrations can also evoke TRPA1-independent responses mediated by TRPV1 or TRPM8 . AITC is extensively used in in vivo experiments to induce pain and inflammation ( Julius , 2013 ) . In such assays , experimental solutions that are injected or topically applied typically contain AITC at concentrations between 10 and 100 mM ( Bautista et al . , 2006; Caterina et al . , 2000 ) . Whereas earlier studies have clearly shown that pain and inflammatory responses under such experimental conditions are largely mediated by TRPA1 and TRPV1 ( Bautista et al . , 2006; Everaerts et al . , 2011; Kwan et al . , 2006 ) , our present results suggest that also TRPM8-positive sensory nerve endings may become activated at these AITC doses . Since activation of TRPM8-expressing neurons can cause analgesia in animal models of acute and chronic pain ( Liu et al . , 2013; Proudfoot et al . , 2006 ) , the effects of AITC on TRPM8 that we describe here need to be taken into account when using AITC as a proalgesic and/or proinflammatory agent . TRPM8 may contribute to the complex psychophysical effects that one experiences upon eating spices containing millimolar concentrations of AITC such as mustard or wasabi ( Nilius and Appendino , 2013 ) . In line herewith , a transient increase in cold sensitivity was observed upon application of 100 mM AITC on the tongue of human volunteers ( Albin et al . , 2008 ) . Although speculative , this may correlate with the transient activation followed by channel inhibition that we observed in human TRPM8 . In voltage-gated ion Na+ , K+ and Ca2+ channels , ligand modulators have since decades been classified based on their distinct state-dependent effects on channel gating ( Hille , 2001 ) , and this mechanistic insight has been key to understanding their physiological impact in for instance neurons and cardiac cells ( Sack and Sum , 2015 ) . In this study , we demonstrate for the first time the existence of two types of agonists with distinct state-dependent effects for a member of the TRP superfamily , the cold-sensitive TRPM8 , and provide a paradigm for their differential effects in sensory neurons . We argue that establishing the state-dependent mode of action of ( ant ) agonists of this and other TRP channels will be essential to clarify their physiological actions as well as to understand their impact on conformational changes in the channel molecule .
HEK293 were grown in DMEM containing 10% ( v/v ) fetal calf serum , 4 mM L-alanyl-L-glutamine , 100 U ml–1 penicillin and 100 μg ml–1 streptomycin at 37°C in a humidity controlled incubator with 10% CO2 . For patch-clamp and calcium imaging , cells were transiently transfected with different human ( NM024080 ) or mouse ( NM134252 ) TRPM8 constructs cloned in the bicistronic pCAGGSM2-IRES-GFP vector using TransIT-293 transfection reagent ( Mirus , Madison , WI ) . Mutations and chimeras were made using the PCR-overlap technique , and verified by Sanger sequencing ( LGC-genomics , Germany ) . Chimeras were made by swapping the N termini ( amino acids 1–336 ) or C termini ( amino acids 993–1004 ) between the orthologues . For TIRF imaging , we used human TRPM8 linked to mCherry at its C-terminal end ( Ghosh et al . , 2016 ) . Trigeminal ganglia ( TGs ) of 10-16-week-old female Trpv1-/-/Trpa1-/- mice were isolated after CO2 euthanasia . Bilateral TGs were collected and digested with 1 mg/ml collagenase and 2 . 5 mg/ml dispase dissolved in ‘basal medium’ ( Neurobasal A medium supplemented with 10% FCS ) ( all from Gibco/Life Technologies , Belgium ) at 37°C for ca . 45–60 min . Digested ganglia were gently washed once in ‘basal medium’ and twice in ‘complete medium’ ( Neurobasal A medium supplemented with 2% B27 [Invitrogene/Life Technologies , Belgium] , 2 ng/ml GDNF [Invitrogen/Life Technologies] and 10 ng/ml NT4 [Peprotech , UK] ) and mechanically dissociated by mixing with syringes fitted with increasing needle gauges . Neurons were seeded on poly-L-ornithine/laminin-coated glass bottom chambers ( Fluorodish WPI , UK ) and cultured at 37°C in complete medium overnight . These experiments were approved by the KU Leuven Ethical Committee Laboratory Animals under project number P192/2014 . Between 16 and 24 hr after transfection , currents were recorded in the whole-cell or inside-out configurations of the patch-clamp technique using an EPC-9 amplifier and PULSE software ( HEKA Elektronik , Germany ) . Data were sampled at 5–20 kHz and digitally filtered off-line at 1–5 kHz . In the whole-cell mode , between 70 and 90% of the series resistance was compensated , and recordings where the estimated voltage error due to uncompensated series resistance exceeded 10 mV were excluded from analysis . Whole-cell recordings were performed using an intracellular solution containing ( in mM ) 150 NaCl , 5 MgCl2 , 5 EGTA and 10 HEPES , pH 7 . 4 . The extracellular solution contained ( in mM ) 150 NaCl , 1 MgCl2 and 10 HEPES , pH 7 . 4 . In inside-out recordings , the extracellular solution was used as pipette solution , and ligands were included in the intracellular bath solution . For intracellular Ca2+ measurements , cells were incubated with 2 µM Fura-2 acetoxymethyl ester for 30 min at 37°C . The fluorescent signal was measured during alternating illumination at 340 and 380 nm using either an CellM ( Olympus , Belgium ) or Eclipse Ti ( Nikon , Belgium ) fluorescence microscopy system . The standard extracellular solution used in ratiometric [Ca2+]i measurements contained ( in mM ) 150 NaCl , 5 KCl , 2 CaCl2 , 1 . 5 MgCl2 , and 10 HEPES , pH 7 . 4 . TIRF images were acquired using a through-the-lens TIRF system that was built around an inverted Axio Observer . Z1 microscope equipped with a X-100 oil objective numerical aperture ( NA ) =1 . 45 ( Zeiss , Germany ) , a Orca-R2 camera ( Hamamatsu , Japan ) , and using a 561-nm laser . Time series of images at 1-s intervals were recorded . Constant focus was maintained using the Definite Focus module ( Zeiss ) . The TIRF angle was set to achieve an evanescent field with a characteristic penetration depth ( i . e . , the distance in the z direction over which the intensity declines e-fold ) of 90 nm . Cells on 25-mm glass coverslips were placed in a custom-made chamber and imaged at 25°C . Chemicals were obtained from Sigma ( Belgium ) , unless indicated otherwise . AITC , menthol , thymol , linalool were dissolved in ethanol to obtain 1-M stock solutions . Icilin was dissolved in DMSO to obtain a 50-mM stock solution . Tetrodotoxin ( TTX; from Alomone labs , Israel ) was dissolved in acetate buffer at a concentration of 31 mM . As a starting point to model the gating of TRPM8 in the absence and presence of ligands , we built on our earlier work describing the effects of temperature and menthol on steady-state TRPM8 currents ( Janssens and Voets , 2011; Voets , 2012; Voets et al . , 2004 , 2007 ) . In the absence of ligands , the transition between the closed and open conformation of the channel is determined by the opening and closing rates: ( 1 ) α0=κkbThe−ΔGα†RT and ( 2 ) β0=κkbThe−ΔGβ†RT , where kb is the Boltzmann constant ( 1 . 38 × 10−23 J K−1 ) , T the absolute temperature , h the Planck constant ( 6 . 63 × 10−34 J s ) , R the universal gas constant ( 8 . 314 × J K−1 mol−1 ) and κ the transmission coefficient , whose value for the studied processes is unknown . ΔG0α† ( ΔG0β† ) represents the difference in free energy between closed ( open ) state and the transition state of the non-liganded channel ( see Figure 5 ) , and depend on temperature and voltage ( V ) according to: ( 3 ) ΔGα†=ΔHα†−TΔSα†−0 . 5zFV and ( 4 ) ΔGβ†=ΔHβ†−TΔSβ†+0 . 5zFV . ΔH0α† and ΔH0β† represent the differences in enthalpy and ΔS0α† and ΔS0β† the differences in entropy between , respectively , the closed and open state and the transition state , z the gating charge , and F the Faraday constant ( 96485 C mol−1 ) . In our experiments , temperature was kept constant at 23°C , yielding: ( 5 ) ΔGα†=ΔG0 mVα†−0 . 5zFV and ( 6 ) ΔGβ†=ΔG0 mVβ†+0 . 5zFV . In the absence of ligands , the voltage-dependent opening and closing rates are then given by: ( 7 ) α0 ( V ) =α0 ( 0 ) ×e0 . 5zFVRT and ( 8 ) β0 ( V ) =β0 ( 0 ) ×e−0 . 5zFVRT , where α0 ( 0 ) and β0 ( 0 ) represent the opening and closing rates at 0 mV . As evidenced in earlier work ( Janssens and Voets , 2011 ) , we consider that TRPM8 has 4 independent and energetically equivalent ligand binding sites ( i . e . one per subunit ) , with an affinity Kd of the open channel determined by ligand-channel association and dissociation rates kon and koff ( Kd = koff/kon ) . The energetic effect of ligand binding on steady-state channel equilibrium can be quantified as ΔΔGligand , which represents the change of the difference in Gibbs free energy between the closed and open state of the channel ( ΔG ) upon binding of one ligand molecule to one of the four subunits . Values for Kd , menthol , ΔΔGmenthol , Kd , AITC and ΔΔGAITC were obtained from concentration-dependent changes in the midpoint of the steady-state voltage-dependent activation curves ( ΔV1/2 ) , according to: ( 9 ) ΔV1/2=−RTzFln ( 1+[L]Kd ) 4 ( 1+[L]Kd×expΔΔGRT ) 4 . Since saturating effects of AITC could not be obtained at the highest concentration tested ( 10 mM; on the limits of solubility ) , values for Kd , AITC and ΔΔGAITC should be considered as approximative . In the presence of a Type I agonist such as menthol , ligand binding stabilizes the open state , without affecting the closed or transition states ( Figure 5 and Figure 5—figure supplement 1 ) . Therefore , the opening rate of a channel with i bound ligands remains unaltered ( 10 ) αi ( V ) =α0 ( V ) , where as the closing rate becomes slower for each bound ligand ( 11 ) βi ( V ) =β0 ( V ) ×e−i×ΔΔGligandRT . In the presence of a Type II agonist such as AITC , ligand binding destabilizes the closed state , without affecting the open or transition states ( Figure 5—figure supplement 2 ) . Therefore , the closing rate of the channel remains unaltered ( 12 ) βi ( V ) =β0 ( V ) , where as the opening rate becomes faster for each bound ligand ( 13 ) αi ( V ) =α0 ( V ) ×ei×ΔΔGligandRT . Procedures were written in Igor Pro 6 . 22 ( Wavemetrics , Lake Oswego , OR ) to numerically solve the set of 10 differential equations describing the transitions between the 10 states of the model at different voltages and in the presence of different ligand concentrations . Briefly , to fit the gating behavior during voltage steps , eigenvalues and corresponding eigenvectors of the transition matrix were numerically solved using the MatrixEigenV operation , and used to calculate the sums of exponential terms describing the time-dependent changes of the probabilities that the channel is in one of the 10 states . The FuncFit and DoNewGlobalFit procedures were then used to find the model parameters that yield the best global fit to current relaxation time courses measured within one cell in the absence and presence of ligand ( Table 1 ) . The global kinetic fit included three free parameters: α0 ( 0 ) , β0 ( 0 ) and kon . Prior to the kinetic fit , Kd , ligand and ΔΔGligand were determined from steady-state currents , according to Equation 9; koff was set as Kd× kon; z was fixed at a value of 0 . 82 , based on earlier work ( Voets et al . , 2007 ) . We further assumed that the rate of ligand binding to the open and closed state of the channel were identical , whereas the rate of ligand unbinding from the closed state was constrained by detailed balance . The integrateODE operation was used to model TRPM8 currents during voltage steps or action potential waveforms , using mean parameters obtained from the fits ( Table 1 ) . Data analysis was performed using Origin 9 . 0 ( OriginLab Corporation , Northampton , MA ) . Group data are presented as mean ± SEM from n cells . Comparison between two groups was done using Student’s unpaired or paired test , as indicated . No explicit power analysis was performed prior to the experiments to determine sample size , since we had no means to reliably estimate the size and variability of the effects of the ligands on parameters of TRPM8 gating . For patch-clamp experiments on HEK cells , typically 5–10 cells were measured for each condition , thereby limiting the SEM to ≤20% of the mean value for the relevant parameters . For the calcium imaging experiments on mouse TG neurons , a maximal number of neurons from nine mice isolated on 5 independent days were analyzed . Since highly significant results were obtained from this set of experiments , no further animals were sacrificed . | Sensory neurons in our skin detect cues from the environment – such as temperature and touch – and pass the information onto other cells in the nervous system . A protein called TRPM8 in sensory neurons is responsible for our ability to detect cool temperatures . TRPM8 sits in the membrane that surrounds the cell and forms a channel that can allow sodium and calcium ions to enter the cell . Cold temperatures activate TRPM8 , which opens the channel and triggers electrical activity in the sensory neurons . Chemicals that cause a cold sensation – such as menthol , the refreshing substance found in mint plants – can also open the TRPM8 channel . Janssens , Gees , Toth et al . investigated how menthol , and another natural compound called mustard oil , influence the opening of TRPM8 . The experiments show that menthol and mustard oil both stimulate sensory neurons by opening the TRPM8 ion channel , but using different mechanisms . Mustard oil forces the channel to open faster than it normally would , whereas menthol prevents the channel from closing . Further experiments show that these mechanisms explain why some compounds stimulate sensory neurons more strongly than others . The findings of Janssens , Gees , Toth et al . will help to understand how chemicals act on this class of ion channels , and how this affects the roles of the ion channels in cells . Altering the activity of TRPM8 and related ion channels may help to reduce pain in humans so a future challenge is to use these new insights to develop drugs that target these channels more efficiently . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics",
"neuroscience"
] | 2016 | Definition of two agonist types at the mammalian cold-activated channel TRPM8 |
The genome forms specific three-dimensional contacts in response to cellular or environmental conditions . However , it remains largely unknown which proteins specify and mediate such contacts . Here we describe an assay , MAP-C ( Mutation Analysis in Pools by Chromosome conformation capture ) , that simultaneously characterizes the effects of hundreds of cis or trans-acting mutations on a chromosomal contact . Using MAP-C , we show that inducible interchromosomal pairing between HAS1pr-TDA1pr alleles in saturated cultures of Saccharomyces yeast is mediated by three transcription factors , Leu3 , Sdd4 ( Ypr022c ) , and Rgt1 . The coincident , combined binding of all three factors is strongest at the HAS1pr-TDA1pr locus and is also specific to saturated conditions . We applied MAP-C to further explore the biochemical mechanism of these contacts , and find they require the structured regulatory domain of Rgt1 , but no known interaction partners of Rgt1 . Altogether , our results demonstrate MAP-C as a powerful method for dissecting the mechanistic basis of chromosome conformation .
The three-dimensional organization of the genome within the nucleus is structured but dynamic ( Bonev and Cavalli , 2016 ) . Although many features of this conformation are largely conserved across cell types and conditions ( Rao et al . , 2014; Schmitt et al . , 2016a ) , some chromatin loops and contacts form specifically in response to signals such as differentiation ( Bonev et al . , 2017; Monahan et al . , 2019; Schmitt et al . , 2016a; Stadhouders et al . , 2018 ) , changes in nutrient availability ( Brickner et al . , 2015; Brickner et al . , 2016 ) , heat shock ( Chowdhary et al . , 2019; Chowdhary et al . , 2017 ) , drugs ( D'Ippolito et al . , 2018 ) , meiosis ( Muller et al . , 2018 ) , or circadian rhythms ( Kim et al . , 2018 ) . This dynamic three-dimensional organization of the genome plays a role in regulating gene expression in diverse organisms . In multicellular organisms , active developmental gene promoters form long-range loops with specific enhancer elements ( Bonev et al . , 2017 ) , and this looping is in some cases sufficient for transcriptional activation ( Deng et al . , 2014 ) . In the budding yeast Saccharomyces cerevisiae , a well-studied model of genome conformation , genes targeted to nuclear pores are activated ( Taddei et al . , 2006 ) , while those at the nuclear periphery are repressed ( Andrulis et al . , 1998 ) . Transcription factors ( TFs ) are attractive candidates for orchestrating such dynamic changes in chromatin conformation , given their site-specific DNA binding and changes in abundance or activity in response to differentiation and cellular signals ( Lambert et al . , 2018 ) . For many conditions , it remains unknown exactly which TFs bind to any given locus . Although binding site motifs are known for many TFs , motif searches poorly predict TF binding ( Guertin and Lis , 2010; Jolma et al . , 2015; Le et al . , 2018; Levo et al . , 2015; Liu et al . , 2006; Slattery et al . , 2014 ) . Even if the set of TFs bound to each locus is known , it is unclear which TFs are capable of forming chromosomal contacts . DNA-bound TFs can also recruit other cofactor proteins that can mediate chromosomal contacts ( Deng et al . , 2012; Monahan et al . , 2019; Song et al . , 2007 ) , but our understanding of TF-cofactor interactions remains incomplete . Among chromosomal contacts and loops , interchromosomal contacts are less well-understood . This is in part due to the relative paucity of interchromosomal contacts in Hi-C and other 3C ( chromosome conformation capture ) data , which results from their greater contact distance ( Maass et al . , 2018 ) and chromosomal self-association into territories ( Cremer and Cremer , 2010 ) . Nevertheless , many distinct classes of interchromosomal contacts are known , including clustering of transcriptionally active genes ( Mitchell and Fraser , 2008; Osborne et al . , 2004; Schoenfelder et al . , 2010 ) , associations with nuclear bodies ( Quinodoz et al . , 2018 ) , interactions among developmental enhancers and promoters ( Lomvardas et al . , 2006; Monahan et al . , 2019 ) , and mitotic homologous chromosome pairing in organisms ranging from yeast ( Burgess et al . , 1999 ) to flies ( Henikoff and Dreesen , 1989; Joyce et al . , 2016; Morris et al . , 1999 ) and mammals ( Xu et al . , 2006 ) . However , our understanding of the molecular mechanisms of these contacts remains incomplete . Many known mechanisms for establishing 3D chromosome conformation may act on both intrachromosomal loops and interchromosomal contacts . The emerging consensus model for such DNA-DNA interactions involves loop extrusion by cohesin and other Structural Maintenance of Chromosomes ( SMC ) factors , which is thought to primarily mediate intrachromosomal loops ( Alipour and Marko , 2012; Rao et al . , 2014; Rowley and Corces , 2018; Sanborn et al . , 2015; Swygert et al . , 2019 ) . However , SMC complexes are also capable of mediating interactions between multiple DNA molecules , such as between sister chromatids ( Michaelis et al . , 1997 ) . Interchromosomal contacts ( Monahan et al . , 2019 ) and intrachromosomal contacts ( Weintraub et al . , 2017; Deng et al . , 2012 ) can also be mediated by dimerization of structured proteins . In addition to these well-defined strong molecular interactions , weak interactions such as those underlying phase separation of nuclear factors such as transcription factors ( Boija et al . , 2018; Chong et al . , 2018 ) , coactivators ( Cho et al . , 2018 ) , RNA polymerase ( Boehning et al . , 2018 ) , and heterochromatin proteins ( Larson et al . , 2017; Strom et al . , 2017 ) may play an important role in shaping 3D genome organization . How these and other mechanisms synergize remains an open question . Mitotic ( or somatic ) homologous chromosome pairing is the preferential association of homologous pairs of loci in mitotically dividing cells . Homolog pairing occurs along the length of the genome in Drosophila ( Joyce et al . , 2016 ) , but is more subtle in yeast and other organisms , where the association is often transient and/or genomically localized ( Xu et al . , 2006 ) . Fluorescence in situ hybridization screens in flies have nominated various pairing and anti-pairing factors that modulate the strength of homolog pairing ( Joyce et al . , 2012 ) , but the precise mechanisms by which these factors regulate pairing are largely unknown . In mammals , X chromosome pairing is mediated by CTCF and Oct4 ( Donohoe et al . , 2009 ) , in conjunction with transcription ( Xu et al . , 2007 ) . However , cases of highly localized homolog pairing remain rare . Furthermore , the distinctions between homolog pairing and non-allelic interactions between repetitive elements ( Gladyshev and Kleckner , 2017; Mirkin et al . , 2014 ) remain unclear . We recently identified a novel example of an inducible , localized interchromosomal contact between homologous copies of the HAS1pr-TDA1pr locus in diploid Saccharomyces yeasts ( Kim et al . , 2017 ) . This interaction occurs in saturated culture conditions , requires the 1 kb intergenic region between the HAS1 and TDA1 coding sequences , and is detectable by both Hi-C and microscopy . The condition-specificity and dependence on intergenic sequence led us to hypothesize that one or more TFs might mediate this pairing . Although yeast TF binding is well-characterized for standard growth conditions ( Badis et al . , 2008 ) , TF binding has not been systematically measured in saturated culture conditions . Meanwhile , computational predictions of TF binding sites are insufficiently specific: for example , within the 1 kb region required for HAS1pr-TDA1pr pairing , dozens of TFs have at least one motif match . Furthermore , even if the TFs bound to this region were known , it would remain unclear which subset played a role in mediating inducible interchromosomal pairing . Here we describe a method that enables the simultaneous testing of hundreds of cis or trans-acting mutations for their effects on a chromosomal contact of interest . As a proof of concept , we applied this method , which we call Mutation Analysis in Pools by Chromosome conformation capture ( MAP-C ) , to characterize the molecular components mediating HAS1pr-TDA1pr pairing . We first perform saturating mutagenesis of the regulatory region that mediates the interchromosomal pairing ( cis MAP-C ) to identify sequence motifs required for pairing that potentially correspond to TF binding sites . We then test the effects of knocking out over one hundred TFs ( trans MAP-C ) , and confirm that three—Leu3 , Sdd4 ( Ypr022c ) , and Rgt1—are necessary for inducible interchromosomal pairing . We verify their binding by chromatin immunoprecipitation , and find that HAS1pr-TDA1pr exhibits the strongest coincident , combined binding by all three factors across the genome in a condition-specific manner . We further use trans MAP-C to interrogate how interaction partners and domains of Rgt1 regulate pairing . Finally , we make an initial attempt to characterize the functional consequences of HAS1pr-TDA1pr pairing . Taken together , our results demonstrate how a combination of TFs can mediate inducible interchromosomal pairing . Furthermore , our study shows the utility of a pooled mutant approach to studying both the cis and trans dependencies of chromosome conformation .
In order to identify and dissect the molecular mechanisms underlying chromosome conformation , experiments involving perturbations ( e . g . mutations ) are needed . However , despite numerous advances in chromosome conformation capture ( 3C ) technology over the last two decades ( de Wit and de Laat , 2012; Schmitt et al . , 2016b ) , each experiment characterizes a single sample , which limits the number of genes or cis-regulatory elements that can be disrupted ( Monahan et al . , 2019; Nora et al . , 2017; Schwarzer et al . , 2017; Weintraub et al . , 2017 ) . To address this limitation and enable systematic screens , we developed MAP-C , an assay in which hundreds of mutations are simultaneously tested for their effects on a single chromosomal contact of interest ( Figure 1A ) . In the cis version of MAP-C , which we describe first , these mutations are targeted to one of the regions involved in the chromosomal contact . In the trans version of MAP-C , mutations can be spread across the genome , as long as they are associated with a unique barcode sequence at the chromosomal contact site . The first step of cis MAP-C is to generate an allelic series of a region of interest , which can be achieved in a cost-effective manner via array-synthesized oligonucleotide pools or error-prone PCR . Alternatively , if desired , variants can be generated individually and then pooled prior to conducting MAP-C . The resulting mutant pool is then integrated into the genome and subjected to the 3C assay ( Dekker et al . , 2002 ) . The region containing the genetic variants is amplified using two different primer pairs: the first ( 3C library ) amplifies a specific ligation product , and the second ( genomic library ) amplifies regardless of ligation . These amplification products are deeply sequenced to measure the abundance of each variant in the 3C library , which is normalized to its abundance in the genomic library . The relative extent to which sequence variants participate in the chromosomal contact of interest is proportional to their normalized representation in the 3C library . As a first test of cis MAP-C , we sought to systematically dissect the conserved pairing between HAS1pr-TDA1pr homologs in diploid Saccharomyces yeasts grown to saturation . We recently used Hi-C of S . cerevisiae x S . uvarum hybrids ( <80% nucleotide identity ) to discover this homolog pairing interaction , and furthermore identified a 1 , 038 bp noncoding region that was necessary and sufficient for pairing ( Kim et al . , 2017 ) . To find a minimal subsequence of the 1 , 038 bp HAS1pr-TDA1pr region that is sufficient to pair with other HAS1pr-TDA1pr alleles , we replaced the native S . cerevisiae HAS1pr-TDA1pr locus with a library containing each of 861 tiling 178 bp subsequences of the 1 , 038 bp region ( along with a G418 resistance cassette and restriction site ) , in S . cerevisiae x S . uvarum hybrid yeast . We then performed cis MAP-C for pairing of the modified locus with the S . uvarum copy of HAS1pr-TDA1pr on a saturated culture of the pool , in two replicates ( Figure 1B ) . Compared to the genomic libraries , the 3C libraries were highly enriched for a narrow region spanning ~ 500 to ~ 700 bp from the HAS1 coding sequence , with a plateau between ~ 525 to ~ 675 bp , consistent with that ~ 150 bp region being the only subsequence shorter than 178 bp sufficient for pairing ( termed the ‘minimal pairing region’ below ) . To confirm that this pattern of enrichment is specific to HAS1pr-TDA1pr homolog pairing rather than underlying all of its chromosomal contacts , we repeated the assay with a pair of primers amplifying a ~ 10 kb intrachromosomal contact ( Figure 1—figure supplement 1A and B ) . In this ‘off-target’ control , coverage from the 3C library matched that of the genomic control , suggesting that most variants are capable of intrachromosomal looping , but only those containing the minimal pairing region are capable of interchromosomal pairing with the other HAS1pr-TDA1pr allele . Since our initial experiment was performed in the endogenous genomic context , other DNA sequences outside but near the HAS1pr-TDA1pr locus could be required in addition to the minimal pairing region . We therefore tested whether inserting a 184 bp sequence that included the minimal pairing region into an ectopic location , the gene FIT1 ( YDR534C ) , would induce pairing with the native HAS1pr-TDA1pr locus in saturated cultures of haploid S . cerevisiae ( Figure 1—figure supplement 1C ) . As a negative control , we inserted an equivalently sized subsequence insufficient for pairing into the same locus ( Figure 1—figure supplement 1C ) . Indeed , insertion of the minimal pairing region led to a > 30 fold increase in 3C signal for pairing with HAS1pr-TDA1pr as compared to the negative control ( Figure 1C ) . We next sought to obtain a base-pair resolution map of the DNA sequences necessary for pairing . We used error-prone PCR to generate variants of a 207 bp region ( 161 bp excluding fixed primer sequences ) containing the minimal pairing region , with an average of 1 . 49 substitutions ( range 0–14 ) per template ( Figure 1—figure supplement 1D ) . We inserted this variant library in place of the native S . cerevisiae HAS1pr-TDA1pr sequence as before , and performed cis MAP-C . The ratio of total substitution abundance in the 3C and genomic libraries can be plotted at each mutagenized position . This identified six clusters of two or more adjacent positions showing strong depletion of substitutions in the 3C libraries , indicating that they are required for HAS1pr-TDA1pr pairing ( Figure 1D ) . We inspected motifs in the region to identify candidate TFs that might mediate the pairing ( Figure 1—figure supplements 2 and 3 ) . Given the abundance of potential matches , we prioritized motifs with high-scoring matches , motifs with conserved positions corresponding to those most important for pairing , and motifs occurring in multiple clusters . The first two clusters together aligned to a Leu3 motif ( Figure 1E ) , the third to several similar motifs , including Sdd4 ( Ypr022c ) and Mig1 ( Figure 1F ) , and the last three to Rgt1 motifs in both orientations ( Figure 1G–I ) . These motifs , which span 47 bp , include 23 of the 24 positions most depleted for mutations in the 3C libraries ( Figure 1—figure supplement 1E ) , and all clusters of two or more adjacent such positions ( Figure 1D ) . None of these mutations had a strong effect on intrachromosomal looping ( Figure 1—figure supplement 1F ) , and all of the clusters were reproduced using an alternative mutagenesis strategy ( programmed 3 bp substitutions ) ( Figure 1—figure supplement 4 ) . Interestingly , a fourth Rgt1 motif and a second Sdd4/Mig1 motif mutagenized only in our validation experiment were not required for pairing , suggesting that either not all motifs in this region are bound by the same TFs or not all bound TFs are involved in mediating homolog pairing at this locus . Thus , using cis MAP-C , we identified a ~ 150 bp subsequence of HAS1pr-TDA1pr sufficient for pairing , containing five required TF motif occurrences . If these TF motifs are together sufficient for pairing , we would expect that 1 ) they are only observed in a cluster in the minimal pairing region and not elsewhere in the HAS1pr-TDA1pr , and 2 ) they are present in a cluster in the S . uvarum copy of this region , and potentially other Saccharomyces as well . Indeed , these motifs are clustered together only in the central region of HAS1pr-TDA1pr; remarkably , this pattern holds across all Saccharomyces species ( Figure 1—figure supplement 5 ) . Although we identified the TF motifs required for HAS1pr-TDA1pr homolog pairing at base-pair resolution with cis MAP-C , the redundancy among TF motifs made it difficult to definitively identify the TFs involved ( Figure 1F ) . To address this , we developed a modified version of MAP-C that is capable of assaying trans mutations spread across the genome , such as gene knockouts , for their effects on a specific chromosomal contact ( trans MAP-C ) . With trans MAP-C , each mutation is uniquely associated with a short barcode sequence near a region involved in the chromosomal contact of interest , which is then assayed by 3C . Mutations that affect pairing frequency modulate the abundance of their corresponding barcodes in the 3C products , which can be readily quantified by deep sequencing . To study the trans requirements of HAS1pr-TDA1pr pairing , we ectopically integrated barcoded versions of the minimal pairing region , wherein the barcode identifies which TF is knocked out , and assayed their capacity to pair with the native copy of HAS1pr-TDA1pr ( Figure 2A ) . We first tested this approach in a pilot set of 10 TF knockouts by inserting the minimal pairing region ( Figure 1—figure supplement 1C ) into the common KanMX drug resistance cassette used to replace each deleted gene in the haploid yeast deletion collection ( Giaever et al . , 2002 ) . We then assayed these constructs for interactions with the native HAS1pr-TDA1pr region , using the existing gene-specific barcodes to measure strain abundances in each library . In this approach , because the pairing sequences are inserted into different genes throughout the genome , the frequency of pairing is confounded by the potential effects of the genomic location of the pairing sequence . Therefore , for each of the 10 TFs we targeted , we included as controls up to six neighboring genes , which should have a similar genomic location effect as the targeted gene ( Figure 2—figure supplement 1A and B ) . We hypothesized that due to the Rabl orientation of yeast chromosomes , in which centromeres are clustered together ( Duan et al . , 2010 ) , centromere-proximal regions would interact less with HAS1pr-TDA1pr , which is centromere-distal . Indeed , the most centromere-distal gene knockouts interacted ~ 4 fold more with the HAS1pr-TDA1pr locus than the most centromere-proximal gene knockouts ( Figure 2—figure supplement 1C ) . Of the 7 TF gene knockouts that were measured in our assay ( three dropped out during library construction , including LEU3 ) , six had no substantial difference in pairing strength compared to their genomic neighbors ( Figure 2—figure supplement 1D ) . However , deletion of RGT1 led to a ~ 20 fold decrease in pairing strength , consistent with Rgt1 binding its cognate motifs in the minimal pairing region ( Figure 2B and C ) . Next , we expanded our trans MAP-C screen to include the majority of known nonessential TFs with known binding motifs ( de Boer and Hughes , 2012 ) . To avoid the confounding effect of genomic location , we inserted a barcoded pairing sequence construct into a fixed locus instead of into the gene knockout locations . We associated each of these barcodes with the cognate knockout by individually transforming each knockout strain with a unique barcode , in a 96-well plate format . We tested a total of 109 TF gene knockouts , as well as 15 nuclear pore complex components , eight fitness neutral negative controls , and a wild-type control , with multiple barcode replicates for controls and expected hits ( Figure 2—figure supplement 2A ) . As expected , most barcoded strains were equally abundant in the 3C and genomic libraries , indicating that the corresponding knockout did not impact HAS1pr-TDA1pr pairing . However , LEU3 and RGT1 knockouts were depleted ~ 4 fold from the 3C libraries , suggesting that they are required for pairing ( Figure 2B ) . In addition , deletion of MOT3 modestly decreased pairing ( ~2 . 5 fold ) ; however , its binding motif is not present in the HAS1pr-TDA1pr region , suggesting an indirect role . Two other knockouts , VHR1 and CBF1 , appeared to also decrease pairing , but were at low abundances and might reflect noise ( Figure 2—figure supplement 2B ) . Surprisingly , none of the TFs required for pairing in either trans MAP-C experiment had a motif matching the sequence CCCCAC ( the third cluster of positions required for pairing; Figures 1F and 2B , and Figure 2—figure supplement 1C ) . However , two putative TFs with high scoring motif matches , YPR022C ( SDD4 ) and YGR067C , were excluded in the initial screens due to their lack of annotations . Therefore , we repeated our fixed-locus TF knockout screen with a limited set of genes , including the two putative TFs and additional replicates for the hits MOT3 , VHR1 , and CBF1 . We found that SDD4 is indeed required for pairing , suggesting that it is the trans-acting factor that binds the CCCCAC motif ( Figure 2C ) . MOT3 once again exhibited modest depletion , suggesting a minor , perhaps indirect , role in pairing , whereas VHR1 and CBF1 displayed no depletion ( Figure 2C ) . The strong concordance between the genomic base-pairs required for pairing ( identified by cis MAP-C ) and the DNA binding motifs of trans factors required for pairing ( identified by trans MAP-C ) suggests that Leu3 , Sdd4 , and Rgt1 bind the HAS1pr-TDA1pr to mediate pairing . To test this hypothesis , we performed chromatin immunoprecipitation ( ChIP ) for the tandem affinity purification ( TAP ) tagged versions of Leu3 , Sdd4 , and Rgt1 ( Ghaemmaghami et al . , 2003 ) , in haploid S . cerevisiae yeast under both saturated and exponential growth conditions . By qPCR using two different primer pairs , all three TFs strongly bound to the HAS1pr-TDA1pr pairing region in saturated conditions , but near background levels in exponential growth ( Figure 2D and E ) . We then performed ChIP sequencing to determine TF binding genome-wide . Consistent with our quantitative PCR measurements , all three TFs showed robust ChIP-seq peaks at HAS1pr-TDA1pr in saturated conditions . In contrast , only Leu3 demonstrated a significant peak in exponential growth , with 20-fold weaker enrichment ( 2 . 5-fold vs . 51-fold in saturated conditions; Figure 2F ) . For all three TFs , saturated conditions produced ChIP-seq peaks with greater enrichments over the input controls and more robust enrichment of the expected motifs , suggesting generally more extensive DNA binding ( Figure 2—figure supplements 3 and 4 ) . These global trends could be a result of technical artifacts as well as biological differences; however , all samples were treated identically using a protocol not optimized for saturated culture conditions , and the HAS1pr-TDA1pr peak showed particularly strong condition-specificity ( Figure 2—figure supplement 3 ) . Based on the convergence of cis MAP-C , trans MAP-C , and ChIP-seq data , we conclude that Leu3 , Rgt1 , and Sdd4 directly bind to both alleles of the HAS1pr-TDA1pr minimal pairing region under saturated growth conditions and thereby mediate inducible interchromosomal contacts between them . We next explored whether the combination of Leu3 , Sdd4 , and Rgt1 binding explains the uniqueness of HAS1pr-TDA1pr pairing . If the clustered binding of Leu3 , Sdd4 , and Rgt1 is necessary and sufficient to cause pairing in saturated culture conditions , either 1 ) no loci other than HAS1pr-TDA1pr should have all three TFs bound on both the S . cerevisiae and S . uvarum copies , or 2 ) other loci that do have all three TFs bound should also exhibit pairing . Because DNA binding data was only available for the S . cerevisiae genome , we tested the first possibility by scanning the S . cerevisiae and S . uvarum genomes for clusters of the three motifs using permissive thresholds for motif matches and allowing up to 200 bp between motifs ( see Materials and methods ) , and then assessed ChIP-seq data at these clusters . The promoters of four genes , TDA1 but also HXT3 , YKR075C , and SKS1 , harbored a motif cluster containing all three motifs in both S . cerevisiae and S . uvarum ( Figure 3A ) . We also assessed two additional loci , MIG1pr and ILV2pr , that lacked one of the three motifs in S . uvarum . Of these six loci , TDA1pr exhibited the strongest total ChIP-seq signal in saturated conditions ( Figure 3B and C ) , even when extending the search to motif clusters with only one or two of the three motifs in S . cerevisiae ( Figure 3—figure supplement 1 ) . Furthermore , YKR075Cpr was the only other locus with robust binding of all three TFs in S . cerevisiae , but the only Leu3 motif in the S . uvarum copy of the region overlaps a stronger Rgt1 motif and may not result in Leu3 binding . We also noticed that the TDA1pr cluster was the most compact ( i . e . shortest maximum distance between motifs ) ; whether this plays a role in pairing is unclear . We infer that the combinatorial binding of Leu3 , Sdd4 , and Rgt1 on both homologs specifies inducible homolog pairing in saturated cultures . Although TDA1pr exhibits the strongest combinatorial TF binding , we wondered whether any other motif clusters pair inducibly like the HAS1pr-TDA1pr locus , albeit perhaps more weakly . To address this question , we leveraged our previously published Hi-C datasets ( Kim et al . , 2017 ) , along with new Hi-C experiments for the high-pairing strain background in which we performed our pooled mutant experiments , to compare the strength of pairing at each homologous motif cluster and assess whether this pairing is condition-specific ( Figure 3—figure supplement 2 ) . We observed that HXT3pr appears to form inducible homologous contacts in saturated culture conditions ( Figure 3D ) , despite weak Sdd4 and no Leu3 binding . We did not detect pairing at any other loci ( Figure 3—figure supplement 2 ) . Across several strain backgrounds , the HXT3 promoters exhibited 1 . 6- to 2 . 7-fold increased interaction frequencies compared to other similar interchromosomal pairs of loci ( at least 480 kb from a centromere and excluding subtelomeric regions ) in saturated culture conditions ( Figure 3E ) , but only at baseline levels during exponential growth in rich medium . This is weaker than the pairing between HAS1pr-TDA1pr alleles ( Figure 3—figure supplement 2 ) , suggesting that combinatorial TF binding is not strictly necessary for inducible pairing but may facilitate particularly strong pairing . The identification of two pairs of loci , HAS1pr-TDA1pr and HXT3pr , exhibiting homolog pairing opened the possibility that there might be cross-pairing ( nonhomologous contacts ) between HAS1pr-TDA1pr and HXT3pr . To test this possibility , we extracted the interaction frequencies among the four loci ( i . e . two pairs ) from our Hi-C data . In all saturated culture datasets where homolog pairing was present , inter-locus interactions were substantially weaker and similar in frequency to those in non-pairing conditions ( Figure 3—figure supplement 1 ) . Together with our previous experiment demonstrating that two identical copies of the minimal pairing region can pair even at non-allelic locations in haploid S . cerevisiae ( Figure 1C ) , these data suggest that the interchromosomal pairing mediated by Leu3 , Sdd4 , and Rgt1 is sequence-specific beyond the simple presence of the same TF binding sites . We next sought to explore the mechanisms that regulate pairing . We hypothesized that TF expression levels might regulate the strength of HAS1pr-TDA1pr pairing . To test this hypothesis , we analyzed RNA-seq data for haploid S . cerevisiae in pairing and non-pairing conditions ( saturated and exponentially growing cultures , respectively ) ( Kim et al . , 2017 ) . RGT1 and SDD4 were upregulated in saturated cultures , ~ 2 fold and ~ 6 fold , respectively , whereas LEU3 transcript levels remained constant ( Figure 4A ) . These results are consistent with the hypothesis that increased transcription of the TFs mediating pairing , particularly RGT1 , regulates the strength of the pairing interaction . Based on these results , we wondered whether overexpression of any one of the three proteins would be sufficient to produce pairing in non-saturated culture conditions . We used the Z3EV estradiol induction system ( McIsaac et al . , 2014 ) to individually overexpress S . cerevisiae Leu3 , Sdd4 , or Rgt1 , and measured pairing between the native HAS1pr-TDA1pr loci using 3C in S . cerevisiae x S . uvarum hybrids growing exponentially in rich medium ( Figure 4B ) . In all three strains , a 2 hr estradiol induction led to no increase in pairing strength despite between 2- and 10-fold increases in transcript levels ( Figure 4—figure supplement 1A ) . As an alternative test , we used galactose induction to overexpress epitope-tagged RGT1 , and observed a decrease in pairing strength relative to a strain lacking the overexpression cassette ( Figure 4—figure supplement 1B ) . These results are consistent with no single TF being sufficient for pairing; however , Leu3 and Rgt1 are both known to change in conformation ( Sze et al . , 1992 ) or phosphorylation state ( Kim et al . , 2003 ) in different conditions , so it remains possible that overexpression of a single TF in the appropriate state suffices for pairing . Rgt1 is known to interact with several cofactors that affect its DNA binding and transcriptional repression activities: the Tup1/Ssn6 co-repressor complex and the proteins Mth1 and Std1 ( Polish et al . , 2005 ) . Our experiments thus far had not distinguished whether Rgt1’s pairing activity is directly mediated by physical interactions among molecules of Rgt1 or indirectly , through these or other interaction partners . To address this , we performed trans MAP-C for individual deletions of these four interacting partners of Rgt1 , along with the same positive and negative controls as before ( Figure 4C ) . In addition , based on the glutamine and asparagine-rich domains present in Sdd4 and Rgt1 , we tested deletion of RNQ1 , a Q/N-rich peptide known to influence the oligomerization of other Q/N-rich proteins ( Derkatch et al . , 2004 ) . Deletion of RNQ1 had no effect on pairing , suggesting that HAS1pr-TDA1pr pairing is not mediated by Q/N-rich domains . Deletion of TUP1 or SSN6 both led to increased pairing , indicating that the recruitment of the Tup1/Ssn6 co-repressor complex inhibits pairing . This is consistent with its known inhibition of Rgt1’s DNA binding activity ( Roy et al . , 2013 ) , which is presumably required for pairing . Deletion of MTH1 or STD1 had a minimal negative effect on pairing . These results suggest that the role of Rgt1 in pairing is not simply to recruit cofactors that mediate pairing; instead , its pairing activity may compete with its transcriptional repression activity . We hypothesized that a particular domain of Rgt1 , coupled with its DNA-binding activity , might be responsible for HAS1pr-TDA1pr pairing . To test this idea , we generated a series of 10 amino acid deletions spanning most of the Rgt1 protein ( positions 91–1030 , excluding several deletions that dropped out during strain construction ) and performed trans MAP-C to test the pairing function of the Rgt1 mutants ( Figure 4D ) . Surprisingly , much of the C-terminal half of Rgt1 was required for HAS1pr-TDA1pr pairing . These ‘pairing domains’ are closely aligned to the regions of Rgt1 predicted to be highly structured , largely through alpha helices ( Figure 4D ) . The regions of Rgt1 required for pairing also correspond loosely to those required for regulation of activation vs . repression function through allosteric changes in protein conformation in response to glucose-regulated phosphorylation ( Polish et al . , 2005 ) ( Figure 4—figure supplement 2 ) . To further compare the roles of Rgt1 protein domains and phosphorylation in HAS1pr-TDA1pr pairing , we generated deletions of the Rgt1 zinc finger , Q/N-rich , and C-terminal domains , and phosphodepletion mutants S88A and S758A , and performed trans MAP-C to test their pairing function ( Figure 4E and F ) . As expected , deletion of the zinc finger DNA-binding domain or the C-terminal domain led to background pairing levels , equivalent to lacking Rgt1 altogether . Surprisingly , deletion of the Q/N-rich domain led to stronger pairing than the wild-type control ( with RGT1 integrated at the same ectopic location ) , suggesting that the Q/N-rich domain inhibits pairing . Both the S88A and S758A mutations had a weak decrease in pairing , indicating that although these mutations are capable of disrupting Rgt1 activator function and intramolecular interactions ( Polish et al . , 2005 ) , they do not individually play major roles in regulation of pairing in saturated cultures . Together , our results suggest multiple potential modes by which the transcription factor Rgt1 regulates HAS1pr-TDA1pr pairing: its own expression level , recruitment of Tup1/Ssn6 , and the competing activities of its Q/N-rich and C-terminal regulatory domains . We have thus far characterized the molecular mechanism and regulation of HAS1pr-TDA1pr homologous pairing in saturated cultures . However , the question of why this pairing occurs ( i . e . what biological function it serves , if any ) remains outstanding . Now that we have identified the precise DNA base pairs and proteins involved in HAS1pr-TDA1pr pairing , we are equipped to test whether Leu3 , Sdd4 , and Rgt1 play a role in transcription at HAS1 or TDA1 in saturated cultures . Furthermore , we can employ S . cerevisiae x S . uvarum hybrids , in which we can discriminate the two homologous copies of each gene , in order to distinguish the effects on transcription of disrupting the pairing DNA sequence in cis ( presumably through impaired cis regulation ) vs . in trans ( presumably through impaired pairing ) . To this end , we generated mutant hybrid strains carrying a wild-type copy of S . uvarum HAS1pr-TDA1pr and a copy of S . cerevisiae HAS1pr-TDA1pr with either a wild-type ( WT ) genotype , mutated Leu3 binding site ( leu3 ) , or two mutated Rgt1 sites ( rgt1 × 2 ) ( Figure 5A ) . We then performed RNA-seq in saturated culture conditions . Of note , neither of these mutants is expected to have complete loss of pairing , since the individual binding site or TF mutants disrupt pairing by ~ 4–10 fold ( Figure 1D , Figure 2B , C ) , as opposed to the ~ 30 fold increase in pairing upon ectopic insertion of the pairing sequence ( Figure 1C ) ; however , if the Leu3 and Rgt1 binding site mutations cause the same effect on HAS1 or TDA1 transcription , the orthogonal nature of these perturbations would add confidence that any resulting molecular phenotype is an effect of disrupting pairing . The Rgt1 binding site mutations led to higher levels of S . cerevisiae TDA1 , consistent with Rgt1 acting as a repressor under low glucose conditions ( Figure 5B ) , but had no significant effect on either S . uvarum HAS1 or TDA1 . Surprisingly , despite the conserved binding site and strong ChIP-seq signal for Leu3 in HAS1pr-TDA1pr , disrupting its binding site had no significant effect on the transcript levels of any gene ( Figure 5C ) . To further characterize the transcriptional roles of Leu3 , Sdd4 , and Rgt1 in saturated culture , we also performed RNA-seq on deletion strains for each of these genes , along with a wild-type control , grown to saturation . As expected , each deleted gene was highly down-regulated in the mutant strains ( Figure 5—figure supplement 1A ) . In addition , many known targets of Rgt1 and genes downstream of ChIP-seq peaks were highly upregulated in rgt1∆ ( Figure 5—figure supplement 2 ) . Focusing on the HAS1-TDA1 locus , none of the deletion strains had altered HAS1 expression , and only the rgt1∆ strain had even slightly altered TDA1 expression ( Figure 5D ) . These results are consistent with our earlier results from mutating TF binding sites , and together indicate that despite the importance of Leu3 , Sdd4 , and Rgt1 binding for pairing at the HAS1pr-TDA1pr locus in saturated culture conditions , they do not play a major role in transcriptional regulation as detected by standard polyA mRNA sequencing .
In summary , we developed Mutation Analysis in Pools by Chromosome conformation capture ( MAP-C ) , a method to simultaneously test hundreds of mutations for their effects on a chromosomal contact of interest . MAP-C can be used to identify the precise sequences that are necessary for a contact ( cis MAP-C ) as well as the factors that are necessary to mediate the contact ( trans MAP-C ) . Here we applied both versions of MAP-C to dissect the mechanism of inducible interchromosomal pairing between HAS1pr-TDA1pr alleles in budding yeast . Using a combination of gain-of-function and loss-of-function screens , we demonstrate that a trio of transcription factors—Leu3 , Sdd4 , and Rgt1—mediates pairing between clusters of binding sites . Our results begin to elucidate the mechanisms of condition-specific interchromosomal contacts and homolog pairing , which have often been elusive ( Mirkin et al . , 2013 ) . We have not yet fully defined the biochemical mechanism of pairing—it is possible the TFs interact directly and/or indirectly—but so far , no known interaction partners or cofactors have proven essential for this interaction . Unlike more prevalent nuclear-pore mediated gene relocalization and homolog pairing ( Brickner et al . , 2012; Randise-Hinchliff and Brickner , 2016 ) , HAS1pr-TDA1pr pairing in saturated cultures does not appear to require the nuclear pore complex ( Figure 2B ) . Instead , the Tup1/Ssn6 repressor complex recruited by Rgt1 negatively regulates pairing ( Figure 4C ) , perhaps by inhibiting the DNA-binding activity of Rgt1 ( Roy et al . , 2013 ) . Within Rgt1 , the zinc finger DNA-binding domain and C-terminal regulatory domain are required for pairing , while the Q/N-rich region , which contains the activation domain ( Polish et al . , 2005 ) , inhibits pairing ( Figure 4F ) . These results point toward specific interactions among structured domains mediating pairing , rather than phase separation by intrinsically disordered regions . Intriguingly , Leu3 also contains a regulatory domain whose conformation responds to environmental cues ( Sze et al . , 1992; Wang et al . , 1997 ) . We speculate that these regulatory domains , which are capable of intramolecular interactions ( Polish et al . , 2005 ) , may mediate interchromosomal pairing . Furthermore , if these regulatory domains are indeed capable of direct dimerization or oligomerization , they may mediate not only interchromosomal contacts , but also cooperative TF binding at clusters of binding sites ( Kim et al . , 2003 ) . We note that of Leu3 , Sdd4 , and Rgt1 , only Leu3 is known to form strong dimers at each of its binding sites . However , even weak homotypic intermolecular interactions could increase the duration of interchromosomal contacts as well as increase local TF concentrations , and thereby increase DNA binding activity . It remains unclear whether heterotypic interactions among Leu3 , Sdd4 , and Rgt1 occur . Another question is whether the contacts are stoichiometric , for example one-to-one , or instead mediated by aggregation of one or more TFs via weak interactions . Notably , we observe homotypic pairing between homologs of HAS1pr-TDA1pr and HXT3pr but not heterotypic pairing ( Figure 3—figure supplement 3 ) , suggesting that Rgt1 does not indiscriminately form contacts between all of its binding sites . However , this is not because the entire homologous context is necessary for pairing , as we found that the pairing sequence from HAS1pr-TDA1pr is sufficient to induce ectopic pairing in haploid S . cerevisiae ( Figure 1C ) . Instead , a possible explanation is that only pairs of loci with a series of motifs in similar orders and orientations form frequent contacts , consistent with a stoichiometric interaction model . This added specificity beyond the presence of TF motifs could explain the lack of pairing between HAS1pr-TDA1pr and other sites of Leu3 , Sdd4 , and Rgt1 binding ( Figure 3—figure supplement 4 ) . Despite our detailed molecular characterization of HAS1pr-TDA1pr homolog pairing , we still do not know its biological function , if any . We hypothesized that it might contribute to transcriptional regulation at either HAS1 or TDA1 . In particular , TDA1 is a kinase that phosphorylates Hxk2 , the main hexokinase in yeast , and thereby inhibits its ability to mediate glucose repression together with Mig1 ( Kaps et al . , 2015; Kettner et al . , 2012 ) We speculated that pairing might help activate TDA1 , which is mildly upregulated at the transcriptional level in saturated conditions ( Kim et al . , 2017 ) . However , we find that Leu3 and Sdd4 play no detectable role in regulating transcript levels of either gene in saturated culture conditions , while Rgt1 mildly represses TDA1 in cis but not in trans ( Figure 5 ) . There are several possible reasons why we may not have detected a phenotypic effect from disrupting the pairing region or factors: ( 1 ) even low levels of pairing ( ~10–25% based on our MAP-C experiments shown in Figure 1D and Figure 2B , C ) are sufficient to mediate its biological function; ( 2 ) its function is not in saturated culture conditions per se , but affects either exit from or re-entry into those conditions , similar to transcriptional memory at GAL1 and INO1 ( Brickner et al . , 2015; Sood et al . , 2017 ) ; ( 3 ) its function only affects the dynamics of transcription ( or other processes ) but not steady-state RNA levels ( Zhang and Bai , 2016 ) . However , we also acknowledge the possibility that there is no function per se , but that pairing is instead a side-effect of conservation of biochemical features of the pairing TFs . If the same domains are required for both cooperative binding and pairing , then evolutionarily conserved pairing could result from selection on cooperative binding . Interestingly , the TF Hsf1 , which forms trimers , and in flies also exhibits cooperativity between trimers ( Xiao et al . , 1991 ) , was also recently shown to mediate interchromosomal contacts in yeast ( Chowdhary et al . , 2019 ) . More studies of the biochemical basis of cooperative TF binding are required to test this hypothesis . Is HAS1pr-TDA1pr pairing unique ? Based on Hi-C , we previously found HAS1pr-TDA1pr to exhibit the strongest interchromosomal interactions in saturated cultures of S . cerevisiae x S . uvarum hybrids , excluding centromeres , telomeres , and the chromosomes carrying the rDNA arrays ( Kim et al . , 2017 ) . Using ChIP-seq and motif analyses , we found that a requirement for robust adjacent DNA-binding by Leu3 , Sdd4 , and Rgt1 on both the S . cerevisiae and S . uvarum copies was sufficient to identify HAS1pr-TDA1pr . However , we also expect that the limited resolution of Hi-C and the sequence divergence in interspecific hybrids would both limit our sensitivity for detecting homolog pairing . We hypothesize that more cases of localized pairing exist , as we found with HXT3pr ( Figure 3D and E ) . We imagine that other TFs capable of interchromosomal contacts , like Hsf1 ( Chowdhary et al . , 2019 ) , are also capable of mediating homolog pairing , and the condition-specificity of these contacts suggests that other conditions may also exhibit similar contacts but have not yet been explored . Although our study was focused on HAS1pr-TDA1pr pairing in budding yeast , both cis and trans MAP-C should be applicable to other loci and organisms . The main constraints in experimental design are that 1 ) the introduced mutations or associated barcodes must be included in the 3C PCR product , and 2 ) the region of interest should not be digested by the restriction enzyme . As we have implemented this approach , the region targeted by saturation mutagenesis was limited to < 250 bp to allow for Illumina sequencing , but this could be extended using either barcode association , similar to our trans knockout screens , or long-read sequencing methods . Also , we focused on a single pair-wise interaction , but it is also possible to assay a mutant pool for multiple interactions , by using multiple primer pairs . MAP-C should be applicable to intrachromosomal contacts as well as interchromosomal ones , albeit with a potentially higher background of nonspecific contacts . It would be interesting to apply MAP-C to dissect the cis and trans regulators of enhancer-promoter loops in mammals , and thereby distinguish the contributions of cohesin and CTCF ( Guo et al . , 2012 ) , general looping factors like YY1 ( Weintraub et al . , 2017 ) , site-specific transcription factors ( Nolis et al . , 2009 ) , and other cofactors . Trans MAP-C could also allow mutational scanning of TFs to clarify the biochemical mechanisms by which these transcription factors mediate chromosomal contacts . MAP-C leverages the high throughput of saturation mutagenesis and mutant collections to allow systematic dissection of chromosome conformation . We tested up to ~ 1000 variants at a time , but with larger-scale experiments , it should be possible to test even more variants . A major potential strength of our approach is that unlike cellular high-throughput genetic screens ( Fowler and Fields , 2014; Gasperini et al . , 2016; Shalem et al . , 2015 ) , it resolves the functional consequences of mutations at the allelic level , and thus is not confounded by heterozygosity ( Patwardhan et al . , 2009 ) . As we continue to map chromosome conformation at high resolution across ever-expanding numbers of cell types and conditions , MAP-C will provide a scalable approach to dissect the molecular mechanisms underlying specific contacts .
Yeast strains used in this study are described in the Key Resources Table . Yeast were cultured at 30C , with the exception of S . uvarum strains , which were grown at room temperature . Cultures were grown shaking overnight to OD600 > 5 for saturated culture samples , or diluted to OD600 ~ 0 . 125 and grown to OD600 = 0 . 5–0 . 8 for exponential growth samples . Estradiol inductions were performed by addition of beta-estradiol to 1 µM final concentration ( or equivalent volume of ethanol for negative control ) to OD600 = 0 . 5 cultures grown in YPD ( 1% w/v yeast extract , 2% w/v peptone , 2% w/v dextrose ) and grown for 2 hr . Galactose induction was performed by growth in synthetic complete medium ( without uracil for selection for overexpression plasmid ) with 2% v/v raffinose to OD600 = 0 . 75 followed by addition of 2% galactose and subsequent growth for 1 . 5 hr . For comparison , yeast were grown in synthetic complete medium with or without uracil with 2% glucose or 2% raffinose . S . cerevisiae x S . uvarum hybrids were generated by standard mating and auxotrophic or drug selection procedures . Yeast transformations were performed using a modified Gietz LiAc method ( Pan et al . , 2004 ) . Cells were crosslinked by addition of 37% formaldehyde to a final concentration of 1% ( v/v ) and incubation at room temperature for 20 min , quenched by addition of 2 . 5M glycine to a final concentration of 150 mM and incubation at room temperature for 5 min , and then washed in 1x Tris-buffered saline ( TBS ) and stored as a pellet at −80C in aliquots of 50–100 µl dry pellets . Cells were lysed by vortexing in lysis buffer ( TBS + 1% Triton X-100 with Pierce EDTA-free protease inhibitor tablet ( Thermo Fisher Scientific , Waltham , MA ) ) with 500 µm acid-washed glass beads for 6 cycles of 2 min , with 2 min on ice between cycles . The lysate was collected by puncturing the bottom of each tube and then centrifuging the tube , stacked on top of an empty tube . The lysate was then washed in lysis buffer , then TBS , and finally resuspended in 10 mM Tris pH 8 . 0 to a volume of ~ 200 µl per 25 µl of starting dry pellet volume . A single 200 µl aliquot of lysate was then precleared by addition of 0 . 2% SDS and incubation at 65C for 10 min , cooled on ice , quenched by addition of 1% Triton X-100 ( v/v ) , and then digested overnight with at least 100U of restriction enzyme ( 200U DpnII for cis experiments and galactose induction , 400U EcoRI-HF for trans experiments , and 100U NlaIII for estradiol inductions ) . The restriction digest was heat-inactivated at 65C for 20 min in the presence of 1 . 3% SDS , and then chilled on ice and added to a dilute ligation reaction in 4 ml volume with 1% Triton X-100 , 1x T4 DNA Ligase Buffer ( NEB ) , and 10 , 000U of T4 DNA ligase ( NEB ) and incubated at room temperature for 4 hr . The ligation products were reverse-crosslinked with proteinase K at 65C overnight , and then purified by phenol-chloroform extraction followed by clean-up on a Zymo DNA Clean and Concentrator-5 column ( Zymo Research , Irvine , CA ) . The resulting 3C DNA was quantified using a Qubit . Each technical replicate was processed separately beginning with cell lysis . 3C libraries were prepared by amplification of up to 8 reactions of 50 ng 3C DNA per replicate , using primer pairs specific to the chromosomal contact of interest ( for pairing library ) or a control off-target chromosomal contact ( for off-target library ) , for 24–32 cycles . Genomic libraries were prepared by amplification of up to 4 reactions of 50 ng of either 3C DNA or genomic DNA using primers flanking the targeted mutations or barcodes , for 17–22 cycles . Reactions for each replicate were pooled , purified by Ampure XP beads ( Beckman Coulter Life Sciences , Brea , CA ) , and then re-amplified with primers flanking the mutagenized or barcode region and including sequencing adapter sequences for 5–8 cycles , and then again with primers adding sample indices and Illumina flow-cell adapters for 6–9 cycles . All reactions were prepared with KAPA HiFi HotStart ReadyMix ( Roche , Basel , Switzerland ) with recommended thermocycling conditions , and included 0 . 5x SYBR Green I to monitor amplification by quantitative PCR and minimize the number of PCR cycles . The final libraries were sequenced on an Illumina MiSeq or Nextseq 500 ( Illumina , San Diego , CA ) using paired-end sequencing . See Supplementary file 2 for detailed information on each library . Paired-end reads were merged and adapter-trimmed using PEAR ( Zhang et al . , 2014 ) , except for trans knockout experiments , in which only read one was used . These reads were then trimmed of the first 4 bp ( corresponding to a randomized region for Illumina clustering purposes ) and mapped using Bowtie 2 ( Langmead and Salzberg , 2012 ) Reads were mapped to the S . cerevisiae HAS1pr-TDA1pr region , and then the read coverage was calculated using bedtools ( Quinlan and Hall , 2010 ) . 3C DNA was amplified using the same conditions as MAP-C libraries , in three replicates per primer pair . The pairing 3C products were normalized to the off-target ( intrachromosomal ) 3C products , assuming 2-fold amplification per cycle . Either 50 ml of exponentially growing ( OD600 = 0 . 7–0 . 8 ) or 12 ml of saturated cultures of TAP-tagged strains ( Ghaemmaghami et al . , 2003 ) were crosslinked and lysed as for 3C , but for 15 min in FA lysis buffer ( 50 mM HEPES-KOH pH 7 . 5 , 140 mM NaCl , 1 mM EDTA , 1% Triton X-100 , 0 . 1% sodium deoxycholate , 1x Pierce EDTA-free protease inhibitor tablet ) . Lysates were pelleted and resuspended in FA lysis buffer , sonicated using a Diagenode Bioruptor ( Diagenode , Liege , Belgium ) for 3 cycles of 10 min on the HIGH power setting , with 30 s cycles on and 30 s on , and cleared by centrifugation at 20000 g for 10 min . An aliquot of 50 µl supernatant was saved for input , and the remaining sample was incubated with 10 µl of Dynabeads Pan Mouse IgG magnetic beads ( pre-washed with FA lysis buffer ) rotating overnight at 4C . The immunoprecipitations were washed twice with FA lysis buffer , once with high salt ( 500 mM NaCl ) FA lysis buffer , twice with RIPA buffer ( 10 mM Tris-HCl pH 8 , 250 mM LiCl , 0 . 5% Igepal CA-630 , 0 . 5% sodium deoxycholate , 1 mM EDTA ) , and once with TE ( 50 mM Tris-HCl pH 8 , 1 mM EDTA ) , and then eluted first with 100 µl TE + 1% SDS at 65C for 15 min , and a second time with 150 µl TE + 0 . 67% SDS . Inputs were diluted with 200 µl of TE + 1% SDS , and all samples were treated with 50 mg RNase A at 37C for 10 min and 100 mg proteinase K at 42C for 1 hr , and then reverse crosslinked overnight at 65C . DNA was purified with a Zymo ChIP DNA Clean and Concentrator-5 kit and eluted in 15 µl of 10 mM Tris-HCl pH 8 . qPCRs were performed using 5 µl of either 1:20 dilution of IP samples or 1:800 dilution of input samples , in 25 µl reactions with KAPA Robust 2G HotStart ReadyMix , using 0 . 5x SYBR Green I and standard cycling conditions for 40 cycles on a Bio-Rad C1000 Touch thermal cycler with a CFX96 Real-Time System ( Bio-Rad Laboratories , Hercules , CA ) . Cq values were calculated using the Bio-Rad CFX Manager 3 . 1 software , using the single threshold mode for Cq calculation and baseline subtracted curve fitting . Primer efficiencies were calculated using a 5-fold dilution series of genomic DNA from S . cerevisiae BY4741 starting from 20 ng . Primer sequences are included in Supplementary file 1 . ChIP-seq libraries were prepared using Swift Accel-NGS 2S Plus ( Swift Biosciences , Ann Arbor , MI ) dual-indexed kits using 10 µl of IP samples or 1 ng of input samples , with 9 cycles of PCR for input samples and 12–15 cycles for IP samples . Libraries were sequenced to ~ 2 . 5–6 million read pairs per sample using 2 × 37 bp reads on an Illumina NextSeq 500 . Sequencing reads were first pre-processed using cutadapt ( Martin , 2011 ) : reads were quality-trimmed ( option -q 20 ) , trimmed of adapter sequences , excluding any read pairs in which either read was shorter than 28 bp after trimming ( option -m 28 ) . Read pairs were mapped to the sacCer3 S . cerevisiae reference genome using Bowtie 2 ( Langmead and Salzberg , 2012 ) with the --very-sensitive parameter set , requiring the reads in each pair to be with 2000 bp of each other ( option -X 2000 ) . Read pairs in which both reads had a mapping quality score of at least 30 were deduplicated using samtools rmdup . Replicates were merged prior to calling peaks and generating fold enrichment tracks using MACS2 ( https://github . com/taoliu/MACS ) ( Zhang et al . , 2008 ) . Fold enrichment tracks were visualized using the UCSC Genome Browser ( Karolchik et al . , 2003 ) . Yeast strains were grown overnight in YPD in biological triplicate ( independent colonies , or for newly generated transformants , independent transformants ) , pelleted and then stored at −80C . RNA was purified using acid phenol extraction , and then treated with Turbo DNase and purified with a Qiagen RNeasy Mini kit ( Qiagen , Hilden , Germany ) . Illumina libraries were prepared from 800 to 900 ng of total RNA , using the Illumina Truseq RNA Library Prep kit v2 ( for TF binding site mutants ) or the Illumina Truseq Stranded mRNA Library Prep kit ( for TF knockouts ) . Libraries were sequenced to ~ 11–13 million read pairs per sample using 2 × 37 bp reads on an Illumina NextSeq 500 . Reads were preprocessed and mapped as with the ChIP-seq libraries , but with the -X 500 option for Bowtie 2 . Read pairs in which both reads had a mapping quality score of at least 30 were overlapped with annotated genes using HTSeq ( Anders et al . , 2015 ) . Global fold-change analyses were using DESeq2 ( Love et al . , 2014 ) . Yeast were grown as described above in Yeast strains and culture , and then pelleted and stored at −80C . RNA was purified using acid phenol extraction , and then treated with Turbo DNase and purified with a Qiagen RNeasy Mini kit . For each sample , 1 ug of total RNA was annealed to oligo ( dT ) 20 ( 13 µl reaction with 1 µl of 50 µM oligo ( dT ) , 1 µl 10 mM dNTP mix ) by incubating at 65C for 5 min and then on ice for 1 min . Reverse transcription was performed with SuperScript IV by adding 4 µl of 5x SuperScript IV buffer , 1 µl SUPERase In RNase inhibitor , 1 µl 100 mM DTT , and 1 µl of SuperScript IV enzyme , and then incubating at 50C for 10 min and then 80C for 10 min . For qPCRs , 2 . 5 µl of each reverse transcription reaction was used for each 25 µl PCR , using KAPA Robust 2G HotStart ReadyMix with standard cycling conditions ( except annealed at 55C ) for 40 cycles on a Bio-Rad C1000 Touch thermal cycler with a CFX96 Real-Time System . Cq values were calculated using the Bio-Rad CFX Manager 3 . 1 software , using the regression mode for Cq calculation and baseline subtracted curve fitting . Primer efficiencies were calculated using a 5-fold dilution series of genomic DNA from S . cerevisiae BY4741 starting from 10 ng . Primers were specific to S . cerevisiae ( at least two substitutions to S . uvarum ) . See Supplementary file 1 for primer sequences . Predicted intrinsic disorder was calculated using IUPred2 long disorder ( Mészáros et al . , 2018 ) . Predicted secondary structure was calculated using Jpred4 ( Drozdetskiy et al . , 2015 ) separately on the first 400 amino acids and on the remaining 770 amino acids , as submissions are capped at 800 amino acids . Systematic scans of motifs were performed using YeTFaSCo ( de Boer and Hughes , 2012 ) with the expert-curated no dubious motif set . Sequence logos were generated using ggseqlogo ( Wagih , 2017 ) . Motif clusters were identified using MCAST ( Grant et al . , 2016 ) with a motif p-value threshold of 0 . 001 , a maximum gap threshold of 200 bp , and an E-value threshold of 2 , using the high-confidence motifs for Leu3 ( #781 ) , Rgt1 ( #2227 ) , and Sdd4 ( #588 ) from YeTFaSCo ( de Boer and Hughes , 2012 ) . Individual motif occurrences for each TF were scored for the S . cerevisiae ( version R64 . 2 . 1 ) and S . uvarum genomes ( as revised in Kim et al . , 2017 ) using FIMO ( Grant et al . , 2011 ) with a p-value threshold of 0 . 001 and the option —max-strand . Motif clusters were named based on the nearest downstream gene ( on - strand if coordinate of gene < coordinate of motif cluster , and on + strand if coordinate of gene > coordinate of motif cluster ) . Motif clusters were defined to be homologous if they were upstream of homologous genes . De novo motif discovery was performed using MEME version 4 . 12 . 0 ( Bailey et al . , 2006 ) . For all analyses , 100 bp centered at each ChIP-seq peak or tRNA gene was used . Motifs were allowed to be between 6 bp and either 10 bp or 20 bp , and the top three motifs were analyzed , with otherwise default settings . Hi-C was performed and analyzed as in Kim et al . ( 2017 ) using the restriction enzyme Sau3AI . Code used to analyze data and generate figures are available at https://github . com/shendurelab/MAP-C ( Kim , 2019; copy archived at https://github . com/elifesciences-publications/MAP-C ) . All sequencing data have been deposited in the Gene Expression Omnibus ( GEO ) under accession number GSE118118 . Hi-C data from Figure 3 and Figure 3—figure supplements 2 , 3 and 4 and RNA-seq data from Figure 4 are from GEO accession number GSE88952 . Processed microarray data of gene expression in TF deletions under exponential growth from Figure 5—figure supplements 1 and 2 are from GEO accession number GSE4654 ( Hu et al . , 2007 ) . | Inside cells , genetic information is stored within molecules of DNA that are folded into three-dimensional structures known as chromosomes . Each fold in a chromosome forms when two points on a single DNA molecule link together to make a loop . DNA in two different chromosomes can also form links with each other ( known as “contacts” ) . Many cells contain two copies of every chromosome and these copies are often able to make contacts with each other . DNA loops and contacts can change in response to the environment and this may help cells switch the right genes on and off at specific times . For example , in budding yeast cells that have used up most of their preferred food source – a sugar called glucose – the two copies of a region of DNA known as the HAS1pr-TDA1pr region stick together . This may help the budding yeast cells switch on genes that are needed to make use of alternative sources of food . Cells contain hundreds of proteins called transcription factors that can bind to specific locations on DNA and can also stick to each other . These proteins are thought to be responsible for anchoring bridges between the DNA at most loops and contacts . One way to find out which transcription factors form specific DNA loops and contacts is to generate many different genetic mutations in the DNA and identify precisely which mutations disrupt the links . However , current methods can only test one mutation at a time , so it remains unclear how and why many segments of DNA stick together . Now , Kim et al . have developed a new method known as MAP-C to test how hundreds of mutations in budding yeast affect a particular DNA contact , in a single experiment . The MAP-C method was used to test which mutations within either the DNA segment involved in the contact , or in genes encoding transcription factors , prevent copies of the HAS1pr-TDA1pr region from forming contacts . This revealed that three transcription factors – Leu3 , Sdd4 , and Rgt1 – bridge contacts between the two copies of HAS1pr-TDA1pr . Mutations that disrupt the three-dimensional structure of chromosomes can cause cancer , developmental disorders and other diseases . The MAP-C method will allow researchers to better understand which transcription factors control how DNA is folded inside the cell , and which mutations change this folding . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"genetics",
"and",
"genomics"
] | 2019 | A combination of transcription factors mediates inducible interchromosomal contacts |
Olfactory associative learning in Drosophila is mediated by synaptic plasticity between the Kenyon cells of the mushroom body and their output neurons . Both Kenyon cells and their inputs from projection neurons are cholinergic , yet little is known about the physiological function of muscarinic acetylcholine receptors in learning in adult flies . Here , we show that aversive olfactory learning in adult flies requires type A muscarinic acetylcholine receptors ( mAChR-A ) , particularly in the gamma subtype of Kenyon cells . mAChR-A inhibits odor responses and is localized in Kenyon cell dendrites . Moreover , mAChR-A knockdown impairs the learning-associated depression of odor responses in a mushroom body output neuron . Our results suggest that mAChR-A function in Kenyon cell dendrites is required for synaptic plasticity between Kenyon cells and their output neurons .
Animals learn to modify their behavior based on past experience by changing connection strengths between neurons , and this synaptic plasticity is often regulated by metabotropic receptors . In particular , neurons commonly express both ionotropic and metabotropic receptors for the same neurotransmitter , where the two may mediate different functions ( e . g . direct excitation/inhibition vs . synaptic plasticity ) . In mammals , where glutamate is the principal excitatory neurotransmitter , metabotropic glutamate receptors ( mGluRs ) have been widely implicated in synaptic plasticity and memory ( Jörntell and Hansel , 2006; Lüscher and Huber , 2010 ) . Given the complexity of linking behavior to artificially induced plasticity in brain slices ( Schonewille et al . , 2011; Yamaguchi et al . , 2016 ) , it would be useful to study the role of metabotropic receptors in learning in a simpler genetic model system with a clearer behavioral readout of synaptic plasticity . One such system is Drosophila , where powerful genetic tools and well-defined anatomy have yielded a detailed understanding of the circuit and molecular mechanisms underlying associative memory ( Busto et al . , 2010; Cognigni et al . , 2018; Hige , 2018 ) . The principal excitatory neurotransmitter in Drosophila is acetylcholine , but , surprisingly , little is known about the function of metabotropic acetylcholine signaling in synaptic plasticity or neuromodulation in Drosophila . Here , we address this question using olfactory associative memory . Flies can learn to associate an odor ( conditioned stimulus , CS ) with a positive ( sugar ) or a negative ( electric shock ) unconditioned stimulus ( US ) , so that they later approach ‘rewarded’ odors and avoid ‘punished’ odors . This association is thought to be formed in the presynaptic terminals of the ~2000 Kenyon cells ( KCs ) that make up the mushroom body ( MB ) , the fly’s olfactory memory center ( Busto et al . , 2010; Cognigni et al . , 2018; Hige , 2018 ) . These KCs are activated by odors via second-order olfactory neurons called projection neurons ( PNs ) . Each odor elicits responses in a sparse subset of KCs ( Campbell et al . , 2013; Lin et al . , 2014 ) so that odor identity is encoded in which KCs respond to each odor . When an odor ( CS ) is paired with reward/punishment ( US ) , an odor-specific set of KCs is activated at the same time that dopaminergic neurons ( DANs ) release dopamine onto KC presynaptic terminals . The coincident activation causes long-term depression ( LTD ) of synapses from the odor-activated KCs onto mushroom body output neurons ( MBONs ) that lead to approach or avoidance behavior ( Aso and Rubin , 2016; Aso et al . , 2014b; Cohn et al . , 2015; Hige et al . , 2015; Owald et al . , 2015; Perisse et al . , 2016; Séjourné et al . , 2011 ) . In particular , training specifically depresses KC-MBON synapses of the ‘wrong’ valence ( e . g . odor-punishment pairing depresses odor responses of MBONs that lead to approach behavior ) , because different pairs of ‘matching’ DANs/MBONs ( e . g . punishment/approach , reward/avoidance ) innervate distinct regions along KC axons ( Aso et al . , 2014a ) . Both MB input ( PNs ) and output ( KCs ) are cholinergic ( Barnstedt et al . , 2016; Yasuyama and Salvaterra , 1999 ) , and KCs express both ionotropic ( nicotinic ) and metabotropic ( muscarinic ) acetylcholine receptors ( Crocker et al . , 2016; Croset et al . , 2018; Davie et al . , 2018; Shih et al . , 2019 ) . The nicotinic receptors mediate fast excitatory synaptic currents ( Su and O'Dowd , 2003 ) , while the physiological function of the muscarinic receptors is unknown . Muscarinic acetylcholine receptors ( mAChRs ) are G-protein-coupled receptors; out of the three mAChRs in Drosophila ( mAChR-A , mAChR-B and mAChR-C ) , mAChR-A ( also called Dm1 , mAcR-60C or mAChR ) is the most closely homologous to mammalian mAChRs ( Collin et al . , 2013 ) . Mammalian mAChRs are typically divided between ‘M1-type’ ( M1/M3/M5 ) , which signal via Gq and are generally excitatory , and ‘M2-type’ ( M2/M4 ) , which signal via Gi/o and are generally inhibitory ( Caulfield and Birdsall , 1998 ) . Drosophila mAChR-A seems to use ‘M1-type’ signaling: when heterologously expressed in Chinese hamster ovary ( CHO ) cells , it signals via Gq protein ( Collin et al . , 2013; Ren et al . , 2015 ) to activate phospholipase C , which produces inositol trisphosphate to release Ca2+ from internal stores . Recent work indicates that mAChR-A is required for aversive olfactory learning in Drosophila larvae , as knocking down mAChR-A expression in KCs impairs learning ( Silva et al . , 2015 ) . However , it is unclear whether mAChR-A is involved in olfactory learning in adult Drosophila , given that mAChR-A is thought to signal through Gq , and in adult flies Gq signaling downstream of the dopamine receptor Damb promotes forgetting , not learning ( Berry et al . , 2012; Himmelreich et al . , 2017 ) . Moreover , it is unknown how mAChR-A affects the activity or physiology of KCs , where it acts ( at KC axons or dendrites or both ) , and how these effects contribute to olfactory learning . Here , we show that mAChR-A is required in KCs for aversive olfactory learning in adult Drosophila . Surprisingly , genetic and pharmacological manipulations of mAChR-A suggest that mAChR-A is inhibitory and acts on KC dendrites . Moreover , mAChR-A knockdown impairs the learning-associated depression of odor responses in an MB output neuron , MB-MVP2 , that is required for aversive memory retrieval . We suggest that dendritically acting mAChR-A is required for synaptic depression between KCs and their outputs .
Drosophila larvae with reduced mAChR-A expression in KCs show impaired aversive olfactory learning ( Silva et al . , 2015 ) , but it remains unknown whether mAChR-A in KCs also functions in learning in adult flies . We addressed this question by knocking down mAChR-A expression in KCs using two UAS-RNAi lines , ‘RNAi 1’ and ‘RNAi 2’ ( see Materials and methods ) . Only RNAi 2 requires co-expression of Dicer-2 ( Dcr-2 ) for optimal knockdown . To test the efficiency of these RNAi constructs , we expressed them pan-neuronally using elav-GAL4 and measured their effects on mAChR-A expression levels using quantitative real-time polymerase chain reaction ( qRT-PCR ) . Both RNAi lines strongly reduce mAChR-A levels ( RNAi 1: 39 ± 8% of elav-GAL4 control , or 61 ± 8% below normal; RNAi 2: 43 ± 10% of normal; mean ± s . e . m . ; see Figure 1A ) . We then examined whether knocking down mAChR-A in KCs using the pan-KC driver OK107-GAL4 affects short-term aversive learning in adult flies . We used the standard odors used in the field ( i . e . 3-octanol , OCT , and 4-methylcyclohexanol , MCH; see Materials and methods ) . Under these conditions , both UAS-RNAi transgenes significantly reduced aversive learning , whether training against MCH or OCT ( Figure 1B , C and Figure 1—figure supplement 1 ) . Interestingly , knocking down mAChR-A did not affect learning when we trained flies with a more intense shock ( 90 V instead of 50 V , Figure 1—figure supplement 1 ) , suggesting that mAChR-A may only be required for learning with moderate intensity reinforcement , not severe reinforcement . Consistent with this , knocking down mAChR-A had no effect on naïve avoidance of MCH and OCT ( Figure 1D; see Materials and methods ) or flies’ reaction to electric shock ( Figure 1—figure supplement 1 ) , showing that the defect was specific to learning , rather than reflecting a failure to detect odors or shock . Given that mAChR-A is expressed in the larval MB and indeed contributes to aversive learning in larvae , it is possible that developmental effects underlie the reduced learning observed in mAChR-A knockdown flies . To test this , we used tub-GAL80ts to suppress RNAi 1 expression during development . Flies were grown at 23°C until 3 days after eclosion and were then transferred to 31°C for 7 days . Adult-only knockdown of mAChR-A in KCs reduced learning ( Figure 1E ) , just as constitutive knockdown did , indicating that mAChR-A plays a physiological , not purely developmental , role in aversive learning . To further verify that GAL80ts efficiently blocks RNAi expression ( i . e . that GAL80ts is not leaky ) , flies were grown at 23°C without transferring them to 31°C , thus blocking RNAi expression also in adults . When tested for learning at 10 days old , these flies showed normal learning ( Figure 1E ) . Kenyon cells are subdivided into three main classes according to their innervation of the horizontal and vertical lobes of the MB: γ neurons send axons only to the γ lobe of the horizontal lobes , while the axons of αβ and α′β′ neurons bifurcate and go to both the vertical and horizontal lobes ( αβ axons make up the α lobe of the vertical lobe and β lobe of the horizontal lobe , while α′β′ axons make up the α′ lobe of the vertical lobe and β′ portion of the horizontal lobe ) . These different classes play different roles in olfactory learning ( Guven-Ozkan and Davis , 2014; Krashes et al . , 2007 ) . To unravel in which class ( es ) mAChR-A functions , we used a Minos-mediated integration cassette ( MiMIC ) line to investigate where mAChR-A is expressed ( Venken et al . , 2011 ) . The MiMIC insertion in mAChR-A lies in the first 5’ non-coding intron , creating a gene trap where GFP in the MiMIC cassette should be expressed in whichever cells endogenously express mAChR-A . Because the GFP in the original mAChR-A MiMIC cassette produced very little fluorescent signal ( data not shown ) , we used recombinase-mediated cassette exchange ( RMCE ) to replace the original MiMIC cassette with a MiMIC cassette containing GAL4 ( Venken et al . , 2011 ) . These new mAChR-A-MiMIC-GAL4 flies should express GAL4 wherever mAChR-A is endogenously expressed . To reveal the expression pattern of mAChR-A , we crossed mAChR-A-MiMIC-GAL4 and 20xUAS-6xeGFP flies . mAChR-A-MiMIC-GAL4 drove GFP expression throughout the brain , consistent with previous reports ( Blake et al . , 1993; Croset et al . , 2018; Davie et al . , 2018; Hannan and Hall , 1996 ) and with the fact that the Drosophila brain is mostly cholinergic . In the mushroom bodies , GFP was expressed in the αβ and γ lobes , but not the α′β′ lobes ( Figure 2A ) . No GFP signal was observed with an inverted insertion where GAL4 is inserted in the MiMIC locus in the wrong orientation ( data not shown ) . Consistent with these MiMIC results , two recently reported databases of single-cell transcriptomic analysis of the Drosophila brain ( Croset et al . , 2018; Davie et al . , 2018 ) confirm that mAChR-A is more highly expressed in αβ and γ KCs than in α′β′ KCs ( Figure 2—figure supplement 1 ) . However , mAChR-A is still clearly present in α′β′ KCs’ transcriptomes , suggesting that mAChR-A-MiMIC-GAL4 may not reveal all neurons that express mAChR-A . The higher expression of mAChR-A in αβ and γ KCs compared to α′β′ KCs suggests that learning would be impaired by mAChR-A knockdown in αβ or γ , but not α′β′ , KCs . To test this , we expressed mAChR-A RNAi in different KC classes . As expected , aversive olfactory learning was reduced by knocking down mAChR-A in αβ and γ KCs together using MB247-GAL4 , but not by knockdown in α′β′ KCs using c305a-GAL4 . To examine if αβ and γ KCs both participate in the reduced learning observed in mAChR-A knockdown flies , we sought to limit mAChR-A RNAi expression to either αβ or γ neurons . While strong driver lines exist for αβ neurons , the γ GAL4 drivers we tested were fairly weak ( H24-GAL4 , MB131B , R45H04-GAL4 , data not shown ) , perhaps too weak to drive mAChR-A-RNAi enough to knock down mAChR-A efficiently . Therefore , we used MB247-GAL4 , which was strong enough to affect behavior , and blocked GAL4 activity in either αβ or γ KCs by expressing the GAL80 repressor under the control of R44E04-LexA ( αβ KCs ) or R45H04-LexA ( γ KCs ) ( Bräcker et al . , 2013 ) . These combinations drove strong , specific expression in αβ or γ KCs ( Figure 2—figure supplement 2 ) . Learning was reduced by mAChR-A RNAi expression in γ , but not αβ , KCs ( Figure 2B ) . These results suggest that mAChR-A is specifically required in γ KCs for aversive olfactory learning and short-term memory . We next asked what effect mAChR-A knockdown has on the physiology of KCs , by expressing GCaMP6f and mAChR-A RNAi 2 together in KCs using OK107-GAL4 ( this driver and RNAi combination was also used for behavior in Figure 1C ) . Knocking down mAChR-A in KCs increased odor-evoked Ca2+ influx in the mushroom body calyx , where KC dendrites reside ( Figure 3 ) . This result is somewhat surprising because mAChR-A is a Gq-coupled receptor whose activation leads to Ca2+ release from internal stores ( Ren et al . , 2015 ) , which predicts that mAChR-A knockdown should decrease , not increase , odor-evoked Ca2+ influx in KCs . However , some examples have been reported of inhibitory signaling through Gq by M1-type mAChRs ( see Discussion ) , and Drosophila mAChR-A may join these as another example of an inhibitory mAChR signaling through Gq . Because mAChR-A is required for aversive learning in γ KCs , not αβ or α′β′ KCs ( Figure 2 ) , we next asked how odor responses in αβ , α′β′ and γ KCs are affected by mAChR-A knockdown . αβ , α′β′ and γ KC dendrites are not clearly segregated in the calyx , so we examined odor responses in the axonal lobes . Indeed , although odor responses in all lobes were increased by mAChR-A knockdown , only in the γ lobe was the effect statistically significant for both MCH and OCT ( Figure 3 ) . This result is consistent with the behavioral requirement for mAChR-A only in γ KCs . However , we do not rule out the possibility that mAChR-A knockdown also affects αβ and α′β′ odor responses in a way that does not affect short-term memory , especially as αβ and α′β′ odor responses were somewhat , although not consistently significantly , increased . Although the ∆F/F traces from the γ lobe had higher signal-to-noise ratio ( SNR ) than some other lobes ( Figure 3—figure supplement 1 ) due to its larger size ( averaging over more pixels ) or shallower z-depth ( less light scattering ) , a power analysis revealed that all lobes had SNRs high enough to detect an effect as large as that observed in the γ lobe ( Figure 3—figure supplement 1 ) . However , note that we do not exclude the possibility that αβ- or α′β′-specific ( as opposed to pan-KC ) knockdown of mAChR-A might significantly increase αβ or α′β′ KC odor responses . Do increased odor responses in γ KCs prevent learning by increasing the overlap between the γ KC population representations of the two odors used in our task ( Lin et al . , 2014 ) ? When GCaMP6f and mAChR-A-RNAi 2 were expressed in all KCs , mAChR-A knockdown did not affect the sparseness or inter-odor correlation of KC population odor responses ( Figure 4A–C ) even though it increased overall calyx responses . To focus specifically on γ KCs , we expressed GCaMP6f and mAChR-A-RNAi 1 only in γ KCs , using mb247-Gal4 , R44E04-LexA and lexAop-GAL80 , the same driver and RNAi combination used in the behavioral experiments in Figure 2B . GCaMP6f was visible mainly in the γ lobe ( Figure 4D ) . γ-only expression of mAChR-A-RNAi 1 increased odor responses in the calyx ( here , dendrites of γ KCs only ) and , in the case of OCT , in the γ lobe ( Figure 4E , F ) . Note that γ KC odor responses were increased by both RNAi 1 ( Figure 3A , B ) and RNAi 2 ( Figure 4E , F ) . As with pan-KC expression , γ-only expression of mAChR-A-RNAi 1 did not affect the sparseness or inter-odor correlation of γ KCs ( Figure 4G–I ) . Thus , mAChR-A knockdown does not impair learning through increased overlap in KC population odor representations . RNAi-based knockdown of mAChR-A might induce homeostatic compensation that obscures or even reverses the primary effect of reduced mAChR-A expression . To test the acute role of mAChR-A in regulating KC activity , we took the complementary approach of pharmacologically activating mAChR-A . Initially , we bath-applied 10 µM muscarine , an mAChR-A agonist ( Drosophila mAChR-B is 1000-fold less sensitive to muscarine than mAChR-A is [Collin et al . , 2013] , and mAChR-C is not expressed in the brain [Davie et al . , 2018] ) . Muscarine strongly decreased odor responses in all subtypes of KCs ( Figure 5A , B , Figure 5—figure supplement 1 ) . However , muscarine did not significantly affect the amplitude of odor responses in PN axons in the calyx ( Figure 5C ) , suggesting that the effect of muscarine on KCs arose in KCs , not earlier in the olfactory pathway . KCs can be silenced by an inhibitory GABAergic neuron called the anterior paired lateral ( APL ) neuron ( Lin et al . , 2014; Masuda-Nakagawa et al . , 2014; Papadopoulou et al . , 2011 ) , so we asked whether muscarine reduces KC odor responses indirectly by activating APL , rather than directly inhibiting KCs . We applied muscarine to flies with APL-specific expression of tetanus toxin ( TNT ) , which blocks inhibition from APL and thereby greatly increases KC odor responses . In these flies , APL is labeled stochastically , so hemispheres where APL was unlabeled served as controls ( Lin et al . , 2014 ) ( see Materials and methods ) . Muscarine decreased KC odor responses both in control hemispheres and hemispheres where APL synaptic output was blocked by tetanus toxin ( Figure 5D ) , and the effect of muscarine was not significantly different between the two cases ( Figure 5E ) . This result indicates that muscarine does not act solely by activating APL or by enhancing inhibition on KCs ( e . g . increasing membrane localization of GABAA receptors ) . To test mAChR-A function even more acutely , we locally applied muscarine to the MB calyx by pressure ejection ( Figure 6 , Figure 6—figure supplement 1 ) . Red dye included in the ejected solution confirmed that the muscarine remained in the calyx for several seconds but did not spread to the MB lobes ( Figure 6B ) . Surprisingly , applying muscarine to the calyx in the absence of odor stimuli increased GCaMP6f signal in the calyx and α lobe , with small increases in the β and γ lobe that were not statistically significant ( Figure 6A , C ) . It also decreased GCaMP6f signal in the α′ and β′ lobes around 1–2 s after application ( Figure 6A ) , although this effect was also not statistically significant . The increased Ca2+ in the calyx most likely did not reflect increased excitability , as applying muscarine to the calyx did not increase the calyx odor response ( Figure 6D , E ) . If anything , it likely decreased the calyx odor response , because the Ca2+ increase induced by muscarine alone ( no odor ) lasted ~6–7 s and thus would have continued into the odor pulse in the muscarine +odor condition . If the odor response was unaffected by muscarine , the muscarine-evoked and odor-evoked increases in GCaMP6f signal should have summed . Instead , the peak calyx ∆F/F during the odor pulse was the same before and after locally applying muscarine , suggesting that the specifically odor-evoked increase in GCaMP6f was decreased by muscarine . Indeed , applying muscarine to the calyx suppressed odor responses in KC axons ( Figure 6D , E ) . Although muscarine did not significantly affect peak ∆F/F during the odor in the α lobe , muscarine most likely did decrease α lobe odor responses , by the same logic as for calyx odor responses ( see above ) . Given that calyx muscarine suppresses α′β′ axonal odor responses , the decrease in α′β′ KC GCaMP6f signal in the absence of odor likely reflects suppression of spontaneous action potentials ( Figure 6A , C ) , as α′β′ KCs have the highest spontaneous spike rate out of the three subtypes ( Groschner et al . , 2018; Turner et al . , 2008 ) . The effect of muscarine on α′β′ KCs is consistent with single-cell transcriptome analyses showing that α′β′ KCs express mAChR-A , albeit at a lower level than αβ or γ KCs ( Figure 2—figure supplement 1 ) ( Croset et al . , 2018; Davie et al . , 2018 ) . The increase in calyx Ca2+ induced by muscarine alone ( without odor ) might reflect Ca2+ release from internal stores triggered by Gq signaling , which then inhibits KC excitability ( thus smaller odor responses ) . Note that muscarine on the calyx is unlikely to reduce KC odor responses via presynaptic inhibition of PNs , because bath muscarine does not affect odor-evoked Ca2+ influx in PNs in the calyx ( Figure 5C ) , although we cannot rule out Ca2+-independent inhibition . We next asked where mAChR-A exerts its effect . To visualize the localization of mAChR-A , we created a new construct with mAChR-A tagged with FLAG on the C-terminus under UAS control . When we overexpressed FLAG-tagged mAChR-A in KCs using OK107-GAL4 , we only observed anti-FLAG staining in the calyx ( Figure 7A ) , suggesting that mAChR-A is localized to the calyx . To test whether the FLAG tag or overexpression might cause the mAChR-A to be mis-localized , we tested whether mb247-GAL4 > mAChR A-FLAG overexpression could rescue learning in a mAChR-A mutant background . The original MiMIC allele with a GFP insertion in the 5’ UTR intron of mAChR-A contains a stop cassette and polyadenylation signal , and indeed , it is a strongly hypomorphic allele: qPCR shows almost total lack of mAChR-A mRNA in the ‘MiMIC-stop’ allele ( Figure 7B ) . Flies homozygous for the ‘MiMIC-stop’ allele are viable but show impaired learning , while learning is significantly improved by using mb247-GAL4 to overexpress mAChR-A-FLAG in αβ and γ KCs ( Figure 7C ) , indicating that overexpressed mAChR-A-FLAG can support learning . These flies ( ‘MiMIC-stop’ , mb247 >mAChR A-FLAG ) also show anti-FLAG staining only in the calyx ( Figure 7—figure supplement 1 ) . These results suggest that mAChR-A exerts its effect on learning in KC dendrites , consistent with the effect of locally applying muscarine to KC dendrites . The finding that mAChR-A functions in KC dendrites raises the question of how mAChR-A can affect learning . While learning-associated plasticity in KC dendrites has been observed in honeybees , In Drosophila , olfactory associative memories are stored by weakening the synapses between KCs and output neurons that lead to the ‘wrong’ behavior . For example , aversive memory requires an output neuron downstream of γ KCs , called MBON-γ1pedc>α/β or MB-MVP2 . MB-MVP2 leads to approach behavior ( Aso et al . , 2014b ) , and aversive conditioning reduces MB-MVP2’s responses to the aversively-trained odor ( Hige et al . , 2015; Perisse et al . , 2016 ) . We tested whether knocking down mAChR-A would prevent this depression . We knocked down mAChR-A in KCs using OK107-GAL4 and UAS-mAChR-A-RNAi 1 , and expressed GCaMP6f in MB-MVP2 using R12G04-LexA and lexAop-GCaMP6f ( Figure 8A ) . We trained flies in the behavior apparatus and then imaged MB-MVP2 odor responses ( 3 hr after training to avoid cold-shock-sensitive memory ) . Because overall response amplitudes were variable across flies , for each fly we measured the ratio of the response to MCH ( the trained odor ) over the response to OCT ( the untrained odor ) . Consistent with previous published results ( Hige et al . , 2015; Perisse et al . , 2016 ) , in control flies not expressing mAChR-A RNAi , the MCH/OCT ratio was substantially reduced in trained flies relative to mock-trained flies ( Figure 8B ) . This was not because the OCT response increased , because there was no difference between trained and mock-trained flies in the ratio of the response to OCT over the response to isoamyl acetate , a ‘reference’ odor that was absent in the training protocol . This was also not because of any general decrease in odor responses , as shown by analyzing absolute response amplitudes to MCH , OCT and isoamyl acetate ( Figure 8—figure supplement 1 ) . In contrast , in flies expressing mAChR-A RNAi in KCs , the MCH/OCT ratio was the same between trained and mock-trained flies ( Figure 8B ) , indicating that the mAChR-A knockdown impaired the learning-related depression of the KC to MB-MVP2 synapse . This result suggests that mAChR-A function in KC dendrites is necessary for learning-related synaptic plasticity in KC axons .
Here , we show that mAChR-A is required in γ KCs for aversive olfactory learning and short-term memory in adult Drosophila . Knocking down mAChR-A increases KC odor responses , while the mAChR-A agonist muscarine suppresses KC activity . Knocking down mAChR-A prevents aversive learning from reducing responses of the MB output neuron MB-MVP2 to the conditioned odor , suggesting that mAChR-A is required for the learning-related depression of KC->MBON synapses . Why is mAChR-A only required for aversive learning in γ KCs , not αβ or α′β′ KCs ? Although our mAChR-A MiMIC gene trap agrees with single-cell transcriptome analysis that α′β′ KCs express less mAChR-A than do γ and αβ KCs ( Croset et al . , 2018; Davie et al . , 2018 ) , transcriptome analysis indicates that α′β′ KCs do express some mAChR-A ( Figure 2—figure supplement 1 ) . Moreover , γ and αβ KCs express similar levels of mAChR-A ( Crocker et al . , 2016 ) . It may be that the RNAi knockdown is less efficient at affecting the physiology of αβ and α′β′ KCs than γ KCs , whether because the knockdown is less efficient at reducing protein levels , or because αβ and α′β′ KCs have different intrinsic properties or a different function of mAChR-A such that 40% of normal mAChR-A levels is sufficient in αβ and α′β′ KCs but not γ KCs . This interpretation is supported by our finding that mAChR-A RNAi knockdown significantly increases odor responses only in the γ lobe , not the αβ or α′β′ lobes . Alternatively , γ , αβ and α′β′ KCs are thought to be important mainly for short-term memory , long-term memory , and memory consolidation , respectively ( Guven-Ozkan and Davis , 2014; Krashes et al . , 2007 ) ; as we only tested short-term memory , mAChR-A may carry out the same function in all KCs , but only its role in γ KCs is required for short-term ( as opposed to long-term ) memory . Indeed , the key plasticity gene DopR1 is required in γ , not αβ or α′β′ KCs , for short-term memory ( Qin et al . , 2012 ) . It may be that mAChR-A is required in non-γ KC types for other forms of memory besides short-term aversive memory , such as appetitive conditioning or other phases of memory like long-term memory . Our finding that mAChR-A is required in γ KCs for aversive short-term memory is consistent with our finding that mAChR-A knockdown in KCs disrupts training-induced depression of odor responses in MB-MVP2 , an MBON postsynaptic to γ KCs required for aversive short-term memory ( Perisse et al . , 2016 ) . However , the latter finding does not rule out the possibility that other MBONs postsynaptic to non-γ KCs may also be affected by mAChR-A knockdown in KCs . mAChR-A seems to inhibit KC odor responses , because knocking down mAChR-A increases odor responses in the calyx and γ lobe , while activating mAChR-A with bath or local application of muscarine decreases KC odor responses . Some details differ between the genetic and pharmacological results . In particular , while mAChR-A knockdown mainly affects γ KCs , with other subtypes inconsistently affected , muscarine reduces responses in all KC subtypes . What explains these differences ? mAChR-A might be weakly activated in physiological conditions , in which case gain of function would cause a stronger effect than loss of function . Similarly , pharmacological activation of mAChR-A is likely a more drastic manipulation than a 60% reduction of mAChR-A mRNA levels . Although we cannot entirely rule out network effects from muscarine application , the effect of muscarine does not stem from PNs or APL ( Figure 5C , D ) and locally applied muscarine would have little effect on neurons outside the mushroom body . How does mAChR-A inhibit odor-evoked Ca2+ influx in KCs ? Given that mAChR-A signals through Gq when expressed in CHO cells ( Ren et al . , 2015 ) , that muscarinic Gq signaling normally increases excitability in mammals ( Caulfield and Birdsall , 1998 ) , and that pan-neuronal artificial activation of Gq signaling in Drosophila larvae increases overall excitability ( Becnel et al . , 2013 ) , it may be surprising that mAChR-A inhibits KCs . However , Gq signaling may exert different effects on different neurons in the fly brain , and some examples exist of inhibitory Gq signaling by mammalian mAChRs . M1/M3/M5 receptors acting via Gq can inhibit voltage-dependent Ca2+ channels ( Gamper et al . , 2004; Kammermeier et al . , 2000; Keum et al . , 2014; Suh et al . , 2010 ) , reduce voltage-gated Na +currents ( Cantrell et al . , 1996 ) , or trigger surface transport of KCNQ channels ( Jiang et al . , 2015 ) , thus increasing inhibitory K+ currents . Drosophila mAChR-A may inhibit KCs through similar mechanisms . What is the source of ACh which activates mAChR-A and modulates odor responses ? In the calyx , cholinergic PNs are certainly a major source of ACh . However , KCs themselves are cholinergic ( Barnstedt et al . , 2016 ) and release neurotransmitter in both the calyx and lobes ( Christiansen et al . , 2011 ) . KCs form synapses on each other in the calyx ( Zheng et al . , 2018 ) , possibly allowing mAChR-A to mediate lateral inhibition , in conjunction with the lateral inhibition provided by the GABAergic APL neuron ( Lin et al . , 2014 ) . What function does mAChR-A serve in learning and memory ? Our results indicate that mAChR-A knockdown prevents the learning-associated weakening of KC-MBON synapses , in particular for MBON-γ1pedc>α/β , aka MB-MVP2 ( Figure 7 ) . One potential explanation is that the increased odor-evoked Ca2+ influx observed in knockdown flies increases synaptic release , which overrides the learning-associated synaptic depression . However , increased odor-evoked Ca2+ influx per se is unlikely on its own to straightforwardly explain a learning defect , because other genetic manipulations that increase odor-evoked Ca2+ influx in KCs either have no effect on , or even improve , olfactory learning . For example , knocking down GABA synthesis in the inhibitory APL neuron increases odor-evoked Ca2+ influx in KCs ( Lei et al . , 2013; Lin et al . , 2014 ) and improves olfactory learning ( Liu and Davis , 2009 ) . The most intuitive explanation would be that mAChR-A acts at KC synaptic terminals in KC axons to help depress KC-MBON synapses . Yet overexpressed mAChR-A localizes to KC dendrites , not axons , and functionally rescues mAChR-A hypomorphic mutants , showing that dendritic mAChR-A suffices for its function in learning and memory . Does this show that mAChR-A has no role in KC axons ? Our inability to detect GFP expressed from the mAChR-A MiMIC gene trap suggests that normally there may only be a small amount of mAChR-A in KCs . It may be that with mAChR-A-FLAG overexpression , the correct ( undetectable ) amount of mAChR-A is trafficked to and functions in axons , but due to a bottleneck in axonal transport , the excess tagged mAChR-A is trapped in KC dendrites . While our results do not rule out this possibility , a general bottleneck in axonal transport seems unlikely as many overexpressed proteins are localized to KC axons ( Trunova et al . , 2011 ) . We feel it is more parsimonious to take the dendritic localization of mAChR-A-FLAG at face value and infer that mAChR-A functions in KC dendrites . How can mAChR-A in KC dendrites affect synaptic plasticity in KC axons ? mAChR-A signaling might change the shape or duration of KC action potentials ( Allen and Burnstock , 1990; Ghamari-Langroudi and Bourque , 2004 ) , an effect that could potentially propagate to KC axon terminals ( Juusola et al . , 2007; Shu et al . , 2006 ) . Such changes in the action potential waveform may not be detected by calcium imaging , but could potentially affect a ‘coincidence detector’ in KC axons that detects when odor ( i . e . KC activity ) coincides with reward/punishment ( i . e . dopamine ) . This coincidence detector is generally believed to be the Ca2+-dependent adenylyl cyclase rutabaga ( Levin et al . , 1992 ) . Changing the waveform of KC action potentials could potentially affect local dynamics of Ca2+ influx near rutabaga molecules . In addition , rutabaga mutations do not abolish learning ( mutants have ~40–50% of normal learning scores ) ( Yildizoglu et al . , 2015 ) , so there may be additional coincidence detection mechanisms affected by action potential waveforms . Testing this idea would require a better understanding of biochemical events underlying learning at KC synaptic terminals . Alternatively , mAChR-A’s effects on synaptic plasticity may not occur acutely . Although we ruled out purely developmental effects of mAChR-A through adult-only RNAi expression ( Figure 1E ) , knocking out mAChR-A for several days in adulthood might still affect KC physiology in a not-entirely-acute way . For example , as with other G-protein-coupled receptors ( Wang and Zhuo , 2012 ) , muscarinic receptors can affect gene expression ( von der Kammer et al . , 1998 ) , which could have wide-ranging effects on KC physiology , for example action potential waveform , expression of key genes required for synaptic plasticity , etc . Another intriguing possibility is suggested by an apparent paradox: both mAChR-A and the dopamine receptor Damb signal through Gq ( Himmelreich et al . , 2017 ) , but mAChR-A promotes learning while Damb promotes forgetting ( Berry et al . , 2012 ) . How can Gq mediate apparently opposite effects ? Perhaps Gq signaling aids both learning and forgetting by generally rendering synapses more labile . Indeed , although damb mutants retain memories for longer than wildtype , their initial learning is slightly impaired ( Berry et al . , 2012 ) ; damb mutant larvae are also impaired in aversive olfactory learning ( Selcho et al . , 2009 ) . Although one study reports that knocking down Gq in KCs did not impair initial memory ( Himmelreich et al . , 2017 ) , the Gq knockdown may not have been strong enough; also , that study shocked flies with 90 V shocks , which also gives normal learning in mAChR-A knockdown flies ( Figure 1—figure supplement 1 ) . Such hypotheses posit that mAChR-A regulates synaptic plasticity ‘competence’ rather than participating directly in the plasticity mechanism itself . Why should synaptic plasticity competence be controlled by an activity-dependent mechanism ? It is tempting to speculate that mAChR-A may allow a kind of metaplasticity ( Abraham , 2008 ) in which exposure to odors ( hence activation of mAChR-A in KCs ) makes flies’ learning mechanisms more sensitive . Indeed , mAChR-A is required for learning with moderate ( 50 V ) shocks , not severe ( 90 V ) shocks . Future studies may further clarify how muscarinic signaling contributes to olfactory learning .
Fly strains ( see below ) were raised on cornmeal agar under a 12 hr light/12 hr dark cycle and studied 1–10 days post-eclosion . Strains were cultivated at 25˚C unless they expressed temperature-sensitive gene products ( GAL80ts ) ; in these cases , the experimental animals and all relevant controls were grown at 23˚C . To de-repress the expression of RNAi with GAL80ts , experimental and control animals were incubated at 31˚C for 7 days . Subsequent behavioral experiments were performed at 25˚C . Experimental animals carried transgenes over Canton-S chromosomes where possible to minimize genetic differences between strains . Details of fly strains are given in the Key Resources Table . UAS-mAChR-A-FLAG plasmid was generated by Gibson assembly of fragments using the NEBuilder HiFi Master Mix ( NEB ) . Fragments were created by PCR using Phusion High-Fidelity DNA Polymerase ( NEB ) . The full-length mAChR-A cDNA was purchased from GenScript ( clone ID OFa11160 ) . The vector was pTWF-attB , a gift from Prof . Oren Schuldiner ( Yaniv et al . , 2012 ) . This vector consists of a FLAG tag in the C-terminal of the inserted gene and an attB site for site-specific integration of the transgene . PCR and Gibson assembly were carried out following the manufacturer’s recommendations with the following primers: For mAChR-A: tgggaattatcgacaagtttgtacaaaaaagcaggctATGGAGCCGGTCATGAGTC and cactttgtacaagaaagctgggtaATTGTAGACGCCGCGTAC For pTWF-AttB: aaagctgggtaCTTGTACAAAGTGGTGAGCTCC and agcctgcttttttgtacAAACTTGTCGATAATTCCC Transgenes were injected into the attP2 landing site using φC31 integration ( by BestGene ) . Total RNA was extracted by EZ-RNA II Total RNA Isolation kit ( Biological Industries , Israel ) from 30 adult heads for each biological replicate . cDNA was generated from 1 µg total RNA with the High-Capacity cDNA Reverse Transcription Kit with RNase Inhibitor ( Applied Biosystems ) . Real-time quantitative PCR was carried with TaqMan Fast Advanced Master Mix ( Applied Biosystems ) and run in technical triplicates on a StepOne Plus Real-Time PCR System ( Applied Biosystems ) . Taqman assays were Dm01820303_g1 for mAChR-A and Dm02151962_g1 for EF1 ( Ef1alpha100E , ThermoFisher ) . The expression levels obtained for mAChR-A were normalized to those of the housekeeping gene EF1 . The fold change for mAChR-A was subsequently calculated by comparing to the normalized value of either ELAV-gal4 parent ( for RNAi experiments ) or w1118 flies ( for MIMiC experiments ) . Behavioral experiments were performed in a custom-built , fully automated apparatus ( Claridge-Chang et al . , 2009; Lin et al . , 2014; Parnas et al . , 2013 ) . Single flies were housed in clear polycarbonate chambers ( length 50 mm , width 5 mm , height 1 . 3 mm ) with printed circuit boards ( PCBs ) at both floors and ceilings . Solid-state relays ( Panasonic AQV253 ) connected the PCBs to a 50 V source . Air flow was controlled with mass flow controllers ( CMOSens PerformanceLine , Sensirion ) . A carrier flow ( 2 . 7 l/min ) was combined with an odor stream ( 0 . 3 l/min ) obtained by circulating the air flow through vials filled with a liquid odorant . Odors were prepared at 10 fold dilution in mineral oil . Therefore , liquid dilution and mixing carrier and odor stimulus stream resulted in a final 100 fold dilution of odors . Fresh odors were prepared daily . The 3 l/min total flow ( carrier and odor stimulus ) was split between 20 chambers resulting in a flow rate of 0 . 15 l/min per half chamber . Two identical odor delivery systems delivered odors independently to each half of the chamber . Air or odor streams from the two halves of the chamber converged at a central choice zone . The 20 chambers were stacked in two columns each containing 10 chambers and were backlit by 940 nm LEDs ( Vishay TSAL6400 ) . Images were obtained by a MAKO CMOS camera ( Allied Vision Technologies ) equipped with a Computar M0814-MP2 lens . The apparatus was operated in a temperature-controlled incubator ( Panasonic MIR-154 ) maintained at 25˚C . A virtual instrument written in LabVIEW 7 . 1 ( National Instruments ) extracted fly position data from video images and controlled the delivery of odors and electric shocks . Data were analyzed in MATLAB 2015b ( The MathWorks ) and Prism 6 ( GraphPad ) . A fly’s preference was calculated as the percentage of time that it spent on one side of the chamber . Training and odor avoidance protocols were as depicted in Figure 1 . The naïve avoidance index was calculated as ( preference for left side when it contains air ) – ( preference for left side when it contains odor ) . During training , MCH was paired with 12 equally spaced 1 . 25 s electric shocks at 50 V ( Tully and Quinn , 1985 ) . The learning index was calculated as ( preference for MCH before training ) – ( preference for MCH after training ) . Flies were excluded from analysis if they entered the choice zone fewer than four times during odor presentation . Brains were imaged by two-photon laser-scanning microscopy ( Ng et al . , 2002; Wang et al . , 2003 ) . Cuticle and trachea in a window overlying the required area were removed , and the exposed brain was superfused with carbogenated solution ( 95% O2 , 5% CO2 ) containing 103 mM NaCl , 3 mM KCl , 5 mM trehalose , 10 mM glucose , 26 mM NaHCO3 , 1 mM NaH2PO4 , 3 mM CaCl2 , 4 mM MgCl2 , 5 mM N-Tris ( TES ) , pH 7 . 3 . Odors at 10−1 dilution were delivered by switching mass-flow controlled carrier and stimulus streams ( Sensirion ) via software controlled solenoid valves ( The Lee Company ) . Flow rates at the exit port of the odor tube were 0 . 5 or 0 . 8 l/min . Fluorescence was excited by a Ti-Sapphire laser centered at 910 nm , attenuated by a Pockels cell ( Conoptics ) and coupled to a galvo-resonant scanner . Excitation light was focussed by a 20X , 1 . 0 NA objective ( Olympus XLUMPLFLN20XW ) , and emitted photons were detected by GaAsP photomultiplier tubes ( Hamamatsu Photonics , H10770PA-40SEL ) , whose currents were amplified and transferred to the imaging computer . Two imaging systems were used , #1 for Figures 3–6 except 5C , and #2 for Figure 5C and Figure 7 , which differed in the following components: laser ( 1: Mai Tai eHP DS , 70 fs pulses; 2: Mai Tai HP DS , 100 fs pulses; both from Spectra-Physics ) ; microscope ( 1: Movable Objective Microscope; 2: DF-Scope installed on an Olympus BX51WI microscope; both from Sutter ) ; amplifier for PMT currents ( 1: Thorlabs TIA-60; 2: Hamamatsu HC-130-INV ) ; software ( 1: ScanImage 5; 2: MScan 2 . 3 . 01 ) . Volume imaging on System 1 was performed using a piezo objective stage ( nPFocus400 , nPoint ) . Muscarine was applied locally by pressure ejection from borosilicate patch pipettes ( resistance ~10 MOhm; capillary inner diameter 0 . 86 mm , outer diameter 1 . 5 mm; concentration in pipette 20 mM; pressure 12 . 5 psi ) using a Picospritzer III ( Parker ) . A red dye was added to the pipette to visualize the ejected fluid ( SeTau-647 , SETA BioMedicals ) ( Podgorski et al . , 2012 ) . Movies were motion-corrected in X-Y using the moco ImageJ plugin ( Dubbs et al . , 2016 ) , with pre-processing to collapse volume movies in Z and to smooth the image with a Gaussian filter ( standard deviation = 4 pixels; the displacements generated from the smoothed movie were then applied to the original , unsmoothed movie ) , and motion-corrected in Z by maximizing the pixel-by-pixel correlation between each volume and the average volume across time points . ∆F/F , activity maps , sparseness and inter-odor correlation were calculated as in Lin et al . ( 2014 ) . Briefly , movies were smoothed with a 5-pixel-square Gaussian filter ( standard deviation 2 ) . Baseline fluorescence was taken as the average fluorescence during the pre-stimulus period . Frames with sudden , large axial movements were discarded by correlating each frame to the baseline image and discarding it if the correlation fell below a threshold value , which was manually selected for each brain by noting the constant high correlation value when the brain was stationary and sudden drops in correlation when the brain moved . ∆F/F was calculated for each pixel as the difference between mean fluorescence during the stimulus period vs . the baseline fluorescence ( ∆F ) , divided by the baseline fluorescence . For pixels where ∆F did not exceed two times the standard deviation over time of that pixel’s intensity during the pre-stimulus period , the pixel was considered non-responsive . We excluded non-responsive flies and flies whose motion could not be corrected . Inter-odor correlations were calculated by first aligning the activity maps of each odor response by maximizing the inter-odor correlations of baseline fluorescence , and then converting image matrices of the activity maps of each odor response into linear vectors and calculating the Pearson correlation coefficients between each ‘odor vector’ . A threshold for baseline fluorescence was applied as a mask to the activity map to exclude pixels with no baseline GCaMP6f signal . Population sparseness was calculated for activity maps using the following equation ( Vinje and Gallant , 2000; Willmore and Tolhurst , 2001 ) :SP=11-1N ( 1-∑i=1NriN2∑i=1Nri2N ) Brain dissections , fixation , and immunostaining were performed as described ( Pitman et al . , 2011; Wu and Luo , 2006 ) . To visualize native GFP fluorescence , dissected brains were fixed in 4% ( w/v ) paraformaldehyde in PBS ( 1 . 86 mM NaH2PO4 , 8 . 41 mM Na2HPO4 , 175 mM NaCl ) and fixed for 20 min at room temperature . Samples were washed for 3 × 20 min in PBS containing 0 . 3% ( v/v ) Triton-X-100 ( PBT ) . The neuropil was counterstained with nc82 ( DSHB ) or monoclonal anti-FLAG M2 antibody ( F3165 , Sigma ) and goat anti-mouse Alexa 647 or Alexa 546 . Primary antisera were applied for 1–2 days and secondary antisera for 1–2 days in PBT at 4˚C , followed by embedding in Vectashield . Images were collected on a Leica TCS SP5 , SP8 , or Nikon A1 confocal microscope and processed in ImageJ . APL expression of tetanus toxin was scored by widefield imaging of mCherry . mCherry expression in APL was distinguished from 3XP3-driven dsRed from the GH146-FLP transgene by using separate filter cubes for dsRed ( 49004 , Chroma: 545/25 excitation; 565 dichroic; 605/70 emission ) and mCherry ( LED-mCherry-A-000 , Semrock: 578/21 excitation; 596 dichroic; 641/75 emission ) . Statistical analyses were carried out in GraphPad Prism as described in figure legends and Supplementary file 1 . In general , no statistical methods were used to predetermine sample sizes , but where conclusions were drawn from the absence of a statistically significant difference , a power analysis was carried out in G*Power to confirm that the sample size provided sufficient power to detect an effect of the expected size . The experimenter was blind to which hemispheres had APL neurons expressing tetanus toxin before post-experiment dissection ( Figure 5 ) but not otherwise . | We can learn a surprising amount about how the brain forms memories by studying the humble fruit fly . These insects can learn to associate odors with positive or negative experiences , allowing them to then seek out ‘rewarded’ odors and avoid ‘punished’ ones . This association takes place in a brain region called the mushroom body , and it involves two types of neurons: Kenyon cells , which detect odors , and MBONs , which lead to approach or avoidance behaviors . When Kenyon cells detect an odor accompanying an unpleasant event , they weaken their connections with the MBONs that trigger approach behaviors . This prevents the fly from coming close to that odor in the future . Kenyon cells exchange signals with other neurons using a chemical called acetylcholine , which attaches onto the cells through two types of receptors: nicotinic and muscarinic . Studies in fruit fly larvae suggest that muscarinic receptors are required in Kenyon cells for the insects to learn how to associate odors with unpleasant experiences . Bielopolski et al . now show that this is also the case in adult flies . Surprisingly , while acetylcholine usually excites fly neurons , activating muscarinic receptors inhibits Kenyon cells rather than exciting them . Labeled muscarinic receptors revealed that the receptors act within the input region of Kenyon cells . Moreover , reducing the levels of muscarinic receptors inside the cells stops flies from associating an odor with a mild electric shock . This manipulation also prevents the learning experience from weakening connections from Kenyon cells onto an MBON that triggers approach behavior . This suggests that allowing these changes in connectivity might be why muscarinic receptors are important for memory . Understanding how memory works in flies can reveal basic principles that apply to many species , including humans . Such knowledge could ultimately help us improve the memory of patients with dementia , but also inspire better algorithms for artificial intelligence . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2019 | Inhibitory muscarinic acetylcholine receptors enhance aversive olfactory learning in adult Drosophila |
MHC class I-related molecule MR1 presents riboflavin- and folate-related metabolites to mucosal-associated invariant T cells , but it is unknown whether MR1 can present alternative antigens to other T cell lineages . In healthy individuals we identified MR1-restricted T cells ( named MR1T cells ) displaying diverse TCRs and reacting to MR1-expressing cells in the absence of microbial ligands . Analysis of MR1T cell clones revealed specificity for distinct cell-derived antigens and alternative transcriptional strategies for metabolic programming , cell cycle control and functional polarization following antigen stimulation . Phenotypic and functional characterization of MR1T cell clones showed multiple chemokine receptor expression profiles and secretion of diverse effector molecules , suggesting functional heterogeneity . Accordingly , MR1T cells exhibited distinct T helper-like capacities upon MR1-dependent recognition of target cells expressing physiological levels of surface MR1 . These data extend the role of MR1 beyond microbial antigen presentation and indicate MR1T cells are a normal part of the human T cell repertoire .
T lymphocytes can detect a diverse range of microbial and self-antigens , including lipids presented by non-polymorphic CD1 molecules ( Zajonc and Kronenberg , 2007 ) , and phosphorylated isoprenoids bound to Butyrophilin 3A1 ( Sandstrom et al . , 2014; Vavassori et al . , 2013 ) , which stimulate TCR Vγ9-Vδ2 cells , the major population of TCR γδ cells in human blood . The heterogeneous phenotypic and functional properties of these T cells support specialized roles in host protection against infections , cancers , and autoimmunity ( Cohen et al . , 2009; Mori et al . , 2016; Tyler et al . , 2015 ) . The repertoire of T cells specific for non-peptide antigens was expanded to include mucosal associated invariant T ( MAIT ) cells , which are restricted to MHC I-related protein MR1 and were originally described within mouse gut ( Treiner et al . , 2003 ) . Elegant functional studies showed that MAIT cells react to different bacteria by recognizing non-proteinaceous antigens ( Gold et al . , 2010; Le Bourhis et al . , 2010 ) . Finally , the stimulatory antigens were identified as small riboflavin precursors produced by a wide range of yeasts and bacteria ( Corbett et al . , 2014; Kjer-Nielsen et al . , 2012; Soudais et al . , 2015 ) . MAIT cells display one of three alternative semi-invariant TCRs in which the TRAV1-2 gene is rearranged with genes encoding TRAJ33 , J12 or J20 ( Gold et al . , 2014; Lepore et al . , 2014; Reantragoon et al . , 2012 ) , and is combined with an oligoclonal TCR-β repertoire ( Lepore et al . , 2014 ) . Cells expressing these evolutionary-conserved TCRs are frequent in human blood , kidney and intestine , and comprise a major fraction of T cells resident in the liver ( Dusseaux et al . , 2011; Lepore et al . , 2014; Tang et al . , 2013 ) . Following activation , MAIT cells release an array of pro-inflammatory and immunomodulatory cytokines , and can mediate direct killing of microbe-infected cells ( Dusseaux et al . , 2011; Kurioka et al . , 2015; Le Bourhis et al . , 2013; Lepore et al . , 2014; Tang et al . , 2013 ) . MAIT cells react to a broad range of microbes , but it remains unknown whether the role of MR1 extends beyond presentation of microbial metabolites to MAIT cells . MR1 is a non-polymorphic MHC I-like protein that is expressed at low to undetectable levels on the surface of many cell types ( Huang et al . , 2008; Miley et al . , 2003 ) . Following bacterial infection , antigen-presenting cells ( APC ) up-regulate surface expression of MR1 due to protein stabilization upon antigen binding ( Huang et al . , 2008; McWilliam et al . , 2016; Miley et al . , 2003 ) . MR1 is highly conserved across multiple species , with human and mouse MR1 sharing >90% sequence homology at the protein level ( Riegert et al . , 1998 ) . To date , the MR1-bound ligands reported to simulate MAIT cells include three ribityl lumazines ( RLs ) , microbial pterin-like compounds displaying a ribitol at carbon 8 ( Kjer-Nielsen et al . , 2012 ) , and two unstable adducts formed by non-enzymatic reaction of microbial 5-ribityl amino uracil ( 5-RAU ) with either host-derived or bacterial carbonyls such as glyoxal or methyl-glyoxal ( Corbett et al . , 2014 ) . RLs bind to MR1 via hydrogen bonds and hydrophobic interactions with amino acids in the antigen binding cleft ( Kjer-Nielsen et al . , 2012; Patel et al . , 2013 ) , whereas binding of unstable 5-RAU-derivatives requires covalent trapping within the MR1 pocket via formation of a Schiff base with Lysine 43 ( Corbett et al . , 2014; Patel et al . , 2013 ) . While three additional molecules have also been demonstrated to bind MR1 via the same mechanism , including 6-formyl-pterin ( 6-FP ) , acetyl-6-formyl-pterin ( Ac-6-FP ) , and acetylamino-4-hydoxy-6-formylpteridine dimethyl acetal ( Kjer-Nielsen et al . , 2012; Patel et al . , 2013; Soudais et al . , 2015 ) , these compounds lack the ribityl moiety necessary for MAIT cell activation and instead inhibit stimulation by microbial ligands ( Corbett et al . , 2014; Kjer-Nielsen et al . , 2012; Soudais et al . , 2015 ) . A recent study reported that 6-FP and Ac-6-FP can also stimulate a second population of ‘atypical’ MR1-restricted T cells of unknown function in the blood of healthy human donors ( Gherardin et al . , 2016 ) , but this population , did not express the canonical TRAV1-2 gene which defines MAIT cells . Together , these findings indicate a degree of MR1 plasticity in accommodating diverse chemical structures and suggest that other MR1-restricted T cell populations may exist in vivo . Here we report a novel population of MR1-restricted T cells ( hereafter termed MR1T cells ) that is readily detectable in blood from healthy individuals . MR1T cells express diverse TCRα and β genes and were not able to recognize previously identified microbial or folate-derived ligands of MR1 . Instead , MR1T cells recognized self-antigens presented by MR1 and expressed a broad selection of cytokines and chemokine receptors . Functionally , MR1T cells were capable of inducing dendritic cell ( DC ) maturation and promoted innate defense in intestinal epithelial cells . These findings demonstrate that MR1 can present non-microbial antigens to a novel population of functionally diverse human T cells with potentially wide-ranging roles in human immunity .
During our previous study on the repertoire of human MAIT cells ( Lepore et al . , 2014 ) , we detected an atypical MR1-restricted T cell clone that did not react to microbial ligands . This T cell clone ( DGB129 ) recognized cell lines constitutively displaying surface MR1 ( CCRF-SB lymphoblastic leukemia cells , or THP-1 monocytic leukemia cells; Figure 1A ) or A375 melanoma cells ( A375-MR1 ) transfected with an MR1-β2m fusion gene construct ( Lepore et al . , 2014 ) ( Figure 1A ) in the absence of any exogenously added antigens ( Figure 1B ) . Sterile recognition of MR1+ target cells was fully inhibited by blocking with anti-MR1 monoclonal antibodies ( mAbs ) ( Figure 1B ) , and thus resembled the MAIT cell response to E . coli-derived antigens assessed in parallel ( Figure 1C ) . Importantly , DGB129 T cells also failed to recognize the synthetic MAIT cell agonist 6 , 7-dimethyl-8-D-ribityllumazine ( RL-6 , 7-diMe; Figure 1D ) ( Kjer-Nielsen et al . , 2012 ) , differently from a control MAIT cell clone , which instead was stimulated in MR1-dependent manner by this compound ( Figure 1E ) . Unlike the canonical semi-invariant TCR typical of MAIT cells , DGB129 cells displayed a TCRαβ heterodimer comprising TRAV29/DV5 rearranged to TRAJ23 and TRBV12-4 rearranged to TRBJ1-1 . Expression of these TCRα and β genes in a TCR-deficient cell line ( SKW-3 T lymphoblastic leukemia ) conferred MR1 reactivity in the absence of exogenous antigens comparable to that displayed by DGB129 cells ( Figure 1F ) , while in control experiments , transduction of TCRα and β genes of a representative MAIT cell clone conferred the ability to recognize the same target cells in MR1-dependent manner only in the presence of E . coli antigens ( Figure 1G ) . Collectively , these data highlight a critical role of the TCR in mediating DGB129 cell recognition of MR1-expressing APCs and suggest that MR1 can present non-microbial antigens to T cells other than MAIT cells . 10 . 7554/eLife . 24476 . 003Figure 1 . Recognition of non-microbial antigens by MR1-restricted T cells . ( A ) Surface expression of MR1 by CCRFSB , THP-1 and A375-MR1 cells . Grey histograms indicate staining with isotype-matched control mAbs . Stimulation of ( B ) T cell clone DGB129 or ( C ) MAIT cell clone SMC3 by the three cell lines in A in the absence ( no Ag ) or presence of E . coli lysate ( E . coli ) and/or anti-MR1 blocking mAbs ( α-MR1 ) . Columns indicate IFN-γ release ( mean + SD ) . Stimulation of ( D ) DGB129 MR1T or ( E ) SMC3 MAIT cells by THP-1 cells , constitutively expressing surface MR1 , loaded with synthetic 6 , 7-dimethyl-8-D-ribityllumazine ( RL-6 , 7-diMe ) with or without anti-MR1 mAbs . Columns indicate mean IFN-γ release + SD . Stimulation of ( F ) SKW-3 cells expressing the DGB129 TCR ( SKW3-DGB129 ) or ( G ) J . RT3-T3 . 5 cells expressing the MAIT MRC25 clone TCR ( J . RT3-MAIT ) with A375 cells that expressed ( A375-MR1 ) or lacked ( A375-WT ) MR1 , with or without E . coli lysate and/or anti-MR1 mAbs . CD69 median fluorescence intensity ( MFI ) ± SD of duplicate cultures of transduced T cells are shown . The CD69 MFI of transduced T cells cultured in the absence of APCs is also shown . Data are representative of four ( A , B and C ) , two ( D and E ) , and three ( F and G ) independent experiments . *p<0 . 05 ( Unpaired Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24476 . 003 To investigate the presence of these unpredicted MR1-restricted T cells in different individuals , we performed combined in vitro stimulation and single cell cloning experiments using total blood T cells . Purified T cells from two healthy donors were labeled with the proliferation marker CellTrace violet ( CTV ) and stimulated with irradiated A375-MR1 cells in the absence of exogenous antigens . The choice of A375-MR1 cells relied on their potent capacity of inducing sterile DGB129 T cell activation ( Figure 1B ) and their high expression of surface MR1 molecules ( Figure 1A ) , which we reasoned could maximize the chance of stimulating and thus expanding DGB129-like MR1-restricted T cells present in the blood . Amongst the T cells of both donors , we observed a significant fraction of proliferating cells that expressed high levels of the activation marker CD137 following re-challenge with A375-MR1 cells . These activated T cells were sorted and cloned by limiting dilution ( Figure 2A ) . Individual T cell clones were then interrogated for their capacity to recognize A375-MR1 and A375 cells lacking MR1 ( A375-WT ) . In both donors we found that a major fraction of T cell clones ( 126/195 and 37/57 , respectively ) displayed specific recognition of A375-MR1 cells ( Figure 2B , D ) , which was potently inhibited by anti-MR1 blocking mAbs ( Figure 2C , E ) . Staining with TCR Vβ-specific mAbs of 12 MR1-reactive T cell clones revealed that they expressed seven different TRBV chains ( TRBV4-3 , 6-5/6-6/6-9 , 9 , 18 , 25–1 , 28 , 29–1 ) with some of the clones sharing the same TRBV gene . Furthermore , none expressed the TRAV1-2 chain , canonical for MAIT cells . These data suggested that MR1-restricted T cells other than MAIT cells exist in the blood of healthy donors , can express diverse TCRs and are able of clonal expansion following in vitro stimulation in the absence of microbial ligands . 10 . 7554/eLife . 24476 . 004Figure 2 . Isolation of non-MAIT MR1-restricted T cell clones after stimulation with A375-MR1 cells in the absence of microbial antigens . ( A ) FACS analysis of purified T cells previously expanded with irradiated A375-MR1 cells following overnight co-culture with A375-MR1 cells . Left dot plot shows CD3 and CellTrace violet ( CTV ) staining in live cells . Upper right and bottom right dot plots show CD69 and CD137 expression on CD3+CTV− and CD3+CTV+ gated cells , respectively . Arrows indicate gating hierarchy . Numbers indicate the percentages of cells within the gates . Cells from Donor A are illustrated as a representative donor . ( B , D ) Cumulative results of T cell clones screening from Donors A and B . T cell clones were generated from CD3+CTV-CD137high sorted T cells as depicted in A . Graphs show the individual clones ( x axis ) and their IFN-γ release ( y axis ) , expressed as ratio between the amount of cytokine secreted in response to A375-MR1 cells vs . A375 WT cells . Each dot represents a single T cell clone , tested at the same time in the indicated experimental conditions . The horizontal red line marks the arbitrary IFN-γ ratio cut-off of two , above which MR1-dependent T cell clone reactivity was set . The intercept of the vertical red line indicates the number of MR1-restricted T cell clones in each donor . Red boxes highlight T cell clones whose reactivity was MR1-dependent . Results are representative of two independent experiments . ( C , E ) IFN-γ release by 14 representative clones from Donor A and 11 clones from Donor B after stimulation with A375 WT , A375-MR1 and A375-MR1 in the presence of blocking anti-MR1 mAbs ( α-MR1 ) . Dots represent the IFN-γ release ( mean ± SD of duplicate cultures ) by each clone . Results are representative of three independent experiments . *p<0 . 05 ( Unpaired Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24476 . 004 Lack of specific markers did not allow univocal identification of these novel T cells ex vivo by standard flow cytometry . Therefore their frequency was estimated by combining flow cytometry analysis after very short-time in vitro stimulation and single T cell cloning experiments . Purified blood T cells from five healthy donors were co-cultured overnight with A375 cells either lacking or over-expressing MR1 and analyzed for the expression of the activation markers CD69 and CD137 ( Figure 3A ) . In all of the five donors screened , the percentage of CD69highCD137+ T cells detected was consistently higher after stimulation with A375-MR1 cells ( range 0 . 034–0 . 072% of T cells ) than after co-culture with A375-WT cells ( range 0 . 015–0 . 032% ) ( Figure 3A , B ) . As the two types of APCs differed for MR1 expression , we assumed that MR1-reactive T cells might account for the increased numbers of activated T cells after stimulation with MR1-positive APCs . Using this approach , we estimated that the circulating T cell pool of the analyzed individuals contained A375-MR1-reactive T cells at frequency ranging between 1:2500 ( 0 . 072–0 . 032 = 0 . 04% ) and 1:5000 ( 0 . 034–0 . 015 = 0 . 019% ) . This estimated frequency , although representing an approximation , is within the range of the frequency of peptide-specific CD4+ T cells after antigen exposure ( Lucas et al . , 2004; Su et al . , 2013 ) . We also performed parallel experiments in which overnight-activated CD69highCD137+ T cells ( 0 . 065% ) were sorted from one of these donors ( Donor C , Figure 3A , right panel ) and were cloned . Indeed , 31 out of 96 screened T cell clones ( 32% ) displayed specific reactivity to A375-MR1 cells ( Figure 3C ) , which was inhibited by anti-MR1 mAbs ( Figure 3D ) . Accordingly , we calculated that the frequency of A375-MR1-responsive T cells among blood T cells of this donor was 1:5000 ( 0 . 065 × 0 . 32 = 0 . 02% ) , a value consistent with the previously estimated range . Detailed analysis of representative T cell clones derived from three donors confirmed that they displayed diverse TCRα and β chains and indicated differential expression of CD4 , CD8 and CD161 ( Table 1 ) . 10 . 7554/eLife . 24476 . 005Figure 3 . Non-MAIT MR1-restricted T cells are readily detectable in the blood of healthy individuals . ( A ) Flow cytometry analysis of purified T cells from a representative donor ( Donor C ) after overnight co-culture with A375 WT or A375-MR1 cells . Dot plots show CD69 and CD137 expression on live CD3+ cells . Numbers indicate the percentage of cells in the gates . ( B ) Frequency of CD69+CD137+ T cells from five different donors after overnight co-culture with A375 WT or A375-MR1 cells . ( C ) Cumulative results of T cell clone stimulation assays from Donor C . T cell clones were generated from CD3+CD69+CD137+ sorted T cells as depicted in A , right dot plot . The graph shows the number of tested clones ( x axis ) and IFN-γ release ( y axis ) expressed as ratio between the amount of cytokine secreted in response to A375-MR1 cells vs . A375-WT cells . Each dot represents a single T cell clone , tested at the same time in the indicated experimental conditions . The horizontal red line marks the arbitrary IFN-γ ratio threshold of two above which MR1-dependent T cell clone reactivity was set . The intercept of the vertical red line indicates the number of MR1-restricted T cell clones . Red box highlights T cell clones whose reactivity was MR1-dependent . Results are representative of two independent experiments . ( D ) Recognition of A375-MR1 but not A375 WT cells in the absence of exogenous antigens by eight representative MR1-restricted T cell clones from Donor C . Inhibition of T cell clone reactivity to A375-MR1 cells by blocking anti-MR1 mAbs ( α-MR1 ) . Dots represent the IFN-γ release ( mean ± SD of duplicate cultures ) by each clone tested in the three experimental conditions . Results are representative of three independent experiments . *p<0 . 05 ( Unpaired Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24476 . 00510 . 7554/eLife . 24476 . 006Table 1 . Phenotype and TCR gene usage of selected MR1-reactive T cell clones . DOI: http://dx . doi . org/10 . 7554/eLife . 24476 . 006CloneCD4CD8αTCRαTCRβCD161DGB129−+TRAV29TRBV12-4−DGB70−−TRAV5TRBV28−DGA28−+TRAV25TRBV29-1+DGA4−−TRAV1-2ND+JMA−+TRAV27TRBV25-1−TC5A87−+TRAV13-1TRBV25-1−CH9A3−+TRAV24TRBV5-5−ND , not determined . Collectively , these findings suggested that identified MR1-responsive T cells are a novel yet readily detectable polyclonal population of lymphocytes in the blood of healthy human individuals ( hereafter termed MR1T cells ) . We next studied the basis of MR1T cell reactivity . Firstly , we tested whether the MR1T cell clones could recognize microbial antigens , in analogy to MAIT cells . While a control MAIT cell clone reacted to A375-MR1 cells only in the presence of E . coli lysate , activation of different MR1T cell clones was not enhanced by the E . coli lysate ( Figure 4A ) . Consistent with these data , MR1-negative A375-WT cells failed to stimulate either type of T cell , irrespective of whether E . coli lysate was added ( Figure 4A ) and importantly , anti-MR1 mAbs efficiently blocked both MR1T and MAIT cell responses ( Figure 4A ) . These findings confirmed that microbial ligands present in E . coli and stimulating MAIT cells do not stimulate the tested MR1T cells . 10 . 7554/eLife . 24476 . 007Figure 4 . MR1T cell clones do not react to microbial ligands or to 6-FP . ( A ) Response of seven MR1T cell clones and one control MAIT cell clone co-cultured with A375 cells expressing ( A375-MR1 ) or not ( A375 WT ) MR1 in the presence or absence of E . coli lysate . Blocking of T cell clone reactivity by anti-MR1 mAbs ( α-MR1 ) is also shown . ( B ) Response of MR1T cell clones to A375 cells expressing either WT MR1 molecules ( A375-MR1 ) or K43A-mutated MR1 molecules ( A375-MR1 K43A ) in the presence of 6-formyl pterin ( 6-FP ) . ( C ) Stimulation of control MAIT cell clone MRC25 or control TCR Vγ9Vδ2 clone G2B9 with A375-MR1 or A375-MR1 K43A cells previously incubated with or without E . coli lysate or zoledronate , respectively , either in the absence or presence of 6-FP . Results are expressed as mean ± SD of IFN-γ measured in duplicate cultures . Results are representative of three independent experiments . *p<0 . 05 ( Unpaired Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24476 . 00710 . 7554/eLife . 24476 . 008Figure 4—figure supplement 1 . MR1T cell clones do not recognize Ac-6-FP . ( A ) Stimulation of three representative MR1T cell clones by A375-MR1 cells in the absence or presence of acetyl-6-formyl pterin ( Ac-6-FP ) . ( B ) Stimulation of two MAIT cell clones ( MRC25 and SMC3 ) by A375-MR1 cells pulsed with E . coli lysate in the absence or presence of Ac-6-FP . ( C ) A375-MR1 cells were treated with zoledronate ( Zol ) in the absence or presence of Ac-6-FP ( 25 µg/ml ) and used to stimulate a TCR Vγ9-Vδ2 cell clone ( G2B9 ) . ( D ) A375 cells expressing K43A mutant MR1 molecules ( A375-MR1 K43A ) were used to stimulate the three MR1T cell clones shown in A , in the absence or presence of Ac-6-FP ( 25 µg/ml ) . ( E ) Stimulation of the two MAIT cell clones used in B by A375-MR1 K43A cells pulsed with E . coli lysate in the absence or presence of Ac-6-FP ( 25 µg/ml ) . Results are expressed as mean ± SD of IFN-γ release assessed in duplicate cultures and are representative of three independent experiments . *p<0 . 05 ( Unpaired Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24476 . 008 We then tested the response of MR1T cells to the known MR1 ligands 6-FP and Ac-6-FP , which have previously been reported to stimulate a rare subset of TRAV1-2-negative T cells ( Gherardin et al . , 2016 ) and inhibit MAIT cell activation by microbial antigens ( Corbett et al . , 2014; Kjer-Nielsen et al . , 2012; Soudais et al . , 2015 ) . MR1T cell stimulation was impaired in the presence of 6-FP or Ac-6-FP ligands , which also impaired E . coli stimulation of control MAIT cells , but did not disrupt TCR γδ cell responses to cognate antigen presented by the same APCs ( Figure 4B , C and Figure 4—figure supplement 1A–C ) . Notably , 6-FP or Ac-6-FP failed to inhibit the activation of MR1T cells or MAIT cells when the target A375 cells were transduced to express mutant MR1 molecules with defective ligand binding capacity ( blockade of Schiff base formation with ligands by mutation of Lysine 43 into Alanine , A375-MR1 K34A; Figure 4B , C and Figure 4—figure supplement 1D , E ) . The specific inhibition observed with 6-FP or Ac-6-FP indicated that MR1T cells ( i ) do not recognize 6-FP and Ac-6-FP; ( ii ) react to MR1-bound cellular antigens; ( iii ) are stimulated by ligands that do not require the formation of a Schiff base with MR1 . To gain information on the origin of the recognized antigens we firstly asked whether the stimulatory capacity of target cells was dependent on culture medium constituents , as some MR1 ligands , e . g . 6-FP , may derive from folate present in RPMI 1640 medium used for cell culture ( Kjer-Nielsen et al . , 2012 ) . Both THP-1 and A375-MR1 cells were extensively washed and cultivated 4 days in phosphate buffered saline solution ( PBS ) supplemented exclusively with 5% human serum . Cells were washed daily before being used to stimulate DGB129 MR1T cells and the T cell activation assays were performed in PBS . Both THP-1 and A375-MR1 cells grown in RPMI 1640 or in PBS showed the same stimulatory capacity ( Figure 5A , B ) , thus indicating that medium constituents are not responsible for MR1T cell activation . To directly investigate whether the stimulatory antigens were present in target cells , we performed T cell activation assays using as source of antigen two types of lysates . The first lysate was obtained from in vitro cultured THP-1 cells , while the second one was prepared from mouse breast tumors immediately after resection . Two hydrophobic and four hydrophilic fractions were obtained and tested using as APCs THP-1 cells that constitutively express low levels of MR1 . The DGB129 clone reacted only to fraction N4 , containing highly hydrophilic compounds isolated from both freshly explanted mouse tumor and in vitro cultured THP-1 cells ( Figure 5C , D ) . These results ruled out the possibility that stimulatory antigens were derived from RPMI 1640 components and indicated their cellular origin . We also tested the fractions generated from THP-1 lysates with DGB70 , another representative MR1T cell clone . DGB70 cells recognized fraction N3 and not N4 , ( Figure 5E ) , suggesting that at least two distinct compounds differentially stimulated the two MR1T clones . The same fractions were also loaded onto plastic-bound MR1 molecules and showed alternative and specific stimulatory capacity , i . e . N3 stimulated only DGB70 cells , while N4 stimulated only DGB129 cells ( Figure 5F ) . In the absence of N3 and N4 fractions , the two clones did not react to MR1 , further indicating the requirement of specific antigens . 10 . 7554/eLife . 24476 . 009Figure 5 . MR1T cells recognize diverse antigens not derived from RPMI 1640 medium . Stimulation of the DGB129 MR1T cell clone by MR1-overexpressing ( A ) A375 cells ( A375-MR1 ) and ( B ) THP-1 cells ( THP1-MR1 ) grown for 4 days in RPMI 1640 or in PBS both supplemented with 5% human serum . Inhibition of T cell clone reactivity by anti-MR1 blocking mAbs ( α-MR1 ) is shown . DGB129 cells recognize APCs loaded with fractions isolated from ( C ) THP-1 cell lysate or from ( D ) in vivo grown mouse breast tumor EMT6 . Fractions E1 and E2 contain hydrophobic molecules; fractions N1-N4 contain hydrophilic molecules . ( E ) DGB70 MR1T cells react to N3 fraction of THP-1 lysate . ( F ) Stimulation of DGB129 and DGB70 T cells by THP-1-derived fractions N3 and N4 loaded onto plastic-bound recombinant MR1 . Shown is T cell release of IFN-γ or GM-CSF mean ± SD of duplicate cultures ( representative of three independent experiments ) . Total cytokine release is shown in panels A , B , F; fold increase over background is shown in panels C , D , E . *p<0 . 05 ( Unpaired Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24476 . 009 In conclusion , these data indicated that MR1T cells recognize MR1 complexed with ligands not derived from culture medium and present also in tumor cells grown in vivo . At least two diverse molecules were stimulatory , whose nature and structure will be the subject of future studies . As DGB129 and DGB70 clones were isolated from the same donor , these findings also suggested that MR1T cells with different antigen specificities are present in the same individual . We next examined the diversity of MR1T cells by comparing the transcriptional response to antigen stimulation of the two representative MR1T cell clones responding to diverse THP-1 lysate fractions ( DGB129 and DGB70 ) . MR1T cells were first rested for three weeks before being stimulated with A375-MR1 cells for 20 hr . The transcriptional profiles of the sorted activated cells ( expressing similar levels of both CD25 and CD137 activation markers ) were subsequently compared with their unstimulated control counterparts ( negative for the two markers ) by RNA-sequencing ( Figure 6—figure supplement 1 ) . Biological replicates were analyzed using stringent cut-off criteria ( FDR < 0 . 05 , minimum log2 fold change >2 ) . The RNA-seq datasets revealed that MR1+ APC stimulated DGB129 cells upregulated 403 genes and downregulated 413 genes , whereas DGB70 cells up-regulated 432 genes and down-regulated 285 genes ( Supplementary file 1 ) . We then identified key transcription factors in each clone using global transcription factor gene regulatory network analysis ( Narang et al . , 2015 ) . A subnetwork was created from the identified genes whose expression was modulated upon antigen stimulation , and the key nodes were defined using centrality measures . This approach showed that some master transcription factors were shared between MR1T cell clones , whereas others appeared to be clone-specific ( Figure 6A , B ) . Whether assessed for betweenness centrality or tested by the PageRank algorithm , both T cell clone responses were identified as being regulated by genes TBX21 , FOXP3 , FOS , RXRA , FOSL2 , IRF4 , BATF and TRIB1 . However , while DGB129 cells were further influenced by expression of MYC , HSP90AB1 and CREM , DGB70 cells were instead regulated by EGR1 , JUNB , and SREBF1 ( Figure 6C ) . 10 . 7554/eLife . 24476 . 010Figure 6 . MR1T cell clones exhibit divergent transcriptional responses to antigen stimulation . Betweeness centrality analysis illustrating the key transcription factors that were differentially expressed in resting ( No Ag ) and antigen-activated ( Ag Act ) MR1T cell clones ( A ) DGB129 and ( B ) DGB70 . Color code represents betweeness centrality score . The size of the nodes ( or hubs ) indicates the relative importance of individual transcription factors within the whole gene network . ( C ) Heat-map comparing key transcription factors that were differentially expressed in resting ( Rest . ) and antigen-activated ( Act . ) DGB129 and DGB70 MR1T cell clones ( analysis performed by PageRank algorithm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24476 . 01010 . 7554/eLife . 24476 . 011Figure 6—figure supplement 1 . FACS analysis of resting and activated DGB129 and DGB70 MR1T cells used for transcriptome studies . Dot-plots display the expression of CD25 and CD137 activation markers on T cells cultured alone ( Resting ) or stimulated with A375-MR1 cells ( Ag stimulated ) . Purity of CD25+CD137+ cells sorted from the Ag stimulated groups ( Sorted Activated ) is shown . Plots are gated on CD3+ DAPI- cells . Resting and Sorted Activated DGB129 and DGB70 MR1T cells were utilized for next-generation sequencing experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 24476 . 011 We next analyzed the cytokine secretion profile of representative MR1T cell clones upon stimulation by A375-MR1 APCs . All clones tested released IFN-γ ( Figure 7—figure supplement 1A ) , probably as a consequence of in vitro expansion ( Becattini et al . , 2015 ) . However , we also observed diverse expression profiles of Th1 ( IL-2 , TNF-α and TNF-β ) , Th2 ( IL-3 , IL-4 , IL-5 , IL-6 , IL-10 , IL-13 ) and Th17 cytokines ( IL-17A , G-CSF , GM-CSF ) , and other soluble factors ( MIP-1β , soluble CD40L , PDGF-AA and VEGF; Figure 7A and Figure 7—figure supplement 1B ) . The variable combinations and quantities of cytokines expressed by MR1T cells suggested considerable functional plasticity within this population . For example , clone DGA4 secreted large quantities of IL-17A , IL-6 , TNF-α and GM-CSF , but failed to secrete the prototypic Th2 cytokines IL-4 , IL-5 , IL-10 or IL-13 , and thus displayed an ‘atypical’ Th17-like phenotype . In contrast , clone TC5A87 released substantial amounts of VEGF and PGDF-AA , but only little Th1 or Th2 cytokines , and no IL-17A . Notably , four of the seven clones studied ( DGB129 , CH9A3 , DGB70 , JMA ) displayed a Th2-skewed profile of cytokine release . In the same experiments , both control MAIT cell clones , although derived from two different donors , showed uniform Th1-biased responses following activation by A375-MR1 cells pulsed with E . coli lysate . Given that the culture conditions and APCs used in these experiments were identical for all clones , these data indicate that MR1T cells exhibit intrinsic functional heterogeneity . 10 . 7554/eLife . 24476 . 012Figure 7 . Functional heterogeneity of MR1T cell clones . ( A ) Heat-map of cytokine expression by seven different MR1T cell clones when stimulated by MR1-expressing A375 cells . Also shown are the cytokine profiles of control MAIT cell clones MRC25 ( MAIT-1 ) and SMC3 ( MAIT-2 ) following activation by A375-MR1 cells pulsed with E . coli lysate . Cytokines were assessed in the supernatants of duplicate cultures . The mean value for each cytokine was used to generate the heat-map . Cluster analysis was performed using Pearson correlation . Graphs displaying the amounts of individual cytokines released by the different clones are shown in Figure 7—figure supplement 1 . ( B ) Flow cytometry analysis of CXCR3 , CCR4 and CCR6 surface expression by resting MR1T cell clones and 2 MAIT control clones ( MRC25 , MAIT-1 and SMC3 , MAIT-2 ) . Graphs show the relative fluorescence intensity calculated by dividing the median fluorescence intensity ( MFI ) of specific mAb staining by the MFI of the corresponding isotype control . Data are representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 24476 . 01210 . 7554/eLife . 24476 . 013Figure 7—figure supplement 1 . Cytokines released by antigen-stimulated MR1T and MAIT cell clones . ( A ) IFN-γ released by 7 MR1T cell clones stimulated with A375-MR1 cells and 2 MAIT cell clones ( MRC25 and SMC3 ) stimulated with A375-MR1 cells pulsed with E . coli lysate . ELISA results are expressed as mean ± SD of IFN-γ release measured in duplicate cultures . ( B ) Analysis of 16 additional cytokines by multiplex cytokine assay performed on the same supernatants for which IFN-γ is shown in A . Results are representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 24476 . 013 To further support these findings , we next investigated the expression of three selected chemokine receptors known to be differentially expressed by T cell subsets with distinct functions ( Becattini et al . , 2015 ) and whose alternative combined expression regulates T cell recirculation and migration to diverse homing sites ( Al-Banna et al . , 2014; Bromley et al . , 2008; Thomas et al . , 2007 ) . Both resting MR1T and MAIT cell clones displayed high levels of CXCR3 ( with the exception of the DGA4 clone only; Figure 7B ) , However , we observed divergent expression patterns of CCR4 and CCR6 ( Figure 7B ) , which further suggested that MR1T cells are heterogeneous and different from MAIT cells . Taken together , these data indicated that the MR1-reactive clones tested here are phenotypically and functionally heterogeneous , thus suggesting that MR1T cells include multiple subsets with diverse recirculation patterns and tissue homing capacity and likely different specialized roles in human immunity . The MR1 expression level on the surface of APCs is physiologically regulated to be low to undetectable in the absence of microbial ligands ( Huang et al . , 2008; McWilliam et al . , 2016 ) . We therefore investigated whether monocyte-derived DCs ( Mo-DCs ) and other types of cells not-overexpressing MR1 , but constitutively displaying low surface MR1 levels ( Figure 1A , Figure 8—figure supplement 1A , C and data not shown ) , were able to stimulate MAIT and MR1T cells . All these cell types , supported MAIT cell activation in the presence of microbial antigens and in an MR1-dependent manner ( Figure 8A ) . The same cells also induced sterile MR1T cell activation to various extents . Mo-DC and THP-1 cells were recognized by the majority of the tested MR1T cell clones , followed by the Huh7 hepatoma cells , the LS 174T goblet-like cells and the HCT116 colon carcinoma cells ( Figure 8B ) . Importantly , all responses were blocked by anti-MR1 mAbs . 10 . 7554/eLife . 24476 . 014Figure 8 . MR1T cells recognize cells constitutively expressing low surface MR1 and show diverse T helper-like functions . ( A ) Recognition of four human cells lines expressing constitutive surface levels of MR1 and Mo-DCs by the representative SMC3 MAIT cell clone in the absence ( no Ag ) or presence of E . coli lysate ( E . coli ) with or without anti-MR1 blocking mAbs ( α-MR1 ) . ( B ) Recognition of the same cell types as in A by thirteen MR1T cell clones with or without anti-MR1 mAbs ( α-MR1 ) . Graphs show IFN-γ release ( mean ±SD of duplicate cultures ) . ( C ) Flow cytometry analysis of co-stimulatory molecules CD83 and CD86 on Mo-DCs after co-culture with DGB129 MR1T cells with or without anti-MR1 mAbs ( α-MR1 ) . A control group consisting of Mo-DCs stimulated with LPS ( 10 ng/ml ) in the absence of T cells is also shown . Numbers indicate percentages of cells in each quadrant . ( D , E ) Stimulation of JMA MR1T cells by LS 174T intestinal epithelial cells with or without anti-MR1 mAbs ( α-MR1 ) . Columns show ( D ) IL-8 ( ng/ml ) and ( E ) IL-13 ( optical density , O . D . ) release . ( F ) Q-PCR analysis of MUC2 gene expression in LS 174 T cells cultured alone or with JMA MR1T cells in the presence or absence of anti-MR1 mAbs ( α-MR1 ) . As control , LS 174 T cells were stimulated with recombinant TNF-α ( rTNF-α , 10 ng/ml ) in the absence of MR1T cells . All data are expressed as mean ± SD of triplicate cultures . Results are representative of at least three independent experiments . *p<0 . 05 ( Unpaired Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24476 . 01410 . 7554/eLife . 24476 . 015Figure 8—figure supplement 1 . MR1 surface expression by monocyte-derived dendritic cells and LS 174 T cells . ( A ) Flow cytometry detection of MR1 surface protein expression on monocyte-derived dendritic cells ( Mo-DCs ) generated from blood of three healthy donors . Grey histograms represent staining with isotype-matched control mAbs . ( B ) Recognition of Mo-DCs from the three donors in A by DGB129 MR1T cell clones in the absence or presence of anti-MR1 ( α-MR1 ) mAbs . IFN-γ release in the supernatants is shown and expressed as mean ± SD . Results are representative of three independent experiments . *p<0 . 05 ( Unpaired Student’s t-test ) . ( C ) MR1 surface protein expression by LS 174T intestinal epithelial cells . Grey histograms represent staining with isotype-matched control mAbs . DOI: http://dx . doi . org/10 . 7554/eLife . 24476 . 015 Having found that MR1 T cells can recognize cells constitutively expressing physiological surface levels of MR1 , we next investigated whether they can also modulate the functions of these target cells . First we focused on Mo-DCs , which stimulated several tested MR1T cell clones . The representative DGB129 MR1T cell clone reacted to Mo-DC from different donors ( Figure 8—figure supplement 1B ) and induced up-regulation of the maturation markers CD83 and CD86 on these target cells ( Figure 8C ) . Remarkably , Mo-DC maturation promoted by DGB129 cells was fully inhibited by anti-MR1 mAbs , thus confirming that recognition of MR1 was required for this T helper-like function . As we observed that three MR1T cell clones reacted also to LS 174T goblet-like cells , we next investigated the outcome of this interaction . LS 174 T cells are used as a model to study mucin gene expression regulation in intestinal epithelial cells ( van Klinken , 1996 ) and express low surface levels of MR1 ( Figure 8—figure supplement 1C ) . The selected JMA MR1T cell clone reacted to LS 174T by secreting in MR1-dependent fashion IL-8 and IL-13 ( Figure 8D , E ) , two cytokines involved in the modulation of human intestinal epithelial cell functions ( Iwashita et al . , 2003; Sturm et al . , 2005 ) . Interestingly , JMA recognition of LS 174 T cells also promoted their reciprocal activation resulting in increased transcription of mucin 2 gene ( MUC2 ) , which was prevented by anti-MR1 blocking antibodies ( Figure 8F ) . In control experiments , similar MUC2 expression upregulation was observed in LS 174 T cells treated with TNF-α ( Iwashita et al . , 2003 ) in the absence of MR1T cells ( Figure 8F ) . These results unraveled the capacity of some MR1T cells to modulate MUC2 expression in intestinal epithelial cells , and therefore suggested their possible contribution to intestinal epithelial barrier homeostasis . Taken together , these data identify MR1T cells as a novel population of human MR1-restricted T lymphocytes that may mediate diverse immunological functions . These findings indicate that the repertoire and reactivity of MR1-restricted T cells in healthy individuals is far broader than was previously appreciated .
Here we report that functionally diverse human T cells respond to MR1 in the absence of microbial antigens , thus extending the role of MR1 beyond the presentation of microbial riboflavin precursors to MAIT cells . While the structure of MR1 resembles that of MHC class I and related proteins , this molecule also displays various unique properties . A key distinguishing feature is the presence of a bulky antigen-binding pocket that binds small molecules via either hydrophilic or hydrophobic interactions . The total volume of the MR1 antigen-binding pocket is far larger than any of the MR1 ligands reported to date , suggesting capacity to bind a wider repertoire of antigens than that known so far . A direct consequence of this is the likely existence of other T cell populations that recognize MR1-presented antigens distinct from the known riboflavin precursors or folate-derivatives . Our study indicates the existence of an unpredicted population of MR1-restricted T cells ( designated MR1T cells ) outside of the MAIT cell compartment . Human MR1T cell clones displayed differential recognition of a variety of target cells expressing constitutive levels of MR1 . The different recognition of target cells might reflect different antigen presentation capacity , although they equally presented microbial antigens to MAIT cells , or the presence of different antigens . Indeed , preliminary antigen purification from a representative target cell line identified two diverse fractions that each stimulated only one of two MR1T cell clones , both derived from the same donor . Although the nature of these molecules remains to be determined , they do not derive from RPMI 1640 culture medium and are present in tumor cells grown in vivo . Furthermore , initial characterization of the molecules showed that they formed stable complexes with plastic-bound MR1 without forming a Schiff base and activated specific MR1T cells without need of APC processing . Using as APCs A375 cells expressing high levels of MR1 , we estimated that MR1T cells represent between 1:2500 and 1:5000 of total circulating T cells . These frequency estimates could not be determined by direct ex vivo identification of MR1T cells due to lack of specific markers , and were obtained using two approaches . In the first , frequencies were calculated after overnight stimulation and subtracting the frequency of T cells responding to MR1-negative APC to the frequency of cells responding to MR1-positive APC . In another set of experiments the overnight-stimulated T cells were immediately cloned without previous in vitro expansion and individual MR1T cell clones were identified with functional assays . These observations indicated that MR1T cell are readily detectable in the blood of healthy individuals with a size that resembles that of MHC-restricted memory T cells recognizing pathogen-derived peptides in pathogen-exposed individuals ( Bacher and Scheffold , 2013; Lucas et al . , 2004; Su et al . , 2013; Yu et al . , 2015 ) . The calculated range of MR1T cell frequency could however be underestimated as A375 cells utilized for the frequency studies might lack some MR1T cell antigens and therefore fail to stimulate a fraction of MR1T cells . Collectively , our data suggest that MR1T cells are not rare T lymphocytes in human blood . In some respects , MR1T cells resemble another population of autoreactive T cells restricted to non-polymorphic antigen presenting molecules and present at high frequency in human blood . These T cells react to self-lipids presented by CD1 molecules ( de Jong , 2010; de Lalla , 2011; Dellabona et al . , 2015 ) . In analogy to MR1T cells , CD1-self-reactive T cells display a polyclonal TCRα and β usage , and are functionally heterogeneous ( de Jong , 2010; de Lalla , 2011; Dellabona et al . , 2015 ) . The physiological role of self-reactive T cells restricted by CD1 and MR1 molecules is not well characterized . It is tempting to speculate that these cells might be activated upon metabolic changes in target cells induced by environmental signals , infection or cell transformation . It is currently unclear whether MR1T cells can be classified into distinct subgroups based on phenotype , gene expression and function . Since MR1T cells exhibit non-invariant TCRs with diverse antigen specificities , express a wide range of different surface markers and release multiple soluble effector molecules , it will be challenging to identify these cells using MR1 tetramer staining combined with conventional phenotypic analyses . Interestingly , some activated MR1T cell clones also displayed atypical functions for T cells , including the release of growth factors VEGF and PDGF , which support mesenchymal cell proliferation , vasculogenesis , and angiogenesis ( Carmeliet , 2003 ) , suggesting a potential role in tissue remodeling processes . The functional variety of MR1T cells was also accompanied by differential expression of selected chemokine receptors , suggesting that this population may exert multiple functions at various sites throughout the body . By comparing the transcriptomes of two select MR1T cell clones , we observed that certain transcription factors were expressed in common by both cell types , whereas others were only expressed in a clone-specific fashion . The DGB70 clone was distinguished by expression of the regulator of T cell polarization and proliferation factor EGR-1 ( Shin et al . , 2009; Thiel and Cibelli , 2002 ) , the mediator of lymphocyte functional differentiation JUN-B ( Li et al . , 1999 ) , and the SREBF1 , a member of the SREBP gene family that regulates sterol synthesis , coordinates tissue-specific gene expression and controls T cell metabolism ( Kidani et al . , 2013; Shao and Espenshade , 2012 ) . In contrast , transcription factors unique to the DGB129 clone were the proliferation and apoptosis regulator MYC ( Kress et al . , 2015 ) , involved in metabolic programming of effector T cells ( Hough et al . , 2015 ) , and the cAMP responsive element binding factor CREM , which modulates T cell effector functions and cytokine gene expression ( Rauen et al . , 2013 ) . DGB129 cells also uniquely upregulated the HSP90AB1 heat shock protein gene , which mediates quality control of cytoplasmic proteins ( Taipale et al . , 2010 ) and is involved in the regulation of diverse cell surface receptor expression in T and NK cells ( Bae et al . , 2013 ) . These clear distinctions between the transcriptional profiles of the two MR1T cell clones indicate the use of different strategies for metabolic programming , cell cycle control and functional polarization and suggest specific roles for these cells . As sorted cells expressed equivalent levels of CD25 and CD137 after Ag stimulation ( see Figure 6—figure supplement 1 ) , and released comparable amounts of IFN-γ ( not shown ) we excluded a different strength of antigen stimulation as possible explanation of the diverse transcriptional profiles observed . Given that the two clones were isolated from the same donor , maintained in identical culture conditions and stimulated with the same type of APC , these divergent gene signatures could reflect unique priming events ( Boltjes and van Wijk , 2014 ) . A key finding of our study is the reactivity of a large fraction of MR1T cell clones to diverse types of target cells expressing constitutive low MR1 surface levels . These results indicated the capacity of MR1T cells to be activated in physiologic conditions , when limiting amounts of MR1-antigen complexes are available on the target cell surface . As consequence , these data raised the issue of what influence MR1T cells could exert on these target cells . Several MR1T clones tested recognized monocyte-derived DCs , thus suggesting that MR1-restricted reactivity to these APCs is quite frequent among MR1T cells . In addition , experiments performed with a representative MR1T cell clone revealed its capacity to promote monocyte-derived DCs maturation . These data suggest that MR1T cells may modulate DC functions as it has already been reported for other T lymphocyte populations including TCR γδ ( Devilder et al . , 2006 ) and CD1-restricted T cells ( Bendelac et al . , 2007; Vincent et al . , 2002 ) . Some MR1T cell clones were also stimulated by human intestinal epithelial LS 174 T cells . In co-culture experiments , we observed reciprocal MR1-dependent activation of both MR1T and LS 174 T cells , with the first releasing IFN-γ , IL-8 and IL13 , and the latter up-regulating mucin 2 gene expression . These data suggested that some MR1T cells might influence epithelial cell function at mucosal sites , perhaps promoting innate defense and/or inflammation . In conclusion , we report that in addition to microbial metabolite-sensitive MAIT cells , human blood contains a novel population of MR1-restricted T cells that can recognize different antigens present in distinct target cells and exhibits a variety of effector functions . Based on the diversity of effector molecules they release following antigen stimulation , MR1T cells might drive inflammatory responses , support B cell function , mediate DC licensing , promote tissue remodeling , and contribute to mucosal homeostasis by enhancing innate defenses at the epithelial barrier . Future studies are therefore likely to uncover the roles of MR1T cells in human diseases .
The following cell lines were obtained from American Type Culture Collection: A375 ( human melanoma ) , THP-1 ( myelomonocytic leukemia ) , J . RT3-T3 . 5 ( TCRβ-deficient T cell leukemia ) , HEK 293 ( human embryonic kidney ) , CCRF-SB ( acute B cell lymphoblastic leukemia ) , Huh7 ( human hepatoma ) , HCT116 ( human colorectal carcinoma ) , LS 174T ( goblet-like cells from colon adenocarcinoma ) , and EMT6 ( mouse breast carcinoma ) . SKW-3 cells ( TCRα- and β-deficient human T cell leukemia ) were obtained from the Leibniz-Institute DSMZ-German Collection of Microorganisms and Cell Cultures . All the used cells were routinely tested for mycoplasma contamination and were negative . None of the cell lines used in this study is present in the database of commonly misidentified cell lines . Cells lines were not authenticated . Two representative MAIT clones ( MRC25 and SMC3 ) were used in this study generated from blood of two different healthy donors and maintained in culture as previously described ( Lepore et al . , 2014 ) . MR1T cells were isolated from the peripheral blood of healthy individuals . Briefly , T cells purified by negative selection were stimulated with irradiated ( 80 Gray ) A375-MR1 cells ( ratio 2:1 ) once a week for three weeks . Human rIL-2 ( 5 U/ml; Hoffmann-La Roche ) was added at day +2 and+5 after each stimulation . Twelve days after the last stimulation cells were washed and co-cultured overnight with A375-MR1 cells ( ratio 2:1 ) . CD3+CD69+CD137high cells were then sorted and cloned by limiting dilution in the presence of PHA ( 1 µg/ml , Wellcome Research Laboratories ) , human rIL-2 ( 100 U/ml , Hoffmann-La Roche ) and irradiated PBMC ( 5 × 105 cells /ml ) . In other experiments , MR1 T cells clones were generated using the same protocol from sorted CD3+CD69+CD137high upon a single overnight stimulation with A375-MR1 cells ( ratio 2:1 ) . T cell clones were periodically re-stimulated following the same protocol ( Lepore et al . , 2014 ) . Monocytes and B cells were purified ( >90% purity ) from PBMCs of healthy donors using EasySep Human CD14 and CD19 positive selection kits ( Stemcell Technologies ) according to the manufacturer instructions . Mo-DCs were differentiated from purified CD14+monocytes by culture in the presence of GM-CSF and IL-4 as previously described ( Lepore et al . , 2014 ) . A human MR1A cDNA construct linked to β2m via a flexible Gly-Ser linker was generated by PCR as previously described ( Lepore et al . , 2014 ) . The K43A substitution in the MR1A cDNA was introduced into the fusion construct using the following primers: MR1K43A_f 5′-CTCGGCAGGCCGAGCCACGGGC and MR1K43A_r 5′GCCCGTGGCTCGGCCTGCCGAG . Resulting WT and mutant constructs were cloned into a bidirectional lentiviral vector ( LV ) ( Lepore et al . , 2014 ) , provided by Jürg Schwaller , Department of Biomedicine , University of Basel . HEK293 cells were transfected with individual LV-MR1A-β2m constructs together with the lentivirus packaging plasmids pMD2 . G , pMDLg/pRRE and pRSV-REV ( Addgene ) using Metafectene Pro ( Biontex ) according to the manufacturer instructions . A375 , THP-1 , CHO and J558 cells were transduced by spin-infection with virus particle containing supernatant in presence of 8 µg/ml protamine sulfate . Surface expression of MR1 was assessed by flow cytometry and positive cells were FACS sorted . β2m-MR1-Fc fusion construct was obtained using human MR1A-β2m construct described above as template . DNA complementary to β2m-MR1A gene was amplified by PCR using primers: β2mXhoI_f 5’- CTCGAGATGTCTCGCTCCGTGGCCTTA and MR1-IgG1_r 5’-GTGTGAGTTTTGTCGCTAGCCTGGGGGACCTG , thus excluding MR1 trans-membrane and intracellular domains . The DNA complementary to the hinge region and CH2-CH3 domains of human IgG1 heavy chain was generated using the following primers: NheI-hinge-f 5’-CAGGTCCCCCAGGCTAGCGACAAAACTCACAC and IgG1NotI_r 5’-GCGGCCGCTCATTTACCCGGAGACAGGGAGA from pFUSE-hIgG1-Fc1 ( InvivoGen ) . The β2m-MR1A and IgG1 PCR products were joined together using two-step splicing with overlap extension PCR and the resulting construct subcloned into the XhoI/NotI sites of the BCMGSNeo expression vector . CHO-K1 cells were transfected with the final construct using Metafectene Pro ( Biontex ) , cloned by limiting dilutions and screened by ELISA for the production of β2m-MR1-Fc fusion protein . Selected clones , adapted to EX-CELL ACF CHO serum-free medium ( Sigma ) , were used for protein production and β2m-MR1-Fc was purified using Protein-A-Sepharose ( Thermo Fisher Scientific ) according to the manufacturer instructions . Protein purity was verified by SDS-PAGE and Western Blot . Protein integrity was assessed by ELISA , using anti-β2M mAb ( HB28 ) as capture and the conformation-dependent anti-MR1 mAb 25 . 6 as reveling ( Kjer-Nielsen et al . , 2012 ) . Cell surface labeling was performed using standard protocols . Intracellular labeling was performed using the True-Nuclear Transcription Factor Buffer Set according to the manufacturers’ instructions . The following anti-human mAbs were obtained from Biolegend: CD4-APC ( OKT4 ) , CD8α-PE ( TuGh4 ) , CD161-Alexa Fluor 647 ( HP-3G10 ) , CD69-PE ( FN50 ) , CD3-PE/Cy7 , Brilliant Violet-711 , or Alexa-700 ( UCHT1 ) , CD137-biotin ( n4b4-1 ) , CXCR3-Brilliant Violet 421 ( G025H7 ) , CD83-biotin ( HB15e ) and TRAV1-2- PE ( 10C3 ) . CD86-FITC ( 2331 ) , CCR4-PECy7 ( 1G1 ) and CCR6-PE ( 11A9 ) mAbs were from BD Pharmingen . All these mAbs were used at 5 µg/ml . Biotinylated mAbs were revealed with streptavidin-PE , -Alexa Fluor 488 , or -Brilliant violet 421 ( 2 µg/ml , Biolegend ) . The MR1-specific mAb clone 26 . 5 ( mouse IgG2a ) was provided by Ted Hansen , Marina Cella and Marco Colonna , Washington University School of Medicine , St . Louis ( MO ) ( Lepore et al . , 2014 ) . Unlabeled MR1-specific mAbs were revealed with goat anti-mouse IgG2a-PE ( 2 µg/ml , Southern Biotech ) . Samples were acquired on LSR Fortessa flow cytometer ( Becton Dickinson ) . Cell sorting experiments were performed using an Influx instrument ( Becton Dickinson ) . Dead cells and doublets were excluded on the basis of forward scatter area and width , side scatter , and DAPI staining . All data were analyzed using FlowJo software ( TreeStar ) . TCRα and β gene expression by MR1T cell clones was assessed either by RT-PCR using total cDNA and specific primers , or by flow cytometry using the IOTest Beta Mark TCR Vβ Repertoire Kit ( Beckman Coulter ) according to the manufacturers’ instructions . For RT-PCR , RNA was prepared using the NucleoSpin RNA II Kit ( Macherey Nagel ) and cDNA was synthesized using Superscript III reverse transcriptase ( Invitrogen ) . TCRα and β cDNAs were amplified using sets of Vα and Vβ primers as directed by the manufacturer ( TCR typing amplimer kit , Clontech ) . Functional transcripts were identified by sequencing and then analyzed using the ImMunoGeneTics information system ( http://www . imgt . org ) . The cDNA-sequencing data set has been deposited in the GenBank repository under accession numbers MF085360-MF085372 . Total cell lysates were generated from a single pellet of 2 . 5 × 109 THP-1 cells via disruption in water with mild sonication . The sonicated material was then centrifuged ( 15 , 000 g for 15 min at 4°C ) and the supernatant collected ( S1 ) . Next , the pellet was re-suspended in methanol , sonicated , centrifuged ( 15 , 000 g for 15 min at 4°C ) , and the supernatant obtained ( S2 ) was pooled with the S1 supernatant . The final concentration of methanol was 10% . The total cell extract ( pool of S1 and S2 ) was then loaded onto a C18 Sep-Pak cartridge ( Waters Corporation ) and the unbound material was collected and dried ( fraction E-FT ) . Bound material was eluted in batch with 75% ( fraction E1 ) and 100% methanol ( fraction E2 ) , collected and dried . The E-FT material was re-suspended in acetonitrile/water ( 9:1 vol/vol ) and loaded onto a NH2 Sep-Pak cartridge ( Waters Corporation ) . Unbound material ( fraction N-FT ) and 4 additional fractions ( N1-N4 ) were eluted with increasing quantities of water . Fraction N1 was eluted with 35% H2O , fraction N2 with 60% H2O , fraction N3 with 100% H2O , and fraction N4 with 100% H2O and 50 mM ammonium acetate ( pH 7 . 0 ) . Fractions N1-N4 were then dried and stored at −70°C until use . All collected dry-frozen fractions were then thawed and re-suspended in H2O 20% methanol ( fractions E1 , E2 and N-FT ) or 100% H2O ( all other fractions ) prior to being tested in T cell activation assays . Mouse EMT6 breast tumors were provided by A . Zippelius and were prepared as described ( Zippelius et al . , 2015 ) . Freshly excised tumors were extensively washed in saline , weighted and 4 g masses were homogenized in 7 ml of HPLC-grade water using a Dounce tissue grinder . Tumor homogenate underwent two freeze-thaw cycles , centrifuged ( 3250 g ) for 10 min at 4°C , and supernatant was collected and stored at −70°C . The pellet was extracted a second time with 2 ml of HPLC-grade water , centrifuged ( 5100 g ) for 10 min at 4°C and the supernatant was collected and stored at −70°C . The pellet was further extracted with 9 ml of HPLC-grade methanol for 5 min at room temperature by vortexing , centrifuged ( 5100 g ) for 10 min at 4°C , and supernatant collected . The three supernatants were pooled , dried , and resuspended in water:methanol ( 10:1 ) . Material was fractionated using C18 and NH2 Sep-Pak cartridges as above . MR1-restricted T cells ( 5 × 104/well unless otherwise indicated ) were co-cultured with indicated target cells ( 5 × 104/well ) in 200 µl total volume in duplicates or triplicates . T cells were cultured with indicated APCs for 24 hr . In some experiments , anti-MR1 mAbs ( clone 26 . 5 ) or mouse IgG2a isotype control mAbs ( both at 30 µg/ml ) were added and incubated for 30 min prior to the addition of T cells . In some experiments A375-MR1 and THP1-MR1 cells were cultured four days in PBS supplemented with 5% human AB serum . Cells were washed every day and then used to stimulate T cell clones upon assessment of target cell viability by trypan blue staining . In such type of experiments , T cell activation assays were also performed in PBS supplemented with 5% human AB serum . E . coli lysate was prepared from the DH5α strain ( Invitrogen ) grown in LB medium and collected during exponential growth . Bacterial cells were washed twice in PBS and then lysed by sonication . After centrifugation ( 15 , 000 g for 15 min ) , the supernatant was collected , dried , and stored at −70°C . APCs were pulsed for 4 hr with E . coli lysate equivalent to 108 CFU/ml ( unless otherwise indicated ) before addition of T cells . In some experiments , APCs were pre-incubated with 6-FP or Ac-6-FP ( Schircks Laboratories ) or pulsed with 6 , 7-dimethyl-8-D-ribityllumazine ( RL-6 , 7-diMe; 150 µM; Otava Chemicals ) for 4 hr before co-culture with T cells . In control experiments with TCR γδ cells , the APCs were first treated for 6 hr with zoledronate ( 10 µg/ml , Sigma ) prior to T cell addition . Activation experiments with plate-bound recombinant human β2m-MR1-Fc were performed by coating β2m-MR1-Fc onto 96 well plates ( 4 µg/ml ) and loading with cartridge-purified cell lysates for 4 hr at 37°C before washing twice and adding T cells . Supernatants were collected after 24 hr and IFN-γ or GM-CSF were assessed by ELISA . Multiple cytokines and chemokines in cell culture supernatants were analyzed using the Milliplex MAP human cytokine/chemokine magnetic bead panel – Premixed 41-plex ( HCYTMAG-60K-PX41; Merck Millipore ) according to the manufacturer’s instructions . Samples were acquired on a Flexmap 3D system ( Merck Millipore ) and Milliplex analyst software was used to determine mean fluorescence intensity and analyte concentration . Mo-DCs were co-cultured with MR1T cell clones at 1:1 ratio . When indicated Mo-DCs were pre-incubated with anti-MR1 mAbs ( clone 26 . 5 , 30 µg/ml ) for 30 min prior addition of T cells . Supernatants were harvested after 48 hr and tested for IFN-γ release by ELISA to assess T cell activation . Cells were collected at the same time and analyzed by FACS for CD83 and CD86 expression . T cells were excluded by the analysis using anti-CD3 mAbs . LS 174 T cells were co-cultured with MR1T cell clones at 1:1 ratio in the presence or absence of anti-MR1 ( clone 26 . 5 ) or mouse IgG2a isotype control mAbs ( both at 30 µg/ml , incubated with APCs 30 min prior addition of T cells ) . MR1T cell activation was evaluated by measuring IFN-γ , IL-8 ( R&D ) and IL-13 ( Pharmingen ) in the supernatants by ELISA . Q-PCR for MUC2 gene expression by LS 174 T cells was performed in a 20 µl reaction volume containing 0 . 5 µM of each primer and 2 . 5 µl of cDNA in Power SYBRgreen MasterMix ( Applied Biosystems ) . β-actin was used as reference gene . All the reactions were carried out in triplicate . The following method was run using an ABI 7500 Fast Real-Time PCR System ( Applied Biosystem ) : initial incubation for 20 s at 50°C , incubation for 10 min at 95°C , 40 cycles of 15 s at 95°C and 1 min at 60°C . The PCR primers used were as follows: β-actin_f 5′-GCCACCCGCGAGAAGATGA , β-actin_r 5′-CATCACGATGCCAGTGGTA; MUC2_f 5′-ACCCGCACTATGTCACCTTC , MUC2_r 5′-GGGATCGCAGTGGTAGTTGT . Changes in gene expression were quantified using the ΔΔCt method . TCRα and β functional cDNA from the MAIT cell clone MRC25 were cloned into the XhoI/NotI sites of the BCMGSNeo expression vectors and the resulting constructs were used to co-transfect J . RT3-T3 . 5 cells by electroporation . Transfectants expressing TRAV1-2 and CD3 were FACS sorted . The TCRα and β functional cDNA from MR1T clones were cloned into the XmaI/SalI sites of a modified version of the p118 lentiviral expression vector ( Amendola et al . , 2005; Lepore et al . , 2014 ) . SKW-3 cells were transduced with virus particle-containing supernatant generated as described above . Cells were FACS sorted based on CD3 expression . Comparative transcriptome analysis between resting and A375-MR1 stimulated MR1T cell clones was performed by RNA-sequencing . T cells were stimulated with A375-MR1 cells and after 20 hr DAPI−CD3+CD25+CD137+ cells were sorted , counted , and immediately frozen . Total RNA was extracted using Arcturus PicoPure RNA Isolation kit ( Applied Biosystems ) according to the manufacturer's protocol . RNA quality , assessed using the Agilent Bioanalyzer , exhibited RNA integrity number ( RIN ) ≥6 . 9 . Next , cDNA libraries were prepared using 200 ng total RNA and a 2 µl volume of a 1:500 dilution of ERCC RNA Spike in Controls ( Ambion ) . Samples were subjected to cDNA library synthesis using the TruSeq Stranded mRNA Library Preparation Kit ( Illumina ) according to the manufacturer's protocol with minor modifications; 13 PCR cycles used , and two additional rounds of purification with Agencourt Ampure XP SPRI beads ( Beckman Courter ) to remove double-stranded cDNA >600 bp in size . The length distribution of the cDNA libraries was monitored using DNA 1 , 000 kits with the Agilent Bioanalyzer . The samples were then subjected to an indexed PE sequencing run of 2 × 51 cycles on an Illumina HiSeq 2000 . Raw RNA-seq reads in fastq format were aligned to the hg19 genome assembly using STAR aligner ( Dobin et al . , 2013 ) . Gene annotations were derived from GENCODE version ( Harrow et al . , 2012 ) and counts of reads mapping over gene features were obtained using the featureCount method of the R/bioconductor subread packages ( Liao et al . , 2014 ) . EdgeR ( Robinson et al . , 2010 ) was used to generate a differentially expressed gene ( DEG ) list . From the DEGs , a global network of transcription factors and their target genes was created using the Encode chip-seq data as described ( Narang et al . , 2015 ) . Using the list of DEGs ( FDR < 0 . 05 , log2 fold-change >2 ) , a subnetwork of the global transcription factors-gene network was created and centrality scores for nodes were generated by both PageRank and betweeness centrality algorithms using NetworkX library in Python . The network was visualized using Cytoscape software and betweeness centrality score were used to define the radius of the nodes ( circles in the network ) . The RNA-sequencing data set has been deposited in the Gene Expression Omnibus repository under accession code GSE81063 . For cytokine secretion assays and Q-PCR data were analyzed using Unpaired Student’s t-test ( Prism 6 , GraphPad software ) . | White blood cells called T cells recognize germs and infected cells , and get rid of other cells in the body that look different to healthy cells – for example , tumor cells . These activities all depend on a molecule called the T cell receptor ( or TCR for short ) , which is found on the surface of the T cells . Each TCR interacts with a specific complex on the surface of the target cell . One of the molecules recognized by the TCR is known as MHC class I-related ( shortened to MR1 ) . This molecule attracts TCRs to infected cells , but it was not know if the MR1 molecule could attract TCRs to cancer cells too . Lepore et al . now show that there are indeed T cells in humans that recognize cancer cells through interaction with the MR1 molecules produced by the cancer cells . This new group of T cells has been named MR1T , and the cells can be easily detected in the blood of healthy individuals . The cells can be classified as a new cell population based on their capacity to recognize MR1 and how they react with different types of cancer cells . Importantly , the MR1 that attracts these TCRs is the same in all people , and so the same TCR may recognize MR1-expressing cancer cells from different patients . The next challenge is to identify MR1T cells that recognize and kill cancer cells from different tissues . These studies will hopefully pave the way for new and broader strategies to combat cancer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"immunology",
"and",
"inflammation"
] | 2017 | Functionally diverse human T cells recognize non-microbial antigens presented by MR1 |
We present results from a longitudinal study conducted over 10 years in a sample of 126 8–33 year olds demonstrating that adolescent development of working memory is supported by decreased variability in the amplitude of expression of whole brain states of task-related activity . fMRI analyses reveal that putative gain signals affecting maintenance and retrieval aspects of working memory processing stabilize during adolescence , while those affecting sensorimotor processes do not . We show that trial-to-trial variability in the reaction time and accuracy of eye-movements during a memory guided saccade task are related to fluctuations in the amplitude of expression of task-related brain states , or brain state variability , and also provide evidence that individual developmental trajectories of reaction time variability are related to individual trajectories of brain state variability . These observations demonstrate that the stabilization of widespread gain signals affecting already available cognitive processes underlies the maturation of cognition during adolescence .
Working memory ( WM ) , the ability to retain information online in order to guide goal directed behavior , is evident in rudimentary form as early as infancy ( Diamond and Goldman-Rakic , 1989; Gilmore and Johnson , 1995 ) , indicating that core WM processes are available throughout development . However , protracted improvements in performance , typically measured as the percentage of correctly performed trials or as changes in mean behavioral response metrics , such as reaction time and accuracy , demonstrate that WM continues to develop across adolescence and into early adulthood ( Gathercole et al . , 2004; Luna et al . , 2004; Alloway et al . , 2006; Crone et al . , 2006; Thomason et al . , 2009; Luna et al . , 2015 ) . Developmental decreases in behavioral variability during cognitive tasks are also evident through adolescence ( McIntosh et al . , 2008; Klein et al . , 2011; Tamnes et al . , 2012 ) , however this has not been directly examined in the context of WM . Behavioral variability , or intra-individual variability , is a sensitive barometer of cognitive function; excessive variability frequently attends disorders such as schizophrenia ( Kaiser et al . , 2008 ) ; ADHD ( Munoz et al . , 2003 ) ; and age-related cognitive decline ( MacDonald et al . , 2006 ) . The association between behavioral variability and cognitive performance suggests that adolescent stabilization of behavior reflects the continuing alteration of fundamental aspects of brain processing that support the transition to adult-like levels of performance . Mechanistically accounting for the stabilization of behavior is critical to our understanding of adolescent neural development . Thus , in the present study , we examine the neural processes that underlie behavioral variability . Behavioral variability has been found to be associated with fluctuations in neural ( Newsome et al . , 1989; Cohen et al . , 2009; Nienborg and Cumming , 2009 ) or blood-oxygen-level dependent ( BOLD ) signals occurring within individual brain areas ( Wagner et al . , 1998; Ress and Heeger , 2003; Yarkoni et al . , 2009 ) , and networks of brain areas ( Rosenberg et al . , 2016 ) . Several lines of research suggest that these brain/behavior relationships are driven , at least in part , by trial-to-trial variability in gain modulating signals ( Fox et al . , 2006; Eldar et al . , 2013; Rabinowitz et al . , 2015 ) that enhance the activity of individual neurons or brain areas experiencing net excitation and further suppress the activity of neurons or areas experiencing net inhibition ( Servan-Schreiber et al . , 1990 ) . Recent electrophysiological evidence indicates that some gain modulating signals are shared across cortical areas and that moment-to-moment variation in such distributed gain signaling may account for a significant portion of neural and behavioral variability ( Rabinowitz et al . , 2015 ) . Additionally , the structure of neural covariance within populations of simultaneously recorded sensory neurons is best explained as the result of multiple ongoing gain modulating signals ( Rabinowitz et al . , 2015 ) , raising the possibility that different sources of gain variability affect neural activity in a functionally targeted way . Widespread or global gain modulating processes would not , by definition , change the spatial distribution or ‘pattern’ of task evoked neural activity . Rather , they would influence the amplitude with which they are expressed , manifesting as trial-to-trial variability in the amplitude of expression of whole-brain patterns of activity that support task-relevant processes . We refer to this hypothesized phenomenon here as ‘brain state’ variability . With this operationalized definition of gain modulation , once an average pattern of task-evoked activity is known , the occurrence of fluctuations in global gain signals may be determined by measuring trial-to-trial differences in the amplitude of expression of the average whole-brain task state . In the present developmental functional magnetic resonance imaging ( fMRI ) experiment , we exploit this anticipated characteristic of global gain modulation to study the relationship between trial-to-trial variability in behavioral responses during a memory-guided saccade ( MGS ) task and trial-to-trial variability in widespread gain signals ( i . e . , brain state variability ) that occur near the time of the behavioral response . We explore the possibility that the reduction of behavioral variability observed during development is the result of stabilizing widespread gain signals . We performed fMRI on an accelerated longitudinal cohort of 126 subjects between the ages of 8 and 33 years ( Figure 1a ) as they performed a variant of the MGS task ( Hikosaka et al . , 1989 ) ( Figure 1b ) . On each trial , subjects first made a visually guided ‘encoding’ saccade to the target stimulus , which appeared at one of six locations: ±3 , 6 , or 9° along the horizontal visual meridian , during an initial visuomotor/encoding ( VME ) epoch . Children typically have difficulty suppressing orienting saccades to target stimuli ( Luna et al . , 2004 ) ; by allowing subjects to make the initial encoding saccade , we removed a possible age-related source of behavioral differences that would be related to response inhibition rather than WM performance . After making a saccade to the target , subjects returned their gaze to a central fixation point , marking the onset of the maintenance epoch . The working memory retrieval epoch began when fixation was extinguished and subjects generated a saccade to the remembered location . We varied the duration of the initial target presentation ( either 1 . 5 or 3 s ) and the duration of the delay ( 1 . 5 or 9 s ) . We measured the subject's gaze location in the scanner with an MR compatible infrared camera and eye-tracking system ( Model R-LRO6 , Applied Science Laboratory , Bedford , MA ) . We applied several levels of analysis: First we characterized developmental changes in mean behavioral performance and behavioral variability and determined the extent to which measures of behavioral variability constitute a distinct metric of developmental status beyond that provided by measures of mean behavioral performance . Second , we identified canonical whole brain patterns of activity ( brain states ) corresponding to visuomotor , WM maintenance , and retrieval processes and determined whether these states were similarly expressed across development . Third , we measured the relationship between trial-to-trial fluctuations in the amplitude of expression of the task-related brain states and single trial behavioral performance . Fourth , we examined developmental trajectories of brain state variability and its relationship with individual developmental trajectories of behavioral variability .
For each trial , we assessed two measures of performance: reaction time ( RT ) , the interval between the extinction of the fixation stimulus at the end of the delay interval and the initiation of the MGS , as well as saccadic error ( SE ) , the signed visual angle separating the horizontal location of the target and the end point of the MGS . For the four task conditions during a session , we computed average RT and the standard deviation of RT . As additional measures of mean behavioral performance and behavioral variability , we defined saccade inaccuracy as the absolute value of the average SE for a given target and saccade imprecision as the standard deviation of SE for each target . Using a linear mixed-effects model to account for the longitudinal nature of the data , we found , consistent with prior findings ( Luna et al . , 2004 ) , that average RT and inaccuracy decreases during development following an age−1 trajectory ( Figure 1c , e ) . We estimated how much each behavioral measure changed at the group level between the ages of 8 and 33 and observed a reduction in average RT amounting to approximately 239 ms , a 39% decline ( t ( 1346 ) =9 . 30; p=6 . 03e-20 Inaccuracy decreased by approximately 0 . 25° , or 47% ( t ( 1346 ) =3 . 11; p=0 . 0019 ) . Importantly , both measures of behavioral variability also decrease with development following an age−1 trajectory ( Figure 1d , f ) . The developmental change in the standard deviation of RT amounts to approximately 156 ms , a 67% reduction ( t ( 1346 ) =7 . 84; p=9 . 4e-15 ) . The imprecision of the MGS decreases with age , resulting in an estimated reduction of 0 . 87° of visual angle , a change , which amounts to roughly 35% ( t ( 1346 ) =4 . 03; p=5 . 83e-5 ) . In order to determine whether behavioral variability provides additional information beyond mean behavioral measures about the developmental status of a subject , we compared the performance of linear models that predicted subject age from either mean RT or inaccuracy ( null models ) to matched linear models containing the corresponding behavioral variability factor ( full models ) . We found that a null model predicting age from only mean RT was significantly improved by including the standard deviation of RT ( null model: DF=4 AIC=−2212 . 7 , Log-Likelihood=1110 . 4; full model: DF=5 AIC=−2220 . 7 , Log-Likelihood=1115 . 6; p=0 . 002 ) ( Burnham and Anderson , 2004 ) . Including imprecision as a term improved model performance , but the difference did not achieve statistical significance ( null model: DF=4 , AIC=−2108 . 4 , Log-Likelihood=1058 . 2; full model: DF=5 AIC=−2110 . 1 , Log-Likelihood=1060 . 0 , p=0 . 052 ) . Thus , of the two measures of behavioral variability , reaction time variability , but not imprecision , appears to measurably reflect a unique aspect of cognitive development . Performance of the MGS task requires coordinated activation of many brain regions involved in visuomotor , WM maintenance , and retrieval processes . Each of these processes is associated with a distinct whole-brain pattern of activity , or ‘brain state’ , which is expressed at the appropriate times during a trial . Developmental changes in average global gain would result in changes in the average amplitude of expression of task-related brain states across trials; subjects with greater mean global gain would express the task-related brain states to a greater extent , while subjects with reduced global gain would express the brain states with reduced amplitude . We therefore sought to estimate the canonical brain state patterns associated with the visuomotor , maintenance , and retrieval processes involved in the MGS task and to determine whether adolescent development is associated with changes in the mean amplitude of their expression . Using a simple dimensionality reduction technique based on linear regression ( see Materials and methods ) , we constructed whole brain patterns of activity , that is , brain states , associated with visuomotor/encoding ( VME ) , WM maintenance , and retrieval processes . These task-related brain state patterns were extracted from idealized time courses of BOLD activity ( see Materials and methods ) observed during the long delay trials ( Figure 2a ) . To represent VME processes we extracted the average patterns of BOLD signal occurring 6 s after the initial ‘encoding’ saccade . To represent retrieval processes , we again extracted patterns of activity present 6 s after the MGS . This delay allowed the BOLD signals associated with these processes to reach their peaks ( Handwerker et al . , 2004 ) . The pattern of activity associated with working memory maintenance was extracted from the TR immediately prior to the execution of the MGS , allowing us the purest estimate of delay period activation ( furthest from the preceding visually-guided saccade , without intruding into the subsequent MGS ) . We orthogonalized each of the brain state patterns to ensure that they captured unique aspects of task activity by regressing the VME-related patterns from maintenance-related patterns , and regressing both VME- and maintenance-related patterns from the retrieval related patterns . This process removed remaining components of VME-related activity from the maintenance activity and importantly , allowed us to remove the pattern of activity associated with visuomotor responses from the retrieval-related pattern occurring during the MGS . Implicit in this procedure is the assumption that VME , maintenance , and retrieval processes are associated with distinct and consistent patterns of whole-brain BOLD activation that are expressed with the time course of a hemodynamic response . Some neuronal gain modulators , particularly those acting through cholinergic pathways , alter gain with hemispheric specificity , similar to the effects of directed spatial attention ( Sarter and Bruno , 1997; Furey et al . , 2000; Bentley et al . , 2004 ) . To account for such potential hemispheric differences in gain signaling , we decomposed each brain state into two component patterns: a target hemifield specific ‘spatial’ component and a target hemifield non-specific , ‘mean’ component . Mean brain state components correspond to the average whole-brain patterns of activity associated with VME , maintenance , or retrieval , regardless of which visual hemifield the target was presented in ( Figure 2b upper panels ) . Spatial components were constructed by computing the difference —right minus left— between brain state patterns determined separately for right and left side targets; they reflect the differences in activity during each task epoch resulting from the changes in the target’s visual hemifield ( Figure 2b lower panel ) . The whole brain patterns of activity observed during different epochs of the task can therefore be approximated with linear combinations of the mean and spatial components of the brain state patterns . For instance , maintenance related activity observed during trials in which the target appears in the right visual hemifield , can be approximated by adding the mean and spatial components of the maintenance brain state patterns , while maintenance activity observed during trials in which the target appears in the left visual hemifield can be approximated by subtracting the spatial component of the maintenance brain state from the mean component . Taken together , the brain states characterize the patterns of engagement of canonical regions underlying the VME epoch ( e . g . , frontal eye fields ) , maintenance ( e . g . , prefrontal and frontal eye fields ) and the non-visuomotor aspects of retrieval and response ( e . g . , preSMA ) ( Anderson et al . , 1994; Fried et al . , 1991 ) . We verified that the resulting brain state components captured the relevant whole-brain patterns of BOLD signal associated with specific task epochs , by projecting each whole-brain volume of a subject’s average trial time course , onto the complete set of brain state components using linear regression ( see Materials and methods ) . We performed this operation separately for each task condition . By temporally ordering the regression weights associated with each brain state component from each TR of the average trial time course , we converted the whole-brain time courses of task activity into time courses of task-related brain state expression ( Figure 3 ) . This procedure is conceptually similar to principle component/independent component analyses in which whole-brain voxel-wise time series are converted into time courses of expression of whole-brain patterns . Here , however , the individual components , that is , brain states , have known behavioral and cognitive processes with which they have been empirically associated . These brain state patterns , although derived from only the long delay trials , also served as an effective basis for describing the whole-brain patterns of BOLD signal evoked during the short delay trials as well ( Figure 3; lower panel ) . As expected , given our approach for constructing the brain states patterns , the time courses of brain state expression for all subjects exhibited similar temporal characteristics . However , it remained possible that there might be age-related differences in mean global gain , which would affect the average amplitude of brain state expression . To determine whether such differences were present , we performed omnibus F-tests on the brain state time courses for each task condition to assess the null hypotheses that all of the coefficients for Age*TR ( Age*TR*TargetHemifield in the case of the spatial brain state components ) were equal to zero . We observed minimal age-related differences in the average time courses of brain state expression . These differences were observed for the spatial , but not mean , component of the VME states across all trial types ( all p<0 . 001 ) . We did not detect any significant age-related differences in the expression of either mean or spatial components of the maintenance state ( all p>0 . 14 ) . Results for age-related differences in the time courses of expression for the retrieval states were mixed: We observed no omnibus age-related differences within either of the long delay conditions , but within the long presentation short delay trials we observed a small age-related difference in the expression of the spatial retrieval state ( F ( 42 , 14784 ) =1 . 6; p=0 . 012 ) . Post-hoc examination of the individual Age*TR and Age*TR*TargetHemifield coefficients at each time point in the trial revealed that this effect was driven by slightly greater expression of the state by adults , across right and left side targets , during the 5th , 6th , and 8th TRs . However , in our post-hoc analysis no single time point reached significance ( minimum ( p ) =0 . 055 ) . We also observed that the mean retrieval state was differentially expressed across age within the short presentation short delay trials ( F ( 40 , 14112 ) =2 . 0; p<0 . 001 ) . Post-hoc analyses revealed that this effect was driven by a slightly greater expression of the mean retrieval state by adults during this condition during the 9th-13th TRs , well after the occurrence of peak expression for this state . From visual inspection it is clear that adults exhibit a slightly prolonged expression of the spatial component of the VME brain state during the different trials ( seen most prominently in Figure 3; lower panel ) . We also wanted to know whether the peak amplitude of spatial VME expression differed with age . From each session we examined the amplitude of peak expression of the spatial VME state for each trial type . Because the sign of expression of the spatial VME state varies depending on target hemifield , we extracted the maximum value of positive expression for right side trials , and we extracted the minimum value of expression for left side trials . If adults expressed the spatial VME state to a greater extent than children and adolescents due to greater average gain , this would result in greater positive expression for right side trials and reduced ( more negative ) peak expression during left side trials . We therefore examined the Age*Target interaction term , which we found did not reach significance ( t ( 2696 ) =1 . 59; p=0 . 111 ) . Combined , these results demonstrate that the set of brain state patterns provide a simplified low dimensional basis for describing BOLD signal changes evoked by the memory-guided saccade task . Importantly , age-related differences in the expression of the brain state patterns during task performance were minimal , and only the spatial component of the VME state exhibited consistent age-related differences in expression across trial types . Even here , however , the age-related differences were not ones of amplitude , but of duration , suggesting that mean global gain does not change during adolescent development . We hypothesized that behavioral performance is affected by fluctuations in global gain signals when they occur around the time of a behavioral response . The signature of variability in global gain signaling would be ‘brain state variability’ , or momentary fluctuations in the amplitude of expression of whole brain states of activity associated with ongoing neural processes . However , due to the delayed and prolonged nature of the BOLD response , the activity measured near the time of the MGS consists of multiple superposed states of activity associated with visuomotor , WM maintenance , and retrieval processes . Each of these processes may be affected differently by global gain variability and variability affecting each process might differentially affect behavioral performance . We therefore examined the trial-wise relationship between behavioral performance and the expression of the mean and spatial components of the VME , maintenance and retrieval brain state patterns in an interval of time centered on the occurrence of each MGS . After removing the mean trial responses from each voxel , we looked for remaining patterns of activity across the brain that matched each of the canonical brain states . To accomplish this , we projected the whole-brain pattern of BOLD signal residual values at each TR onto the set of canonical brain state patterns using linear regression . This approach converted the whole-brain residual time series into a time series of task-related brain state fluctuations , and allowed us to determine the extent to which each brain state component was over- or under-expressed at a particular point in time . We divided the resulting fluctuation time series for each brain state component into snippets , intervals of time centered on each MGS and extending ±15 TRs ( each TR=1 . 5 s ) before and after ( Figure 4a and b ) . Each snippet was associated with the RT and SE of a particular trial . After aligning snippets from each trial to the TR in which the MGS occurred ( TR 0 ) , we used regression models to measure the relationship between both trial-to-trial differences in RT ( z-scored ) and SE ( z-scored and rectified ) and trial-to-trial differences in the expression of each brain state component at different times relative to the MGS . In order to measure the contribution of brain state variability to behavioral variability , we first estimated the fraction of trial-to-trial behavioral variance accounted for by a null model that included terms for non-neural trial-to-trial factors ( e . g . target eccentricity ) and performance factors ( e . g . RT when testing SE and vice versa ) . Then we measured the additional fraction of behavioral variability that could be explained by including the measures of brain state expression from each TR in the snippets . The difference in explained variance between the full and null models indicates the fraction of trial-to-trial behavioral variability uniquely associated with brain state variability occurring at a particular time relative to the execution of a MGS ( Figure 5a and b [top panels] ) . As hypothesized , the relationship between brain state expression and both RT and SE peaked around the time that the MGS was executed . For trial-wise RT , brain state/behavior associations were significant beginning with the TR when the MGS was executed , peaking 1 TR after the saccade , and lasting for a total of 6 TRs . The relationship between trial-wise SE and brain state expression was similar , but much less prominent and evident during only the third TR following the MGS . Trials with faster RTs were associated with greater early expression of the mean VME and maintenance brain states ( TRs 0–2 ) , and reduced later expression of all mean states ( TRs 3–5 ) . Greater SE was associated with greater expression of the mean VME state ( TR 3 ) . RT and SE also covaried with fluctuations in the amplitude of expression of the spatial components of the brain state patterns ( bottom three rows of Figure 4a and b ) . The fastest RTs occurred when whole-brain patterns of task-related activity were biased in the direction of the target ( VME: TRs 1–3; maintenance and retrieval: TRs 1–2 ) . That is , the fastest right side trials were those in which brain states were expressed most ‘rightwardly’ , and the fastest left side trials were those in which the spatial brain states exhibited the greatest ‘leftward’ expression . Interestingly , greater target hemifield appropriate expression of the spatial VME state ( TRs 2–3 ) was associated with increased SE , indicating that increasing the amplitude of its expression does not improve all aspects of task performance . Increased hemifield appropriate expression of the spatial component of the maintenance states ( TR 3 ) , in contrast , was associated with reduced SE . That greater expression of VME brain states was associated with faster RT and greater SE , prompted us to examine the behavioral data for signs of a speed-accuracy trade-off . We found a significant quadratic relationship between z-scored RT and SE at the trial level indicating that , within a session , excessively fast and slow responses were associated with reduced accuracy ( t ( 16754 ) =4 . 64; p=3 . 52e-6 ) . Collectively , these results demonstrate that , trial-to-trial , both reaction time and accuracy of subjects’ responses covary with the amplitude of expression of whole-brain patterns of task related activity . Greater early expression and reduced later expression of the mean brain states for fast RT trials ( represented by the transition from blue to red in some rows of the lower panel of Figure 4a ) , could possibly be explained by trivial correlation between the timing of a saccade and the latency of the expression of the brain state . That is , on trials with longer RTs , the BOLD activity would be shifted later , causing the brain state to appear under-expressed early and over-expressed late relative to the average time course . To determine whether this relationship could account for the observed correlation with behavior , we performed a set of simulations to compare the temporal patterns of BOLD signal residuals for fast and slow RT trials that would result from timing- , amplitude- , and timing and amplitude-based relationships ( see Materials and methods ) . As depicted by Figure 5 , the contributions of timing and amplitude variability to brain state variability are distinguished by their effect on the temporal structure of the residual brain state time series . A purely timing based explanation of the mean brain state/RT relationship predicts that the integral of the mean brain state residual time series , for both fast and slow reaction time trials should converge to zero ( Figure 5 , column 1 ) . A relationship between reaction time and brain state expression mediated by fluctuations in the amplitude of expression of the brain state patterns predicts that the same time integrals converge to non-zero values of opposite sign ( Figure 5 , column 2 ) . A combination of timing- and amplitude-based relationships predicts an initial bifurcation of the time integrals of the fast and slow reaction time residuals that then a partial re-convergence ( Figure 5 , column 3 ) . We found that the trial-wise relationship between RT and the expression of the mean VME brain state —the state associated with eye-movements— was inconsistent with both a purely amplitude-based mechanism , and purely timing-based mechanism and instead is likely to reflect a mixture of both the trivial time shifted BOLD response , and a true relationship between gain fluctuation and performance ( Figure 5 , column 4 ) . If the stabilization of behavior during adolescence is related to a reduction in brain state variability , then the proportion of BOLD signal variability associated with brain state variability will decrease with age . We examined the subset of TRs ( 0–5; the highlighted region Figure 4a ) around each correct MGS that showed a significant relationship with trial-to-trial behavioral performance and computed SSbrain , the sum of whole-brain squared error associated with all brain state patterns across each TR as well as SSerror , the sum of the remaining squared error . For each session , we computed the ratio SSbrain/ ( SSbrain + SSerror ) , producing values we refer to here as total brain state variability , which corresponds to the fraction of residual whole-brain BOLD signal variability associated with brain state variability . We found that total brain state variability decreases with age ( Figure 6a ) after controlling for mean frame-wise displacement ( FD ) ( Power et al . , 2012; Siegel et al . , 2014 ) and the number of correctly performed trials ( t ( 333 ) =-3 . 35; p=9 . 0e-4 ) . The measure of total brain state variability , SSbrain , can be simply decomposed into a sum of contributions from the mean and spatial components of VME , maintenance , and retrieval brain state patterns; SSbrain = SSVME + SSMaint + SSRetrieval . We computed the ratio of each brain state component sum of squared error to ( SSbrain + SSerror ) , and found that the stability of each of the three sets of brain state components exhibit distinct developmental trajectories ( Figure 6b ) . The proportion of VME-related brain state variability did not significantly decrease with age ( t ( 333 ) =-1 . 47; p=0 . 14 ) . However , maintenance- and retrieval-related brain state variability both show significant age-related decreases ( t ( 333 ) =-3 . 8; p=1 . 8e-4 and t ( 333 ) =-5 . 53; p=6 . 27e-8 respectively ) . Pairwise comparisons of slopes reveal that trajectories of VME and retrieval-related variability can not be significantly distinguished from one another ( t ( 670 ) =0 . 60; p=0 . 54 ) , while maintenance-related brain state variability decreases more slowly through adolescence compared to both VME and retrieval-related variability ( t ( 670 ) = 2 . 44; p=0 . 014 and t ( 670 ) =-3 . 71; p=2 . 2e-4 respectively ) . Next , we sought to determine whether a systematic relationship exists between in-scanner motion and measures of brain state variability . We reasoned that if brain state variability was unrelated to movement-related artifacts , then our finding that brain state variability was reduced in older subjects would still hold if we selectively sub-sampled our data so that we compared a group of adults who moved excessively to a group of children who moved relatively little . The biased subsampling routine that we employed is based on a mean-matching technique that has been described in detail elsewhere ( Churchland et al . , 2010 ) . Briefly , we divided our data into two sets , split at the median age of our sample . We selectively drew samples from the two data sets such that , on average the older sample exhibited greater mean FD than the younger group . We found that reversing the relationship between motion and age does not significantly alter our finding that older subjects exhibited less brain state variability than younger subjects ( Figure 7 ) . Our analyses to this point have shown that trial-to-trial differences in behavioral performance are associated with brain state variability , which stabilizes during adolescent development . If the stabilization of behavior during adolescence is the result of reduced brain state variability , then subjects exhibiting the greatest longitudinal reduction in behavioral variability should be those that exhibit the greatest decline in brain state variability . To determine whether this is the case , we leveraged the longitudinal design of our dataset to examine how individual differences in the developmental trajectories of RT variability and saccade imprecision are related to individual developmental trajectories of brain state variability . We selected a subset of 29 subjects for whom we had at least four complete sessions of data , and estimated individual developmental slopes of total brain state variability . We modeled developmental changes in total brain state variability as a linear effect of age after observing superior performance compared to a model in which total brain state variability was fit with a age-1 term ( simulated likelihood ratio test with 5000 iterations , age−1 model: DF=9 AIC=−3116 . 2 , Log-Likelihood=15567 . 1; linear age model: DF=9 AIC=−3120 . 2 , Log-Likelihood=1569 . 1; p=0 . 015 ) ; . We also estimated individual regression weights for the developmental trajectories of reaction time variability and imprecision after controlling for task condition and mean reaction time . Here we used an age−1 term to model individual trajectories . Subjects exhibiting the greatest decreases in behavioral variability given their age have the greatest age−1 regression coefficients . In contrast , subjects exhibiting the greatest decrease in total brain state variability have the smallest age regression coefficients . Evidence consistent with the hypothesis that the developmental stabilization of behavior is driven by a reduction in brain state variability would be the existence of a negative correlation between the brain state and behavioral regression coefficients . We observed just such a negative relationship for reaction time variability ( r = −0 . 48; p=0 . 008 ) ( Figure 8a ) . This result remained significant ( p<0 . 05 ) when we limited our data set to subjects with 3–5 or more sessions as well . However , the within-subject relationship between brain state variability and saccade imprecision was not significant ( r=0 . 28; p=0 . 14 ) ( Figure 8b ) . Given the modest relationship between brain state variability and SE at the single trial level , we considered the possibility that we might be underpowered to detect the longitudinal relationship between total brain state variability and saccadic imprecision in our smaller sample size . We expanded our analyses to investigate whether individual differences in saccade imprecision were related to individual differences in brain state variability using our entire sample and including age−1 terms as covariates . Here we did observe a significant positive relationship between total brain state variability and saccade imprecision ( t ( 1344 ) =3 . 2; p=0 . 001 ) ( Figure 8c ) .
Reduced variability is a key component of the behavioral improvements that are observed during adolescent development . We demonstrated an example of this stabilization using a working memory task in which subjects’ performance of memory guided saccades improved on average and became less variable with age . To understand the neural basis of developmentally stabilized behavior , we investigated the relationship between variability in the reaction times and accuracies of eye-movements and fluctuations of global gain signals hypothesized to affect the amplitude of expression of whole-brain states of activity underlying distinct task-related processes . We found that while the average amplitude of expression of whole-brain task states was similar across subjects , regardless of their age , trial-to-trial variability in the amplitude of their expression decreased during adolescence and was correlated with trial-to-trial variability in the reaction time and accuracy of memory-guided saccades . Importantly , this brain state variability represented fluctuations in the amplitude of brain state expression across trials , not simply variability in the timing of their expression or global fluctuations in mean activity ( see Materials and methods ) . Additionally , variability occurring specifically in the expression of the mean and spatial components of the VME brain states associated with visuomotor processes mirrored the higher-order phenomenon of speed-accuracy trade-off , that is , greater VME expression was associated with faster responses and increased saccadic error . Greater expression of the spatial component of the working memory maintenance state , in contrast , was associated with faster responses and reduced saccadic error . These findings are broadly consistent with recent theoretical models ( Standage et al . , 2011 ) and empirical data from non-human primates ( Heitz and Schall , 2012 ) suggesting gain modulation plays a role the speed-accuracy trade-off . Appropriately balanced , independent variability in gain signals affecting VME and maintenance brain state expression , may explain the quadratic speed accuracy trade-off that we observed in our data . We hypothesized that developmental decreases in the variability of global gain signaling would result in more stable expression of task-related brain states . Accordingly , we determined whether the expression of brain states associated with visuomotor/encoding ( VME ) , maintenance , and retrieval processes , exhibited similar or different trajectories of variable expression across development . We found that the variability of the VME states did not decrease with age although they were significant predictors of single trial performance . Our task design did not allow us to dissociate the activity involved strictly in working memory encoding from that involved strictly in the visuomotor response , however , the re-expresssion of the VME states during the memory-guided saccade suggests that they are largely dominated by visuomotor activity . In contrast , working memory maintenance and retrieval processes , whose fluctuations were also related to trial-wise performance , showed significant decreases in the variability of their expression . Perhaps most significantly , we found a relationship between individual longitudinal changes in total brain state variability and changes in reaction time variability as well as a relationship between total brain state variability and memory-guided saccade imprecision after covarying for age . Combined , our findings provide evidence that adolescent developmental changes in behavioral variability are driven by the stabilization of gain signals specifically affecting cognitive processes while gain signals affecting sensorimotor processes continue to vary greatly across all ages . A complex interplay between top-down control ( Moore and Armstrong , 2003; Larkum et al . , 2004 ) and a mixture of contributions from several interconnected neuro-modulatory systems , each exerting its particular influence on ongoing sensorimotor , and cognitive processes ( Furey et al . , 2000; Luciana et al . , 1998; Rokem et al . , 2010; Chandler et al . , 2014 ) may underlie these developmental changes in brain state variability . Recent fMRI studies have shown that fluctuations in the activity of midbrain and brain stem nuclei affect resting state connectivity in what appears to be a functionally organized way ( Bär et al . , 2016 ) . Similarly , cholinergic modulation has been shown to amplify the spatially selective effects of perceptual processing and attention in a manner analogous to fluctuations in our spatial brain state components ( Furey et al . , 2000; Bentley et al . , 2004; Rokem et al . , 2010 ) . Finally , myelination and synaptic pruning , which continue to progress in critical brain systems ( Simmonds et al . , 2014 ) , occurring at different rates for different brain regions , ( Petanjek et al . , 2011 ) may also affect neural signal to noise ratios and play a role in the stability of gain signals that contribute to behavioral variability . Differing rates of development in any of these systems could produce distinct developmental trajectories for the components of brain state variability . The presence of brain state variability also bears upon the interpretation of brain/behavior correlations in general . In studies of single unit and population activity in non-human primates , correlations between the trial-to-trial fluctuations of neuronal activity and behavioral responses , often termed choice-probability ( CP ) or detect-probability ( DP ) , have been interpreted as signifying a neuron's causal role in the behavior ( Shadlen et al . , 1996 ) . It has been proposed , however , that brain/behavior relationships like CP and DP , might reflect a neuron’s covariation with a neuronal gain signals , such as attention , rather than direct causal involvement ( Nienborg and Cumming , 2009; Cohen and Kohn , 2011 ) . Brain state variability is consistent with this hypothesis and expands upon it in two ways 1 ) That brain state variability is the covariation of many task-related ( and presumably behaviorally relevant ) brain regions suggests that brain/behavior correlations like CP and DP should be wide-spread throughout task-related brain areas; and 2 ) Our finding of distinct developmental trajectories of brain state variability affecting different task-related processes suggests that fluctuations in multiple functionally specific global gain signals contribute to observed brain behavior correlations . This interpretation also gains support from recent electrophysiological evidence that multiple independent gain modulating signals are apparent within the activity of populations of neurons in sensory cortex ( Rabinowitz et al . , 2015 ) . An attractive model that would provide synthesis for our findings and those discussed above is that stable behavioral performance requires similarly stable allocation of top-down control processes , like attention . Such top-down processes partly exert their influence through multiple widespread gain signals that are functionally targeted , differentially affecting sensorimotor and cognitive processes . In this model , the stabilizing of working memory behavioral performance that we observed during adolescent development is the result of stabilizing those gain signals that affect working memory maintenance and retrieval processes . Our results , in sum , provide compelling evidence that core cognitive functions are online by childhood and what underlies cognitive development through adolescence is a fine-tuning of the ability to stabilize the expression neural activity associated with those cognitive processes .
We tested 152 subjects between the ages of 8 and 33 . Subjects were initially recruited between the ages of 8 and 30 years and were scanned approximately annually for 1–10 years . Subjects were included based on two criteria: ( 1 ) Mean frame-wise displacement ( FD ) was less than 0 . 15 mm; and ( 2 ) at least 50% of the trials from each of the four trial types had to be measurably correct . Here , incorrect trials are those for which measurements of reaction time and endpoints for both visually- and memory-guided saccades were unavailable due to blink artifacts , noisy data , or transient loss of pupil- or corneal reflection-lock . After applying these exclusion parameters our dataset consisted of 126 subjects ( 60 female ) . We applied no further outlier control for our analyses . Participants and/or their legal guardians provided informed consent before participating in this study . Experimental procedures for this study complied with the Code of Ethics of the World Medical Association ( 1964; Declaration of Helsinki ) and the Institutional Review Board at the University of Pittsburgh . Subjects were paid for their participation in the study . Eye-movements were recorded in the scanner with an infrared camera system equipped with long-range optics and sampling at 60 Hz ( Model R-LRO6 , Applied Science Laboratories , Bedford MA ) . Subjects’ compliance with instructions was assessed and eye-movements were monitored via remote video during task performance . We used a nine-point calibration procedure to estimate the transformation from the eye-tracker's native encoding space to on-screen pixel location . Saccadic events were detected using an in-house suite of automated routines . Individual saccade candidate events were detected from local maxima in the eye-movement velocity trace . Saccade start and end times were determined by searching backward and forward in time in the velocity trace to find the sample where velocity dropped below 1/10th of the peak velocity ( Gitelman , 2002 ) . Imaging data were acquired using a Siemens 3-Tesla MAGNETOM Allegra ( Erlangen , Germany ) system with a standard radio-frequency ( RF ) head coil at the Brain Imaging Research Center , University of Pittsburgh , Pittsburgh , PA . Structural images were acquired using a sagittal magnetization prepared rapid gradient echo ( MPRAGE ) T1-weighted pulse sequence with 224 slices with 0 . 7825 mm slice resolution . Functional images were acquired using a gradient echo echo-planar ( EPI ) sequence sensitive to blood-oxygen-dependent ( BOLD ) contrast ( T2* ) ( TR=1 . 5 s , TE=25 ms , flip angle=70° , voxel size=3 . 125×3 . 125×4 mm slice resolution , 229 volumes ) . Twenty-nine slices per volume were collected with no gap and aligned to the anterior and posterior commissure ( AC-PC ) plane . Structural and functional fMRI data are available through the Dryad repository ( Montez et al . , 2017 ) . T1-weighted anatomical images were reconstructed from raw DICOM files and converted to NIFTI format . We estimated the bias field corrections using smoothed and highpass filtered anatomical data analyzed with FSLs fast algorithm . After bias field correction we constructed a skull stripped anatomical data set for the subject , which we used to estimate the 12 degree-of-freedom affine transformations that would align the subjects data with the MNI152 anatomical template . Finally , we computed the non-linear transformation that would bring the subject’s affine-aligned anatomical data set into registration with the MNI152 template . We saved final combined linear/non-linear transformation for later use in registering the subjects’ functional data to the standard space . fMRI data were preprocessed using a combination of AFNI ( Analysis of Functional NeuroImages , RRID:SCR_005927 ) and FSL software ( FSL , RRID:SCR_002823 ) . In our pre-processing pipeline , raw data was converted from DICOM format to NIFTI volumes and slice-timing correction was applied using AFNI tools . We performed motion estimation and correction in two phases . First we pre-aligned each frame of a subject’s functional data to a volume created by taking the temporal mean of the 4-D functional time series . Then , a second , ‘true’ , average functional volume was computed from the pre-aligned functional data , producing a reference functional volume that was less affected by motion artifacts . We then aligned each frame of the original function time series to this second reference volume using sinc-function interpolation and estimating the time course of translational and rotational motion throughout the run . We used these estimated time series throughout our later analyses of the functional data . Next , using FSL’s brain extraction tool , we stripped the skull and superfluous tissues from the subject’s motion corrected mean functional EPI images , afterward aligning the resulting mean EPI volumes to their anatomical MPRAGE volume using a six degree-of-freedom rigid-body transformation estimated using spline interpolation . To align each frame of the motion corrected EPI sequence to the subjects structural image , we applied the translation estimated in the previous step to each frame of the motion corrected functional time series and then removed the skull and extraneous tissues from each frame of the functional time series . Tissue remaining within the mean functional volume after the skull stripping procedure was removed by applying a dilated binary mask to the mean aligned functional volume that removed extreme voxels whose values did not reside in middle 98th percentile . We then removed voxel-wise temporal extrema using AFNI’s 3dDespike software . To align a subjects functional data to a standard MNI152 ( Montreal Neurological Institute; MNI ) template in a single transformation step , we used FSL convertwarp , and applywarp functions to combine the estimated motion correction , functional-to-structural , and linear and non-linear subject-to-MNI152 transformations into a single operator , which we applied separately to each frame of the original slice time-corrected functional data . We performed minimal spatial smoothing on the aligned functional data , using a SUSAN algorithm with a 5 mm FWHM kernel , followed by a conservative high-pass filtering of the voxel-wise time series , which removed or attenuated BOLD signal frequencies below 0 . 0083 Hz ( corresponding to fewer than three cycles per task run ) . Finally , we rescaled all voxel values by a value defined to be 10 , 000 divided by the global median . In our analyses , measuring brain state variability requires determining the spatial structure of canonical whole-brain patterns of BOLD signal associated with distinct task-related processes . The flowchart in Figure 9 depicts an outline of the processing steps used to transform individual preprocessed fMRI time series into the average time courses of brain state expression and brain state variability as well as the relationship between the major processing steps and certain key analyses . From each session's data we extracted eight voxel-wise average time courses of BOLD activity corresponding to each of the four task conditions when stimulus presentations occurred in wither the left or right visual hemifields . We estimated these time courses with a finite impulse response ( FIR ) regression model . FIR design matrices were constructed manually and applied to the voxel-wise time series using 3dDeconvolve ( AFNI ) . All trials , including incorrect responses and blinks , for each stimulus type were modeled over an interval consisting of the duration ( from initial stimulus presentation to the execution of the memory-guided saccade ) plus an additional 22 . 5 s ( 15 TRs ) . The design matrix included nuisance regressors to account for the effects of signal drift , subject motion , and global signal changes as captured by white matter and cerebrospinal fluid ( CSF ) signals and their derivatives . Signal drift for each run was modeled as a third order Legendre polynomial time series . Head motion was computed along six affine components corresponding to translation in the three cardinal directions and rotations about three orthogonal axes . In addition we computed a time course of total displacement for each session based on the Euclidean norm of the time derivative of the movement time series at each time point . To account for the prolonged effect of autocorrelated movement on the BOLD signal , we included temporally leading ( −1TR ) and lagging ( +1–2 TRs ) copies of each of the seven motion regressors ( Friston et al . , 1996 ) . Each of the seven motion time courses therefore contributed four motion regressors to the deconvolution design matrix . After deconvolution , we scaled the resulting whole-brain average trial time courses at each voxel , normalizing them to the standard deviation of the regression residuals at the same voxel location . Idealized voxel-wise trial time courses for the long delay conditions were estimated from the scaled average trial time course estimates . We modeled these separately for each condition and target hemifield using 3dLME ( AFNI ) , a linear mixed-effects framework . Each time point was modeled as a separate categorical fixed effect and we did not include an intercept term in the model . To account for any bias due to the over representation of subjects who participated in more scans , we included subject identity as a random effect component in the regression model . For each trial type we computed the total displacement undergone by each subject's brain during the BOLD signal measurement intervals ( trial durations plus 15 TRs ) and included it as a fixed-effect component in the regression analysis . We calculated subjects' average age for all of their sessions and , after centering by the global mean , included it as a subject level fixed-effects regressor . We included a mean age by time interaction term to capture age-related differences in the voxel-wise time courses . We included the subjects' age at each session , after subject level mean-centering , as a second age-related random-effects regressor . Within a given voxel , a single whole trial time course may include independent contributions from visually- and memory-guided saccade events . To account for potential differences due to variability in the number of correct saccades , we included the proportion of unclassifiable and incorrect visually- and memory-guided saccades and their interactions with time as fixed-effect components of the model . We produced idealized trial time courses by generating the voxel-wise model estimates for a subject of mean age , mean in-scanner displacement , and perfect trial performance . This process generated four idealized whole-brain time series corresponding to both long-delay conditions in which targets were located in either the left or right visual hemifield . We used these idealized BOLD time series in our construction of the canonical brain states Including lagged motion regressors during deconvolution accounts for the prolonged linear effects of in-scanner motion up to 2 TRs after a given movement ( Friston et al . , 1996 ) . To control for any effects of motion that may continue beyond that time and bias measures of brain state variability , we developed a method to account for temporally prolonged linear motion artifacts from fMRI data based on motion template volumes that model the spatial pattern of artifacts associated with the linear effects of motion ( Figure 10a ) . We normalized the regression coefficients associated with each lagged motion regressor at each voxel by the temporal standard deviation of the voxel’s post-deconvolution residuals and computed their resulting mean spatial patterns across all subjects . We used these whole-brain patterns of normalized regression coefficients to construct motion artifact templates . For each of the 28 templates ( 7 motion components with 4 temporal lead/lags ) , we subtracted the spatial mean of all voxel values and scaled the resulting volumes to a common vector magnitude . Then , using principle component decomposition , we found a set of 11 motion templates that captured >90% of the variability in the set , which were then converted back into 3D volumes . We verified that the set of motion templates captured components of linear motion related BOLD signal variability that remained after deconvolution by computing SSmotion , the sum of squared error associated with all motion templates across all TRs in the whole-brain residual BOLD time series . We then computed the ratio SSmotion/ ( SSbrain + SSerror ) where SSbrain is the sum of squared error associated with brain state variability and SSerror is unexplained sum of squared error . We refer to this ratio as motion template variance . We found that estimates of mean FD and motion template variance are highly related ( t ( 337 ) =10 . 3; p=1 . 06e-21 ) , indicating that even with rigorous motion controls during deconvolution ( see Materials and methods ) , significant linear motion artifacts remain ( Figure 10b ) . We found that while brain state variability is not significantly related to mean FD ( t ( 337 ) =1 . 39; p=0 . 17 ) , it is significantly greater in subjects exhibiting greater motion template variability ( t ( 337 ) =2 . 01; p=0 . 045 ) , ( c-d ) . The usage of motion templates as an additional control for motion-related artifacts arose from an abundance of caution and the need for high dimensional nuisance regressors that could be fit simultaneously to each volume alongside the set of brain state patterns . While these preliminary analyses suggest promise for the approach , we acknowledge that it has yet to be fully validated . Although the relationship between brain state variability and movement was small and inconsistent across different estimates of motion , we sought a more rigorous control by comparing brain state variability in a group of high motion adults to low motion children and adolescents ( see Results ) . To remove trivial components of whole-brain BOLD signal variability associated with TR-to-TR shifts in the global mean across the entire imaged volume , we included a constant offset template , which consisted of a whole-brain binary mask . To account for TR-to-TR shifts in mean BOLD signal that are limited to the gray matter ( the sort of variability which in some resting state studies is removed via global gray matter signal regression ) we included a second binary template defined by the brain state mask ( see Materials and methods Estimating canonical brain states ) . In addition , we constructed a set of spatial gradient templates to account for other trivial modes of whole-brain BOLD signal variability consisting of simple linear spatial gradient patterns . We first created 3 spatial gradient volumes whose voxel values were equal to their x , y , and z coordinates relative to the volume's center of mass . We set each voxel that fell outside of a whole-brain MNI mask to zero . We then computed a set of 3 ‘interaction’ templates that corresponded to each pair-wise product of the spatial gradient templates . As a whole , this set of 19 templates constituted the set of spatial nuisance regressors that we used to capture and remove remaining unwanted spatial modes of whole-brain BOLD signal variability associated with motion and trivial variability in global signal . We included the motion templates and spatial constant and gradient templates as additional nuisance regressors in all analyses that involved projecting whole brain volumes of BOLD signal onto the canonical set of brain state components . We derived the canonical brain states from whole-brain patterns of BOLD activity extracted from different time points within the idealized trial time courses . To constrain the brain state patterns to swathes of gray matter that were reliably imaged across sessions , we constructed a mask defined by the intersection of a probabilistic MNI gray-matter mask ( threshold≥0 . 5 ) and a mask constructed from the group average of the model R-squared maps from each session's initial deconvolution ( threshold≥0 . 27 , selected to removed unreliably imaged brain regions ) . Combined these masks served to eliminate white matter; CSF; the hind-brain; cerebellum , the inferior portion of which was not consistently imaged across sessions; artifact-prone basal forebrain; and infero-temporal lobes ) . The extent of the resulting brain state mask can be observed in ( Figure 2b ) . Each canonical brain state consists of a mean and spatial component . The mean component represents the average pattern of whole-brain BOLD activity evoked by a specific task epoch , regardless of target location . The spatial component represents the difference between the patterns of whole-brain BOLD activity evoked by specific task epochs for right side versus left side targets . To reduce the extent to which BOLD signal resulting from a trial's initial visually-guided saccade contaminated our estimate of maintenance- and retrieval-related brain states , we maximized the temporal distance between the task epochs by using time courses from long delay trials only . The canonical mean and spatial VME brain states were derived from the whole-brain pattern of activity occurring within the brain state mask around the time of the visually guided saccade . To account for hemodynamic lag , we extracted four volumes of BOLD activity ( corresponding to the left and right side targets from both long delay conditions ) from the idealized long delay time courses six seconds ( 4 TRs ) after the visually guided saccade was performed to ensure that most regions would have achieved their peak BOLD response ( Handwerker et al . , 2004 ) . We constructed the mean VME brain state by computing the voxel-wise average of normalized BOLD activity across all four volumes . To construct the spatial VME state we separately computed the voxel-wise average of the VME volumes for right and left side targets . The spatial state was defined as the difference –right minus left– of the resulting volumes . Canonical maintenance-related brain states were based on the patterns of normalized BOLD signal that occurred during the TR immediately prior to a subjects’ execution of the memory-guided saccade . Since brain states were estimated from the long delay trial types only , the maintenance brain state patterns were naturally separated from the BOLD response evoked by the initial stimulus presentation and the subjects' subsequent visually guided saccade . To completely remove any remaining stimulus and visuomotor contributions , as well as global average signal , we extracted all volumes from the idealized whole-brain trial time courses that occurred before or included a trial's initial visually guided saccade and regressed these patterns from the maintenance state patterns . States were then constructed for maintenance intervals as described for the VME states above . Construction of the retrieval-related mean and spatial brain states followed a similar course as that used in the construction of the VME and maintenance-related brain states . Like the VME states , retrieval-related states were based on the normalized BOLD responses occurring during the 4th TR after the execution of the memory-guided saccade . We removed all components of VME and maintenance activity by regressing out every pattern of activity within the brain state mask that occurred before the TR at which the subject began a memory-guided saccade . The resulting mean and spatial brain states , as expected , exhibited a high degree of mirror symmetry across the midline plane . However , we observed noise and artifacts were introduced into the brain state patterns by accumulating error in our regression-based orthogonalization procedure . To correct for these artifacts , we leveraged symmetry of the brain state patterns by averaging them with corresponding brain state patterns derived from a left/right mirrored version of the idealized BOLD time series . To do this , we constructed left-right mirrored versions of the idealized trial time courses , performed an identical set of operations to define each brain state and combined them with the original set of brain state patterns . We first applied a mirror matrix to the idealized trial time course volumes . We aligned the mirrored volumes to the standard MNI template using a 12-degrees of freedom affine transformation . We then applied a non-linear transformation that warped the mirrored and aligned volumes to the standard MNI template . The final mean brain state components were constructed by computing the voxel-wise average of the mirrored and non-mirrored mean brain states . We inverted the sign of all voxel values of the mirrored spatial volumes before averaging them with the non-mirrored spatial volumes to produce the final spatial brain state components . In practice the mirroring procedure had only a minor effect on our results , visibly removing noise and improving the relationship that we observed between trial-to-trial behavioral performance and brain state expression ( Figure 4 ) . We converted the average whole-brain trial time series and whole-brain residual time series into average time courses of brain state expression and variability respectively . For each TR , we extracted the whole-brain pattern of activity that we then vectorized and modeled using a linear regression . Our design matrix consisted of vectorized versions of the six brain states ( the mean and spatial components of the VME , maintenance and retrieval states ) as well as the 19 nuisance regressors templates described above . For each TR we extracted the regression weights for the six brain states , motion , and nuisance components and ordered them into a time series . When this procedure is performed on the whole-brain average trial time series , the result is a time course of expression of each of the brain states during a trial . When performed on the whole-brain residual time series , the result is a time course of brain state fluctuations , where positive values indicate that a particular brain state was present to a greater extent than average and negative values indicate that a state was expressed less than average . For each session , temporally z-scored the time series of variability for each brain state component . For each session we separately transformed reaction time and saccadic error from each of the four main task conditions into z-scores . SE was rectified such that high SE values reflect greater error in memory-guided saccadic endpoints on a trial . We excluded all trials for which measurements of reaction time and endpoints for both visually- and memory-guided saccades were unavailable due to blink artifacts , noisy data , or transient loss of pupil- or corneal reflection-lock . We related trial-to-trial variability in reaction time and accuracy to variability in the expression of each brain state across a range of times ( ±15 TRs ) relative to the TRs that contained the memory-guided saccades from each trial ( Figure 4 ) . Using all correct trials across all sessions , we extracted our z-scored measurements of brain state fluctuation derived from the whole-brain residual time series . We then constructed a regression model that included terms for the measured values of each brain state at the relative TR . We also included terms for the spatial brain state interaction with target hemifield . Each model contained terms that varied across trials but did not vary across relative TRs . These included terms for run number ( coded as 1–3 ) , target hemifield ( coded as −1 , 1 ) , target location ( eccentricity , coded from least to most eccentric as 1–3 ) , and the square of target location term . Because it is possible that RT and SE are correlated on a trial-to-trial basis , a true relationship between brain state variability and RT may result in a trivial relationship between brain state variability and SE , or vice versa . To account for this possibility , in the RT regression models , we included a term for trial-to-trial SE and its square . Similarly , for the SE regression models we included a term for RT and its square . This set of regressor terms served as a null model against which the full brain state model was compared . The trial-wise reaction time and accuracy models were fit using a linear mixed-effects framework ( MATLAB ) to account for the different numbers repeated measurements for many of the subjects . Subject identity was modeled as a random effect . We used the difference between the ordinary R2 values for full and null models at each relative TR to assess the amount of unique behavioral variability accounted for by trial-to-trial fluctuations in the expression of different brain states . At each relative TR we compared the null and full modes using simulated maximum-likelihood estimation procedure with 5000 iterations ( MATLAB ) . To perform these simulations , we compiled a distribution of reaction times for all correct memory-guided saccades across our subject database . We then simulated 400 trials with reaction times selected to produce a distribution that was identical to the compiled empirical distribution . To simulate the simple timing-based effect of BOLD signal variability we generated an impulse function for each simulated trial , a vector where all but one element is equal to zero , where each consecutive element refers to 60 ms time bin after the extinction of the central fixation cross ( the signal to perform a memory-guided saccade ) . For each draw from the reaction time distribution we generated new impulse function vector by inserting a 1 into the vector at the index corresponding to the reaction time on that trial . We convolved each of the 400 impulse functions with a canonical HRF modeled at the same 60 ms resolution . This produced a set of HRF time series whose time of peak amplitude varied with reaction time Amplitude-based simulations were performed similarly but with two key differences: ( 1 ) for each trial we inserted 1 into all impulse function vectors at the same time index , corresponding to mean reaction time , for all trials; and ( 2 ) we added or subtracted from the 1 a linearly interpolated value between ±0 . 25 where+0 . 25 corresponded to the fastest reaction time and −0 . 25 corresponded to the slowest reaction time . Mixed amplitude and timing based simulations were a hybrid of the two described above . The index of the 1 for each trial's impulse function was selected to coincide with the reaction time on that trial . An additional amplitude modulation factor , as above , was added to the impulse index . Separately for timing- , amplitude- , and timing and amplitude-based simulations , we computed the mean HRF time series across all trials and simulated residual time series by subtracting the mean HRF time series from the individual trial time series . Next we divided the simulated residuals in to fast and slow RT sets , defined by median split and calculated their average . Lastly , we computed the time integral of the mean residual time series for fast and slow trials . To compare the simulated pattern of high temporal resolution residuals to the actual data we extracted equivalent snippets ( beginning at the TR containing the MGS ) of the mean VME brain state fluctuation time course for all trials and all subjects . We selected the mean VME state because of its close relationship with the visuomotor processes which makes it most likely to reflect a trivial timing based relationship between RT and brain state expression . As in the simulated data , for each subject we divided the snippets in to fast and slow RT trials based on a median split and then calculated the group average time series . Finally , we interpolated the resulting time series to a matched temporal resolution using shape preserving piece-wise cubic interpolation ( MATLAB ) . | Adolescence is a period of change: physically , socially and intellectually . During the teenage years , the brain undergoes changes in structure and connectivity that lead to improvements in areas such as self-control , social skills and cognition . Adolescence is also a time during which cognitive skills , such as problem solving and memory , become more stable . Whereas a child will perform a task markedly better on some days or trials than others , adolescents become increasingly consistent . But why does cognitive performance fluctuate at all ? Studies in monkeys suggest that momentary fluctuations in processes like attention and alertness are linked to changes in the level of activity within brain regions that are active during a task . Areas of the brain that are relatively active tend to become even more active , whereas those that are relatively inactive reduce their activity even further . These changes lead to variable accuracy and reaction times . Montez et al . hypothesized that as adolescents become better at controlling processes such as attention and alertness , they show fewer and/or smaller fluctuations in brain-wide activity during a task . This in turn leads to more stable performance . To test this idea , Montez et al . asked healthy volunteers aged 8 years and above to perform a memory task while lying inside a brain scanner . Over the next 10 years , the volunteers returned about once a year to perform the task again , thereby revealing how their brain activity changed as they grew older . Over the course of adolescence , the volunteers performed the task increasingly accurately and consistently . As predicted , their overall level of brain activity during the task also became less variable over the same period . These findings challenge the current view of adolescent development , which assumes that teenagers acquire new cognitive skills with age . The results of Montez et al . suggest instead that improvements in cognitive performance reflect teenagers’ increasing ability to stably engage skills that they have possessed since childhood . This difference has implications for education , healthcare , parenting , and even for the juvenile justice system . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | The expression of established cognitive brain states stabilizes with working memory development |
Uveitis describes a heterogeneous group of inflammatory eye diseases characterized by infiltration of leukocytes into the uveal tissues . Uveitis associated with the HLA haplotype B27 ( HLA-B27 ) is a common subtype of uveitis and a prototypical ocular immune-mediated disease . Local immune mechanisms driving human uveitis are poorly characterized mainly due to the limited available biomaterial and subsequent technical limitations . Here , we provide the first high-resolution characterization of intraocular leukocytes in HLA-B27-positive ( n = 4 ) and -negative ( n = 2 ) anterior uveitis and an infectious endophthalmitis control ( n = 1 ) by combining single-cell RNA-sequencing with flow cytometry and protein analysis . Ocular cell infiltrates consisted primarily of lymphocytes in both subtypes of uveitis and of myeloid cells in infectious endophthalmitis . HLA-B27-positive uveitis exclusively featured a plasmacytoid and classical dendritic cell ( cDC ) infiltrate . Moreover , cDCs were central in predicted local cell-cell communication . This suggests a unique pattern of ocular leukocyte infiltration in HLA-B27-positive uveitis with relevance to DCs .
Uveitis describes a heterogeneous group of inflammatory diseases involving uveal tissues in the intraocular cavity of the eye . Non-infectious uveitis is regarded as an autoimmune disorder and is associated with various immune-mediated systemic diseases . According to the Standardization of Uveitis Nomenclature ( SUN ) working group , uveitis is classified based on the anatomical location of uveitis as anterior , posterior , intermediate , and panuveitis ( Jabs et al . , 2005 ) . Anterior uveitis ( AU ) primarily affects the iris and ciliary body and constitutes its most frequent type ( approximately 80% of cases ) resulting in permanent vision loss through , for example , secondary cataract , glaucoma , or macular edema ( Rothova et al . , 1996; Thorne et al . , 2016 ) . Acute anterior uveitis ( AAU ) associated with the HLA haplotype B27 ( B27+ AAU ) is the most common and often severe form of uveitis . Its typical clinical features include acute onset of discomfort , eye redness , tearing , visual impairment , and excessive cellular infiltration in the aqueous humor ( AqH; ie , the intraocular liquid of the anterior chamber ) that is devoid of immune cells under non-diseased conditions . The prevalence of the HLA-B27 haplotype is approximately 8–10% among Caucasians and 40–70% among AAU patients ( Kopplin et al . , 2016 ) and thus represents a strong genetic risk factor for AAU ( Brewerton et al . , 1973; Huhtinen and Karma , 2000 ) . The HLA-B27 allele also conveys an increased genetic risk for other immune-mediated diseases , including spondyloarthropathies ( SpA ) and inflammatory bowel disease ( IBD ) . B27+ AAU is thus regarded as a prototypic ocular immune-mediated disease and , accordingly , its treatment is based on corticosteroids ( Khan et al . , 2015; Rosenbaum , 2015 ) and classical and biological disease-modifying anti-rheumatic drugs ( Heiligenhaus et al . , 2012; Bou et al . , 2015 ) . Notably , although T-cell inhibition ( eg , by cyclosporine A ) is beneficial in other uveitis entities ( Nussenblatt et al . , 1991 ) , it provides limited efficacy in B27+ AAU ( Gómez-Gómez et al . , 2017 ) , suggesting a yet poorly defined role of innate immune cells in its pathogenesis . Immune cells infiltrating the anterior chamber of the eye can be observed clinically , but are difficult to obtain for further analyses . In fact , the invasive sampling of AqH is rarely clinically justified . AqH fine-needle aspirates are primarily applied for the verification of , for example , infectious uveitis ( Chronopoulos et al . , 2016 ) and rarely to unravel the pathogenesis of non-infectious uveitis ( Greiner and Amer , 2008; Chen et al . , 2015; de Groot-Mijnes et al . , 2015; Zhao et al . , 2015 ) . Our knowledge of AqH-infiltrating leukocytes and underlying mechanisms of uveitis thus remains superficial , despite its frequency and severity . In AAU , a role of bacterial triggers has been proposed ( Huhtinen et al . , 2002a; Huhtinen et al . , 2002b ) . Therefore , innate immune cells such as monocytes and dendritic cells ( DCs ) that phagocytose and process foreign antigens are of special interest . In fact , phenotyping of circulating monocytes in patients with immune-mediated uveitis emphasized differences during the disease course ( Liu et al . , 2015; Walscheid et al . , 2016; Walscheid et al . , 2018; Kasper et al . , 2018 ) . The microbiome , intestinal barrier dysfunction , and immune response have also been suggested to contribute to the pathogenesis of AAU ( Rosenbaum and Asquith , 2018 ) . Soluble mediators ( Bauer et al . , 2020; Abu El-Asrar et al . , 2020; Bauer et al . , 2018; Bonacini et al . , 2020 ) and infiltrating leukocytes have been analyzed ( Denniston et al . , 2012 ) previously in the AqH of uveitis patients , but access to samples and technical challenges remain the main bottlenecks toward better understanding the pathomechanisms in uveitis . Single-cell RNA-sequencing ( scRNA-seq ) studies have provided unprecedented high-resolution insights into immune mechanisms in various tissues ( Heming et al . , 2021; Wolbert et al . , 2020 ) . Therefore , we here combined scRNA-seq with flow cytometry and protein analysis of soluble cytokines and thereby provide the first partially unbiased characterization of leukocytes in the AqH from patients with B27+ AAU compared to patients with HLA-B27-negative AU ( B27-AU ) and acute infectious endophthalmitis . We found that lymphocytes predominate in the intraocular infiltrate in B27+ AAU and specifically showed an elevated frequency of plasmacytoid DCs ( pDCs ) and classical DCs ( cDCs ) . In B27+ AAU , DCs featured the most predicted intercellular interactions and increased expression of AAU- and SPA-related genome-wide association study ( GWAS ) risk genes that distinguished this uveitis from active B27-AU . This suggests a specific involvement of DCs in B27+ AAU , and subtypes of AU thus exhibit specific patterns of local leukocyte responses .
We here aimed for an unbiased characterization of leukocytes in the AqH in AU flares . We screened 4980 total patients with any intraocular inflammation seen at our uveitis center . We included 11 patients with current onset of uveitis flare ( n = 8 ) or endophthalmitis ( n = 3 ) into this study , being untreated with topical corticosteroids for this flare ( Supplementary file 1a ) , corresponding to a recruitment rate of approximately 0 . 18% . In 10 of the AqH samples , cytokine analysis was performed . Cellular infiltrates were analyzed via scRNA-Seq in AqH samples of four patients with active B27+ AAU , two patients with active B27-AU , and one patient with active bacterial endophthalmitis ( Streptococcus pneumoniae ) ; five of those samples were analyzed in parallel via flow cytometry ( Supplementary file 1a ) . Deep characterization of AqH-infiltrating leukocytes is thus feasible . The endophthalmitis patients were significantly older than B27+ AAU patients ( analysis of variance ( ANOVA ) , p = 0 . 0157; Supplementary file 1a ) . There were four females in the B27-AU cohort , and three males and one female in the B27+ AAU cohort . All B27+ AAU patients had associated SpA , and two of them received adalimumab therapy . Both uveitis groups did not differ significantly regarding age ( ANOVA , p = 0 . 566 ) , Antinuclear antibodies ( ANA ) status , frequency of systemic anti-inflammatory treatment , topical medication , previous ocular surgery , or time since uveitis onset . We then performed scRNA-seq of ocular infiltrating cells from fresh AqH fine-needle aspirates ( Supplementary file 1a ) . We defined 5000 input cells as our maximum intended input per sample and used excess cells beyond that number for flow cytometry analysis . Thereby , we obtained transcriptional information of 13 , 550 total individual cells and 1936 average cells ( ± 1411 SD ) per patient with 830 average genes ( ± 402 SD ) detected per cell ( Supplementary file 1b ) . This is in accordance with the expected cell recovery rate of 50% of the scRNA-seq technique we employed ( Zheng et al . , 2017 ) . After quality control , we clustered the single-cell ( sc ) transcriptomes of all patients combined and identified 13 individual cell clusters ( Figure 1A and B ) . We manually annotated these clusters based on the expression of marker genes ( Figure 1C , Figure 1—figure supplement 1 , Supplementary file 1c ) . As previously shown in AqH aspirates of human uveitis patients ( Denniston et al . , 2012 ) , only hematopoietic cell clusters were identified . Clusters were broadly classified into cells of myeloid ( 40%; cDCa , cDCb , pDC , mature DC ( matDC ) , granulo , myeloid ) and of lymphoid origin ( 60%; natural killer ( NK ) , γδ T cells ( gdTC ) , CD8 , regulatory T ( Treg ) , CD4 , naive B cells ( Bc ) , plasma ) . Myeloid clusters separated into cDC clusters tentatively named cDCa ( ITGAX , CLEC7A ) and cDCb ( CLEC10A , MRC1 ) , pDC ( IL3RA/CD123 , CLEC4C/CD303 ) and matDC ( TMEM176B , IDO1 , FSCN1 , LAMP3 , CD83 ) , granulocytes ( granulo; S100A12/ S100 A8 high , CCL2 low ) , and myeloid cells with unclear assignment ( myeloid; S100A12/S100 A8 low , CCL2 high ) ( Figure 1C ) . The marker genes expressed by the cDC sub-clusters did not fully overlap with previously described cDC type 1/2 markers used to identify DC subsets from cerebrospinal fluid ( DC1: CLEC9A , XCR1 , BATF3; DC2: CD1C , FCER1A , CLEC10A ) ( Heming et al . , 2021 ) or peripheral blood ( DC1: CLEC9A , C1ORF54 , HLA-DPA1 , CADM1 , CAMK2D; DC2: CD1C , FCER1A , CLEC10A , ADAM8 ) ( Villani et al . , 2017 ) . We therefore intentionally named these clusters cDCa/b to prevent ambiguity . Lymphoid clusters were classified as CD4+ T cells ( CD4; IL7R , CD3G ) , CD8+ T cells ( CD8; CD8A , CD3G ) , gdTC ( NKG7 , TRDC ) , Treg cells ( IL2RA , FOXP3 ) , NK cells ( GNLY , NKG7 ) , naive Bc ( MS4A1 , CD19 , IGHD ) , and plasma cells ( plasma; IGHG1 , CD38 , SDC1/CD138 ) ( Figure 1C ) . We thus identified all major leukocyte lineages in AqH fine-needle aspirates ( Figure 1B , Supplementary file 1d ) . We next sought to understand how intraocular inflammation differed between uveitis entities . First , we assessed the endophthalmitis control patient and found almost exclusively myeloid lineage clusters with predominating granulocytes in accordance with an acute anti-bacterial response ( Figure 2B; Figure 2—figure supplement 1; Engstrom et al . , 1991 ) . In contrast , the cellular infiltrates in B27-AU and B27+ AAU patients’ AqH were of more lymphoid origin , reflecting their autoimmune etiology ( Figure 2A and B ) . When systematically comparing B27+ AAU vs B27-AU samples , inter-patient variability was high ( Figure 2A and B ) , but several populations still substantially differed between uveitis subtypes ( Figure 2C , Supplementary file 1d ) . While the CD8 and naive Bc clusters were reduced , the pDC and cDCa clusters were more abundant in B27+ AAU patients ( Figure 2C and D ) , overall indicating an influx or expansion of DC in this uveitis entity . Next , we sought to confirm our findings using flow cytometry ( Figure 2E; Figure 2—figure supplement 2 ) . The expression of canonical lineage markers differs between mRNA and protein quantification ( Peterson et al . , 2017 ) , and we therefore used widely applicable pan-lineages in flow cytometry . AqH-derived cells of some of the patients ( Supplementary file 1a ) were analyzed to distinguish granulocytes ( CD3-CD11b+HLA-DR- ) , monocytes/macrophages/DCs ( CD3-CD11b+HLA-DR+ ) , CD4+ and CD8+ T cells ( CD3+CD11b- ) , and NK cells ( CD3-CD11b-CD56+ ) . This confirmed the mainly myeloid infiltrate in the endophthalmitis patient ( Figure 2E ) . B27-AU patients showed an infiltrate dominated by T cells , low NK cells , and low cells of myeloid origin . In B27+ AAU patients , myeloid cells were more abundant than in B27-AU ( Figure 2E ) . Also , the abundance of broad cell classes quantified by scRNA-seq and flow cytometry showed a positive correlation ( Figure 2E , Figure 2—figure supplement 2B ) . The higher frequency of granulocytes detected in flow cytometry than in the scRNA-seq might be due to the higher fragility of these cells during processing for scRNA-seq ( Zilionis et al . , 2019 ) . Overall , subtypes of uveitis were thus characterized by a unique pattern of local inflammatory cells . Next , we sought to understand how local leukocytes differ transcriptionally between uveitis entities . To analytically account for low total cell numbers , we merged clusters into five broad cell type of ‘meta-clusters’ for differential expression analysis ( Figure 3A , Supplementary file 1e ) : helper T cells ( help; Treg , CD4 ) , cytotoxic cells ( toxic; CD8 , NK , gdTC ) , merged DCs ( mergeDC; matDC , pDC , cDCa , cDCb ) , other myeloid cells ( myeloidLin; myeloid , granulo ) , and B-cell lineage ( BcLin: naive Bc , plasma ) . We then tested for differentially expressed ( DE ) genes between B27+ AAU and B27-AU . Across all clusters , multiple major histocompatibility complex ( MHC ) class I and class II related genes ( HLA-A , HLA-DPA1 , HLA-DRA , HLA-DRB1 , B2M ) were expressed at lower levels in B27+ AAU samples ( Figure 3B–F ) . Furthermore , in B27+ AAU , the help and BcLin meta-clusters downregulated one cytokine receptor ( CXCR4 ) . The toxic meta-cluster downregulated signs of cytotoxicity ( GZMK , GZMH , LTA ) . The BcLin meta-cluster featured an increase of macrophage-inhibitory factor ( MIF ) , known to inhibit NK cell activity ( Apte et al . , 1998 ) . The myeloid meta-cluster reduced expression of several interferon ( IFN ) -regulated genes ( IFITM genes; Figure 3B–F ) . Elevated CTNN1 expression in mergeDC ( Figure 3B ) pointed to an involvement of the WNT/catenin pathway as previously described for SpA ( Xie et al . , 2016 ) . Increased expression of Lyz and its cognate antisense AC020656 . 1 , and CALR ( Liu et al . , 2016 ) and its antisense AC092069 . 1 , within the mergeDC cluster suggested an activation of myeloid cells and simultaneously local counter regulating mechanisms . Elevated expression of B4GALT1 in mergeDC , in detail cDCa and cDCb ( Supplementary file 1f ) , potentially reflected the migratory capacity of DC ( Johnson and Shur , 1999 ) . Notably , the mergeDC , help , and toxic meta-clusters all upregulated C1ORF56 , an oncogene previously found induced in activated lymphocytes in IBD ( Uniken Venema et al . , 2019 ) and splenic NK cells ( Crinier et al . , 2018 ) . We also found elevated expression of prefoldin 5 ( PFDN5 ) in myeloid , toxic , and help clusters ( Supplementary file 1e ) , which was recently described as a specific marker for B27+ AAU associated with SpA ( Kwon et al . , 2019 ) . Furthermore , as previously shown in joint biopsies of SpA patients ( Carlberg et al . , 2019; Lam et al . , 2019 ) and in association with autoinflammatory diseases ( Carlberg et al . , 2019; Lam et al . , 2019 ) , elevated expression of CDC42 was detected in mergeDC , myeloid , and help clusters and of CDC42SE1 in mergeDC , myeloid , help , and toxic clusters ( Figure 3B–E ) . The risk for developing HLA-B27-associated autoimmune diseases is partially determined by variants in non-HLA genetic loci ( Robinson et al . , 2015; Lin et al . , 2011; Australo-Anglo-American Spondyloarthritis Consortium ( TASC ) et al . , 2010; Li et al . , 2019; Evans et al . , 2011; International Genetics of Ankylosing Spondylitis Consortium ( IGAS ) et al . , 2013; Trochet et al . , 2019; Ellinghaus et al . , 2016; Huang et al . , 2020 ) . We therefore interlinked our transcriptional data with existing genetic information , by testing which ‘meta-clusters’ differentially expressed AAU/SpA risk genes between B27+ AAU and B27-AU ( Figure 3G , Supplementary file 1g ) . Both uveitis-related and SpA-related risk genes were included in the analysis because all B27+ AAU patients had systemic SpA ( Supplementary file 1 ) . Many SpA-related risk genes were highly expressed in the BcLin cluster ( eg , CLEC16A , IFNLR1 ) , and in the mergeDC cluster ( eg , IFNGR2 , AHR ) in B27+ AAU ( Figure 3G , Supplementary file 1g ) . Compared to B27-AU , expression of genes involved in the detection of pathogens ( eg , CARD9 , TLR4 ) showed lower expression in the mergeDC cluster in B27+ AAU . Notably , most AU-related risk genes were preferentially expressed in the B27+ AAU mergeDC cluster ( eg , EYS , HLA-DRB5 , ERAP1 , and IL18R1 ) . This indicates the relevance of DC in B27+ AAU . We also attempted to understand how uveitis controlled the local inter-cellular signaling circuitry . We therefore used a computational tool ( CellPhoneDB; Efremova et al . , 2020 ) to predict cell-cell interactions between human leukocytes in uveitis from scRNA-seq data . The pDC and plasma clusters had to be excluded because of their small size in B27-AU . In the resulting analysis , the granulo and myeloid clusters had the highest number of predicted interactions ( Figure 4—figure supplement 1 ) . In both uveitis groups , most interactions were between myeloid , granulo , gdTC , and multiple DC clusters . Myeloid lineage clusters thus express the highest capacity for cell-cell interaction . Next , we tested for differences of predicted interactions between both subtypes of uveitis . We calculated the number of predicted interactions in B27+ AAU minus that in B27-AU ( Figure 4A ) . B27-AU displayed many interactions between T/NK clusters with myeloid , matDC , and cDCb clusters ( Figure 4A , Figure 4—figure supplement 1A ) . In contrast , the cDCa cluster showed widespread interactions in B27+ AAU ( Figure 4A , Figure 4—figure supplement 1B . ) . This suggests that preferential interactions differ between uveitis entities and that the cDCa cluster might be involved in intraocular inflammation in B27+ AAU . Notably , when focusing on individual interactions of the cDCa cluster ( Figure 4B ) , MHC class II-related transcripts ( eg , CD74 , HLA-E ) were predicted to interact with surface molecules such as MIF or KLRC1 with immunomodulatory capacity and known association to autoimmunity ( Figueiredo et al . , 2018; Borrego et al . , 1998 ) . The predicted interaction of cDCa with pDC clusters in B27+ AAU samples also included the LAIR-LILRB4 and AXL-GAS6 interaction pairs ( Figure 4B ) that exert immunosuppressive phagocytosing functions ( Scutera et al . , 2009; Brown et al . , 2009 ) . Overall , our findings indicate subtype-specific local inter-cellular leukocyte signaling and indicate that DC forms a central signaling node in uveitis . Next , we sought to characterize the intraocular immune response through the analysis of soluble mediators . We therefore quantified a predefined set of cytokines in the AqH of these patient groups ( Supplementary file 1h ) . Several cytokines were below detection limits , and the most notable feature of all these samples was a pronounced presence of interleukin ( IL ) -6 and IL-1 receptor antagonist ( IL-1RA ) . High levels of immunosuppressive IL1-RA in AqH of uveitis/endophthalmitis patients have also been shown in B27+ AAU patients ( Zhao et al . , 2015; de Vos et al . , 1994; Planck et al . , 2012 ) and may reflect the immune-privileged microenvironment in the anterior chamber ( Zhao et al . , 2015; de Vos et al . , 1994; Planck et al . , 2012; Dana et al . , 1998 ) . The cytokine milieu in endophthalmitis was characterized by innate-related IL-6 , tumor necrosis factor ( TNF ) -α , and IL-1β ( Supplementary file 1h ) , which are involved in ocular barrier breakdown and leukocyte recruitment into ocular tissue ( Hao et al . , 2016; Feys et al . , 1994; Giese et al . , 1998 ) . Across uveitis entities , the interindividual heterogeneity of cytokine patterns was high ( Figure 4C and D ) . When comparing uveitis entities , the B27+ AAU group showed significantly increased levels of IL-2 , IL-6 , IL-18 , IL-22 , IL-1β , and interferon ( IFN ) -γ ( Figure 4C and D ) . When comparing this with the scRNA-seq dataset , cytokine expression was identified across all ‘meta-clusters’ ( Figure 4—figure supplement 2A ) , but not all cytokines were detected ( eg , IL31 , IL5 , IL9 , IL21 , and IFNA1 ) . Single-cell transcriptomics is generally prone to false negatives ( Vieth et al . , 2019 ) , and stromal cells and resident immune cells in the iris can also express proinflammatory cytokines ( Miyamoto et al . , 1999 ) , which may explain the discrepancy between techniques . We additionally performed analysis of B27+ AAU and B27-AU serum samples , recruited during the active uveitis disease stage in the context of another project ( Kasper et al . , 2018; Kasper , 2020 ) . There , we found significantly elevated IL-1RA and IFN-γ levels in the serum of B27+ AAU as compared to B27-AU patients ( Figure 4—figure supplement 2B , C; Supplementary file 1i ) . This indicates a potential role of IFN-γ in both intraocular and systemic immune response during uveitis activity in patients with SpA .
The main limitation of the study is its small sample size combined with a high inter-patient variability influenced by multiple potential confounders ( eg , age , sex , disease duration , medication , comorbidities ) . However , considering the invasiveness of the procedure , the rarity of the disease , and the high costs of sc transcriptomics , our study is relatively sizable . Since both flow cytometry and scRNA-seq were performed with fresh material and since AqH is precious material with limited cell numbers , we were not able to verify several of the hall markers identified with the unbiased scRNA-seq approach . Additional studies that verify the identified genes in a larger cohort will be necessary . Furthermore , the study lacks matched blood data , and we observed an inter-assay variability in some cell populations ( eg , granulocytes ) between scRNA-seq and flow cytometry . In future studies , multiplex staining with antibodies conjugated to a feature barcode oligonucleotide could improve phenotyping of cells , and an unbiased low-input proteomics analysis will be beneficial to link protein and gene-expression analyses . Despite these limitations , our study demonstrates the proof-of-concept feasibility of scRNA-seq of inflammation in AqH , and provides a snapshot of the differences in cellular infiltrate of different uveitis entities and insights into local immune-mediated mechanisms .
Inclusion criteria ( all must be fulfilled ) were as follows: patients with clinically non-granulomatous AU with an anterior chamber ( AC ) cell grade >2 + according to SUN guidelines or with infectious endophthalmitis ( Jabs et al . , 2005 ) . Even though patients received topical corticosteroids in previous recurrences , at the time of sampling , patients received no topical anti-inflammatory medication . Patients included in the study were classified into the following groups ( Supplementary file 1a ) . ( 1 ) Patients with HLA-B27-associated AAU ( B27+ AAU ) and with typical clinical signs of AAU . ( 2 ) Patients with HLA-B27-negative AU ( B27-AU ) , with typical clinical signs but without inflammatory/immune-mediated systemic disease associated with uveitis . ( 3 ) Patients with infectious endophthalmitis . Patients with ages , gender , and systemic medical therapies typical for these three entities were chosen . The following standard laboratory parameters were tested in all patients: differential blood count , liver and kidney function tests , C-reactive protein , angiotensin-converting enzyme , soluble interleukin 2-receptor , and serological testing for Treponema pallidum . The patient was excluded from the study if any of those were remarkable . Patients were analyzed for their HLA-B27 status using established olymerase chain reaction ( PCR ) ( licensed lab standard ) . In addition , patients underwent chest x-ray and consultation with a specialist for internal medicine or rheumatology to identify any associated systemic immune-mediated disease . Patients were classified as having HLA-B27-associated uveitis ( eventually with associated systemic disease ) if none of the tests except HLA-B27 positivity produced any further findings indicating non-related associated systemic disease . Patients with a clinical appearance of infectious ( eg , herpes simplex virus ( HSV ) - or varicella-zoster virus ( VZV ) -induced ) uveitis or uveitis syndromes ( eg , Fuchs uveitis syndrome ) were not included in the study . A standardized ophthalmic database was applied for the analysis that included the following parameters: clinical ophthalmic observations on uveitis in the involved eyes were documented according to the SUN criteria ( Jabs et al . , 2005 ) . Briefly , best-corrected visual acuity testing ( in LogMAR ) , slit-lamp examination , Goldmann tonometry , and funduscopy were performed by two independent observers . Any uveitis-related intraocular complications were recorded ( Supplementary file 1a ) . AqH ( 100–150 µl ) was collected from each study subject using a 30 G needle under local anesthesia and immediately shipped at 4°C to the department of neurology at the University Clinic Muenster ( Germany ) for scRNA-seq and/or flow cytometry analysis . Freshly isolated cells were centrifuged once , counted , and up to 5000 of the input cells were used for scRNA-seq , and the remaining cells were used for flow cytometry . For protein ( luminex ) analysis , 60 µl of the AqH was centrifuged for 5 min at 12 , 000×g and stored at –80°C until analysis . Freshly isolated sc suspensions were loaded onto the Chromium Single Cell Controller using the Chromium Single Cell 3' Library & Gel Bead Kit v2 or v3 chemistry ( both from 10x Genomics ) . Sample processing and library preparation were performed according to the manufacturer’s instructions using AMPure XP beads ( Beckman Coulter ) . Sequencing was either carried out on a local Illumina Nextseq 500 using the High-Out 75 cycle kit with a 26-8-0-57 read setup and 150 cycles or commercially ( Microanaly , China ) on a NovaSeq 6000 using the 300 cycle kit with a paired-end 150 read setup . Sample library kits version and sequencing information are shown in Supplementary file 1b . Processing of sequencing data was performed with the cellranger pipeline v3 . 0 . 2 ( 10x Genomics ) according to the manufacturer’s instructions . Raw bcl files were de-multiplexed using the cellranger mkfastq pipeline . Subsequent read alignments and transcript counting were done individually for each sample using the cellranger count pipeline with standard parameters . The cellranger aggr pipeline was employed to generate an sc barcode matrix containing all the samples without normalization . The normalization of each library was subsequently performed in Seurat ( see below ) . Subsequent analysis steps were carried out using Seurat v3 . 1 . 5 ( Stuart et al . , 2019 ) using R v4 . 0 . 2 as recommended by the Seurat tutorials . Briefly , cells were filtered to exclude cell doublets and low-quality cells with few genes ( <200 ) , high genes ( >900–3500 ) , or high mitochondrial percentages ( 5–7% ) in each patient individually . After quality control , the total cell number used for the analysis was 12 , 305 ( Supplementary file 1b ) . To account for technical variation , data were normalized using regularized negative binomial regression ( Hafemeister and Satija , 2019 ) , taking into account mitochondrial percentage and cycle score . Dimensionality reduction was done by principal component analysis . The number of principal components used for the further analysis was determined using an elbow plot . Cells were clustered using the ‘FindNeighbors’ ( based on k-nearest neighbor ( KNN ) graphs ) and ‘FindCluster’ ( Louvain algorithm ) functions in Seurat . To account for batch effects , different samples were aligned using Harmony ( Korsunsky et al . , 2019 ) . The UMAP was then used to visualize cells in a two-dimensional space . Clusters were annotated based on known marker genes . The ‘FindMarker’ function in Seurat , which used the Wilcoxon rank sum test , was applied to normalized and aligned data . The threshold of the adjusted p-value was set to 0 . 05 . Volcano plots were created with the R package EnhancedVolcano . DE genes identified by Seurat were used as the input . The threshold for the average log fold change was set at 0 . 5 and that for p-values at 0 . 001 . Summary statistics were downloaded from the NHGRI-EBI GWAS Catalog ( Buniello et al . , 2019 ) for the studies GCST007362/GCST007361 ( Robinson et al . , 2015 ) , GCST001345 ( Lin et al . , 2011 ) , GCST000563 ( Australo-Anglo-American Spondyloarthritis Consortium ( TASC ) et al . , 2010 ) , GCST007361 ( Robinson et al . , 2015 ) , GCST007844 ( Li et al . , 2019 ) , GCST001149 ( Evans et al . , 2011 ) , GCST005529 ( International Genetics of Ankylosing Spondylitis Consortium ( IGAS ) et al . , 2013 ) , GCST008910 ( Trochet et al . , 2019 ) , GCST003097 ( Ellinghaus et al . , 2016 ) , and GCST010481 ( Huang et al . , 2020 ) downloaded on 04/09/2020 . Cellular interactions were analyzed using CellPhoneDB ( Efremova et al . , 2020 ) . Normalized and aligned scRNA-seq data with the clusters identified by Seurat separated by diagnosis were used for analysis . Clusters with less than 10 cells were excluded . Statistical iterations were set at 1000 and genes expressed by less than 10% of cells in a cluster were removed . Resulting interactions are based on the CellPhoneDB repository . Heatmaps were produced by using the integrated heatmap function and then calculating the difference of the count of significant interactions in condition 1 and condition 2 . Dot plots were created with the integrated dot plot function . Flow cytometry analysis was performed on the maximum of recovered cells from the AqH samples ( ≤106 cells ) . Cells were first blocked with FcR anti-human blocking reagent ( Miltenyi ) . Afterwards , cells were stained for 30 min at 4°C in the dark with a combination of the following anti-human antibodies: CD3 ( perCp-Cy5 . 5 , clone OKT3 ) , CD4 ( BV510 , clone OKT4 ) , CD8 ( APC , clone SK1 ) , CD11b ( FITC clone M1/70 ) , CD11c ( Pacific Blue , clone 3 . 9 ) —all from Biolegend—and CD56 ( Pe-Cy7 , clone N901 ) and HLA-DR ( ECD , clone Immu-357 ) from Beckman Coulter . Samples were measured on a Gallios ( 10 colors , 3 lasers; Beckman Coulter ) flow cytometer using FACS Kaluza software v2 . 1 . 1 ( Beckman Coulter ) . Data were analyzed with FlowJo v10 . 6 . 1 ( BD Biosciences ) . The gating strategy is illustrated in Figure 2—figure supplement 2 . Cytokines in AqH samples were quantified via luminex analysis using a ProcartaPlex Human Cytokine-Panel 1B ( Thermo Fisher Scientific , Waltham , Massachusetts , USA ) that quantifies granulocyte macrophage-colony stimulating factor ( GM-CSF ) , IFN-α , IFN-γ , IL-1α , IL-1β , IL-1RA , IL-2 , IL-4 , IL-5 , IL-6 , IL-7 , IL-9 , IL-10 , IL-12p70 , IL-13 , IL-15 , IL-17A , IL-18 , IL-21 , IL-22 , IL-23 , IL-27 , IL-31 , TNF-α , and TNF-β , according to the manufacturer’s instructions . Standards and samples were measured in duplicates using Bio-Plex MAGPIX Multiplex Reader ( BioRad , Hercules , California , USA ) and cytokines were quantified in ( pg/µl ) using ProcartaPlex Analyst 1 . 0 software ( Thermo Fisher Scientific ) . | Uveitis is a form of inflammation in the eye . It can occur in response to infection , or when the immune system mistakenly attacks the eye , in what is known as autoimmune uveitis . In approximately 80 percent of cases , the front part of the eye is affected . During an inflammatory episode , the liquid inside the front part of the eye fills with immune cells , but the nature of these cells remains unknown . This is because uveitis is rare , and doctors cannot routinely take samples from inside the eyes of affected individuals to diagnose the disease . This lack of samples makes research into this disease challenging . There are two main groups of immune cells that could be responsible for uveitis: myeloid cells and lymphoid cells . Myeloid cells form the first line of immune defense against infection by non-specifically attacking and removing pathogens . Lymphoid cells form the second line of immune defense , attacking specific pathogens . Lymphoid cells also have long-term memory , meaning they can ‘remember’ previous infections and fight them more effectively . Lymphoid cells receive instructions from a type of myeloid cell called a dendritic cell about what to attack . Dendritic cells relay their instructions to lymphoid cells using molecules called human leukocyte antigens ( HLA ) . Autoimmune uveitis affecting the front part of the eye is common in individuals with an HLA type called HLA-B27 , suggesting that communication between dendritic and lymphoid cells plays an important role in this type of inflammation . To make the most of limited patient samples , Kasper et al . used single cell techniques to examine the immune cells from the fluid inside the eye . Six samples came from people with autoimmune uveitis , and one from a person with an eye infection . The infection sample contained mainly myeloid cells that might attack bacteria responsible for the infection . In contrast , the autoimmune uveitis samples contained mainly lymphoid cells . Of these samples , four were from individuals with the gene that codes for the HLA-B27 molecule . These samples had a unique pattern of immune cells , with more dendritic cells than the samples from individuals that did not have this gene . This study included only a small number of individuals , but it shows that analysing single immune cells from the eye is possible in uveitis . This snapshot could help researchers understand the local immune response in the eye , and find an optimal treatment . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"immunology",
"and",
"inflammation"
] | 2021 | Intraocular dendritic cells characterize HLA-B27-associated acute anterior uveitis |
The introduction of direct electron detectors with higher detective quantum efficiency and fast read-out marks the beginning of a new era in electron cryo-microscopy . Using the FEI Falcon II direct electron detector in video mode , we have reconstructed a map at 3 . 36 Å resolution of the 1 . 2 MDa F420-reducing hydrogenase ( Frh ) from methanogenic archaea from only 320 , 000 asymmetric units . Videos frames were aligned by a combination of image and particle alignment procedures to overcome the effects of beam-induced motion . The reconstructed density map shows all secondary structure as well as clear side chain densities for most residues . The full coordination of all cofactors in the electron transfer chain ( a [NiFe] center , four [4Fe4S] clusters and an FAD ) is clearly visible along with a well-defined substrate access channel . From the rigidity of the complex we conclude that catalysis is diffusion-limited and does not depend on protein flexibility or conformational changes .
It has long been recognized that electron cryo-microscopy ( cryo-EM ) has the potential to solve protein structures at near-atomic resolution ( Henderson , 1995 ) . Since the first sub-nanometer resolution cryo-EM structures of icosahedral viruses , which have the advantage of multi-MDa size and high symmetry ( Böttcher et al . , 1997; Conway et al . , 1997 ) , steady advances in instrumentation have yielded a number of virus structures at better than 4 Å resolution ( Grigorieff and Harrison , 2011 ) . For smaller and less symmetrical structures , progress has been slower , but several complexes have yielded structures at around 6 Å resolution where secondary structure is recognizable and X-ray structures of subcomplexes can be reliably fitted ( Ludtke et al . , 2004; Armache et al . , 2010; Gipson et al . , 2010 ) . In a few cases , side chain densities have been resolved ( Cong et al . , 2010; Ludtke and Baker , 2008; Zhang et al . , 2010 , 2013; Mills et al . , 2013 ) . Radiation damage limits the electron dose that can be used in any individual EM image and the resulting low signal-to-noise ratio ( SNR ) has meant that until recently high-resolution maps have required the equivalent of 106 images of asymmetric units . The recent introduction of direct electron detection cameras with much better detective quantum efficiency ( DQE ) has meant that images having higher SNR can be obtained with the same , or lower , total electron exposure . In addition , these cameras have frame rates which make it possible to collect videos of the particles in the thin film of vitreous water during an exposure , and thus partially correct for the effects of beam-induced particle movement and specimen drift . Using a 70-MDa virus that can be accurately aligned even at short exposures , it was shown that electron irradiation causes random translations and rotations of the particles , which are typically largest in the first few frames ( Brilot et al . , 2012; Campbell et al . , 2012 ) . In two recent studies , image processing schemes were developed to correct for these beam-induced motions . Using the 700 kDa archaeal 20S proteasome , which is too small to reliably detect and align in images recorded with a short exposure , a protocol was developed to align entire frames or subareas of >2000 × 2000 pixels to each other ( Li et al . , 2013 ) ( The motion correction software is available to download from http://www . nature . com/nmeth/journal/v10/n6/extref/nmeth . 2472-S2 . zip ) . For the much larger ribosome ( ∼4 MDa ) a statistical video processing approach was developed , which acts on individual particles using a user-defined running average of video frames ( Bai et al . , 2013 ) . Both studies produced maps of unprecedented resolution , 3 . 3 Å for the D7 proteasome from 126 , 000 particles and 4 . 5 Å from 35 , 000 asymmetric 80S ribosomes . Recently we determined the structure of the 1 . 2 MDa Frh complex , the F420-dependent hydrogenase from Methanothermobacter marburgensis , ab initio from cryo-EM data collected on photographic film ( Mills et al . , 2013 ) . Frh is a key enzyme in the metabolism of methanogenic archaea , where the reduction of carbon dioxide to methane involves the oxidation of four molecules of H2 by a number of different hydrogenases . The reduced form of the F420 coenzyme , which is the electron donor in several of these steps , is regenerated by Frh ( Thauer et al . , 2010 ) . The central role of Frh in the metabolism of methanogens is reflected in its abundance ( ∼2% ) in the soluble cell protein ( Fox et al . , 1987 ) . Frh is a heterotrimeric enzyme composed of the 43 kDa subunit FrhA that contains a [NiFe]-center , the 26 kDa subunit FrhG with three [4Fe4S] clusters , and the 31 kDa iron–sulphur flavoprotein FrhB , which contains the F420-binding site and has one [4Fe4S] cluster and an FAD . Our cryo-EM study showed that the Frh complex is a dodecamer with tetrahedral symmetry . The map had ∼5 Å resolution which was sufficient to show secondary structure and density for many side chains as well as the cofactors forming the electron transfer chain , making it possible to trace the three protein chains , one of them without a template . We have now collected a new dataset of Frh as videos on a back-thinned Falcon II detector and reconstructed a cryo-EM using a ‘gold standard’ refinement approach . The resulting map from 26 , 000 particles has a resolution of 3 . 36 Å , which enabled us to refine the model obtained from film data . We can now trace the three proteins in the complex completely , and localize most side chains with confidence . We compare and discuss different processing schemes to deal with the video data .
Images of the Frh complex were collected using the back-thinned FEI Falcon-II in video mode with 18 frames per second . Using videos allows long exposures in which the individual particles and Thon rings are clearly visible , while retaining the radiation-sensitive higher resolution information only present during the initial exposure . The particles were clearly distinguishable below 1 μm defocus with a 1 . 5 s exposure ( ∼90 e/Å2 ) ( Figure 1A ) . A dataset of 33 , 590 particles ( 403 , 080 asymmetric units ) at a defocus range of 0 . 8–3 . 5 µm was collected from 235 images in two sessions on a single grid . Areas of very thin ice were selected that we estimate to be only slightly thicker than the particle diameter of ∼175 Å; often similar holes nearby had no particles in the center but a densely packed outer rim , so apparently the thin ice had squeezed the complexes out to the edge of the hole . 10 . 7554/eLife . 01963 . 003Figure 1 . Cryo-EM data collection of the Frh complex and CTF correction . ( A ) A typical electron micrograph recorded with the Falcon II camera on the FEI Tecnai Polara operated at 300 kV . The defocus was determined by CTFFIND3 to be 0 . 9 µm ( Mindell and Grigorieff , 2003 ) . The particles are clearly visible and easy to box . Scale bar , 25 nm . ( B ) CTF of boxed particles . The inset shows a zoom of the high-resolution range . Thon rings are visible beyond 80% of the Nyquist frequency at 0 . 3 Å−1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 003 The Frh dodecamer was refined from a low-pass filtered model to high resolution using the RELION gold standard refinement procedure . In this protocol , two half datasets are refined completely independently throughout , and after each cycle the new reference volumes are low-pass filtered to the resolution where the Fourier shell correlation ( FSC ) between the two volumes drops to 0 . 143 , thus preventing overfitting of noise ( Scheres and Chen , 2012 ) . A first refinement using the sum of the first 20 video frames ( discarding the last 4 frames because of likely radiation damage [Baker and Rubinstein , 2010] ) yielded a map at 3 . 94 Å resolution ( Figure 2A ) that looked considerably better than our previous map from 84 , 000 particles on film . Also , the map was virtually noise-free due to the RELION refinement procedure ( Scheres , 2012 ) . Next we applied the frame alignment software ( Li et al . , 2013 ) and aligned all 24 frames to each other . In the majority of cases , the first frame showed a large movement in comparison with the others ( Figure 3A ) . We refined the sum of the aligned frames with or without the first frame and found that discarding the first frame results in a higher-resolution map ( Figure 3B ) . As found also by others ( Li et al . , 2013 ) , the correction cannot deal with the fast motion of the particles during the first frame . On this basis we discarded the first frame of every video for all subsequent refinements . The refinement of the 20 aligned frames resulted in a 3 . 74 Å map ( Figure 2A ) . In another refinement approach , we used the statistical alignment procedure ( Bai et al . , 2013 ) implemented in RELION ( Scheres , 2012 ) . This method follows the movement of the particles within the videos ( particle-based method ) using a running average of frames . This procedure also improved the resolution relative to the unaligned particles in a very similar way to the whole-frame alignment ( Figure 2A ) ; the best resolution was obtained when 5 frames were averaged . In yet another approach we combined the two alignment procedures ( whole-frame and particle-based ) . The resolution curve of this refinement gave a better FSC than the others over all frequencies and the resulting reconstruction had a resolution of 3 . 69 Å ( Figure 2A ) , indicating that for the intermediate-sized particle investigated here the combination of the two alignment procedures works best . 10 . 7554/eLife . 01963 . 004Figure 2 . Fourier shell correlation ( FSC ) curves for different refinement strategies . All refinements were performed with the gold standard procedure in RELION ( Scheres and Chen , 2012 ) . A post-processing procedure ( Chen et al . , 2013 ) was applied unless otherwise indicated . The dotted line is at FSC 0 . 143 , used to determine the resolution from comparing two independently refined half data sets ( Rosenthal and Henderson , 2003 ) . ( A ) Comparison of different alignment procedures . Blue , average of 20 unaligned frames . Purple , 20 frames aligned with the statistical video processing procedure ( Bai et al . , 2013 ) . Red , refinement after aligning frames with the area-motion correction software ( Li et al . , 2013 ) . Gold , combination of the latter two alignment procedures . ( B ) Effect of radiation damage . Blue , 20 frames ( 3 . 69 Å ) ; purple , 16 frames ( 3 . 69 Å ) ; red , 12 frames ( 3 . 60 Å ) ; gold , 8 frames ( 3 . 43 Å ) ; green , 6 frames ( 3 . 39 Å ) . All curves were obtained by the combination approach described above . ( C ) Data set quality . Particle images of sub-standard quality were omitted from the original data set of 33 , 590 images , yielding a smaller dataset of 26 , 635 particles ( ‘Results’ ) . Refinement using 8 frames of the reduced data set ( green , 3 . 39 Å ) . The improvement is clear in comparison with the full dataset ( blue , 3 . 43 Å ) . A further improvement ( red , 3 . 36 Å ) resulted from using 6 instead of 8 frames . ( D ) Post-processing to determine the resolution and B-factor ( Chen et al . , 2013 ) of the final map from the 6-frame refinement . The raw unmasked map ( purple ) indicates a resolution of 3 . 52 Å; the map masked with a soft mask ( red ) indicates the final resolution of 3 . 36 Å . The gold curve shows the FSC for the two half data sets with randomized phases beyond 4 . 5 Å . Subtraction of the gold curve from the red curve yields the green curve , which indicates the true map resolution corrected for over-aggressive masking . The close correspondence of the red and green curves shows that the used soft mask did not introduce spurious correlation and the true map resolution is 3 . 36 Å . A B factor of −156 Å2 was determined and applied to sharpen the map . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 00410 . 7554/eLife . 01963 . 005Figure 3 . Specimen movement as detected by recording images in video mode . The motion correction software indicates a large movement at the beginning of the exposure . ( A ) Video frame alignment for three separate micrographs . Each spot represents one frame . The drift plots show that the movement between the first and second frame is considerably higher than in subsequent frames , although some micrographs ( red and green ) indicate much higher drift than others ( blue ) . ( B ) FSC curves of a refinement with and without the first video frame . The violet curve represents a refinement of frames 1–17 , the green curve frames 2–17 . Although at FSC 0 . 143 the resolution is the same for for both maps ( 3 . 69 Å ) , the green curve without the first frame clearly shows a higher FSC in the whole resolution range . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 005 We had already discarded the last 4 frames of the 24-frame videos , because of likely radiation damage . We then continued to assess the effect of radiation damage on the quality of the map by successively reducing the number of frames . Every omitted frame improved the map resolution until only 6 frames were averaged . This gave the best resolution of 3 . 39 Å , compared to 3 . 74 Å for all 20 frames ( Figure 2B ) . Reducing the number of averaged frames further to 4 or 5 made the resolution worse . We attempted to improve the resolution further by adding more data . A third microscope session on a different grid yielded another 15 , 000 particles . Adding them , however , reduced the map resolution and these images were therefore not included in further steps . We assume the data quality was worse because of the thicker ice on this grid , which had made it necessary to use a higher defocus , never less than 1 . 2 µm , to distinguish the particles . We therefore decided to discard suboptimal images from the first dataset to improve the map . We removed the images with high defocus ( >2 . 5 µm ) and those that were of poor quality as judged by the visibility and symmetry of the Thon rings . This reduced the dataset by 20% from 33 , 590 to 26 , 635 particles ( 319 , 620 asymmetric units ) . Refinement of this smaller dataset under otherwise identical conditions resulted in maps with better resolution than the full dataset , and the best map , with 6 frames averaged , extended to 3 . 36 Å resolution ( Figures 2C and 4 ) . The accumulated dose for this data was ∼24 e−/Å2 . We tested the local resolution of the map with the program ResMap ( Kucukelbir et al . , 2014 ) to identify possible flexible protein regions . The local resolution map was featureless ( not shown ) , indicating that the dodecameric Frh complex is completely rigid . 10 . 7554/eLife . 01963 . 006Figure 4 . The 3 . 36 Å map with each of the 12 heterotrimers in a different colour . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 006 To assess the effects of radiation damage further , we also created maps of the last 6 and the middle 6 frames of the 20-frame videos of the reduced dataset , averaging particles that were in effect pre-irradiated with ∼24 and 49 e−/Å2 . This resulted in maps of 3 . 94 and 4 . 16 Å resolution , respectively , which was clearly reflected in the poorer visibility of side chain densities ( Figure 5 ) . 10 . 7554/eLife . 01963 . 007Figure 5 . Effect of radiation damage . Helix 124–144 of FrhA in the map calculated from ( A ) video frames 1–6 ( 3 . 36 Å ) , ( B ) frames 8–13 , pre-irradiated by ∼24 e/Å2 ( 3 . 94 Å ) , and ( C ) frames 15–20 , pre-irradiated by ∼49 e/Å2 ( 4 . 16 Å ) . Note that side chain density is lacking for Asp125 and Glu132 already in the first map . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 007 We fitted the model of the FrhABG heterotrimer based on the model built into the ∼5 Å map obtained previously from film data ( Mills et al . , 2013 ) . The overall features of the 3 . 36 Å map were similar , but the higher resolution of the new map and the low noise level made the interpretation of the electron densities unambiguous ( Figure 6; Video 1 ) . Alpha helices ( Figure 7 ) and beta sheets ( Figure 8 ) were easily recognizable and the large majority of side chains had good density ( Figure 9; Video 2 ) . Most of the protein could be fitted to the map using the ‘real space refine’ feature in Coot , which had not been possible for the lower resolution map from film images . The FSC between this map and a map calculated from the fitted model is 0 . 5 at 3 . 56 Å resolution , close to the resolution of the final map , confirming the correctness of the fit ( Figure 10 ) . 10 . 7554/eLife . 01963 . 008Figure 6 . EM map of Frh at 3 . 36 Å resolution . ( A ) Slice through an FrhABG heterotrimer at the level of the electron transfer chain . ( B ) The same slice with atomic model . In this and other figures the carbons of FrhA are green , FrhG magenta , and FrhB blue . Green and orange spheres indicate the [NiFe] center in FrhA; the [4Fe4S] clusters in FrhG and FrhB are shown as orange and yellow spheres . The FAD in FrhB is shown as a stick model with yellow carbons . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 00810 . 7554/eLife . 01963 . 009Video 1 . A slice through the map and model of an FrhABG heterotrimer . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 00910 . 7554/eLife . 01963 . 010Figure 7 . Two helices of the 4-helix bundle in FrhA ( Leu92-Ala114 and Val276-Glu300 ) without and with the model . Note the absence of side chain density for glutamate and aspartate side chains . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 01010 . 7554/eLife . 01963 . 011Figure 8 . Beta sheet 319-348 of FrhA . ( A ) Top view , ( B–D ) side views of individual strands 341–348 , 328–339 and 319–327 , rotated by 90° relative to ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 01110 . 7554/eLife . 01963 . 012Figure 9 . Resolved amino acid side chains in the 3 . 36 Å map . The histogram shows the number of residues on the y-axis and the amino acid ( one letter code ) on the x-axis . Blue bars indicate fully resolved side chains , red bars indicate side chains without or with ambiguous map density . Glycine residues ( no side chain ) are indicated in green for completeness . Negatively charged side chains of aspartate ( D ) and glutamate ( E ) are almost all missing . In contrast , side chains of hydrophobic residues like valine ( V ) , leucine ( L ) , isoleucine ( I ) , phenylalanine ( F ) , tyrosine ( Y ) , and tryptophan ( W ) , are nearly all visible . Excluding glycine , aspartate and glutamate , 86% of side chains are well visible . Unresolved side chains other than aspartate and glutamate are mostly located on the surface . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 01210 . 7554/eLife . 01963 . 013Video 2 . The quality of the 3 . 36 Å map shown for alpha-helix 88–114 of FrhA . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 01310 . 7554/eLife . 01963 . 014Figure 10 . Correlation between map and model . The blue line shows the FSC between the final cryo-EM map and a map calculated from the fitted model; the red line is the FSC between maps from independent halves of the data . Dotted lines indicate the 0 . 5 FSC criterion for the map/model comparison and 0 . 143 for the half datasets . The cryo-EM map was filtered to 3 . 36 Å , causing the map-to-model correlation to drop to 0 at that resolution . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 014 All three subunits were traced completely except for a few N- or C-terminal residues , accounting for 893 out of 903 amino acids ( Figure 11 ) . We found that all three proteins had been traced essentially correctly in the lower-resolution film map ( Video 3 ) . The largest differences between the two models were found in surface loops that had been difficult to trace in the film map . The only unresolved part in the previous model involved a 12-residue stretch in FrhG ( 188–199 ) , for which no density was found and which was thought to form a flexible surface loop . In the new map , two nearby density features , modelled as extended loops before , were clearly recognizable as alpha-helices ( 194–201 and 208–215 ) that accommodate the missing residues . The resulting shift of residues brings the cysteine residues C206 and C208 , which are conserved in the FrhG family ( Figure 11B ) , near a strong density on the dimer axis between two FrhG subunits . The cysteines of the dimer partner are just a few Å away , and the density in between the four cysteine side chains is strongly suggestive of a coordinated ion ( Figure 12 ) . Although the cysteine side chains are not completely resolved , their tetrahedral arrangement suggests that it is a zinc ion ( Dokmanić et al . , 2008; Harding et al . , 2010; Rulíšek and Vondrášek , 1998; Zheng et al . , 2008 ) . 10 . 7554/eLife . 01963 . 015Figure 11 . Polypeptide sequence and secondary structure . α-helices are highlighted in green , β-strands in blue . The second line shows a consensus sequence of the protein families ( Mills et al . , 2013 ) with fully conserved amino acids in capitals and partly conserved residues in lower case ( h: hydrophobic; s: small [GAS]; l: large [LIFYHW]; a: aromatic [FYWH]; z: T or S; n: negative , D or E; p: positive , R or K ) . ( A ) FrhA . In the consensus sequence , the [NiFe] ligands are highlighted in orange and the ligands of the third ion in red . ( B ) FrhG . Ligands of the proximal , medial , and distal [4Fe4S] cluster are shown in yellow , orange , and red , respectively . The cysteines coordinating a putative zinc ion on the FrhG dimer interface are highlighted in magenta . ( C ) FrhB . The residues for coordination of the iron–sulphur cluster and FAD are highlighted in green and cyan , respectively , and residues lining the F420 access channel in yellow . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 01510 . 7554/eLife . 01963 . 016Video 3 . Comparison of the Frh models based on the film map ( Mills et al . , 2013 ) and the Falcon II map from the present study . Light colours: film , dark colours: Falcon II . FrhA is shown in green , FrhG magenta , FrhB blue . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 01610 . 7554/eLife . 01963 . 017Figure 12 . The ferredoxin domains ( residue 206-260 ) of an FrhG dimer containing the medial and distal FeS clusters . The two protomers are shown in shades of purple . A high density on the dimer axis between two copies of Cys206 and Cys208 is interpreted as an ion ( grey sphere ) , most likely Zn2+ . The ion is ∼9 Å away from the medial FeS cluster ( top left and bottom right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 017 The FAD cofactor in FrhB has continuous density in the map , and its folded conformation with the adenine and isoalloxazine moieties in close proximity ( Mills et al . , 2013 ) is confirmed ( Figure 13; Video 4 ) . As noted before ( Mills et al . , 2013 ) , the FAD is completely surrounded by conserved residues , including loops 23–30 , 71–77 , and 131–138 ( which includes Cys134 , a ligand of the nearby FeS cluster i . e . , the electron donor to FAD ) ( Figure 11C ) . The binding pocket is well-defined in the new map , with the pyrophosphate moiety at the N-terminal end of helix 27–39 and the conserved 24QDGG as an extended chain around it ( Figure 13A ) . There is no density for the substrate , F420 , but the FAD is accessible from the surface through a ∼5 Å gap between the main body of FrhB and a domain formed by residues 128–188 containing an alpha helix and a three-stranded beta sheet ( Video 5 ) . This putative substrate access channel is , like the FAD itself , completely lined by conserved residues ( Video 6 ) . The model for the FrhABG trimer and the dodecamer is shown in Figure 14 . 10 . 7554/eLife . 01963 . 018Figure 13 . FAD cofactor in FrhB with part of its binding pocket . Conserved residues are labelled in black . Other residues mentioned in the text are grey . ( A ) The phosphate moiety ( orange ) sits in a pocket formed by A23–T30 at the C-terminal end of helix 28–39 . The adenine moiety ( left ) is coordinated by the loop A72–N81 . ( B ) In this view , the loop I132–F138 surrounding the isoalloxazine ring can be seen as well as the highly conserved loop G73–T77 . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 01810 . 7554/eLife . 01963 . 019Video 4 . FAD and its binding pocket . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 01910 . 7554/eLife . 01963 . 020Video 5 . The 3 . 36 Å complex map with trimer colours , sliced and rotating to see FAD coloured with FrhA green , FrhG magenta , FrhB blue , and the ligands gold . The substrate access channel from the surface to FAD is clearly visible between two domains of FrhB . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 02010 . 7554/eLife . 01963 . 021Video 6 . Conserved residues in FrhB . The ligands FAD and FeS are shown in magenta and green , respectively . Conserved residues and corresponding densities are shown in red and partially conserved residues in orange . Both ligands and the F420 access channel are completely surrounded by conserved residues . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 02110 . 7554/eLife . 01963 . 022Figure 14 . Cartoon of the FrhABG heterotrimer ( top ) and the tetrahedral complex of 12 trimers . FrhA is green , FrhG magenta , and FrhB blue . The [NiFe] center in FrhA is shown as green and orange spheres , the three [4Fe4S] clusters as orange and yellow spheres , and the FAD in FrhB as a stick model with yellow carbons , the ion in FrhA is orange , and the putative zinc ion on the twofold axis of the FrhG dimer is grey . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 022
Cryo-EM has long had the potential of reaching near-atomic resolution , but until very recently this has only been achieved for very large , highly symmetrical icosahedral viruses . The fundamental problem is the inevitable radiation damage to the biological sample caused by the electron beam . Reaching high resolution requires low exposures to maximize the SNR of rapidly destroyed high-resolution information . The resulting images are noisy , which makes it both difficult to detect particles and to determine the defocus parameters accurately . The high frame rate of direct electron detectors allows data collection in video mode , which removes these restrictions and allows beam-induced motion of particles to be identified and partially corrected for . Just as importantly , direct electron detectors offer increased detective quantum efficiency at high resolution and so while individual images are still noisy they contain more information . The combination of better images and improved data processing is now leading to better reconstructions which in turn allows more information to be extracted from the images due to better alignment of individual particles . The advantages are illustrated in recently published studies of two very different specimens , the 4 MDa asymmetric Saccharomyces cerevisiae ribosome and the 700 kDa D7-symmetric proteasome . These studies used two different detectors , the FEI Falcon II and the Gatan K2 , respectively , and employed different strategies to align the video frames ( Bai et al . , 2013; Li et al . , 2013 ) . We have now reconstructed the 1 . 2 MDa tetrahedral Frh complex from ∼26 , 000 particles collected on the Falcon II , and determined the structure of the three proteins in the complex at high resolution . The Falcon II data quality is far superior to that obtained with photographic film . With film we obtained a 5 Å map from 84 , 000 Frh particles ( Mills et al . , 2013 ) , but with Falcon II images of the same sample and the same electron microscope we were able to achieve a map with better than 4 Å resolution from 70% fewer particles . The higher DQE of the detector makes it possible to reliably detect particles at very low defocus . This is illustrated by Figure 1A , showing a micrograph taken at 900 nm defocus in which the particles are clearly visible , whereas with film we were not able to detect the particles at less than 1500 nm defocus . The high-resolution signal for the low-defocus images is very much better than that for images recorded at higher defocus , which was a critical factor in attaining the final map . This is illustrated by the fact that adding more particles recorded at higher defocus ( from a grid with thicker ice ) actually degraded the overall resolution , and that omitting particles with more than 2500 nm defocus , which did not contribute to the signal at high resolution , improved the map ( Figure 2C ) . A decisive advantage of direct electron detectors over film is the possibility of collecting video data . We found that two issues are important here: aligning the video frames to reduce the effects of beam-induced movement , and selecting the frame sequence used in the final map reconstruction to eliminate data affected by radiation damage . We used two approaches for video alignment . The first was the procedure developed by Li et al . ( 2013 ) for aligning whole frames ( or large fractions of frames ) independent of the visibility of the particles of interest in the individual frames . This was developed for the 700 kDa 20S proteasome and is most useful for relatively small protein complexes that are hard to detect on images recorded at low electron exposure . The second method is the statistical alignment procedure developed by Bai et al . ( 2013 ) , which works on the particle level on user-defined running averages of substacks and requires the particles to be visible on the substacks . This was developed for ribosomes , which for a given dose are easier to detect due to their large size ( several MDa ) and high RNA content . The 1 . 2 MDa Frh complex , which contains 48 [4Fe4S] clusters and several other metal ions , is intermediate in density and size between the proteasome and the ribosome . For this complex we found that both video alignment methods gave similar improvements in final map resolution compared to unaligned frames ( Figure 2A ) . However , the combination of the two methods , running the statistical particle alignment on pre-aligned stacks , yielded another improvement of similar magnitude . We used a running average of 5 frames for the particle-based alignment , so it is not unexpected that a pre-alignment of the frames was advantageous . Collecting data in video mode also allows elimination of sub-optimal frames from the reconstruction . As was found in earlier studies ( Brilot et al . , 2012; Campbell et al . , 2012; Li et al . , 2013 ) , beam-induced movement was much more noticeable in the first frame than in subsequent frames ( Figure 3A ) . This large movement during the first fraction of the exposure would blur the first frame , and by omitting it , we obtained an improvement in resolution ( Figure 3B ) . The later frames are increasingly affected by radiation damage . Reducing the number of frames used in the reconstruction to only 6 ( out of the total 24 ) produced the best map ( Figure 2B ) . The accumulated dose in these frames was ∼24 e/Å2 , very similar to the accumulated dose ( ∼21 e/Å2 ) used for the 3 . 3 Å proteasome map ( Li et al . , 2013 ) . A 4 . 16 Å reconstruction after a refinement of only the last 6 frames of the 20-frame videos still showed partial side chains and clear helices ( Figure 5C ) . In case of smaller particles than the 1 . 2 MDa Frh complex , it would be advantageous to record videos over longer periods with higher cumulative doses , so that the particles are more visible and easier to align . The later , radiation-damaged frames can then be omitted from the reconstruction if necessary , as suggested before ( Bai et al . , 2013 ) . Thus , direct electron detectors provide a way to optimize a posteriori the electron exposure of the data used for reconstruction . Our 3 . 36 Å resolution map made it possible to determine the structure of the Frh complex unambiguously . All elements of secondary structure are obvious , and side chain density is clearly visible for most residues in the protein interior . Side chains on the surface are in general not seen , probably due to their flexibility in the aqueous solvent . The only conspicuous exceptions are the side chains of glutamates and aspartates , which are almost all absent ( Figure 9 ) , even in maps reconstructed from the earliest , least damaged frames ( Figure 5A ) . An absence of aspartate and glutamate side chain densities can also be observed in other high-resolution cryo-EM maps , including the electron crystallography maps of LHC-II ( Kühlbrandt et al . , 1994 ) and bacteriorhodopsin ( Kimura et al . , 1997 ) and the recent 3 . 3-Å single particle map of the 20S proteasome ( Li et al . , 2013 ) . In X-ray crystallography it was noted that carboxylate side chains have higher B-factors after extended exposure to intense synchrotron radiation , possibly due to decarboxylation ( Weik et al . , 2000 ) . A quantification of synchrotron radiation damage on carboxylate side chains ( Fioravanti et al . , 2006 ) shows that this damage already occurs at radiation doses equivalent to less than 1 e/Å2 ( Henderson , 1990 ) , a much lower dose than is feasible in cryo-EM . This implies that the decarboxylation would have occurred well within the first frame of our video data set and aspartate and glutamate side chains would in general not be visible in cryo-EM maps . However , not all are absent . Among the few visible carboxylate side chains in Frh , two are involved in ligand binding: FrhG Asp60 in the coordination of the proximal FeS cluster , and FrhA Glu44 in liganding a possible Fe or Mg ion ( Mills et al . , 2013; Figure 15 ) . In X-ray crystallography it is also found that not all glutamate and aspartate side chains are equally susceptible to radiation damage , but the dependence on the structural and chemical environment is not well understood ( Fioravanti et al . , 2006 ) . In the 3-Å bacteriorhodopsin EM map , the side chains of Asp85 and Asp212 , which from spectroscopy are known to be deprotonated , were invisible , whereas the protonated side chains of Asp96 and Asp115 had good density ( Kimura et al . , 1997 ) . Active sites are often found to contain the most radiation-sensitive residues in X-ray crystallography , but the protonation state was not found to be essential , and a relation to pKa was also not observed ( Fioravanti et al . , 2006 ) . As more high-resolution cryo-EM maps become available , more data on the relative radiation sensitivity of side chains in different environments may give new insights in protein structure and dynamics . 10 . 7554/eLife . 01963 . 023Figure 15 . ( A ) The proximal FeS cluster of FrhG is coordinated by three cysteine residues and an aspartate ( Asp60 ) with clear density . ( B ) An ion in FrhA ( grey sphere ) is coordinated by the C-terminal His386 , the main chain oxygen of Ala347 and by Glu44 , one of the few carboxylate residues with clear density . DOI: http://dx . doi . org/10 . 7554/eLife . 01963 . 023 The structure of Frh was traced ab initio in our previous cryo-EM map from film data , which had a nominal overall resolution of 5 . 5 Å as determined by gold-standard FSC , but a significantly higher resolution of ∼4 . 5 Å in well-defined regions of the map ( Mills et al . , 2013 ) . There were many indications that our original Frh structure was essentially correct , for example well-resolved hydrophobic side chains in the protein interior and the observation of secondary structure in regions where it was expected from prediction programs . It was nevertheless gratifying to see how well the model fits the new , higher-resolution map . Note that it would hardly be possible to trace an electron density map obtained by X-ray crystallography at a nominal resolution of 5 . 5 Å . This clear difference reflects the quality of the phase information , which is conserved in the phase contrast electron micrographs and determined directly by image processing , whereas in X-ray crystallography it is determined indirectly by isomorphous replacement or anomalous scattering . Although the overall chain trace is the same , in detail the model based on the 3 . 36 Å map is of course much improved . The function of Frh is the hydrogenation of F420 by molecular hydrogen . Electrons are extracted from hydrogen by the [NiFe] cluster in FrhA , and transferred via the three [4Fe4S] clusters in FrhG and another one in FrhB to FAD and from there to the substrate . The electron transfer chain is clearly recognizable . The [NiFe] cluster has a high density ( Figure 15B ) and is coordinated by conserved residues , as in the large subunits of other [NiFe] hydrogenases ( Figure 11A ) . Nearby is another ion , also conserved , coordinated among others by the C-terminal histidine of FrhA , His386 ( Figure 15B ) . The four FeS clusters were easily localized in the film map because of their high density; moreover , they are arranged in a chain with distances of ∼10 Å ( Mills et al . , 2013 ) . In the Falcon II map , the clusters are not just featureless blobs , but their tetrahedral coordination ( mostly by cysteine residues , or , in one case , an aspartate ) can be clearly seen ( Figure 6 , Figure 15A ) . Thus their orientation can be modelled correctly , with the irons facing the sulphurs of the cysteine side chains . One unexpected feature found in the new map was the density for an ion on the dimer interface between two FrhG subunits ( Figure 12 ) . This density feature was actually visible in the film map , but since the flanking conserved cysteines from the two FrhG subunits were not recognized , its interpretation as an ion was not obvious . The function of this ion is most likely to stabilize the dodecameric complex , but its location between the two medial FeS clusters at a distance of only 9 Å , which is less than the distance of ∼10 Å between adjacent FeS clusters in the electron transfer chains , means that a role in electron transfer cannot be ruled out . The final part of the electron transfer chain , the FAD in FrhB , which is close to the last FeS cluster , has unambiguous , continuous density in the Falcon II map ( Figure 13; Video 4 ) . FAD acts as a one-electron/two-electron redox switch between the FeS clusters and the F420 substrate ( Fox et al . , 1987; Alex et al . , 1990; Thauer et al . , 2010 ) . This means that the substrate needs to approach the FAD to within van der Waals distance to enable hydride transfer ( Ceh et al . , 2009 ) . In our previous study , we collected data from Frh with and without the substrate . In the map obtained in the presence of substrate we had identified a density near the FAD isoalloxazine ring that we interpreted as part of F420 at low occupancy ( Mills et al . , 2013 ) . For the new Falcon-II map , we used an Frh sample with a large excess of F420 , but the F420 binding pocket is clearly empty . As the same sample was used in both studies , we conclude that the density in the film map must have been due to noise . An access channel for the substrate can easily be recognized ( Video 5 ) and its significance is confirmed by the high conservation of the amino acid residues lining the channel ( Video 6 ) . There is no indication of even a low occupancy of substrate or of flexibility in this map region . A map of local resolution ( Kucukelbir et al . , 2014 ) confirmed that the protein region around this pocket is rigid . This suggests that conformational changes do not play a role for the access of F420 to the hydride donor , consistent with the finding that F420 reduction by Frh is very rapid and most likely diffusion-limited ( Livingston et al . , 1987 ) . The advent of direct electron detection cameras opens new horizons for cryo-EM . It is now possible to obtain higher-resolution maps with many fewer images . Using the video data collection mode , the optimal data in terms of SNR or radiation damage can be extracted a posteriori , and frame alignment schemes can be fine-tuned to the object under investigation .
The purification of the Frh complex from Methanothermobacter marburgensis and the preparation of the grids for the electron microscope data collection were performed as described ( Mills et al . , 2013 ) . Briefly , 3 μl of a 0 . 7 mg/ml Frh sample in the presence of 10 mM F420 was applied to freshly glow discharged Quantifoil R1/4 holey carbon grids ( Quantifoil Micro Tools , Jena , Germany ) . The grids were blotted in an FEI Vitrobot plunge-freezer . Data was collected on an FEI Tecnai Polara operating at 300 kV , using a back-thinned FEI Falcon II direct electron detector . The microscope was carefully aligned as previously described ( Mills et al . , 2013 ) and the Falcon II camera was calibrated at the desired nominal magnification of 78 , 000× . The calibrated magnification on the 14 µm pixel camera was 106 , 000 , resulting in a 1 . 32 Å pixel size at the specimen . The camera system was set up to record 18 frames/sec as previously described ( Bai et al . , 2013 ) . Videos were collected for 1 . 5 s with a total of 24 frames with a calibrated dose of 3 . 5 e−/Å2 per frame , at various defocus values in the range between 0 . 8 and 3 . 8 μm . Particle picking was carried out using the semi-automatic procedure of EMAN Boxer ( Ludtke et al . , 1999 ) , and the contrast transfer function of every image was determined using CTFFIND3 ( Mindell and Grigorieff , 2003 ) in the RELION workflow ( Scheres , 2012 ) . If necessary , the CTF values were double-checked using the particle-based CTF procedure of EMAN2 ( Tang et al . , 2007 ) . Four independent refinements were launched with the gold standard refinement procedure of RELION ( Scheres and Chen , 2012 ) starting from our previous Frh map from film data ( Mills et al . , 2013 ) low-pass filtered to 60 Å , using 20 frames ( from frame 2 to frame 21 ) , testing four different approaches:The 20 unaligned frames of each video were added up without motion correction;The 20 frames of each video were aligned using the whole-image motion correction method described in Li et al . , 2013 ( the authors kindly provided the necessary support to install the software ) ;The 20 unaligned frames were processed by the statistical video refinement procedure described in Bai et al . ( 2013 ) ( particle-based ) ;A combination of approaches 2 and 3 , applying the statistical video refinement procedure ( Bai et al . , 2013 ) to the 20 frames pre-aligned using the whole-image motion correction software ( Li et al . , 2013 ) . Procedure number 4 , which combines the area-based and particle-based frame alignment , gave the best results . All further maps as described in the Results section were obtained by RELION refinement as in procedure 4 , unless less than 8 video frames were used , in which case procedure 2 was followed . Tetrahedral symmetry was applied in all refinements . A post-processing procedure implemented in RELION ( Scheres , 2012 ) was applied to the final maps for appropriate masking , B-factor sharpening and resolution validation to avoid over-fitting ( Chen et al . , 2013 ) . In this procedure , the appropriate B-factor is determined according to Rosenthal and Henderson ( 2003 ) , after correction for the Falcon II MTF . In addition , a soft mask is applied to the last two unfiltered models before convergence and a new FSC curve is calculated . The procedure also measures any spurious correlation due to too tight masking , by subtracting the FSC curve between the two masked half datasets where the phases beyond a chosen resolution were randomized . For all maps , B factors between −150 and −230 Å2 were found . All resolutions stated are at FSC 0 . 143 ( Rosenthal and Henderson , 2003 ) after applying this post-processing procedure . The local resolution of the map was estimated with the ResMap software ( available at http://resmap . sourceforge . net ) ( Kucukelbir et al . , 2014 ) . The protein structure was built in Coot ( Emsley and Cowtan , 2004 ) into the high-resolution map using real-space refinement , starting with the earlier model from cryo-EM data recorded on film ( Mills et al . , 2013 ) . Torsion angle , planar peptide , and Ramachandran restraints were applied throughout . The final model for the FrhABG heterotrimer contains 893 amino acid residues out of a possible 903 , a [NiFe] cluster , four [4Fe4S] clusters , one FAD , and two metal ions , one of them coordinated by two trimers . 96 . 5% of residues have backbone dihedral angles in the most favored region of the Ramachandran plot and the remaining 3 . 5% are in the generously allowed regions . Figures were made using Chimera ( Pettersen et al . , 2004 ) . | Many microbes rely on enzymes known as hydrogenases to catalyse the metabolic reactions that generate energy . These enzymes cleave hydrogen molecules to release electrons that go on to participate in further reactions . In order to fully understand how hydrogenases and other enzymes work it is necessary to work out their structure at the atomic level . Last year a technique known as electron cryo-microscopy ( cryo-EM ) was used to show that Frh—a hydrogenase that is crucial for many different steps in the metabolic process of microbes that produce methane—had a tetrahedral structure . Cryo-EM involves freezing the molecule of interest in a layer of ice to preserve its structure as it is imaged with an electron beam . Unfortunately , the signal-to-noise ratio in each image is low , so researchers must combine many separate images in order to determine the structure of the molecule . The use of a new type of electron detector can improve the performance of an electron cryo-microscope in several ways . Higher frame rates can be used , which makes it possible to correct for movement of the molecule caused by the electron beam . The new electron detectors are also more efficient , so samples can be exposed to lower doses of electrons , reducing damage to the sample . Using the new direct electron detectors , Allegretti et al . were able to work out the structure of Frh in greater detail than before . The results confirm that the previously reported structure is correct . Furthermore , several new structural features were seen for the first time , including a previously unseen ion located between two protein subunits . Allegretti et al . also revealed that the structure of Frh is highly rigid , and so the process by which it catalyses reactions involving its substrate , the coenzyme F420 , does not involve changes in its shape . Instead , the reaction rate depends on the rate at which F420 diffuses to the correct position in the enzyme , where the reaction occurs very rapidly . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | Atomic model of the F420-reducing [NiFe] hydrogenase by electron cryo-microscopy using a direct electron detector |
Comprehensively elucidating the molecular mechanisms of human immunodeficiency virus type 1 ( HIV-1 ) latency is a priority to achieve a functional cure . As current 'shock' agents failed to efficiently reactivate the latent reservoir , it is important to discover new targets for developing more efficient latency-reversing agents ( LRAs ) . Here , we found that TRIM28 potently suppresses HIV-1 expression by utilizing both SUMO E3 ligase activity and epigenetic adaptor function . Through global site-specific SUMO-MS study and serial SUMOylation assays , we identified that P-TEFb catalytic subunit CDK9 is significantly SUMOylated by TRIM28 with SUMO4 . The Lys44 , Lys56 and Lys68 residues on CDK9 are SUMOylated by TRIM28 , which inhibits CDK9 kinase activity or prevents P-TEFb assembly by directly blocking the interaction between CDK9 and Cyclin T1 , subsequently inhibits viral transcription and contributes to HIV-1 latency . The manipulation of TRIM28 and its consequent SUMOylation pathway could be the target for developing LRAs .
Despite the suppressive combined antiretroviral therapy ( cART ) , the persistence of HIV-1 in the latent reservoirs is the major obstacle to achieve a cure ( Chun et al . , 1997; Finzi et al . , 1997; Wong et al . , 1997 ) . To completely eradicate the reservoir , it needs almost 73 . 4 years of cART due to its long half-life in resting CD4+ T cells ( Siliciano et al . , 2003 ) . Although over 200/106 resting CD4+ T cells contain proviruses , only 1/106 resting CD4+ T cells ( or 1/200 of them ) contain inducible replication-competent proviruses and 40/106 resting CD4+ T cells contain intact non-inducible proviruses ( Eriksson et al . , 2013; Ho et al . , 2013 ) . Most of the proviruses are defective , some of which can be induced to produce functional viral proteins and exposed to immunosurveillance ( Ho et al . , 2013; Pollack et al . , 2017 ) . Most of the integration sites locate in the intron of actively transcribed genes ( Schröder et al . , 2002 ) . Some integration hotspots were found in latently infected clonally expanded CD4+ T cells in HIV-1 patients on cART ( Cohn et al . , 2015; Maldarelli et al . , 2014; Wagner et al . , 2014 ) . To decrease the latent reservoirs , several functional cure strategies which are defined as a long-term control of HIV-1 replication and remission of the symptoms of HIV-1 infection without cART , have been proposed ( Katlama et al . , 2013 ) . The latently infected resting CD4 +T cells do not produce sufficient viral antigens which are recognized by immune system . Thus , the infected cells can hardly be eradicated . To this end , the ‘shock and kill’ strategy , which is one of the functional cure strategies , has been introduced and extensively performed these years . ( Deeks , 2012; Geng et al . , 2016b; Liu et al . , 2016; Liu et al . , 2015 ) . Based on the ‘shock and kill’ strategy , the inducible proviruses are ‘shocked’ out by latency reversing agents ( LRAs ) . Then the immune surveillance system recognizes and ‘kills’ these HIV-1-expressing cells utilizing various ways which include CTL response and antibody-dependent cell-mediated cytotoxicity ( ADCC ) . However , some infected cells harbor non-inducible proviruses which can hardly be reactivated by LRAs . Permanent silence of proviruses , accompanied by potent anti-HIV-1 immune surveillance , have been proposed as another strategy to inactivate proviruses in infected cells ( Gallo , 2016; Kessing et al . , 2017; Liu et al . , 2015; Mousseau et al . , 2012; Mousseau et al . , 2015; Shan et al . , 2012 ) . Further elucidating the mechanisms of HIV-1 latency will help us to better understand the formation and maintenance of viral reservoirs and develop new therapeutic interventions . Epigenetic regulations contribute to the establishment and maintenance of HIV-1 latency . Both histone deacetylases including HDAC1 and HDAC2 , and histone methyltransferases including G9a , Suv39H1 , GLP , EZH2 and SMYD2 , have been found to be responsible for ‘’writing’ repressive marks on HIV-1 long terminal repeat ( LTR ) ( Boehm et al . , 2017; Ding et al . , 2013; du Chéné et al . , 2007; Friedman et al . , 2011; Imai et al . , 2010; Marban et al . , 2007; Ruelas and Greene , 2013 ) . Suppressive epigenetic marks are further maintained by ‘reader’ proteins HP1γ and L3MBTL1 ( Boehm et al . , 2017; du Chéné et al . , 2007 ) . In addition , multiple miRNAs including miR-28 , miR-125b , miR-150 , miR-223 and miR-382 , and lncRNAs including NEAT1 and NRON , were also found to target viral RNAs and viral proteins to mediate transcriptional or posttranscriptional regulations of HIV-1 latency ( Huang et al . , 2007; Li et al . , 2016; Zhang et al . , 2013 ) . Apart from the above epigenetic mechanisms of HIV-1 latency , another barrier to successfully reactivate latent HIV-1 depends upon transcriptional control ( Mbonye and Karn , 2014 ) . In transcription initiation level , HIV-1 latency is contributed by both the insufficiency of transcription factors including NF-κB , Sp1 , AP-1 , NFAT and TFIIH , and the accumulation of transcription suppressors including LSF , YY1 and CTIP2 ( Mbonye and Karn , 2017 ) . For the escaped RNA Polymerase II ( RNAP II ) which passed through initiation , the absence of HIV-1 Tat and the presence of negative elongation factors NELF and DSIF facilitate promoter-proximal pausing of RNAP II on HIV-1 LTR ( Ping and Rana , 2001; Razooky et al . , 2015 ) . To further escape from promoter-proximal pausing and turn to transcriptional elongation , RNAP II must be extensively phosphorylated at Ser2 residues by positive transcription elongation factor b ( P-TEFb ) , which consists of cyclin-dependent kinase 9 ( CDK9 ) and Cyclin T1 ( Ott et al . , 2011 ) . However , the expression of Cyclin T1 is downregulated in latently infected cells ( Budhiraja et al . , 2013 ) . CDK9 is also inactive because of the dephosphorylation of its T-loop at Thr186 and sequestered in the 7SK small nuclear ribonucleoprotein ( snRNP ) complex by HEXIM1 or HEXIM2 ( Budhiraja et al . , 2013; Nguyen et al . , 2001; Yang et al . , 2001 ) . Another two studies indicate that CDK9 is acetylated at Lys44 by p300 to fully perform its kinase activity ( Cho et al . , 2010; Fu et al . , 2007 ) . Acetylation of Lys48 by GCN5 negatively regulates CDK9 activity ( Sabò et al . , 2008 ) . Although many work have unveiled the epigenetic and transcriptional mechanisms of HIV-1 latency , some important questions remain . For instance , there could be a versatile factor responsible for both mechanisms . The mechanism of promoter-proximal pausing has not been fully elucidated . In addition , how the P-TEFb is appropriately sequestered , released and targeted to HIV-1 promoter . More realistically , we have not yet found a powerful LRA which can efficiently reactivate the latent HIV-1 ( Spivak and Planelles , 2018 ) . To find more cellular factors as potential targets for LRAs , we designed and screened a custom siRNA library targeting multiple cellular epigenetic and non-epigenetic modification pathways in the nucleus . We found that a SUMOylation E3 ligase tripartite motif-containing protein 28 ( TRIM28 ) , also known as transcriptional intermediary factor 1β ( TIF1β ) and KAP1 ( KRAB-associated protein-1 ) , binds to CDK9 and mediates the SUMOylation of CDK9 , resulting in the disassociation of CDK9 with Cyclin T1 and the inhibition of CDK9 kinase activity . Consequently , its depletion significantly reactivates HIV-1 transcription and reverses HIV-1 latency .
To identify cellular targets which may contribute to HIV-1 suppression and latency , we started from the design and high-throughput screening of a custom siRNA library which targeted several cellular pathways within the nucleus including chromatin binding , epigenetic modification , chromatin remodeling , ubiquitination , SUMOylation , and chromosome organization ( Supplementary file 1 ) . We knocked down each gene in a TZM-bl cell line which harbors an integrated copy of luciferase under the control of HIV-1 promoter ( Platt et al . , 1998 ) . We found that many proteins restricted the activity of HIV-1 promoter based on the expression level of luciferase upon knockdown each target ( Figure 1A ) . The top hit proteins included HP1α , GLP , SUZ12 and CYLD , which have been identified to inhibit HIV-1 transcription ( Ding et al . , 2013; Khan et al . , 2018; Manganaro et al . , 2014 ) . Intriguingly , we found that knockdown of two less-defined SUMOylation pathway genes TRIM28 and SUMO4 significantly upregulated HIV-1 promoter activity ( Figure 1A , Figure 1—figure supplement 1A–B ) . The overexpression of TRIM28 inhibited the basal level of HIV-1 promoter activity and rescued HIV-1 repression in dose-dependent manner ( Figure 1—figure supplement 1C ) . The upregulation was more significant when combined with HIV-1 Tat and TNFα ( Figure 1—figure supplement 1D ) . We measured the expression of TRIM28 in different cells and found that TRIM28 is ubiquitously overexpressed in multiple cell lines and primary cells ( Figure 1—figure supplement 1E ) . As a complemental experiment to search for latency contributors , we compared gene expression in unstimulated and PHA-stimulated primary CD4+ T cells utilizing RNA-Seq ( Figure 1—figure supplement 1F ) . We found that TRIM28 was highly expressed in unstimulated primary CD4+ T cells and down regulated upon activation by PHA ( Figure 1—figure supplement 1G ) . The expression of TRIM28 was upregulated again when the activated CD4+ T cells returned to resting status ( Figure 1—figure supplement 1H , Figure 1—figure supplement 2 ) . To test whether TRIM28 contributes to HIV-1 latency , we knocked down TRIM28 in HIV-1 latency cell line J-Lat 10 . 6 and found that the depletion of TRIM28 upregulated HIV-1 expression ( Figure 1B–C ) ( Jordan et al . , 2003 ) . Besides , HIV-1 reactivation was enhanced much higher when supplemented with histone deacetylase ( HDAC ) inhibitor suberoylanilide hydroxamic acid ( SAHA , vorinostat ) or Bromodomain and Extra-Terminal ( BET ) domain inhibitor JQ-1 , both of which have been widely described as LRAs ( Spivak and Planelles , 2018 ) . These results were well repeated in other latency model cell lines including J-Lat 6 . 3 , 8 . 4 , 9 . 2 , and 15 . 4 ( Figure 1—figure supplement 3A–E ) . TRIM28 was previously identified to inhibit HIV-1 integration by recruiting HDAC1 to deacetylate HIV-1 integrase ( Allouch et al . , 2011 ) . However , its roles in the expression of integrated HIV-1 and HIV-1 latency have not been clearly elucidated . To this end , we performed chromatin immunoprecipitation ( ChIP ) assay of TRIM28 in TZM-bl and J-Lat 10 . 6 cell lines to examine its possible association with integrated HIV-1 DNA ( Supplementary file 2 ) . We found that TRIM28 was significantly enriched on HIV-1 LTR compared to the regions of host-provirus junction and viral coding region ( Figure 1D–E and Figure 1—figure supplement 3F ) . The enrichment of TRIM28 on HIV-1 LTR was not influenced by TNFα signaling ( Figure 1—figure supplement 3G ) . Because TRIM28 was identified as an epigenetic adaptor recruiting HP1 , SETDB1 and NuRD complex to maintain suppressive epigenetic environment , we then tested whether the depletion of TRIM28 would influence the epigenetic status of HIV-1 LTR ( Iyengar and Farnham , 2011 ) . We observed significant decrease of H3K9me2 and H3K9me3 , as well as significant increase of H3K4me3 and H3K9Ac after knocking down TRIM28 ( Figure 1F–I , Figure 1—figure supplement 3H–J ) . The depletion of TRIM28 also induced slight H3K27me3 downregulation ( Figure 1J , Figure 1—figure supplement 3K ) . These results indicate that TRIM28 suppresses HIV-1 expression and contributes to HIV-1 latency by manipulating suppressive epigenetic modifications . Having identified the suppressive epigenetic adaptor role of TRIM28 on HIV-1 latency , we next attempted to search for new mechanism ( s ) of TRIM28 by function-based mutation . TRIM28 is a muti-functional protein containing seven different domains ( Ivanov et al . , 2007 ) . The C-terminal bromodomain ( BR ) , which is SUMOylated by the adjacent plant homeodomain ( PHD ) , recruits SETDB1 and NuRD complex in a SUMOylation-dependent manner . The N-terminal tripartite motif RBCC region is composed of a RING finger domain ( RING ) , two B-box domians ( BB ) , and a coiled-coil domain ( CC ) . The RING of TRIM28 functions as an intermolecular SUMO E3 ligase , while PHD is important for the intramolecular SUMO E3 ligase activity ( Ivanov et al . , 2007; Liang et al . , 2011; Neo et al . , 2015 ) . We constructed different TRIM28 mutants by depleting each of the seven domains ( Figure 2A ) . Then we knocked down the endogenous TRIM28 with siRNA targeting 3’UTR of TRIM28 mRNA and supplied with the wild-type TRIM28 construct and the mutants , respectively . Reactivation of HIV-1 expression by the knockdown of endogenous TRIM28 was re-suppressed to the basal level by the wild-type TRIM28 overexpression ( Figure 2B ) . Theoretically , none of the HP1BD , NHD , or BR mutants , especially the mutant of PHD which harbor the intramolecular SUMO E3 ligase activity , was able to significantly rescue the suppression , but the results showed they did . Nevertheless , the mutant without RING or RBCC domains totally aborted the re-suppression , which might be due to the loss of the Krüppel-associated box domain zinc fingers ( KRAB-ZNFs ) binding ability . We tested a mutant containing only RBCC . Interestingly , it still resumed the suppression . We also tested whether the two E3 ligase domains contributed to the epigenetic suppression of HIV-1 promoter by knocking down endogenous TRIM28 , followed by the overexpression of wild type or mutated TRIM28 . The results showed that the wild-type TRIM28 was able to rescue the suppressive epigenetic marks H3K9me3 and H3K27me3 and suppress the active epigenetic mark H3K9Acetyl , however , the mutant without RING or PHD domain was only able to rescue partial of the suppressive marks ( Figure 2C–E , Figure 2—figure supplement 1 ) . As the RING within RBCC domain plays a key role for the intermolecular SUMO E3 ligase activity of TRIM28 , we therefore hypothesize that TRIM28 may utilize the RING domain to SUMOylate cellular protein ( s ) which is ( are ) vital for HIV-1 expression ( Liang et al . , 2011 ) . To identify candidate substrates SUMOylated by TRIM28 , we conducted a modified global site-specific SUMOylation Mass Spectrometry ( SUMO-MS ) ( Figure 3A ) . We generated SUMO1-Q92R , SUMO2-Q88R and SUMO4-Q88R mutants mimicking yeast SUMO Smt3 to enable efficient identification of SUMO-acceptor lysines by MS ( Supplementary file 3 ) and co-expressed the SUMO mutants with TRIM28 and SUMO E2 UBC9 followed by the enrichment of SUMO conjugated substrates ( Hendriks et al . , 2014 ) . To increase the coverage and mapping possibility of targeted proteins , we used SDS-PAGE to separate the enriched proteins and excised the entire gel lane into 16 slices which were subjected to separate in-gel digestions . The digested peptides were analyzed by nanoscale LC-MS/MS . Finally , we identified 1 , 329 SUMOyalted proteins at significance threshold below 10−7 ( Supplementary file 4 ) . Based on the STRING network analysis , the SUMOylated proteins exerted a large complex network at the interaction confidence of 0 . 7 ( Figure 3B ) . We further performed MCODE analysis on SUMOylated proteins and found that the STRING core network could be clustered into 12 subclusters with interconnectivity scores ranging from 14 to 96 ( Figure 3B , Figure 3—figure supplement 1A and Supplementary file 5 ) . Through Gene Ontology ( GO ) analysis , we found that cellular and metabolic processes were the top two biological processes which the SUMOylated proteins could be involved in ( Figure 3—figure supplement 1B and Supplementary file 6 ) . Most SUMOylated targets have the catalytic activity and DNA binding function . Many transferases and transcription factors were also among the SUMOylated candidates . We specifically clustered the transferases and transcription factors by k-means clustering and visualized with STRING analysis . Interestingly , we found that many candidates were pivotal for HIV-1 expression , such as JUN , JUNB , JUND , mTOR , STAT3 , Cyclin T1 ( CCNT1 ) and CDK9 ( Figure 3C ) . Especially , CDK9 and CCNT1 were also found in MCODE Cluster 8 ( Figure 3—figure supplement 1A ) . Recently , it has been identified that the SUMOylation of transcription factor STAT5 was inactivated by benzotriazoles , resulting in the reactivation of latent HIV-1 ( Bosque et al . , 2017 ) . SUMOylation may participate in transcription more generally . We further narrowed down the significance threshold below 10−8 to find the more extensively SUMOylated targets . CDK9 was still among the top protein candidates ( Supplementary file 7 ) . Then , we co-overexpressed SUMO system proteins ( SUMO1 , SUMO2 , SUMO4 , UBC9 and TRIM28 ) with 10 transcription factor candidates , respectively . Several transcription factors were SUMOylated , such as NFKB1A , RelA , CCNT1 , CDK9 , SKIP , MEN1 and JUN , which verified the reliability of our global site-specific SUMO-MS ( Figure 3D ) . Nevertheless , the SUMOylation signals were much more significant for CDK9 , which merited being further studied . To further verify that CDK9 is SUMOylated by TRIM28 , we conducted several in vivo and in vitro SUMOyaltion assays . In vertebrates , there are four well-studied SUMO paralogs , SUMO1 , SUMO2 , SUMO3 , and SUMO4 . Because SUMO2 and SUMO3 share highly sequence-homolog and have similar functions , they are often referred to as SUMO2/3 ( Cubeñas-Potts and Matunis , 2013 ) . It is worthy to note that the depletion of SUMO4 was able to upregulate the HIV-1 promoter activity more significantly than the depletion of the other SUMO paralogs in our siRNA library screening ( Figure 1A ) . The upregulation was more significant when combined with HIV-1 Tat , the phenomenon of which was similar as we observed for TRIM28 ( Figure 4A–B , Figure 4—figure supplement 1A ) . The knockdown or knockout of SUMO4 was able to reactivate latent pseudotyped HIV-1 in J-Lat 10 . 6 as well ( Figure 4C ) . SUMO4 is also ubiquitously overexpressed in multiple cell lines and primary CD4+ T cells ( Figure 4—figure supplement 1B ) . After PHA stimulation in primary CD4+ T cells , the expression of SUMO4 was downregulated ( Figure 1—figure supplement 1G , Figure 4—figure supplement 1C ) . The expression SUMO4 returned to basal level when activated primary CD4+ T cells re-entered to resting status ( Figure 4—figure supplement 1C ) . As the SUMOylation of TRIM28 and associated epigenetic modifiers participates in the regulation of epigenetic patterns , we next testified whether SUMO4 could influence the function of TRIM28 and the epigenetic status of HIV-1 promoter ( Iyengar and Farnham , 2011 ) . We found that more than half of TRIM28 was lost from HIV-1 LTR upon SUMO4 knockdown , which indicated that the enrichment of TRIM28 on HIV-1 LTR may be partially SUMOylation-dependent apart from the Krüppel-associated box domain zinc fingers ( KRAB-ZNFs ) –dependent binding ( Figure 4D , Figure 4—figure supplement 1D–H ) . We also found that H3K9me , H3K9me2 and H3K9me3 were significantly decreased on HIV-1 LTR in the absence of SUMO4 , as well as the H3K9 methylation ‘writer’ SETDB1 and ‘reader’ HP1α ( Figure 4E–G , Figure 4K–L ) . Moreover , we observed significant upregulation of H3K9acetyl and H3K4me3 and downregulation of HDAC1 , which was consistent with previous reports that TRIM28 recruited SETDB1 , HP1α and HDAC1 in a SUMOylation-dependent manner ( Figure 4H–I , Figure 4M ) ( Iyengar and Farnham , 2011 ) . Besides , we found that the H3K27me3 was also decreased on HIV-1 LTR upon SUMO4 knockdown ( Figure 4J ) . It is possible that some polycomb repressive complex 2 ( PRC2 ) components such as EZH2 and SUZ12 , the major ‘writers’ of H3K27me3 , may be SUMOylated by SUMO4 , resulting in the enhancement of modifier function . As SUMO4 was able to mediate HIV-1 suppression and latency , possibly through the epigenetic control of HIV-1 promoter , we next attempted to identify the underlying mechanism by investigating its role in TRIM28-mediated CDK9 SUMOylation . We co-overexpressed CDK9 with SUMO1 , SUMO2 and SUMO4 , respectively . We found that CDK9 was mainly SUMOylated with SUMO1 and SUMO4 ( Figure 4—figure supplement 1I ) . The SUMO4-CDK9 amount was much more abundant than the SUMO1-CDK9 amount . Besides , SUMO E3 ligase TRIM28 utilized more SUMO4 compared with SUMO1 and SUMO2 ( Figure 4—figure supplement 1J ) . After the supplement of TRIM28 , the SUMO-CDK9 amount turned to be more abundant . However , the SUMOyaltion did not increase if we only co-overexpressed CDK9 with TRIM28 but without SUMO E2 UBC9 , which indicated that TRIM28-mediated SUMOylation was UBC9-dependent ( Figure 5A ) . The SUMO-CDK9 amount was increased dose-dependently when the TRIM28 increased gradually ( Figure 5B ) . We then conducted in vitro SUMOylation assay . Only when SUMO4 , E1 SAE1/UBA2 , E2 UBC9 and TRIM28 were supplied into the SUMO conjugation reaction buffer together , was SUMO4 conjugated to CDK9 ( Figure 5C ) . After knocking down TRIM28 in HeLa cells , the SUMOylated CDK9 decreased ( Figure 5D ) . In our previous siRNA screening , we noticed that the absence of several SUMO-specific isopeptidases ( SENPs ) , which deSUMOylated substrates , prevented the expression of HIV-1 , especially SENP3 ( Figure 5—figure supplement 1A–B ) . We then co-overexpressed SENP3 with TRIM28 and found that SENP3 prevented TRIM28-mediated CDK9 SUMOylation ( Figure 5E ) . To investigate whether TRIM28-mediated SUMOylation of CDK9 by SUMO4 exist in primary CD4+ T cells , we firstly confirmed that the conjugation of SUMO4 to cellular proteins frequently occurs ( Figure 5—figure supplement 1C ) . We also immunoblotted the endogenous CDK9 in primary CD4+ T cells and found that a small portion of CDK9 was SUMOylated by SUMO4 ( Figure 5—figure supplement 1D ) . The SUMO4-SUMOylated endogenous CDK9 increased significantly after the overexpression of SUMOylation components including SUMO4 , UBC9 and TRIM28 ( Figure 5—figure supplement 1E ) . Taken together , our data indicates that TRIM28 mediates the conjugation of SUMO4 to CDK9 , which is reversed by SENP3 . To identify whether TRIM28 binds to CDK9 , we used the super-resolution continuous STochastic Optical Reconstruction Microscopy ( cSTORM ) to investigate the three dimensional ( 3D ) co-localization in the resolution of 20 nm . We found that TRIM28 existed in many small clusters and large bodies in the nucleus and co-localized with dotted SUMO4 ( Figure 6A , first panel ) . From amplified view and 3D-cSTORM , we found that SUMO4 proteins were enriched by TRIM28 and shaped big spots ( Figure 6A , second and third panels; Video 1 ) . Although CDK9 existed in dispersed dots all within the nucleus , we still found that CDK9 co-localized with TRIM28 ( Figure 6B , first panel ) . Similarly to SUMO4 , CDK9 proteins were enriched by and surrounded TRIM28 bodies ( Figure 6B , second and third panels; Video 2 ) . The lateral resolution of cSTORM imaging can be up to 20 nm and the axial resolution is 50 nm , which is within the range to distinguish protein complexes , even single protein molecules ( Lagache et al . , 2015 ) . Thus , we transformed the cSTORM-imaged protein molecules and complexes into small or large spots based on their diameter ( Figure 6C–D , Figure 6—figure supplement 1A–B , left panel; Video 3 , Video 4 ) . The direct interaction between spots and spots was measured in compliance with the criterion of maximal distance of 10 nm ( Figure 6C–D , Figure 6—figure supplement 1A–B , middle panel ) . The indirect interaction between complexes and spots was measured in compliance with the criterion of maximal distance of 100 nm ( Figure 6C–D , Figure 6—figure supplement 1A–B , right panel ) . Finally , we found that nearly 80% of TRIM28 spots or complexes were co-localized with 94% of SUMO4 spots ( Figure 6E ) . Similarly , 88% of TRIM28 spots or complexes were co-localized with 76% of CDK9 spots ( Figure 6E ) . Through co-immunoprecipitation ( Co-IP ) assay , we found that CDK9 bound to TRIM28 , even in the presence of RNase ( Figure 7—figure supplement 1A ) . To identify which region of TRIM28 bound to CDK9 , we examined various TRIM28 deletion mutants to enrich CDK9 . The depletion of RING aborted the binding of CDK9 as well as the SUMOylation of CDK9 ( Figure 7A–B ) . Further , we co-transfected GFP-TRIM28 and several GFP-TRIM28 mutants with RFP-CDK9 in HEK293T cells and utilized the super-resolution Structured Illumination Microscopy ( SIM ) to investigate the co-localization . Exogenously expressed TRIM28 also co-localized with CDK9 with Pearson’s coefficient of 0 . 7336 and thresholded Mander’s coefficient of 0 . 5846 , which indicated a highly co-localization . However , the mutant of RING domain deletion was not capable ( Figure 7C–D ) . We also inspected the SUMOylation status of each TRIM28 mutants and found that all the mutants was SUMOylated , which coincided with previous reports that both the RING and PHD had the E3 ligase activity and enriched UBC9 ( Figure 7—figure supplement 1B ) ( Ivanov et al . , 2007; Liang et al . , 2011 ) . Collectively , our results indicate that TRIM28 binds to CDK9 and SUMOylates CDK9 through its RING domain . After confirming CDK9 is indeed SUMOylated by TRIM28 , and also because the RBCC domain contributes to HIV-1 suppression , we next tried to examine whether the function of CDK9 is influenced by TRIM28-mediated SUMOylation . We firstly utilized ATAC-Seq to probe the chromatin accessibility of HIV-1 promoter upon TRIM28 elimination . We found that most of the increased accessible regions across the genome lied on promoters and distal intergenic regions upon the depletion of TRIM28 in J-Lat 10 . 6 or TZM-bl cell lines ( Figure 8—figure supplement 1A–B ) . Through GO analysis and Clusters of Orthologous Groups of proteins ( COGs ) analysis , we found that the chromatin accessibility variation happened in genes related to various biological processes and cellular components upon TRIM28 depletion ( Figure 8—figure supplement 1C–D ) . Most of the influenced general functional genes had the DNA or protein-binding abilities and catalytic activities ( Figure 8—figure supplement 1C–D , Figure 8—figure supplement 2A–B ) . To inspect whether the chromatin accessibility of HIV-1 genome was influenced upon TRIM28 depletion , we separately aligned the sequencing reads to HIV-1 reference genome . We found that the accessible region indicated by transposable tag density increased on HIV-1 LTR when TRIM28 was knocked out from J-lat 10 . 6 cell lines , as well as when TRIM28 was knocked down in TZM-bl cell lines , which indicated significantly enhanced promoter activity ( Figure 8A–B ) . The promoters of genes within which the integrated pseudotyped HIV-1 or HIV-1 reporter provirus located and housekeeping gene GAPDH were not influenced ( Figure 8—figure supplement 2C–F ) . Alternatively , we also observed significant enrichment of CDK9 and Ser2 super-phosphorylated RNAP II on HIV-1 LTR upon the knockdown of either TRIM28 or SUMO4 , which was in agreement with the results that the depletion of TRIM28 or SUMO4 reactivated HIV-1 expression ( Figure 8C–D ) . Interestingly , through Co-IP assay , we found that Cyclin T1 only bound to wild-type CDK9 to form P-TEFb complex , not the SUMOylated CDK9 ( Figure 8E ) . To investigate whether TRIM28-mediated SUMOylation of CDK9 affects the kinase activity of CDK9 , we conducted in vitro CDK9 SUMOylation assay followed by CDK9 kinase assay ( Figure 8—figure supplement 3A ) . We found that the kinase activity of CDK9 significantly decreased when SUMOylated by TRIM28 . However , the kinase activity of CDK9 was not influenced without the addition of TRIM28 , although the other SUMOylation components have been added ( Figure 8F ) . Collectively , TRIM28-mediated SUMOylation impairs both the binding ability of CDK9 to Cyclin T1 and the kinase activity of CDK9 to RNAP II , resulting in the dysfunction of transcription elongation . To elucidate the mechanisms that SUMOylation weakens the interaction between CDK9 and Cyclin T and the CDK9 kinase activity , we next attempted to identify the CDK9 SUMOylation sites which should occur on lysine residues . In order to narrow down the search scope , we equally grouped the sequence of CDK9 into three parts . Each part was given a mutant version that all the lysines were mutated to arginines . Then , we combined these six sequences and obtained eight constructs including the wild type CDK9 ( Figure 9—figure supplement 1A and Supplementary file 3 ) . The construct named CDK9-K0R , which contained the mutation that all the lysines were changed to arginines , totally aborted the capability of CDK9 to be SUMOylated ( Figure 9—figure supplement 1B ) . However , the other CDK9 mutants still were able to be SUMOylated by TRIM28 , which indicated that multiple SUMOylation sites might exist across the whole CDK9 sequence . To locate all the suspicious SUMOylation sites , we adopted reversing mutation strategy based on CDK9-K0R construct . Each of the 29 arginines of CDK9-K0R was mutated back to lysine separately ( Supplementary file 3 ) . Finally , we found that several lysines on CDK9 were significantly SUMOylated ( Figure 9A ) . Among them , multiple SUMOylation sites were adjacent to CDK9 C-terminal autophosphorylation sites which have been reported to be required for high-affinity binding of Tat–P-TEFb to TAR RNA ( Baumli et al . , 2008; Garber et al . , 2000 ) . SUMOylation may decrease the binding ability through preventing the neighboring phosphorylation . It was notable that , although the remove of endogenous TRIM28 significant downregulated SUMO-CDK9 , slightly residual SUMO-CDK9 still occurred , implying that other CDK9 SUMOylation E3 ligases may exists and some of the SUMOylation sites are not the TRIM28 targets ( Figure 5D ) . To further identify which sites are indeed SUMOylated by TRIM28 only , we knocked down the endogenous TRIM28 and tested the SUMOylation potential of the candidate sites identified above . We found that the SUMOylation signals of Lys44 , Lys56 and Lys68 totally disappeared in the absence of endogenous TRIM28 , further supporting that these sites are specifically SUMOylated by TRIM28 ( Figure 9B ) . The target-specific SUMO-MS for directly analyzing the enriched SUMO-CDK9 also confirmed this result ( Figure 9—figure supplement 1C–E ) . As the acetylation of Lys44 is required for its kinase activity , it is not surprising that the kinase activity of CDK9 was weakened when CDK9 was SUMOylated ( Cho et al . , 2010; Fu et al . , 2007 ) . Interestingly , other two SUMOylated sites Lys56 and Lys68 are within the interaction region of CDK9 and Cyclin T1 based on the co-crystal structure ( PDB ID: 4EC8 ) ( Baumli et al . , 2012 ) ( Figure 9C ) . Because SUMO protein is a polypeptide macromolecule , its presence can form steric hindrance which prevents the formation of P-TEFb complex . To verify whether TRIM28 could be a safe target for developing new LRAs , we firstly evaluated the possible toxicities associated with depleting TRIM28 in Hela cells , Jurkat cells as well as resting CD4+ T cells isolated from aviremic participants . We conducted several experiments which included cytotoxicity assay , cell viability assay , cell number counting and cell proliferation assay . The results showed that the depletion of TRIM28 was non-toxic to cell viability and proliferation ( Figure 10—figure supplement 1 , Figure 10—figure supplement 2 ) . Afterwards , we tried to determine whether the knockdown of TRIM28 reactivated latent HIV-1 in resting CD4+ T cells from HIV-1-infected individuals who received suppressive cART for at least 6 months . Stimulation with αCD3/αCD28 significantly induced the expression of HIV-1 based on the quantitation of intracellular HIV-1 RNAs ( Figure 10A and Figure 10—figure supplement 3A–B ) . The depletion of TRIM28 reactivated similar amount of HIV-1 RNA as suberanilohydroxamic acid ( SAHA ) . After we combined the knockdown of TRIM28 with SAHA , the reactivation was more significant ( Figure 10A ) . To provide evidence that SUMO4-mediated modification of CDK9 by TRIM28 is one of the mechanisms used by TRIM28 to contribute to HIV-1 latency in cells isolated from aviremic participants , we also tested whether the depletion of SUMO4 could reactivate latent HIV-1 in resting CD4+ T cells isolated from HIV-1-infected individuals . The results showed that the depletion of SUMO4 reactivated substantial productions of HIV-1 RNAs which were even slightly higher than those activated by SAHA . The combination use of SUMO4 knockdown and SAHA addition could reactivate more HIV-1 RNAs than those reactivated by them separately ( Figure 10—figure supplement 4 ) . We next examined whether the knockdown of TRIM28 reactivated more genetically-diversified HIV-1 , as we described previously ( Figure 10—figure supplement 3A ) ( Geng et al . , 2016b ) . Although TRIM28 depletion alone reactivated similar amount of genetically diversified HIV-1 with SAHA , the combination of TRIM28 knockdown and SAHA reactivated much more genetically-diversified HIV-1 ( Figure 10B ) . To determine whether the reactivated HIV-1 was replication-competent , we co-cultured the PHA-stimulated , SAHA-induced , or TRIM28-deficient resting CD4 +T cells from HIV-1-infected individuals , with PHA-activated CD4 +T cells from heathy donors ( Figure 10—figure supplement 3A ) . The accumulating production of p24 antigen indicates the reactivated HIV-1 viral particles were replication-competent . The knockdown of TRIM28 reactivated replication-competent viruses in all the three samples ( Figure 10C and Figure 10—figure supplement 3C ) . Similarly , the combination of SAHA with TRIM28 knockdown reactivated more replication-competent viral particles . These results indicate that TRIM28 contributes to HIV-1 latency in HIV-1-infected individuals . Targeting TRIM28 is well-tolerated for HIV-1-infected CD4+ T cells .
Post-translational modifications of CDK9 have been studied extensively , most of which focus on phosphorylation and acetylation ( Cho et al . , 2010 ) . Interestingly , many CDK9 SUMOylation sites which we identified here are highly related to phosphorylation and acetylation . The acetylation of Lys44 is vital for CDK9 phosphorylation activity on RNAP II . The SUMOylation of Lys44 masks the kinase activity . The acetylated Lys44 can also be deacetylated by NuRD complex which recruited by TRIM28 . Although multiple sites on CDK9 can be SUMOylated by TRIM28 , the percentage of SUMOylated CDK9 is only a small proportion ( less than 5% ) . This phenomenon has been observed for most of the identified SUMOylation targets ( Gareau and Lima , 2010; Impens et al . , 2014 ) . How the small portion triggers extensive effect on target substrate remains a mystery . Two models have been proposed to explain the small fraction of SUMOylation mediated transcriptional suppression , respectively ( Hay , 2005; Johnson , 2004 ) . Both models suggest transcriptional suppression is initiated by SUMOylation . However , the maintenance of suppression is SUMOylation-independent . In our co-localization experiment , we found that CDK9 is extensively recruited to the sub-compartment shaped by TRIM28 , although the SUMOylated CDK9 is only a small proportion based on the western blotting data . We propose that SUMOylation is a transient signal for CDK9 to enter to silent status or silent complex . The SUMOylated CDK9 may recruit other suppressive modifiers to stabilize the suppressive complex . After the remove of SUMO peptide by ubiquitous SENPs , CDK9 might be still sequestered in the suppressive complex . In recent years , TRIM28 was identified to form a large repressive complex with other epigenetic silencing complex such as the human silencing hub ( HUSH ) complex which also recruits SETDB1 to HIV-1 LTR to maintain H3K9me3 ( Robbez-Masson et al . , 2018; Tchasovnikarova et al . , 2015 ) . In rapid growing cells , 90% of P-TEFb is sequestered in suppressive complex 7SK snRNP ( Zhou et al . , 2012 ) . Whether TRIM28 is part of 7SK snRNP and whether TRIM28 complex shares overlap with 7SK snRNP or other CDK9 suppressive complexes in primary CD4+ T cells need to be further elucidated . TRIM28 has previously been found to stabilize the RNAP II promoter-proximal pausing ( Bunch et al . , 2014 ) . However , the detailed mechanism is largely unknown . Our findings here could potentially explain this phenomenon . The largest barrier for RNAP II to escape from transcriptional-pausing to effective elongation is the recruitment of P-TEFb to super-phosphorylate RNAP II . TRIM28 is bound to upstream of transcription start sites ( TSSs ) and SUMOylates the invaded CDK9 , resulting in the disconnection of CDK9 with Cyclin T1 and inhibition of CDK9 kinase activity . This hypothesis is also consistent with our finding that the depletion of TRIM28 or SUMO4 induces more significant HIV-1 expression when combining the use of HIV-1 Tat . Without the constraint of TRIM28-mediated CDK9 SUMO4-SUMOylation , HIV-1 Tat utilizes more functional CDK9 to facilitate RNAP II on transcribing HIV-1 RNA . Another mechanisms which TRIM28 may manipulate is TRIM28-mediated suppressive epigenetic modifications on nucleosomes downstream of RNAP II pausing sites , which further stabilizes transcriptional-pausing . One report showed that SENP3 deSUMOylates RbBP5 , one of the subunits of MLL1/MLL2 complexes , resulting in the complexes stabilization , H3K4me3 accumulation and RNAP II recruitment ( Nayak et al . , 2014 ) . We found that SENP3 prevents TRIM28-mediated CDK9 SUMOylation , which facilitates the transcriptional-pausing release of recruited RNAP II . More work needs to further identify the upstream signaling pathway which determines when to release TRIM28-mediated transcriptional-pausing of RNAP II on HIV-1 LTR . Until now , nearly all the shock agents have failed to decrease the latent HIV-1 reservoir based on several clinical trials ( Spivak and Planelles , 2018 ) . The only effective LRAs across multiple latency model cell lines and ex vivo patient cells are protein kinase C ( PKC ) agonists ( Bullen et al . , 2014 ) . However , PKC agonists induce some degree of T cell activation which is toxic to global T cells . Several lines of evidence have shown that both epigenetic regulation and transcriptional control are two barriers which we need to overcome when we develop novel LRAs ( Mbonye and Karn , 2017 ) . Interestingly , we found that TRIM28 bridges both suppressive epigenetic modifications and RNAP II transcriptional-pausing to contribute to HIV-1 latency . Besides , LRAs which target the SUMOylation of transcription factor result in the reactivation of latent HIV-1 ( Bosque et al . , 2017 ) . TRIM28-mediated RNAP II transcriptional-pausing on HIV-1 promoter is also SUMOylation-dependent as we have elucidated extensively above . Developing next-generation LRAs targeting TRIM28 may release both epigenetic and transcriptional restrictions , which also provides a new direction to search dual-function candidates .
Chronically HIV-1-infected participants sampled by this study were recruited from Department of Infectious Diseases in Guangzhou 8th People’s Hospital , Guangzhou . The Ethics Review Board of Sun Yat-Sen University and the Ethics Review Board of Guangzhou 8th People’s Hospital approved this study . All the participants were given written informed consent with approval of the Ethics Committees . The enrollment of HIV-1-infected individuals was based on the criteria of prolonged suppression of plasma HIV-1 viremia on cART , which is undetectable plasma HIV-1 RNA levels ( less than 50 copies/ml ) for a minimum of 6 months , and having high CD4+ T cell count ( at least 350 cells/mm3 ) . Blood samples from healthy individuals were obtained from Guangzhou Blood Center . We did not have any interaction with the healthy individuals or protected information , and therefore no informed consent was required . HEK293T ( CVCL_0063 ) and HeLa ( CVCL_0030 ) cells which were obtained from ATCC , and TZM-bl ( 8129 ) cells , which were obtained from NIH AIDS Reagent Program , were cultured in DMEM supplemented with 1% penicillin-streptomycin ( ThermoFisher ) , 1% L-glutamine ( ThermoFisher ) , and 10% FBS ( ThermoFisher ) . J-Lat 6 . 3 , 8 . 4 , 9 . 2 , 10 . 6 and 15 . 4 cell lines , which were originally generated from Dr . Eric Verdin ( The Buck Institute for Research on Aging , Novato , CA ) Laboratory , were obtained from Dr . Robert F . Siliciano ( Department of Medicine , Johns Hopkins University School of Medicine , Baltimore , MD ) Laboratory . All the J-Lat cell lines were cultured in RPMI 1640 supplemented with 1% penicillin-streptomycin , 1% L-glutamine , and 10% FBS . Peripheral blood mononuclear cells ( PBMCs ) and primary CD4+ T cells , which were isolated and purified from study participants , were cultured in RPMI 1640 supplemented with 1% penicillin-streptomycin , 1% L-glutamine , and 10% FBS . 1/1000 Recombinant human interleukin 2 ( IL-2 ) ( R and D ) was supplied for primary CD4+ T cells to maintain proliferation . All cells have been tested for mycoplasma using a PCR assay and confirmed to be mycoplasma-free . All cells cultured in sterile incubator at 37°C and 5% CO2 . SiRNA library targeting 182 human genes , negative control siRNA ( siNC ) and siRNA targeting TRIM28 3’UTR ( 5’-GCTCTGTTCTCTGTCCTGT-3’ ) were purchased from RiboBio ( Guangzhou , China ) ( Supplementary file 1 ) . Three siRNAs were synthesized for each gene . The siRNAs targeting each gene were transfected as a mixture and have been validated by company to insure that at least one siRNA was able to knock down target gene mRNA up to 70% . The siRNA library covered six cellular pathways within the nucleus , which were chromatin binding , epigenetic modification , chromatin remodeling , ubiquitination , SUMOylation , and chromosome organization . Evenly mixed TZM-bl cell suspension was added into each well of 96-well plates with a Tecan Freedom EVO150 ( Tecan , Männedorf , Schweiz ) to insure that the cell confluency was 60% when the cells were transfected . Twelve hours post-seeding , cells from each well were transfected with siRNAs targeting each gene using Lipofectamine RNAiMAX ( ThermoFisher ) according to the manufacturer’s instruction . Each gene was set three biological replicates . Forty-eight hours post-transfection , cell samples from each well were removed culture medium and washed twice with PBS . Fifty microliter passive lysis buffer ( Promega ) was added into each well and lysed for 30 min with shaking . The cell lysates were clarified with centrifugation at 12 , 000 g for 3 min . Luciferase in the cell lysates was measured with luciferase-reporter assay ( Promega ) using a multiwell plate luminometer with an auto-injector ( Promega ) and analyzed by GloMax 96 Microplate Luminometer Software ( Promega ) . Fold changes were calculated for each gene compared with siNC according to the light units . ShRNA targeting luciferase ( shluc: 5’-ACCGCCTGAAGTCTCTGATTAA-3’ ) was set as negative control ( Rousseaux et al . , 2018 ) . The shRNA target sequence against TRIM28 CDS was 5’-CCAGCCAACCAGCGGAAATGTGA-3’ ( Ivanov et al . , 2007 ) . Target sequences were cloned into pLKO . 3G-RFP which was derived from pLKO . 3G . The GFP-tag was replaced with RFP-tag in pLKO . 3G-RFP . Pseudotyped viral stocks were produced in HEK293T cells by co-transfecting 3 μg of VSV-G glycoprotein-expression vector , 6 μg of lentiviral packaging construct pCMVΔR8 . 2 , and 6 μg shRNA-expression lentiviral construct using Lipofectamine 2000 ( ThermoFisher ) according to the manufacturer’s instruction . VSV-G glycoprotein-expression vector was abtained from Addgene ( Addgene plasmid # 12259 ) . pCMVΔR8 . 2 was a kindly gift from Dr . Didier Trono ( School of Life Sciences , Ecole Polytechnique Fédérale de Lausanne , Lausanne , Switzerland ) ( Zufferey et al . , 1997 ) . Virus supernatants from each 10 cm dish were concentrated into 1 ml RPMI 1640 by PEG 6000 . J-Lat 6 . 3 , 8 . 4 , 9 . 2 , 10 . 6 and 15 . 4 cell lines were spin-infected with shRNA virus . Forty-eight hours later , infected cells were treated with 500 nM SAHA ( Selleckchem ) or 1 μM JQ-1 ( Selleckchem ) . Another 24 hr later , the percentages of GFP positive cells from each group were determined by BD LSRFortessa cell analyzer ( BD Biosciences ) and analyzed by FlowJo V10 ( Tree Star ) . The infection efficiency was measured based on the percentage of RFP-positive cells using flow cytometry . The knockdown efficiency was confirmed by both qPCR and western blot . For knocking out TRIM28 , CRISPR-CAS9 system was used . SgRNA targeting dummyguide ( sgNT: 5’-ACGGAGGCTAAGCGTCGCAA-3’ ) was set as negative control ( Sanjana et al . , 2014 ) . The sgRNA target sequence against TRIM28 CDS was 5’-CACCGATTGAGCTGGCAGTCTCGGC-3’ ( Sanjana et al . , 2014 ) . Target sequences were cloned into lentiCRISPRv2 ( Sanjana et al . , 2014 ) . Pseudotyped viruses were produced and concentrated as shRNA viruses . J-Lat 10 . 6 cells were spin-infected with sgRNA virus and cultured for 48 hr followed by puromycin ( Sigma-Aldrich ) selection . Three days post-selection , the supernatant of infected cells was replaced with fresh RPMI 1640 and infected cells were went on culturing for 2 to 7 days . The knockout efficiency was confirmed both western blot . The percentages of GFP-positive cells were determined by flow cytometry . Chromatin immunoprecipitation ( ChIP ) was performed according to the manufacturer’s instruction ( CST ) . Approximately 4 × 106 cells were prepared for each immunoprecipitation ( IP ) . Briefly , TZM-bl cells were treated with siNC , siTRIM28 or TNFα ( PeproTech ) for 48 hr followed by crosslinking proteins to DNA with 1% formaldehyde ( Sigma-Aldrich ) for 10 min at room temperature . The fixation was quenched with 125 mM glycine for 5 min at room temperature followed by centrifuging at 1 , 500 rpm for 5 min at 4°C . The supernatants were removed immediately . Cell pellets were resuspended in ice-cold Buffer A ( CST ) supplemented with DTT and Protease inhibitor cocktail ( PIC ) and incubated on ice for 10 min . The nuclei were enriched by centrifugation at 3000 rpm for 5 min at 4°C and resuspended in ice-cold Buffer B ( CST ) supplemented with DTT . Nuclei pellets were centrifuged again , removed supernatants and resuspended in 100 μl Buffer B supplemented with DTT and 0 . 5 μl micrococcal nuclease ( CST ) per IP preparation . The digestion was conducted at 37°C for 20 min . Incubation tubes were inverted several times per 5 min . After digestion , the reaction was stopped by adding 50 mM EDTA followed by centrifugation at 13 , 000 rpm for 1 min at 4°C . Nuclei pellet was resuspended in 100 μl ChIP Buffer ( CST ) supplemented with PIC per IP preparation and incubated for 10 min on ice . The nuclei pellet was further lysed by sonication with 3 sets of 20 s pulses at 40% amplitude . Pellet was incubated on ice for 30 s between pulses . The lysates were clarified by centrifugation at 10 , 000 rpm for 10 min at 4°C . The supernatants which contained digested chromatin were transferred into new tube . One-tenth of the chromatin sample was proceeded to analyze the size and concentration . Briefly , 50 μl chromatin sample was removed RNA by RNase A ( CST ) and reversed cross-linking by 200 mM NaCl and Proteinase K ( CST ) . DNA from samples were purified by DNA purification spin columns ( CST ) . Concentration was determined by measuring OD260 . The size range was analyzed by electrophoresis on a 1% agarose gel , which should be between 150 and 900 bp . For each IP preparation , approximately 10 μg chromatin was diluted into ChIP Buffer . Ten microliter diluted chromatin , which was 2% input sample , was transferred to a new tube and stored at −20°C . Immunoprecipitation antibodies normal rabbit IgG ( CST , 2729 ) , anti-TRIM28 antibody ( Proteintech , 15202–1-AP ) , anti-H3K9me2 antibody ( Abcam , ab1220 ) , anti-H3K9me3 antibody ( Abcam , ab8898 ) , anti-H3K4me3 antibody ( Abcam , ab8580 ) , anti-H3K27me3 antibody ( Abcam , ab6002 ) , anti-H3K9Acetyl antibody ( Abcam , ab4441 ) , anti-CDK9 antibody ( CST , 2316 ) , and anti-RNA polymerase II CTD repeat YSPTSPS ( phospho Ser2 ) antibody ( Abcam , ab5095 ) were separately added to siNC and siTRIM28 groups , respectively . The immunoprecipitation was carried out overnight at 4°C while rotating . ChIP-Grade Protein G Magnetic Beads ( CST ) were added to the each IP reaction and incubated with IP samples for another 2 hr at 4°Cwhile rotating . The protein G magnetic beads were pelleted by placing the IP tubes in a magnetic separation rack and washed with 3 times low-salt washes and one time high-salt wash . Each wash was conducted at 4°C for 5 min while rotating . DNA enriched by protein G magnetic beads was eluted by ChIP Elution Buffer ( CST ) . All the DNA samples including 2% input samples were reversed cross-linking with 200 mM NaCl and Proteinase K and purified as above . ChIP primers targeting the HIV-1 mini-model in TZM-bl cell line were used to quantitate each target by Real-Time Quantitative PCR . The quantitation regions were shown below . G5: Cellular DNA and viral 5’LTR junction; A: Nucleosome 0 assembly site; B: Nucleosome free region; C: Nucleosome one assembly site; V5: Viral 5’LTR and gag leader sequence junction; L: Luciferase region; V3: Viral poly purine tract and 3’LTR junction; G3: Viral 3’LTR and cellular DNA junction . Primers which amplified each region were shown in Supplementary file 2 . All the ChIP-qPCR DNA signals were normalized to siNC IgG of G5 . ChIP-qPCR in J-Lat 10 . 6 cell line was conducted as in TZM-bl cell line . In J-Lat 10 . 6 , G5’ represented cellular DNA and viral 5’LTR junction; E represented envelop; G3’ represented viral 3’LTR and cellular DNA junction; A , B , C , V5 and V3 represented as in Figure 1D . The identities of unstimulated primary CD4+ T cells , PHA-stimulated primary CD4+ T cells and resting CD4+ T cells were confirmed by flow cytometry with antibodies against human CD4 ( ThermoFisher , 11-0048-42 ) , CD45RA ( BioLegend , 304127 ) , CD45RO ( BD Biosciences , 560607 ) , CD62L ( ThermoFisher , 25-0629-42 ) , CD69 ( ThermoFisher , 25-0699-42 ) and CD25 ( ThermoFisher , 15-0259-42 ) . RNAs from indicated numbers of cells were isolated with TRIzol reagent ( ThermoFisher ) and proceeded to cDNA synthesis with PrimeScript RT reagent Kit ( Takara ) . For the samples which quantitated the expression of TRIM28 , Real-time PCR was performed with SYBR Ex-taq premix ( Takara ) in a CFX96 Real-time PCR Detection System ( Bio-Rad ) . Human β-actin mRNA was measured as internal control ( Li et al . , 2016 ) . Primer pairs were shown as below: β-actin qPCR Forward Primer: 5’-GCATGGAGTCCTGTGGCA-3’ , β-actin qPCR Reverse Primer: 5’-CAGGAGGAGCAATGATCTTGA-3’; TRIM28 qPCR Forward Primer: 5’-CTACTCAAGTGCAGAGCCCC-3’ , TRIM28 qPCR Reverse Primer: 5’-GGGAAGACCTTGAAGACGGG-3’ . The relative expression of each gene was calculated as 2[Ct ( Control-TRIM28 ) -Ct ( Control-β-Actin ) ]-[Ct ( Treatment-TRIM28 ) -Ct ( Treatment-β-Actin ) ] . For the quantitation of HIV-1 expression , a specific reverse primer was used to reversely transcribe HIV-1 RNA: 5’- GCTTCAGCAAGCCGAGTCCTGCGTC-3’ . QPCR was performed for specific reverse-transcribed HIV-1 cDNA with primer pairs: HIVTotRNA Forward Primer: 5’-CTGGCTAACTAGGGAACCCACTGCT-3’ and HIVTotRNA Reverse Primer: 5’-GCTTCAGCAAGCCGAGTCCTGCGTC-3’ ( Liu et al . , 2016 ) . After quantitation , an in vitro transcribed HIV-1 RNA was used as the external control for measuring cell-associated viral RNAs . The Ct of each group was converted to mass and further converted to copies . The final expression of intracellular HIV-1 RNA was represented as 103 copies viral RNA per million CD4 +T cells . His-tagged SUMO mutants SUMO1-Q92R , SUMO2-Q88R and SUMO4-Q88R were co-overexpressed with E2 UBC9 and E3 TRIM28 in HeLa cells . Forty-eight hours post-transfection , cell pellets were lysed by guanidine lysis buffer ( 6 M guanidine-HCl , 100 mM sodium phosphate , and 10 mM Tris , buffered at pH 8 . 0 ) . Lysates were sonicated for 15 s with 5 s pulse at a power of 30 W . Subsequently , prewashed anti-His Ni-NTA agarose beads ( QIAGEN ) , 50 mM imidazole and 5 mM β-mercaptoethanol were added into the lysates and tumbled overnight at 4C . After overnight incubation , beads were centrifuged at 500 r . c . f . and washed for 30 min at 4C with the following wash buffers in order: wash buffer A ( 6 M guanidine-HCl , 0 . 1% Triton X-100 , 10 mM imidazole , 5 mM β-mercaptoethanol , 100 mM sodium phosphate , and 10 mM Tris , buffered at pH 8 . 0 ) , wash buffer B ( 8 M urea , 0 . 1% Triton X-100 , 10 mM imidazole , 5 mM β-mercaptoethanol , 100 mM sodium phosphate , and 10 mM Tris , buffered at pH 8 . 0 ) , wash buffer C ( 8 M urea , 10 mM imidazole , 5 mM β-mercaptoethanol , 100 mM sodium phosphate , and 10 mM Tris , buffered at pH 6 . 3 ) , wash buffer D ( 8 M urea , 5 mM β-mercaptoethanol , 100 mM sodium phosphate , and 10 mM Tris , buffered at pH 6 . 3 ) , and wash buffer E ( same as wash buffer D ) . After washing , proteins were eluted three times from beads with elution buffer ( 7 M urea , 500 mM imidazole , 100 mM sodium phosphate , and 10 mM Tris , buffered at pH 7 . 0 ) for 30 min at 4C . All the eluates were combined together and filtered with 0 . 45 μm filter ( Millipore ) . The clarified proteins were concentrated with a 10 kDa-cutoff filter ( Millipore ) and washed with PBS for three times . Concentrated proteins were transferred to new tubes and boiled with 4 × protein SDS-PAGE loading buffer ( Takara ) at 100C for 15 min . Samples were separated with 4–12% protein gel ( ThermoFisher ) . The gel was dyed with silver stain kit ( Sigma-Aldrich ) . Sixteen gel slices were cut out and proceeded to in-gel digestion . Briefly , gel slices were destained and treated with 10 mM DTT followed by the treatment of 55 mM iodoacetamide . The gels were washed with 25 mM NH4HCO3 and 25 mM NH4HCO3 in 50% ACN followed by desiccation with vacuum . One hundred nanogram trypsin ( ThermoFisher ) which was dissolved in 25 mM NH4HCO3 was added to each gel and incubated overnight at 37C . Twenty four hours later , digested peptides were extracted with the following extraction solutions in order: 50% ACN containing 5%TFA , 75% ACN containing 0 . 1% TFA , and 100% ACN . The extracts were subjected to vacuum for 3 hr to remove the solvent . The peptides were desalted and enriched by C18 ZipTip ( Millipore ) , and redissolved in 50% ACN containing 0 . 1% TFA , followed by vacuum to remove the solvent . Twelve microliter of 0 . 01% formic acid was used to resolve the peptides and proceeded to nanoscale LC-MS/MS with an EASY-nLC system ( ThermoFisher ) connected to a Q-Exactive ( ThermoFisher ) with higher collisional dissociation ( HCD ) fragmentation . Peptide were separated by 20-cm-long analytical columns ( ID 75 μm , Polymicro Avantes ) packed in house with Luna 3 . 0u C18 ( 2 ) 100A ( Phenomenex ) with a 90-min gradient from 3% to 90% acetonitrile in 0 . 1% formic acid and a flow rate of 300 nL/min . Data-dependent acquisition mode with a top-ten method was used to operate the mass spectrometer . Full-scan MS spectra were obtained with a target value of 3E6 , a resolution of 70 , 000 , with a scan range from 300 to 1 , 800 m/z . HCD tandem MS/MS spectra were obtained with a target value of 1E6 , a resolution of 17 , 500 , and a normalized collision energy of 25% . Unknown charges , or charges lower than two and higher than eight were rejected . To confirm the SUMOylation sites on CDK9 by SUMO-MS , two different tagged SUMO4 mutants were used to co-overexpressed with HA-tagged CDK9 , respectively , which were Flag-tagged SUMO4-Q88R and His-tagged SUMO4-Q88R . Anti-HA-tag beads ( Sigma-Aldrich ) were used to immunoprecipitate CDK9 and corresponding SUMO-CDK9 . Enriched target proteins were eluted from beads by boiling with 4 × protein SDS PAGE loading buffer at 100°C for 15 min . The supernatants containing target proteins were transferred to new tubes after centrifugation at 12 , 000 rpm for 3 min . One part of the samples was proceeded to western blot with antibodies against HA-tag , Flag-tag and His-tag to determine the SUMOylation efficiency . The left samples were separated with 4–12% SDS-PAGE protein gel and developed with silver staining . Stained bands which indicated the SUMOylated CDK9 were cut out and proceeded to in-gel digestion as above . LC-MS/MS was used to analyze the SUMOylated peptides as we have described in Global site-specific SUMO-MS . For all the SUMOylation-related co-immunoprecipitation ( Co-IP ) , different tagged protein-expression constructs were transfected into Hela cells which were cultured in 6 cm dishes . Forty-eight hours post-transfection , cells were washed twice with PBS and lysed with NP-40 lysis buffer ( 10 mM Tris-HCl buffered at pH 7 . 5 , 150 mM NaCl , 0 . 5% NP-40 , 1% Triton X-100 , 10% Glycerol , 2 mM EDTA , 1 mM NaF , 1 Mm Na3VO4 ) supplemented with 1/100 protease inhibitor cocktail ( PIC ) ( Sigma-Aldrich ) and 2 M N-Ethylmaleimide ( NEM ) ( Selleckchem ) for 30 min on ice . Every 10 min , the incubation tubes were inverted several times . The lysates were clarified by centrifugation at 12 , 000 rpm for 10 min at 4°C , followed by incubating with anti-HA-tag beads ( Sigma-Aldrich ) , anti-Flag-tag beads ( Sigma-Aldrich ) or anti-His-tag beads ( Abcam ) for 4 hr to overnight at 4°C while rotating . The next day , proteins which were enriched by beads were washed for five times with ice-cold STN IP wash buffer ( 10 mM Tri-HCl buffered at pH 7 . 5 , 150 mM NaCl , 0 . 5% NP-40 , 0 . 5% Triton X-100 ) and eluted by boiling with 4 × protein SDS-PAGE loading buffer at 100°C for 15 min . The supernatants containing target proteins were transferred to new tubes after centrifugation at 12 , 000 rpm for 3 min , followed by western blot with antibodies against HA-tag ( MBL , PM020 ) , Flag-tag ( MBL , M180-3 ) , His-tag ( Proteintech , 66005–1-Ig ) or other indicated antibodies . GAPDH ( Proteintech , 10494–1-AP ) was set as internal reference . 680RD goat anti-mouse IgG antibody ( LI-COR Biosciences , 926–68070 ) and 800CW goat anti-rabbit IgG antibody ( LI-COR Biosciences , 926–32211 ) were used as secondary antibodies . The western blot membranes were developed with Odyssey CLX Imager ( LI-COR Biosciences ) and analyzed by Image Studio Lite Ver 4 . 0 ( LI-COR Biosciences ) . For a given protein , the SUMOylated form is only a small proportion . To enhance the SUMOylation signals , we conducted several SUMOylation assay by co-overexpression target proteins with SUMOylation system components which were SUMOs , E1 SAE1/UBA2 , E2 UBC9 , and E3 TRIM28 . In vertebrates , there are four well-studied SUMO paralogs , SUMO1 , SUMO2 , SUMO3 , and SUMO4 . Because SUMO2 and SUMO3 share highly sequence identity and have similar functions , they are referred to as SUMO2/3 . In preliminary data , we found the overexpression of E1 had little influence on the SUMOylation due to the high expression of endogenous E1 . Therefore , we omitted E1 in the following SUMOylation assays . Besides , there are lots SUMO-specific isopeptidases ( SENPs ) which deSUMOylate substrates . Thus we used mature SUMO polypeptides instead of immature ones . For CDK9 SUMOylation assay , 2 μg HA-tagged wild type or mutated CDK9-expression plasmids , 4 μg Flag-tagged SUMO4-expression plasmids , 500 ng Flag-tagged UBC9 and 500 ng Flag-tagged TRIM28 were co-transfected into Hela cells which cultured in 6 cm dishes . Forty-eight hours post-transfection , cells were harvested in NP-40 lysis buffer containing 2 M NEM which was used to prevent deSUMOylation . Co-IP and western blot against HA-tagged CDK9 was performed according to the procedure which we mentioned above . For SENP3-mediated deSUMOylation assay , 500 ng or 1 μg SENP3-expression plasmids were additionally co-overexpressed with indicated amount of CDK9 , SUMO4 , UBC9 and TRIM28 . Specific antibodies against SENP3 ( Proteintech , 17659–1-AP ) was used in western blot to confirm the expression . In vitro SUMOylation assay was performed by co-culturing in vitro-purified 1 μg CDK9 ( Abcam ) with in vitro-purified 4 μg SUMO4 ( This paper ) , 500 ng E1 ( SAE1/UBA2 ) ( R and D ) , 500 ng UBC9 ( R and D ) or 500 ng TRIM28 ( Abcam ) in SUMO conjugation reaction buffer ( R and D ) . The reaction was initiated by adding 1 mM Mg-ATP solution and incubated for 3 hr at 30°C , followed by adding stop buffer to terminate the reaction . Samples were boiled with SDS-PAGE loading buffer supplemented with 1 M DTT for 15 min at 100°C and proceeded to western blot with specific antibodies against CDK9 ( CST , 2316 ) , SUMO4 ( Abcam , ab126606 ) , SAE1 ( Proteintech , 10229–1-AP ) , UBA2 ( Abclonal , A4363 ) , UBC9 ( Abclonal , A2193 ) , and TRIM28 ( Proteintech , 15202–1-AP ) . For samples used for super-resolution Structured Illumination Microscopy ( SIM ) imaging , HEK293T cells were plated into Lab-Tek II chambered coverglass ( ThermoFisher ) which was pretreated with poly-lysine ( Sigma-Aldrich ) . Twelve hours later , cells were transfected with GFP-tagged TRIM28 or GFP-tagged TRIM28-dRING with RFP-tagged CDK9 . Twenty-four hours post-transfection , cells were washed with PBS once and fixed with 3% paraformaldehyde ( Electron Microscopy Sciences ) /0 . 1% glutaraldehyde ( Electron Microscopy Sciences ) for 10 min at room temperature ( RT ) . Fixed samples were reduced with 0 . 1% NaBH4 ( Sigma-Aldrich ) for 7 min at room temperature while shaking , followed by washing with PBS for 3 times at room temperature , 5 min per wash . Cells were further permeabilized with 0 . 2% Triton X-100 ( Sigma-Aldrich ) for 15 min and blocked with 10% normal donkey serum ( NDS ) ( Jackson ImmunoResearch ) /0 . 05% Triton X-100 for 90 min at RT . After blocking , samples were washed with 1% NDS/0 . 05% Triton X-100 for 15 min at RT for five times . Then , samples were wash with PBS once for 5 min , followed by post-fixation for 10 min with 3% paraformaldehyde/0 . 1% glutaraldehyde . After post-fixation , samples were washed with PBS for three times , 5 min per wash . 4' , 6-Diamidino-2-Phenylindole , Dihydrochloride ( DAPI ) ( ThermoFisher ) solution was added into samples to dye DNA for 10 min while shaking . Finally , samples were washed with PBS for three times and imaged on an Eclipse Ti inverted microscope equipped with a CFI Apo TIRF objective ( NA 1 . 49 , oil immersion ) and NIS-Elements AR software , an sCMOS camera ( Hamamatsu Flash 4 . 0 , 6 . 5 μm × 6 . 5 μm pixel size ) , and four lasers named SIM 405 , SIM 488 , SIM 561 and SIM 647 . The original images were acquired with 512 × 512 resolution and reconstructed to form the SIM images with 1024 × 1024 resolution . The lateral resolution of the SIM image is 115 nm and the axial resolution is 300 nm . Z-step size was set to 0 . 20 μm . For each focal plane , 15 images ( five phases , three angles , 3D-SIM mode ) were captured with the NIS-Elements software . SIM images were reconstructed and analyzed with the N-SIM module of the NIS-Elements Advanced Research software ( Nikon ) . For the quantitation of co-localization , SIM images were further analyzed with Imaris software ( Version 9 . 2 ) ( BITPLANE ) using Coloc toolbar . Percentages of each channel voxels above threshold co-localized were calculated . Both Pearson`s coefficient and thresholded Mander’s coefficient were calculated to indicate the qualities of co-localization . For Pearson’s coefficient , a value of 1 represents perfect co-localization , 0 no co-localization , and −1 perfect inverse co-localization . For thresholded Mander’s coefficient , a value of 1 represents perfect co-localization and 0 no co-localization . For samples used for super-resolution continuous STochastic Optical Reconstruction Microscopy ( cSTORM ) imaging , cells were plated , fixed , reduced , permeabilized , blocked and washed as in SIM samples preparation . After blocking , primary antibodies against TRIM28 ( Proteintech , 66630–1-Ig ) , SUMO4 ( Abcam , ab126606 ) and CDK9 ( CST , 2316 ) were incubated with cells for 60 min at RT in 5% NDS/0 . 05% Triton X-100 . Samples were washed for five times with 1% NDS/0 . 05% Triton X-100 at RT , 15 min per wash . Then , cells were incubated with secondary antibodies diluted in 5% NDS/0 . 05% Triton X-100 for 30 min at RT while shaking . Two sets of secondary antibody pairs were used to confirm the specificity , which were: Donkey Anti-Mouse IgG H and L ( Alexa Fluor 647 ) Antibody ( Abcam , ab150107 ) combining with Donkey Anti-Rabbit IgG H and L ( CF 568 ) Antibody ( Biotium , 20803–500 μl ) , Donkey Anti-Rabbit IgG H and L ( Alexa Fluor 647 ) Antibody ( Abcam , ab150075 ) combining with Donkey Anti-Mouse IgG H and L ( CF 568 ) Antibody ( Biotium , 20802–500 μl ) . After incubation , cells were washed as above followed by another wash with PBS for 5 min . Post-fixation was performed with 3% paraformaldehyde/0 . 1% glutaraldehyde for 10 min without shaking . Then , cells were washed with PBS for three times , 5 min per wash , followed by washing with water for two times , 3 min per wash . Of note , DAPI and Hoechst were not allowed to dye DNA according to cSTORM protocol . cSTORM imaging buffer was freshly prepared as below . GLOX solution was compounded by mixing 100 μl of 70 mg/ml Glucose Oxidase ( Sigma-Aldrich ) diluted in Buffer A ( 10 mM Tris-HCl buffered at pH 8 . 0 , 50 mM NaCl ) with 25 μl of 17 mg/ml Catalase ( Sigma-Aldrich ) diluted in Buffer A . One mole per liter of Cysteamine ( MEA ) ( Sigma-Aldrich ) was compounded by diluting 77 mg of MEA into 1 ml 0 . 25 N HCl . On ice , cSTORM imaging buffer was compounded by mixing 7 μl of GLOX , 70 μl of 1M MEA , and 620 μl of Buffer B ( 50 mM Tris-HCl buffered at pH 8 . 0 , 10 mM NaCl , 10% Glucose ) . Each well of Lab-Tek II chambered coverglass was added 700 μl of imaging buffer which was able to be used for 2 hr . Samples were imaged under a Nikon N-STORM super-resolution microscope equipped with a high-numerical-aperture ( high-NA ) 100 × oil immersion objective ( Nikon CFI SR Apochromat TIRF 100 × oil , 1 . 49 NA ) , a high-sensitivity and high-resolution sCMOS camera ( Hamamatsu Flash 4 . 0 , 6 . 5 μm × 6 . 5 μm pixel size , and an 0 . 4 × relay lens to match the pixel size under STORM mode ) , and four lasers with excitation wavelengths of 405 , 488 , 561 and 647 nm . For cSTORM which we used here , 405 nm laser was used as activation laser . 488 nm , 561 nm and 647 nm lasers were used as reporter lasers . The lateral resolution of the cSTORM image is 20 nm and the axial resolution is 50 nm . The z position was maintained during the acquisition by a Nikon ‘perfect focus system’ . 20 , 000 to 25 , 000 frames were taken for each image . Single molecule localization was obtained by Gaussian fitting using the STORM plug-in of NIS-Elements Advanced Research software taking into account both drift and chromatic aberrations . For the quantitation of co-localization , cSTORM images were further analyzed with Imaris software ( Version 9 . 2 ) ( BITPLANE ) by measuring the distance of spots-spots center . cSTORM-imaged protein molecules and complexes were transformed into small or large spots based on their diameter . The spots-spots co-localization was defined by the criterion of maximal distance of 10 nm . The complexes-spots co-localization was defined by the criterion of maximal distance of 100 nm . The percentages of co-localization were calculated for both total proteins-proteins co-localization , spots-spots co-localization and complexes-spots co-localization for each protein . In vitro SUMOylation assay was performed for CDK9 as described above . Five groups were set: Group 1 ( G1 ) : CDK9 only; Group 2 ( G2 ) : CDK9 and SUMO4; Group 3 ( G3 ) : CDK9 , SUMO4 and E1 ( SAE1 and UBA2 ) ; Group 4 ( G4 ) : CDK9 , SUMO4 , E1 and E2 ( UBC9 ) ; Group 5 ( G5 ) : CDK9 , SUMO4 , E1 , E2 and E3 ( TRIM28 ) . The reaction was terminated by stop buffer . To initiate the CDK9 kinase assay , CDK9 substrate PDKtides and ATP were added into each samples according to the manufacturer’s instruction ( Promega ) . The reaction was incubated for 120 min at room temperature followed by ADP-Glo kinase assay ( Promega ) . Briefly , ADP-Glo reagent was added into the reaction to deplete the remaining ATP . Samples were incubated at room temperature for 40 min . After ATP depletion , kinase detection reagent was added into samples to convert the ADP which was consumed during CDK9 kinase assay to ATP . This reaction was performed by incubating samples at room temperature for 30 min . Finally , the newly synthesized ATP was quantitated using luciferase/luciferin reaction . The luminescence generated during luciferase/luciferin reaction was recorded with integration time of 0 . 5 to 1 s . The relative light units were calculated by normalizing to untreated wild-type CDK9 group . TRIM28 in Hela cells and HIV-1-infected CD4+ T cells was knocked down by siRNA targeting TRIM28 . ShRNA and sgRNA lentiviruses targeting TRIM28 were used to knock down TRIM28 and knock out TRIM28 in J-Lat 10 . 6 , respectively . The cytotoxicity assay was conducted by incubating Cell Counting Kit-8 ( CCK-8 ) reagents ( Dojindo , CK04 ) with wild type and TRIM28-deficient cells for 3 hr followed by measuring the absorbance at 450 nm using a microplate reader . The cell viability assay was conducted by measuring the percentage of amine-reactive fluorescent dye ( BioLegend , 423113 ) non-permeant cells , which indicated the percentage of viable cells . Cell numbers were recorded every 2 days for both wild-type and TRIM28-deficient cells . The proliferation assay was conducted by staining live cells with CFSE ( ThermoFisher , C34554 ) . On Day 0 , cells from each group were stained with CFSE . The percentage and mean fluorescence intensity ( MFI ) of CFSE-positive cells were analyzed by flow cytometry every 2 days . Resting CD4 +cells were isolated from HIV-1-infected individuals who underwent suppressive cART for at least 6 months with undetectable plasma HIV-1 RNA ( less than 50 copies/ml ) and high CD4+ T cell count ( at least 350 cells/mm3 ) ( Human Lymphocyte Separation Kit , TBDsciences; BD IMag Human CD4+ T Lymphocyte Enrichment Set-DM , BD Biosciences ) . These CD4+ T cells were nucleofected with siRNAs targeting negative control and TRIM28 respectively , and cultured in Super T Cell Medium ( STCM ) consisting of RPMI 1640 supplemented with 1% penicillin-streptomycin , 1% L-glutamine , 10% FBS , 100 U/ml IL-2 , and 2% T-cell growth factor ( TCGF ) from the supernatants of mitogen-activated healthy PBMCs treated with 2 μg/ml PHA-M and 5 ng/ml PMA for 4 hr . Six hours post-transfection , supernatants were replaced with new culture medium . Twenty-four hours later , half of siNC-treated cells were separated and supplemented with 0 . 5 μg/ml PHA-M ( Sigma-Aldrich ) . Half of siTRIM28-treated cells were separated and supplemented with 500 nM Vorinostat ( SAHA ) ( Selleckchem ) . Another 24 hr later , supernatants from each group were changed with fresh culture medium to prevent the toxicity of the PHA-M or SAHA . Seventy-two hours post-transfection , cells were exposed to 20 Gy X-ray irradiation for 5 min and supplemented with PHA-activated healthy CD4 +T cells . The supernatants were collected and half-changed with fresh STCM every 3 days . Cell suspension was half-changed with PHA-stimulated healthy CD4+ T cell suspension every 6 days . All the supernatants from each time points and each groups were measured for the presence of HIV-1 antigen with HIV-1 p24 ELISA kit ( Abcam ) according to the manufacturer’s instruction by SkanIt SW for Microplate Readers ( ThermoFisher ) . The genetic diversity of HIV-1 quasispecies under different conditions was evaluated by sequencing the envelope V1-V3 region . HIV-1 RNAs from each group were reverse-transcribed by specific primer ES8B: 5’-CACTTCTCCAATTGTCCCTCA-3’ . Two rounds of nested PCR were performed to amplify V1-V3 region with the following primer pairs: 1st round Nest PCR Forward Primer ( E00 ) : 5’-TAGAAAGAGCAGAAGACAGTGGCAATGA-3’ , 1st round Nest PCR Reverse Primer ( ES8B ) : 5’-CACTTCTCCAATTGTCCCTCA-3’; 2nd round Nest PCR Forward Primer ( E20 ) : 5’-GGGCCACACATGCCTGTGTACCCACAG-3’ , 2nd round Nest PCR Reverse Primer ( E115 ) : 5’-AGAAAAATTCCCCTCCACAATTAA-3’ ( Geng et al . , 2016b ) . For each PCR reaction , Phanta Max Super-Fidelity DNA Polymerase ( Vazyme ) was used to amplify the V1-V3 region of HIV-1 envelope in order to ensure the fidelity . The amplification error rate of Phanta Max is 53-fold lower than that of Taq and 6-fold lower than that of Pfu according to the manufacturer’s instruction . After two rounds of nested PCR utilizing Phanta Max , the PCR products were proceeded to deoxyadenosine ( A ) -tailing at the 3'-end of the PCR products utilizing Ex Taq DNA polymerase ( Takara ) without thermal cycling as follows: 95°C , 5 min; 72°C , 30 min; 4°C hold . The A-tailed PCR products were TA-ligated into pMD-18T vector . To minimize the sampling bias , single genome amplification method was performed by obtaining 30 independent PCR products from each sample . At least 60 single clones were picked from each group and proceeded to Sanger sequencing . The sequences from each group were aligned using MUSCLE . The sequences with ambiguous positions were removed . The average genetic distance between one give clone and the relevant entire population were calculated by MEGA seven and represented as genetic diversity index . The Mann-Whitney U-test was performed to compare the genetic diversity indexes between different groups using Prism 5 . The phylogenetic bootstrap consensus trees were generated for each samples using neighbor-joining method with 1000 bootstrap replications implemented in MEGA seven to depict the global landscape of HIV-1 diversity . Freshly isolated CD4+ T cells were stimulated with PHA for 2 days or left untreated . Total RNAs from each group were extracted by TRIzol Reagent ( ThermoFisher ) according to the manufacturer’s instruction . The quality of RNA samples were evaluated by Nanodrop 2000 ( ThermoFisher ) and BioAnalyzer 2100 ( Aglient ) . The RNA-Seq library were built with TruSeq Stranded mRNA Library Prep Kit ( Illumina ) and sequenced with HiSeq X Ten ( Illumina ) at BioMarker ( Beijing , China ) under the PE150 protocol . RNA-Seq reads were trimmed , filtered and quality-controlled by FastQC ( Babraham Institute ) tool . The reads were aligned to human reference genome NCBI build 38 ( GRCh38 ) by Hisat2 ( Kim et al . , 2015 ) , followed by calculating the reads per kilobase per million mapped reads ( RPKM ) . Differentially expressed genes were filtered by DEGseq ( Bioconductor ) tool with log2FC of 1 and PvalueFDR cutoff of 0 . 05 , and plotted as heatmap or volcanoplot by gplots ( R Foundation ) . TRIM28-defective ( sgTRIM28 ) J-Lat 10 . 6 cell line was generated by CRISPR-CAS9 technique . ATAC-Seq was conducted with sgNT and sgTRIM28 J-Lat 10 . 6 cell lines , as well as siNC and siTRIM28 TZM-bl cell lines . The ATAC-Seq library was built with TruePrep DNA Library Prep Kit V2 ( Vazyme ) as previously described ( Buenrostro et al . , 2013 ) . Briefly , approximately 30 , 000 cells were harvested , washed with ice-cold PBS , and lysed with 50 μl of ice-cold lysis buffer ( 10 mM Tris-HCl buffered at pH 7 . 4 , 10 mM NaCl , 3 mM MgCl2 , 0 . 1% Igepal CA-630 ) for 10 min on ice . The lysates were centrifuged for 5 min at 500 G , 4°C . The supernatants were carefully removed . Transposition reaction mix , which consisted of 10 μl of 5 × TTBL , 5 μl of TTE Mix V50 and 35 μl of ddH2O , was used to resuspend nuclei pellet and incubated at 37°C for 30 min . The transposed DNA was purified by VAHTS DNA Clean Beads ( Vazyme ) and PCR-amplified with the following mixture: 24 μl of purified DNA , 10 μl of 5 × TAB , 5 μl of PPM , 5 μl of N5 primer , 5 μl of N7 primer , and 1 μl of TAE . Thermal cycle was as follows: 72°C for 3 min; 98°C for 30 s; and thermocycling at 98°C for 15 s , 60°C for 30 s and 72°C for 3 min; following by 72°C 5 min . The amplified ATAC-Seq library was purified with VAHTS DNA Clean Beads and eluted with 30 μl ddH2O . The library quality was evaluated by Qubit 3 . 0 Fluorometer ( ThermoFisher ) and BioAnalyzer 2100 ( Aglient ) , and sequenced with HiSeq X Ten ( Illumina ) at BioMarker ( Beijing , China ) under the PE150 protocol . ATAC-Seq reads were trimmed , filtered and quality-controlled by FastQC tool . Then the reads were aligned to GRCh38 by Bowtie2 ( Langmead and Salzberg , 2012 ) , followed by rearranging with Samtools ( Li et al . , 2009 ) . The reads were also separately aligned to HIV-1 reference genome K03455 , M38432 ( Version K03455 . 1 ) by Bowtie2 , followed by rearranging with Samtools . Igvtools ( Broad Institute ) was used to visualize the tag peaks . Specific gene loci was amplified . Tag density from different groups was calculated by normalizing to the total mapped reads . The highest tag density was set as 100 . Relative tag densities of two kilobases range centered HIV-1 5’LTR integration sites were calculated and compared with sgNT or siNC . Triplicates data were presented as mean ±SEM . A value of p<0 . 05 was considered to be statistically significant and represented as asterisk ( * ) . Value of p<0 . 01 was considered to be more statistically significant and represented as double asterisks ( ** ) . Value of p<0 . 001 was considered to be the most statistically significant and represented as triple asterisks ( *** ) . For the comparison of ChIP , the GFP-positive percentages and qPCR experiments , standard t test was used . For the comparison of genetic diversity index experiment , Mann-Whitney U-test was used . Statistical analyses were conducted with Prism 5 ( GraphPad ) . The network analysis and clustering analysis were conducted with STRING and MCODE in Cytoscape ( Cytoscape Consortium ) . Co-crystal structure of Cyclin T1 and CDK9 ( PDB ID: 4EC8 ) were reconstituted in PyMOL ( Schrödinger ) ( Baumli et al . , 2012 ) . Both ribbon models and surface models were used to present the structure . | The human immunodeficiency virus-1 , or HIV-1 , infects certain human cells , including white blood cells . One reason the infection is incurable is because the virus can integrate its genetic information into its host , and essentially ‘sleep’ within the host cell , a process called latency . This helps to hide HIV-1 from the immune system and stops it getting destroyed . Latency represents a critical challenge in treating and curing HIV-1 . One proposed cure for HIV-1 involves ‘shocking’ the viruses out of latency so that they can be eliminated . Applying this so-called shock and kill approach means scientists need to understand more about how latency is maintained . Previous evidence shows that latency requires proteins known as histone deacetylases and histone methyltransferases . Certain gene-silencing proteins called transcription suppressors are also involved . Ma et al . have now examined latent HIV-1 in several kinds of human cells grown in the laboratory . The cells were modified to make certain proteins at much lower levels than normal . The experiments showed that the loss of a protein called TRIM28 ‘wakes up’ latent HIV-1 . TRIM28 attaches chemical marks called SUMOylations to gene regulators in the cell . These SUMOylations restrict the activity of HIV-1’s genes , which is important to maintain latency . Specifically , TRIM28 adds SUMOylations to a protein named CDK9 at three key positions . Reducing the levels of TRIM28 made it easier to shock many HIV-1 in infected cells out of latency . With further investigation , targeting TRIM28 may one day be used to treat HIV-1 infection through a shock and kill method . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2019 | TRIM28 promotes HIV-1 latency by SUMOylating CDK9 and inhibiting P-TEFb |
The only property of reinforcement insects are commonly thought to learn about is its value . We show that larval Drosophila not only remember the value of reinforcement ( How much ? ) , but also its quality ( What ? ) . This is demonstrated both within the appetitive domain by using sugar vs amino acid as different reward qualities , and within the aversive domain by using bitter vs high-concentration salt as different qualities of punishment . From the available literature , such nuanced memories for the quality of reinforcement are unexpected and pose a challenge to present models of how insect memory is organized . Given that animals as simple as larval Drosophila , endowed with but 10 , 000 neurons , operate with both reinforcement value and quality , we suggest that both are fundamental aspects of mnemonic processing—in any brain .
What are the fundamental capacities of insect brains ? To date , little use has been made of insect memory experiments to reveal these capacities , in particular regarding reinforcement . For example , after experiencing an odor with a sugar reward , fruit flies ( Drosophila melanogaster ) approach that odor in a later test . All the known circuitry ( Heisenberg , 2003; Perisse et al . , 2013 ) of such learned search behavior suggests that this is because the odor has acquired positive value; that is , that the flies are expecting to find ‘something good’ in its vicinity ( Heisenberg , 2003; Gerber and Hendel , 2006; Schleyer et al . , 2011; Perisse et al . , 2013 ) . Likewise , Drosophila can associate an odor with an electric-shock punishment . This supports their learned escape in a later test because they may expect ‘something bad’ with the odor . In other words , the only feature of reinforcement processing that insects are granted is value . We show that larval Drosophila are in a defined sense richer than this in their mnemonic capacity: they also recall of what particular quality that good or bad experience was . Given the numerical simplicity of the larval brain , this is suggested to be a more basic property of brains than hitherto assumed . To address this question , we exploit an established assay for Pavlovian conditioning ( Gerber et al . , 2013 ) that allows reinforcers of various strength and quality to be used . In this Petri dish assay , larvae are placed onto a tasteless agarose substrate . This substrate is supplemented with a fructose sugar reward—if odorant A is presented . Odorant B is presented without the reward ( A+/B ) . For a companion group of larvae , contingencies are reversed ( A/B+ ) . In a binary choice test , the larvae then systematically approach the previously rewarded odorant ( Figure 1—figure supplement 1 ) . This behavior , quantified as a positive associative performance index ( PI ) ( Figure 1A ) , can best be grasped as a memory-based search for reward: if the test is performed in the presence of fructose , the learned approach is abolished ( Figure 1A ) ( Gerber and Hendel , 2006; Schleyer et al . , 2011 ) ( olfactory behavior per se is not affected: see below ) . This is adaptive as learned search behavior is indeed obsolete in the presence of a sought-for item . We have previously shown ( Schleyer et al . , 2011 ) that regardless of the absolute concentration of fructose , such an abolishment is seen if the fructose concentration in the test substrate is equal to or higher than that used in training . This means that learned behavior is based on a relative assessment: the larvae recall how strong the training reward was and compare this remembered strength to the current testing situation . Only if that comparison promises a gain ( remembered strength > current strength ) do they search for what they can thus expect to gain at the odor source . We note that widely applied formal learning models of the Rescorla-Wagner type ( Rescorla and Wagner , 1972 ) propose that memory acquisition will only occur if something new and unexpected happens , specifically if the experienced reward is stronger than predicted on the basis of memory ( current strength > remembered strength ) . Thus , the same two pieces of information are compared during memory acquisition on the one hand and the expression of learned search behavior on the other hand—yet in a ‘swapped’ way . This can inform the animals respectively about what is new or what there is to be gained . Here , we ask whether these processes are integrated across different qualities of reward into one common scale of appetitive value or whether separate systems exist to confer mnemonic specificity for the ‘quality’ of reward . 10 . 7554/eLife . 04711 . 003Figure 1 . Reward processing by quality and value . ( A ) Larvae are trained to associate one of two odors with either 2 M fructose or 10 mM aspartic acid as reward . Subsequently , they are tested for their choice between the two odors—in the absence or in the presence of either substrate . For example , in the left-most panel , a group of larvae is first ( upper row ) exposed to n-amyl acetate ( blue cloud ) together with fructose ( green circle ) , and subsequently ( middle row ) to 1-octanol ( gold cloud ) without any tastant ( white circle ) . After three cycles of such training , larvae are given the choice between n-amyl acetate and 1-octanol in the absence of any tastant ( lower row ) . A second group of larvae is trained reciprocally , that is , 1-octanol is paired with fructose ( second column from left , partially hidden ) . For the other panels , procedures are analogous . Aspartic acid is indicated by brown circles . ( B ) Data from ( A ) plotted combined for the groups tested on pure agarose ( ‘Mismatch’ in both value and quality ) , in the presence of the respectively other reward , or of the training reward . Learned search behavior towards the reward-associated odor is abolished in presence of the training reward because both reward value and reward quality in the testing situation are as sought-for ( ‘Match’ in both cases ) , yet search remains partially intact in the presence of the other quality of reward , because reward value is as sought-for ( ‘Match’ ) but reward quality is not ( ‘Mismatch’ ) . Please note that value-memory is apparently weaker than memory for reward quality , and is revealed only when pooling across tastants . Sample sizes 15–19 . Shaded boxes indicate p < 0 . 05/6 ( A ) or p < 0 . 05/3 ( B ) from chance ( one-sample sign-tests ) , asterisks indicate pairwise differences between groups at p < 0 . 05/3 ( A ) or p < 0 . 05/2 ( B ) ( Mann–Whitney U-tests ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04711 . 00310 . 7554/eLife . 04711 . 004Figure 1—figure supplement 1 . Preference scores for the reciprocally trained groups of Figure 1 . Sample sizes: 15–19 . Shaded boxes indicate pairwise differences between groups ( p < 0 . 05/6 , Mann–Whitney U-tests ) . For a detailed description of the sketches , see legend of Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04711 . 00410 . 7554/eLife . 04711 . 005Figure 1—figure supplement 2 . ( A , B ) Larvae are trained in a one-odor version of the learning paradigm , using different concentrations of aspartic acid as reward . Increasing the concentration increases the associative performance index . For the highest concentration used , levels of learned behavior are not different from those obtained with 2 M fructose as reward . ( B ) Preference scores underlying the associative performance indices in ( A ) . Sample sizes: 11–17 . Shaded boxes in ( A ) indicate difference from chance ( p < 0 . 05/4 , one-sample sign-tests ) , in ( B ) pairwise differences between groups ( p < 0 . 05/4 , Mann–Whitney U-tests ) . Asterisk: difference between groups ( p < 0 . 05 , Kruskal–Wallis test ) . For a detailed description of the sketches , see legend of Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04711 . 005
We introduce aspartic acid , a proteinogenic amino acid , as a novel quality of reward ( Figure 1—figure supplement 2 ) . Concentrations of aspartic acid and fructose are chosen such that their reward value is equal ( Figure 1A ) . This allows us to study the larvae under test conditions that are of equal reward value but that either match or do not match the ‘quality’ of reward that has been employed during training . If the larvae merely searched for a reward of remembered value , learned search behavior should cease in both cases—because current strength matches remembered strength in both cases . If reward quality were the sole determinant for learned search , in contrast , learned search should be abolished only when test and training substrate match in quality , but should remain intact when quality does not match . If the larvae searched for a reward specified both by its value and by its quality , scores in the mismatch case should be partially abolished: in that case , the reward's value is as sought yet its quality is not . We find that learned search is fully abolished when the training and test reward match in both value and quality but remains partially intact ( by 68% ) if there is a mismatch in reward quality between training and test ( Figure 1A , B ) . We conclude that after odor-fructose training the larvae approach the odor both in search of something ‘good’ ( value ) and in search of what is specifically fructose ( quality of reward ) . Likewise , after odor-aspartic acid training , they search both for something ‘good’ and for aspartic acid . In other words , if during the test the larvae , for example , have sugar anyway but remember where aspartic acid can be found , they will still go for aspartic acid in addition . Regarding the aversive domain , pairing an odor with quinine as punishment leads to aversive memory ( Gerber and Hendel , 2006; Schleyer et al . , 2011; El-Keredy et al . , 2012; Apostolopoulou et al . , 2014 ) . In this case , learned behavior can best be understood as an informed escape that is warranted in the presence but not in the absence of quinine . Accordingly , in the presence but not in the absence of quinine one observes that the larvae run away from the previously punished odor ( Gerber and Hendel , 2006; Schleyer et al . , 2011; El-Keredy et al . , 2012; Apostolopoulou et al . , 2014 ) ( see also Niewalda et al . , 2008; Schnaitmann et al . , 2010; Eschbach et al . , 2011; Russell et al . , 2011 for reports using other aversive reinforcers and/or adult flies ) . Notably and in accordance with our earlier results in the appetitive domain ( Schleyer et al . , 2011 ) , such learned escape does not merely depend on the concentration of quinine in the test; rather , learned escape lessens as the quinine concentration in the test is reduced relative to that in training ( Figure 2—figure supplement 1 ) . Given that high-concentration salt can also serve as punishment ( Gerber and Hendel , 2006; Niewalda et al . , 2008; Russell et al . , 2011 ) , we ask whether an odor-quinine memory is specific in prompting escape from quinine—but not from salt . For concentrations of quinine and salt that are of equal value as punishment , this is indeed the case ( Figure 2A ) ( Figure 2—figure Supplement 2 ) ; likewise , odor-salt memories are specific in prompting learned escape from salt but not from quinine ( for a summary see Figure 2B ) . Such specificity shows that larvae have a memory specific to the quality of punishment , a memory that can specifically be applied in the appropriate situation . We stress that the present results do not provide proof of the absence of aversive ‘common currency’ value processing . Indeed , in cases of unequal punishment value , larvae may use this information ( Eschbach et al . , 2011 ) . 10 . 7554/eLife . 04711 . 006Figure 2 . Punishment processing by quality . ( A ) Larvae are trained to associate one of two odors with either 5 mM quinine ( red circle ) or 4 M sodium chloride ( purple circle ) as punishment and asked for their choice between the two odors—in the presence of either substrate . The larvae show learned escape from the punishment-associated odor only if a matching quality of punishment is present during the test as compared to training . ( B ) Data from ( A ) combined according to ‘Match’ or ‘Mismatch’ between test- and training-punishment . We note that value-memory would reveal itself by negative scores upon a match of punishment value despite a mismatch in punishment quality , which is not observed ( right-hand box plot , based on second and fourth box plot from A ) . Sample sizes: 25–32 . Shaded boxes indicate p < 0 . 05/4 ( A ) or p < 0 . 05/2 ( B ) from chance ( one-sample sign-tests ) , asterisks indicate pairwise differences between groups at p < 0 . 05/3 ( A ) or p < 0 . 05 ( B ) ( Mann–Whitney U-tests ) . For a detailed description of the sketches , see legend of Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04711 . 00610 . 7554/eLife . 04711 . 007Figure 2—figure supplement 1 . Quinine memory includes quinine strength . ( A ) Larvae are trained to associate one of two odors with a quinine punishment , using either a low ( light red ) or a high ( dark red ) concentration of quinine . Learned escape behavior is expressed if the testing conditions are as ‘bad’ as the training-punishment , or worse . In other words , neither the value of the training-punishment alone nor the value of the testing situation alone , determines the levels of learned behavior—but a comparison of them does . ( B ) Preference scores underlying the associative performance indices in ( A ) . Sample sizes: 22–23 . Shaded boxes in ( A ) indicate p < 0 . 05/5 from chance ( one-sample sign-tests ) , in ( B ) pairwise differences between groups at p < 0 . 05/5 ( Mann–Whitney U-tests ) . Asterisks in ( A ) indicate pairwise differences between groups at p < 0 . 05/3 ( Mann–Whitney U-tests ) . For a detailed description of the sketches , see legend of Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04711 . 00710 . 7554/eLife . 04711 . 008Figure 2—figure supplement 2 . Preference scores for the reciprocally trained groups of Figure 2 . Sample sizes: 25–32 . Shaded boxes indicate differences between groups ( p < 0 . 05/4 , Mann–Whitney U-tests ) . For a detailed description of the sketches , see legend of Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04711 . 008 Taken together , within both the appetitive and aversive domain , experiencing an odor with a taste reinforcement can establish an associative olfactory memory that is specific to the quality ( fructose , aspartic acid , quinine , high-concentration salt ) of taste reinforcement . The experimental twist to reveal such quality-of-reinforcement memory is accomplished by flagrantly breaking the first rule of associative memory research: namely never , ever , to test for learned behavior in the presence of the reinforcer . We would like to stress that innate olfactory behavior per se is not affected by the presence of any of the tastant reinforcers ( Hendel et al . , 2005; Schleyer et al . , 2011 ) ( Figure 3 ) . Also , the mere presence of any given tastant reinforcer during the memory test is not a critical determinant for whether learned behavior is observed: learned behavior can be observed ( or not ) in the presence of any of the tastant reinforcers in this study—what matters is how closely it matches the one used during training in quality and/or in value ( Figures 1B and 2B ) . 10 . 7554/eLife . 04711 . 009Figure 3 . Innate odor preference is not influenced by taste processing . Larvae are tested for their olfactory preference regarding ( A ) n-amyl acetate ( blue cloud ) , ( B ) 1-octanol ( gold cloud ) , or ( C ) for their choice between n-amyl acetate and 1-octanol . This is done in the presence of pure agarose ( white circle ) , 2 M fructose ( green circle ) , 10 mM aspartic acid ( brown circle ) , 5 mM quinine ( red circle ) , or 4 M sodium chloride ( purple circle ) . We find no differences in odor preferences across different substrates ( p > 0 . 05 , Kruskal–Wallis tests ) . Sample sizes: 20–26 . DOI: http://dx . doi . org/10 . 7554/eLife . 04711 . 009
The mushroom bodies , a third-order ‘cortical’ ( Tomer et al . , 2010 ) brain region in insects , are canonically proposed to feature distinct regions harboring appetitive and aversive olfactory memory traces , respectively ( Heisenberg , 2003; Perisse et al . , 2013; see also Schleyer et al . , 2011 ) ( Figure 4A , B ) . Only recently has the possibility of different neuronal substrates underlying different qualities of reinforcement come to be considered . These studies have so far not yielded a double dissociation between different dopaminergic mushroom body input neurons for different qualities of reinforcement:For the aversive domain Galili et al . ( 2014 ) suggested that the set of dopaminergic mushroom body input neurons responsible for heat-punishment in adult Drosophila is nested within that for electric-shock punishment . Similarly , electric-shock punishment and punishment with the insect repellent DEET appear to be signaled towards the mushroom body by largely if not completely overlapping sets of dopamine neurons ( Das et al . , 2014 ) . For the appetitive domain , a set of dopaminergic mushroom body input neurons ( included in the 0104-Gal4 strain ) that was previously found to be required for sugar-learning in adult Drosophila ( Burke et al . , 2012 ) turned out to be dispensable for water-reward learning ( Lin et al . , 2014 ) . Whether in turn dopaminergic mushroom body input neurons included in the R48B04 strain , which were discovered by Lin et al . ( 2014 ) to be required for water-reward learning , are dispensable for sugar-learning remains to be tested . 10 . 7554/eLife . 04711 . 010Figure 4 . Working hypotheses of reinforcement processing by value-only or by value and quality in larval Drosophila . ( A ) Simplified overview ( based on e . g . , Heisenberg , 2003; Perisse et al . , 2013 ) . Odors are coded combinatorially across the olfactory sensory neurons ( OSN , blue ) . In the antennal lobe , these sensory neurons signal towards local interneurons ( not shown ) and projection neurons ( PN , deep blue ) . Projection neurons have two target areas , the lateral horn ( LH , orange ) mediating innate approach , and the mushroom body ( MB , yellow ) . Reinforcement signals ( green and red for appetitive and aversive reinforcement , respectively ) from the gustatory system reach the mushroom body , leading to associative memory traces in simultaneously activated mushroom body neurons . In the present analysis , this sketch focuses selectively on five broad classes of chemosensory behavior , namely innate odor approach , learned odor search and escape , as well as appetitive and aversive innate gustatory behavior . The boxed region is displayed in detail in ( B–C ) . The break in the connection between mushroom body output and behavior is intended to acknowledge that mushroom body output is probably not in itself sufficient as a ( pre- ) motor signal but rather exerts a modulatory effect on weighting between behavioral options ( Schleyer et al . , 2013; Menzel , 2014; Aso et al . , 2014 ) . ( B ) Reinforcement processing by value ( based on e . g . , Heisenberg , 2003; Schleyer et al . , 2011; Perisse et al . , 2013 ) : a reward neuron sums input from fructose and aspartic acid pathways and thus establishes a memory allowing for learned search for ‘good’ . In a functionally separate compartment , a punishment neuron summing quinine and salt signals likewise establishes a memory trace for learned escape from ‘bad’ . This scenario cannot account for quality-of-reinforcement memory . ( C ) Reinforcement processing by both value and quality: in addition to a common , value-specific appetitive memory , fructose and aspartic acid drive discrete reward signals leading to discrete memory traces in at least functionally distinct compartments of the Kenyon cells , which can be independently turned into learned search . For aversive memory , there may be only quality-specific punishment signals . This scenario is in accordance with the present data . DOI: http://dx . doi . org/10 . 7554/eLife . 04711 . 010 Thus , the nuanced memory of at least two qualities of appetitive and two qualities of aversive taste reinforcers as shown in the present study is unexpected . Appropriate to such nuanced memories , the mushroom bodies show a fairly complex substructure , even in larval Drosophila . At least 10 mushroom body regions are recognized , defined by the tiled innervation of input and output neurons ( Pauls et al . , 2010b ) . Our behavioral data suggest that at least five such tiles of the mushroom body would be required to accommodate learned search for fructose or aspartic acid , learned escape from quinine or from high-concentration salt and in addition a less specific appetitive value-memory ( Figure 4C ) . Clearly the things worth remembering for a larva include many more than these five ( Niewalda et al . , 2008; Pauls et al . , 2010a; Eschbach et al . , 2011; Khurana et al . , 2012; Rohwedder et al . , 2012; Diegelmann et al . , 2013 ) . Likewise , the behavioral repertoire of larvae may be considerably greater than thought ( Vogelstein et al . , 2014 ) . Using our current approach , it will now be possible to systematically determine the limits of specificity in the processing of reinforcement quality . This may reveal signals of intermediate specificity to inform the animals about , for example , edibility , caloric value , proteinogenic value , suitability for pupariation , toxicity , or even acutely and situationally modulated matters of concern ( Simpson et al . , 2015 ) . We note that the distinction between fructose and aspartic acid memory implies that the sensory neurons mediating the rewarding effects of these stimuli cannot be completely overlapping and that the sensory neurons mediating the punishing effects of quinine and high-concentration salt likewise cannot ( for reviews of the taste system in Drosophila , see Cobb et al . , 2009; Gerber et al . , 2009 ) . For identifying these neurons , it is significant that they may be distinct from those mediating innate choice behavior ( Apostolopoulou et al . , 2014; König et al . , 2014 ) . In the vertebrate literature , the processing of reward by value has been regarded as a matter of sophistication because an integrated , higher-order value signal can be generated from sensorially distinct qualities of reward ( e . g . , Lak et al . , 2014 ) . On the other hand , reward expectations can apparently also be processed in a quality-specific manner ( e . g . , Dickinson and Balleine , 1994; Watanabe , 1996 ) . In terms of the minimally required number of cells , the processing by reinforcer quality is more demanding than value-only processing ( Figure 4B , C ) . The fact that even the humble , 10 , 000-neuron brain of a larva operates with both reward value and quality may suggest that they both represent fundamentally important , indispensable aspects of reward processing .
We used third-instar feeding-stage larvae from the Canton-Special wild-type strain , aged 5 days after egg laying . Flies were maintained on standard medium , in mass culture at 25°C , 60–70% relative humidity and a 12/12 hr light/dark cycle . Before each experiment , we removed a spoonful of food medium from a food vial , collected the desired number of larvae , briefly rinsed them in distilled water , and started the experiment . For experiments , we used Petri dishes of 90-mm inner diameter ( Sarstedt , Nümbrecht , Germany ) filled with 1% agarose ( electrophoresis grade; Roth , Karlsruhe , Germany ) . As reinforcers fructose ( FRU; CAS: 57-48-7; Roth , Karlsruhe , Germany ) , aspartic acid ( ASP; CAS: 56-84-8; Sigma–Aldrich , Seelze , Germany ) , quinine ( QUI; CAS: 6119-70-6; Sigma–Aldrich ) , or sodium chloride ( SAL; 7647-14-5; Roth , Karlsruhe , Germany ) were used in concentrations given in the results section . As odors , we used n-amyl acetate ( AM; CAS: 628-63-7; Merck , Darmstadt , Germany ) , diluted 1:50 in paraffin oil ( Merck , Darmstadt , Germany ) and 1-octanol ( OCT; CAS: 111-87-5; Sigma–Aldrich ) . Prior to experiments , odor containers were prepared: 10 µl of odor substance was filled into custom-made Teflon containers ( 5-mm inner diameter with a lid perforated with seven 0 . 5-mm diameter holes ) . Before the experiment started , Petri dishes were covered with modified lids perforated in the center by 15 holes of 1-mm diameter to improve aeration . For training , 30 larvae were placed in the middle of a FRU-containing dish with two odor containers on opposite sides , both filled with AM . After 5 min , larvae were displaced onto an agarose-only dish with two containers filled with OCT , where they also spent 5 min . Three such AM+/OCT training cycles were performed , in each case using fresh dishes . In repetitions of the experiment , in half of the cases training started with a reinforcer-added dish ( AM+/OCT ) and in the other half with an agarose-only dish ( OCT/AM+ ) . For each group of larvae trained AM+/OCT ( or OCT/AM+ , respectively ) , a second group was trained reciprocally , that is , AM/OCT+ ( or OCT+/AM , respectively ) . Following training , larvae were transferred to a test Petri dish that , as specified for each experiment , did or did not contain a reinforcer and given the choice between the two trained odors . After 3 min , larvae were counted and a preference score calculated as: ( 1 ) Preference = ( #AM − #OCT ) /#Total . In this equation , # indicates the number of larvae on the respective half of the dish . Thus , PREF values are constrained between 1 and −1 with positive values indicating a preference for AM and negative values indicating a preference for OCT . From two reciprocally trained groups of animals , we calculated an associative performance index ( PI ) as: ( 2 ) Performance Index = ( PreferenceAM+/OCT − PreferenceAM/OCT+ ) /2 . Thus , performance index values can range from 1 to −1 with positive values indicating appetitive and negative values indicating aversive conditioned behavior . Preferences and performance indices for other reinforcers were calculated in an analogous way . A group of 30 experimentally naïve larvae were placed on a Petri dish filled with pure agarose ( PUR ) or agarose containing FRU , ASP , QUI , or SAL . Animals were given the choice between an odor-filled and an empty Teflon container; as odor , either AM or OCT was used . After 3 min , the position of the larvae was scored to calculate their preference as: ( 3 ) Preference = ( #AM − #EM ) /#total ( this equation was used in Figure 3A ) . ( 4 ) Preference = ( #OCT − #EM ) /#total ( this equation was used in Figure 3B ) . To measure choice , one container was loaded with AM and the other with OCT and preference calculated as: ( 5 ) Preference = ( #AM − #OCT ) /#total ( this equation was used in Figure 3C ) . Preference values and performance indices were compared across multiple groups with Kruskal–Wallis tests . For subsequent pair-wise comparisons , Mann–Whitney U-tests were used . To test whether values of a given group differ from zero , we used one-sample sign tests . When multiple tests of the same kind are performed within one experiment , we adjusted significance levels by a Bonferroni correction to keep the experiment-wide error rate below 5% . This was done by dividing the critical p value 0 . 05 by the number of tests . We present our data as box plots which represent the median as the middle line and 25%/75% and 10%/90% as box boundaries and whiskers , respectively . | Actions have consequences; positive consequences or rewards make it more likely that a behavior will be repeated , while negative consequences or punishments can stop a behavior occurring again . Neuroscientists commonly refer to such rewards and punishments as ‘reinforcement’ . Fruit flies that are given a reward of sugar when they experience an odor will move towards the odor in later tests . However , in 2011 , research revealed that if the flies were given at least the same amount of sugar in the tests as they were rewarded with during the earlier training , the flies stopped moving towards the odor . This suggests that fruit flies can recall how strong a reward was in the past and compare this remembered strength to the current reward on offer; fruit flies will only continue searching if they expect to gain a larger reward by doing so . Insects were commonly thought to only learn the amount or ‘value’ of reinforcement , but not recall what kind or ‘quality’ of reward ( or punishment ) they had experienced . Now Schleyer et al . —including some of the researchers involved in the 2011 work—challenge and extend this notion and show that fruit fly larvae can remember both the value and quality of rewards and punishments . Fruit fly larvae were trained to expect a reward of sugar when exposed to one odor and nothing when exposed to a different odor . Consistent with the previous results , the larvae moved towards the first odor in the tests where no additional reward was provided . Moreover , the larvae did not move towards the odor in later tests if an equal or greater amount of sugar was provided during the testing stage . Schleyer et al . then took larvae that had been trained to expect a sugar reward and gave them a different , but equally valuable , reward during the testing stage—in this case , the reward was an amino acid called aspartic acid . These experiments revealed that most of the larvae continued to move towards the sugar-associated odor in search of the sugar reward . This indicates that the larvae were able to remember the quality of the reward , namely that it was sugar rather than aspartic acid . Schleyer et al . performed similar experiments , and observed similar results , when using two different punishments: bitter-tasting quinine and high concentrations of salt . These findings show that experiencing an odor along with taste reinforcement could set up a memory specific to the quality of reinforcement in fruit fly larvae . Given the numerical simplicity of a larva's brain—which contains only 10 , 000 neurons—it is likely that other animals can also recall both the value and quality of a reward or punishment . However , understanding how such specificity comes about should be easier in the larva's simple brain . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"neuroscience"
] | 2015 | Learning the specific quality of taste reinforcement in larval Drosophila |
Catching primal functional changes in early , ‘very far from disease onset’ ( VFDO ) stages of Huntington’s disease is likely to be the key to a successful therapy . Focusing on VFDO stages , we assessed neuronal microcircuits in premanifest Hdh150 knock-in mice . Employing in vivo two-photon Ca2+ imaging , we revealed an early pattern of circuit dysregulation in the visual cortex - one of the first regions affected in premanifest Huntington’s disease - characterized by an increase in activity , an enhanced synchronicity and hyperactive neurons . These findings are accompanied by aberrations in animal behavior . We furthermore show that the antidiabetic drug metformin diminishes aberrant Huntingtin protein load and fully restores both early network activity patterns and behavioral aberrations . This network-centered approach reveals a critical window of vulnerability far before clinical manifestation and establishes metformin as a promising candidate for a chronic therapy starting early in premanifest Huntington’s disease pathogenesis long before the onset of clinical symptoms .
Huntington’s disease is caused by the expansion of a CAG repeat in the open-reading frame of the huntingtin gene ( HTT ) , which translates into an expanded glutamine stretch in the aberrant , mutant protein ( mHTT ) . Huntington’s disease has primarily been described as a late-onset neurodegenerative disease . However , it is preceded in its premanifest period by a prolonged presymptomatic phase followed by a prodromal phase with hardly detectable and unspecific symptoms occurring far before classical Huntington’s disease becomes apparent ( Ross et al . , 2014 ) . These symptoms include reduced impulse control , social disengagement , low conversational participation , reduction of the concentration span and decline of clearly defined cognitive domains ( Stout et al . , 2011; Predict-HD Investigators of the Huntington Study Group et al . , 2007; Ross et al . , 2014; Labuschagne et al . , 2016 ) and are accompanied by slight changes of cortical network topology and functional connectivity in resting state fMRI measures ( PREDICT-HD investigators of the Huntington Study Group et al . , 2015; Wolf et al . , 2012 ) . Importantly , such subtle network dysregulations may occur even earlier than the described prodromal symptoms in a very far from disease onset ( VFDO ) premanifest stage ( Figure 1a ) . This stage reaches back more than 15–20 years before motor symptoms become visible in patients and far before protein aggregates and neurodegeneration are observed . In Huntington’s disease occurrence of such very early changes is supported by the observation of early deficits in premanifest Huntington’s disease mutation carriers , such as loss of phosphodiesterase 10A in the occipital lobe up to 47 years prior to disease onset ( summarized in Wilson et al . , 2017 ) . Also , the ability to perform complex visuospatial orientation , such as visual search , seems to be altered even in pre-manifest stages far from clinical diagnosis ( Labuschagne et al . , 2016 ) . We hypothesize that primal functional changes in the VFDO stage of premanifest Huntington’s disease open up very early vulnerable windows for disease development and preventive therapy prior to neuronal loss and may also provide promising early biomarkers for therapy development ( Mehler and Gokhan , 2000; Kerschbamer and Biagioli , 2015; Humbert , 2010; Cepeda et al . , 2003 ) . It seems that the visual cortex is one of the first regions that are functionally affected during disease development in Huntington’s disease ( Labuschagne et al . , 2016; Dogan et al . , 2013 ) . In order to identify primal network changes , we have here established a network-centered approach and focused on microcircuit function in layer 2/3 of the visual cortex at an early premanifest stage in a mouse model of Huntington’s disease corresponding to the VFDO stage in premanifest Huntington’s disease . We have used two-photon imaging using fluorescent indicators of intracellular Ca2+ in order to resolve the functional architecture of intact cortical microcircuits in vivo with single neuron resolution ( Grienberger and Konnerth , 2012 ) . We show that even at the early age of 10 – 15 weeks ( young adults compared to humans ) , the entire cortical microcircuit shifts towards a more excitable state characterized by a complex change in neuronal activity pattern and hyperactive neurons . These findings are accompanied by changes in animal behavior , including a decrease in anxiety . No effective and curative treatment has been developed for Huntington’s disease so far ( Frank , 2014 ) . A chronic drug therapy that commences early on in VFDO stages in premanifest Huntington’s disease and ameliorates early dysregulations as the potential origin of pathogenic processes and disease spreading is therefore a promising and necessary strategy . In Huntington’s disease animal models , even short-term reduction of protein load through RNA interference and antisense strategies has beneficial effects on disease phenotypes and progression lasting several months after intervention ( Stanek et al . , 2014; DiFiglia et al . , 2007; Keiser et al . , 2016 ) . We have recently shown that mRNAs carrying CAG repeats bind to a protein complex containing the ubiquitin ligase midline 1 ( MID1 ) in a repeat size-dependent manner . Through ubiquitination MID1 regulates PP2A ( protein phosphatase 2A ) and mTOR ( mechanistic target of rapamycin ) activities and the translation of associated mRNAs ( Krauss et al . , 2013; Griesche et al . , 2016 ) . Disruption of the MID1/PP2A/mTOR protein complex leads to an increase of PP2A activity , a decrease of mTOR activity and a reduction of translation rates of mRNAs with expanded CAG repeats ( Krauss et al . , 2013 ) . We show here that the type II diabetes drug metformin interferes with the MID1/PP2A/mTOR protein complex and significantly reduces the translation rate of Htt mRNA , resulting in a reduction of aberrant Htt protein production in vitro and in vivo in the Hdh150 mouse model . Notably , in Hdh150 mice in vivo metformin , when given early in the VFDO stage , and chronically in the drinking water , fully reverses both early neuronal network dysregulations and behavioral aberrations .
Since the visual cortex is one of the first regions affected by the disease ( Dogan et al . , 2013; Labuschagne et al . , 2016 ) , we focused on layer 2/3 of the visual cortex of ∼12 weeks old , heterozygous knock-in mice expressing expanded Htt with 150 glutamine repeats ( Hdh150 ) , in the lightly anesthetized mouse . This time corresponds to the VFDO in presymptomatic Huntington’s disease ( Figure 1a , b ) . We focused our analysis on male mice , thereby minimizing the influence of hormonal fluctuations on network activity . This approach is in line with a recent study in the field of Alzheimer’s disease ( Iaccarino et al . , 2016 ) , using males only . We employed two-photon Ca2+ imaging in vivo using the synthetic Ca2+ indicator Oregon-Green BAPTA1 ( OGB-1 ) AM to monitor the suprathreshold activity of a neuronal microcircuit with cellular resolution , typically comprising around 200 neurons ( Kerr et al . , 2005 ) ( Figure 1c , Figure 1—figure supplement 1a , b ) . Events from astrocytes , which are also stained by OGB-1 , were excluded from this neuronal network based on their temporal dynamics ( Figure 1—figure supplement 1c–f , Figure 1—source data 1 ) . By imaging at different depths , we found a similar spatial extent of OGB-1 staining in the visual cortex of Hdh150 and control wild-type ( WT ) mice ( Figure 1—figure supplement 1b ) . Figure 1—video 1 shows two-photon images acquired from the pial surface illustrating a homogenous OGB-1 staining in layers 1 , 2 and 3 of mouse visual cortex . We observed no difference in the density of stained cells ( Figure 1c and e , Figure 1—source data 1 , WT: 1413 ± 74 cells/mm2 ( n = 11 mice ) , Hdh150 1456 ± 90 cells/mm2 ( n = 10 mice ) , Mann-Whitney test p=0 . 8 ) . To identify changes in microcircuit activity , we assessed spontaneous activity in vivo in the cortical microcircuit of WT and Hdh150 mice , which reliably reflects the functional microarchitecture of sensory cortical areas ( Miller et al . , 2014 ) . Figure 1—video 2 shows an example of in vivo two-photon Ca2+ imaging exhibiting ongoing spontaneous activity in layer 2/3 of mouse visual cortex with single-cell resolution . Both WT and Hdh150 mice exhibited silent and active cells ( Figure 1d ) , but a significantly higher proportion of active neurons was detected in Hdh150 mice , indicating a more active network ( Figure 1f , Figure 1—source data 1 , WT: 65 . 1 ± 3 . 5% ( n = 8 mice ) , Hdh150: 77 . 1 ± 3 . 5% ( n = 6 mice ) , unpaired t-test p<0 . 05 ) . No spatial clustering of silent or active cells could be observed in either group ( Figure 1d ) . Next , we analyzed the frequency of Ca2+ transients and activity patterns in the population of active neurons ( Figure 1—source data 2 ) . Notably , the frequency of Ca2+ transients was significantly higher in Hdh150 compared to WT mice ( Figure 1g–h , Figure 1—source data 1 , WT: 0 . 74 ± 0 . 06 trans/min ( n = 8 mice ) , Hdh150: 1 . 2 ± 0 . 14 trans/min ( n = 6 mice ) , Mann-Whitney test , p<0 . 01 ) . No difference was found in the mean area under the curve ( AUC ) of Ca2+ transients between WT and Hdh150 mice ( Figure 1i , Figure 1—source data 1 , WT: 39 . 8 ± 3 . 9 ( dF/f ) *s , Hdh150: 39 . 9 ± 3 . 6 ( dF/f ) *s , unpaired t-test , p=0 . 98 ) indicating that on average , the mean number of underlying action potentials for each individual calcium transient was not different in Hdh150 mice . We furthermore analyzed the distribution of neurons according to their Ca2+ transient frequency . The cumulative probability distribution of the activity of individual neurons in Huntington’s disease mice was shifted toward higher frequencies indicating that the overall neuronal activity transitioned toward a more excitable network ( Figure 1j , Figure 1—source data 1 , two-way ANOVA p<0 . 0001 ) . We sub-classified the active neurons into three functional subgroups according to their transient frequency: low , medium and hyperactive ( Figure 1k , Figure 1—source data 1 ) . Notably , we identified a unique subgroup in VFDO Hdh150 mice: hyperactive neurons ( Figure 1l , Figure 1—source data 1 , Chi-square test p<0 . 01 ) . This subgroup was accompanied by a reduction in the number of neurons with low activity , which corroborates the shift of the microcircuit activity . In later stages of Alzheimer’s disease , hyperactive cells were shown to cluster near amyloid plaques ( Busche et al . , 2008 ) . A similar scenario might occur in Huntington’s disease . To clarify whether the cortex of VFDO Hdh150 mice is affected by mHtt aggregates - a later stage Huntington’s disease hallmark - we performed immunohistochemistry in Hdh150 and WT animals . We observed only diffuse and non-aggregated Htt immunoreactivity in the cytoplasm of neurons in cortical areas of both WT and Hdh150 mice ( Figure 1—figure supplement 2a ) in accordance with previous studies ( Lin et al . , 2001 ) . In addition , no reactive astrocytes were observed ( Figure 1—figure supplement 2b ) and only few cells that were stained for activated caspase-3 , an apoptotic marker , in cortical areas of WT and Hdh150 mice ( Figure 1—figure supplement 2c ) as previously observed ( Yu et al . , 2003 ) . In order to test whether hyperactive cells cluster in the Hdh150 mouse model , spatial distance between every pair of neurons was quantified ( Figure 1—figure supplement 3c , d , Figure 1—source data 1 ) . We observed no significant differences between the permutations of functional subgroups in premanifest Hdh150 and control mice , reflecting a rather homogenous distribution of all functional subgroups ( Figure 1—figure supplement 3c , d , Figure 1—source data 1 , Mann-Whitney , not significant , see Table 1 for p-values ) . This finding was confirmed in randomized data in which cell location was kept but functional identity was randomly permutated ( Figure 1—figure supplement 3a , b and d , Figure 1—source data 1 ) . Taken together , clustering of hyperactive cells could not be observed in VFDO Hdh150 animals and cortical hyperactivity is independent of mHtt aggregation , apoptotic cell death or astrogliosis at this presymptomatic VFDO stage . An important aspect of neuronal information processing is the optimization of encoding strategies . In the visual cortex , encoding of information is characterized by sparse and precisely timed neuronal activity . Cortical activity is defined by transiently co-active ensembles of neurons acting as a functional unit ( Miller et al . , 2014 ) . To capture these spatiotemporal dynamics in the microcircuit , we analyzed synchronicity of the transients between all pairs of neurons ( Figure 2a ) by calculating Pearson’s correlation coefficient ( Pearson’s r ) for every pair . First , we confirmed that the level of synchronicity within a healthy cortical microcircuit is drastically higher than the random synchronicity in shuffled data ( Figure 2b , Figure 2—source data 1 , WT: 0 . 024 ± 0 . 005 , WT rand: −0 . 0003 ± 0 . 0003 , Mann-Whitney test p<0 . 0001 ) . Notably , the synchronicity was even higher in Hdh150 mice compared to healthy controls ( Figure 2b , Figure 2—source data 1 , WT: 0 . 02 ± 0 . 005 , Hdh150: 0 . 04 ± 0 . 006 , Mann-Whitney test p<0 . 05 ) . Next , we compared the synchronicity level in all functional subgroups ( Figure 2c , Figure 2—source data 1 ) . The pairs involving the low activity subgroups ( LL and LM ) showed no differences in WT and Hdh150 mice ( Mann-Whitney test , in LL p=0 . 5 and in LM p=0 . 7 ) . However , Pearson’s r increased significantly for the medium-to-medium ( MM ) pairs in Hdh150 . Pearson’s r was even higher for medium-hyperactive ( MH ) and hyperactive-hyperactive ( HH ) compared to low-low ( LL ) pairs ( Mann-Whitney test compared to LL in Hdh150 mice: in MM p<0 . 05 , in MH and HH p<0 . 01 ) . To assess whether the increased synchronicity in the VFDO Hdh150 mice occurred merely due to the higher number of transients , we used randomized data with unchanged frequency but temporally shuffled transients ( Figure 1—figure supplement 3a and b ) . Pearson’s r in the randomized data was nearly zero for all , including the high frequency subgroups in WT and Hdh150 mice ( Figure 2—figure supplement 1a and b , Figure 2—source data 1 ) . This finding argued against the possibility that the increased Pearson's r in the experimental data occurred by chance . We next asked whether functional ensembles with a high level of synchronicity are located in spatial vicinity to each other , by testing whether Pearson's r changed with physical distance between the pairs of neurons ( Figure 2d ) . An inverse linear relationship was observed between Pearson’s r and the pairwise distance in both the WT and Hdh150 mice ( Figure 2d , Figure 2—source data 1 , two-way ANOVA p<0 . 0001 , WT vs Hdh150 ) suggesting that two closely located neurons have a higher probability to fire together . This is consistent with similar findings in the forelimb motor cortex of head-restrained mice ( Dombeck et al . , 2009 ) . Randomization of the data abolished the inverse relationship between Pearson's r and distance ( Figure 2e , Figure 2—source data 1 , two-way ANOVA p=0 . 3 , WT rand vs Hdh150 rand ) . Since many studies , especially in premanifest mutation carriers , have linked Huntington’s disease pathology to changes in metabolism ( Damiano et al . , 2010; Duan et al . , 2014; Jin and Johnson , 2010; Labbadia and Morimoto , 2013; Mochel and Haller , 2011 ) , hyperactivity might mirror metabolic dysregulation in subgroups of cortical cells . We used mitochondrial respiration as a marker of metabolic functionality and quantified mitochondrial respiration using high-resolution respirometry of cortical tissues . Mitochondrial respiration was unchanged in cortical tissue of Hdh150 mice suggesting that the observed neuronal hyperactivity occurs prior to metabolic changes ( Figure 2—figure supplement 2a–c , Figure 2—source data 1 , Mann-Whitney test , not significant , see Table 1 for p-values ) . Next , we asked whether hyperactivity and increased synchronicity of cortical networks are associated with behavioral changes in the VFDO animals . A visual discrimination task did not show any aberrations in the Hdh150 animals ( Figure 3—figure supplement 1a–d; Figure 3—video 1 , Figure 3—source data 1 , two-way ANOVA , not significant , see Table 1 for p-values ) . In contrast , in an open-field test premanifest VDFO Hdh150 animals moved significantly more to the center than WT littermates suggesting anxiolytic effects of the VFDO changes ( Figure 3a , b , Figure 3—source data 1 , Mann-Whitney test , p=0 . 03 ) . Distance travelled ( as a measure of motility ) did not differ between groups ( Figure 3—figure supplement 1e , Figure 3—source data 1 , Mann-Whitney test , not significant , see Table 1 for p-values ) . Taken together , we have found network hyperactivity in the cortex of VFDO Hdh150 mice combined with anxiolytic behavior . Based on our previous observations that the MID1/PP2A/mTOR protein complex regulates the translation of mHtt protein ( Krauss et al . , 2013 ) and that treatment with metformin interferes with the MID1 complex ( Kickstein et al . , 2010 ) , we hypothesized that metformin inhibits the MID1/PP2A/mTOR-mediated protein synthesis of mutant mHtt and is therefore a promising candidate molecule to reduce mHtt load and reverse symptoms associated with Huntington’s disease . In order to test for an effect of metformin on mHtt protein load and aggregation , HEKT cells stably expressing FLAG-tagged exon 1 of human mHTT carrying 83 CAG repeats were treated with 1 mM or 2 . 5 mM metformin , or with vehicle for 48 hr . Aggregation was quantified in a filter retardation assay . Metformin reduced the amount of aggregated FLAG-HTT in a concentration-dependent manner ( Figure 4a , Figure 4—source data 1 , Mann-Whitney test , control vs 2 . 5 mM metformin p=0 . 02 ) . To test whether the metformin effect on human exon 1 mHTT aggregates is mediated by a blockade of the MID1 protein complex , we depleted MID1 by siRNA-mediated knockdown in the cell line expressing FLAG-HTT exon 1 with 83 CAG repeats , in presence or absence of metformin . While depletion of MID1 reduced mHTT aggregation , no additive effect of metformin treatment on mHTT aggregates was observed , suggesting that MID1 and metformin indeed act through the same pathway ( Figure 4b , Figure 4—source data 1 , Mann-Whitney test , control siRNA vs MID1 siRNA p=0 . 009; control siRNA vs MID1 siRNA +metformin p=0 . 02 ) . We had shown previously that the MID1/PP2A/mTOR protein complex regulates the translation efficiency of the human HTT mRNA in a repeat-dependent manner ( Krauss et al . , 2013 ) . We therefore looked at a possible influence of metformin on the protein synthesis rate of mHTT exon one protein using a previously described FRAP ( Fluorescence recovery after photo bleaching ) - based assay that allows monitoring of protein translation rates in living cells ( Krauss et al . , 2013 ) . We detected a clear reduction in the protein synthesis rate of a GFP-Htt fusion protein carrying 49 repeats in the metformin-treated samples in a concentration-dependent manner in primary neurons ( Figure 4c , Figure 4—source data 1 , RM two-way ANOVA p=0 . 008 ) . This effect was confirmed in N2A cells ( Figure 4—figure supplement 1a , Figure 4—source data 1 , RM two-way ANOVA p=0 . 03 ) . To further support the contribution of the MID1/PP2A/mTOR protein complex and PP2A activity to this effect , the GFP-Htt transfected cells were subsequently either ( i ) mock treated , ( ii ) treated with only metformin , ( iii ) treated with ocadaic acid ( OA ) , or ( iv ) co-treated with metformin and OA . OA is an inhibitor of PP2A activity . As expected , OA significantly increased translation rates of the GFP-Htt reporter mRNA and metformin did not have a reducing effect on the translation rates in cells co-treated with OA suggesting that indeed the metformin effect is mediated by PP2A activity ( Figure 4d , Figure 4—source data 1 , RM two-way ANOVA p=0 . 002 ) . To analyze metformin effects on early signs of pathology in vivo , VFDO Hdh150 mice were fed with , or without 5 mg/ml metformin in the drinking water . Metformin did not significantly reduce drinking volume ( Figure 4—figure supplement 1b , Figure 4—source data 1 , RM two-way ANOVA not significant , see Table 1 for p-values ) . After 3 weeks of treatment , we looked at phosphorylation patterns of the PP2A/mTOR target S6 and the amount of mHtt protein relative to wild-type Htt in whole brain tissue . The metformin-treated group showed a significant reduction of S6 phosphorylation suggesting an increase in PP2A activity ( Figure 4e and f , Figure 4—source data 1 , unpaired t-test p=0 . 05 ) . At the same time , a slight tendency ( not significant ) of reduced mHtt was detected suggesting that metformin has an influence on mHtt expression ( Figure 4—figure supplement 1c and d , Figure 4—source data 1 , unpaired t-test not significant , see Table 1 for p-values ) . A significant reduction of mHtt expression , however , became clearly visible after 11 weeks of treatment in cortical tissue ( Figure 4g–j , Figure 4—source data 1 , unpaired t-test , mHtt/wtHtt p=0 . 05 , mHtt/Gapdh p=0 . 002 , wtHtt/Gapdh p=0 . 9 ) . Our data suggest metformin as a promising molecule to interfere with VFDO Huntington’s disease biochemical and cellular pathological changes . We initially used a C . elegans model of polyQ-mediated diseases to test whether metformin effectively ameliorates disease symptoms in an easily controllable model with short lifespan . The C . elegans worms carry a transgene encoding YFP-tagged Q40 polypeptide in body wall muscle cells ( Morley et al . , 2002 ) . Adult Q40::YFP nematodes exhibit intracellular aggregates of polyglutamine-containing protein and develop progressive paralysis over time , which is reflected in significantly reduced motility . We counted aggregates in metformin-treated and untreated nematodes . Moreover , we assessed their motility in a liquid thrashing experiment , in which worms are placed in liquid and the frequency of lateral swimming ( thrashing ) movement is analyzed as a measure of motility . We found that 5 days of metformin treatment reduced the number of intracellular inclusion bodies significantly and rescued motility impairment ( Figure 4—figure supplement 2a and b , Figure 4—source data 1 , Mann-Whitney , p<0 . 0001 ) . Since bacteria can metabolize metformin ( Cabreiro et al . , 2013 ) , we confirmed the results on heat-inactivated OP50 bacteria ( Figure 4—figure supplement 2c , Figure 4—source data 1 , Mann-Whitney test , control vs 5 mM metformin p=0 . 008 , control vs 10 mM metformin p<0 . 0001 ) . siRNA-mediated knock-down of arc-1 , the C . elegans MID1 homolog , leads to a reduction of inclusion bodies and improved motility similar to the metformin effects ( Figure 4—figure supplement 2d and e , Figure 4—source data 1 , Mann-Whitney test , p<0 . 0001 ) and confirms that the MID1/PP2A/mTOR protein complex underlies metformin effects . We then asked whether metformin could rescue the altered cortical activity in vivo in Hdh150 mice . Metformin treatment did not affect the density of OGB-1 stained cells ( Figure 5—figure supplement 1a , b , Figure 5—source data 1 , WT: 1413 ± 74 cells/mm2 ( n = 11 mice ) , Hdh150: 1456 ± 90 cells/mm2 ( n = 10 mice ) , WT met: 1351 ± 65 cells/mm2 ( n = 9 mice ) and Hdh150 met: 1448 ± 60 cells/ mm2 ( n = 6 mice ) , unpaired t-test , not significant , see Table 1 for p-values ) . Notably , 3 weeks of metformin treatment in the drinking water completely restored the proportion of active cells ( Figure 5b , Figure 5—source data 1 , Hdh150: 77 . 2 ± 3 . 5% ( n = 6 mice ) , Hdh150 met: 64 . 4 ± 4 . 1% ( n = 6 mice ) , Mann-Whitney test p<0 . 05 ) and the average frequency of Ca2+ transients ( Figure 5a , c , Figure 5—source data 1 , Hdh150: 1 . 2 ± 0 . 1 trans/min ( n = 6 mice ) , Hdh150 met: 0 . 7 ± 0 . 06% ( n = 6 mice ) , Mann-Whitney test p<0 . 05 ) . The individual traces of treated Hdh150 animals were indistinguishable from untreated WT animals ( Figure 5a , Figure 5—figure supplement 1c ) . Importantly , in these experiments , metformin acts specific on dysregulated network components . Only the AUC of calcium transients was slightly yet significantly affected in WT mice which might be due to an increase of baseline calcium concentration induced by the activation of MID1/PP2A/mTOR signaling pathway ( Figure 5—figure supplement 1c , Figure 5—source data 1 ) . One of the hallmarks of the VFDO Hdh150 mice was the distinct functional subgroup of hyperactive neurons ( Figure 1 ) . We thus assessed the effect of metformin treatment on the relative proportions of low , normal and hyperactive subgroups . Treatment with metformin in Hdh150 mice led to the complete abolishment of the hyperactive subgroup ( Figure 5d , Figure 5—source data 1 , two-way ANOVA p<0 . 0001 and Figure 5e , Figure 5—source data 1 , Chi-square test not significant , see Table 1 for p-values ) and restored the relative proportion of functional subgroups . The spatial distribution of functional subgroups remained unchanged by metformin treatment ( Figure 5—figure supplement 1f , Figure 5—source data 1 , Mann-Whitney test , not significant , see Table 1 for p-values ) . This complete restoration of dysregulated microcircuit activity was also evident in the cumulative frequency distribution ( Figure 5d , Figure 5—source data 1 ) and all other network measures found to be aberrant in young Hdh150 animals , such as synchronicity ( Figure 5f and g , Figure 5—figure supplement 1d–e , Figure 5—source data 1 ) . Accompanying the network dysfunction , we observed anxiolytic behavior in the VFDO Hdh150 mice ( see Figure 3a–b ) . Treatment with metformin fully reversed the anxiolytic behavior ( Figure 5h and i , Figure 5—source data 1 , Mann-Whitney test , WT vs Hdh150 p=0 . 002 , Hdh150 vs Hdh150 met p=0 . 002 , Hdh150 vs WT met p=0 . 03 , WT vs Hdh150 met p=0 . 8 ) . In conclusion , our data suggest that metformin , by interfering with the translation rate of mHtt protein , reduces protein load in cell culture and in vivo and reverses early Huntington’s disease-related network dysregulations as well as anxiety-related behavioral aberrations in mice . Furthermore , usage of metformin in adult C . elegans , a model of polyQ disease , also significantly influences inclusion bodies formation and motility .
Here , we have identified a dysregulation of spontaneous neuronal activity in the visual cortex in a very early stage of a mouse model of Huntington’s disease that corresponds to a very early stage in the premanifest stage in Huntington’s disease mutation carriers ( Figure 1a , b ) , which might indeed be a distinct pathological disease stage which is very far from disease onset ( VFDO ) . Correspondingly , the visual cortex is one of the first structures affected in patients ( Labuschagne et al . , 2016; Dogan et al . , 2013 ) . This dysregulation is characterized by an increase in cortical network activity patterns , the emergence of a functional subgroup of hyperactive neurons , and enhanced synchronicity . Overall visual cortex functioning seems to be preserved at this early time point of Huntington’s disease course . Network changes are accompanied by subtle behavior alterations including an anxiolytic phenotype , suggesting that at least part of the brain-wide circuitry exhausted its compensational reserve . Anxiety-related abnormalities including anxiolytic behavior have previously been described in the preclinical phase in several other rodent Huntington’s disease models ( ( Nguyen et al . , 2006 ) , reviewed in [Pouladi et al . , 2013] ) . So far , changes described here represent the earliest identified abnormalities in cortical pathophysiology and behavior in heterozygous Hdh150 animals , which closely resemble the human disease ( Lin et al . , 2001; Heng et al . , 2007; Brooks et al . , 2012; Tallaksen-Greene et al . , 2005 ) . Furthermore , we show that the type II diabetes drug metformin inhibits the translation of mHtt protein and thereby decreases mHtt protein load in vitro and in vivo . Promisingly , this leads to a complete restoration of VFDO network activity patterns as well as behavior abnormalities under chronic metformin therapy . Our data report primal changes in cortical network function in the VFDO stage of Huntington’s disease . A recognition of the network as a pathophysiological entity has recently been suggested in the context of Alzheimer’s disease ( Busche et al . , 2008; Iaccarino et al . , 2016 ) . Moreover , focus has shifted away from the mechanisms accompanying neuronal and network degeneration and instead moved toward small and subtle functional changes at very early stages of the disease when irrevocable damage to the network has not yet occurred ( Busche and Konnerth , 2016 ) . Indeed , hyperactive neurons are associated with both advanced and early stages of Alzheimer’s disease , independent of plaque formation ( Busche et al . , 2012; Busche et al . , 2008 ) . In addition , evidence points toward hyperactive neurons preventing the cortex-wide propagation of slow oscillations in early Alzheimer’s disease ( Busche et al . , 2015 ) . We here describe a similarly distinct hyperactive phenotype in very early stages of Huntington’s disease . We conclude that neuronal hyperactivity may be a principle mechanism that develops early not only in Alzheimer’s disease but also in other neurodegenerative diseases . Thus , the notion of early network dysregulation as a therapeutic target may have broad implications . Similar to early stages of Alzheimer’s disease hyperactive neurons in Hdh150 animals emerge in the absence of aggregate formation . Also hyperactive cells do not cluster , and reactive astrocytes or cells with activated caspase-3 as a marker of early apoptosis are not found in these early stages of Huntington’s disease . Furthermore , cortical neurons did not exhibit metabolic dysregulation as measured in a mitochondrial respiration assay . Therefore , we may postulate that the cortex merely responds to early pathophysiological events already commencing in subcortical regions , e . g . the striatum . This is well in line with the current emerging hypotheses of disease progression in Alzheimer’s disease . Young Alzheimer’s disease animals develop hyperactivity in a plaque-independent fashion first in the hippocampus ( possibly due to higher vulnerability of the hippocampus in Alzheimer’s disease ) , followed by a similar hyperactivity pattern in the cortex later in the disease process ( Busche et al . , 2012; Busche et al . , 2008 ) . Furthermore , in Alzheimer’s disease patients , degeneration of cortical projection targets of the hippocampus is associated with hippocampal hyperactivity indicating a connectivity-based spread of network dysregulations eventually leading to neurodegeneration ( Putcha et al . , 2011 ) . Our data indeed suggest an altered activity in subcortical drivers , since unspecific alteration of excitability in individual neurons is unlikely to lead to an increase in synchronicity , but rather would result in a random increase in firing . Our data was collected at very early stages of the disease in the Hdh150 mouse model , which corresponds to the VFDO stage in Huntington’s disease patients . The importance of expression of mutant Htt protein during very early phases for disease development has been demonstrated in the BACHD:CAG-CreERT2 mouse ( Molero et al . , 2016 ) . With the help of tamoxifen treatment expression of mutant Htt was turned off early postnatally . Still the typical symptoms of Huntington’s disease at three and nine months of age were observed in the animals . We conclude that only at the VFDO stages , when cellular and network degeneration have not yet been established , preventive strategies will be most effective; only then can we still rescue small homeostatic shifts , prevent spreading and potentially stabilize network function . Phenotype reversal could be demonstrated in a tetracyclin-dependent conditional mouse model for Huntington’s disease . Both neuropathological findings and behavior aberrations were found to disappear when mHtt protein production was stopped through a tet-off regulation mechanism in the adult animal ( Yamamoto et al . , 2000 ) . Additional support for the beneficial effect of suppression of aberrant protein production on the Huntington’s disease phenotype stems from several studies with RNA interference ( siRNA ) , or antisense oligonucleotides showing that gene suppression reducing mHtt protein load by 40% or more , is sufficient to significantly ameliorate the Huntington’s disease phenotype ( HD iPSC Consortium , 2017; Harper et al . , 2005; Keiser et al . , 2016; Stanek et al . , 2014 ) . These studies have demonstrated , that ( i ) the earlier suppression takes place the more robust and beneficial effects on behavior phenotypes are [reviewed in ( Keiser et al . , 2016 ) ] and ( ii ) that even transient suppression of Htt protein during early disease stages was sufficient to obtain long-term effects on the disease phenotype lasting for months , far beyond the treatment period ( Lu and Yang , 2012; Kordasiewicz et al . , 2012 ) . However , developing siRNA and antisense oligonucleotides technologies into therapeutics for clinical use is difficult and a long way to go . Difficulties here include toxicity and modes of delivery: so far oligonucleotides have to be regularly injected into the cerebrospinal fluid , which is a huge effort for patients and physicians . Furthermore , a short N-terminal fragment of the mHtt protein , mHttexp1p , that is produced by incomplete exon one splicing and a short poly-adenylated mRNA in several animal models as well as in Huntington’s disease patients rather than full-length mHtt protein was found to be particularly pathogenic . Its occurrence correlates well with age of onset and severity of the disease . This short mRNA is difficult to target by oligonucleotide strategies ( Neueder et al . , 2017; Sathasivam et al . , 2013 ) . We show here that a well-known , widely used , orally delivered small compound , metformin , suppresses mHtt production by targeting both , full-length and mHttex1p , in vitro and in vivo and thereby significantly reduces mHtt protein load , which makes it a very promising candidate for chronic early onset Huntington’s disease therapy . Metformin is an FDA-approved , inexpensive biguanide that has been used in patients with Type II diabetes for decades and is under discussion for cancer preventive therapy ( Demir et al . , 2014; Micic et al . , 2011 ) . Very recently , metformin has been shown to rescue core phenotypic features in a mouse model for fragile X-syndrome , a neurodevelopmental disorder , by normalizing ERK signalling ( Gantois et al . , 2017 ) . Metformin had been brought in as a promising compound in Huntington’s disease previously . It has been found to protect cells from the toxicity of mutant Huntingtin protein in a cell culture model ( Jin et al . , 2016 ) . In a study on R6/2 animals , a very aggressive model for Huntington’s disease , Ma and colleagues had found a significant effect on survival rates and hind clasping in male animals only when given metformin in the drinking water starting from week 5 ( Ma et al . , 2007 ) . In relation to the phenotype in the R6/2 animals that develops severe aberrations from 4 weeks on this is a late time point and would be placed in the motor phase stage when projected to the phenotypic time line given in Figure 1a . We hypothesize here that preventive treatment at a very early stage is important to substantially and stably influence the disease . This is in agreement with observations in a mouse model for spinocerebellar ataxia I , another neurodegenerative disease based on CAG expansion and studies with antisense oligonucleotides in a Huntington’s disease model . Both studies show that gene suppression has more stable effects on the phenotype when performed early enough ( Kordasiewicz et al . , 2012; Rubinsztein and Orr , 2016; Zu et al . , 2004 ) . In line with that only two phenotypic features had been found to react on late metformin treatment- survival rates and frequency of clasping- in the R6/2 animals . Also and again as expected , effect size on animal survival was quite small ( p=0 . 02 ) . Gender differences in response rate seen in this study can possibly be explained by gender differences in disease development and progression at the motor stage , which had been observed in several mouse models ( Menalled et al . , 2009 ) . When disease progression differs , differences in blood brain barrier permeability can be expected ( reviewed in [Sweeney et al . , 2018] ) which then is likely to lead to gender specific variation in bioavailability of metformin in the brain . In contrast to this study we here show that in vivo metformin has highly significant effects ( p=0 . 03 to p<0 . 0001 ) already on primal changes in the very early , VFDO phases of Huntington’s disease making metformin a promising compound for the development of a therapeutic scheme that is based on early prevention of pathology development . While we focused on male mice in this study , to reduce physiological variability due to hormone fluctuations , at the VFDO stage brain barrier changes are not expected to influence bioavailability of metformin . Metformin was suggested to lead , through AMPK activation , to a reduction of mHtt aggregates in vitro ( Walter et al . , 2016; Vázquez-Manrique et al . , 2016 ) . In our previous work , however , we have shown that in cortical neurons metformin does not induce phosphorylation of the AMPK target ACC at all and only when given chronically it induces phosphorylation of AMPK itself in vitro . When giving 5 mg/ml metformin in the drinking water for 16 – 24 days to wildtype animals , while phosphorylation of S6 is significantly reduced , AMPK phosphorylation does not change in whole brain extracts ( Kickstein et al . , 2010 ) . This indicates that AMPK activation is likely to depend on the dose . In WT animals as in preclinical Huntington’s disease animals the blood-brain barrier is intact , which limits bioavailability of metformin in the brain . Metformin concentrations needed to influence mTOR/PP2A activity seem to be significantly lower than those needed to influence AMPK activity . Like in the Kickstein et al . paper , in the present study , we used 5 mg/ml metformin in the drinking water , a concentration at which AMPK activation is not expected , but as we show here in the Hdh150 animals , metformin has a significant effect on the phosphorylation of the mTOR/PP2A target S6 . The effect of Htt loss on brain function is still under debate . SiRNA studies suggest that postnatal reduction of endogenous Htt protein is well tolerated ( summarized in [Keiser et al . , 2016] ) . However , conditional knock-out animals with a perinatal loss of around 40% of Htt protein in the forebrain show a neurodegenerative phenotype ( Dragatsis et al . , 2000 ) . Likewise , depletion of Htt protein in the adult brain leads to progressive behavior deficits ( Dietrich et al . , 2017 ) . We demonstrate here that metformin has a very specific effect on the expression of mHtt protein only , leaving wild-type Htt that is produced from the non-mutated allele in dominant Huntington’s disease untouched ( Figure 4 ) . This makes metformin the only compound available at present with a specific effect on only mutant but not wild-type Htt protein . In support of an effect of metformin in Huntington’s disease patients an in silico comparison of cognitive performance of Huntington’s disease patients treated with metformin was performed . In this study , using the Enroll patient cohort , it was shown that diabetic Huntington’s disease patients in the manifest stage treated with metformin had a better cognitive performance than Huntington’s disease patients not treated with metformin ( Hervás et al . , 2017 ) . Our data indicate that metformin treatment reverses all cortical network dysregulations in vivo in the premanifest VFDO Hdh150 mice including functional sub-group distribution , frequency and synchronicity . Regaining network stability shows promise for ameliorating the molecular pathophysiology , probably by activating intrinsic repair mechanisms as shown in the context of Alzheimer’s disease ( Iaccarino et al . , 2016; Keskin et al . , 2017 ) . Following this network-centered view , restoration of network functions might also prevent secondary damage to the neuronal microcircuit . We therefore propose a shift in experimental treatment strategies: rather than exploring single pathways for target , we might also consider re-balancing network function in the VFDO stages of the disease . Taken together , our data provide evidence for the existence of a pathophysiological entity very far from onset of the manifest disease ( VFDO ) characterized by early homeostatic changes of network activity and , associated with that , subtle behavior alterations . The data also strongly support the observation that , similar to humans , the disease in mice develops over a long period of time . This provides an early critical window of vulnerability and gives opportunities for early therapeutic interference with disease development . So far , all attempts to develop a causative therapy for Huntington’s disease have been unsuccessful [summarized in ( Crook and Housman , 2011; Clabough , 2013 ) ] . In terms of therapeutic intervention , consideration should be given to a chronic treatment of mutation carriers , which covers the critical windows of vulnerability , as early as in the VFDO stages . Such a strategy avoids delaying intervention until clinical signs of the disease are evident , implying that substantial brain damage has already occurred . Our data suggest that metformin has the potential to reduce mHtt protein load and to substantially influence the early development of pathology and , as seen in a C . elegans model , protein aggregation and movement aberrations which are pathognomonic for later disease stages . It is an inexpensive substance , well known in long-term clinical usage and has a defined , relatively benign spectrum of side effects . Prescription to mutation carriers from young adulthood on ( or even earlier ) is possible and will cover these newly discovered critical windows of opportunity for therapy .
All experimental procedures were performed in accordance with institutional animal welfare guidelines and were approved by the state government of Rhineland-Palatinate , Germany ( G14-1-010 and G14-1-017 ) . WT littermates and heterozygous Hdh ( CAG ) 150 mice ( Hdh150 , RRID:IMSR_JAX:004595 ) carrying an extended CAG sequence ( ∼150 ) replacing the normal length CAG sequence in mouse Htt gene were obtained by crossing Hdh150 heterozygous with WT mice ( Lin et al . , 2001 ) . Male mice at 10 – 15 weeks of age were used to examine the change in neuronal network activity prior to disease onset and at 14 – 17 weeks of age to examine the effect of in vivo metformin treatment . Male mice at 12 – 16 weeks of age were used for behavior studies . Male mice at 13 weeks of age were used for immunohistochemistry . The mice were kept under specific-pathogen-free conditions on a 12 hr light/12 hr darkness cycle with free access to water and food . Mice were genotyped using the primers 5’-CCC ATT CAT TGC CTT GCT GCT AGG-3’ and 5’-CCT CTG GAC AGG GAA CAG TGT TGG-3’ ( Sigma-Aldrich ) producing 379- and 829-bp-long fragments for WT and mutant alleles , respectively . Mice were prepared for in vivo imaging under isoflurane ( 1 – 1 . 5% in pure O2 , AbbVie ) . Anesthesia depth was assessed by monitoring pinch withdrawal and respiration rate . Body temperature was kept at 37°C with a heating pad ( ATC 200 , World precision instruments ) . Local anesthesia ( 2% xylocaine , AstraZeneca ) was applied to the scalp . A custom-made recording chamber was glued to the skull with cyanoacrylic glue ( UHU ) followed by dental cement ( Paladur , Heraeus ) . Then , a craniotomy of 1 . 5 × 1 . 5 mm was performed using stainless steel drill bits . The position of the primary visual cortex was located according to brain atlas coordinates ( Bregma −3 to −4 . 5 mm , 2 – 3 mm lateral to the midline ) ( Paxinos and Franklin , 2001 ) . After surgery , the mouse was subjected to the two-photon imaging setup . The fluorescent Oregon-Green BAPTA1 AM ( OGB-1 AM , O6807 , Molecular Probes ) was bulked-loaded in the visual cortex as described previously ( Stosiek et al . , 2003 ) . Anesthesia level was continuously monitored by keeping the breathing rate at 100 – 110 breaths/min . High-speed two-photon Ca2+ imaging was performed in layer 2/3 ( 150 to 350 µm from the pial surface ) with an upright LaVision BioTec TrimScope II resonant scanning microscope , equipped with a Ti:sapphire excitation laser ( Chameleon Ultra II , Coherent ) and a 25x ( 1 . 1 N . A . , MRD77220 , Nikon ) or 40x ( 0 . 8 N . A . , NIRAPO , Nikon ) objective . The laser was tuned to 800 nm and fluorescence emission was isolated using a band-pass filter ( 525/50 , Semrock ) and detected using a GaAsP photomultiplier tube ( PMT; H7422-40 , Hamamatsu ) . The TriM Scope II scan head , equipped with a resonant scanner , imaged time-lapses ( 512 × 512 pixels , ~440 × 440 μm field of view ) at a maximum frame rate of 30 . 4 Hz . Time lapses were recorded for 5 – 8 min on average . Imspector software ( LaVision BioTec ) was used for microscope control and image acquisition . First , the number of cells loaded with OGB-1 was manually counted in ImageJ ( National Institutes of Health ) . The area containing all the cells was traced freehand and calculated by the software . Functional data were analyzed using custom-written functions in MATLAB R2011a ( Mathworks , Natick , MA ) and Igor Pro 6 . 22 – 6 . 37 ( Wavemetrics , Inc . , Lake Oswego , OR ) . The code is attached as Figure 1—source data 2 . Regions of interest ( ROIs ) were hand-drawn by tracing the outlines of OGB1-positive neurons . Fluorescence intensities were quantified by averaging pixels inside each ROI for every image in a sequence . The fluorescence values were normalized by user-defined baseline . Specifically , dF/F was defined as the following:dFF= mean fluorescence inside an ROImean user-defined baseline-1*100where the baseline is defined as a mean fluorescence from a 1–3 s silent period in the same ROI . The peak of a Ca2+ transient was defined as the first derivative to crossed zero , and the second derivative to be negative , and where the amplitude to be greater than three standard deviations ( SD ) above the mean . The peak location was corrected manually where necessary . Each dF/F trace , sampled at 15 . 2 – 30 . 4 Hz sampling frequency , was preprocessed by binomial Gaussian smoothing ( 20 – 40 iterations ) followed by a high pass filter . The baseline was estimated as the median of activity-free 10 s period preceding each peak . The foot and the tail of Ca2+ transients were determined as the first data point that fell within 0 . 5 SD of the baseline before and after the peak , respectively . The area under the curve was trapezoidal and measured between the foot and the tail . A distribution histogram of neurons according to their Ca2+ transient frequency was used to segregate neurons into three functional subgroups . The definition of hyperactive neurons ( >3 trans/min ) was determined by the absence of neurons above this limit in WT ( 0 . 4 ± 0 . 3% ) . The criterion used for low active neurons was set to comprise ~25% of the WT neuronal population ( 25 . 8 ± 3 . 9% ) . The distance between two neurons was calculated by Pythagorean theorem after the x , y coordinate was determined for each ROI . The randomization of experimental data comprised two steps: first , each raster plot was reassigned to a randomly selected ROI; then , the location of the individual raster event was shuffled randomly , except no spikes were allowed to occur within 1 s of each other . Mice were anesthetized with a mixture of ketamine/xylocaine and perfused transcardially with 4% paraformaldehyde ( #15710 , Life technologies ) in 0 . 1M phosphate buffer and brains were post-fixed . 50-µm-thick sections were sliced using a HM650 V vibratome ( ThermoFisher ) and collected in phosphate buffer saline ( PBS; Life technologies ) . Floating sections were incubated for 1 hr with PBS containing 4% natural goat serum ( NGS , ab7481 , Abcam ) or 4% natural donkey serum ( NDS , ab7475 , Abcam ) and 1% Triton X-100 ( Sigma-Aldrich ) at room temperature ( RT , 22˚C ) . Slices were then incubated for 48 hr at 4°C with primary antibodies against the apoptotic marker-cleaved caspase-3 ( 1:500; rabbit polyclonal; 9661 , Cell signaling ) , neuronal marker NeuN ( 1:500; mouse monoclonal; MAB377 , Millipore ) or astrocytic marker GFAP ( 1:1500; rabbit polyclonal; Z0334 , Dako ) . Slices were incubated for 2 hr at RT with secondary antibody Alexa 546 goat anti-rabbit ( 1:300 , A11035 , Invitrogen ) , Alexa 647 goat anti-mouse ( 1:300 , A21235 , Invitrogen ) or Cy2 donkey anti-rabbit ( 711-225-152 , Jackson Immuno Research ) . Primary and secondary antibodies were diluted in PBS containing 2% NGS or 2% NDS and 0 . 2% Triton X-100 . After staining , brain slices were mounted with Fluoroshield Mounting Medium ( ab104135 , Abcam ) . For Htt staining , brains were embedded in tissue tek ( Sakura ) and frozen on dry ice with 100% ethanol . 5 – 10 µm sagittal sections were sliced and subjected to antigen retrieval by being placed in 10 mM sodium citrate buffer at 84°C or 90 – 95°C for 15 – 20 min and rinsed with TBS-Triton-X ( 0 . 3% , Roth ) . Subsequently , sections were blocked with 20% sheep or horse serum in TBS-Triton-X for 1 hr at RT . Primary antibody ( Htt: 1:200 , rabbit monoclonal , ab109115 , Abcam ) was diluted in TBS-Triton-X and incubated overnight at 4°C . Secondary antibody ( 1:200 , goat anti-rabbit AlexaFluor 488 , A11008 , Life technologies ) in TBS-Triton-X was incubated for 2 hr at RT . Afterwards , sections were embedded in fluoromount ( Sigma-Aldrich ) including Hoechst33342 ( 1:1000 , B2261 , Sigma-Aldrich ) . Mounted slices were analyzed with a confocal laser-scanning microscope ( Leica SP8 ) . Experimental animals were sacrificed by cervical dislocation immediately before OXPHOS analysis . Brains were micro-dissected on ice and specimens weighed on an analytical balance ( Sartorius , CPA1003S; Germany ) . The micro-dissected brain regions were directly transferred into ice-cold mitochondrial respiration medium MiR05 ( EGTA 0 . 5 mM , MgCl23 mM , K-lactobionate 60 mM , taurine 20 mM , KH2PO410 mM , HEPES 20 mM , sucrose 110 mM , BSA 1 g/L , adjusted to pH 7 . 1 ) ( Kuznetsov et al . , 2000 ) . The tissue was then homogenized in a pre-cooled 1 . 5 ml tube with a motorized pestle in MiR05 medium with 10 strokes . Resulting homogenates containing 10 mg tissue wet weight were suspended in 100 µl of ice-cold MiR05 and later 20 µl ( 2 mg ) from the 100 µl tissue suspension was added to each chamber of the Oxygraph-2k , Oroboros Instrument containing 2 ml of MiR05 for OXPHOS analysis ( Holmström et al . , 2012 ) . All chemicals were purchased from Sigma-Aldrich , Germany . The optimized motorized pestle preparation of brain tissue yields a high degree of permeabilization as evident by the minimal effect of digitonin titrations on OXPHOS capacity . Therefore , digitonin is not necessary for this protocol . Tissue homogenates were transferred into calibrated Oxygraph-2k ( O2k , Oroboros Instruments , Innsbruck , Austria ) 2 ml chambers . Oxygen polarography was performed at 37 ± 0 . 001°C ( electronic Peltier regulation ) in O2k chambers and oxygen concentration ( μM ) as well as oxygen flux per tissue mass ( pmol O2·s− 1·mg− 1 ) were recorded real-time using DatLab software ( Oroboros Instruments Innsbruck , Austria ) . A multisubstrate protocol was used to sequentially explore the various components of mitochondrial respiratory capacity . The homogenate was suspended in MiR05 , added to the Oxygraph-2k glass chambers and the O2flux was allowed to stabilize . A substrate-uncoupler-inhibitor titration ( SUIT ) protocol was applied to assess qualitative and quantitative mitochondrial changes in Hdh150 transgenic mice and unaffected controls . After stabilization , LEAK respiration was evaluated by adding the complex I ( CI ) substrates malate ( 2 mM ) , pyruvate ( 10 mM ) and glutamate ( 20 mM ) . The maximum oxidative phosphorylation ( OXPHOS ) capacity with CI substrates was attained by the addition of ADP+Mg2+ ( 5 mM ) ( CIOXPHOS ) . For evaluation of maximum OXPHOS capacity of the convergent input from CI and complex II ( CII ) at saturating ADP-concentration , the CII substrate succinate ( 10 mM ) was added ( CI +CIIOXPHOS ) . We then uncoupled respiration to examine the maximal capacity of the electron transport system ( ETS or CI +IIETS ) using the protonophore , carbonylcyanide 4 ( trifluoromethoxy ) phenylhydrazone ( FCCP; successive titrations of 0 . 2 μM until maximal respiration rates were reached ) . We then examined consumption in the uncoupled state solely due to the activity of complex II by inhibiting complex I with the addition of rotenone ( 0 . 1 μM; ETS CII or CIIETS ) . Finally , electron transport through complex III was inhibited by adding antimycin A ( 2 μM ) to obtain the level of residual oxygen consumption ( ROX ) due to oxidating side reactions outside of mitochondrial respiration . The O2 flux obtained in each step of the protocol was normalized by the wet weight of the tissue sample used for the analysis and in addition ROX was subtracted from the fluxes in each run to correct for non-mitochondrial respiration ( Hollis et al . , 2015 ) . All respiration experiments comprise 2 – 3 counterbalanced blocks across days . All substrates and inhibitors used were obtained from Sigma . WT and presymptomatic VFDO Hdh150 mice ( 13 – 15 weeks of age ) were isolated and food deprived for 24 hr . Subsequently , they were placed into an operant chamber with a touchscreen including two windows and a food dispenser on the opposite wall ( Med Associates Inc; St . Albans ) . In order to keep animals motivated to perform the task , their daily food intake was adjusted to maintain body weight at 75 – 80% of their initial body weight during the course of the experiment . The experiment consisted of three phases: 1 ) mice were trained to collect a food pellet from the dispenser twice on day 1 , 2 ) mice were trained to nose poke the touchscreen to obtain food pellet reward . They needed to collect and consume the pellet to proceed to the next trial . One daily session was either 30 min or 70 trials . The touch training was over when mice reached 70 trials on three consecutive days . 3 ) For the visual discrimination task , the screen presented two stimuli ( pair 1: black vs . white or pair 2: black vs . grey ) , one correct , one false , randomly presented left or right . The mice were trained to nose poke the correct stimulus , whereupon a pellet was released . Again , they needed to collect and consume the pellet to proceed to the next trial . One daily session was either 30 min or 100 trials . The task was successful when the mice reached 70% correct trials on three consecutive days . Mice were not habituated to the set-up . Each mouse was removed from its home cage and put into a holding box next to the testing box . Subsequently , the mouse was put into the testing box facing the rear wall . The mice had time to explore the area for 5 min . Time in the center , which was determined as 10 cm away from each wall of the box , was measured automatically by EthoVision XT 8 . 5 , when the center-point of the mouse moved into it . For all cell lines used in this study the identity has been authenticated and mycoplasma contamination has been tested and excluded . HEKT cell lines ( ATCC , RRID:CVCL_1926 ) stably expressing FLAG-tagged HTT-exon 1 with either 51 or 83 CAG repeats under the control of a Tet-off promotor as well as the filter retardation assay were described previously ( Scherzinger et al . , 1997 ) . For the filter retardation assay cells were either transfected with a pool of MID1 specific siRNA oligonucleotides ( AATTGACAGAGGAGTGTGATC , CACCGCAUCCUAGUAUCACACTT , CAGGAUUACAACUUUUAGGAATT , TTGAGTGAGCGCTATGACAAA , AAGGTGATGAGGCTTCGCAAA , TAGAACGTGATGAGTCATCAT ) or non-silencing control oligonucleotides ( AATTCTCCGAACGTGTCACGT ) using Oligofectamine ( Thermo Fisher Scientific ) or treated with metformin at a final concentration of 1 mM and 2 . 5 mM for 24 hr . Cell lysates were soaked through a filter membrane and aggregates were detected using monoclonal anti-FLAG M2-Peroxidase ( HRP ) antibodies ( Sigma-Aldrich ) . Signals were quantified using the Fiji Software . Neuro-2A ( a mouse neuroblastoma cell line , ATCC , RRID:CVCL_0470 ) cells or murine primary cortical neurons ( prepared from NMRI mice E14 . 5 as described previously [Kickstein et al . , 2010] were transfected with constructs expressing Htt exon1 with 49CAG repeats fused to GFP ( vector pEGFP-C1-HTTex1; an N-terminal GFP tag ) the day before analysis . Cells were analyzed in a previously established FRAP-based assay to monitor translation in living cells ( Krauss et al . , 2013 ) using a Zeiss LSM700 confocal microscope . In brief , in contrast to standard FRAP , the GFP-signal of the entire cell was bleached using a 488 nm argon laser and fluorescence recovery was imaged over a time frame of 4 hr . The fluorescence signal was quantified as the sum of the pixel over the cell area , and the resulting total cell fluorescence was normalized to the post-bleach signal , which was set to 100% . Fluorescence recovery curves represent mean ± SEM of at least 35 cells . The following C . elegans strains was used: strain AM141 , genotype rmIs133 [unc-54p::Q40::YFP] ( RRID:WB-STRAIN:AM141 ) . AM141 worms express YFP that is linked to a polyglutamine stretch of 40 glutamines ( Q40 ) in the muscle cells of the body wall . In the early lifetime of the worms YFP-Q40 is soluble and it aggregates gradually over time . This strain is used as a model for polyglutamine diseases like Huntington´s disease , since the 40Q represents a pathological range of the polyglutamine stretch . For the treatment with metformin , NGM plates were seeded with OP50 bacteria and dried overnight . For heat inactivation OP50 bacteria were incubated at 70°C for 30 min . Metformin in a concentration of 5 mM , 10 mM or 500 mM was added the next day and dried again before usage . Worms were then put onto the plates and aggregates and liquid trashing were quantified after 5 days . Nematodes were synchronized by hypochloride treatment . At day 5 of adulthood the worms were analyzed . Aggregates were counted under a fluorescence stereo microscope after anesthetizing the worms with 25 mM Levamisol on a coverslip . For each experiment 15 – 20 animals were counted . In addition , liquid thrashing was analyzed in 10 – 15 animals per experiment . Therefore , single worms were transferred into one drop of M9 buffer ( 3 g KH2PO4 , 6 g Na2HPO4 , 5 g NaCl , 1 ml 1 M MgSO4 , H2O to 1 liter ) and the rhythmic bending of the worm around its body axis was counted for 30 s . Each experiment was conducted at least three times . For western blotting mice were sacrificed and brains were grinded and shaked in magic mix ( 48% urea , 15 mM TRIS-HCl ( pH7 . 5 ) , 8 . 7% glycerin , 1% SDS , 1% mercaptoethanol , complete protease inhibitors ( Roche ) , Phosstop ( Roche ) ) at 4°C with add-on homogenization ( QIAshredder ) . Afterwards , samples were boiled at 95°C and 30 µg protein lysate ( 40 µg for Htt blot ) was loaded on a 10% SDS PAGE gel , resolved ( overnight at 100 V for separation of mHtt from wtHtt ) and blotted onto a PVDF membrane ( BioRad ) using TransBlot Turbo ( BioRad ) . Membranes were then blocked with 1% BSA ( pS6 ) or 5% milk ( Actin , Gapdh , Htt ) in PBS-Tween20 ( Roth ) and incubated with primary antibody ( Actin: 1:2000 , A2066 , Sigma Aldrich; pS6: 1:2000 , 2215 , cell signaling; Gapdh: 1:2000 , ab8245 , abcam; Htt: 1:850 , ab109115 , abcam ) in blocking buffer overnight . Membranes were then washed three times with PBS-Tween20 and subsequently incubated with secondary antibody ( 1:6000 , for Htt 1:4000 , Donkey anti-rabbit or anti-mouse , Jackson Immuno Research ) for 1 hr in blocking buffer . Subsequently , membranes were again washed three times with PBS-Tween20 . Chemiluminescent detection was done by using Western Lightning Plus-ECL ( PerkinElmer ) . Visualisation was performed on a ChemiDoc MP Imaging System ( Biorad ) . Quantification of resulting bands was performed using Image Lab ( version 5 . 2 . 1 ) . Both the Hdh150 and WT mice received chronic metformin ( MP Biomedicals , LLC; France ) administration ( 5 mg/ml in the drinking water ) freshly prepared every day for 3 weeks starting from an age of 9 – 10 weeks . Statistical significance was tested in GraphPad Prism ( GraphPad Software Inc . , La Jolla , CA ) . Table 1 contains all details concerning statistical tests used ( name of the test , p-values , F values and degree of freedom ) . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 and ****p<0 . 0001 . For all data , we first tested for normal distribution using the one-sample Kolmogorov-Smirnov test . In case that the null hypothesis of a normal distribution could not be rejected ( for p>0 . 05 ) , we employed a parametric test , if H0 could be rejected ( for p<0 . 05 ) we used non-parametric tests: Mann-Whitney U test for non-parametric data and t test for parametric data . Pearson’s correlation coefficient was used on raster plots that were temporally binned ( 328 ms per bin ) to compare activity patterns between pairs of neurons . Box-and-whisker plots indicate the median ( line ) of average values from multiple time-lapses , the 25-75th percentiles ( box ) and the 10-90th percentiles ( whiskers ) . Graphs show mean ±s . e . m ( standard error of the mean ) . Values used in figures are available on Dryad Digital repository ( doi:10 . 5061/dryad . g3b5272 ) and the code used for the analysis of calcium imaging is in Figure 1—source data 2 . | Huntington’s disease is a devastating brain disorder that causes severe mood disorders , problems with moving , and dementia . Most people develop the condition between their thirties and fifties , and die a decade or two after the symptoms first appear . The disease emerges because of a mutation in the gene for the Huntingtin protein , which leads to neurons slowly dying in the brain . While genetic testing can reveal who carries the faulty gene , no treatment addresses the root of the disorder or prevents it from appearing . Instead , most therapies for Huntington’s disease aim to reduce brain damage once the telltale symptoms are already present . However , the disease-causing protein is expressed early during the life of a patient , which could give it time to damage the brain long before neurons die and the disorder reveals itself . Treatments that start after the first signs of the disease may be too late to reverse the damage . Detecting and preventing early brain changes in people that carry the mutation may thus help to stop the disease from progressing . Here , Arnoux , Willam , Griesche et al . set out to detect the minute changes that the faulty Huntingtin protein may cause in the brain network of young mice with the mutation . State-of-the-art imaging tools helped to examine individual neurons in the brain area that processes visual information . These experiments revealed that a group of brain cells had become hyperactive; once this change had occurred , the mutant animals were less anxious than is typical for mice . Metformin is a drug used to treat diabetes , but it also interferes with a structure that is required to produce the disease-causing Huntingtin protein . Arnoux et al . therefore explored whether the compound could rescue the early brain alterations observed in mutant mice . Adding metformin in the water of the animals for three weeks halted the production of the mutant protein , reversed the brain changes and stopped the abnormal behavior . Further work is now required in humans to confirm that Huntington’s disease starts with a change in the activity of networks in the brain , and to verify that metformin can stop the disorder in its track . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Metformin reverses early cortical network dysfunction and behavior changes in Huntington’s disease |
To monitor nitrate and peptide transport activity in vivo , we converted the dual-affinity nitrate transceptor CHL1/NRT1 . 1/NPF6 . 3 and four related oligopeptide transporters PTR1 , 2 , 4 , and 5 into fluorescence activity sensors ( NiTrac1 , PepTrac ) . Substrate addition to yeast expressing transporter fusions with yellow fluorescent protein and mCerulean triggered substrate-dependent donor quenching or resonance energy transfer . Fluorescence changes were nitrate/peptide-specific , respectively . Like CHL1 , NiTrac1 had biphasic kinetics . Mutation of T101A eliminated high-affinity transport and blocked the fluorescence response to low nitrate . NiTrac was used for characterizing side chains considered important for substrate interaction , proton coupling , and regulation . We observed a striking correlation between transport activity and sensor output . Coexpression of NiTrac with known calcineurin-like proteins ( CBL1 , 9; CIPK23 ) and candidates identified in an interactome screen ( CBL1 , KT2 , WNKinase 8 ) blocked NiTrac1 responses , demonstrating the suitability for in vivo analysis of activity and regulation . The new technology is applicable in plant and medical research .
Quantitatively , nitrogen is the single most limiting nutrient for plants . Thus , not surprisingly , maximal crop yield depends critically on nitrogen fertilizer inputs . Current practices require production of ∼1 . 5 × 107 tons of N-fertilizer per annum , consuming ∼1% of the world’s annual energy production . Plants absorb only a fraction of the fertilizer applied to the field , leading to leaching into groundwater , polluting the environment , and damaging human health . Improvements in nitrogen use efficiency of crops are urgently required; although potential targets including uptake transporters and metabolic enzymes have been identified , successful improvements in N-efficiency are rare ( McAllister et al . , 2012; Jiang , 2012b; Xu et al . , 2012; Schroeder et al . , 2013 ) . Overexpression of an alanine amino transferase or the transporter OsPTR9 are two of the few examples of improved nitrogen use efficiency ( Shrawat et al . , 2008; Fang et al . , 2013 ) . Ammonium , nitrate , amino acids , and di- and tripeptides serve as the major forms of inorganic and organic nitrogen for plants . Uptake occurs predominantly from the soil/rhizosphere into roots , although aerial parts of the plant are also capable of absorbing nitrogen ( McAllister et al . , 2012 ) . Nitrogen availability and distribution in soil vary both spatially and temporally . Inorganic nitrogen uptake is complex and involves multiple ammonium and nitrate uptake systems , typically grouped into low-affinity/high-capacity and high-affinity/low-capacity systems ( Siddiqi et al . , 1990; Wang et al . , 1994; von Wirén et al . , 1997 ) . Their relative activity is influenced by both exogenous and endogenous factors . The exact sites of uptake of the various forms of nitrogen along the length of the root , the cells that are directly involved , and in vivo regulation are not well understood . Also , the exact intercellular path towards the stele is not experimentally proven . The reasons for this lack of knowledge lie in the fact that nitrogen transport is difficult to measure . Some studies rely on the analysis of the depletion of the medium , others use stable isotopes , or the 13N-isotope , which has a short half-life time of ∼10 min and requires access to a suitable supply source ( Wang et al . , 1993; Clarkson et al . , 1996 ) . Most of these techniques lack spatial resolution , that is information on which cell layers and which root zones absorb the nutrient . Electrophysiological assays can provide spatial information; however , they are mostly used at accessible surfaces . Spatial information has been provided in a few studies by methods such as vibrating electrodes ( Henriksen et al . , 1990 , 1992 ) , positron-emitting tracer imaging systems ( Matsunami et al . , 1999; Kiyomiya et al . , 2001 ) , or secondary ion mass spectrometers ( Clode et al . , 2009 ) . We also know little about differences in the distribution of the nitrogen forms in different root cell types or zones and with respect to cellular compartmentation . Classical approaches average total ion/metabolite levels over all cells in the sample , for example , in whole roots . Nitrate levels differ dramatically between root cell types ( Zhen et al . , 1991; Karley et al . , 2000 ) . Recently , a GFP-labeled protoplast-sorting platform was used to compare metabolomes of individual cell types in roots ( Moussaieff et al . , 2013 ) . This study found that the levels of small oligopeptides were comparatively higher in the epidermis and endodermis compared to other root cell types . Compartmental analyses indicated that the nitrate concentration of root vacuoles is ∼10-fold higher compared to the cytosol ( Zhen et al . , 1991 ) . Transporters are placed in strategic positions to control which and how much of a specific nitrogen form can enter a given cell at a given point of time . The progress in identifying transporter genes provided a new handle for addressing the mechanisms and the spatial and temporal regulation of nitrogen acquisition from a new level of detail . Three major families of transporters for inorganic nitrogen uptake ( and distribution ) have been identified: the NPF/POT nitrate transporter family ( Leran et al . , 2013 ) , NRT2 nitrate transporters ( Kotur et al . , 2012 ) , and the ammonium transporters of the AMT/MEP/Rh family ( von Wirén et al . , 2000; Andrade and Einsle , 2007 ) . In addition to their role in nitrate uptake , members of the NPF/POT ( Leran et al . , 2013 ) family play important roles also in the transport of histidine , dicarboxylates , oligopeptides , and glucosinolates , and surprisingly at least three major plant hormones: auxin , ABA , and gibberellin ( Krouk et al . , 2010; Kanno et al . , 2012; Boursiac et al . , 2013 ) . Genes are valuable tools for exploring physiological functions . Analysis of RNA levels allows us to study gene regulation ( Gazzarrini et al . , 1999 ) , for example , transcriptional GUS-fusions for determining organ and cell type specific expression and translational GFP-fusions for subcellular localization . Both classical and novel methods , including cell-specific transcriptional profiles and ‘translatomes’ provide us with new insights into differences in the expression of transporters in roots ( Brady et al . , 2007; Mustroph et al . , 2009 ) . Analysis of cell type-specific expression profiles showed that the majority of changes in nitrate-induced gene expression are cell-specific ( Gifford et al . , 2008 ) . Expression and purification of the proteins followed by reconstitution in vesicles or expression in heterologous systems to interrogate biochemical properties include Km and transport mechanisms . The genes can be used as a basis for structure function studies ( Loqué et al . , 2007 ) and to obtain crystal structures ( Andrade et al . , 2005; Doki et al . , 2013 ) . We can use the genes to identify interacting proteins ( Lalonde et al . , 2010 ) . Importantly , the availability of genes enables us to generate specific mutants ( Yuan et al . , 2007; Wang et al . , 2009 ) that provide insights into their physiological roles . However , even with this massive amount of detailed data , the key information is missing , namely the information on the activity state of a given protein in vivo . In vivo activity depends mainly on two additional parameters beyond protein abundance at a given membrane: the local concentration of the substrate/s , the status of the cell ( e . g . , the membrane potential and local pH as key determinants for ion transporter activity ) , and the status of cellular regulatory networks required for the activity of the protein in question . Again , genes can help us to find regulators and study the effect of mutations on nitrogen acquisition , but ultimately we need to be able to quantify the activity of the transporters in individual cells in vivo . Nitrogen uptake is controlled by many factors , such as nitrogen level , energy status of the plant , assimilation status of imported nitrogen , N-demand , and involved mobile signals between shoots and roots as well as between different parts of the root system ( von Wirén et al . , 1997 ) . Nitrate transporters are regulated through phosphorylation , mediated by calcium-dependent calcineurin-like kinases ( Calcineurin B-like , CBL and CBL-interacting protein kinase , CIPK ) ( Ho et al . , 2009; Hu et al . , 2009; Wang et al . , 2009 ) . Major breakthroughs were findings that indicate both the members of the AMT and NPF/POT family function as transporters and receptors ( transceptors ) ( Ho et al . , 2009; Lima et al . , 2010; Rubio-Texeira et al . , 2010 ) . However , despite broad progress , at present , we have only a limited understanding of signaling pathways that control nitrogen acquisition . It is important to develop tools for monitoring the activity of individual transporters in specific locations in individual cells of plant roots in a minimally invasive manner . A minimally invasive tool that has proven valuable for monitoring ions and metabolite levels with high spatial and temporal resolution is genetically encoded fluorescent nanosensors ( Okumoto , 2012 ) . These sensors rely on substrate binding–dependent conformational rearrangements in a sensory domain . The rearrangements are reported by changes in Förster Resonance Energy Transfer ( FRET ) efficiency between two fluorescent proteins , which act as FRET donor and acceptor due to spectral overlap . Sensors for glucose , sucrose , and zinc have successfully been used in Arabidopsis to monitor steady state levels as well as accumulation and elimination under both static and dynamic conditions where roots were exposed to pulses of the respective analytes ( Deuschle et al . , 2006; Chaudhuri et al . , 2008; Okumoto et al . , 2008; Chaudhuri et al . , 2011; Lanquar et al . , 2014 ) . The recent progress in obtaining crystal structures for transporters , and more importantly the availability of transporter structures in multiple configurations , has provided insights into the conformational rearrangements occurring during the transport cycle ( Doki et al . , 2013; Guettou et al . , 2013; Henderson and Baldwin , 2013; Madej et al . , 2013 ) . Biochemical and structural analyses have shown that many transporters undergo conformational changes during the transport cycle ( Shimamura et al . , 2010; Jiang , 2012a; Krishnamurthy and Gouaux , 2012 ) . Important in this context is that such rearrangements have been observed for many members of the MFS superfamily , including members of the NPF/POT family ( Doki et al . , 2013 ) . We therefore hypothesized that it should be possible to ‘record’ the conformational rearrangements that occur during the transport cycle in a similar manner as used for the engineering of the FRET sensors . The first prototype for transport activity sensors , named AmTrac , uses ammonium transporters as sensory domains for engineering transport activity sensors by inserting a circularly permutated EGFP ( cpEGFP ) into a conformation-sensitive position of an ammonium transporter ( De Michele et al . , 2013 ) . Addition of ammonium to yeast cells expressing the AmTrac sensor triggers concentration-dependent and reversible changes in fluorescence intensity ( De Michele et al . , 2013 ) . Whether this approach is transferable to other family proteins in different species remained to be shown . To create nitrate and peptide transport activity sensors , we fused CHL1 and four PTRs to fluorescent protein pairs , expressed the fusions in yeast , and tested their response to substrate addition ( named NiTrac for nitrate transport activity and PepTrac for peptide transport activity ) . The five sensors responded to the addition of nitrate or peptides , respectively . The kinetics of the NiTrac1 sensor response was strikingly similar to the transport kinetics of the native CHL1; the response was specific and reversible . The new sensors were used to study structure/function relationships , to correlate effects of mutations in CHL1 and NiTrac1 on activity and sensor responses , and to observe the effect of potential regulators on the conformation of the transporter . The successful use of the sensors in yeast indicates that these new tools can be used for in planta analyses .
It is likely that the nitrate transceptor CHL1 undergoes conformational rearrangements during its transport cycle . To measure substrate-dependent conformational rearrangements , CHL1 was sandwiched between a yellow acceptor ( Aphrodite ) and cyan donor fluorophore ( mCerulean ) ( Rizzo et al . , 2006; Figure 1A ) . This chimera , named NiTrac1 , was expressed in yeast , followed by spectral analysis of yeast cultures in a spectrofluorimeter ( Figure 1B ) . The fluorophores were in Förster distance , as evidenced by significant resonance energy transfer . If conformational rearrangements were induced by substrate addition , one might expect a change in the energy transfer rate . To our surprise , and in contrast to typical FRET sensors ( e . g . , glucose or glutamate [Fehr et al . , 2003; Okumoto et al . , 2005] ) , we observed an overall reduction in the emission intensities of both donor and acceptor , but no obvious change in FRET efficiency . Cyan FPs are typically robust compare to the yellow variants; specifically , they are less sensitive to pH changes or other ions compared to yellow variants ( here Venus encoded by codon-modified Aphrodite gene sequence [Deuschle et al . , 2006] ) . However , Aphrodite emission was unaffected by nitrate when excited directly ( Figure 1B , inset ) , indicating that the external nitrate triggers donor quenching in the cytosol . The sensor response can be expressed as a ratio change between the emission intensity of the sensor at CFP excitation relative to YFP emission obtained from acceptor excitation . As one may have expected , the nitrate analog chlorate that lead to the naming of CHL1 ( chlorate resistance of the chl1 mutant ) ( Tsay , 1993 ) also triggered NiTrac quenching ( Figure 1C ) . The response of NiTrac1 is nitrate- and chlorate-specific; other compounds such as chloride , ammonium , divalent cations , and dipeptide had no significant effect ( Figure 1D ) . When mCerulean was replaced by the corral-derived cyan fluorescent protein mTFP ( Ai et al . , 2008 ) , we observed FRET , but nitrate addition had no effect on the emission of this variant ( Figure 1E ) . The mTFP variant , named NiTrac1c ( control ) , therefore can serve as a control sensor for in vivo measurements . Replacement of mCerulean with eCFP , another jellyfish variant , retained the donor-quenching response to nitrate ( Figure 1F ) . Although we do not understand the mechanism by which nitrate triggers donor quenching , the effect is likely related to a specific property common to mCerulean and eCFP and lacking in mTFP . 10 . 7554/eLife . 01917 . 003Figure 1 . Design and development of NiTrac sensors . ( A ) Schematic representation of the NiTrac1 sensor construct . Aphrodite , yellow; mCerulean , light blue; CHL1/NRT1 . 1/NPF6 . 3 dark blue; TM , transmembrane domain . ( B ) Emission spectra for NiTrac1 expressed in yeast cells; excitation at 428 nm: addition of 5 mM potassium nitrate ( red; control 5 mM KCl , blue ) lead to a reduction in fluorescence intensity of donor and acceptor emission , caused by donor quenching . Inset: emission of Aphrodite in NiTrac1 when excited at 505 nm . Aphrodite emission was unaffected . ( C ) Nitrate and its analog chlorate both trigger quenching at 5 mM concentrations . Nitrate-induced ratio change ( peak fluorescence intensity of Aphrodite excited at 505 nm over emission spectrum at 485 nm obtained with excitation at 428 nm ) . Data are normalized to buffer-treated control c . ( D ) Substrate specificity: yeast cells expressing NiTrac1 were treated with the indicated compounds at 5 mM concentrations . Only nitrate and chlorate triggered responses that were significantly different from control c ( *p<0 . 05 , t test ) . Experiment performed as in Figure 1C . ( E ) Absence of quenching of NiTrac1 when mCerulean was exchanged for mTFP ( excitation at 440 nm ) . Inset: emission of Aphrodite in NiTrac1 when excited at 505 nm . ( F ) Donor quenching is retained when mCerulean is exchanged for eCFP in NiTrac1 in response to addition of 5 mM potassium nitrate ( red; control 5 mM KCl , blue; excitation at 428 nm ) . Inset: emission of Aphrodite in NiTrac1 when excited at 505 nm . ( G ) No detectable effect on the fluorescence properties of nitrate addition to yeast cells coexpressing a cytosolically localized free mCerulean and the CHL1 transceptor . Mean ± SD; n = 3 . DOI:http://dx . doi . org/10 . 7554/eLife . 01917 . 003 It is conceivable that nitrate is taken up by CHL1 into the cytosol where it binds to mCerulean or eCFP , leading to quenching . However , addition of nitrate to yeast cells expressing CHL1 alone had no effect on the fluorescence of a cytosolically expressed mCerulean ( Figure 1G ) . One could argue that quenching occurs locally at the exit pore of the transporter directly at the plasma membrane and thus requires tethering of mCerulean to the transporter . To test whether quenching is specifically caused by nitrate , we created similar constructs for the oligopeptide transporters PTR1 , 2 , 4 , and 5 from Arabidopsis ( Komarova et al . , 2012; Tsay et al . , 2007; Leran et al . , 2013; Figure 2A ) . These proteins share between 39% and 74% homology with CHL1 . PepTrac1 , PepTrac2 , and PepTrac5 sensors all responded with donor quenching to the addition of 0 . 5 mM diglycine ( Figure 2B–D ) . Interestingly , PepTrac4 responded to substrate addition with a ratio change that is consistent with a change in the energy transfer rate rather than donor quenching ( Figure 2E ) . Further characterization will be necessary to explore the molecular basis of donor quenching and how conformational rearrangements cause donor quenching in NiTrac1 by nitrate or PepTrac1 by peptides and how they induce resonance energy transfer in PepTrac4 . 10 . 7554/eLife . 01917 . 004Figure 2 . PepTrac sensors . ( A ) Schematic representation of the PepTrac sensor constructs . AtPTR1 , 2 , 4 , and 5 were used for PepTrac sensor creation . Three-dimensional model of AFP-PTRs-mCerulean chimeric protein based on the crystal structure of bacteria peptide transporters ( ‘Materials and methods’ ) . PTR1 is shown in rainbow cartoon; AFP in yellow; mCerulean in blue . ( B–D ) Donor quenching of PepTrac1 , 2 , and 5 expressed in yeast in response to addition of 0 . 5 mM diglycine ( red; control 5 mM KCl , blue; excitation at 428 nm ) . Inset: emission of Aphrodite in PepTrac1 , 2 , and 5 when excited at 505 nm . ( E ) FRET ratio change for PepTrac4 ( red; control 5 mM KCl , blue; excitation at 428 nm ) . Mean ± SD; n = 3 . DOI:http://dx . doi . org/10 . 7554/eLife . 01917 . 004 The conformational rearrangements in the sensors could be induced by substrate binding or reflect rearrangements that occur during the transport cycle . Because binding and transport typically have different kinetic constants , we analyzed the response kinetics of NiTrac1 . CHL1 is unusual in that it shows biphasic nitrate uptake kinetics ( Figure 3A; Liu and Tsay , 2003 ) . The observed dual-affinity in oocytes had been attributed to phosphorylation of T101 by endogenous kinase ( Liu and Tsay , 2003 ) . The phosphorylation hypothesis would suggest that about half of the transporter molecules are phosphorylated . Interestingly , we observed that the kinetics of the fluorescence response of NiTrac1 in yeast were also biphasic ( Figure 3A ) . Since it is unlikely that yeast also partially phosphorylates the transporter , the biphasic kinetics are more likely an intrinsic property of the protein . Mutation of T101 to alanine had been shown to eliminate the high-affinity component ( Figure 3B; Liu and Tsay , 2003 ) . Introduction of T101A into NiTrac1 also eliminated the high-affinity component , intimating that NiTrac1 is a transport activity sensor , and that conformational rearrangements during the transport cycle affect mCerulean emission ( Figure 3B ) . Interestingly , the transport Kms of both high- and low-affinity phases matched the values obtained for the fluorescence response , supporting the hypothesis that NiTrac measures the transport activity . Measurement of the sensor response in individual yeast cells demonstrated rapid nitrate-induced quenching and reversibility of the fluorescence intensity after removal of nitrate ( Figure 3C ) , indicating that the sensor can be used effectively for in planta analyses . 10 . 7554/eLife . 01917 . 005Figure 3 . Biphasic kinetics of the NiTrac1 response . ( A ) Biphasic nitrate uptake kinetics of the fluorescence response of NiTrac1 ( red ) and biphasic nitrate uptake transport kinetics of CHL1/NRT1 . 1 ( Black ) . ( B ) Monophasic nitrate uptake kinetics of the fluorescence response of NiTrac1-T101A ( red ) and monophasic low-affinity transport kinetics of CHL1/NRT1 . 1-T101A ( Black , oocyte uptake data from ( Liu and Tsay , 2003 ) . The Kms of NiTrac1 for nitrate are ∼75 . 1 ± 21 μM and 3 . 8 ± 2 . 6 mM; for NiTrac1-T101A is 3 . 5 ± 3 . 7 mM . Excitation and emission as Figure 1C . The amount of decreased fluorescence intensity by addition of indicated nitrate concentration in Figure 3A , B were normalized to water-treated control ( 0 ) ( mean ± SD; n = 3 ) . ( C ) Analysis of the NiTrac1 response in individual yeast cells trapped in a Cellasic microfluidic plate . Cells were initially perfused with 50 mM MES buffer pH 5 . 5 , followed by a square pulse of 10 mM KNO3 in MES buffer for 6 min ( blue frame ) . Data were normalized to the initial value ( mean ± SD; n = 3 ) . DOI:http://dx . doi . org/10 . 7554/eLife . 01917 . 005 To study the NiTrac mechanism in more detail and to identify residues important for the transport function of the transporter and sensor , we generated a homology model for CHL1 on the basis of crystal structures of bacterial proton-dependent oligopeptide transporter homologs ( ‘Materials and methods’ ) and predicted potentially functionally important residues structurally close to the substrate binding pocket from the predicted structure and from sequence alignments . We specifically targeted residues that might be important for substrate specificity , residues involved in proton cotransport , and salt bridges possibly involved in dynamic movements during the transport cycle ( Figure 4A ) . As one may have expected , different mutants showed different energy transfer ratios , consistent with conformational differences ( altered distance and/or orientation of the fluorophores in the absence of substrate; Figure 4B ) . Interestingly , we not only observed cases in which donor quenching was lost , but also changes that are consistent with changes in FRET efficiency in response to ligand addition , as well as mixtures of donor quenching and change in the FRET efficiency ( Figure 4B ) . However , without knowledge of the effect of the mutations on transport activity , the data are difficult to interpret . Therefore , we introduced the corresponding mutations into CHL1 , expressed the mutants in Xenopus oocytes and used two-electrode voltage clamp ( TEVC , Figure 5 ) , and 15N-uptake ( Figure 6 ) to measure the transport activity . In response to nitrate addition , CHL1 expressing oocytes showed an inward current , consistent with the proposed 2H+/NO3− cotransport mechanism . CHL1 contains a highly conserved motif E41-E44-R45 in TM1 predicted to play a role in proton coupling ( Newstead , 2011; Newstead et al . , 2011; Solcan et al . , 2012; Doki et al . , 2013 ) . Mutations in this motif in the oligopeptide transporters PepTSt ( from Streptococcus thermophiles [Solcan et al . , 2012] ) , PepTSo , PepTSo2 ( both from Shewanella oneidensis ) ( Newstead , 2011 ) , and GkPOT ( from Geobacillus kaustophilus ) ( Doki et al . , 2013 ) typically lost proton-driven transport activity . We therefore tested the role of residues in this motif using NiTrac expressed in yeast and CHL1 expressed on oocytes . Mutation in any of the three residues ( E41A , E44A , and R45A , TM1 ) led to a loss of nitrate-induced currents and 15N-uptake in both the high- and low-affinity range ( 0 . 5/0 . 25 and 10 mM ) ( Figures 5 , 6 ) . The corresponding mutant of NiTrac1 also lost the sensor response to nitrate addition ( Figures 4B , 7 ) , indicating that the conserved motif is also used for proton cotransport of nitrate . Interestingly , the mutant was characterized by higher FRET compared to wild-type CHL1 , indicating that the mutation leads to a conformational change in the protein ( Figure 4B ) . Structural and functional analyses of the bacterial peptide transporter PepTSt had implicated a salt bridge between a conserved K126 in TM4 and E400 in TM10 in peptide recognition and/or structural movements during the transport cycle ( Solcan et al . , 2012 ) . This lysine is conserved throughout the POT family ( K164 of CHL1 , TM4 ) ( Newstead , 2011; Newstead et al . , 2011; Solcan et al . , 2012; Doki et al . , 2013 ) . Consistent with results from the bacterial PepTSt and GkPOT , mutation of K164 to alanine or aspartate in CHL1 completely abolished nitrate uptake in both the high- and low-affinity range ( 0 . 25 and 10 mM ) ( Figure 6 ) ; however , the nitrate-dependent inward currents were retained ( Figure 5 ) . Mutation K136A in GkPOT and K126A in PepTSt both abolished completely proton-driven uptake but still had counterflow activities ( Doki et al . , 2013; Solcan et al . , 2012 ) . Both CHL1-K164 mutants either function as nitrate-dependent proton channels or have lost selectivity , and consistent with the shift of the reversal potential to more negative values transport other anions such as chloride . NiTrac1-K164A surprisingly showed a different response mode , that is upon addition of nitrate the mutant not only showed donor quenching but apparently also a change in FRET efficiency , underlining the exceptional sensitivity of NiTrac1 to effects of mutations on conformation ( Figure 4B ) . Mutation of the salt bridge acceptor E476A in TM10 of CHL1 led to loss of both the nitrate-induced inward current and 15N-uptake ( Figures 5 , 6 ) and NiTrac lost the sensor response to addition of nitrate ( Figure 4B ) ; by contrast , and as one might expect , the conservative mutation E476D had no significant effect on transport properties and sensor response ( Figures 4B , 5 , and 6 ) . Alanine substitutions were introduced into corresponding sites predicted to be in the vicinity of the substrate-binding pocket ( L49 , Q358 , and Y388 in TM1 , TM7 , and TM8 , respectively ) . Consistent with the results obtained for the corresponding residue ( N342 in TM8 ) in GKPOT ( Doki et al . , 2013 ) , Y388A had no detectable effect on the nitrate-induced inward currents , 15N-uptake , and sensor response ( Figures 4B , 5 , and 6 ) , indicating the residue Y388 may not be involved in nitrate binding or transport cycle of CHL1 . Mutation of L49 in TM1 and Q358 in TM7 of CHL1 to alanine had no significant effect on nitrate-induced inward currents and 15N-uptake ( Figures 5 , 6 ) , but NiTrac responses were characterized by a mixture of donor quenching and FRET change ( Figure 4B ) . Based on the protein sequence alignments , CHL1 carries an extended cytoplasmic loop connecting the N- and C-terminal six helical bundles . To test the role of this loop , a triple mutant E264A-E266A-K267A was analyzed . The triple mutant lost specifically the low-affinity component nitrate-induced inward current and 15N uptake but retained the high-affinity component ( Figures 5 , 6 ) , implicating the charged residues in the extended loop in the regulation of nitrate uptake affinity . Similarly , the corresponding NiTrac1 mutant also lost the sensor response to high nitrate concentrations ( Figure 4B ) . Together , these data show that NiTrac1 is a sensitive tool for reporting conformational changes in mutants and further support the hypothesis that NiTrac1 reports activity states of the transporter . 10 . 7554/eLife . 01917 . 006Figure 4 . Response of NiTrac1 mutants to nitrate addition . ( A ) Three-dimensional model of CHL1 protein based on the crystal structures of bacteria ( ‘Materials and methods’ ) . Red square , potential substrate binding pocket . Right panel , enlarged potential substrate binding pocket . ( B ) . Fluorescence response of NiTrac1 mutants expressed in yeast in response to addition of 10 mM potassium nitrate ( red; control 10 mM KCl , blue; excitation at 428 nm ) . To compare the differences in fluorescence intensity between wild type and mutants of CHL1 as well as the differences after addition of nitrate , all data from wild type and mutants were normalized to the intensity of KCl-treated controls at 470 nm . Mean ± SD; n = 3 . DOI:http://dx . doi . org/10 . 7554/eLife . 01917 . 00610 . 7554/eLife . 01917 . 007Figure 5 . Current and voltage curve of CHL1/NRT1 . 1 mutants using TEVC . Oocytes were voltage clamped at −40 mV and stepped into a test voltage between −20 and −180 mV for 300 ms , in −20-mV increments . The currents ( I ) shown here are the difference between the currents flowing at +300 ms in the cRNA-injected CHL1 mutants and water-injected control of the indicated substrates . The curves presented here were recorded from a single oocyte . Similar results were obtained using another two different batches of oocytes . DOI:http://dx . doi . org/10 . 7554/eLife . 01917 . 00710 . 7554/eLife . 01917 . 008Figure 6 . 15NO3− uptake activity of various CHL1 mutants in oocytes . The injected oocytes with various cRNA of CHL1 mutants were incubated with 0 . 25 mM and 10 mM K15NO3 buffer at pH 5 . 5 for about 1 . 5∼2 hr , and their 15N content was determined as described in ‘Materials and methods’ . The values are mean ± SD ( n = 5∼6 for all three experiments ) . Data are normalized to the 0 . 25 mM treated CHL1-injected oocytes . + , significant difference ( p<0 . 05 , t test ) compared with water-injected oocytes . An asterisk indicates a significant difference ( p<0 . 05 , t test ) compared with the CHL1-injected oocytes . Similar results were obtained using another two batches of oocytes . DOI:http://dx . doi . org/10 . 7554/eLife . 01917 . 00810 . 7554/eLife . 01917 . 009Figure 7 . Summary of nitrate uptake and TEVC measurements for CHL1 mutants and relative fluorescence emission and change in apparent energy transfer efficiency of NiTrac1 mutants in presence and absence of nitrate . CHL1 column: wild type and various mutants of CHL1 . 15N column: nitrate uptake measured in oocytes using 15NO3− ( checkmark ✓ indicates transport activity detected , whereas × indicates no significant , or dramatically reduced , uptake activity; red HA: measured at low nitrate concentration to analyze high-affinity component; blue LA: measured at high nitrate concentration to analyze low-affinity component ) . TEVC column: effect of mutations on current voltage relationships measured by two-electrode voltage clamping; checkmark ✓ indicates nitrate-induced current observed , whereas × indicates no significant , or dramatically reduced , current induced by nitrate; red HA and blue LA as defined above . FRET* column: crude classification of the apparent energy transfer efficiency observed in NiTrac and NiTrac mutants in the absence of substrate; multiple arrows ( ↑ ) indicate a relative higher energy transfer efficiency . Blue arrows , reduced apparent energy transfer; red arrows , increased apparent energy transfer . SR column: type of response of NiTrac or mutants to substrate addition: DQ , donor quenching; iET , increased energy transfer; rET , reduced energy transfer . DOI:http://dx . doi . org/10 . 7554/eLife . 01917 . 009 The transceptor CHL1 plays important roles in nitrate uptake , transport , sensing , and must therefore be subject to regulation of its activity by posttranslational regulation on the one hand; on the other hand , CHL1 must interact with intracellular proteins in order to control downstream transcription by signaling pathways . We hypothesized that binding of regulatory proteins or signaling proteins might affect the fluorescence properties of NiTrac1 . Therefore , we tested whether coexpression of the known interactor CIPK23 , which can phosphorylate CHL1 at T101 in in vitro assays , would affect the properties of NiTrac1 ( Figure 8 ) . CIPK23 did not change the energy transfer between the fluorophores in the absence of nitrate , but blocked the fluorescence response of NiTrac1 to nitrate addition ( Figure 8B ) , either by stoichiometric binding or by phosphorylation of T101 . The coactivator CBL9 , which did not affect CHL1 transport activity on its own but enhanced the CIPK23-mediated phosphorylation of CHL1 ( Ho et al . , 2009 ) , had no detectable effect on the fluorescence response of NiTrac1 by itself ( Figure 8C ) . By contrast , CIPK8 , which is nitrate inducible in a CHL1-dependent fashion , did not affect the Nitrac1 response . However , CBL1 on its own also blocked the Nitrac1 response to nitrate addition ( Figure 8D ) . The analysis of coexpression of NiTrac1 with combinations of CIPKs and CBLs will require a different approach since episomal expression of three partners likely will create high variability due to copy number variance . 10 . 7554/eLife . 01917 . 010Figure 8 . Effects of the fluorescence response of NiTrac1 by interacting proteins . Known interactors or regulators , such as CIPK8 , CIPK23 , CBL1 , and CBL9 as well as other interactors identified in a large-scale membrane protein interaction screen were co-expressed with NiTrac1 in yeast cells . ( A ) Donor quenching response of NiTrac1 with vector as control . ( B ) and ( C ) Fluorescence response of NiTrac1 in CIPK23 and CBL9 coexpressing yeast , respectively . ( D ) The fluorescence response indicates that CIPK23 , CBL1 , At2g40540 , and At5g41990 affect the conformation NiTrac1 , whereas no detectable change is observed for CBL9 , CIPK8 , At1g71140 , and At1g25510 . Experiment performed as in Figure 1C . ( A–C ) Nitrate-induced ratio change ( peak fluorescence intensity of Aphrodite excited at 505 nm over emission spectrum obtained with excitation at 428 nm ) . Data are normalized to KCl-treated control at 470 nm . An asterisk indicates a significant difference ( p<0 . 05 , t test ) compared with the KNO3-treated control . Mean ± SD; n = 3 . DOI:http://dx . doi . org/10 . 7554/eLife . 01917 . 010 A large-scale interactome screen recently identified novel CHL1 interactors ( Lalonde , 2010 ) . To test whether some of these interactors affect NiTrac1 fluorescence , we coexpressed four candidate proteins with NiTrac1 in yeast . Although two of the four did not show significant effects on NiTrac1 or the response to nitrate addition , we found that the potassium transporter KT2 and the WNK kinase WNK8 blocked NiTrac1 responses ( Figure 8D ) . Further experiments will be required to characterize the role of these new interactions; however , the results demonstrate the suitability of NiTrac1 for analyzing the effect of known and novel interactors on CHL1 conformation and activity .
The activity sensors can provide three types of reports: ( i ) the basic ratio provides information on structure , specifically conformation of the population of sensors that can be compared between , for example , mutants or in response to coexpression of a regulator; ( ii ) the intensity of donor or acceptor can be subject to substrate-induced changes that lead to quenching as seen in NiTrac and PepTrac . At present , we do not understand the molecular basis of nitrate-induced donor quenching , which appears to affect mCerulean and CFP , but not mTFP . The fact that three PepTracs show a similar quenching effect when dipeptides are added may indicate that the quenching is caused by a conformational rearrangement in the transporter . A more detailed biophysical characterization may shed light on this unexpected behavior of the sensors . ( iii ) The change in the emission ratio of the two fluorophores upon substrate addition in PepTrac4 is likely caused by a change in the resonance energy transfer as had been observed for small molecule sensors ( Okumoto , 2012 ) . In certain cases , that is , NiTrac1 mutants L49A and Q358A ( Figure 7 ) , we observed a mixture of donor quenching and FRET changes . We thus hypothesize that both NiTracs and all four PepTracs have the potential to report in two different modes , that is , donor quenching , a FRET change or a combination thereof . Structural rearrangements triggered by mutations , by binding of a regulator or by mutations apparently lead to a variety of changes in the fluorescence output . One of the most striking features of NiTrac1 is that it reflects the biphasic kinetics of CHL1 and that even the transport and fluorescence response constants are highly similar . Mutagenesis of T101 to alanine , which had been shown to specifically affect the high-affinity component of nitrate uptake , also specifically eliminated the high-affinity response in NiTrac1 . These findings strongly supported the notion that NiTrac reports the processes that occur in the transceptor , when it binds and/or transports nitrate . The observations also intimate that the dual-affinity is not caused by partial phosphorylation of CHL1 when expressed in oocytes , as suggested by Tsay’s group ( Liu and Tsay , 2003 ) , but more likely represent an intrinsic property of CHL1 since they also occur when NiTrac1 is expressed in yeast . CHL1 also functions as nitrate sensor to regulate transcription of a variety of genes including that of the high-affinity nitrate transporter NRT2 ( Ho et al . , 2009 ) . Interestingly , the transport and signaling activities of CHL1 can be decoupled by Pro492L in the loop connecting TM10 and 11 ( Ho et al . , 2009 ) . It will thus be interesting to introduce this mutation into NiTrac1 and monitor the effect on the sensor output . Taking the advantage of a homology model , we introduced mutations into NiTrac1 and studied the effects on the transport activity by TEVC recording and 15N-uptake into oocytes , and compared the effects to fluorescence readouts from the corresponding NiTrac1 mutants . Specifically , we analyzed the role of the putative proton-coupling motif 41ExxER45; the role of charged residues in the extended loop R264R266K267; and residues in the substrate binding pocket as well as a predicted a salt bridge L49 , K164 , Q358 , Y388 , and E476 ( Newstead , 2011; Newstead et al . , 2011; Solcan et al . , 2012; Doki et al . , 2013 ) . We observed three main types of response ( Figure 7 ) : ( i ) loss of both nitrate uptake activity and loss of the sensor response in E41A , E44A , R45A , and E476A; ( ii ) loss of either high- or low-affinity uptake activity and correlated loss of the respective sensor response in T101A and R264A/R266A/K267A; and ( iii ) maintenance of the nitrate uptake activity and sensor response in L49A , Y388A , Q358A , and E476D . Relative to NiTrac1 , more than half of the mutants show a change in FRET between the two fluorophores in the absence of substrate addition . Consistent with the role of E41 and E44 in proton coupling for bacterial peptide transporters , the fluorescence response of E41A , E44A , and R45A NiTrac1 variants support a similar role in the nitrate transporter CHL1 . Interestingly , L49A , which did not show detectable differences in transport activity , showed a mixture of donor quenching and FRET change in response to nitrate addition , demonstrating that NiTrac1 is exquisitely sensitive for detecting changes in the overall protein conformation . R45A lost transport activity but retained a FRET change response after the addition of substrate rather than showing a quenching response , indicating an overall conformational change due to binding of nitrate in the absence of a functional transport cycle . Interestingly , mutation of charged residues in the extended cytoplasmic loop of CHL1 ( R264A/R266A/K267A ) specifically affected the low-affinity component in sensor and uptake response , implicating the loop , potentially through interacting proteins that can tune activity . How T101 phosphorylation , which affects the behavior of CHL1/NiTrac1 at low nitrate levels cooperates with the cytosolic loop , which appears to specifically affect the behavior in high nitrate conditions will be interesting to address in future experiments . Based on our studies , we presume that K164 , Q358 , and E476 may participate in nitrate binding . Consistent with data from bacterial peptide transporters , E476A lost both sensor response and uptake activity . This conserved residue aspartate likely plays a role in the binding pocket and/or salt bridge formation that is important for the substrate transport cycle . Mutants carrying K164A/D and Q385A mutations were both characterized by significantly increased nitrate-dependent inward currents . It will be interesting to further explore the cause for the increased conductivity with respect to transported ion species . Although the data from the limited number of mutants do not allow us to draw conclusions on the exact molecular nature of the conformational changes , we nevertheless provide the first evidence that activity sensors are highly sensitive and simple tools for probing structure-function relationships in heterologous and homologous systems without the necessity to purify the transporters . The interaction of proteins likely affects the conformation of both partners , either directly or as a consequence of modifications such as phosphorylation . Here we show that activity sensors can be used to probe such interactions with exquisite sensitivity . As a proof of concept , we demonstrate that coexpression of the calcium-dependent kinase CIPK23 , which is phosphorylating T101 of CHL1 and thereby inhibiting the low-affinity component of CHL1 , can block the fluorescence response when coexpressed with NiTrac1 . Although typically CIPKs are thought to require a CBL for substrate recognition and derepression of the autoinhibition , CIPK23 had been shown to be able to interact with CHL1 on its own and trigger at least partial phosphorylation of T101 in vitro ( Ho et al . , 2009 ) . Interestingly , although CBL9 had been shown to enhance the CIPK-mediated phosphorylation of T101 , we did not observe an effect of coexpression of CBL on NiTrac1 . Surprisingly , and despite the high sequence identity between AtCBL9 and AtCBL1 ( ∼89% identity ) , CBL1 but not CBL9 inhibited the nitrate response of NiTrac1 . AtCBL1 and 9 have been shown to regulate a variety of processes including potassium uptake , pollen germination , as well as sugar- , hormone- and ROS-signaling ( Sagi and Fluhr , 2006; Xu et al . , 2006; Cheong et al . , 2007; Hashimoto and Kudla , 2011; Drerup et al . , 2013; Kimura et al . , 2013; Li , 2013a ) . Even though CBL1 and 9 have apparent overlapping functions , they can have specific effects , for example , the AtCBL1-AtCIPK1 complex is involved in ABA-dependent stress responses , whereas the AtCBL9-AtCIPK1 complex plays roles in ABA-independent stress responses ( Drerup et al . , 2013 ) . In general , CIPKs depend on their coactivator-CBLs to activate CIPK kinase activity . However , recent studies showed that full-length CIPK23 , CIPK16 , or CIPK6 alone can activate the AKT1 potassium channel system ( Li et al . , 2006; Lee et al . , 2007; Fujii et al . , 2009 ) . Also , AtCBL10 interacts with AKT1 to regulate potassium homeostasis without binding to any AtCIPKs ( Ren et al . , 2013 ) . The assays deployed here use strong promoters and high copy number plasmids . It will therefore be important to test whether low levels of the kinase are sufficient for inhibiting NiTrac1 . It will also be interesting to compare the responses of NiTrac1 when expressed in mutant plants lacking components of the CBL-CIPK machinery . In addition , we tested whether NiTrac1 can be used to monitor conformational rearrangements caused by interacting proteins , specifically we tested interactors detected in a large-scale membrane protein/signaling protein interaction screen ( Lalonde et al . , 2010 ) . Surprisingly , we found an interaction of CHL1 with the potassium transporter AtKT2/KUP2/SHY3 , which plays a role in potassium uptake . Coexpression of KT2 with NiTrac1 led to a block of the nitrate response . Whether this interaction plays a role in crosstalk between nitrogen and potassium uptake remains to be shown . In addition , we had found an interaction with the ‘no lysine ( K ) kinase 8’ WNK8 . Also , WNK8 blocked the nitrate-induced fluorescence response of NiTrac1 . WNK8 had been shown to interact specifically with and phosphorylate subunit C of the vacuolar H+-ATPase AtVHA-C ( Hong-Hermesdorf et al . , 2006 ) , as well as with the calcineurin B-like 1 calcium sensor AtCBL1 ( Li , 2013b ) . It will be interesting to further explore the network between CBL1 , WNK8 , and CHL1 . Obviously , NiTrac1 is highly sensitive to conformational changes that occur during the transport cycle , effects of mutations , and to changes caused by interaction with other proteins . Thus , analyses performed with these sensors in plants will have to differentiate between responses caused by substrate and regulatory interactions . The use of controls , for example , the mTFP sensor and elimination of FRET by exchanging the acceptor with a non-fretting fluorophore , as well as the use of mutant sensors may be a way to dissect the relative contribution of substrate and protein interactions . These new tools are complementary to the classical tools set including electrophysiology and tracer studies but have the clear advantage of allowing measurements deep inside plant or animal tissues and organs , domains largely inaccessible to other technologies . In summary , we developed a set of five sensors that can report the activity of nitrate and peptide transporters in vivo . At the same time , such activity sensors prove to be sensitive tools for studying the effect of mutations on the conformation of the transporter or to detect the regulatory interactions with other proteins . The next step will be to deploy NiTrac1 and its mutants as well as the PepTracs in Arabidopsis plants to characterize the activity of the transporters and their regulation in vivo . The plant peptide transporters are close homologs of the human SLC15 peptide transporters . The SLC15 transporter PepT1 has pathophysiological relevance in processes like intestinal inflammation and inflammatory bowel disease ( Ingersoll et al . , 2012 ) , and it serves as a key transport mechanism for uptake of drugs ( Agu et al . , 2011 ) . Given the success in engineering five members of the plant transporter family , we envisage that the approach can be implemented also for measuring the activity of the human transporters in situ and to use such sensors , for example , for drug screens .
All transporter and sensor constructs were inserted by Gateway LR reactions , into the yeast expression vectors pDRFlip30 , 34 , and 39 . pDRFlip30 is a vector that sandwiches the insert between an N-terminal Aphrodite t9 ( AFPt9 ) variant ( Deuschle et al . , 2006 ) , with nine amino acids truncated of C-terminus and a C-terminal monomeric Cerulean ( mCer ) ( Rizzo et al . , 2006 ) . pDRFlip39 sandwiches the inserted polypeptide between an N-terminal enhanced dimer Aphrodite t9 ( edAFPt9 ) and C-terminal fluorescent protein enhanced dimer , seven amino acids and nine amino acids truncated of N-terminus and C-terminus of eCyan ( t7 . ed . eCFPt9 ) , respectively . pDRFlip34 carries an N-terminal AFPt9 and a C-terminal t7 . TFP . t9 ( t7 . TFP . t9 ) ( Rizzo et al . , 2006 ) . All plasmids contain the f1 replication origin , a GATEWAY cassette ( attR1-CmR-ccdB-attR2 ) , positioned between the pair of fluorescent proteins , the PMA1 promoter fragment , an ADH terminator , and the URA3 cassette for selection in yeast . Vector construction has been described ( Jones , In press ) . The full length ORF of CHL1 , PTR1 , PTR2 , PTR4 , and PTR5 from Arabidopsis and different mutants of NRT1 . 1 in the TOPO GATEWAY Entry vector were used as sensory domains for creating the nitrate sensor NiTrac1 and the peptide sensors PepTrac1 , PepTrac2 , PepTrac4 , and PepTrac5 . The yeast expression vectors were then created by GATEWAY LR reactions between different forms of pTOPO-NRT/PRT and different pDRFlip-GWs , following manufacturer’s instructions . For functional assays in Xenopus oocytes , the cDNAs of CHL1 and all mutants of CHL1 were cloned into the oocyte expression vector pOO2-GW ( Loqué et al . , 2009 ) . Point mutations for studying characterization of CHL1 in oocyte and NiTrac1 in yeast were generated by QuikChange Lightning Site-Directed Mutagenesis Kit ( Agilent Technologies , Santa Clara , CA ) . For the coexpression assays with interactors in yeast , putative interactors were inserted , by LR reaction , in the yeast expression vector pDR-XN-GW vector , which replaced URA3 with LEU2 in pDRf1 containing the f1 replication origin , GATEWAY cassette ( -attR1-CmR-ccdB-attR2 ) , PMA1 promoter fragment , and ADH terminator in yeast ( Loqué et al . , 2007 ) . The yeast BJ5465 [MATa , ura3–52 , trp1 , leu2Δ1 , his3Δ200 , pep4∷HIS3 , prb1Δ1 . 6R , can1 , GAL+] was obtained from the Yeast Genetic Stock Center ( University of California , Berkeley , CA ) . Yeast was transformed using the lithium acetate method ( Gietz et al . , 1992 ) , and transformants were selected on solid YNB ( minimal yeast medium without nitrogen; Difco ) supplemented with 2% glucose and -ura/-ura-leu DropOut medium ( Clontech , Mountain View , CA ) . Single colonies were grown in 5 ml liquid YNB supplemented with 2% glucose , and -ura/-ura-leu drop out under agitation ( 230 rpm ) at 30°C until OD600nm ∼ 0 . 5 was reached . The liquid cultures were subcultured by dilution to OD600nm 0 . 01 in the same liquid medium and grown at 30°C until OD600nm ∼ 0 . 2 . Fluorimetric analyses are described in more detail at Bio-protocol ( Ho and Frommer , 2016 ) . In brief , fresh yeast cultures ( OD600nm ∼ 0 . 2 ) were washed twice in 50 mM MES buffer , pH 5 . 5 , and resuspended to OD600nm ∼0 . 5 in the same MES buffer supplemented with 0 . 05% agarose to delay cell sedimentation . Fluorescence was measured in a fluorescence plate reader ( M1000; TECAN , Austria ) , in bottom reading mode using a 7 . 5 nm bandwidth for both excitation and emission ( Bermejo et al . , 2010; Bermejo et al . , 2011 ) . Typically , emission spectra were recorded ( λem 470–570 nm ) . To quantify fluorescence responses of the sensors to substrate addition , 100 µl of substrate ( dissolved in MES buffer , pH 5 . 5 as 500% stock solution ) was added to 100 µl of cells in 96-well flat bottom plates ( #655101; Greiner , Monroe , NC ) . Fluorescence from cultures harboring pDRFlip30 ( donor: mCER ) and 39 ( donor: t7 . ed . eCFPt9 ) was measured by excitation at λexc 428 nm; cell expressing from pDRFlip34 ( donor t7 . TFP . t9 ) was excited at λexc = 440 nm . Quantitative fluorescence intensity data from individual yeast cells expressing the sensors ( Figure 3C ) were acquired on an inverted microscope ( Leica , Wetzlar , Germany ) . To be able to record fluorescence intensities in single cells over time , yeast cells were trapped as a single cell layer in a microfluidic perfusion system ( Y04C plate , Onyx , Cellasic , Hayward , CA , USA ) and perfused with either 50 mM MES buffer , pH 5 . 5 , or buffer supplemented with 10 mM KNO3 ( Bermejo et al . , 2010; Bermejo et al . , 2011 ) . Briefly , imaging was performed on an inverted fluorescence microscope ( Leica DMIRE2 ) with a QuantEM digital camera ( Photometrics ) and a 40×/NA ( numerical aperture ) 1 . 25–0 . 75 oil-immersion lens ( IMM HCX PL Apo CS ) . Dual-emission intensity ratios were simultaneously recorded using a DualView unit with a Dual CFP/YFP-ET filter set ( ET470/24m and ET535/30m; Chroma ) and Slidebook 4 . 0 software ( Intelligent Imaging Innovations ) . Excitation ( filter ET430/24x; Chroma ) was provided by a Lambda LS light source ( Sutter Instruments; 100%lamp output ) . Images were acquired within the linear detection range of the camera at intervals of 20 s . The exposure time was typically 1000 ms with an EM ( electron-multiplying ) gain of 3 °ø at 10 MHz and an electron multiplying charge coupled device ( EMCCD ) camera ( Evolve , Photometrics , Tucson , AZ , USA ) . Measurements were taken every 10 s , with 100 ms exposure time using Slidebook 5 . 4 image acquisition software ( Intelligent Imaging Innovations , Denver , CO , USA ) . Fluorescence pixel intensity was quantified using Fiji software; single cells were selected and analyzed with the help of the ROI manager tool . Protein structure prediction for CHL1 and PTR1 was performed using Phyre ( Kelley and Sternberg , 2009 ) . Full-length CHL1 ( At1g12110 ) and AtPTR1/NPF8 . 1 ( At3g54140 ) amino acid sequences were used for the 3D structure prediction on the website . The analysis made use of four solved crystal structures of nitrate/peptide homologs ( PDB ID: 4iky , 2xut , 4aps , 4lep ) ( Newstead , 2011; Newstead et al . , 2011; Solcan et al . , 2012; Doki et al . , 2013 ) . The homologs shared 16–27% identity with CHL1 or PTR1 . The predicted potentially functionally important residues were from the predicted structure ( 3DLigandSite , [Wass et al . , 2010] ) and from sequence alignments . After structural prediction of CHL1 , 41-ExxER-45 , in TM1 , the conserved sequence motif involved in proton cotransport ( 22-ExxER-26 , 21-ExxER-25 , 21-ExxER-25 , 32-ExxER-36 in PepTSt , PepTSo , PepTSo2 , and GtPOT , respectively ) , putative residues involved in substrate binding pocket L49 in TM1 ( Y30 , Y29 , Y29 , and Y40 in PepTSt , PepTSo , PepTSo2 , and GtPOT , respectively ) , Q358 in TM7 ( Q289 , Q317 , Q291 , and Q311 in PepTSt , PepTSo , PepTSo2 , and GtPOT , respectively ) , and Y388 in TM8 ( N328 , N321 , and N342 in PepTSt , PepTSo2 , and GtPOT , respectively ) , and putative residues of salt bridges K164 in TM4 ( K126 , K127 , K121 , and K136 in PepTSt , PepTSo , PepTSo2 , and GtPOT , respectively ) , E476 in TM10 ( E400 , E419 , E402 , and E413 in PepTSt , PepTSo , PepTSo2 , and GtPOT , respectively ) , and residues R264/R266/K267 in the lateral helices loop between TM6 and TM7 were selected for mutagenesis . TEVC in oocyte was performed essentially as described previously ( De Michele et al . , 2013 ) . In brief , for in vitro transcription , pOO2-CHL1 and respective mutants were linearized with MluI . Capped cRNA was in vitro transcribed by SP6 RNA polymerase using mMESSAGE mMACHINE kits ( Ambion , Austin , TX ) . Xenopus laevis oocytes were obtained from the laboratory of Miriam Goodman by surgery manually or ordered from Ecocyte Bio Science ( Austin , TX ) . The oocytes were injected via the Roboinjector ( Multi Channel Systems , Reutlingen , Germany; [Lemaire et al . , 2004; Pehl et al . , 2004] ) with distilled water ( 50 nl as control ) or cRNA from CHL1 or CHL1 mutants ( 50 ng in 50 nl ) . Cells were kept at 16°C 2–4 days in ND96 buffer containing 96 mM NaCl , 2 mM KCl , 1 . 8 mM CaCl2 , 1 mM MgCl2 , and 5 mM HEPES , pH 7 . 4 , containing gentamycin ( 50 μg/μl ) before recording experiments . Recordings were typically performed at day three after cRNA injection . Electrophysiological analyses of injected oocytes were performed as described previously ( Huang et al . , 1999; De Michele et al . , 2013 ) . Reaction buffers used recording current ( I ) -voltage ( V ) relationships were ( i ) 230 mM mannitol , 0 . 3 mM CaCl2 , and 10 mM HEPES and ( ii ) 220 mM mannitol , 0 . 3 mM CaCl2 , and 10 mM HEPES at the pH indicated plus 0 . 5 or 10 mM CsNO3 . Typical resting potentials were ∼−40 mV . Measurements were recorded by oocytes that were voltage clamped at −40 mV and a step protocol was used ( −20 to −180 mV for 300 ms , in −20 mV increments ) and measured by the two-electrode voltage-clamp ( TEVC ) Roboocyte system ( Multi Channel Systems ) ( Pehl et al . , 2004; Lemaire et al . , 2004 ) . Nitrate uptake assays were performed using 15N-labeled nitrate ( Ho et al . , 2009 ) , and oocytes injected with CHL1 cRNA were used as positive controls . After 2–4 days cRNA injection , the oocytes were incubated for 90∼120 min in 15NO3− medium containing 230 mM mannitol , 0 . 3 mM CaCl2 , 10 mM HEPES , and pH 5 . 5 . Then , oocytes were rinsed five times with ND96 buffer and individually dried at 80°C for 1–2 days . 15N content was analyzed in an ECS 4010 Elemental Combustion System ( Costech Analytical Technologies Inc . , Valencia , CA , USA ) whose output was connected to a Delta plus Advantage mass spectrometer ( Thermo Fisher Scientific Inc . , Waltham , MA , USA ) . For statistical analyses of 15N-nitrate uptake into oocytes ( Figure 6 ) and the effect of treatments on the fluorescence responses ( Figures 1C and 8D ) , we used analysis of deviance ( ANOVA ) ; factors ( sample , treatment ) were treated as fixed factors . ANOVAs were performed using the analysis of variance ( ANOVA ) calculator—one-way ANOVA from Summary Data ( www . danielsoper . com/statcalc ) . All experiments were performed at least with three biological repeats . The reported values represent mean and standard deviation . Student’s t test was used in Figures 1 , 6 and 8 to determine significance . | About 1% of global energy output is used to produce nitrogen-enriched fertiliser to improve crop yields , but much of this energy is wasted because plants absorb only a fraction of the nitrogen that is applied as fertiliser . Even worse , the excess nitrogen leaches into water sources , poisoning the environment and causing health problems . However , to date , most efforts to increase the efficiency of nitrogen uptake in plants have been unsuccessful . The key to improving the uptake efficiency of a nutrient is to identify obstacles in its journey from the soil to cells inside the plant . The first obstacle that nitrate ions encounter is the membrane of the cells on the surface of the roots of the plant . Many researchers believe that it would be possible to increase the amount of nitrogen absorbed by the plant if more was known about the ways that plants control how nitrate ions and other chemicals enter cells . The cell membrane contains gated pores called transporters that allow particular molecules to pass through it . Although the transporters responsible for the uptake of nitrate ions , peptides , and ammonium ions ( the main nitrogen compounds that plants acquire ) have been identified , current experimental techniques cannot determine when and where a specific transporter is active within a living plant . This makes it difficult to know where to target modifications and to determine how effective they have been at each step . The nitrate transporter also acts as an antenna that measures nitrate concentration to ensure it is used optimally in the plant , but current techniques cannot show how this actually works . Now , Ho and Frommer have exploited the fact that a transporter changes shape as it does its job to create sensors that can track the movement of nitrate and peptides through the cell membrane . By using fluorescent proteins to monitor how the shape of the transporter changes , Ho and Frommer were able to measure how structural mutations and regulatory proteins influenced the movement of nitrate and peptides through the membrane . For efficiency , all of this work was performed in yeast cells . The next goal is to use the technique in plants to uncover how they adjust to changes in nutrient levels in the soil . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | Fluorescent sensors for activity and regulation of the nitrate transceptor CHL1/NRT1.1 and oligopeptide transporters |
The assembly and maintenance of all cilia and flagella require intraflagellar transport ( IFT ) along the axoneme . IFT has been implicated in sensory and motile ciliary functions , but the mechanisms of this relationship remain unclear . Here , we used Chlamydomonas flagellar surface motility ( FSM ) as a model to test whether IFT provides force for gliding of cells across solid surfaces . We show that IFT trains are coupled to flagellar membrane glycoproteins ( FMGs ) in a Ca2+-dependent manner . IFT trains transiently pause through surface adhesion of their FMG cargos , and dynein-1b motors pull the cell towards the distal tip of the axoneme . Each train is transported by at least four motors , with only one type of motor active at a time . Our results demonstrate the mechanism of Chlamydomonas gliding motility and suggest that IFT plays a major role in adhesion-induced ciliary signaling pathways .
Cilia and flagella are microtubule-based organelles that power the locomotion of many organisms , generate fluid flow over multiciliated surfaces , and mediate cell signaling ( Liem et al . , 2012 ) . In order to assemble and maintain cilia , ciliary proteins are transported from cytoplasm to the tip by IFT along axonemes ( Kozminski et al . , 1993 ) . In IFT , linear arrays of multiprotein complexes ( IFT trains ) are transported by kinesin-2 and dynein-1b in anterograde and retrograde directions , respectively ( Cole et al . , 1998; Porter et al . , 1999 ) . IFT is a universal mechanism for nearly all eukaryotic cilia and flagella , and defects in this process are linked to a wide range of human diseases , including polycystic kidney disease , retinal degeneration ( Rosenbaum and Witman , 2002; Ishikawa and Marshall , 2011 ) , and Bardet-Biedl syndrome ( Ou et al . , 2005; Lechtreck et al . , 2009 , 2013; Wei et al . , 2012 ) . Several studies have suggested that IFT not only transports material between the cell body and the flagellar tip , but also interacts dynamically with the flagellar membrane ( Kozminski et al . , 1993 ) to regulate diverse ciliary functions including motility , mating , sensing extracellular signals and influencing developmental decisions ( Huangfu et al . , 2003; Snell et al . , 2004; Pedersen and Rosenbaum , 2008; Ishikawa and Marshall , 2011 ) . However , it has remained unclear how transport of IFT trains underneath the flagellar membrane transmits force to components at the exterior of the flagellar membrane . In order to investigate interactions between IFT and the ciliary surface , we used Chlamydomonas reinhardtii gliding motility as a model system . In Chlamydomonas , the flagellar surface is highly dynamic; polystyrene microspheres and other inanimate small objects adhere to and are moved bidirectionally along the flagellar surface , and cells glide over solid surfaces via adhesion of their flagella ( Lewin , 1952; Bloodgood , 1981 ) . Gliding motility is central to understanding the function and evolution of cilia , as it may have existed in early cilia before the establishment of axonemal beating . There are indications that gliding and FMG1-B motility are driven by the same process ( Bloodgood , 2009 ) , because they move at comparable speeds and both require ligation of the major flagellar surface protein , FMG1-B , into large clusters ( Bloodgood and Workman , 1984 ) , accompanied by a Ca2+-dependent signaling pathway ( Bloodgood and Salomonsky , 1994 ) . Microsphere movement is considered a more easily assayed and quantitated surrogate for the force transduction system that drives whole cell gliding motility . However , gliding is driven by the pulling motion of the leading flagellum , whereas FMG1-B movement is bidirectional ( Bloodgood , 2009 ) , implying that the two motilities could employ different motors . It has been proposed that IFT provides the force for gliding motility ( Bloodgood , 2009 ) . IFT trains make multiple connections with the flagellar membrane ( Pigino et al . , 2009 ) and carry several ciliary and flagellar membrane proteins ( Qin et al . , 2005; Huang et al . , 2007; Lechtreck et al . , 2009 , 2013 ) . Inactivation of kinesin-2 in the temperature-sensitive ( ts ) mutant fla10ts stops both IFT and gliding motility ( Kozminski et al . , 1995 ) . While these results suggest that kinesin-2 serves as the anterograde motor responsible for both microsphere movement and gliding motility ( Kozminski et al . , 1995; Laib et al . , 2009 ) , the retrograde motor for these motilities has not been clearly identified . Mutations in the LC8 subunit of dynein do not abolish FMG1-B movement ( Pazour et al . , 1998 ) , and other flagellar motors , such as the minus-end directed kinesin KCBP ( Dymek et al . , 2006 ) , have been proposed to drive FSM ( Bloodgood , 2009 ) . Several studies have raised arguments against this model . IFT motility differs significantly from FSM in that trains move faster and more processively along the length of the flagellum ( Kozminski et al . , 1993; Bloodgood , 2009 ) . FSM requires micromolar levels of free calcium , whereas IFT is Ca2+-independent ( Kozminski et al . , 1993; Bloodgood , 2009 ) . Sexual agglutinins were observed to migrate from the cell body into flagella in the absence of IFT ( Pan and Snell , 2002 ) . Therefore , evidence supporting the role of IFT in gliding motility is indirect and the precise functions of IFT , molecular motors , FMG1-B and Ca2+ in FSM remain unclear .
To dissect the mechanism of FSM , we directly observed IFT , FMG1-B and gliding motilities using single-molecule imaging techniques . We monitored the movement of individual IFT trains by using total internal reflection fluorescence ( TIRF ) illumination to image paralyzed-flagella ( pf ) mutant cells that had adhered to the glass surface with both flagella . To establish a link between FMG1-B transport and IFT , we simultaneously tracked the movement of FMG1-B antibody-coated fluorescent beads ( 200 nm diameter , dark red ) and IFT27-GFP in the pf18 strain . The beads performed short processive runs with reversals of direction whereas IFT trains moved in a regular and unidirectional manner . Multicolor kymography analysis shows that the beads transiently dissociated from one IFT train , diffused for a period of time and then bound to another IFT train ( Figure 1A , Video 1 ) similar to the movement of extraflagellar particles observed along the flagellar membrane ( Dentler , 2005 ) . Diffraction-limited images of beads and IFT trains were fitted to a two-dimensional Gaussian to achieve a higher localization precision ( Yildiz et al . , 2003 ) . Figure 1B shows the colocalization of a membrane-attached bead with an individual IFT train as it moves unidirectionally for >500 nm . The movements of the bead and the colocalized IFT trains correlate strongly ( >0 . 99 ) with each other during the processive run ( N = 30 , see Figure 1—figure supplement 1 for more examples ) , excluding the possibility that microspheres coincidentally moved together with IFT trains . We measured the distance between microspheres and the nearest IFT trains moving in the same direction for both the experimental data and randomly generated kymographs . The Kolmogorov-Smirnov test ( p=7 × 10−9 , Figure 1C ) demonstrates that IFT trains colocalize with FMG1-B along the flagellar membrane ( Figure 1D ) . 10 . 7554/eLife . 00744 . 003Figure 1 . IFT transports FMG1-B . ( A ) ( Top ) Simultaneous tracking of anti-FMG1-B beads ( red ) and IFT27-GFP ( green ) . ( Bottom ) Kymographs show that bead motility colocalizes with IFT trains during processive runs . Between the runs , the bead transiently attaches to and detaches from IFT trains . The white arrows indicate the IFT trains transporting the bead . ( B ) Two-dimensional Gaussian fitting of the bead and IFT trains show that bead motility correlates strongly ( >0 . 99 ) with the movement of individual anterograde ( green shaded region ) and retrograde ( blue shaded region ) IFT trains . The bead moves at similar speeds to IFT trains in both anterograde and retrograde directions . ( C ) Comparison of distances from beads to the closest IFT train moving in the same direction ( grey bars ) to the predicted distribution without correlation ( null hypothesis , black line ) . Kolmogorov-Smirnov statistics indicate that bead and IFT train positions strongly correlate with each other . ( D ) A model for IFT particles transporting FMG1-B . The bead is attached to FMG1-B in the flagellar membrane through antibody linkages ( not to scale ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 00310 . 7554/eLife . 00744 . 004Figure 1—figure supplement 1 . Additional examples of simultaneous tracking of bead motility and IFT . ( A–C ) Three additional examples for colocalization and correlated movement of IFT trains ( IFT27 GFP ) and beads . ( D ) Two-dimensional Gaussian fitting reveals the position of the bead and IFT individual trains as a function of time . Bead motility correlates strongly with the movement of individual IFT trains . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 00410 . 7554/eLife . 00744 . 005Video 1 . Simultaneous imaging of bead motility and IFT . The left channel shows fluorescent beads coated with anti-FMG1-B . One bead displays bidirectional movement along the flagellum . The middle channel shows the movement of IFT trains within the IFT27-GFP pf18 cell . The right channel is the superimposed image of the two channels ( IFT: green , bead: red ) . Movement of the bead colocalizes with that of individual IFT trains . The data was collected at 5 frames/s . The size of a single channel is 6 . 8 × 19 . 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 005 We next investigated whether IFT plays a direct role in gliding motility by monitoring individual IFT trains in IFT20-GFP and IFT27-GFP pf18 cells as they glided on coverslips . A small fraction of IFT trains displayed rare pauses ( 0 . 125 s−1 per cell , Ncells = 23 , Npauses = 247 ) as they moved either in the retrograde or anterograde direction . Remarkably , the pausing of IFT trains was required for the initiation of whole-cell movements ( Figure 2A , Figure 2—figure supplements 1 and 2 ) . We did not observe the start of a gliding event without a paused IFT train . We were able to determine the directionality of 60% of the paused IFT trains , and all of the trains correspondent to the initiation of gliding motility stopped moving during retrograde transport ( N = 148 ) . To rule out the possibility that IFT pausing events and initiation of gliding motility are simply coincidental , we quantified the lag time between the pausing of the last retrograde IFT train and the initiation of gliding motility . The average lag time was 0 . 48 ± 0 . 05 s . In comparison , we observed individual IFT pausing events ( including both anterograde and retrograde pauses ) every 8 . 25 ± 0 . 96 s per cell ( Figure 2—figure supplement 3 ) . The Student’s t-test excludes the null hypothesis that the timing of retrograde IFT pausing and gliding motility are independent of each other ( p<0 . 0001 ) . Anterograde trains also displayed rare pauses ( ∼0 . 02 s−1 per cell , N = 27 ) , but we never observed pausing of an anterograde train before the initiation of gliding motility . 10 . 7554/eLife . 00744 . 006Figure 2 . Dynein-1b drives gliding motility . ( A ) ( Left ) Kymograph of a gliding IFT20-GFP cell . A single retrograde IFT train transiently pauses ( red arrow ) and initiates the gliding movement of the cell toward the paused train . ( Right ) A schematic representing the timing and trajectory of the paused IFT train ( red curve ) in the gliding cell shown on the left . ( B ) ( Left ) Kymograph of an IFT27-GFP cell , pseudo-colored to show the corresponding velocity of each IFT train . Multiple IFT trains ( red arrows ) pause ( green color ) prior to gliding motility . The cell glides until it reaches the paused IFT trains . ( Right ) A schematic representing the timing and trajectories of the paused IFT trains ( red curves ) in the gliding cell shown on the left . ( C ) Gliding of uniflagellate cells under bright field illumination . A uniflagellate pf18 cell glides unidirectionally toward its flagellum . A uniflagellate dhc1b-3ts cell displays bidirectional gliding at the restrictive temperature . Red and blue arrowheads represent forward ( flagellum in the lead ) and backward ( cell body in the lead ) gliding directions . ( D ) All of the uniflagellate pf18 cells glided with the flagellum in the lead . Heat inactivation of dynein-1b in dhc1b-3ts cells at 37°C resulted in a 46% reduction in gliding frequency ( N = 35 ) and led to bidirectional gliding motility . 8% of the cells glided with the flagellum leading the cell body ( 0 . 34 ± 0 . 09 µm/s , mean ± SEM ) , while 23% glided with the cell body leading the flagellum ( 0 . 47 ± 0 . 16 µm/s ) . 23% of the cells displayed saltatory gliding movement . ( E ) Uniflagellate cells glided ∼20% faster than biflagellate cells . Inhibition of dynein-1b ( dhc1b-3ts ) resulted in a fivefold decrease in gliding speed , whereas inhibition of kinesin-2 ( pf1 fla10-1ts ) led to a twofold speed increase ( mean ± SEM ) . In temperature-insensitive paralyzed ( pf18 ) and WT cells , changes in gliding speed between permissive and restrictive temperatures were negligible . ( F ) A model for gliding motility . Retrograde IFT trains adhere to the glass surface through FMG1-B , and the surface-tethered dynein motors pull the cell body through microtubules toward the flagellar tip . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 00610 . 7554/eLife . 00744 . 007Figure 2—figure supplement 1 . Three additional examples for kymographs of gliding IFT27-GFP cells . ( A–C ) Multiple retrograde IFT trains pause in the leading flagellum prior to gliding motility . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 00710 . 7554/eLife . 00744 . 008Figure 2—figure supplement 2 . The procedure for Fourier space direction analysis . ( A ) The original kymograph of an IFT27-GFP cell during gliding motility . ( B ) Magnitude of the fast Fourier transform of the kymograph . ( C ) Mask with colors assigned to different slopes . ( D ) The pattern in Fourier space was color-coded using the mask . Green and red colors represent anterograde and retrograde IFT , blue represents paused IFT trains . Only the magnitude is shown . ( E ) Inverse Fourier transform of the colored Fourier pattern in ( C ) shows the kymograph with trajectories colored by slope . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 00810 . 7554/eLife . 00744 . 009Figure 2—figure supplement 3 . Lag time between IFT pausing and initiation of gliding motility . ( A ) A kymograph of IFT trains in a uniflagellate gliding cell . ( B ) A schematic representing the timing and trajectories of paused IFT trains in the gliding cell shown in ( A ) . Arrows show the time interval between the pausing of the retrograde IFT train and the start of gliding motility . ( C ) The histogram shows the time between the start of gliding motility and the pausing time of the last retrograde IFT train in the leading flagellum . The histogram was fitted to a single exponential decay ( blue curve ) . Average lag time is 0 . 48 ± 0 . 05 s ( N = 36 ) , which is significantly shorter than the average time between individual IFT pausing events ( 8 s ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 00910 . 7554/eLife . 00744 . 010Figure 2—figure supplement 4 . Three examples for kymograph of gliding uniflagellate IFT27-GFP cells . ( A ) – ( C ) Multiple retrograde IFT trains pause prior to gliding motility . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 010 In 24% of all cases ( N = 50 ) , a single paused train appeared sufficient to pull the entire cell ( Figure 2A , Video 2 ) . In other cases , multiple trains paused before the start of gliding motility ( Figure 2B , Video 2 ) . Pauses occurred in the leading flagellum , and cells moved in the opposite direction relative to the transport trajectory of the paused trains . In uniflagellate IFT27-GFP pf18 cells , we also always observed pausing of single or multiple IFT trains before the initiation of gliding motility ( Figure 2—figure supplement 4 ) . Gliding motility either stopped when paused IFT trains resumed movement ( Figure 2A ) , or continued until the cell bodies reached the paused trains ( Figure 2B ) . We never observed the cell body gliding further than the position of the paused IFT train ( s ) . 10 . 7554/eLife . 00744 . 011Video 2 . Pausing of IFT trains during gliding motility . A , Gliding motility of an IFT20-GFP cell on a glass surface . A single IFT train tethered to the surface ( arrow ) immediately preceding the initiation of gliding motility . The size of the window is 24 . 1 × 24 . 9 µm . The data was collected at 5 frames/s . B , Gliding motility of an IFT27-GFP cell on a glass surface . Multiple IFT trains became surface-tethered ( arrows ) , providing force for gliding motility . The size of the window is 24 . 5 × 24 . 9 µm . The data was collected at 5 frames/s . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 011 We found compelling evidence that dynein-1b is the primary motor responsible for gliding motility . First , uniflagellate pf18 cells always glided with the flagellum leading the cell body at 1 . 49 ± 0 . 10 µm/s ( mean ± SEM ) ( Figure 2C–E , Video 3 ) ( Bloodgood , 2009 ) , suggesting that gliding motility is driven by pulling forces generated through a minus end directed microtubule motor . In contrast to pf18 cells , uniflagellate dynein-1b ts cells ( dhc1b-3ts ) ( Engel et al . , 2012 ) were unable to display robust gliding at restrictive temperatures ( Figure 2C , Video 3 ) . We observed only short-range ( <500 nm ) gliding-like motion in 54% of these cells , compared to robust unidirectional gliding motility observed over 5 µm in pf18 cells . These short-range motions were bidirectional ( Figure 2D ) and significantly slower ( forward: 0 . 35 ± 0 . 09 µm/s , backward: 0 . 47 ± 0 . 16 µm/s ) than the gliding speed of pf18 cells ( Figure 2E ) . Second , inactivation of dynein-1b in biflagellate dhc1b-3ts cells decreased the fraction of gliding cells from 100% to 52% and reduced the gliding speed from 0 . 86 ± 0 . 08 µm/s to 0 . 18 ± 0 . 03 µm/s . In contrast , inactivation of kinesin-2 in fla10ts cells resulted in a twofold increase in gliding speed ( Figure 2E ) , indicating that kinesin plays an inhibitory role during dynein-1b driven gliding motility . These results agree with our observations that the pausing of anterograde trains does not initiate gliding motility . 10 . 7554/eLife . 00744 . 012Video 3 . Gliding motility of uniflagellate WT and dhc1b-3 cells . Uniflagellate WT and dhc1b-3 Chlamydomonas cells glide with the flagellum in the lead . The size of the whole window is 47 . 0 × 11 . 9 µm . The data was collected at 10 frames/s under a bright-field microscope . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 012 To further test the role of dynein-1b in gliding motility , we treated the cells with a small molecule inhibitor of dynein , ciliobrevin D ( Firestone et al . , 2012 ) . We varied ciliobrevin D concentration between 0–150 µM and monitored IFT in cells adhered both of their flagella to surface 2 min after drug treatment ( Figure 3A , Figure 3—figure supplement 1 , Video 4 ) . Both anterograde and retrograde IFT train frequencies dropped with increasing concentrations of ciliobrevin D ( Figure 3B ) , and at >100 µM ciliobrevin D we observed accumulation of IFT trains at the flagellar tip ( see example kymograph in Figure 3A ) . At 150 µM ciliobrevin D , retrograde IFT frequency was reduced by 92% compared to the 70% decrease observed after 6 hr of heat inactivation of dhc1b-3ts cells ( Engel et al . , 2012 ) . The velocities of retrograde and anterograde trains also decreased by 60% and 36% , respectively ( Figure 3C ) . Importantly , inhibition of dynein-1b resulted in a significant reduction in the speed ( 79% ) and frequency ( 79% ) of gliding motility ( N = 50 , Figure 3D ) . These results further support our conclusion that dynein-1b motors produce the force for gliding motility . 10 . 7554/eLife . 00744 . 013Figure 3 . Ciliobrevin D inhibits dynein-1b and stops gliding motility . ( A ) Kymographs of IFT20-GFP cells treated with varying concentrations of ciliobrevin D . Images were acquired 5 min after addition of ciliobrevin D . ( B and C ) Frequency and speed of retrograde and anterograde IFT trains at different ciliobrevin D concentrations . The frequency of retrograde IFT was reduced by 92% at 150 µM ciliobrevin D . ( D ) The speed and fraction of gliding cells decreased by 79% at 150 µM ciliobrevin D . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 01310 . 7554/eLife . 00744 . 014Figure 3—figure supplement 1 . Additional example for kymographs of IFT20-GFP cells treated with varying concentrations of ciliobrevin D . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 01410 . 7554/eLife . 00744 . 015Video 4 . IFT motility at different concentrations of ciliobrevin D . IFT27-GFP pf18 cells were immobilized on a glass coverslip . 0–150 µM ciliobrevin D was added to the cell culture and the movies were recorded at 10 frames/s within 2–10 min of drug treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 015 Based on these results , we propose a model to describe the functions of IFT motors , IFT trains and FMG1-B transport in gliding motility ( Figure 2F ) . Surface adhesion of the FMG1-B cargo through its large extracellular carbohydrate domain ( Bloodgood , 2009 ) stops the retrograde IFT train . Dynein motors previously engaged in transporting the paused IFT train exert force towards the microtubule minus end , causing the whole cell to move toward the plus-end flagellar tip . Thus , gliding motility works similarly to microtubule gliding assays , in which surface-immobilized dyneins glide microtubules with their plus-end tips in the lead . To investigate how cells reverse gliding direction , we developed a kymography method for monitoring IFT trains as the cells reorient their flagella during gliding ( Figure 4A–B , Video 5 ) . During 60% of reversals ( N = 40 ) , the cell lifted one of its flagella and the paused IFT trains on the surface-adhered flagellum drove the motility ( Bloodgood , 2009 ) . In other cases , single or multiple paused IFT trains accumulated in one flagellum , and the cell body glided toward this cluster until the paused trains either detached from the glass surface or reached the flagellar base . We also observed cases with paused IFT trains in both flagella where the cell remained immotile , likely due to the balance of forces between bound dynein-1b motors ( Figure 4C–D ) . There were no indications of coordination of the pausing events between the two flagella . Different modes of reversals in gliding motility may allow cells to search through the environment by a random walk when they adhere both of their flagella , and to move in unidirectional manner by lifting one of the flagella . 10 . 7554/eLife . 00744 . 016Figure 4 . Mechanisms of reversal in gliding motility . ( A ) The average intensity of all the frames in the first 60 s of Video 5 of an IFT27-GFP pf18 cell , showing the path of the gliding flagella . ( B ) To monitor IFT trains while the cells alter their flagellar orientation in gliding motility , different curves were plotted before and after cells reoriented their flagella ( red dotted lines ) . Each curve shares at least one common point with previous and subsequent curves . Intermediary frames were analyzed by interpolation of the assigned curves , and the intensity along the line in each frame was stacked according to the fixed point ( yellow dot ) . ( C ) A kymograph generated by this method reveals multiple ways that the cell can control gliding direction . ( i ) When the cell lifts one flagellum ( red arrow ) , the cell body moves toward the surface-attached flagellum . When both flagella are attached to the surface , gliding direction is determined by the balance of forces exerted by surface-tethered IFT trains . ( ii ) The cell glides toward the flagellum with more paused IFT particles ( yellow arrow ) . ( iii ) When there are equal numbers of surface-tethered IFT particles in both flagella ( cyan arrow ) , opposite forces cancel out and the cell remains stationary . ( D ) A schematic model representing the three different modes ( i , ii , iii ) of gliding direction reversal observed in ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 01610 . 7554/eLife . 00744 . 017Video 5 . Reversal of direction in gliding motility . Gliding motility of an IFT27-GFP cell on a glass surface . The cell changes the direction and speed of gliding motility either by raising one of its flagella or by the pausing of one or more IFT trains relative to the glass surface . The data was collected at 10 frames/s . The size of the window is 33 . 9 × 23 . 7 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 017 Gliding motility in C . reinhardtii is controlled by a Ca2+-calmodulin regulated kinase and requires micromolar levels of Ca2+ in the media ( Bloodgood and Spano , 2002 ) . We tested whether the pausing of IFT trains depends on Ca2+ ( Figure 5A–B ) by analyzing the kymographs of IFT at different Ca2+ concentrations . To quantify pausing in flagella , we used Fourier space direction analysis ( FSDA , see Figure 2—figure supplement 2 ) to separate the traces of paused IFT trains and moving IFT trains into two kymographs . Kymographs of paused IFT trains in Figure 5A , B show that pausing along the length of the flagellum was significantly reduced in Ca2+-deprived cells ( see additional examples in Figure 5—figure supplement 1 ) . After Ca2+ depletion , IFT trains rarely paused in the middle regions of flagella and accumulated at the base ( Figure 5B , middle ) . Figure 5C plots the total fluorescence intensity of paused IFT trains ( Ncells = 30 for each case ) along the lengths of flagella at different Ca2+ levels . The frequency of pausing in Ca2+-deprived cells was significantly lower than IFT pausing in cells at a normal Ca2+ concentration . In regular TAP media ( free [Ca2+] = 0 . 34 mM ) , 95% of all surface-adhered cells displayed gliding motility and individual pausing events were observed every ∼8 s per flagellum , on average . In contrast , both the fraction of gliding cells and IFT pausing frequency were reduced by ∼10-fold at <1 µM Ca2+ ( Figure 5D ) . Next , we compared the IFT pausing frequencies in gliding and non-gliding cells at different Ca2+ levels . The pausing frequency in non-gliding cells was low and independent of Ca2+ concentration . In contrast , the pausing frequency in gliding cells was high at normal Ca2+ levels and gradually declined to match that of non-gliding cells as the Ca2+ concentration decreased below 1 nM ( Figure 5E ) . 10 . 7554/eLife . 00744 . 018Figure 5 . Ca2+ is required for the pausing of IFT trains at flagellar adhesion sites . ( A ) ( Left ) Kymograph of an IFT27-GFP cell adhering both of its flagella in the presence of 0 . 34 mM free Ca2+ . Pausing ( middle ) and moving ( right ) IFT trains were split into separate kymographs by FSDA analysis . IFT trains pause frequently along the length of the flagellum and drive gliding motility . ( B ) When cells are deprived of free Ca2+ , immotile IFT trains accumulate near the cell body and do not display frequent pauses between the flagellar base and the tip . ( C ) The average fluorescence intensity of ‘pausing IFT trains’ in kymographs relative to the length of the flagellum ( N = 6 cells for each condition ) . Cells in 0 . 34 mM Ca2+ show robust pausing uniformly along the length of the flagellum . In contrast , Ca2+-deprived cells show significantly reduced pausing events . Background fluorescence was excluded from the analysis . ( D ) IFT pausing frequency and the fraction of gliding cells as a function of free Ca2+ concentration . The free Ca2+ concentration was controlled through Ca2+-EGTA buffering . ( E ) IFT pausing frequency of gliding and non-gliding cells as a function of free Ca2+ . At 0 . 34 mM Ca2+ , the IFT pausing frequency in gliding cells is very high compared to non-gliding cells . In Ca2+-deprived cells , the pausing frequency in gliding cells is reduced to the residual level of pausing events observed in non-gliding cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 01810 . 7554/eLife . 00744 . 019Figure 5—figure supplement 1 . The analysis of IFT pausing in the presence and absence of free Ca2+ in media . ( A ) Two additional examples for kymographs of IFT27-GFP cells adhering both flagella to the surface in the presence of 0 . 34 mM free Ca2+ . Pausing ( middle ) and moving ( right ) IFT trains were split to separate kymographs by FSDA analysis . IFT trains pause frequently along the length of flagellum and drive gliding motility . ( B ) Two additional examples for kymographs of IFT27-GFP cells adhering both flagella to the surface when cells are deprived of free Ca2+ . IFT trains accumulate near the cell body and do not display frequent pauses between the flagellar base and the tip . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 019 In Ca2+-deprived cells , the beads freely diffused on the flagellar membrane but did not move processively by IFT trains ( not shown ) . Thus , we propose that Ca2+ is required for attachment of FMG1-B to IFT trains , not for the activation of a motor protein that provides force for gliding , as previously suggested ( Bloodgood , 2009 ) . As the flagellar membrane is enriched with a PKD2-like Ca2+ channel ( Pazour et al . , 2005; Huang et al . , 2007 ) , Ca2+ signaling at flagellar adhesion sites may play a role in controlling the directionality and timing of gliding movement . To measure the forces exerted by motors bound to individual IFT trains , we tracked the movement of FMG1-B-antibody coated bead motility using an optical trap . To rule out the possibility that loads exerted by the trap might disrupt the linkage between IFT and FMG1-B , we performed TIRF imaging of IFT27-GFP and optical trapping of beads simultaneously ( Figure 6A ) . We observed that IFT trains remained colocalized with trapped beads when subjected to external forces ( Figure 6B; see Figure 6—figure supplement 1 for additional examples ) . The average offset between the positions of the trapped beads and colocalized IFT trains was 280 ± 10 nm ( N = 11 ) . This is due to the fact that the beads ( 920 nm diameter ) and IFT trains ( 200–1000 nm in length ) ( Pigino et al . , 2009 ) are large objects relative to the resolution of conventional fluorescence imaging ( approximately 200–250 nm ) and the forces that stretch the bead-FMG1-B-IFT linkage displace the center of the bead from the IFT train ( see the schematics in Figure 6—figure supplement 2 ) . The trap measurements correspond to forces generated by IFT motors and provide direct evidence that IFT transports FMG1-B . Since beads are outside the cell , but are physically linked to the action of IFT trains , the assay combines the advantages of precise in vitro trapping with the ability to manipulate IFT motility under load . 10 . 7554/eLife . 00744 . 020Figure 6 . Stall force measurements on single IFT trains . ( A ) Schematic representation of combined optical trapping of bead motility and fluorescent tracking of IFT . ( B ) Simultaneous tracking of IFT27-GFP and bead motility . At t = 3 s , the microscope stage was moved to bring the flagellum underneath the trapped bead . An IFT train stalls at the trap position ( t = 5–8 s ) and stays adhered to the bead until the bead escapes the trap ( t = 10 s ) . The CCD camera is toggled between fluorescence and bright field imaging to monitor IFT trains as well as the bead when it is out of the detection range . ( C ) Displacement records of bead motility show successive runs including stalling ( inset ) and releasing events . ( D ) The peak force distributions and statistics ( mean ± SEM ) of stalling and releasing events in pf18 cells . IFT particles exert 25–30 pN of peak forces with slightly less force produced in the anterograde direction . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 02010 . 7554/eLife . 00744 . 021Figure 6—figure supplement 1 . Three additional examples of simultaneous bead trapping and IFT27-GFP fluorescence tracking . ( A ) A single anterograde IFT train stalls directly underneath the trapped bead for 20 s . ( B and C ) The bead shows multiple runs and escapes the trap . The optical trapping beam was on throughout the duration of the kymograph . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 02110 . 7554/eLife . 00744 . 022Figure 6—figure supplement 2 . The offset between the bead position and IFT trains . ( A ) Schematic for an in vitro fixed-trap assay with a single motor bound to a microtubule . ( B ) Schematic for the optical trap assay of Chlamydomonas bead motility ( roughly to scale ) . In both ( A and B ) , there is a large offset between the bead center and the motor/cargo . The size of the offset depends on tether length , the bead diameter and the length of IFT train . ( C ) Histogram for the separation between the bead center and IFT position in the combined optical trap and fluorescence experiment . We observe up to 500 nm offset between an IFT train ( ∼400 nm in length ) and the polystyrene bead ( 900 nm in diameter ) . The average separation between the bead and IFT is 280 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 02210 . 7554/eLife . 00744 . 023Figure 6—figure supplement 3 . Additional examples for bead motility under fixed trap . ( Top ) Displacement records of bead motility in fla10ts cells show successive runs , stalling and releasing events along the anterograde direction at permissive temperatures . ( Bottom ) The bead motility in fla10ts cells displays runs , stalling and releasing events mostly along the retrograde direction at restrictive temperatures . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 02310 . 7554/eLife . 00744 . 024Figure 6—figure supplement 4 . The effect of antibody concentration on peak forces measurements . ( A ) The fraction of beads moving along the flagellar surface as a function of the antibody concentration on the bead surface . ( B ) The fraction of stall , release and escape events along the anterograde and retrograde directions in optical trapping assays as a function of antibody concentration used to coat the bead surface . ( C ) Histograms of peak forces measured in the optical trap assay along the anterograde and retrograde directions in WT cells . 1 µm diameter polystyrene beads were coated with 0 . 05 , 0 . 5 and 10 mg/ml FMG1-B antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 024 In a fixed trap assay , ∼50% of processive bead movements terminated with a stall before returning to the trap center ( Figure 6C , Figure 6—figure supplement 3 ) . We did not observe a significant difference between the force values of stall and release events . This suggests that motors attached to an IFT train may not be able to reach their maximum stall force , defined as the stall force of a single motor multiplied by the number of bound motors ( Shubeita et al . , 2008 ) . Histograms of peak forces ( Figure 6D ) show that anterograde and retrograde IFT trains moved against 21 . 4 ± 0 . 7 pN and 25 . 2 ± 1 . 3 pN ( SEM ) , respectively . Forces exerted on IFT trains well exceed the force-generation capability of a single motor . Previous studies demonstrated that multiple motors produce larger forces , move with higher velocities under load and carry cargos further than a single kinesin or dynein motor ( Mallik et al . , 2005; Vershinin et al . , 2007 ) . It remains controversial whether motor stall forces are additive at low motor copy numbers ( Vershinin et al . , 2007 ) or whether multiple motors tend to transport their cargo using only one load-bearing motor at a time ( Jamison et al . , 2010 ) . Therefore , we believe that peak forces in our experiment represent a lower boundary for estimates of motor copy number . Assuming that IFT motors produce 6–7 pN forces ( Gennerich et al . , 2007; Brunnbauer et al . , 2010 ) , we estimate that at least four motors transport IFT trains at a time , in agreement with previously reported values ( Engel et al . , 2009b ) . It is possible that multiple motor engagements enhance the run length of individual IFT trains , allowing them to traverse the length of a flagellum . In addition , the fast transport of IFT trains in a viscous cellular environment and IFT’s role in FSM may require forces larger than single motor stall forces . The measured forces in both directions only marginally changed ( <20% ) under different bead antibody-coating conditions ( Figure 6—figure supplement 4 ) . This result argues against the possibility that antibody-coating of the beads leads to the crosslinking of more than one IFT trains , which would be expected to multiply the average peak force . 20% of retrograde runs escaped the trap by producing more than 80 pN of force , presumably via clustering of FMG1-B ( Bloodgood , 2009 ) , while <1% escape events were observed in the anterograde direction . We next investigated why the speed of gliding motility ( 1 . 49 ± 0 . 10 µm/s , SEM ) is significantly slower than that of retrograde IFT ( 2 . 96 ± 0 . 14 µm/s ) , even though both processes are powered by dynein-1b motors actively transporting IFT trains . IFT trains that carry 1-μm beads moved ∼30% slower than IFT trains that were not associated with the beads ( two-tailed Student’s t-test , p = 2 . 1 × 10−9 ( anterograde ) and 1 . 8 × 10−11 ( retrograde ) , Figure 7A ) . We reasoned that the viscous drag of the membrane may slow down IFT trains transporting FMG1-B clusters and led to the observed differences in speed between IFT and FSM . To estimate the drag constant of the bead-IFT complex , we analyzed individual stalling events in the optical trap assay ( Figure 7B ) and measured the recoil time of the bead after a stall ( Figure 7C ) . The average drag constant of the bead-IFT complex was found to be 8 . 0 ± 2 . 8 pN s/μm ( SEM , N = 30 ) . 10 . 7554/eLife . 00744 . 025Figure 7 . Viscous drag of the membrane slows down the motility of IFT trains that carry beads . ( A ) The average speed of FMG1-B antibody-coated beads and IFT trains . Coating of the coverslip surface with 0 . 7 mg/ml polylysine led to a ∼20% reduction in the speed of IFT motility ( two tailed t-test , p=5 . 2 × 10−14 [anterograde] and 5 . 6 × 10−6 [retrograde] ) . IFT trains that carry 0 . 2-μm beads move at 10–30% slower speeds than IFT trains that are not associated with the beads ( p=1 . 1 × 10−2 [anterograde] and 7 . 2 × 10−5 [retrograde] ) . IFT trains that carry 0 . 2-μm beads move at similar speeds to beads ( p=0 . 4 [anterograde] and 0 . 6 [retrograde] ) ( error bars represent SEM ) . ( B ) A typical example for stalling of bead motility by the optical trap . The bead recovers slowly to the trap center after a stall ( rectangular box ) . ( C ) The recoiling of a trapped bead to the trap center after a stall was fitted to single exponential decay . The mean drag constant of the bead-FMG1-B-IFT complex is 8 . 0 ± 0 . 28 pN s/µm . ( D ) The IFT-bead complex experiences large drag forces as the IFT train moves several micrometers per second inside the flagellum ( not to scale ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 02510 . 7554/eLife . 00744 . 026Figure 7—figure supplement 1 . Measurement of viscous drag on bead movement along the surface of the flagellar membrane . ( A ) A polystyrene bead was trapped by the laser beam and brought on top of a surface-immobilized flagellum . After the bead was bound to FMG-1B , the trap was oscillated 500 nm back and forth along the flagellar surface in a square wave pattern . Trap stiffness was set to 0 . 07 pN/nm . Due to the viscous drag of the membrane , the bead displays a relaxation curve when it follows the trap . ( B ) The recoiling of a trapped bead to the trap center after a stall was fitted to single exponential decay . The mean drag constant of the bead-FMG1-B-IFT complex is 8 . 4 ± 4 . 2 pN . s/µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 026 We hypothesized that the large drag constants we measured were due to the interaction between the bead and the membrane . Our results rule out the possibility that microtubule motors step backwards during the recoiling of the bead , as the average bead velocity was 20 µm/s ( an order of magnitude faster than that of dynein and kinesin ) and there were no detectable backward steps . To rule out other possible interactions between IFT trains and axonemes , we oscillated individual beads on a flagellar membrane surface in a square wave pattern and measured the recoiling time when the beads are decoupled from IFT ( Figure 7—figure supplement 1 ) . The average viscous drag constant was 8 . 4 ± 4 . 2 pN . s/µm , which is similar to the drag that beads experience when they interact with IFT trains . Anterograde and retrograde trains moving at 2–3 µm/s would experience 16–24 pN resistive forces , comparable to the total motor force exerted on a single IFT train ( Figure 7D ) . Therefore , IFT is subjected to a high drag force when it carries a large bead along the flagellar surface , which leads to the reduction of transport velocity . We next investigated how kinesin-2 activity reduces the speed of gliding motility ( Figure 2E ) . This could theoretically be explained by a tug-of-war between active kinesin and dynein motors that are both present on the same IFT train . Alternatively , kinesin-2 and dynein-1b motors may be exclusively active on anterograde and retrograde cargos , respectively , and pausing of anterograde trains could produce forces that oppose the gliding forces of paused retrograde trains . To distinguish between tug-of-war and coordinated transport mechanisms , we examined how inhibition of one class of motors affected the forces exerted on IFT trains traveling in the opposite direction ( Laib et al . , 2009 ) . At permissive temperatures , the peak forces of IFT in fla10-1ts and dhc1b-3ts were in close agreement with wild-type ( WT ) cells . Heat-inactivation of kinesin-2 reduced the frequency of anterograde runs by 66% , but did not alter the peak forces on retrograde runs ( Laib et al . , 2009 ) ( Figure 8A ) . Similarly , heat inactivation of dynein-1b significantly reduced the ratio of retrograde to anterograde transport events ( 0 . 15 ) , but had minimal effect on the peak forces of anterograde runs ( Figure 8B ) . These results exclude a tug-of-war mechanism in IFT , which would lead to an increase of force from motors walking in one direction upon inactivation of the motors walking in the opposite direction . Instead , only one type of a motor remains active on IFT trains at a time , which is consistent with the result that retrograde speed of fla10-1ts does not change after kinesin-2 is inactivated . We propose that kinesin-2 on paused anterograde trains slows down gliding motility by exerting forces in the opposite direction to that of dynein-1b on paused retrograde trains . 10 . 7554/eLife . 00744 . 027Figure 8 . Force measurements on temperature-sensitive mutants . ( A ) Peak force histograms for IFT movement in pf1 fla10ts cells at 22°C and 34°C . Stalling events are less common than the release of the bead . The frequency of anterograde IFT trains is reduced after switching the temperature to 34°C . Forces along the retrograde direction remain unaltered after the inhibition of kinesin-2 ( mean ± SEM ) . ( B ) Peak force histograms for IFT in dhc1b-3ts . The frequency and force production of retrograde IFT trains are reduced after switching the temperature to 34°C . Forces along the anterograde direction remain unaltered after the inhibition of dynein-1b ( mean ± SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00744 . 027
We present direct evidence for the mechanism of IFT-mediated cell-surface interactions using Chlamydomonas gliding motility and FMG1-B transport as a model system . In Chlamydomonas , IFT trains carry FMG1-B as a cargo . The interaction between IFT trains and FMG1-B clusters is transient , as FMG1-B boards moving trains but usually unloads and diffuses away before the trains arrive at the flagellar base or tip . During gliding motility , IFT trains pause due to attachment of the FMG1-B cargo to the surface , and the dynein-1b motors engaged to the paused IFT trains generate pulling forces along the microtubule long axis . While the majority of organismal motilities rely on the actin cytoskeleton or axonemal beating , gliding motility in C . reinhardtii is distinct in that it is powered by intracellular transport machinery along microtubules . By providing surface adhesion points , FMG1-B performs an analogous function to integrins in mammalian cell motility ( Bloodgood , 2009; Lecuit et al . , 2011 ) . Our results also demonstrate how different types of flagellar motility in Chlamydomonas have distinct characteristics despite being driven by the same forces . First , we showed that viscous drag of the membrane slows down the motility of IFT trains that carry FMG1-B-conjugated beads , causing gliding motility and FMG1-B transport to proceed slower than IFT . Second , previous work showed that IFT is unaffected by changes in environmental Ca2+ concentration ( Bloodgood and Salomonsky , 1990; Kozminski et al . , 1993 ) , whereas gliding motility is Ca2+-dependent ( Bloodgood and Salomonsky , 1990; Kozminski et al . , 1993 ) . This raised the possibility that Ca2+ may be required to activate a specific motor protein that provides force for gliding ( Bloodgood , 2009 ) . Our results are inconsistent with this hypothesis . We found that the presence of free Ca2+ leads to frequent pauses in IFT motility at flagellar adhesion sites . In the absence of Ca2+ , IFT moves unidirectionally without interruptions , suggesting that attachment of FMG1-B to IFT trains is regulated by a Ca2+-dependent signaling pathway . The signaling pathway responsible for linking FMG1-B to IFT remains unclear . There is evidence that surface adhesion or crosslinking of FMG1-B into large clusters induces dephosphorylation of a transmembrane protein that co-immunoprecipitates with FMG1-B ( Bloodgood and Salomonsky , 1994 , 1998 ) . The complex may be regulated by a calcium-calmodulin dependent gliding associated kinase ( GAK ) , which is required for the gliding motility ( Bloodgood and Spano , 2002 ) . Further investigation is required to identify the rest of the signaling pathway regulating IFT-flagellar surface interactions . Our force measurements show a number of similarities and differences with earlier published work . In a previous trapping assay , beads nonspecifically adsorbed to the flagellar membrane were found to exert 60 pN average peak forces ( Laib et al . , 2009 ) . In contrast , our measured peaked forces are significantly lower ( 20–30 pN ) than these values . Laib et al . could not determine whether their force measurements reflect the forces produced by motors attached to IFT trains or by motors that are directly bound to FMG1-B clusters . Our experiments directly link optical trap recordings of bead motility to forces generated by flagellar motors attached to IFT trains ( Figure 1 ) , enabling us to propose physical models for the coordination of IFT and its role in FSM . In agreement with Laib et al . , we observe reciprocal coordination of motors during anterograde transport . Additionally , our measurements of dynein-1b inactivation verify that motors do not engage in a tug-of-war while travelling in the retrograde direction . The mechanism we described for IFT pausing and force generation has broad implications for the traffic of ciliary sensory proteins as well as cell signaling at ciliary membrane adhesion points ( Wang et al . , 2006 ) . For example , IFT is required for signal transduction during the mating of ciliated organisms . The initial interaction between gametes of two mating types in Chlamydomonas can occur at any point along their flagellar membranes , but the tips of their flagella must be aligned and locked before activating gametic fusion ( Homan et al . , 1987 ) . Since IFT trains transport sexual agglutinins within the flagellar membrane of gametic cells ( Ferris et al . , 2005 ) , adhesion between sexual agglutinins of opposite cell types may stall retrograde IFT movement . As a result , forces produced by IFT trains would move the flagellar contact points until a balance of forces is achieved when the flagella are properly aligned with respect to each other . Indeed , microspheres have been observed to move and accumulate at the flagellar tips during mating ( Hoffman and Goodenough , 1980 ) . We propose that the absence of retrograde bead movement may be related to the formation of adhesion contacts between the flagella and pausing of retrograde IFT trains . Cilia in the retina , liver , and kidney cells were recently observed to make direct physical contacts , which may serve as ‘bridges’ for signaling networks between many cells ( Ott et al . , 2012 ) . These contacts are tight adhesions between the ciliary membranes , mediated by N-linked glycoproteins . It is possible that IFT exerts force on cell-cell adhesion sites and determines the positioning of these adhesion sites by moving them along the length of the cilium . Ca2+ signaling at flagellar adhesion sites may play a major role in regulating attachment to IFT trains and controlling the direction of force generation . We believe that the assay we developed will be a starting point for deciphering the role of IFT in the signaling , mating and development of ciliated cells .
Vegetative C . reinhardtii cells were grown in Tris-acetate-phosphate ( TAP ) media in an illuminated plant growth chamber at 22°C . WT mt+ ( cc125 ) , pf18 mt− ( cc1297 ) , and pf18 mt+ ( cc1036 ) strains were obtained from the Chlamydomonas Resource Center . The IFT20-GFP ΔIFT20 strain was provided by K Lechtreck and G Witman . The IFT27-GFP mt+ strain and pf1 fla10-1ts were provided by J Rosenbaum . IFT27-GFP pf18 strain was generated through crosses ( Engel et al . , 2009b ) . pf1 fla10-1ts cells were incubated at 37°C for 15 min for heat inactivation of kinesin-2 , and the movies were recorded within 30 min at 37°C , before the complete impairment of IFT and the start of flagellar resorption ( Kozminski et al . , 1993 ) . In the dhc1b-3ts mt- mutant strain ( cc4422 ) ( Engel et al . , 2012 ) , dynein-1b is inhibited after 6 hr incubation at 37°C . Paralyzed flagella ( pf ) strains and paralyzing compounds ( Engel et al . , 2011 ) were utilized to impair swimming motility . Ciliobrevin D was dissolved in DMSO and then added to the pf18 cell culture to inhibit dynein activity . IFT and gliding motility were recorded simultaneously 2 min after addition of 0–150 µM ciliobrevin D to the flow chamber . The IFT and FSM imaging assays were performed with an objective-type TIRF microscope ( DeWitt et al . , 2012 ) . The GFP signal was recorded by an electron multiplied charge-coupled device ( EM-CCD ) camera , with an effective pixel size of 106 nm . IFT27-GFP pf18 and IFT20-GFP ΔIFT20 strains were used for tracking individual IFT trains ( Engel et al . , 2009a ) . The sample chamber was pre-treated with 0 . 7 mg/ml poly-L-lysine to adhere flagella on a glass surface . 15 µl cells in TAP media were applied to cover glass and inverted onto slides . Gliding motility assays were performed without polylysine treatment and cells were imaged under bright-field illumination . ts mutants were assayed at permissive ( 22°C ) and restrictive ( 37°C ) conditions by controlling the temperature with an objective heater ( Bioptechs ) . Because paralyzed cells glide 30–40% faster than non-paralyzed cells ( Bloodgood , 1995 ) , gliding speeds of ts mutants of kinesin-2 ( pf1 fla10-1ts ) and dynein-1b ( dhc1b-3ts ) were compared to those of both pf18 and WT cells at permissive ( 22°C ) and restrictive ( 37°C ) temperatures . To visualize bead movement along the flagellar surface , 200 nm carboxyl-modified nile-red fluorescent beads ( Invitrogen ) were coated with anti-FMG1-B antibody ( Bloodgood et al . , 1986 ) by EDC-NHS cross linking . 1 ml pf18 cell culture was spun down at 400g for 1 min and resuspended in 100 µl fivefold diluted TAP media . 1 µl bead stock solution ( 25 mg/ml ) was then added to 20 µl of resuspended cell culture and the mixture was incubated in ice for 10 min , in order to slow down IFT movement for bead attachment . 1 µl of 20 mM CaCl2 was then added , the cell culture was incubated at room temperature for 10 min for bead movement to recover . To immobilize cells on a glass surface , the cover glasses ( 18 × 18 mm ) were pre-treated with 0 . 7 mg/ml poly-L-lysine ( MW = 150–300 kDa; Sigma ) for 5 min . The nile-red bead fluorescence was recorded with Andor iXon 128 × 128 EM-CCD at 400 µs per frame . To prevent deflagellation of cells under intense ( 100 mW ) laser illumination , only the flagellar regions of immobilized cells were excited . Flagellar membrane adsorption of carboxylated 100 nm-beads was significantly reduced without antibody crosslinking ( ∼50-fold ) , and only two bead motility events were observed in 500 cells . To simultaneously monitor the movement of IFT trains and FMG1-B proteins , IFT27-GFP pf18 cells were incubated with anti-FMG1-B-coated dark-red beads . The GFP and dark-red bead fluorescence were simultaneously recorded at 200 ms per frame . The image was split into two fluorescent channels , which were registered to a sub-pixel accuracy ( DeWitt et al . , 2012 ) with respect to each other , prior to live-cell imaging . The crosstalk between the two fluorescent channels was below 0 . 1% . Various concentrations of EGTA were added to IFT27-GFP pf18 cells grown in TAP media ( contains 0 . 34 mM of Ca2+ ) and the movies were recorded within 15 min after EGTA addition . The concentration of free Ca2+ in the assay buffer as a function of added EGTA was calculated from the Chelator program ( http://maxchelator . stanford . edu ) . Only the cells with fully grown flagella ( 8–12 µm in length ) were analyzed . Immotile IFT trains from the beginning to the end of the image acquisition were excluded from data analysis . The pausing frequency was calculated from dividing the total number of pausing events by the image acquisition time . Force measurements on bead motility were carried out with a custom-built optical trap microscope ( Gennerich et al . , 2007 ) , with single molecule fluorescence detection ability . A 1064 nm laser ( Coherent , Compass ) was mounted on an inverted microscope equipped with a 100× 1 . 49 NA oil immersion objective ( Nikon ) . The trapping beam was steered by a computer-controlled acousto-optic deflector ( AA Electronics ) at 20 kHz . Trap stiffness was calibrated for each bead from the amplitude of its thermal diffusion . The beads were trapped by a 400 mW 1064 nm laser beam to achieve an average spring constant of ∼0 . 4 pN/nm . The bead displacement was detected by a quadrant photodiode ( QPD ) and recorded at 2 kHz . Experiments were carried out with the trap position fixed . Stall forces were defined as the magnitude of the opposing force that reduces the mean velocity of the cargo to 0 for more than 100 ms . Returning events with stalling periods shorter than 100 ms at peak forces were scored as releases . Movements of beads beyond the linear range of the detector ( ±200 nm , ±80 pN ) were scored as escapes . The trapping assays at restrictive conditions were performed at 34°C . Optical trap microscope was equipped with a 488 nm laser beam with near-TIRF excitation to track individual trains in IFT27-GFP cells . Bright-field illumination was turned on for ∼1 s at every 1–2 min to monitor the bead position along the flagella and to synchronize the PSD signal and CCD image . Carboxylated latex beads ( 0 . 92 µm diameter , Invitrogen ) were coated with 0 . 05 mg/ml anti-carbohydrate mouse monoclonal antibody to FMG1-B for flagellar membrane attachment . Cover glasses were pre-treated with 0 . 7 mg/ml polylysine for 5 min . 10 µl pf18 cell culture was flowed into a flow chamber and incubated for 30 s to allow cells to attach to the cover glass . 10 µl solution including 0 . 2× TAP , 1 mM CaCl2 , 0 . 3 mg/ml BSA , and 0 . 1 g/l beads was then flowed into the chamber to replace the buffer . Trapped beads were positioned over surface immobilized flagella ( Guilford and Bloodgood , 2013 ) . The fraction of moving beads was measured by resting beads on the flagellar surface for 1 min . Viscous drag constant between the bead and the flagellar membrane was measured by oscillating a flagellar surface attached bead ±500 nm along the length of flagellum in a square wave pattern at 0 . 2 Hz . The connection between the bead and FMG-1B was verified by moving the trap away from the flagellum . If the bead displayed active movement powered by IFT trains during bead oscillation measurements , the bead was lifted for 0 . 5–1 . 0 μm to dissociate it away from IFT and then brought back to flagella . Speeds were measured from linear fits to the displacement traces observed in kymographs . In order to distinguish between IFT trains moving in anterograde and retrograde directions and IFT trains pausing over a period of time , we assigned different colors to individual IFT trains as a function of their velocity by Fourier space directional analysis ( Figure 2—figure supplement 2 ) . The image was then transformed back into real space to recover the kymographs . To perform IFT kymography analysis during gliding motility , we separately determined the position of IFT trains relative to the cell body , and the position of the cell body relative to the glass surface ( Figure 4 ) . Correlation between the movements of beads and IFT trains was demonstrated by rejecting the null hypothesis ( IFT and bead movement was uncorrelated ) using Kolmogorov-Smirnov statistics . | Cilia and flagella protrude like bristles from the cell surface . They share the same basic ‘9+2’ axoneme structure , being made up of nine microtubule doublets that surround a central pair of singlet microtubules . Flagella are generally involved in cell propulsion , whereas motile cilia help to move fluids over cell surfaces . Maintaining cilia and flagella is a challenge for cells , which must find a way to send new proteins all the way along the axoneme to the site of assembly at the flagellar tip . Cells achieve this via a process called intraflagellar transport , in which proteins are carried back and forth by kinesin and dynein motors along the axonemal doublet microtubules . Intraflagellar transport has been proposed to influence other functions of cilia and flagella , including the propulsion of cells over surfaces . However , the details of these interactions are unclear . Through a combination of biophysical and microscopy approaches , Shih et al . describe the mechanism that the green alga Chalmydomonas uses to power flagellar gliding over surfaces . By tracking single fluorescently tagged molecules , Shih et al . observed that flagellar membrane glycoproteins are carried along the axoneme by the intraflagellar transport machinery . During transport , flagellar membrane glycoproteins transiently adhere to the surface , and dynein motors that were previously engaged in carrying these glycoproteins now transmit force that moves the axonemal microtubules . This process , which is dependent on the concentration of calcium ions in the extracellular environment , generates the force that propels the alga's flagella along the surface . Gliding motility is thought to have been one of the initial driving forces for the evolution of cilia and flagella . How the intricate mechanism of flagellar beat motility could have evolved has been the subject of much discussion , as it would require the flagellum to have evolved first . In demonstrating that gliding motility is powered by the same intraflagellar transport mechanism that is required for flagellar assembly , Shih et al . provide strong evidence for the evolution of primitive flagella before the evolution of flagellar beating . Furthermore , since algal flagella have essentially the same structure as the cilia of human cells , these findings could ultimately aid in the development of treatments for diseases that result from defects in intraflagellar transport , including polycystic kidney disease and retinal degeneration . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2013 | Intraflagellar transport drives flagellar surface motility |
We show that multiple , functionally specialized cohesin complexes mediate the establishment and two-step release of sister chromatid cohesion that underlies the production of haploid gametes . In C . elegans , the kleisin subunits REC-8 and COH-3/4 differ between meiotic cohesins and endow them with distinctive properties that specify how cohesins load onto chromosomes and then trigger and release cohesion . Unlike REC-8 cohesin , COH-3/4 cohesin becomes cohesive through a replication-independent mechanism initiated by the DNA double-stranded breaks that induce crossover recombination . Thus , break-induced cohesion also tethers replicated meiotic chromosomes . Later , recombination stimulates separase-independent removal of REC-8 and COH-3/4 cohesins from reciprocal chromosomal territories flanking the crossover site . This region-specific removal likely underlies the two-step separation of homologs and sisters . Unexpectedly , COH-3/4 performs cohesion-independent functions in synaptonemal complex assembly . This new model for cohesin function diverges from that established in yeast but likely applies directly to plants and mammals , which utilize similar meiotic kleisins .
In all organisms , faithful segregation of chromosomes during cell division is essential for genome stability . Accurate chromosome transmission is required both for the proliferative cell divisions that occur during mitosis and the sequential divisions that occur during meiosis to reduce genome copy number from two in diploid germline stem cells to one in haploid gametes . Approximately 30% of human zygotes have abnormal chromosomal content at conception due to defects in meiosis . Such aneuploidy is a leading cause of miscarriages and birth defects ( Hassold and Hunt , 2001 ) , and is thought to result , in part , from defects in sister chromatid cohesion ( SCC ) ( Chiang et al . , 2012; Jessberger , 2012; Nagaoka et al . , 2012 ) . SCC tethers replicated sister chromatids during mitosis and meiosis and is critical for accurate chromosome segregation . SCC is mediated by an evolutionarily conserved protein complex called cohesin . The cohesin complex is composed of two long coiled-coil proteins of the Structural Maintenance of Chromosomes ( SMC ) family , called Smc1 and Smc3 , a non-SMC protein called Scc3 , and a fourth subunit called the α-kleisin ( Nasmyth and Haering , 2009 ) . Smc1 , Smc3 and the kleisin form a tripartite ring proposed to mediate SCC by encircling sister chromatids . The kleisin subunit differs between mitotic and meiotic cohesin complexes . During yeast meiosis , the mitotic kleisin Scc1 is replaced by the meiosis-specific kleisin Rec8 ( Klein et al . , 1999 ) . This substitution is crucial for the reduction of ploidy . We recently showed that the dual kleisin model derived for yeast is insufficient to explain how cohesin complexes facilitate the reduction of genome copy number in all organisms , since Rec8 is not the sole meiotic kleisin in many organisms ( Severson et al . , 2009 ) . Here , we establish a new model: multiple , functionally specialized cohesin complexes that differ in their kleisin subunit perform distinct roles in reducing ploidy . The kleisin influences nearly all aspects of meiotic cohesin function , including how a cohesin complex loads onto meiotic chromosomes , how a complex becomes cohesive once loaded , and when , where and how a complex is removed from chromosomes in meiotic prophase . We first summarize the known roles of meiotic kleisins to provide context for these findings . Analysis of rec8 mutants in numerous sexually reproducing organisms showed that Rec8 cohesin is essential for the three key events that are unique to meiosis and underlie the production of haploid gametes ( DeVeaux and Smith , 1994; Bhatt et al . , 1999; Klein et al . , 1999; Watanabe and Nurse , 1999; Buonomo et al . , 2000; Pasierbek et al . , 2001; Yokobayashi et al . , 2003; Bannister et al . , 2004; Parra et al . , 2004; Chelysheva et al . , 2005; Severson et al . , 2009; Tachibana-Konwalski et al . , 2010; Shao et al . , 2011 ) . First , homologous chromosomes become covalently linked through reciprocal exchange of DNA during the process of crossover ( CO ) recombination . COs promote accurate homolog segregation during anaphase of meiosis I , and Rec8 cohesin is required for efficient CO formation and maintenance . Second , sister chromatids attach to microtubules from the same spindle pole ( co-orient ) in meiosis I to ensure that spindle forces pull homologs apart but not sister chromatids . Sister chromatids then attach to microtubules from opposite spindle poles ( bi-orient ) in meiosis II , as they do in mitosis . Rec8 cohesin facilitates co-orientation . Third , spatially-regulated release of meiotic SCC must occur in two steps to allow the sequential separation of homologs in anaphase I and then sisters in anaphase II . Rec8 cohesin is essential for the linkages that tether sisters until anaphase II . The widely conserved meiotic defects of rec8 mutants reinforced the view of Rec8 as the sole meiotic kleisin . Our recent work challenged this prevalent view by demonstrating that Caenorhabditis elegans gametogenesis requires two nearly identical and functionally redundant predicted α-kleisins , called COH-3 and COH-4 ( hereafter , COH-3/4 ) , in addition to REC-8 ( Severson et al . , 2009 ) . REC-8 and COH-3/4 together mediate meiotic SCC , and severe disruption of SCC occurs only when all three kleisins are removed , suggesting the formation of cohesin complexes that differ in their kleisin subunit . Moreover , REC-8 and COH-3/4 are required for CO recombination . CO recombination fails in rec-8 single mutants and in coh-4 coh-3 double mutants , causing homologs to remain detached . Although REC-8 and COH-3/4 are both required for CO formation and act in concert to mediate SCC , they perform distinct roles in meiotic chromosome segregation ( Severson et al . , 2009 ) . Unlike REC-8 , COH-3/4 cannot co-orient sisters or mediate SCC that persists until anaphase II . Consequently , in rec-8 mutants , sister chromatids are tethered by COH-3/4-dependent SCC until anaphase I , when they segregate prematurely toward opposite spindle poles ( equational division ) . In contrast , in coh-4 coh-3 mutants , REC-8 cohesin co-orients sisters during meiosis I and tethers sisters until anaphase II . Consequently , sister chromatids remain together while homologs segregate randomly during anaphase I . Subsequent to the discovery of COH-3/4 in C . elegans , meiotic kleisins similar to COH-3/4 were identified in plants and mammals ( Herran et al . , 2011; Ishiguro et al . , 2011; Lee and Hirano , 2011; Llano et al . , 2012; Yuan et al . , 2012 ) . The involvement of these kleisins in meiotic SCC likely explains why cohesion persists in rec8 mutants of Arabidopsis , maize , and mouse , as it does in C . elegans ( Bhatt et al . , 1999; Bannister et al . , 2004; Chelysheva et al . , 2005; Xu et al . , 2005; Golubovskaya et al . , 2006; Severson et al . , 2009 ) . The involvement of multiple kleisins in gametogenesis is therefore widely conserved , and our current study dissects the mechanisms by which the kleisin subunit influences cohesin function . Here , we show that REC-8 and COH-3/4 are bona fide kleisin subunits of meiotic cohesin complexes , and that the mechanisms that regulate cohesin loading , sister chromatid cohesion , and cohesin removal are strongly affected by the kleisin subunit . We identify factors required for association of REC-8 cohesin , but not COH-3/4 cohesin , with meiotic chromosomes , providing strong evidence of complex-specific loading mechanisms . We show that COH-3/4 cohesin is triggered to become cohesive , and thereby establish SCC , independently of DNA replication and requires the programmed , SPO-11-dependent double-strand DNA breaks ( DSBs ) that initiate meiotic recombination . This result was not expected , because prior work showed that yeast mitotic cohesin loads onto chromosomes during telophase or G1 of the cell cycle and becomes cohesive only during S phase ( Uhlmann and Nasmyth , 1998; Nasmyth and Haering , 2009; Wood et al . , 2010 ) . The sole example of replication-independent SCC establishment occurs in mitotically proliferating yeast that suffer DNA damage in G2 or M of the cell cycle ( Ström et al . , 2007; Unal et al . , 2007 ) . The SCC formed in response to DSBs is thought to reinforce the cohesion generated during S phase . Since Rec8 cannot generate SCC in response to DNA damage , damage-induced SCC was thought to occur only in proliferating cells ( Heidinger-Pauli et al . , 2008 ) . Our data indicate that damage-induced SCC is an essential feature of meiosis . Finally , we show that prior to homolog separation in anaphase I , REC-8 and COH-3/4 cohesins become selectively removed from complementary domains that flank the single CO of each worm chromosome in a separase-independent manner , consistent with their distinct roles in meiotic chromosome segregation: COH-3/4 becomes enriched where SCC is released at anaphase I and REC-8 becomes enriched where sister chromatids co-orient and SCC persists until anaphase II . Because REC-8 alone can co-orient sisters and mediate SCC that persists after anaphase I , this reciprocal pattern of cohesin removal may facilitate or underlie the stepwise separation of homologs and sister chromatids . This finding contrasts with the two-step cohesion release mechanism of yeast that utilizes only Rec8 and factors like Mei-S332/Shogushin to protect centromeric Rec8 cohesin from degradation during anaphase I , thereby ensuring sister cohesion until anaphase II . Our findings not only reveal unanticipated features of meiosis in C . elegans , but also establish models of meiotic cohesin function applicable to gametogenesis in plants and mammals .
During C . elegans meiosis , the α-kleisin paralogs REC-8 and COH-3/4 function in sister chromatid cohesion ( SCC ) but perform specialized functions ( Severson et al . , 2009 ) , suggesting their participation in independent cohesin complexes ( Severson et al . , 2009 ) . To test this hypothesis , we generated antibodies that recognize both COH-3 and COH-4 and assessed whether COH-3/4 associate with meiotic chromosomal axes , as expected for cohesin subunits . Indeed , REC-8 and COH-3/4 co-localized with the chromosomal axis protein HTP-3 ( Goodyer et al . , 2008 ) in pachytene nuclei of wild-type animals ( Figure 1A ) . The COH-3/4 antibody recognizes COH-3 and COH-4 specifically , since staining was undetectable in coh-4 coh-3 double mutants , but strong staining persisted in both single mutants ( Figure 1—figure supplement 1A ) , as expected from the complete genetic redundancy of the coh-4 and coh-3 genes ( Severson et al . , 2009 ) . 10 . 7554/eLife . 03467 . 003Figure 1 . Multiple cohesin complexes that differ in their kleisin subunit bind to C . elegans meiotic chromosomes . Interdependent loading of REC-8 and COH-3/4 with cohesin SMC proteins is demonstrated . Shown are Z-projected confocal sections through pachytene nuclei ( A–D ) , the distal region of the gonad ( E ) , and entire dissected gonads ( F ) . ( A ) The predicted α-kleisins REC-8 and COH-3/4 are present along synapsed homologs in pachytene nuclei of wild-type animals and co-localize with the axis protein HTP-3 , as expected for meiotic kleisins . COH-3/4 ( B ) and REC-8 ( C ) both require SMC-1 for their association with meiotic chromosomes but bind chromosomes independently . ( D ) SMC-3 associates with chromosomes of rec-8 and coh-4 coh-3 mutants , but SMC-3 staining is undetectable in kleisin triple mutants and smc-1 ( RNAi ) animals . ( E ) The distal region of the gonad holds nuclei undergoing mitotic proliferation and premeiotic DNA replication ( Premeiotic Zone ) and nuclei that have entered prophase of meiosis I ( Transition Zone ) . REC-8 is strongly expressed in all germline nuclei , including S phase nuclei , which express GFP::PCN-1 . In contrast , COH-3/4 staining is undetectable in GFP::PCN-1 positive nuclei and first appears on meiotic chromosomes in the transition zone , indicating that COH-3/4 cohesin becomes cohesive independently of DNA replication . ( F ) glp-1 ( gf ) mutations prevent initiation of meiosis; consequently , the gonad fills with mitotically proliferating germ cell nuclei . Robust expression of REC-8 , but not COH-3/4 , is detected in the mitotic nuclei of glp-1 ( gf ) worms , indicating that COH-3/4 is first expressed during meiosis . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 00310 . 7554/eLife . 03467 . 004Figure 1—figure supplement 1 . REC-8 accumulates in hermaphrodite gonads prior to the initiation of meiosis , while COH-3/4 becomes detectable only in meiotic nuclei . ( A ) Rabbit polyclonal antibodies recognize both COH-3 and COH-4 . COH-3/4 antibodies label meiotic chromosomes in wild-type animals as well as coh-3 and coh-4 single mutants . Antibody staining is undetectable only in coh-4 coh-3 double mutants . ( B ) Cartoon showing that premeiotic nuclei and nuclei in various stages of meiosis occupy distinct , predictable regions of the gonad in wild-type animals . ( C ) Confocal micrographs of germline nuclei stained with DAPI and antibodies to REC-8 and COH-3/4 . COH-3/4 are undetectable in the distal-most region of the gonad , which contains mitotically-cycling germline stem cells and nuclei in premeiotic S phase . Arrowheads indicate a metaphase figure . COH-3/4 appear abruptly on chromosomes at the onset of meiotic prophase ( transition zone , TZ ) , and COH-3/4 are detected along the entire chromosomal axis in pachytene . By prometaphase of meiosis I , COH-3/4 are detected only at the short arm . The pattern of COH-3/4 localization differs from that of REC-8 in two important ways . First , REC-8 is expressed in premeiotic nuclei , although it is unclear whether REC-8 is bound to chromosomes at this stage . Second , while COH-3/4 become restricted to the short arm by prometaphase I , REC-8 is removed from the short arm but persists at the long arm . ( D ) Gain-of-function alleles of glp-1 prevent initiation of meiosis , and consequently , the germline fills with mitotically proliferating nuclei . High magnification confocal images of regions of the gonad that would contain premeiotic , pachytene , or diakinesis nuclei in wild-type animals are shown . REC-8 is highly expressed in all of these nuclei , but COH-3/4 are not . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 004 As subunits of distinct cohesin complexes , COH-3/4 and REC-8 are expected to bind chromosomes independently of each other and to require the SMC subunits for their chromosomal association . Moreover , SMC staining should persist in rec-8 single and coh-4 coh-3 double mutants but not in rec-8; coh-4 coh-3 triple mutants ( hereafter called kleisin triple mutants ) . These expectations were met ( Figure 1B–D ) . Long tracks of COH-3/4 ( Figure 1B ) and SMC-3 ( Figure 1D; Chan et al . , 2003 ) staining were evident on meiotic chromosomes of rec-8 single mutants , and both REC-8 ( Figure 1C ) and SMC-3 ( Figure 1D ) persisted on chromosomes of coh-4 coh-3 mutants . However , levels of REC-8 and SMC-3 were reduced in coh-4 coh-3 double mutants compared to wild-type animals , and both proteins appeared in dispersed puncta rather than in linear structures , as previously noted for the axial element HTP-3 ( Severson et al . , 2009 ) . The disorganized localization and reduced staining intensity of REC-8 and SMC-3 reflect the failure to form continuous chromosomal axes . Loss of COH-3/4 binding also reduces SMC-3 levels . In contrast , SMC-3 was nearly undetectable on meiotic chromosomes of kleisin triple mutants ( Figure 1D ) . In converse experiments , binding of REC-8 ( Figure 1C ) and COH-3/4 ( Figure 1B ) to meiotic chromosomes was severely disrupted in smc-1 ( RNAi ) animals , as was binding of SMC-3 ( Figure 1D ) . These results provide strong evidence that REC-8 and COH-3/4 associate with chromosomes as subunits of independent meiotic cohesin complexes that differ in their kleisin subunit ( Figure 1A ) . In proliferating cells , cohesin loading and SCC establishment are temporally separate events: Scc1 cohesin loads onto chromosomes prior to S phase and becomes cohesive during DNA replication ( Guacci et al . , 1997; Michaelis et al . , 1997; Losada et al . , 1998; Uhlmann and Nasmyth , 1998 ) . C . elegans REC-8 cohesin appears to behave similarly , since REC-8 accumulates before premeiotic S phase ( Pasierbek et al . , 2001; Hayashi et al . , 2007; Severson et al . , 2009 ) . To our surprise , COH-3/4 cohesin behaves differently: COH-3/4 become detectable during meiotic prophase , after completion of premeiotic replication . To determine the precise timing of REC-8 and COH-3/4 accumulation , we examined the staining of each kleisin during the different stages of germ-cell development . In C . elegans , changes in chromosomal morphology and nuclear position distinguish germ-cell nuclei undergoing mitotic proliferation or premeiotic DNA replication from nuclei in meiotic prophase I ( Figure 1—figure supplement 1B ) . Chromosomes are dispersed in premeiotic nuclei , which occupy the most distal region of the gonad . Upon initiation of meiosis , chromosomes cluster in a crescent on one side of the nucleus , opposite the nucleolus , in a region of the gonad called the transition zone ( Francis et al . , 1995; MacQueen and Villeneuve , 2001 ) . Nuclei in this region are in the leptotene and zygotene stages of prophase I , when synaptonemal complexes ( SCs ) assemble between homologs . In leptotene , linear structures called axial elements ( AEs ) form along the length of meiotic chromosomes . In zygotene , central region ( CR ) proteins assemble between homologous AEs , tethering homologs along their lengths in a process called synapsis . Chromosomes remain clustered until pachytene , when the fully-synapsed homologs redistribute around the nuclear periphery . As shown previously , we detected REC-8 in all germ-cell nuclei of the gonad , including premeiotic nuclei , consistent with SCC establishment in premeiotic S phase ( Figure 1E , Figure 1—figure supplement 1C ) ( Pasierbek et al . , 2001; Hayashi et al . , 2007; Severson et al . , 2009 ) . In contrast , COH-3/4 was not detected in premeiotic nuclei , but intense COH-3/4 staining appeared abruptly on meiotic chromosomal axes in the transition zone and persisted through prophase I , suggesting that COH-3/4 first accumulates at the onset of meiosis ( Figure 1E , Figure 1—figure supplement 1C ) . Two additional lines of evidence supported this conclusion . First , COH-3/4 staining was not detected in nuclei that expressed PCNA , an S phase-specific DNA polymerase processivity factor used for mitotic and premeiotic DNA replication ( Figure 1E ) . Second , in animals carrying a glp-1 gain-of-function allele that blocks the initiation of meiosis ( Berry et al . , 1997 ) , REC-8 , but not COH-3/4 , was strongly expressed in the mitotically proliferating nuclei that filled the entire gonad ( Figure 1F , Figure 1—figure supplement 1D ) . Thus , COH-3/4 associates with chromosomes after premeiotic S phase is complete , suggesting that COH-3/4 and REC-8 cohesins may load onto chromosomes and become cohesive by different mechanisms . The cohesin loading factors identified to date , exemplified by the heterodimeric Scc2/Scc4 complex , are required for the loading of all cohesin complexes examined , regardless of subunit composition ( Ciosk et al . , 2000; Gillespie and Hirano , 2004; Takahashi et al . , 2004; Lightfoot et al . , 2011 ) . Our finding that REC-8 accumulates in both premeiotic and meiotic nuclei , but COH-3/4 accumulates only in meiotic nuclei , suggested that REC-8 and COH-3/4 cohesins might load by different mechanisms . We therefore examined COH-3/4 loading in two mutant strains in which REC-8 was undetectable on meiotic chromosomes but SCC persisted: htp-3 mutants lacking the HORMA domain AE protein HTP-3 ( Goodyer et al . , 2008; Severson et al . , 2009 ) , and tim-1 mutants lacking the C . elegans TIMELESS homolog TIM-1 ( Chan et al . , 2003 ) ( Figure 2 ) . TIMELESS was initially identified in Drosophila as a factor involved in circadian rhythms , but TIMELESS orthologs were subsequently shown to function during DNA replication ( Gotter et al . , 2007 ) . In both mutants , REC-8 was detected at normal levels in premeiotic nuclei but failed to associate with meiotic chromosomes ( Chan et al . , 2003; Severson et al . , 2009 ) . In contrast , COH-3/4 loading appeared normal in htp-3 and tim-1 mutants ( Figure 2 ) . Thus , specific factors likely promote the differential loading of cohesin complexes , and the kleisin subunit specifies the mechanism by which a complex initially associates with meiotic chromosomes . 10 . 7554/eLife . 03467 . 005Figure 2 . The kleisin subunit determines mechanisms of cohesin loading . Confocal micrographs of pachytene nuclei reveal that the axial element protein HTP-3 ( A ) and the Timeless ortholog TIM-1 ( B ) are both essential for REC-8 cohesin loading , but neither protein is needed for COH-3/4 cohesin loading . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 005 To determine whether REC-8 and COH-3/4 cohesins also differ in their requirements for triggering SCC , we first developed a reliable assay for evaluating SCC . We assessed sister-chromatid tethering in strains carrying an array of lac operator repeats ( lacO ) that had been integrated into one of the two chromosome V homologs ( Figure 3C , D ) ( Gonzalez-Serricchio and Sternberg , 2006 ) . LacI::GFP binding to lacO repeats uniquely marks the two sister chromatids of the homolog harboring the array ( Belmont and Straight , 1998; Gonzalez-Serricchio and Sternberg , 2006 ) . The ability to identify a specific pair of sister chromatids and measure the distance between them permits more accurate quantification of SCC defects than previous methods reliant on fluorescence in situ hybridization or counting of DAPI-staining bodies . 10 . 7554/eLife . 03467 . 006Figure 3 . A conserved mechanism initiates SCC in response to programmed DSBs in C . elegans meiosis and exogenous DNA breaks in budding yeast mitosis . ( A ) In S . cerevisiae , DNA damage in G2/M activates ATR and Chk1 , resulting in Scc1 phosphorylation and SCC establishment . ( B ) A model for SCC establishment by COH-3/4 cohesin . DSBs created by SPO-11 activate ATM/ATR and CHK-2 , leading to COH-3/4 phosphorylation and generation of SCC . ( C–G ) Data supporting the model in ( B ) . Images on the left show projected Z-sections through entire diakinesis nuclei stained with LacI::GFP ( green ) and DAPI ( red ) . LacI::GFP bound to a heterozygous lacO array integrated into chromosome V reveals whether sisters are held together by SCC . Charts on the right show quantification of distances between LacI::GFP foci . 0 µm indicates that discrete GFP foci could not be resolved . no . = number of nuclei scored . ( C ) LacI::GFP labels a single bivalent in wild-type animals , and the two sisters of a single univalent in rec-8 mutants . ( D ) Sister chromatids are held together by REC-8-dependent SCC in coh-4 coh-3 double mutants , but are apart in kleisin triple mutants . ( E ) Sister chromatids are held together by SCC in spo-11 mutants , but not in spo-11 rec-8 mutants . DSBs induced by γ-irradiation restore SCC in spo-11 rec-8 mutants . ( F ) Sisters are apart and extensive chromosomal fragmentation and rearrangement occurs in atm-1; rec-8; atl-1 animals , but not in atm-1; coh-4 atl-1 coh-3 mutants . ( G ) Cohesion between sisters is not established in rec-8; chk-2 mutants . Irradiation of rec-8; chk-2 mutants does not restore SCC but does induce chromosome fragmentation and rearrangement , demonstrating a role for CHK-2 in SCC establishment that is downstream of DSB formation . CHK-1 is not required for COH-3/4-dependent SCC ( Figure 3—figure supplement 3B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 00610 . 7554/eLife . 03467 . 007Figure 3—figure supplement 1 . Beeswarm plots show individual distances between LacI::GFP foci in diakinesis nuclei . Red horizontal lines indicate the median , and blue horizontal lines indicate the interquartile range . A distance of >1 . 5 microns indicates defective SCC . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 00710 . 7554/eLife . 03467 . 008Figure 3—figure supplement 2 . Table of significance values for distances between LacI::GFP foci in diakinesis nuclei . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 00810 . 7554/eLife . 03467 . 009Figure 3—figure supplement 3 . REC-8 and COH-3/4 cohesin tether SCC by different mechanisms . Shown are projected confocal images of entire diakinesis nuclei stained with DAPI ( red ) and LacI::GFP ( green ) . SCC was assessed by the distribution of LacI::GFP bound to a lacO array integrated into one ( A ) or both ( B ) chromosome V homologs . Bar charts show quantification of distances between LacI::GFP foci . 0 µm indicates that discrete GFP foci could not be resolved . no . = number of nuclei scored . ( A ) SPO-11-dependent DSBs trigger SCC mediated by COH-3/4 but not REC-8 cohesin . In rec-8 single mutants , the sister chromatids of each homolog are tethered together by COH-3/4-dependent SCC . Similarly , sisters are held together by REC-8-dependent SCC in coh-4 coh-3 double mutants . Sisters are detached in diakinesis nuclei of spo-11 rec-8 double mutants , rec-8; coh-4 coh-3 triple mutants , and spo-11 rec-8; coh-4 coh-3 quadruple mutant animals , and the frequency of detachment and the distance between sisters are similar in all three genotypes . In contrast , disrupting SPO-11 function in coh-4 coh-3 double mutants does not lead to cohesion defects . ( B ) COH-3/4-dependent SCC requires CHK-2 , but not CHK-1 . Diakinesis nuclei of rec-8; chk-1 ( RNAi ) animals resemble those of rec-8 single mutants , while sisters are detached in diakinesis nuclei of rec-8; chk-2 ( RNAi ) worms . Thus , CHK-2 , but not CHK-1 , is required for the establishment of COH-3/4 dependent SCC . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 009 We validated the assay and established a quantitative metric for SCC by measuring the distance between sister chromatids in nuclei of ( 1 ) wild-type animals in which both REC-8 and COH-3/4 cohesins tether sisters , ( 2 ) kleisin triple mutants in which SCC is not established ( Severson et al . , 2009 ) , and ( 3 ) rec-8 single or coh-4 coh-3 double mutants in which SCC is achieved by a single type of cohesin . Because REC-8 and COH-3/4 cohesins are both sufficient to tether sisters in diakinesis nuclei ( Severson et al . , 2009 ) , the contribution of REC-8 to SCC could only be assessed in coh-4 coh-3 mutants , and the contribution of COH-3/4 could only be assessed in rec-8 mutants . Our initial analysis focused on nuclei in late diakinesis , the final stage of meiotic prophase I , because diakinesis chromosomes are highly compacted and the nuclear volume is much greater than that in earlier stages of meiosis . As expected , a single GFP focus was detected in 100% of diakinesis nuclei of wild-type worms and coh-4 coh-3 double mutants ( Figure 3C , D and Figure 3—figure supplement 1 ) , consistent with the finding that REC-8 cohesin is sufficient to tether and co-orient sister chromatids ( Severson et al . , 2009 ) . LacI::GFP also labeled a single detached homolog ( univalent ) in most diakinesis nuclei of rec-8 worms; however , two discrete GFP foci could usually be detected within the univalent ( Figure 3C and Figure 3—figure supplement 1 ) . This finding was also anticipated because sister chromatids bi-orient in rec-8 mutants , and the univalents adopt a dumbbell shape in which the two sister chromatids can usually be resolved ( Figure 3C; Severson et al . , 2009 ) . In 84% of rec-8 nuclei , either a single GFP focus was detected or two foci were detected within one univalent at a spacing of ≤1 . 5 µm ( Figure 3C ) . In contrast , only 10% of diakinesis nuclei in kleisin triple mutants had LacI::GFP foci separated by ≤ 1 . 5 µm ( Figure 3D and Figure 3—figure supplements 1 and 2 ) ( Severson et al . , 2009 ) . Based on these measurements , we defined diakinesis nuclei with GFP foci ≤1 . 5 µm apart as having sisters tethered by meiotic SCC , and nuclei with GFP foci >1 . 5 µm apart as having sisters separated due to defective SCC . The number and morphology of DAPI-staining bodies in diakinesis nuclei of each genotype examined indicated that other chromosomes behaved similarly to chromosome V ( Figure 3C , D ) . Thus , the distance between LacI::GFP spots is a reliable measure of the global status of SCC in diakinesis nuclei . Using the SCC assay , we assessed whether COH-3/4 cohesin becomes cohesive independently of DNA replication , as suggested by the lack of detectable COH-3/4 in PCNA-positive nuclei ( Figure 1E ) . Because the sole example of replication-independent SCC establishment occurs during yeast G2/M phase in response to DNA breaks ( Ström et al . , 2007; Unal et al . , 2007 ) , we hypothesized that the programmed , SPO-11-dependent DSBs used to initiate CO recombination might trigger COH-3/4 to become cohesive . We found sisters to be apart in 90% of diakinesis nuclei of spo-11 rec-8 double mutants , indicating that COH-3/4 requires SPO-11 to tether sister chromatids ( Figure 3E and Figure 3—figure supplements 1 and 2 ) . In contrast , sisters could not be resolved in any nuclei of spo-11; coh-4 coh-3 triple mutants , indicating that REC-8 cohesin becomes cohesive independently of SPO-11 ( Figure 3—figure supplement 3A ) . The SCC disruption in spo-11 rec-8 mutants did not result from a defect in COH-3/4 cohesin loading , since levels of bound COH-3/4 were similar in wild-type animals , spo-11 rec-8 double mutants , and spo-11 or rec-8 single mutants ( Figure 4A ) . 10 . 7554/eLife . 03467 . 010Figure 4 . Analysis of COH-3/4 cohesin loading and DSB formation and repair in SCC-defective worms . ( A ) Imaging of pachytene nuclei stained with antibodies to COH-3/4 and HTP-3 demonstrated that COH-3/4 associates with meiotic axes in most mutants that fail to establish COH-3/4-dependent SCC . Although COH-3/4 associates with chromosomes of him-3 rec-8 animals , the intensity of COH-3/4 signal is less than that detected in him-3 single mutants , which , in turn , is less than that in wild-type animals ( See also Figure 4—figure supplement 2 ) . A reduction in signal is also true of DAPI and HTP-3 , which loads onto chromosomes independently of HIM-3 ( Goodyer et al . , 2008; Severson et al . , 2009 ) . Thus , the strong staining of COH-3/4 and HTP-3 observed in wild-type nuclei likely results from the close association of the four chromatid axes via synapsis and SCC , while the reduced staining in him-3 mutants likely results from homolog separation due to defective synapsis , and in him-3 rec-8 mutants from homolog and sister separation due to defective synapsis and SCC . Consistent with this model , a similar reduction in the intensity of COH-3/4 and HTP-3 staining was detected in rec-8 animals also lacking the CR protein SYP-1 , which is dispensable for chromosomal loading of all known AE proteins ( MacQueen et al . , 2002; AFS unpublished data ) . ( B and C ) Confocal images of early ( B ) and late ( C ) pachytene nuclei stained with DAPI ( red ) and antibodies to DSB marker RAD-51 ( green ) . Abundant RAD-51 foci are detected in him-3 rec-8 and rec-8; syp-1 mutants , indicating that DSBs are formed . RAD-51 staining persists abnormally late in these mutants , and chromosomal fragmentation and fusions are evident in diakinesis nuclei ( D ) stained with DAPI ( red ) and LacI::GFP ( green ) as in Figure 3C . Thus , establishment of COH-3/4-dependent SCC is essential for homology-directed DSB repair in animals homozygous for a rec-8 deletion . Remarkably , such rearrangements are not detected in kleisin triple mutants . Explaining this finding , few RAD-51 foci form in kleisin triple mutants . Those that do appear in late pachytene , well after DSBs are repaired in wild-type animals . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 01010 . 7554/eLife . 03467 . 011Figure 4—figure supplement 1 . Enlargements of early and late pachytene nuclei of wild-type and mutant animals stained with RAD-51 antibodies . Abundant RAD-51 foci are present in nuclei of all genotypes , except in rec-8; coh-4 coh-3 mutants , which have a severely reduced number of foci . RAD-51 foci are very rarely present in early pachytene nuclei , and occasional foci appear in late pachytene nuclei . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 01110 . 7554/eLife . 03467 . 012Figure 4—figure supplement 2 . Enlargements of pachytene nuclei from wild-type animals and him-3 or him-3; rec-8 mutants stained with DAPI and antibodies to COH-3/4 , HTP-3 . COH-3/4 and HTP-3 associate with pachytene chromosomes but the levels are reduced in mutants relative to wild-type animals , likely reflecting a failure of homolog synapsis in him-3 single mutants and defective synapsis and SCC in him-3 rec-8 mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 012 The requirement for SPO-11 in tethering sister chromatids in rec-8 mutants reflects a need for DSBs , since SCC was restored in spo-11 rec-8 mutants treated with γ-irradiation . Prior experiments showed that γ-irradiation restored CO recombination in spo-11 mutants and suppressed lethality ( Dernburg et al . , 1998 ) . In our experiments , 12 Gy of γ-irradiation triggered CO recombination between most homolog pairs of spo-11 mutants without causing chromosomal fragmentation or fusion ( Figure 3E ) . That dose also restored SCC: sisters were tethered in 75% of irradiated spo-11 rec-8 double mutants ( Figure 3E and Figure 3—figure supplements 1 and 2 ) . Thus , programmed meiotic DSBs are essential for COH-3/4-dependent SCC , and the kleisin subunit determines whether a cohesin requires DSBs to establish cohesion . Moreover , our data suggest that meiotic SCC is established in successive waves: during premeiotic DNA replication , the nascent sister chromatids are tethered by Watson-Crick base pairing in unreplicated regions and by REC-8 cohesin in replicated regions . Subsequently , DSBs trigger COH-3/4 cohesin to become cohesive during meiosis . This second wave of SCC establishment reinforces the cohesion generated during DNA replication . Evidence supporting this model is described in the following paragraphs . The essential role for DSBs in COH-3/4-dependent meiotic SCC suggested that the pathway for establishing SCC in response to DNA damage during yeast G2/M phase might similarly trigger COH-3/4-mediated SCC during nematode meiosis . In the yeast model for DSB-induced SCC ( DI-SCC ) , DNA breaks activate the Mec1/ATR kinase . Mec1 in turn stimulates the Chk1 kinase to phosphorylate the mitotic kleisin Scc1 on serine 83 , which triggers cohesin to become cohesive ( Figure 3A; Heidinger-Pauli et al . , 2008 ) . Yeast Rec8 lacks the Chk kinase phosphorylation site and consequently cannot become cohesive in response to DNA damage . Thus , DI-SCC was thought to occur only during G2/M of the mitotic cell cycle . To determine whether a similar signaling cascade occurs in C . elegans meiosis ( Figure 3B ) , we first asked whether ATM and the related kinase ATR ( ATM-1 and ATL-1 , respectively ) are required for the COH-3/4-dependent SCC that tethers sisters in rec-8 mutants . Because ATM-1 and ATL-1 perform partially redundant functions in C . elegans ( Garcia-Muse and Boulton , 2005 ) , we examined cohesion in atm-1; atl-1 double mutants . Sisters were apart in 59% of diakinesis nuclei in atm-1; rec-8; atl-1 ( RNAi ) mutants ( Figure 3F and Figure 3—figure supplements 1 and 2 ) . The persistence of cohesion between some chromatids likely results from incomplete ATL-1 depletion . Unfortunately , SCC cannot be examined in atm-1; rec-8 animals carrying an atl-1 null allele due to severe defects in gonadal development . Importantly , the SCC defects in atm-1; rec-8; atl-1 ( RNAi ) mutants did not result from impaired COH-3/4 cohesin loading ( Figure 4A ) . Furthermore , sisters were always tethered in coh-4 coh-3 mutants deficient in ATM-1 and ATL-1 ( 100% , Figure 3F and Figure 3—figure supplement 1 ) . We conclude that ATM-1 and ATL-1 are required to trigger the cohesiveness of COH-3/4 cohesin , but not REC-8 cohesin . We next asked whether CHK-1 or the paralogous kinase CHK-2 is required for SCC establishment by COH-3/4 cohesin . Diakinesis nuclei of rec-8; chk-1 ( RNAi ) worms had 12 univalents , and LacI::GFP staining showed no sister separation , indicating that CHK-1 kinase is not required ( Figure 3—figure supplement 3B ) . In contrast , sisters were apart in 100% of rec-8; chk-2 nuclei ( Figure 3G , Figure 3—figure supplements 2 , 3 and 3B ) . Because CHK-2 is required for the formation of SPO-11-dependent DSBs ( Alpi et al . , 2003; Martinez-Perez and Villeneuve , 2005 ) , the failure to establish COH-3/4-dependent SCC in rec-8; chk-2 animals could have resulted from the absence of DSBs rather than a failure to respond to DSBs . However , exposure of rec-8; chk-2 mutants to the dose of γ-irradiation that restored SCC in spo-11 rec-8 worms failed to restore SCC in rec-8; chk-2 mutants . Sisters remained apart ( 100% ) , and chromosome fragments were evident , indicating defective repair of DSBs by homologous recombination ( Figure 3G and Figure 3—figure supplements 1 and 2 ) . Thus , DSBs , ATM-1/ATL-1 and CHK-2 , but not CHK-1 , are required for COH-3/4-dependent SCC . The lower levels of COH-3/4 in chk-2 single and rec-8; chk-2 double mutants compared to those in wild-type , spo-11 rec-8 , and atm-1; rec-8; atl-1 animals also suggest the possibility that CHK-2 is required for COH-3/4 cohesin loading in addition to triggering SCC . Our results provide strong evidence that a conserved pathway establishes SCC in response to DSBs in yeast mitosis and in C . elegans meiosis . However , the DI-SCC response during yeast G2/M is initiated by stochastic DSBs and is a secondary mechanism of SCC establishment by Scc1 cohesin , while the DI-SCC response during meiosis is triggered by programmed DSBs that initiate CO recombination and is the primary mechanism for SCC establishment by COH-3/4 cohesin . Analysis of SCC in diakinesis demonstrated the essential role of REC-8 and COH-3/4 in tethering sister chromatids . We next analysed SCC in pachytene nuclei to assess the role of REC-8 and COH-3/4 in triggering SCC . Our analysis confirmed the central role of these meiotic kleisins in establishing SCC and also revealed the unexpected finding that some sister chromatid linkages can also be formed independently of REC-8 and COH-3/4 . Due to the smaller size and different chromosomal structure of pachytene nuclei , distance measurements were divided into 1 µm bins , and sisters were scored as tethered if separated by ≤ 1 µm . In most pachytene nuclei of wild-type worms ( >99% ) , rec-8 single mutants ( 95% ) , and coh-4 coh-3 double mutants ( >99% ) , sisters were tethered ( Figure 5A , B and Figure 5—figure supplement 1 ) . In contrast , sisters were apart in most nuclei of meiotic kleisin triple mutants ( 55% ) , revealing that REC-8 and COH-3/4 cohesins establish SCC . However , the persistence of linkages in 45% of pachytene nuclei in kleisin triple mutants indicated that factors other than REC-8 and COH-3/4 cohesins can also tether sisters ( Figure 5A , B and Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 03467 . 013Figure 5 . Cohesin-dependent and cohesin-independent SCC holds sisters together in pachytene nuclei . ( A ) Projections of confocal Z-sections through entire pachytene nuclei . A single LacI::GFP focus is detected in pachytene nuclei of wild-type , coh-4 coh-3 , and spo-11 mutant worms , indicating that sister chromatids are tethered by SCC . In contrast , sisters are separated in most pachytene nuclei of kleisin triple mutants but still remain close together , suggesting that residual SCC persists . Partial depletion of SCC-1 in kleisin triple mutant animals increases both the frequency of sister separation and the distance between sisters , demonstrating a meiotic role for SCC-1 . Surprisingly , sisters could be resolved in only ∼10% of nuclei in rec-8 and spo-11 rec-8 worms ( white circles ) . The robust synaptonemal complex ( SC ) assembly in spo-11 rec-8 worms suggested that SC proteins may tether sister chromatids independently of cohesin . Indeed , disrupting the axial element ( AE ) protein HIM-3 severely compromised SCC in both rec-8 and spo-11 rec-8 mutants . Disrupting the central region ( CR ) protein SYP-1 had a lesser effect , suggesting that AE proteins can tether sisters together independently of CR proteins and cohesin . ( B ) Quantification of sister separation in pachytene nuclei . no . = number of nuclei scored . ( C ) Z-projected confocal images of wild-type gonads stained with DAPI and antibodies to SCC-1 . Similar to REC-8 , SCC-1 was detected in premeiotic nuclei and became enriched in axial structures of transition zone and pachytene nuclei . Nucleoplasmic staining obscured any chromosomal signal from pachytene exit until prometaphase; however , SCC-1 was undetectable following nuclear envelope breakdown in prometaphase , indicating that SCC-1 cohesin was removed from chromosomes during diplotene or diakinesis . ( D ) Similar sets of kleisins function during meiosis in C . elegans , mammals and plants . ( E ) A schematic of SC structure . Studies in worms have identified four components of the axial/lateral element , or LE ( HTP-3 , HIM-3 , and the functionally redundant proteins HTP-1 and HTP-2 ) and four components of the CR ( SYP-1 , SYP-2 , SYP-3 , and SYP-4 ) . ( F and G ) Two models of SC-dependent linkages between sisters . ( F ) CR proteins link AEs formed along each sister . ( G ) AE proteins hold sisters together independently of CRs . ( H ) REC-8 cohesin and COH-3/4 cohesin load onto chromosomes at different times and establish SCC by different mechanisms . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 01310 . 7554/eLife . 03467 . 014Figure 5—figure supplement 1 . Synaptonemal complex ( SC ) proteins associate with pachytene chromosomes in rec-8 and spo-11 rec-8 animals , but they do not tether homologous chromosomes . Shown are confocal images of nuclei stained with LacI::GFP , which labels lacO arrays integrated into both chromosome V homologs ( top panels ) and antibodies to the axial/lateral element protein HTP-3 and the central region protein SYP-1 ( bottom panels ) ( A ) SC proteins associate with pachytene chromosomes of wild-type animals , rec-8 and spo-11 single mutants , and spo-11 rec-8 double mutants . A single focus of LacI::GFP was detected in pachytene nuclei of wild-type and spo-11 worms , as expected since homologs are fully synapsed in these animals . In contrast , two widely separated GFP foci were detected in rec-8 and spo-11 rec-8 mutants , suggesting that synapsis occurs between non-homologous chromosomes or sister chromatids rather than homologs . ( B ) SC proteins form polycomplexes in pachytene nuclei of rec-8; coh-4 coh-3 triple mutants and spo-11 rec-8; coh-4 coh-3 quadruple mutants . 3–4 LacI::GFP foci can be detected in most nuclei of these animals , consistent with a role for SC proteins in cohesin-independent SCC . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 01410 . 7554/eLife . 03467 . 015Figure 5—figure supplement 2 . Beeswarm plots show individual distances between LacI::GFP foci in pachytene nuclei . Red horizontal lines indicate the median , and blue horizontal lines indicate the interquartile range . A distance of >1 micron indicates defective SCC . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 01510 . 7554/eLife . 03467 . 016Figure 5—figure supplement 3 . Table of significance values for distances between LacI::GFP foci in pachytene nuclei . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 016 We found that the C . elegans mitotic kleisin SCC-1 , previously thought to function only during mitosis , contributes to cohesion in pachytene nuclei ( Figure 5A , B and Figure 5—figure supplements 1 and 2 ) . Since SCC-1 is required for mitotic proliferation of germline precursors , we used RNAi to partially deplete SCC-1 in kleisin triple mutants , then scored SCC in pachytene nuclei . Although LacI::GFP revealed extensive aneuploidy in premeiotic and transition zone nuclei of these animals , likely due to defective SCC during mitotic proliferation of the germline stem cells , only one or two LacI::GFP foci were detected in pachytene nuclei ( Figure 5A , B , data not shown ) , indicating that nuclei with abnormal chromosomal number had not yet progressed into pachytene . Only 12% of pachytene nuclei exhibited cohesion in SCC-1-depleted kleisin triple mutants , indicating that SCC-1 cohesin can mediate REC-8 and COH-3/4-independent linkages ( Figure 5 A , B and Figure 5—figure supplements 1 and 2 ) . However , the finding that two LacI::GFP foci could be resolved in ∼89% of pachytene nuclei of kleisin triple mutants with wild-type levels of SCC-1 demonstrates that , unlike SCC mediated by REC-8 or COH-3/4 cohesin , SCC mediated by SCC-1 cohesin is insufficient to maintain the close association of sisters along their entire lengths . Thus , SCC-1 by itself is not likely to establish SCC in wild-type animals . Consistent with the involvement of SCC-1 cohesin in tethering chromosomes during pachytene but not diakinesis , we observed that SCC-1 associates with axial structures of transition zone and pachytene nuclei of wild-type animals ( Figure 5C ) . SCC-1 staining was detected between homologs , suggesting that SCC-1 cohesin associates with the meiotic chromosomal axis , similar to REC-8 and COH-3/4 cohesins . Diffuse nuclear staining obscured any chromosomal signal in diplotene/diakinesis nuclei; however , SCC-1 was undetectable on chromosomes following nuclear envelope breakdown , when the nucleoplasmic signal dissipated . Thus , any role played by SCC-1 in meiotic SCC of wild-type animals or kleisin triple mutants is likely to occur during pachytene , but not prometaphase I ( See Discussion ) . Unexpectedly , the absence of DSBs in spo-11 rec-8 animals did not abrogate SCC in pachytene nuclei ( 6 . 8% lacking SCC ) , unlike in diakinesis nuclei ( 90% lacking SCC ) ; however , we reasoned that axial element proteins and synaptonemal complex ( SC ) proteins ( Figure 5E ) might mediate the linkages that persist between sisters in pachytene nuclei of spo-11 rec-8 mutants and thereby obscure the role of DSBs . In kleisin triple mutants , unlike in spo-11 rec-8 mutants , SC proteins cannot account for the residual SCC , because SC assembly fails completely in these mutants , and all known SC proteins form nucleoplasmic aggregates called polycomplexes ( Figure 5A , Figure 5—figure supplement 1B ) . However , SC assembly still occurs in both spo-11 rec-8 double mutants and rec-8 single mutants , likely between sister chromatids or non-homologous chromosomes , unlike in wild-type animals , and polycomplexes fail to form ( Figure 5A , Figure 5—figure supplement 1A ) ( Pasierbek et al . , 2001; Martinez-Perez et al . , 2008; Severson et al . , 2009; Rog and Dernburg , 2013 ) . We therefore asked whether axial element ( AE ) proteins ( Figure 5G ) alone , or AE proteins bridged by SC central region ( CR ) proteins ( Figure 5F ) could tether sister chromatids ( Figure 5A , B ) . SCC was severely compromised in pachytene nuclei of animals lacking REC-8 and AE protein HIM-3 . Sister chromatids were apart in ∼70% of pachytene nuclei of him-3 rec-8 animals regardless of whether DSBs were made ( Figure 5A , B and Figure 5—figure supplements 2 and 3 ) , suggesting the involvement of AE proteins in tethering sister chromatids . The SCC defect did not result from a failure to form DSBs or to load COH-3/4 cohesin . Using RAD-51 , a RecA homolog that binds to nascent recombination intermediates just after DSB formation , as a marker for DSBs , we found abundant RAD-51 foci in him-3 rec-8 mutants in early and late pachytene ( Figure 4B , C and Figure 4—figure supplement 1 ) . Furthermore , COH-3/4 associated with meiotic chromosomes of him-3 rec-8 animals , although the intensity of the COH-3/4 signal was less than that in wild-type animals due to defective synapsis and SCC ( Figure 4A and legend and Figure 4—figure supplement 2 ) . We therefore propose that HIM-3 , or a protein that depends on HIM-3 for its loading , can tether sister chromatids during pachytene , independently of cohesin , thereby accounting , in part , for the SCC in spo-ll rec-8 mutants . We also discovered that disrupting the gene encoding the SC central region protein SYP-1 further reduces SCC in spo-11 rec-8 animals ( Figure 5A , B and Figure 5—figure supplements 2 and 3 ) . In spo-11 rec-8; syp-1 triple mutants , 36% of pachytene nuclei lacked SCC compared to 7% of nuclei in spo-11 rec-8 double mutants and 71% of nuclei in him-3 spo-11 rec-8 , indicating that SYP-1 assists in linking sisters independently of cohesin , but AE protein HIM-3 plays a more prominent role . The more minor involvement of SYP-1 in tethering sister chromatids provided the opportunity to assess whether DSBs trigger COH-3/4-dependent SCC . We found that the frequency of sister separation in spo-11 rec-8; syp-1 triple mutants ( 36% ) was greater than in rec-8; syp-1 double mutants ( 14% ) ( p < 0 . 001 ) indicating that DSBs play an important role in triggering COH-3/4-dependent SCC ( Figure 5A , B and Figure 5—figure supplements 2 and 3 ) . Further indication of the key role played by DSBs in establishing COH-3/4-dependent SCC came from analysis of pachytene nuclei in rec-8 single and coh-4 coh-3 double mutants that were also defective in the ATM/ATR signaling cascade . If DSBs are important for triggering COH-3/4 to be cohesive , the rec-8 mutants should exhibit greater sister separation than the coh-4 coh-3 mutants when this signaling cascade is defective . Indeed , while 0% of atm-1; coh-4 coh-3; atl-1 mutants showed sister separation in pachytene nuclei , 21% of atm-1; rec-8; atl-1 mutants exhibited separation ( p < 0 . 001 ) ( Figure 5B and Figure 5—figure supplements 2 and 3 ) . These results indicate that DSBs trigger COH-3/4 to become cohesive . The participation of AE and SC components in tethering sisters during pachytene in spo-11 rec-8 mutants made it difficult to discover this role . Although CR proteins are unlikely to tether sisters during wild-type meiosis , AE-dependent linkages between sisters may be a normal feature of meiosis ( See Discussion ) . As expected , we found chromosome fragments and fusions in 100% of nuclei in animals with severely compromised REC-8- and COH-3/4-dependent SCC ( e . g . him-3 rec-8 and atm-1; rec-8; atl-1 mutants ) as a consequence of defective DSB repair ( Figure 3F; Figure 4D ) . Unexpectedly , we found very few RAD-51 foci ( Figure 4B , C ) and virtually no chromosome fragments or fusions ( Figure 3D and Figure 4D ) in all 30 of the gonads examined from meiotic kleisin triple mutants , suggesting a nearly complete absence of early DSB repair intermediates and little or no defective DSB repair . We conclude that cohesin , but not SCC per se , is necessary for the timely formation of RAD-51 foci , and likely DSBs . Dependence of DSB formation on REC-8 and COH-3/4 could reflect either a direct requirement for cohesin in the formation of SPO-11-dependent DSBs or instead the known requirement for REC-8 and COH-3/4 cohesin in loading HTP-3 ( Severson et al . , 2009 ) , which is essential for DSB formation ( Goodyer et al . , 2008 ) . In C . elegans , a single , asymmetrically positioned CO forms between each homolog pair . The CO divides the pair into long and short arms ( Barnes et al . , 1995; Albertson et al . , 1997; Nabeshima et al . , 2005 ) ( Figure 6A ) . The holocentric chromosomes of C . elegans lack a localized centromere; however , the long arms share features with centromeres of monocentric chromosomes during meiosis . Co-orientation occurs at the long arms to ensure that the two sister chromatids interact with microtubules from the same spindle pole , and SCC is maintained at the long arms until anaphase II to keep sisters together . Cohesion at the short arms holds homologs together , and SCC release at short arms in anaphase I triggers disjunction of homologous chromosomes . In other words , the CO site , not a centromere , defines a region of each homolog in which co-orientation is implemented and SCC persists until anaphase II ( Reviewed in Schvarzstein et al . , 2010 ) . 10 . 7554/eLife . 03467 . 017Figure 6 . CO recombination triggers removal of REC-8 and COH-3/4 cohesins from reciprocal domains in late prophase/prometaphase of meiosis I . ( A ) In worms , CO position determines where SCC will be removed in anaphase I . A single , asymmetrically positioned CO forms between each homolog pair in pachytene , dividing the homologs into long and short arms . In diplotene , each homolog pair is restructured around the CO to form a cruciform bivalent . At anaphase I , SCC is released at the short arm to allow homologs to separate . SCC persists at the long arm until anaphase II . ( B ) Confocal micrographs showing that REC-8 and COH-3/4 adopt complementary patterns on meiotic chromosomes by metaphase . In pachytene , REC-8 and COH-3/4 overlap with HTP-3 along the entire meiotic axis . In diplotene , HTP-3 and REC-8 persist along the length of the axis , but COH-3/4 staining diminishes at long arms . By diakinesis , COH-3/4 levels are substantially reduced at long arms but not at short arms . In contrast , REC-8 levels usually remain equal at long and short arms until late diakinesis or prometaphase . Diakinesis nuclei shown are from the third oldest oocyte . In prometaphase/metaphase I , REC-8 and COH-3/4 occupy reciprocal domains . REC-8 is reduced or undetectable at short arms , while COH-3/4 is detectable only at short arms . Arrowheads indicate bivalents viewed from the ‘front’ , that is with both long and short arms in the image plane . In these bivalents , HTP-3 staining is cruciform and long and short arms can usually be distinguished by their relative lengths . Pink arrowheads indicate the bivalent shown at higher magnification in the inset . Arrows indicate bivalents viewed from the ‘side’ , that is with short arms perpendicular to the image plane . In these bivalents , HTP-3 staining resembles a ‘figure 8’ , with two loops of uniform staining ( the long arms ) meeting at a region of more intense staining ( the short arms ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 017 During diplotene and diakinesis , a condensin-dependent process restructures the short and long arms of each recombined homolog pair around the CO to form a cruciform ( Figure 6A; Chan et al . , 2004 ) . We found that during this reorganization , COH-3/4 was removed from the long arm and became restricted to the short arm by prometaphase ( Figure 6B ) . The opposite pattern of removal was noted for REC-8 ( e . g . , Figure 3E , F in de Carvalho et al . , 2008; Supplemental Figure 1D in Harper et al . , 2011; Figure 5 in Rogers et al . , 2002 ) . We confirmed that REC-8 is progressively removed from the short arm of diakinesis bivalents and often becomes undetectable by metaphase I ( Figure 6B ) . The redistribution of COH-3/4 precedes that of REC-8 , which is usually not apparent until late diakinesis or prometaphase I ( Figure 6B ) . Thus , the kleisin subunit determines both when and where a cohesin complex will be removed from chromosomes during late prophase and prometaphase . The partitioning of REC-8 and COH-3/4 into reciprocal domains that flank the CO site suggests that CO recombination triggers removal of REC-8 and COH-3/4 cohesins from complementary regions of the bivalent . Indeed , REC-8 and COH-3/4 persist along the entire axis of desynapsing chromosomes in diplotene nuclei of spo-11 mutants ( Figure 7A ) , and both kleisins associate with the midunivalent of the detached homologs in diakinesis nuclei ( Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 03467 . 018Figure 7 . CO recombination triggers separase-independent removal of REC-8 and COH-3/4 from complementary chromosomal territories . ( A–D ) Projected images of entire nuclei in pachytene and diplotene ( A ) or diakinesis ( B–D ) . ( A ) In spo-11 mutants , CO recombination fails and REC-8 and COH-3/4 are present along the length of meiotic axes in pachytene and diplotene nuclei . In diakinesis , both kleisins are detected at the mid-univalent ( Figure 7—figure supplement 1 ) . ( B ) Depletion of the separase ortholog sep-1 does not impede removal of REC-8 or COH-3/4 . ( C ) AIR-2 associates with short arms of diakinesis bivalents . ( D ) In air-2 ( RNAi ) animals , REC-8 persists on both long and short arms of prometaphase bivalents , indicating that AIR-2 is required for removal of REC-8 from short arms . COH-3/4 still persists at the midbivalent , indicating that AIR-2 is not required for removal of COH-3/4 from long arms or maintenance of COH-3/4 at short arms . ( E ) A model demonstrating how establishing reciprocal domains of REC-8 cohesin and COH-3/4 cohesin could facilitate sequential separation of homologs and then sisters . REC-8 cohesin ( red ) can co-orient sister chromatids and mediate SCC that persists until anaphase II . COH-3/4 cohesin ( green ) cannot . Restricting REC-8 cohesin to long arms would ensure that co-orientation and persistent SCC occur only in this domain . Co-orientation of long arms would ensure that sister chromatids are pulled to the same spindle pole in anaphase I following proteolytic cleavage of COH-3/4 . Proteolysis of REC-8 in meiosis II would allow sisters to separate . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 01810 . 7554/eLife . 03467 . 019Figure 7—figure supplement 1 . SPO-11-dependent CO recombination triggers the removal of REC-8 and COH-3/4 from complementary domains . Projected confocal micrographs show the distribution of COH-3/4 and REC-8 on diakinesis bivalents in wild-type worms and univalents of spo-11 mutants . ( A ) COH-3/4 become substantially reduced or undetectable at the long arms of wild-type bivalents by diakinesis , but persists at high levels at short arms . In contrast , the AE protein HTP-3 is present at uniform levels at both long and short arms . ( B ) In contrast , REC-8 becomes reduced at short arms but persists at high levels at long arms . CO recombination fails in spo-11 mutants , and homologs remain apart as discrete univalents . COH-3/4 ( A ) and REC-8 ( B ) both associate with univalents in diakinesis nuclei of spo-11 mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 03467 . 019 During mitosis in many organisms , cohesin complexes are removed from chromosomes by two pathways ( Nasmyth and Haering , 2009 ) . In prophase , cohesin complexes are removed from chromosome arms via the prophase pathway , a non-proteolytic pathway that involves phosphorylation by Polo and Aurora B kinases . At anaphase onset , cohesin is removed from centromeres by proteolysis of the kleisin by the cysteine protease separase ( Buonomo et al . , 2000; Uhlmann et al . , 2000 ) . The timing of removal of REC-8 and COH-3/4 suggested that cohesin complexes are triggered to dissociate from chromosomes in prophase , independently of separase . Consistent with this , removal of REC-8 from the short arm did not require the worm separase homolog SEP-1 ( Figure 7B ) but did require the Aurora B kinase AIR-2 , which accumulates at the short arm in late diakinesis , prior to the reduction of REC-8 in this region ( Figure 7C , D; Rogers et al . , 2002 ) . These data are consistent with removal of REC-8 cohesin from the short arm by a meiotic prophase pathway . Dissociation of COH-3/4 from the long arm was also independent of SEP-1 . However , neither removal of COH-3/4 from the long arm nor maintenance of COH-3/4 at the short arm required AIR-2 ( Figure 7D ) . These data are consistent with our finding that COH-3/4 begins to disappear from the long arm in diplotene , well before the accumulation of AIR-2 at the short arm . Thus , the kleisin determines not only when and where , but also how a cohesin complex is removed from chromosomes . Kleisin-specified cohesin removal in late prophase could promote the stepwise separation of homologs and sisters , a model consistent with the mutant phenotypes of rec-8 single and coh-4 coh-3 double mutants .
We showed that multiple , functionally distinct cohesin complexes mediate sister chromatid cohesion during meiosis . The cohesins differ in their α-kleisin subunit , and the kleisin influences nearly all aspects of meiotic cohesin function: the mechanisms for loading cohesins onto chromosomes , for triggering DNA-bound cohesins to become cohesive , and for releasing cohesins in a temporal- and location-specific manner during prophase I . Our findings establish a new model for cohesin function in meiosis: the choreographed actions of multiple cohesins , endowed with specialized functions by their kleisins , underlie the stepwise separation of homologs and sisters essential for the reduction of genome copy number . Our work demonstrated a critical and unexpected role for DSBs in triggering meiotic SCC . The importance of DSBs in inducing SCC was shown previously only in proliferating yeast cells that suffered DNA damage during G2/M . Because Rec8 cohesin failed to establish DSB-induced cohesion when replacing Scc1 in mitotic cells , the DSBs used to initiate CO recombination during meiosis were presumed not to trigger SCC ( Heidinger-Pauli et al . , 2008 ) . To the contrary , we found meiotic DSBs to be the essential trigger that induces COH-3/4 cohesin to tether sisters . Mutants deficient in SPO-11 failed to form DSBs and failed to generate COH-3/4–dependent cohesion unless subjected to ionizing radiation . Thus , DSB-induced SCC is an essential , conserved process that functions not only in proliferating yeast cells suffering DNA damage , but also in nematode germ cells undergoing normal gametogenesis . Unexpectedly , although SCC was severely disrupted in diakinesis nuclei of spo-11 rec-8 mutants , indicating that DSBs are essential for tethering sister chromatids , the sisters were together in most pachytene nuclei of spo-11 rec-8 mutants , raising the question of whether DSBs are essential for establishing COH-3/4-mediated cohesion . Several lines of evidence demonstrate a requirement for DSBs in triggering COH-3/4-mediated SCC . First , COH-3/4 was not detected in PCNA-positive nuclei , suggesting that a replication-independent mechanism initiates the process by which COH-3/4 cohesin becomes cohesive . DSBs are the only known trigger for establishing SCC outside of S phase . Second , the AE protein HIM-3 can tether sisters in pachytene nuclei independently of REC-8 in both rec-8 single and spo-11 rec-8 double mutants , thereby obscuring the role of DSBs in establishing COH-3/4-mediated SCC . Third , synaptonemal complexes form between sisters in rec-8 and spo-11 rec-8 animals . Disrupting DSB formation in rec-8 mutants lacking the SC central region protein SYP-1 ( i . e . , spo-11 rec-8; syp-1 triple mutants ) increased the frequency of sister separation in pachytene nuclei relative to that of rec-8; syp-1 double mutants proficient in DSB formation . This result strengthens the view that DSBs are required to establish SCC , but the requirement is partially masked by sister linkages mediated by SC proteins . Finally , sister separation is greater in pachytene nuclei of rec-8 mutants with a defective ATM/ATR signaling pathway than in rec-8 mutants with a functional pathway . In Saccharomyces cerevisiae , DNA damage in G2/M activates ATM and Chk1 kinases , leading to Chk1-dependent phosphorylation of Scc1 on serine 83 . Unphosphorylatable S83A mutants fail to establish DI-SCC , while phosphomimetic S83D mutants establish DI-SCC independently of Chk1 , showing the importance of S83 phosphorylation in inducing cohesion in response to DNA damage ( Heidinger-Pauli et al . , 2009 , 2008 ) . Yeast Rec8 lacks an equivalent Chk1 consensus site , explaining its failure to establish DI-SCC . SPO-11-dependent DSBs appear to activate a similar signaling cascade to trigger meiotic DI-SCC through COH-3/4 cohesin in C . elegans . This cascade requires ATM , ATR , and CHK-2 and likely culminates in kleisin phosphorylation . Because rec-8; chk-2 mutants exhibited reduced COH-3/4 staining and complete failure to tether sisters , CHK-2 may function in regulating cohesin loading as well as in triggering sister linkages . To assess whether the loading and SCC establishment defects are separable , the predicted CHK kinase phosphosites in COH-3/4 are being changed to alanine . S81 and/or T82 in COH-3/4 are the most plausible sequence homologs of S83 in yeast Scc1 , but COH-3 and COH-4 also have four sites with a perfect match to the Chk kinase consensus sequence that might serve as targets of CHK-2 phosphorylation required for SCC establishment by COH-3/4 cohesin . The crucial role of DSBs in establishing COH-3/4-dependent cohesion allowed us to answer a longstanding question: does SC assembly require only the chromosomal binding of cohesin or also the conversion of bound cohesin into a cohesive state ? The strikingly different SC assembly defects in mutants that retain COH-3/4 binding but lack both REC-8- and COH-3/4-mediated cohesion ( spo-11 rec-8 ) compared to mutants that lack COH-3/4 binding as well as REC-8- and COH-3/4-dependent cohesion ( rec-8; coh-4 coh-3 ) revealed the answer . Robust SC assembled along chromosomes of spo-11 rec-8 mutants , between sister chromatids or non-homologous chromosomes , but SC assembly failed on chromosomes of kleisin triple mutants . Thus , SC assembly requires COH-3/4 binding but not its conversion to a cohesive state , revealing that cohesin functions in cohesion-independent processes . These data also demonstrated that SC assembly is more sensitive to disruption of COH-3/4 cohesin than REC-8 cohesin , since we observed severe SC structural defects in coh-4 coh-3 double mutants but not in rec-8 single mutants ( See also Severson et al . , 2009 ) . Thus , the kleisin determines the role of cohesins in synaptonemal complex ( SC ) assembly . It was previously believed that meiosis-specific cohesin complexes were both necessary and sufficient to establish and maintain the linkages that tether sisters during gametogenesis . Eliminating REC-8- and COH-3/4-dependent SCC enabled us to discover the participation of mitotic kleisins and axial proteins in chromosome tethering during meiosis . In diakinesis nuclei , REC-8 and COH-3/4 cohesins are indispensable for holding sister chromatids together . In contrast , although REC-8 and COH-3/4 are critical for establishing meiotic SCC , weak linkages can occur between sisters in pachytene nuclei of mutants lacking the three meiotic kleisins . These linkages are mediated by SCC-1 cohesin , previously believed to function only during mitosis . SCC-1 associates with meiotic chromosomal axes in transition-zone and pachytene nuclei of wild-type animals , suggesting that SCC-1 has the capacity to mediate SCC during wild-type meiosis . However , the role of SCC-1 during wild-type meiosis is not easy to determine . No obvious meiotic phenotypes were detected following depletion of SCC-1 in wild-type animals ( data not shown ) ; but this analysis was limited by the need to use partial loss-of-function conditions due to the essential role of SCC-1 in mitotic proliferation of germline precursors . Mitotic kleisins also associate with meiotic chromosomes of budding yeast and mice ( Klein et al . , 1999; Xu et al . , 2004; Ishiguro et al . , 2011; Lee and Hirano , 2011 ) . The Rad54 paralog Tid1 is necessary to remove Scc1-dependent linkages from yeast meiotic chromosomes to facilitate normal chromosome segregation ( Kateneva et al . , 2005 ) , indicating that Scc1 cohesin can tether sisters during meiosis . Meiotic roles for mammalian Rad21/Scc1 have not yet been defined . Nevertheless , the association of Scc1 orthologs with meiotic chromosomes of widely diverged species suggests they play important roles in gametogenesis . Although SCC-1 can tether sisters in pachytene nuclei of rec-8; coh-4 coh-3 mutants , SC proteins do not associate with meiotic chromosomes . Thus , unlike REC-8 and COH-3/4 , SCC-1 appears unable to promote even partial SC assembly . These data demonstrate that establishment of cohesion between sisters is insufficient to promote the formation of SC , consistent with our finding that SC assembly requires chromosomally bound COH-3/4 cohesin , but not conversion of COH-3/4 cohesin into a cohesive state . We also found that axial element ( AE ) proteins can mediate cohesin-independent linkages between sister chromatids when rec-8- and DSB-dependent cohesion fail to occur . That is , HIM-3 tethers sisters independently of cohesin in spo-11 rec-8 mutants , which have chromosome-bound COH-3/4 that is not cohesive . The HIM-3-dependent linkages are not mediated by an SC-like structure , given that disrupting the CR protein SYP-1 only partially weakened SCC in spo-11 rec-8 double mutants , while mutations in him-3 severely abrogated SCC . HIM-3-dependent linkages between sisters may not be essential in animals with wild-type REC-8 , because REC-8-dependent SCC established during premeiotic DNA replication is likely sufficient to tether sisters until anaphase II ( Severson et al . , 2009 ) . However , HIM-3-dependent linkages may prevent newly replicated sister chromatids from drifting apart in rec-8 mutants before DSBs trigger COH-3/4 to become cohesive , thus explaining the severe SCC defects observed in him-3 rec-8 double mutants with wild-type spo-11 . Single molecule experiments have shown that purified Hop1 , a HORMA domain protein that is the yeast ortholog of HIM-3 , HTP-3 , and HTP-1/2 , can mediate trans interactions that bridge linear double-stranded DNA molecules and promote their restructuring and compaction ( Khan et al . , 2012 ) . These interactions were proposed to facilitate pairing or synapsis of homologous chromosomes , but our data suggest a different role: HORMA-domain AE proteins such as HIM-3 may directly tether sister chromatids independently of cohesin in wild-type animals , as we have shown for cohesion-defective mutants . The mutant analysis described here reveals the unexpected molecular complexity of meiotic SCC and suggests that the linkages that tether sisters in the germline are not formed synchronously during DNA replication , as occurs during mitosis , but rather through the sequential action of proteins that bind to chromosomes at temporally distinct times and generate cohesion by different mechanisms . Our data support the following model . REC-8 cohesin mediates the first wave of SCC establishment . SCC-1 may function together with REC-8 during that process . Both REC-8 and SCC-1 are present in all premeiotic nuclei and likely bind to chromatids prior to premeiotic S phase and become cohesive during DNA replication , a pattern similar to that described for cohesin complexes in mitotically proliferating cells . HIM-3 may mediate a second wave of cohesion , but through a mechanism distinct from that of cohesin . HIM-3 is expressed and loads onto chromosomes in leptotene , at the onset of meiosis ( Zetka et al . , 1999; MacQueen et al . , 2002 ) , and our data suggest that HIM-3 tethers sisters independently of REC-8 and COH-3/4 cohesin . More definitive evidence for the involvement of HIM-3 in mediating cohesion will require a separation-of-function allele that disrupts SCC but not SC assembly or crossover recombination . A final wave of SCC establishment is mediated by COH-3/4 cohesin , which loads onto chromosomes at the onset of meiosis but is not triggered to become cohesive until the formation of SPO-11-dependent DSBs during leptotene and/or zygotene . Sequential mutational analysis in kleisin-defective animals enabled us to strip away the layers of chromosome tethering that are mediated by mitotic kleisin SCC-1 , axial protein HIM-3 , and SC to uncover the central roles of DSBs , REC-8 , and COH-3/4 in establishing SCC . Our data also suggest that the removal of linkages between sisters occurs in stages . In diplotene ( our unpublished data ) and diakinesis nuclei , sisters are held together by REC-8 and COH-3/4 cohesin alone , indicating that the HIM-3 and SCC-1-dependent linkages we demonstrated in pachytene are transient and are removed during late pachytene or early diplotene . In contrast , SCC mediated by COH-3/4 and REC-8 persists until anaphase I and II , respectively ( see below ) . Prior to this study and our earlier demonstration that COH-3/4 and REC-8 mediate meiotic SCC ( Severson et al . , 2009 ) , the triggering and release of meiotic cohesion were thought to depend entirely on Rec8 cohesin . All models of eukaryotic meiotic chromosome segregation asserted that the stepwise cleavage of Rec8 by separase initiated the successive separation of homologs and sisters , and thus the production of haploid gametes . In organisms with monocentric chromosomes , cleavage of Rec8 along chromosome arms was proposed to trigger homolog separation in meiosis I , while cleavage of Rec8 at centromeres was proposed to allow sister separation in meiosis II ( Klein et al . , 1999; Watanabe and Nurse , 1999; Buonomo et al . , 2000; Watanabe , 2004; Kudo et al . , 2006 ) . In the holocentric nematode C . elegans , where the asymmetric CO site rather than the centromere defines the bivalent short and long arms , cleavage of REC-8 at the short arm was believed to elicit homolog segregation in meiosis I , while cleavage of REC-8 at the bivalent long arm was believed to cause sister separation in meiosis II ( Pasierbek et al . , 2001; Siomos et al . , 2001; Kaitna et al . , 2002; Rogers et al . , 2002 ) . Consistent with this model , prior C . elegans studies identified the Aurora B kinase AIR-2 as a key factor that regulates SCC release , and thus chromosome segregation , during meiosis and mitosis . In meiosis I , AIR-2 accumulates at bivalent short arms and is essential for homolog separation ( Kaitna et al . , 2002; Rogers et al . , 2002 ) . In meiosis II , AIR-2 accumulates between sisters and is required for sister separation . AIR-2 can phosphorylate REC-8 in vitro ( Rogers et al . , 2002 ) . Thus , the distribution of AIR-2 predicts where SCC will be released at anaphase in meiosis I and II . AIR-2 is prevented from accumulating at the long arms during meiosis I by the partially redundant AE proteins HTP-1 and HTP-2 ( HTP-1/2 ) and the novel , Caenorhabditis-specific protein LAB-1 ( de Carvalho et al . , 2008; Kaitna et al . , 2002; Martinez-Perez et al . , 2008; Rogers et al . , 2002 ) . HTP-1/2 and LAB-1 accumulate along the entire length of meiotic chromosomal axes in early pachytene , but CO recombination triggers their removal from short arms and thereby allows the accumulation of AIR-2 at the short arms . HTP-1/2 and LAB-1 are undetectable on chromosomes after anaphase I; consequently , AIR-2 accumulates between sisters in meiosis II . In htp-1 htp-2 and lab-1 mutants , AIR-2 associates with both long and short arms . As a consequence , sister chromatids separate prematurely during anaphase I . The correlation between the presence of AIR-2 and the release of SCC during meiosis of wild-type animals and htp-1 htp-2 and lab-1 mutants , together with the finding that AIR-2 can phosphorylate REC-8 in vitro , led to the model that AIR-2 induces the stepwise separation of homologs and sisters by phosphorylating REC-8 , first at the short arm to trigger separase-dependent cleavage in anaphase I , then at the long arm to trigger separase-mediated cleavage in anaphase II ( Kaitna et al . , 2002; Rogers et al . , 2002 ) . However , it has never been determined whether AIR-2-dependent phosphorylation is required for separase to cleave REC-8 or whether cleavage of REC-8 by separase is required for homolog separation at anaphase I . Thus , this model has never been put to a rigorous test . Together , our discovery of COH-3/4 in C . elegans and our demonstration that CO recombination consistently triggers separase-independent removal of REC-8 and COH-3/4 cohesins from reciprocal domains of meiosis I bivalents indicate that prior models are insufficient to explain how linkages are removed between sister chromatids to permit the separation of homologs and then sisters . Any model of C . elegans meiotic chromosome segregation must incorporate not only the removal of REC-8 cohesin by separase , but also the separase-dependent removal of COH-3/4 cohesin and the role of separase-independent cohesin removal that occurs in late prophase . Our finding that different mechanisms trigger separase-independent removal of REC-8 from short arms and COH-3/4 from long arms indicates that AIR-2 is not the sole factor to direct the stepwise separation of homologs and sisters . REC-8 persists at high levels at the long arm but becomes markedly reduced and often undetectable at the short arm ( our work and de Carvalho et al . , 2008; Rogers et al . , 2002; Harper et al . , 2011 ) . Removal of REC-8 from the short arm begins in late diakinesis or prometaphase and requires AIR-2 . In contrast , COH-3/4 cohesin is removed from the long arm of wild-type animals beginning in diplotene , prior to AIR-2 accumulation at the midbivalent and coincident with the removal of SC from the long arm ( Nabeshima et al . , 2005 ) . Both removal of COH-3/4 from the long arm and persistence of COH-3/4 at the short arm are independent of AIR-2 function . The factors that restrict COH-3/4 cohesin to the short arm are not known , but could include HTP-1/2 , LAB-1 , and PP1; however , if these factors regulate the distribution of COH-3/4 cohesin , they must do so in parallel with their regulation of AIR-2 . We propose two models for how separase-independent cohesin removal in prophase could promote the separation of homologs before sisters in C . elegans . In the first model , REC-8 cohesin is eliminated from the short arm by an AIR-2-dependent mechanism that is independent of kleisin proteolysis , allowing homolog disjunction at anaphase I to be triggered by separase-dependent cleavage of COH-3/4 ( Figure 7E ) . The separase-independent partitioning of REC-8 and COH-3/4 cohesins into reciprocal domains could explain why factors like Mei-S332/Shugoshin , which protect centromeric cohesin from separase-mediated cleavage during anaphase I in monocentric organisms , are not required in C . elegans ( de Carvalho et al . , 2008; Severson et al . , 2009 ) . In the second model , selective removal of REC-8 from the short arm does not determine the timing of separation for homologs vs sisters , but rather restricts co-orientation to the long arm . Once homologs have made proper attachments to microtubules and aligned on the metaphase plate , separase-dependent cleavage of COH-3/4 and any REC-8 remaining at the short arm would allow homologs to segregate toward opposite poles . Because REC-8 and COH-3/4 are both required for CO recombination , and hence the formation of bivalents with differentiated long and short arms , a direct test of these models will require versions of REC-8 and COH-3 or COH-4 that can be removed from chromosomes after COs have formed . Our work reveals the unexpected degree to which kleisin variants influence virtually all facets of meiotic cohesin function . It establishes a new model for cohesin function during gametogenesis in higher eukaryotes . The orchestrated actions of multiple cohesins , endowed with specialized functions by their kleisins , reduce genome copy number to produce haploid gametes . The kleisin determines the mechanisms by which cohesin loads onto meiotic chromosomes , establishes SCC , and is removed from chromosomes prior to proteolytic cleavage by separase at anaphase I . Plants and mammals require similar sets of meiotic kleisins as those in C . elegans , demonstrating the widely conserved involvement of multiple kleisins in gametogenesis and highlighting the importance of understanding the mechanisms by which kleisins influence cohesin function . Our work represents a major stride toward achieving that goal . The phenotypes of kleisin-deficient mice and the published localization patterns of mammalian kleisins suggest that the models established for C . elegans will apply to mammals . For example , as in C . elegans ( Figure 5A and Severson et al . , 2009 ) , SC proteins in mice assemble between sister chromatids in Rec8 single mutants but not in Rad21L Rec8 double kleisin mutants , demonstrating the involvement of multiple kleisins in SC assembly ( Llano et al . , 2012; Ishiguro et al . , 2014 ) . Moreover , the idea that factors other than known meiosis-specific cohesin complexes contribute to SCC during spermatogenesis is suggested by the persistence of partial cohesion in meiotic nuclei of mouse Rad21L Rec8 double mutants ( Llano et al . , 2012; Ishiguro et al . , 2014 ) . The mitotic kleisin RAD21 appears to associate with meiotic chromosomes of both wild-type and Rad21L Rec8 mutants , suggesting that ‘mitotic’ cohesin might tether sisters during gametogenesis in mammals , as in worms ( Prieto et al . , 2002; Parra et al . , 2004; Xu et al . , 2004; Ishiguro et al . , 2011; Lee and Hirano , 2011; Llano et al . , 2012 ) . The published data for mice are also consistent with our findings that the kleisin determines the mechanisms of cohesin loading and SCC establishment . High levels of REC8 were detected in premeiotic mouse nuclei , consistent with REC8 cohesin becoming cohesive during DNA replication ( Eijpe et al . , 2003; Ishiguro et al . , 2014 ) . In contrast , RAD21L staining was faint in PCNA-positive premeiotic testicular cells but greatly increased on meiotic chromosomes during leptotene and zygotene ( Herran et al . , 2011; Lee and Hirano , 2011; Ishiguro et al . , 2014 ) , indicating that substantial amounts of RAD21L cohesin load after meiotic entry . Thus , different mechanisms may promote the premeiotic loading of REC8 cohesin and the post-replicative loading of RAD21L cohesin , a model consistent with a role for programmed meiotic DSBs in triggering replication-independent SCC establishment by RAD21L cohesin . Indeed , a Spo11 disruption exacerbated the SCC defects of Rec8 knockout mice ( Ishiguro et al . , 2014 ) . However , the partial sister separation observed in Spo11 Rec8 double mutants could have resulted from defective SC assembly rather than a failure to establish DSB-induced SCC , since DSBs promote formation of SC between homologs of wild-type animals and between sister chromatids of Rec8 mutant mice ( Romanienko and Camerini-Otero , 2000; Xu et al . , 2005; Ishiguro et al . , 2014 ) . On the other hand , while DSBs may be critical for RAD21L cohesin to establish SCC , the sisters may have remained tethered in Spo11 Rec8 mutants by RAD21 cohesin , thereby obscuring the essential role of DI-SCC . Thus , establishing whether DSBs are essential for mammalian meiosis will require an assessment of whether AE and/or SC CR proteins can tether sisters in mice , as they do in worms , and whether RAD21 cohesin contributes to meiotic SCC . Finally , studies hint that the kleisin subunit of mammalian cohesin complexes may determine whether a complex will be removed during late prophase I via a separase-independent mechanism . Although REC8 persists at high levels at centromeres and chromosome arms until anaphase I in mouse spermatocytes and oocytes , RAD21L and RAD21 proteins progressively diminish in abundance during late prophase at chromosome arms of spermatocytes and at arms and centromeres of oocytes . ( Prieto et al . , 2002 , 2004; Lee et al . , 2003; Parra et al . , 2004; Tachibana-Konwalski et al . , 2010; Ishiguro et al . , 2011; Lee and Hirano , 2011 ) . Understanding the mechanisms by which the kleisin subunit influences cohesin function to reduce genome copy number during meiosis of plants and mammals will require a more complete understanding of the factors that mediate cohesion . The rigorous experimental approaches we developed to elucidate the contributions of C . elegans meiotic kleisins can be applied to define the precise contributions of kleisin subunits in these other species .
Worms strains were cultured using standard methods ( Brenner , 1974 ) . N2 Bristol was used as wild-type; other strains used in this study are listed in Supplementary file 1 . Many mutants used in this study produce viable progeny with polyploid genomes; experiments using these alleles were performed on homozygous worms produced by known diploid , heterozygous parents . The template for air-2 dsRNA production was PCR amplified from the cDNA clone yk483g8 with T7 and T7_T3 primers ( Supplementary file 2 ) . The templates for chk-1 and chk-2 dsRNA production were amplified from the Open Biosystems RNAi library ( Fisher Scientific , Pittsburgh , PA ) clones GHR-10020 and GHR-11002 , respectively . Other templates for dsRNA production were amplified from genomic DNA with gene-specific primers that included 5′ T7 sequences ( Supplementary file 2 ) . In all cases , PCR products were gel purified , then reamplified with T7 primers . dsRNA was prepared by in vitro transcription ( Ambion , Austin , TX ) . Young adult hermaphrodites were injected with dsRNA at concentrations of 2 . 5–5 mg/ml , then mated with him-8; mIs10 males at 20°C . Worms and embryos were fixed and stained 72 hr post injection for depletion of AIR-2 , ATL-1 , and SMC-1 , 60 hr post injection for depletion of CHK-1 and CHK-2 , and 48 hr post injection for depletion of SCC-1 . Immunofluorescence analysis was performed as described previously ( Chan et al . , 2004 ) . The following antibodies were used: rabbit anti-AIR-2 ( Schumacher et al . , 1998 ) , rat anti-SMC-3 ( Chan et al . , 2003 ) , rabbit anti-REC-8 and SCC-1 ( Novus Biologicals , Littleton , CO ) , and mouse anti-REC-8 ( CIM , Arizona State University ) . Anti-COH-3/4 antibodies were raised in rabbits ( Covance , Princeton , NJ ) immunized with a mixture of the peptides CGGNIDLLSTDDSEDIDDLAMADF and CGGNIDLLSTDDIEDIDDLAMADF ( synthesized by D . King of the HHMI Mass Spectrometry Facility , University of California , Berkeley , CA ) . Peptides were coupled to Sulfolink ( Thermo Fisher Scientific , Rockford , IL ) for affinity purification . The staining pattern with this antibody was identical to that obtained with a commercial COH-3 antibody ( Novus Biologicals , Littleton , CO ) except that the nucleoplasmic background was much lower with our antibody . LacI-His6-GFP ( Darby and Hine , 2005 ) was expressed in BL21 Escherichia coli and purified on TALON resin ( Clontech , Mountain View , CA ) . Animals heterozygous for the integrated lacO array syIs44 were generated by crossing syIs44 males to hermaphrodites that lacked the array . This crossing scheme allows self progeny of the hermaphrodite to be identified by the absence of GFP::LacI staining , ensuring that all examples of two GFP foci in the same nucleus resulted from defective SCC establishment in syIs44 heterozygotes rather than a recombination defect in syIs44 homozygotes . For quantification of distances between sister chromatids , Z-stacks of 1024 × 1024 pixel , unbinned images were acquired at 0 . 2 µm axial spacing on a Deltavision microscope ( Applied Precision , Issaquah , WA ) equipped with a 100x/1 . 4 NA objective lens . The X , Y , and Z positions of GFP foci were marked by hand in ImageJ and the distance between spots calculated by the three dimensional generalization of the Pythagorean theorem . Distances were measured in the −1 and −2 oocyte and in mid and late pachytene nuclei . For the analyses presented here , data were combined into ‘diakinesis’ and ‘pachytene’ datasets . In synapsis-defective mutants , distances were measured in nuclei that occupied similar positions in the gonad as mid and late pachytene nuclei in wild-type animals . To determine the effects of γ-irradiation on meiotic SCC , L4 hermaphrodites were exposed to 12 Gy from a sealed 137Cs source as described previously ( Mets and Meyer , 2009 ) , then fixed and stained 18 hr after exposure . | Most plant and animal cells have a pair of each chromosome: one copy is inherited from the father , the other from the mother . When a cell divides , each daughter cell must receive a copy of all of the original cell's genetic information . To this end , the chromosomes are replicated to form so-called ‘sister chromatids’ , which are then segregated equally between the two daughter cells . In contrast , sex cells such as eggs and sperm ( also called gametes ) have a single copy of each chromosome . When an egg and a sperm fuse to form a single cell ( called a zygote ) , the zygote ends up with a full set of chromosomes . Gametes are formed by two successive rounds of cell division that occur after the chromosomes are replicated . The first round separates the pairs of chromosomes , and the second separates the sister chromatids to produce the gametes , each of which has half the original amount of genetic information . If something goes awry in the production of gametes , a zygote can end up with the wrong number of chromosomes . Almost one-third of human zygotes inherit an aberrant complement of chromosomes , and many of these zygotes either fail to survive or develop into offspring with birth defects and developmental disorders . To ensure that gametes receive the correct number of chromosomes , the sister chromatids remain bound together by a ring-shaped protein complex during the first cell division . Previous studies on how this protein complex—called cohesin—tethers the sister chromatids together were conducted on yeast and mammalian cells . Now , Severson and Meyer show that , in a microscopic worm called Caenorhabditis elegans , cohesin functions differently from how it functions in the simpler yeast cells . Severson and Meyer found that rather than using a single cohesin complex like in yeast , the worms use multiple cohesin complexes that have different versions of one key protein subunit . Changing this single subunit has a major impact on cohesin's function . Consequently , each complex plays a specific role in tethering and then releasing sister chromatids . One of the cohesin complexes is triggered to tether the sister chromatids when the chromosomes replicate . Unexpectedly , another complex only tethers the sisters once breaks occur in the DNA . These breaks allow sister chromatids that are produced from maternally- and paternally-derived chromosomes to cross over and swap genetic material—which increases the genetic diversity of any future offspring . After these genetic swaps occur , the cohesin complexes are then selectively removed by different mechanisms , first to release the pairs of chromosomes and then the sister chromatids . The findings of Severson and Meyer establish a new model for the mechanisms of chromosome segregation during gamete production . Further studies are now needed to determine the roles and regulation of these protein complexes in other species—including plants and mammals , which use similar cohesin complexes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"cell",
"biology"
] | 2014 | Divergent kleisin subunits of cohesin specify mechanisms to tether and release meiotic chromosomes |
Rift Valley fever phlebovirus ( RVFV ) is a clinically and economically important pathogen increasingly likely to cause widespread epidemics . RVFV virulence depends on the interferon antagonist non-structural protein ( NSs ) , which remains poorly characterized . We identified a stable core domain of RVFV NSs ( residues 83–248 ) , and solved its crystal structure , a novel all-helical fold organized into highly ordered fibrils . A hallmark of RVFV pathology is NSs filament formation in infected cell nuclei . Recombinant virus encoding the NSs core domain induced intranuclear filaments , suggesting it contains all essential determinants for nuclear translocation and filament formation . Mutations of key crystal fibril interface residues in viruses encoding full-length NSs completely abrogated intranuclear filament formation in infected cells . We propose the fibrillar arrangement of the NSs core domain in crystals reveals the molecular basis of assembly of this key virulence factor in cell nuclei . Our findings have important implications for fundamental understanding of RVFV virulence .
Rift Valley fever phlebovirus ( RVFV ) , of the genus Phlebovirus , is one of the most clinically significant members of the Phenuiviridae family , of the Bunyavirales order ( Elliott and Brennan , 2014; Plyusnin et al . , 2012; Adams et al . , 2017 ) . RVFV is an arbovirus spread by many mosquito vector species as well as by exposure to infected tissues . It causes recurring epidemics in livestock and humans ( Ikegami and Makino , 2011 ) . The most prominent effect of RVFV infection in ruminants is a high rate of abortions . In humans infections are usually self-limiting , but can develop into hepatitis , retinitis , encephalitis and hemorrhagic fever with fatality rates of 0 . 5–2% ( Pepin et al . , 2010 ) . Although originally endemic to sub-Saharan Africa , RVFV has recently appeared in Madagascar , the Comoros and the Arabian Peninsula ( Balkhy and Memish , 2003 ) . Increasing spread of competent mosquito vector species due to climate change could facilitate emergence of this virus in new ecosystems , including Europe and the United States ( Chevalier , 2013; Elliott , 2009; Golnar et al . , 2014; Rolin et al . , 2013 ) . In 2017 the World Health Organization ranked RVFV among the ten most dangerous pathogens most likely to cause wide epidemics in the near future , requiring urgent attention ( http://www . who . int/blueprint/priority-diseases/en/ ) . While an animal vaccine for RVFV exists , there is no treatment available for human use ( Boshra et al . , 2011; Lihoradova and Ikegami , 2014 ) . RVFV is a zoonotic arbovirus evolved to evade the immune response of mammals and insects . Innate immune antagonism of RVFV is primarily mediated by its main virulence factor , the 30 kDa non-structural protein ( NSs ) , which is transcribed and translated from a subgenomic mRNA derived from the antigenomic S segment early during infection . NSs impedes interferon production through three known mechanisms ( Billecocq et al . , 2004; Bouloy et al . , 2001 ) . In infected host cells , NSs is transported into the nucleus where it interferes with the assembly of the RNA polymerase II preinitiation complex transcription factor II H ( TFIIH ) by binding to the p44 subunit ( Le May et al . , 2004 ) and directing the p62 subunit for degradation ( Kainulainen et al . , 2014; Kalveram et al . , 2011 ) , thereby halting global cellular transcription . NSs also prevents activation of the interferon-β promoter specifically by binding to factor SAP30 , a member of a multisubunit histone deacetylase transcriptional repression complex that regulates interferon-β expression ( Le May et al . , 2008 ) . Furthermore , NSs blocks host recognition of viral dsRNA by targeting the RNA-dependent protein kinase ( PKR ) for degradation ( Ikegami et al . , 2009a , 2009b ) . RVFV with a deletion of NSs cannot establish viremia and is not pathogenic in the mouse model ( Bouloy et al . , 2001 ) . A characteristic feature of RVFV pathogenesis is formation of 0 . 5 μm thick proteinaceous filaments formed by bundles of thin fibrils in the nuclei of infected cells ( Swanepoel and Blackburn , 1977 ) . The filaments are composed primarily of NSs protein ( Struthers and Swanepoel , 1982; Yadani et al . , 1999 ) but co-localize with p44 , XPB , SAP30 and YY1 , as well ( Le May et al . , 2004 , 2008 ) . It is not clear if filament formation is required for NSs function , that is if any interaction sites are present on the monomeric form of the protein , or are only present in the multimerized , fibrillar NSs . In the absence of structural information it is difficult to dissect NSs function from filament formation . Potential functions of nuclear NSs filaments include sequestering NSs binding partners , causing cell cycle arrest and defects in chromosome cohesion and segregation ( Mansuroglu et al . , 2010 ) . This could contribute to the high rate of fetal deformities and abortions observed in RVFV-infected ruminants ( Baer et al . , 2012; Mansuroglu et al . , 2010 ) . NSs proteins are encoded in the S segment of the tripartite negative sense single strand RNA genome of most members of the Phlebovirus genus , and other families of the Bunyavirales order such as the Peribunyaviridae , Nairoviridae , and some members of the Hantaviridae ( Ly and Ikegami , 2016 ) . They constitute notoriously difficult targets for structural characterization due to an inherent tendency to multimerize , as well as the putative presence of substantial intrinsically unfolded regions . While high-resolution structures of glycoproteins ( Dessau and Modis , 2013 ) and nucleoproteins ( Ferron et al . , 2011; Raymond et al . , 2010 ) from RVFV and several other bunyaviruses have been determined , no structural information is currently available for any NSs protein , significantly limiting the current understanding of key bunyaviral virulence mechanisms . Structure and function can also not be predicted based on sequence , as NSs proteins are ‘ORFans’ , with very little sequence similarity in members of the Bunyavirales order and the Phlebovirus genus . Despite large sequence diversity , NSs proteins examined thus far have been found to play roles in the suppression of the host innate immune response . However , this and the diverse mechanisms of NSs function have been studied only for a few viruses . Here we describe the first high-resolution structure of a bunyaviral NSs protein . A truncated stable , soluble core domain spanning residues 83–248 of RVFV NSs ( NSs-ΔNΔC ) was identified using NMR spectroscopy . The corresponding crystal structure represents a novel all-helical fold . Notably , NSs molecules packed in crystals in double-helical fibrils stabilized by multiple interfaces . Dimensions of these fibrils are in good agreement with fibrils constituting nuclear NSs filaments in RVFV-infected cells . Cells infected with recombinant RVFV encoding NSs-ΔNΔC produced nuclear filaments indistinguishable from filaments observed in cells infected with parental RVFV . Furthermore , recombinant viruses encoding full-length NSs with mutations predicted to destabilize significant fibril interfaces , as suggested by crystal packing , did not induce nuclear filaments . Combination of structural biology with cell infection assays using reverse genetics to tailor the main RVFV virulence factor identified a RVFV NSs core domain sufficient for filament formation in vivo , and revealed the structural basis for NSs nuclear filament formation .
Full-length NSs protein ( residues 1–265 ) was expressed as a cleavable MBP fusion yielding only a small amount ( about 0 . 1 mg per litre of culture ) of soluble protein after removal of MBP . Circular dichroism spectroscopy showed secondary and tertiary structure in mature NSs ( Figure 1—figure supplement 1 ) , however the protein formed large soluble aggregates , eluting in the void volume of size exclusion chromatography ( mass over 200 kDa ) ( Figure 1A ) . This construct was therefore redesigned in order to obtain a stable , non-aggregating and soluble fragment of NSs . Secondary structure predictions suggested an N-terminal region with a propensity to form β sheet structure ( residues 1–70 ) , connected by an unstructured linker to a central α-helical domain ( 85-230 ) . The C-terminal region was predicted to be natively unfolded ( 231-265 ) . Deletion of the N-terminal 82 amino acids ( resulting in truncated NSs hereafter referred to as NSs-ΔN , residues 83–265 ) significantly improved expression yields , solubility and stability . In solution , the NSs-ΔN protein existed in equilibrium between monomeric and multimeric forms ( Figure 1A ) . Near-UV CD spectra indicated a helical protein , with characteristic minima at 208 and 220 nm ( Figure 1—figure supplement 1 ) . NSs-ΔN was highly soluble ( up to 30 mg/mL ) in near-physiological buffer conditions . A 2D 1H-15N heteronuclear single-quantum coherence ( HSQC ) NMR spectrum showed well-dispersed crosspeaks , indicative of folded , monomeric protein ( Figure 1B ) . The central region of the spectrum , spanning chemical shifts of 7 . 5 to 8 . 5 ppm in the 1H dimension , was dominated by 17 intense and narrow backbone amide signals , indicating significant flexibility of corresponding residues . Based on secondary structure and disorder predictions , we concluded the intrinsically unfolded residues were present at the C-terminus of NSs-ΔN . Subsequent removal of the C-terminal 17 amino acids ( resulting in doubly truncated NSs hereafter referred to as NSs-ΔNΔC , residues 83–248 ) yielded protein present as a monomer in solution ( Figure 1A ) . Its 1H-15N HSQC spectrum was devoid of all but two intense crosspeaks observed for NSs-ΔN ( Figure 1B ) . All other signals overlapped well with crosspeaks in the NSs-ΔN spectrum , indicating absence of residues 249–265 does not influence the structure of NSs-ΔNΔC . The NSs-ΔNΔC protein was crystallized at conditions optimized following routine sparse-matrix crystallization screens . The crystals were found to be extremely fragile , and diffracted synchrotron X-rays to 2 . 2 Å resolution . Crystals belonged to space group P6422 . Calculation of the Matthews coefficient indicated a high solvent content of 76% , assuming two molecules in the asymmetric unit . Due to the absence of any molecular replacement models , experimental phasing had to be undertaken . Because of the higher-than-average sulfur content of NSs-ΔNΔC ( 12 sulfurs per chain of 175 residues ) and the highly symmetrical P6422 space group of the crystals , S-SAD phasing was pursued . The structure was successfully phased from the anomalous sulfur signal in a single native crystal and refined to high quality ( Table 1 ) . The NSs-ΔNΔC monomer , molecular mass 19 , 029 Da , measures approximately 30 × 30 × 50 Å . It is composed of eight α-helices with helices α1-α6 and α8 arranged around a long central helix α7 . Two short 310 helices are also present ( Figure 2 ) . Residues 234–244 are devoid of secondary structure and traverse approximately 30 Å of the NSs surface , packed against helices α3 and α7 . Two NSs-ΔNΔC monomers constitute the asymmetric unit , with an interface formed by their α8 helices ( Figure 2—figure supplement 1 ) . There was no visible electron density for the first N-terminal and the five C-terminal amino acids; these residues are unaccounted for in the crystal structure . Structural similarity searches with PDBeFold ( Krissinel and Henrick , 2004 ) and DALI ( RRID:SCR_013433 ) ( Holm and Rosenström , 2010 ) did not identify significant structural homologues . We therefore conclude the NSs-ΔNΔC structure represents a novel fold . In the crystal lattice NSs-ΔNΔC molecules were found to pack into a highly organized network of two fibril-like assemblies , referred to here as F1 and F2 ( Figure 3 ) , creating large solvent channels . The F1 and F2 crisscross pattern is produced by F1 fibrils running in parallel to each other and the longest unit cell edge , while F2 fibrils are oriented at an angle of 60° with respect to each other , perpendicular to F1 ( Figure 3—figure supplement 1 ) . F1 and F2 fibrils themselves are best described as being composed of repeating NSs tetramer units . Four different , partially overlapping ( or alternative ) tetramers T1 , T2 , T3 and T4 can be distinguished that differ in stability and interface areas , according to PISA interface analyses ( Krissinel and Henrick , 2007 ) , with T1 being the most stable tetramer ( Table 2 ) . F1 is an assembly of T1 tetramers stacked at 60° angles , while F2 is composed of alternating T1 and T2 tetramers arranged at 90° angles with respect to each other . The T1 interface that lies parallel to the F1 fibril is largely determined by hydrophobic packing of the α6 helices . The T1 interface perpendicular to the F1 fibril axis is defined by an extensive network of hydrogen bonds and salt bridges between the basic face of α1 ( Lys84 , Gln87 , Arg88 ) and the acidic face of α8 ( Glu220 , Glu221 , Ser228 ) , leading to a large buried interface area of 4 × 852 Å2 ( Figure 4 ) . The superficially very similar tetramer T2 in F2 fibrils is less stable . Here the molecules are rotated by about 8° with respect to the orientation of the equivalent monomers in T1 , and form a twisted tetramer ( Figure 4—figure supplement 1 ) . The register of the α6-α6 interface is shifted , decreasing the buried surface area by 169 Å2 compared to the same interface in T1 . The total buried surface area is nearly 1000 Å2 ( 17% ) smaller in T2 than in T1 . The B factors for T2 are also notably higher than for T1 , suggesting that interfaces in T1 , and therefore F1 , possess a higher degree of conformational order , contributing to a higher stability of this assembly ( Figure 4—figure supplements 1 , 2 ) . B factors have been used to distinguish biologically relevant interfaces from crystal packing interfaces , which have been found to be characterized by relatively high B factors ( Liu et al . , 2014 ) . NSs-ΔNΔC T1 tetramers align at 60° angles in the F1 fibril , held by a network of hydrogen bonds clustered around the loop between helices α5-α6 ( Figure 4—figure supplement 1 ) . The 60° turn produces a left-handed double helix with twelve monomers per turn and a 34 . 8 nm pitch ( Figure 3 ) . To gain insight into the architecture of nuclear NSs filaments , we infected Vero-E6 cells with the recombinantly produced parental RVFV strain MP12 as previously described ( Billecocq et al . , 2008; Brennan et al . , 2014 ) and imaged cell nuclei containing filaments with transmission electron microscopy ( TEM ) . Thick filaments of NSs , 0 . 5–1 μm in diameter , started forming about four hours post-infection and averaged to 1–2 filaments per nucleus . TEM reveals a substructure of these filaments ( Figure 5 ) . They appear to be composed of a bundle of tightly packed , extended parallel fibrils with diameters roughly ranging from 8 to 15 nm . Recombinant RVFV encoding NSs-ΔNΔC ( rMP12NSsΔNΔC ) was rescued . Working stock of recombinant virus achieved similar titers ( 1 . 15 × 108 pfu/mL ) to parental MP12 ( 2 . 00 × 108 pfu/mL ) and no difference in plaque morphology was observed . Reverse transcription PCR analysis was performed to confirm the presence of the NSs truncation . Vero-E6 cells were infected with the recombinant rMP12NSsΔNΔC strain . Twenty-four hours after infection NSs-specific immunofluorescence staining showed prominent filaments in nuclei of infected cells ( Figure 6 ) . The morphology of these NSs-ΔNΔC filaments was indistinguishable from structures observed for the parental MP12 strain encoding full-length NSs . Fibrillation was observed only in about 30% of cells infected with rMP12NSsΔNΔC , compared to occurrence of filaments in nearly all cells infected with parental virus . To test if the fibrillar crystal packing observed for NSs-ΔNΔC represents a biologically relevant assembly underpinning intranuclear NSs filament formation , combinations of two key residues in three different crystal interfaces were mutated ( Figure 4 ) . Residues were chosen that were predicted to be critical for assembly of tetramers present in both or either F1 and F2 fibrils , but not essential for monomer stability . Three NSs variants were generated: NSs-ΔNΔC-muT1 ( Arg88Asp , Ser228Ala ) , NSs-ΔNΔC-muT3 ( Lys150Gly , Thr152Gly ) and NSs-ΔNΔC-muT4 ( Ile216Asp , Met219Ala ) . Disruption of the polar T1 interface , perpendicular to F1 , would interrupt both T1 and T2 , and therefore F1 and F2 . Residue Arg88 , forming two salt bridges to Glu220 and Glu221 , was mutated to an aspartate . Ser228 , forming a side-chain hydrogen bond with Gln87 , was replaced by alanine . The T3 interface , unique to F1 , was targeted through replacement of Lys150 and Thr152 with glycines . The hydrophobic T4 interface , formed by the α8 helices of the two molecules in the asymmetric unit of the crystals , is unique to F2; it was changed by mutating Ile216 to aspartate and Met219 to alanine . These two residues contribute about 50% of the total buried surface area to the α8-α8 interface . NMR spectroscopy was used to assess the effect of these mutations on the folded state of NSs-ΔNΔC in solution . 1H spectra agreed with conservation of structure in all mutated proteins ( Figure 7—figure supplement 1 ) . All spectra showed the same signal dispersion and no indications of significant changes in structure , such as partial unfolding . High-field shifted methyl signals between −1 and 0 ppm were observed at the same chemical shifts in all spectra . Such signals belong to buried aliphatic side-chains , and conservation of their chemical shifts reflects preservation of the hydrophobic packing in the protein variants . Three recombinant viruses encoding full-length NSs containing interface mutations were rescued , rMP12muT1NSs , rMP12muT3NSs , and rMP12muT4NSs . Recombinant viruses grew to similar titers to that of the parental virus rMP12 and had indistinguishable plaque morphologies . Segment specific RT-PCR followed by Sanger sequencing was used to confirm the genomic composition of recombinant viruses . NSs-specific immunofluorescence staining showed the presence of filaments in nuclei of rMP12mut4NSs infected Vero-E6 cells ( Figure 7 ) . In contrast , no filaments were observed in nuclei of cells infected with rMP12muT1NSs and rMP12muT3NSs . In both cases NSs was detected as speckles or diffusely dispersed throughout the nuclei ( Figure 7 ) . These data indicate that all three NSs variants had retained the ability to be imported into nuclei . Mutating interfaces observed in the F1 crystal fibril clearly affected intranuclear NSs filament formation , whereas mutating an interface unique to the F2 fibril did not . Therefore , we conclude that the F1 crystal fibril is not only more stable than F2 , but that it also represents the architecture of NSs in intranuclear filaments .
Innate immune antagonism is a common function of many viral non-structural proteins ( Randall and Goodbourn , 2008 ) , including NSs proteins from bunyaviruses ( Walter and Barr , 2011 ) . Non-structural viral proteins have been previously structurally characterized , such as the NS1 protein of influenza virus ( Bornholdt and Prasad , 2008; Liu et al . , 1997 ) , VP35 of Ebola virus ( Leung et al . , 2009 ) or NS5A from hepatitis C virus ( Feuerstein et al . , 2012; Tellinghuisen et al . , 2005 ) . These studies have contributed to understanding both viral pathology and the innate immune system . The NSs protein is the main virulence factor of RVFV and most bunyaviruses . RVFV lacking NSs cannot inhibit the interferon response and causes asymptomatic infection in mice ( Bouloy et al . , 2001; Muller et al . , 1995; Vialat et al . , 2000 ) . Naturally occurring RVFV clone 13 encoding NSs lacking residues 16–198 ( Muller et al . , 1995 ) , and a recombinant virus deficient of NSs , are currently candidates for live-attenuated vaccines for RVFV ( Bird et al . , 2008; Bird et al . , 2011 ) . The mode of action of RVFV NSs is understood to an extent . Several binding partners have been identified , and models for NSs function proposed ( Ly and Ikegami , 2016 ) . The underlying mechanisms of these interactions are however unknown , arguably because so far NSs proteins have largely evaded molecular and structural characterization . Here we report the first structure of a bunyaviral NSs protein . Recombinant full-length RVFV NSs was found to form large aggregates in solution , and was not suitable for NMR or crystallography . Therefore , we designed a stable construct amenable to crystallization through double deletion of 82 N-terminal and 17 C-terminal residues ( NSs-ΔNΔC ) . Both these regions had previously been linked to biological functions of NSs , particularly its intranuclear localization and fibrillation ( Yadani et al . , 1999; Billecocq et al . , 2004; Cyr et al . , 2015 ) . However , we demonstrate that a recombinant virus containing NSs-ΔNΔC successfully induces filament assembly in nuclei of infected cells ( Figure 6 ) , indicating neither the truncated N- or C-terminal regions are required for fibrillation . NSs does not contain any nuclear localization signals , thus nuclear import likely relies on host proteins such as the p44 subunit of the TFIIH transcription complex ( Le May et al . , 2004 ) . Interaction sites for any such binding partners are predicted to be near the N-terminus , since mutation of a PxxP putative protein interaction motif ( residues 29–32 ) results in retention of NSs in the cytoplasm ( Billecocq et al . , 2004 ) . Our data , however , show that the N-terminal 82 residues of NSs ( including the 29–32 motif ) are dispensable for nuclear localization . Deletion of 17 C-terminal residues was based on 2D NMR of NSs-ΔN showing the presence of an intrinsically unfolded region . Notably , this truncation matches one reported in a study where RVFV NSs lacking these C-terminal residues ( encoded by recombinant Semliki Forest virus ) was localized in nuclei of infected cells , but did not form filaments ( Yadani et al . , 1999 ) . A more recent study suggests the 261-FVEV-264 motif is the binding site for the TFIIH subunit p62 , and this interaction is required for filament formation ( Cyr et al . , 2015 ) . While the C-terminal tail ( including the FVEV motif ) is missing in the NSs-ΔNΔC construct , we nevertheless observed intranuclear filaments indistinguishable from filaments in rMP12-infected cells . Therefore , neither the intrinsically disordered C-terminal nor the N-terminal regions of NSs are required for filament formation . The discrepancies between published results and our findings regarding the terminal regions of NSs may be reconciled if a role for the N-terminal domain of NSs in self-association is considered . The N-terminal domain is included in the NSs constructs used in both studies cited above ( Cyr et al . , 2015; Yadani et al . , 1999 ) that suggest a critical role of the C-terminal tail for NSs self-association . However , removal of the N-terminal domain had a clear effect on the oligomeric state of NSs-ΔN , where the C-terminal tail is present ( Figure 1A ) . In crystal fibrils , NSs-ΔNΔC N- and C-termini are close to an apparent cleft in the middle of tetramer T1 , which could accommodate the N-terminal domain , assuming full-length NSs fibrils contain the same core-domain arrangement . In which case , intermolecular interactions between monomer terminal regions are likely ( Figure 3—figure supplement 2 ) . Arguably , deletion of either N- or C-terminal regions results in non-native interactions that destabilize fibrils , artifacts abrogated in a doubly truncated construct . Our data have implications for understanding the interaction between NSs and TFIIH . The model for NSs function previously proposed ( Cyr et al . , 2015 ) assumes initial binding of NSs to p62 , with subsequent disintegration of the TFIIH complex , followed by sequestration of p44 into NSs filaments ( Le May et al . , 2004 ) . This hypothesis is not supported by our findings , as NSs-ΔNΔC lacks the putative p62-binding motif and still yields filaments . The apparent similarity in size and shape of NSs-ΔNΔC filaments to full-length NSs does not allow conclusions about their composition . We cannot rule out the possibility that NSs-ΔNΔC filaments observed in vivo are lacking other proteins such as p44 possibly associated with full-length NSs filaments . The NSs core domain structure reported here represents a novel fold . Comparison of the secondary structure distribution in NSs-ΔNΔC and secondary structure predictions for other phleboviral NSs proteins shows very good agreement , suggesting that , despite low to moderate sequence identity of this core region ( 15–32% ) , all phleboviral NSs proteins share a similar core domain fold ( Figure 8 ) . They do however vary greatly in their N- and C-terminal regions , which is particularly evident when comparing NSs from viruses transmitted by mosquitos and sandflies ( e . g . RVFV , sandfly fever viruses ) with viruses transmitted by ticks ( e . g . Heartland phlebovirus and SFTS phlebovirus ) ( Figure 8 ) . The finding of highly organized fibrils in the NSs-ΔNΔC crystal lattice is intriguing due to the natural propensity of NSs for fibrillation in nuclei of RVFV-infected cells . Mutagenic analyses of interfaces defining the two types of crystal fibrils suggested that F1 likely represents the core architecture of NSs filaments , while F2 is a crystal packing artifact . T1 and T3 interface NSs variants were both detected in nuclei of infected cells . In immunofluorescence staining the muT3 interface variant appeared to be much more abundant than the muT1 variant , suggesting that the muT3 variant of NSs may be more stable than the muT1 interface mutant . This agrees with the predicted stability of the T1 tetramer ( Table 2 ) , making it less prone to degradation . To our knowledge , no other Phlebovirus causes intranuclear filament formation , and none of the residues stabilizing interfaces among NSs-ΔNΔC crystal fibrils is conserved in NSs sequences of phleboviruses other than RVFV . Biological relevance of the F1 fibril is also supported by accessibility of the α8 helix in this assembly ( Figure 3—figure supplement 3 ) , which is partially buried in the F2-specific T1-T2 interface ( Figure 4 ) . This helix harbors a binding site for SAP30 , a component of a repressor complex that regulates transcription of the interferon-β promoter ( Le May et al . , 2008 ) . NSs arranged in the manner observed in NSs-ΔNΔC crystals would be able to bind to SAP30 and modulate SAP30 function . PISA interface analysis indicates that potentially some NSs-ΔNΔC oligomers , but neither F1 nor F2 fibrils would be stable in solution . This agrees with the observation that NSs is distributed throughout the cytoplasm at early time points in cell infection experiments , and only forms distinct filaments inside nuclei . Fibrillation therefore depends on factors present in the nucleus , but not the cytosol . Essential factors could be binding partners , but also other conditions that shift the association equilibrium of NSs towards fibrils such as the excluded volume effect , or molecular crowding , which plays a role in compartmentalization and architecture of cell nuclei ( Hancock , 2004; Richter et al . , 2008 ) . It is tempting to speculate that intranuclear crowding conditions are mimicked in crystallization where an excluded volume effect is induced by the precipitant polyethylene glycol ( PEG ) . RVFV is one of the most dangerous human pathogens with the potential to cause wide epidemics in the near future . The first high-resolution structure for the main RVFV virulence factor NSs and other data presented here will facilitate a deeper understanding of NSs function . Ongoing research will help determine if NSs filament assembly is required for virulence , a question that this study does not attempt to answer . NSs proteins of phleboviruses and other members of the Bunyavirales order are highly diverse in sequence , but all play roles in suppressing the innate immune response through various mechanisms ( Ly and Ikegami , 2016 ) . In infected cells , NSs proteins are located in the nuclei and cytosol , or the cytosol only . Currently no other NSs protein is known to cause filaments characteristic for RVFV . To establish whether this unique property of RVFV NSs is related to virulence may require animal models , since NSs is dispensable for viral replication and in vitro cell infection , but is required to cause pathology ( Bouloy et al . , 2001 ) . The structure-based insights presented here should facilitate such studies . Development of modulators of fibril formation may be an attractive option for new therapeutics , especially given the conservation of interface residues in RVFV isolates .
Expression constructs were prepared by cloning RVFV strain MP12 cDNA comprising full-length ( residues 1–265 ) , ΔN ( residues 83–265 ) , and ΔNΔC ( residues 83–248 ) regions of NSs into a modified pMal-C2x vector ( New England Biolabs , MA , USA ) containing an N-terminal hexahistidine tag downstream of the malE ( MBP ) gene and a tobacco etch virus ( TEV ) protease sequence . Proteins were expressed using Escherichia coli BL21 ( DE3 ) . Cells were grown at 37°C in Luria-Bertani broth to an OD600 of 0 . 6 . Protein expression was induced with a final concentration of 0 . 5 mM isopropyl β-d-1-thiogalactopyranoside and allowed to proceed overnight at 18°C . Cells were pelleted by centrifugation and lysed in the presence of 10 mM phosphate buffer , 300 mM NaCl , 20 mM imidazole , pH 7 . 2 using sonication . Soluble proteins were separated from cell debris by centrifugation . Soluble protein was purified by nickel metal-affinity chromatography ( IMAC ) . After binding , resin was washed extensively with 10 mM phosphate buffer , 2 M NaCl , 20 mM imidazole , pH 7 . 2 in order to dissociate any bound DNA . MBP-NSs was eluted in 10 mM phosphate buffer , 300 mM NaCl , 250 mM imidazole , pH 7 . 2 , and dialyzed overnight at 4°C in 10 mM phosphate buffer , 300 mM NaCl , 2 mM DTT , pH 7 . 2 with simultaneous cleavage by TEV protease . MBP was separated from NSs by reverse nickel IMAC . NSs was further purified by size exclusion chromatography on a 16/60 S75 Superdex column ( GE Healthcare , UK ) in 10 mM phosphate , 300 mM NaCl , 1 mM DTT , pH 7 . 2 . Interface mutation variants of NSs-ΔNΔC were generated through site-directed mutagenesis . NSs-ΔNΔC-muT1 ( Arg88Asp , Ser228Ala ) was generated in two stages of amplification , while NSs-ΔNΔC-muT3 ( Lys150Gly , Thr152Gly ) and NSs-ΔNΔC-muT4 ( Ile216Asp , Met219Ala ) were made in single steps . Interface mutation variants were expressed and purified as previously described for wild-type NSs-ΔNΔC . NSs 83–248 ( NSs-ΔNΔC ) was dialyzed into 20 mM HEPES , 150 mM NaCl , 1 mM DTT , pH 7 . 2 , concentrated to 30 mg mL−1 . Protein crystals were grown at 4°C by sitting-drop vapor diffusion in 0 . 1 M Tris-HCl pH 8 . 5 , 8% ( w/v ) PEG 10 , 000 with protein and reservoir combined in a 2:1 ratio . Crystals of NSs-ΔNΔC grew as elongated hexagons and reached their maximum size of about 0 . 3 × 0 . 1 × 0 . 1 mm after two weeks . Crystals were cryo-protected by brief immersion in reservoir supplemented with 30% ( v/v ) glycerol then flash frozen in liquid nitrogen . Native and anomalous data were collected using a 2 . 0 Å wavelength beam at the Diamond Light Source ( I03 beamline ) . Diffraction over a 500° angle yielded a high-multiplicity dataset with maximum resolution of 2 . 2 Å . Crystals indexed in space group P6422 with cell dimensions a = 123 . 8 Å , b = 123 . 8 Å , c = 174 Å; α = 90° , β = 90° , γ = 120° . Data were indexed and integrated using XDS , scaled using XSCALE ( Kabsch , 2010 ) , and merged with Scala ( Evans , 2006 ) . The structure was solved by single anomalous dispersion ( SAD ) ; phase determination from native sulfur anomalous signal was performed using SHELX ( Sheldrick , 2010 ) . PHENIX ( RRID:SCR_014224 ) AutoBuild was used for initial automated model-building ( Terwilliger et al . , 2008 ) . Density modification and model refinement was performed with REFMAC5 ( RRID:SCR_014225 ) ( Murshudov et al . , 1997 ) and Coot ( RRID:SCR_014222 ) ( Emsley et al . , 2010 ) . Structure validation was conducted with the MolProbity ( RRID:SCR_014226 ) server ( Chen et al . , 2010 ) . Data collection and refinement statistics are listed in Table 1 . NSs constructs were isotopically labeled by expression in M9 minimal media , supplemented with 1 g L−1 of 15NH4Cl and purified as described above for native proteins . NMR samples typically contained 0 . 3 mM protein in 10 mM phosphate , 50 mM NaCl , 0 . 25 mM DTT , pH 7 . 2 , 5% ( v/v ) D2O . 1H-15N HSQC spectra were recorded on a Bruker DRX500 spectrometer equipped with a 5 mm TXIz probe at 15°C . Spectra were processed with NMRPipe ( Delaglio et al . , 1995 ) and analyzed with CCPN Analysis 2 ( Vranken et al . , 2005 ) . For interface mutation variants of NSs-ΔNΔC , samples contained 0 . 03–0 . 25 mM protein in 10 mM phosphate , 150 mM NaF , 1 mM DTT , pH 7 . 2 , 5% ( v/v ) D2O . One-dimensional 1H spectra were recorded with a spectral resolution of 1 . 7 Hz on a Bruker Ascend 700 MHz spectrometer equipped with a Prodigy TCI probe at 22°C . The spectra were processed and analyzed using Bruker Topspin 3 ( RRID:SCR_014227 ) . Vero-E6 cells ( RRID:CVCL_0574 ) were purchased from the European Collection of Authenticated Cell Cultures ( ECACC ) , and tested negative for mycoplasma . Cells were grown in Dulbecco’s modified Eagle’s medium ( DMEM ) supplemented with 10% ( v/v ) fetal calf serum . Cell lines were grown at 37°C with 5% CO2 unless otherwise stated . Recombinant viruses containing truncated or mutated NSs proteins were generated using reverse genetics as previously described ( Brennan et al . , 2014; Le May et al . , 2008 ) . Stocks of recombinant viruses were grown in BHK-21 cells at 33°C by infecting at a multiplicity of infection ( MOI ) of 0 . 01 and harvesting the culture medium at 5–7 days post infection ( p . i . ) All experiments with infectious virus were conducted under containment level 3 conditions . RT-PCR analysis of p1 stock viruses was performed to confirm the correct configuration of the recombinant virus S RNA . BHK-21 ( clone 13 ) cells ( RRID:CVCL_1915 ) were purchased from the European Collection of Authenticated Cell Cultures ( ECACC ) , and tested negative for mycoplasma . Cells were infected with serial dilutions of virus and incubated under an overlay consisting of Glasgow minimum essential medium ( GMEM ) supplemented with 2% ( v/v ) newborn calf serum and 0 . 6% Avicel ( w/v ) ( FMC BioPolymer , Philadelphia , PA ) at 37°C for 4 days . Cell monolayers were fixed with 4% ( w/v ) formaldehyde and plaques were visualized by Giemsa staining . Vero-E6 cells were grown on glass coverslips ( 13 mm diameter ) , infected with recombinant or parental RVFV at an MOI of 5 and fixed at 24 h p . i . in 4% ( w/v ) formaldehyde in PBS . After permeabilization with 0 . 1% ( v/v ) Triton X-100 in PBS , cells were stained with specific primary antibodies , followed by secondary antibody conjugates . Localization of the fluorescently labeled proteins was examined using a Zeiss LSM-710 confocal microscope . Images in Figures 6 and 7 are representatives of two independent infection experiments , with two repeats for each antibody . Monolayer cultures of Vero-E6 cells were seeded in 6-well plates and infected with RVFV MP12 for 24 h p . i . at an MOI of 3 , and subsequently fixed with 2 . 5% ( v/v ) glutaraldehyde overnight at 4°C . Cells were scraped and pelleted by centrifugation followed by fixation with 1% ( w/v ) osmium tetroxide ( TAAB Labs , UK ) and staining with 2% ( w/v ) aqueous uranyl acetate for 1 hr at room temperature . Cells were then harvested into PBS and pelleted through 1% ( w/v ) SeaPlaque agarose ( Sigma , UK ) at 45°C . The agar was set at 4°C and cell pellets were cut into ~1 mm cubes , which were dehydrated through a graded alcohol series ( 30–100% ( v/v ) ) and embedded in Epon 812 resin ( TAAB Labs , UK ) followed by polymerization for 3 days at 65°C . Thin sections of 120 nm were cut with a UC6 ultramicrotome ( Leica Microsystems , Germany ) and examined with a JEOL 1200 EX II electron microscope and images were recorded on a Gatan Orius CCD camera . The NSs-ΔNΔC protein structure , and the data used to derive these , have been deposited at PDBe ( RRID:SCR_004312 ) with accession number 5OOO . | Rift Valley fever phlebovirus ( RVFV ) is a virus of humans and livestock , transmitted by mosquitos and contact with infected animals . Infection can cause severe disease , including hemorrhagic fever , and may lead to death . Historically , the virus was only found in central Africa but it has spread for instance to the Arabian Peninsula . There is a risk that the virus may appear in temperate regions including Europe because global warming is allowing the mosquitos that carry the virus to extend their geographic range . There are no vaccines or treatments available for use in humans so if there is a serious outbreak of the virus it could become an epidemic and cause great economic losses and severe human disease . RVFV relies on a protein called NSs to cause disease . In cells of infected animals and humans NSs forms filaments inside the nucleus , the control center of the cell , and disarms the immune system . However , it is not known precisely how NSs works . To address this question , Barski , Brennan et al . used a technique called X-ray crystallography to study the atomic three-dimensional structure of NSs . This revealed that the center of the protein contains a core domain that causes the filaments to form . Further experiments identified how the NSs core comes together to build the filaments inside the cell nucleus . These findings represent an important step towards understanding how the NSs protein helps RVFV to cause disease in humans and livestock . In the future , this work may aid the development of much needed drugs and vaccines against RVFV . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics",
"microbiology",
"and",
"infectious",
"disease"
] | 2017 | Rift Valley fever phlebovirus NSs protein core domain structure suggests molecular basis for nuclear filaments |
Memory , on multiple timescales , is critical to our ability to discover the structure of our surroundings , and efficiently interact with the environment . We combined behavioural manipulation and modelling to investigate the dynamics of memory formation for rarely reoccurring acoustic patterns . In a series of experiments , participants detected the emergence of regularly repeating patterns within rapid tone-pip sequences . Unbeknownst to them , a few patterns reoccurred every ~3 min . All sequences consisted of the same 20 frequencies and were distinguishable only by the order of tone-pips . Despite this , reoccurring patterns were associated with a rapidly growing detection-time advantage over novel patterns . This effect was implicit , robust to interference , and persisted for 7 weeks . The results implicate an interplay between short ( a few seconds ) and long-term ( over many minutes ) integration in memory formation and demonstrate the remarkable sensitivity of the human auditory system to sporadically reoccurring structure within the acoustic environment .
Memory is a crucial component of sensory perception , on multiple processing levels ( Bale et al . , 2017; Muckli and Petro , 2017 ) . In the auditory modality , the ability to identify essentially any sound source , from footsteps to musical melody , requires the capacity to hold consecutive events in memory so as to link past and incoming information into a coherent emerging representation ( Koelsch et al . , 2019; McDermott et al . , 2013; Winkler et al . , 2009 ) . Whilst traditional models of sensory memory ( e . g . Cowan , 1998 ) argued that such sensory traces are characterized by short retention times and computational encapsulation , a large body of work has since revealed that observers can retain detailed sensory information implicitly , over long periods ( Arciuli and Simpson , 2012; Chun , 2000; Jiang et al . , 2005; Kim et al . , 2009; Vogt and Magnussen , 2007; Winkler and Cowan , 2005 ) . A compelling instance was demonstrated by Agus et al . , 2010; ( see also Agus and Pressnitzer , 2013; Kang et al . , 2017 who showed that naive listeners readily remembered certain spectro-temporal features of random noise bursts , such that reoccurring snippets were recognized weeks after initial exposure . Here , we focus on long-term memory formation for arbitrary frequency patterns within rapidly unfolding sequences of discrete sounds . We ask whether naïve listeners can become sensitized to sparsely reoccurring tone sequences and investigate the conditions under which such memories are formed . To formalize the underlying psychological mechanisms , we simulate human performance with a probabilistic model of sequential memory ( Harrison et al . , 2020; Pearce , 2018 ) . The experimental design ( Figure 1 ) capitalizes on a paradigm developed by Barascud et al . , 2016 for measuring listeners’ sensitivity to complex acoustic patterns . Using fast sequences of short tones , they showed that listeners can rapidly detect the transition to a regularly repeating pattern ( REG ) from a sequence of random tones ( RAN ) . Sequences were novel and too rapid to allow for conscious tracking , but on most trials , participants were able to respond soon after the onset of the second cycle of regularity , implicating an efficient memory for the immediate sequence context . Here , we ask how this memory is affected if the tone pattern was already experienced in the past . Reaction times in Barascud et al . , 2016 were consistent with those obtained from an ideal-observer model based on prediction by partial matching ( PPM; Pearce , 2005; Pearce , 2018 ) . Shown to be an effective model of human auditory sequence learning on multiple time scales ( Agres et al . , 2018; Di Liberto et al . , 2020; Harrison and Pearce , 2018; Pearce , 2018; Pearce and Wiggins , 2006 ) , this model proposes that listeners acquire an internal representation of the sound input by keeping track of multiple-order Markovian transition probabilities . This context is then used to evaluate the ( un ) expectedness of ensuing sounds by deriving a measure of surprisal ( information content – IC; negative log probability ) . RAN and REG sequences differ in unexpectedness ( high for RAN , low for REG ) . The transition from a random to a regular pattern ( RANREG stimulus ) can therefore be detected as a salient drop in information content in the model output ( Figure 1 ) which reflects increasing compatibility between the incoming sounds and the stored context . The pattern of behavioural reaction times as well as brain response latencies recorded from naive , passively listening participants ( Barascud et al . , 2016; Southwell et al . , 2017; Southwell and Chait , 2018 ) suggest that listeners indeed identify the emergence of regularity by detecting the associated drop in information content and that such tracking of instantaneous expectedness constitutes an automatic , inherent aspect of auditory sequence processing . We used a combination of behavioural manipulation and modelling to examine the durations over which these memory representations are maintained by introducing rare pattern reoccurrences . One might expect that detection of regularities benefits not only from immediate sequence context , but also from traces accumulated over a longer period . Participants listened to RAN and RANREG sequences ( as shown in Figure 1 , see stimulus examples: 'Sound - RAN' , and 'Sound - RANREG' ) , and were instructed to press a keyboard button as soon as possible when a transition to REG was detected . New sequences were generated on each trial , but unbeknownst to participants , a few different regular patterns reoccurred very sparsely ( every ~3 min ) across trials ( RANREGr ) . We hypothesized that , if the stored representation of a pattern strengthens through repetition , the information content associated with a transition to a familiar regularity will dip earlier than that associated with a novel regular pattern ( Figure 1 , yellow line in the cartoon model ) , reaching the putative detection threshold more quickly . Behaviourally , this should be revealed as faster reaction times to reoccurring patterns ( ‘RT advantage’ in Figure 1 ) . The size of this effect may provide a window into the latent variables associated with the retention of sensory information in memory . Several properties render this paradigm attractive . First , all sequences consist of the same 20 frequency ‘building blocks’ . This simplifies parametrization and modelling of the task , while retaining sufficient pattern complexity ( there are more than a trillion permutations of 20 frequencies ) . Second , these 20 frequencies are isochronous and occur with equal probability and roughly equal temporal density in all conditions: stimuli are thus matched in terms of long-term spectrum , average statistics and time patterning . The only difference between RAN and REG patterns and , importantly , between REG and REGr patterns , is the specific arrangement of these tone-pips over time . To distinguish a familiar regularity from a novel one , the specific tone-pip permutation must be remembered ( we confirm this explicitly in Experiment 1B ) . Third , the task does not require listeners to memorize sounds explicitly: the emergence of the regularity readily pops out perceptually ( see stimulus examples in supplementary materials ) . The task thus taps the process by which we automatically glean acoustic information from an ongoing sound-stream . Lastly , the sporadic presentation of REGr prevents them from becoming apparent to the listener , thereby allowing us to focus on putative implicit processes which underlie memory formation . Across the experiments presented here , we ask whether human listeners form implicit long-term memories of sparsely reoccurring regular patterns ( yes ) , whether this memory is robust to interference ( yes ) , and whether it can be formed through passive exposure ( partially ) . Through a combination of behavioural manipulation and modelling , we also demonstrate the interplay between short ( a few seconds ) and long ( over many minutes ) integration in the process of long-term memory formation . Overall , the results highlight the remarkable attunement of the auditory system to exceedingly sparse repeating patterns within the unfolding acoustic environment .
Figure 2A-D plots the mean and individual results of the regularity detection task performed in three sessions: five blocks on day 1 , one block after 24 hr ( ‘24 hr’ ) and one block after 7 weeks , ( ‘7 w’ ) . Participants were highly accurate in detecting regularities ( Figure 2A ) : d’ plateaued at near ceiling performance after the first block . No difference was observed between hit rates for RANREG and RANREGr [no main effect of condition: F ( 1 , 19 ) = . 39 , p = 0 . 539 , ηp2 = . 02; no main effect of block: F ( 5 , 90 ) = 0 . 46 , p = 0 . 804 , ηp2 = . 02; no interaction between condition and block: F ( 5 , 90 ) = 1 . 10 , p = 0 . 367 , ηp2 = . 06] . Despite the ceiling effects associated with pattern detection ( mean hit rate = 97 . 3% ) , faster RTs in RANREGr than in RANREG ( ‘RT advantage’ ) were observed in all participants by the end of the first session ( block 5; Figure 2D ) , indicating a clear implicit memory for the reoccurring patterns . A repeated measures ANOVA on RTs with condition ( RANREG and RANREGr ) and block as factors yielded a main effect of condition [F ( 1 , 18 ) = 34 . 09 , p < 0 . 001 , ηp2 = . 65] , main effect of block [F ( 5 , 90 ) = 9 . 24 , p < 0 . 001 , ηp2 = . 3] and an interaction between condition and block [F ( 5 , 90 ) = 6 . 88 , p < 0 . 001 , ηp2 = . 28] . Specifically , in the first block of the first session , performance did not differ between RANREG and RANREGr [t ( 18 ) = 0 . 794 , p = 1] . By the end of the second block ( after 6 REGr reoccurrences ) , a significant difference ( ~140 ms; 2 . 8 tones ) between RTs was observed [REG – REGr: t ( 18 ) = 3 . 964 , p = 0 . 006] . This difference grew over the following blocks ( all ps < 0 . 001 ) , plateauing after block 3 ( 233 ± 0 . 17 ms; 4 . 7 tones ) . The RT advantage on the third block did not differ from the fourth [t ( 18 ) = −0 . 907 , p = 1] nor from the fifth block [t ( 18 ) = −0 . 0003 , p = 1] ) . In Experiment S1 ( Appendix 1—figure 1 ) , we demonstrate that similar effects are obtained when doubling the number of REGr patterns to be memorised ( six different patterns per participant ) . In Experiments S2A and S2B ( Appendix 1—figure 2 ) , we further demonstrate that the memory trace is not abolished by introducing ‘interrupting blocks’ ( in which REGr were not presented ) between ‘standard blocks’ ( in which REGr patterns reoccurred every ~3 min ) . Critically , implicit memory for reoccurring regularities persisted after 24 hr and after 7 weeks: the RT difference between novel and reoccurring sequences remained constant between the last block of day 1 ( block 5 ) and after 24 hr [t ( 18 ) = 0 . 139 , p = 0 . 891] , as well as between 24 hr and 7 weeks later [t ( 13 ) = −0 . 668 , p = 0 . 515] . An inspection of intra-block reoccurrences ( Figure 2—figure supplement 1 ) revealed that the RT advantage for REGr was similar between the third ( last ) intra-block presentation of day 1 and the first intra-block presentation after 24 hr [t ( 18 ) = 0 . 123 , p = 0 . 903]; similarly , in the session conducted after 7 weeks , the RT advantage measured after the first intra-block presentation did not differ from the third ( last ) presentation in the session conducted after 24 hr [t ( 13 ) = 0 . 958 , p = 0 . 356; ( Figure 2—figure supplement 1 ) ] . This suggests that the effect observed after 24 hr and 7 weeks reflects the presence of a lasting memory trace of reoccurring regularities rather than rapid within-block re-learning . Further , we examined the correlation of individual participants’ RT advantage across the three sessions ( Figure 2C ) . A robust correlation was found between the end of the first day ( block 5 ) and the measurement taken after 24 hr ( spearman’s rho = 0 . 635 , p = 0 . 004 ) – participants who exhibited a larger RT advantage at the end of the first day were also those showing a larger advantage 24 hr later . A similar correlation was found with performance after 7 weeks ( spearman’s rho = 0 . 740 , p = 0 . 003 ) . This confirms strong reliability of individual effects . Explicit memory for reoccurring regularities was examined at the end of each session by means of a familiarity task . Only regular sequences were presented: REGr ( one presentation of each pattern ) were intermixed with previously unheard REG patterns . Participants were instructed to indicate which patterns sounded ‘familiar’ . Classification was evaluated using the MCC score ( see Materials and methods ) which ranges between 1 ( perfect classification ) to −1 ( total misclassification ) . Whilst low overall , the mean MCC on each testing session indicated above chance performance [day 1: mean = 0 . 231; t ( 18 ) = 4 . 214 , p < 0 . 001; 24 hr: mean = 0 . 44 , t ( 18 ) = 7 . 044 , p < 0 . 001; 7 w: mean = 0 . 360 , t ( 13 ) = 5 . 204 , p < 0 . 001] ( see Figure 2—figure supplement 2 ) . An improvement in MCC scores was observed between day 1 and 24 hr later [t ( 18 ) = −3 . 635 , p = 0 . 004] , suggesting potential consolidation . There was no change in MCC scores between the 24 hr session and 7 weeks later [t ( 13 ) = 0 . 348 , p = 1] . Importantly , MCC scores did not correlate with the RT advantage: MCC on day 1 did not correlate with the RT advantage observed in block 5 ( spearman’s Rho = 0 . 307; p = 0 . 201; a similar result was also obtained when pooling across participants from Exp . 1A and Exp . S1 ( which used 6 REGr patterns , see Appendix 1 ) ( Spearman’s Rho = 0 . 114; p = 0 . 493; N = 38 ) . Though a weak correlation between RT advantage and MCC was measured after 24 hr ( uncorrected; Spearman’s Rho = 0 . 459 , p=0 . 048 , N = 19 ) , it disappeared after 7 weeks ( Spearman’s Rho = −0 . 024 , p = 0 . 934 , N = 14 ) . Therefore , implicit memory for reoccurring patterns , observed in nearly all participants , is not linked to explicit awareness of reoccurrence . To confirm that the RT advantage effects are driven by memory of sequential structure , we tested whether implicit memory for reoccurring patterns is tolerant to time reversal of the originally learned patterns ( Figure 2E–G ) . Participants performed the regularity detection task as in Exp . 1A over six experimental blocks . The first four were identical to those in Exp . 1A . In the fifth block , REGr sequences were replaced by time-reversed versions . In block 6 , the original REGr were introduced again . Participants were naive to the experimental manipulation . It was expected that , if implicit memory is specific to the sequential structure of regularity , the RT advantage should disappear in the time-reversed block ( see also Kang et al . , 2017 ) . Blocks 1–4 revealed the same effects as in Exp . 1A ( Figure 2F ) [main effect of condition: F ( 1 , 19 ) = 71 . 96 , p < 0 . 001 , ηp2 = . 79; main effect of block: F ( 3 , 5 ) = 9 . 90 , p < 0 . 001 , ηp2 = . 34; interaction condition by block: F ( 3 , 57 ) = 5 . 67 , p < 0 . 001 , ηp2 = . 23] . Specifically , in the first block RTs in the RANREGr condition were similar to those in RANREG [t ( 19 ) = 0 . 725 , p = 1] , but became progressively faster ( 114 ms; 2 . 27 tones ) in the second block [t ( 19 ) = 3 . 56 , p = . 01] , and across the remaining blocks ( all ps < 0 . 001 ) ( 203 ms; 4 . 1 tones in the 4th block ) . Importantly , this RT advantage was abolished in the time-reversed block , but restored in the subsequent block containing the originally learned REGr: a repeated measures ANOVA with condition ( RANREG and RANREGr ) and the last two blocks as factors yielded a main effect of condition ( F ( 1 , 19 ) = 25 . 57 , p < 0 . 001 , ηp2 = . 57 ) , a main effect of block ( F ( 1 , 19 ) = 18 . 09 , p < 0 . 001 , ηp2 = . 49 ) , and an interaction condition by block ( F ( 1 , 19 ) = 40 . 03 , p < 0 . 001 , ηp2 = . 68 ) , demonstrating the significantly greater RT advantage ( RANREG novel – RANREGr ) in the last than in the time-reversed block [t ( 19 ) = 6 . 33 , p < 0 . 001] . The RT advantage for REGr in the third intra-block presentation of block 4 ( Figure 2—figure supplement 3 ) was greater than in the first intra-block presentation of the time-reversed block [t ( 19 ) = −2 . 261 , p = 0 . 035] , but similar to the first intra-block presentation of the last block reintroducing the original REGr [t ( 19 ) = 0 . 788 , p = 0 . 440] . These results constrain the nature of the observed memory effect to sequential information . We tested whether adjacent repetition of patterns ( as is inherently the case for REG sequences ) is required for implicit memory to be formed ( Figure 3 ) . Over four blocks , listeners were exposed to RAN , RANREG and RANREGr trials as in previous experiments . We also introduced a new condition , PATinRAN ( Figure 3A ) , which consisted of two identical non-adjacent 20-tone patterns ( PAT ) embedded within a random sequence of tone-pips . The second appearance always occurred at the end of the sequence . The first appearance was embedded partway through the sequence at an average distance of 1 . 7 s ( range 0 . 5–2 . 9 s ) . To understand whether memories of non-adjacent patterns ( PAT ) can be formed during listening , three different PAT reoccurred three times within block ( PATinRANr; the random parts of the sequences as well as the separation between the two PAT patterns remained random on each trial ) . Both non-adjacent ( PATinRAN , PATinRANr ) and adjacent ( RANREG , RANREGr ) trials included two repetitions of each pattern with the only difference being that they were contiguous in the latter and separated by random tones in the former . Participants were instructed to respond if they detected two identical , not necessarily contiguous , 20-tone patterns within a trial; 50% of the trials consisted of fully random patterns . In order to make sure that participants paid equal attention to the ( harder ) PATinRAN sequences , accuracy was emphasized over response speed . In the last block ( block 5; ‘test' block ) , we tested whether , following a comparable amount of exposure through block 1 to 4 , PATinRANr and RANREGr patterns were similarly remembered . To equate difficulty of pattern detection in this block , PATinRANr sequences were replaced by versions where the two cycles were set adjacent . We refer to these conditions as RANREGr* . Participants were instructed to respond as quickly as possible . We compared the magnitude of the RT advantage associated with RANREGr* to that associated with RANREGr . Figure 3B shows the detection performance during the exposure blocks ( 1 to 4 ) . Despite having practised the PATinRAN condition , detection performance was overall worse , and substantially more variable in PATinRAN ( mean over blocks 1–4: 47 . 36 ± 16 . 5% ) relative to RANREG ( 88 . 47 ± 11 . 6% ) , and improved less across blocks [main effect of condition: F ( 1 , 29 ) = 419 . 01 , p < 0 . 001 , ηp2 = . 94; main effect of block: F ( 3 , 87 ) = 9 . 24 , p < 0 . 001 , ηp2 = . 24; interaction of condition per block: F ( 3 , 87 ) = 4 . 83 , p = 0 . 004 , ηp2 = . 14] . Thus , whilst a pattern is highly detectable when contiguously repeated , performance drops substantially when the repetition is not adjacent , presumably due to limits on short-term memory . Focusing on the 4th block ( Figure 3C ) : a repeated measures ANOVA with the factors reoccurrence ( novel/reoccurring patterns ) and adjacency ( adjacent/non-adjacent patterns ) yielded a significant main effect of adjacency [F ( 1 , 29 ) = 205 . 99 , p < 0 . 001 , ηp2 = . 88] . As expected , whilst participants were very apt at detecting RANREG patterns , performance on PATinRAN was substantially more variable and lower overall . Interestingly a main effect of reoccurrence [F ( 1 , 29 ) = 21 . 74 , p < 0 . 001 , ηp2 = . 43] , was also observed , with no interaction between the two factors [F ( 1 , 29 ) = 3 . 95 , p = 0 . 056 , ηp2 = . 12] . Therefore , detection data showed an increase in accuracy for reoccurring patterns in both adjacent and non-adjacent conditions . The emergence of this effect for RANREGr , despite its absence in Exp . 1A , is presumably driven by the below ceiling performance observed here ( mean hit rate = 93% relative to 97 . 5% in Exp . 1A ) – likely a consequence of the extra behavioural strain introduced by the PATinRAN stimuli . Critically , the finding of increased hit rates for PATinRANr ( a mean increase of 15% ) demonstrates that , through repeated exposure , listeners formed a memory trace for the non-adjacent patterns . RT results across block 1 to 4 are shown in Figure 3D . To allow for a comparison across conditions , RTs here are measured relative to the onset of the second regularity cycle ( indicated with a red line in Figure 3A ) . Since participants were encouraged to prioritise accuracy over speed in these blocks , the RT data in blocks 1–4 were not statistically analysed . However , an RT advantage ( reaching 131 ms , 2 . 63 tones in block 4 ) is clearly visible for RANREGr relative to RANREG stimuli . Test block: as a critical test for the formation of memory traces , we assessed the presence of an RT advantage in the 1st intra-block presentation of RANREGr and RANREGr* ( Figure 3E ) . The RT advantage was significantly different from zero in RANREGr [one-sample t-test: t ( 29 ) = 3 . 724 , p = 0 . 001] , but not in the RANREGr* condition [one-sample t-test: t ( 29 ) = . 419 , p = 0 . 678] . A paired t-test further confirmed a greater RT advantage in the RANREGr than in the RANREGr* condition [t ( 29 ) = 3 . 169 , p = 0 . 003] . This indicates that , as a group , participants did not demonstrate an immediate RT advantage to RANREGr* patterns . As seen in Figure 3E , an RT advantage in RANREGr* emerged following the second intra-block presentation . This effect may be associated with learning within the test block . A repeated measures ANOVA on RT advantage in the test block with the factors condition ( REGr / REGr* ) and intra-block presentation ( 1st / 2nd / 3rd ) revealed a main effect of condition [F ( 1 , 29 ) = 9 . 09 , p = 0 . 005 , ηp2 = . 24] but no main effect of intra-block presentation [F ( 2 , 58 ) = 0 . 67 , p = 0 . 515 , ηp2 = . 02] , or interaction [F ( 2 , 58 ) = 1 . 27 , p = 0 . 287 , ηp2 = . 04] , consistent with an overall smaller RT advantage to RANREGr* . As an exploratory analysis , we tested whether higher detection accuracy for non-adjacent patterns ( hit rates for PATinRANr / PATinRAN in block four ) predicted a greater RT advantage when the patterns were set adjacently in the test block ( REGr* ) . We observed a significant moderate correlation between the detection accuracy of PATinRANr in block four and the RT advantage in the 1st intra-block presentation of REGr* ( spearman’s rho = 0 . 429 , p = 0 . 018 ) such that those participants who exhibited a higher detection accuracy for PATinRANr in block 4 , also demonstrated a higher RT advantage for REGr* in the test block . This correlation with RT advantage was specific to PATinRANr , in that it did not extend to PATinRAN ( spearman’s rho = 0 . 017 p = 0 . 927 ) and held when the effect of detection accuracy for PATinRAN was accounted for ( spearman’s rho = 0 . 465 , p = 0 . 011 ) . The specificity to PATinRANr suggests that the link is not simply related to some property of short-term memory ( in which case we would have expected a correlation with PATinRAN as well ) , but it is specific to the memory advantage for PATinRANr stimuli which developed over the first four blocks . Overall , these results suggest the presence of measurable ( though small ) memory traces for reoccurring , non-adjacent patterns ( PATinRANr ) . However , it is clear that the formation of robust implicit memory traces for sound sequences depends on short-term memory ( and hence benefits from immediate repetition of patterns ) such that introducing a gap of even 2 s results in substantially weakened storage in memory . We constructed a ‘memory constrained’ computational model , based on ‘prediction by partial matching’ ( PPM; see Materials and methods ) to provide a formal simulation of the psychological mechanisms underlying the process of memory trace formation , as observed in Experiments 1A ( Figure 2 ) , 2 ( Figure 3 ) and S2A ( Appendix 1—figure 2K ) . These experiments reflect critical manipulations of the effect of long- and short-term memory decay . Although the existence of memory decay in humans is in general well established , ways of incorporating memory decay into probabilistic computational models of sequences processing is very much an active topic of research . Our PPM model implemented a single set of values ( Table 1 ) that fully accounted for the dynamics of memory formation observed across experiments . As a benchmark , we also report the results for an equivalent unconstrained model ( i . e . with perfect memory ) , as employed in previous research using the same paradigm ( Barascud et al . , 2016 ) . The following cognitive hypotheses were instantiated: Overall , the memory constrained model shows close qualitative correspondence to the pattern of RTs observed in Experiments 1 and 2 , and specifically to the dynamics of the emergence of the RT advantage . Figure 5A shows model outputs for experiment 1A using an unconstrained ( left ) and constrained ( right ) PPM model . The imposed memory constraints are able to reproduce the slow dynamics of REGr memory formation: like the human participants , the constrained PPM model experiences a moderate facilitation effect that grows over successive presentations of identical regular patterns . Figure 4B illustrates this effect in more detail , plotting average information content profiles for RANREGr trials in block five as compared to RANREGr trials in block 1 . It is important to note that the steady long-term decay , which is a key feature of the memory constrained model predicts that the performance facilitation should disappear after 24 hr , and certainly after 7 weeks . After such time periods , the memory traces for the reoccurring patterns should decay to zero , and the corresponding facilitation effect should disappear . Remarkably , the participants exhibited unaltered performance facilitation . This suggests that the memory traces of these reoccurring patterns are somehow ‘fixed’ at a certain point during testing . One way of simulating this effect would be to change the asymptote of the exponential memory decay , such that the memory trace asymptotically approaches a small but non-zero value as time tends to infinity . However , we found that incorporating such an asymptote caused the performance facilitation for RANREGr trials to increase constantly from block to block , in contrast to the slow plateau shown in the behavioural data . It seems likely , therefore , that there remains a non-trivial ‘fixing’ effect that may reflect consolidation processes , not accounted for by the current model ( to our knowledge there is no other statistical learning model that accounts both for learning dynamics and long-term fixed effects ) . Experiment 2 investigated the effect of pattern adjacency on pattern detection and memory formation . We trained unconstrained and constrained models on blocks 1–4 , and report their performance for the ‘test’ block ( block 5 ) . As expected , the unconstrained PPM model is unaffected by adjacency ( Figure 5B left ) . The memory-decay PPM model ( Figure 5B right ) fully reproduces the behavioural data ( Figure 3E ) . Overall , the modelling successfully replicated the slow dynamics of memory formation exhibited by human listeners demonstrating that memory constrained transition-probability learning is a plausible computational underpinning of sequential pattern acquisition . Does memorization of a new set of REGr interfere with the representation of a previously memorized set ? Participants performed the same transition detection task as in Exp . 1A . They were exposed to a set of three reoccurring patterns ( REGr1 ) in the first three blocks , followed by three blocks in which another set of patterns ( REGr2 ) reoccurred . Blocks 7 and 8 then re-tested memory for the reoccurring regularities of set 1 and set 2 , respectively . After 24 hr , memory for the two sets was tested again . Clear implicit memory for the first set of targets ( REGr1 ) , as indicated by an RT advantage , was observed after the 3rd block ( Figure 6B ) [main effect of condition: F ( 1 , 28 ) = 41 . 01 , p < 0 . 001 , ηp2 = . 59; main effect of block: F ( 3 , 84 ) = 15 . 69 , p < 0 . 001 , ηp2 = . 36; condition by block interaction: F ( 3 , 84 ) = 6 . 83 , p < 0 . 001 , ηp2 = . 20] . As expected , after three blocks of exposure the RT advantage in the RANREGr1 condition ( 163 ms – 3 . 3 tones ) was similar to that observed in Exp . 1A above . Critically , this RT advantage for RANREGr1 was not perturbed after the presentation of the second set of regularities ( REGr2 ) [RT advantage in block three vs . block 7: t ( 28 ) = . 877 , p = 0 . 387] . It also lasted after 24 hr [RT advantage in block seven vs . after 24 hr: t ( 28 ) = −0 . 553 , p = 0 . 584] , and was similar to the 24 hr RT advantage observed in Exp . 1A [no main effect of experiment: F ( 1 , 50 ) = . 33 , p = 0 . 567 , ηp2 = . 01] . These results indicate that , once formed , memory traces are neither overwritten nor weakened by ‘interfering’ new sets of reoccurring patterns . In blocks 4–6 presenting the second set of reoccurring regularities ( REGr2 ) also showed an RT advantage , as demonstrated by the emerging separation between the RT to novel and reoccurring regularities . A repeated measures ANOVA on the RT advantage with ‘experimental stage’ ( blocks 1–3 , blocks 4–6 ) and block number ( 1st , 2nd or 3rd ) showed a main effect of block number [F ( 2 , 56 ) = 20 . 13 , p < 0 . 001 , ηp2 = 0 . 42; consistent with a growing RT advantage across blocks] , and stage [F ( 1 , 28 ) = 15 . 70 , p < 0 . 001 , ηp2 = 0 . 36] with no interactions . The main effect of stage suggests an overall larger RT advantage for the first set ( REGr1 ) . The noisier RT pattern observed in blocks 4–6 may be indicative of an order / fatigue effect . Importantly , at the end of day 1 the RT advantage for the two sets of reoccurring regularities did not differ ( block seven vs . block 8: t ( 28 ) = 1 . 721 , p = 0 . 096] . The RT advantage for the second set was maintained when tested after 24 hr ( RT advantage of last block of day one vs . after 24 hr: t ( 28 ) = −0 . 277 , p = 0 . 784 ) , and did not differ from that of the first set [RT advantage after 24 hr for RANREGr1 vs . RANREGr2 t ( 28 ) = 1 . 848 , p = 0 . 075] . In all the previous experiments reoccurring regularities were always presented at the same phase of the REG cycle . Here we asked whether the resulting memory trace was anchored to this fixed boundary – that is , whether listeners remembered the pattern as a specific ‘chunk’ ( Dehaene et al . , 2015; Thiessen , 2017 ) . If so , the RT advantage should reduce when REGr are phase shifted . Listeners were presented with six reoccurring regularities ( REGr ) over three blocks . In block 4 , identical REGr were presented but each presentation was with a shifted onset relative to the originally presented pattern ( see Figure 7A , and Materials and methods ) . Figure 7C shows the progressive emergence of the RT advantage associated with the memorization of the reoccurring patterns [main effect of condition: F ( 1 , 19 ) = 21 . 12 , p < 0 . 001 , ηp2 = . 53; main effect of block: F ( 3 , 57 ) = 18 . 52 , p < 0 . 001 , ηp2 = . 49; condition by block interaction: F ( 3 , 57 ) = 10 . 64 , p < 0 . 001 , ηp2 = . 36] . Specifically , whilst in the first block performance did not differ between RANREG and RANREGr [t ( 19 ) = −0 . 876 , p = 1] , a faster RT to the RANREGr condition developed across ensuing blocks . This effect continued into block 4 , where phase-shifting was introduced ( Figure 7C bottom plot ) . The RT advantage for phase-shifted RANREGr ( 167 ms – 3 . 35 tones ) in block 4 was greater than the RT advantage in block 3 ( 100 ms; 2 tones ) [block three vs . block 4: t ( 19 ) = −13 . 111 , p < 0 . 001] in the majority of participants ( Figure 7D ) , demonstrating a strengthening ( rather than disappearing ) memory effect . The immediate robustness to phase shifting was confirmed by comparing the RT advantage in the first intra-block presentation in block 4 , to that in the third ( last ) intra-block presentation in block 3 ( Figure 7—figure supplement 1 ) . No significant difference was observed [t ( 19 ) = 1 . 069 , p = 0 . 298] , supporting the conclusion that the RT advantage persisted despite phase shifting . Further tests confirmed that the RT advantage for REGr in block 4 was similar across small and large phase shifts: a repeated measures ANOVA with factor phase shift ( small / large , namely 1–5 and 16–19 vs . 6–15 tones from the original onset ) yielded no significant effect of phase shift on the RT advantage [F ( 1 , 19 ) = 0 . 74 , p = 0 . 400] . These results suggest that sequences are not represented as a fixed chunk of sequential items which is retrieved as a single unit , but more likely as a collection of sequential predictions that are flexibly retrieved from memory according to the available sensory information . As a further probe into the nature of the representation of the pattern in memory , in Experiment S3 ( Appendix 1—figure 3 ) we investigated listeners’ tolerance to small frequency transpositions . We reveal a transfer of the RT advantage to the transposed pattern , suggesting that the formed representation is not of an exact echoic nature . It is possible that tolerance to frequency transposition reflects a ‘fuzzy’ spectral representation , although we note that the spacing in the present pool – 12% – is generally larger than the just noticeable difference ( JND ) for frequency typically exhibited by non-musically trained listeners ( Tervaniemi et al . , 2005 ) . Alternatively , the tolerance to transposition may suggest that instead of the specific frequency pattern , the auditory system maintains a representation of the contour , or inter-tone interval within the pattern . We asked whether memories for reoccurring patterns are formed when sequences are not behaviourally relevant . Naïve participants were exposed to three blocks of the same kind as in Exp . 1A , but instructed to detect the STEP changes only , and ignore the other sounds . In the fourth block ( ‘test’ block ) , they were instructed to also detect the RANREG transitions . We analysed the performance in the test block of the pre-exposed group in comparison to the performance of a non pre-exposed ‘control’ group , formed by pooling block one data from several other experiments ( Pooled data-block1 , N = 147 , see Materials and methods ) . Sensitivity to transitions in the test block ( Figure 8A ) was high overall ( mean d’ = 2 . 77 ± . 73 ) , but lower than in the first block of the control group [independent sample t ( 163 ) = −2 . 028 , p = 0 . 044] . This is likely because , in order to keep them naive , participants did not receive training on RANREG detection . In the test block ( Figure 8B ) , the mean RT to RANREGr was significantly faster than that to novel RANREG [t ( 17 ) = 3 . 1 , p = 0 . 006] , consistent with the presence of an RT advantage . The RT advantage in the pre-exposed group ( ~157 ms , 3 . 14 tones ) was substantially greater than in the control group ( ~30 ms , 0 . 6 tones ) [independent sample t ( 163 ) = 3 . 023 , p = 0 . 003] , indicating a beneficial effect of pre-exposure . As a critical test for the presence of a memory trace after pre-exposure , we examined RT in each intra-block presentation of REGr . If memories for reoccurring patterns are formed during pre-exposure , an RT advantage should be exhibited immediately - at the first presentation of REGr in the test block . One sample t-tests demonstrated that an RT advantage was absent at the first and second intra-block presentations [t ( 16 ) = 0 . 377 , p = 0 . 711; t ( 17 ) = 1 . 691 , p = 0 . 109] , but emerged at the third presentation of REGr [t ( 17 ) = 3 . 954 , p = 0 . 001] . We also compared the RT advantage , across intra-block presentations , between the pre-exposed and control groups . A bootstrap approach ( see Materials and methods ) was used to generate a distribution of performance over subsets of 20 participants drawn from the control group and to compare with the actually observed performance in the pre-exposed group ( Figure 8D ) . The plots in Figure 8D show distributions of the RT advantage for the 1st , 2nd and 3rd REGr presentation in the control group . The mean RT advantage of the ‘pre-exposed’ group is shown by the red dots . This analysis revealed that the RT advantage to the 1st presentation did not differ from the control group . However , a difference emerged after the 2nd presentation . This suggests that , by the 2nd appearance of REGr in the ‘test’ block , the passively pre-exposed group exhibited substantially faster responses than non pre-exposed participants . The difference between the passively pre-exposed group and the control group grew further by the 3rd presentation . Overall , these results demonstrate that implicit memory was not present at the onset of the test block ( as evidenced by the lack of an RT advantage ) ; however , learning occurred more rapidly in the pre-exposed listeners such that by the end of the test block , they exhibited a substantially higher RT advantage than that shown by the control group . Explicit memory was poor ( mean MCC = 0 . 064 ) and did not correlate with the RT advantage measured in the test block [Spearman’s Rho = 0 . 235; p = 0 . 347] . We quantified the robustness of the memory effect for reoccurring patterns across the different experiments reported here . Figure 9A shows the distribution of RTs for RANREG vs . RANREGr pooled from block three data , ( i . e . after nine presentations of each REGr; approx . 25 min of listening ) where most data from different experiments were available ( the pilot experiment , Experiment 1A , 1B , 3 , 4 , S1 , and S3 ) . In Figure 9B each dot represents the mean RT for RANREG vs . RANREGr of an individual participant ( N = 147 ) . 88 . 4% of participants exhibited an RT advantage , which we interpret as revealing implicit memory for REGr . We also tested the generality , across patterns , of the observed memory effect . It is important to note that all REGr were similar in the sense that all are composed from the same set of tones and only differed in the specific permutation of their order . Figure 9C plots a distribution of the RT advantage per unique REGr ( 558 overall ) . Though the data are inherently noisy ( RT is quantified as an average over only three presentations in block 3 ) , RT advantage appears to be normally distributed with 75 . 6% of patterns exhibiting a memory effect . This demonstrates that the observed effects are not driven by particularly ‘memorable’ REGr sequences . The same analysis run over block five data ( not shown; N unique REGr = 165 ) showed that 84 . 4% of REGr were associated with an RT advantage after 15 reoccurrences . Figure 9D plots the distributions of group RT advantage per block , based on performance observed across all of the experiments reported ( see Materials and methods ) . A gradual build-up of RT advantage is seen across blocks reaching a mean of 5 . 5 tones by the end of block 5 . Overall the results demonstrate that the memory effect generalizes to most ( healthy , young ) listeners and is not driven by particular memorable stimuli .
The general behavioural pattern revealed here is reminiscent of the ‘noise memory’ effect first shown by Agus et al . , 2010 ( see also Agus and Pressnitzer , 2013; Andrillon et al . , 2015; Gold et al . , 2014; Keller and Sekuler , 2015; Luo et al . , 2013 ) . In that study naïve listeners readily remembered reoccurring white-noise snippets presented amongst novel noise bursts . The learning was unsupervised , rapid , implicit and lasted upwards of 2 weeks . Inspections of the nature of this memory revealed that it was robust to time reversal and even to scrambling into bins as small as 10–20 ms , indicating that the remembered features reflect local spectro-temporal idiosyncrasies within the reoccurring noise snippet ( Agus et al . , 2010; Viswanathan et al . , 2016 ) . The apparent dependence of this memory on certain local features of the noise signal may also explain the high inter-sample variability often seen with this paradigm ( e . g . , the distinction between ‘memorable’ and ‘not memorable’ patterns; Agus et al . , 2010; Viswanathan et al . , 2016; Kang et al . , 2017 ) . In contrast , here we focus on fast memory formation for sequences of discrete tones , distinguishable only by their specific order , and presented in a surrounding context of highly similar patterns ( all sequences consisted of the same 20 ‘building blocks’ ) . We showed that the vast majority of patterns were learned , revealing high sensitivity to reoccurring arbitrary frequency patterns despite the exceedingly rare reoccurrence rate ( every ~3 min; 5% of trials; in contrast to the much more frequent reoccurrence , <~15 s in Agus et al . , 2010 and Kang et al . , 2017 ) . An important question for future work will be to determine whether these effects draw on similar or distinct neural systems ( discussed further below ) . Signals based on tone-pip patterns have long been used to understand the effect of auditory memory on listeners’ perception of sound sequences ( e . g . Watson et al . , 1975; Atienza and Cantero , 2001; Näätänen et al . , 1993; Schröger et al . , 1992; Tervaniemi et al . , 2001; Moldwin et al . , 2017 ) . However , these paradigms are predominantly based on extensive exposure ( in the order of hundreds of consecutive repetitions ) to a single pattern . Of particular relevance is a large body of work , broadly referred to as ‘statistical learning’ , which has demonstrated the brain’s capacity to discover repeating structure in random stimulus sequences ( Conway and Christiansen , 2005; Frost et al . , 2019; Kim et al . , 2009; Saffran et al . , 1999; Saffran and Kirkham , 2018 ) . The classic paradigm ( Saffran et al . , 1996; Santolin and Saffran , 2018 ) involves a small set of syllables arranged into short ‘words’ ( e . g . , three syllables each ) . A few minutes’ exposure to such structured streams leads to learning of the statistical structure of the unfolding sequence such that subjects can distinguish the repeatedly occurring ‘words’ from a random arrangement of syllables . Our results can be interpreted as reflecting similar implicit learning processes . However , in contrast to the demonstrations above which usually involved one or a small number of stimuli that are repeated many times , we show that a very sparse presentation of long patterns , which are intermixed with many highly similar sequences , is sufficient for robust memories to be formed . Note that to focus on implicit memory formation , we placed our listeners in rather extreme conditions , both in terms of presentation rate of reoccurring targets and their complexity . It is possible that relaxing these constraints would result in stronger ( but perhaps more explicit ) memories . We showed that listeners can learn at least six concurrently presented REGr patterns ( Exp . 4 and Exp . S1 in Appendix 1 ) . Important questions for future work involve understanding the capacity limits on this memory and the factors which might affect subsequent forgetting . Overall , we demonstrate that the brain is tuned to retain repeated structure in our acoustic environments , even when such reoccurrences are exceedingly infrequent and the signals are highly similar . Preserving as much information as possible from the unfolding sensory input is important for an organism because the relevance of any single event is not always immediately apparent , but is rather inferred post-hoc , e . g . , through repetition ( “I’ve heard this exact pattern twice within 3 min , therefore it might reflect an individual sound source rather than random noise in the forest"; e . g . , McDermott et al . , 2011; Woods and McDermott , 2018 ) . Our results hint at the heuristics utilized by the brain in determining how representations of statistical structure in the sensory environment are converted from transient to stable forms of memories ( Leimer et al . , 2018; Li and van Rossum , 2020 ) . We used reaction time ( RT ) as a proxy for memory formation . RT allowed us to determine how much information was required for listeners to detect repeating ( REG ) structure and to compare these measures with formal models of sequence processing . We hypothesized that reoccurrence would increase the weight of sequence components in memory resulting in faster detection of regularity . RT thus provided a sensitive means for tracking the formation and maintenance of such memories over time . We observed that the RT to REGr steadily shortened with increasing number of reoccurrences , allowing us to measure the dynamics of memory trace establishment . The ‘RT advantage’ , defined as the difference in RT between novel and reoccurring REG patterns , grew rapidly over the first three blocks ( 9 reoccurrences ) and then stabilized , though evidence from Figure 9D suggests a continuous slow growth throughout the experimental session . The absence of correlation between the familiarity test and the RT advantage suggests a dissociation between implicit memory and explicit recall abilities . The basic behavioural task required participants to detect the transitions from RAN to REG – namely the emergence of repeating structure . As such it fundamentally relied on auditory short-term memory: in order to detect REG patterns , the listener must compare incoming tones to those that occurred at least a cycle ago . The effect of reoccurrence suggested that listeners also draw on much longer-term memory whereby information about previously encountered sounds is retained over minutes between successive REGr presentations . Due to practical issues related to providing breaks , all of the reported experiments were subject to fixed presentation parameters: the experimental session was divided into blocks of roughly 8 min and REGr were presented three times per block . We therefore only have a relatively coarse estimate of the properties of the long-term memory store . Lengthening of inter-reoccurrence intervals was probed by introducing interrupting blocks where only novel patterns were presented ( see Exp . S2A-B in Appendix 1 ) . Memory was largely maintained over ~10 min intervals indicating a very slow long-term decay . In conjunction , the results of Exp . 2 suggested that the short-term memory store is critical for long-term memory formation . Participants were markedly impaired at detecting pattern repetition when the two cycles were separated by a brief series of random tones ( about 2 s ) . Those conditions were also associated with substantially reduced long-term memories for the reoccurring patterns , indicating that immediate reinforcement is critical for the formation of lasting memory traces . These observations point to an integral interplay between a short ( few seconds ) and much longer ( at least a few minutes ) integration in the process of formation of robust , implicit memories for reoccurring arbitrary sound sequences . The persistence of a stable RT advantage 24 hr and 7 weeks after initial exposure demonstrates the establishment of a long-term memory representation , possibly through a process of consolidation involving long-lasting synaptic changes ( Phan et al . , 2017; Redondo and Morris , 2011 ) . It may also be tempting to interpret the resistance to interruption , observed in early stages of memory formation ( Exp . 3 , Exp . S2 in Appendix 1 ) , as a hint that a form of consolidation might have occurred already after a few initial presentations . In animal models , repetitive exposure to sound tokens ( though , notably at a much higher rate than that used here ) has been shown to evoke a process of long-lasting adaptation manifested as sparser firing and increased response specificity . These effects , persisting for hours to days after the initial stimulation , have been observed in primary and secondary auditory areas in song birds ( Caudal Medial Nidopallium; Cazala et al . , 2019; Honda and Okanoya , 1999; Lu and Vicario , 2014; Menyhart et al . , 2015; Takahasi et al . , 2010; Chew et al . , 1996; Soyman and Vicario , 2019 ) and in secondary auditory cortex in ferrets ( Lu et al . , 2018 ) . The hypothesis that similar processes might back the behavioural effects we report is appealing . Agus et al . proposed that mechanisms based on spike-timing-dependent plasticity ( STDP; Markram et al . , 1997; Masquelier et al . , 2008; Masquelier et al . , 2009 ) may be possible neural underpinnings for rapid noise memory formation: repeatedly presented , but relatively temporally confined , spectro-temporal ‘constellations’ within the noise snippets may evoke coincident firing among auditory afferents leading to rapidly emerging selectivity for this feature in subsequent presentations of the same noise burst . Kang et al . , 2017 suggested that including a degree of temporal integration can also account for similar effects observed with temporal patterns ( Kang et al . , 2017; Karmarkar and Buonomano , 2007; Lim et al . , 2017; see also Bi and Poo , 2001 ) . As will be discussed below , the behavioural pattern observed here is consistent with sequential information being stored as short sub-sequences ( n-grams ) , that is without retaining the full 20-item sequence . Therefore , a form of STDP , incorporating an integration time of several hundred milliseconds , may underpin the representation of n-grams and implement their increased weight through reoccurance , thus supporting memory for discrete tone sequences . On a systems level , accumulating evidence suggests that an interaction between auditory cortex and the hippocampus may play a role in memory formation . Previous work has implicated the hippocampus in sensitivity to sensory patterns across rapid time scales ( Aly et al . , 2013; Stachenfeld et al . , 2017; Yonelinas , 2013 ) and specifically in the process of discovering RAN-REG transitions ( Barascud et al . , 2016 ) . There is also some evidence that hints at its possible role in supporting long-term memory for acoustic patterns ( Kumar et al . , 2014 ) . The RT advantage to REGr reflects an implicit memory of sequential structure ( Exp . 1B ) . But what , specifically , is remembered ? Clearly participants did not perfectly memorize the full pattern , in that this would have been associated with much faster RTs ( e . g . as exhibited by the observer with unconstrained memory , Figure 5A ) . Instead , by the end of block 3 , the distribution of RT to REGr shifted leftwards by about four tones , without otherwise changing ( Figure 9A ) . Modelling suggests that this performance is consistent with a statistical-learning effect whereby the participants retained imperfect memory of patterns presented earlier in the experiment . These memories are not strong enough to prompt immediate recognition of a pattern heard in a past trial , but they are sufficiently strong to speed the recognition of that pattern once it begins repeating in the new trial . Similar to other models of statistical learning ( Bröker et al . , 2018; Harrison et al . , 2011; Meyniel et al . , 2016 ) , our memory-constrained PPM model explicitly assumes that listeners represent the unfolding sequences in the form of n-gram sub-sequences of variable length , from which transition probabilities are computed . Previous computational , behavioural and neuroimaging studies Bianco et al . , 2020; Conklin and Witten , 1995; Di Liberto et al . , 2020; Egermann et al . , 2013; Pearce et al . , 2010; Pearce and Wiggins , 2004; Pearce and Wiggins , 2006 demonstrated that PPM successfully generalizes to prediction of musical sequences and effectively accounts for psychophysiological responses to melodies . In particular , PPM provided a good match to brain response latencies evoked by transitions between RAN and REG patterns ( Barascud et al . , 2016; Southwell and Chait , 2018 ) , suggesting that listeners may rely on similar memory representations as those proposed by the model . Here , the memory constrained version of PPM was able to successfully simulate human performance - concretizing how the interplay between short- and long-term decay might give rise to the progressive emergence of a memory trace across presentations . Whether listeners do indeed represent auditory patterns in this way is a matter of ongoing debate ( e . g . Thiessen , 2017 ) . Additional support for an n-gram-like representation is provided in Exp . 4 , which demonstrated that the REGr RT advantage is robust to pattern phase shifts . This finding indicates that REG patterns are not encoded in memory as rigid chunks of sequential items ( Perruchet and Pacton , 2006; Thiessen , 2017 ) , but are instead represented as a transition rule which allows for flexible retrieval . Whilst further empirical evidence is essential to determine the nature of the memory representation , the insight into single-trial level dynamics derived from the present modelling ( Figure 4 ) may be useful for constraining the search for the physiological underpinnings of these phenomena . Furthermore , the model can readily be applied to statistical learning in other modalities ( reviewed by Frost et al . , 2019 ) and even in other species , including songbirds such as finches , known to be capable of statistical learning ( Menyhart et al . , 2015; Takahasi et al . , 2010 ) . A related question pertains to the generalizability of the present model to natural sounds beyond quantized sequences , such as those used here . In order to relate listeners’ performance to a measure of statistical information within unfolding signals , simplifying assumptions are necessary . This includes the presence of a prior stage of category formation which converts a continuous sound into discrete units that form the model’s ‘alphabet’ . Accumulating evidence is indeed revealing that unsupervised segmentation of unfolding sounds into basic elements , perhaps using envelope-based cues , may be an inherent feature of listening ( Ding et al . , 2017; Doelling et al . , 2014; Hickok and Poeppel , 2007; Poeppel , 2003 ) . The short-term memory mechanisms which allow listeners to discover the emergence of repeating structure ( RANREG ) in rapid tone sequences have been demonstrated to operate automatically , even when listeners’ attention is directed away from sound: brain activity recorded from naïve , distracted listeners reveals robust responses to RAN-REG transitions with latencies consistent with those expected from an ideal observer ( Barascud et al . , 2016; Southwell et al . , 2017; Southwell and Chait , 2018 ) . In contrast , in Exp . 5 , we demonstrated that longer term memory trace formation appears to require attentive processing in that there was no evidence for an immediate RT advantage when listeners moved from the exposure blocks , in which patterns were behaviourally irrelevant , to the active detection ( ‘test’ ) block . This suggests that the formation of lasting memories for sound patterns is not fully automatic , or does not immediately translate to behaviour . Whether this is driven by absence of attention per se or other factors is difficult to determine . For example , it is possible that the reduced memory effect when sounds are not behaviourally relevant is driven by decreased arousal or reward , known to substantially modulate learning ( Beste and Dinse , 2013; Braun et al . , 2018; Polley , 2006; Yebra et al . , 2019 ) , and which likely distinguish active detection ( where feedback was provided after each trial ) from passive listening . Importantly , we showed that though implicit memory was not present at the onset of the test block , learning occurred more rapidly in the pre-exposed listeners , hinting at the presence of pre-exposure-related latent traces that may contribute to faster instantiation of representations in memory once the sequences become behaviourally relevant ( Frankland et al . , 2019 ) . Uncovering how memory traces are encoded and preserved by the brain is crucial for understanding subsequent learning operations which drive pattern recognition and generalization . We showed that representations of sporadically reoccurring rapid sound patterns are retained in memory , thus facilitating detection when previously encountered patterns reoccur . In spite of the fact that the patterns were relatively featureless and undistinctive compared to real-world stimuli , this memory was robust , implicit , remarkably resistant to interruption , and persisted over long durations , revealing human listeners’ astonishing sensitivity to reoccurring structure in the auditory environment . Important questions for future work include understanding the neurobiological foundations of these behavioural effects , the limits on the capacity of the memory store ( s ) involved and the factors which might affect subsequent forgetting .
Stimuli were sequences of 50 ms tone-pips of different frequencies generated at a sampling rate of 44 . 1 kHz and gated on and off with 5 ms raised cosine ramps; the total sequence duration was 7 s ( 140 tones ) . Frequencies were drawn from a pool of twenty values logarithmically spaced between 222 and 2000 Hz; 12% steps . The order in which these frequencies were successively distributed defined different conditions that were otherwise identical in their spectral and timing profiles ( see Figure 1 ) . RAN sequences consisted of tone-pips arranged in random order , with the constraint that adjacent tones were not of the same frequency . Each frequency occurred equiprobably across the sequence duration . The RANREG condition contained a transition between a random ( RAN ) , and a regularly repeating pattern ( REG ) : sequences with initially randomly ordered tones changed into regularly repeating cycles of 20 frequencies ( an overall cycle duration of 1000 ms; new on each trial ) . The change occurred between 3000 and 4000 ms after sequence onset such that each RANREG sequence contained between 3 to 4 REG cycles ( only two in Exp . 2 , see below ) . RAN and RANREG conditions were generated anew for each trial and occurred equiprobably . Thus , each trial contained exactly the same frequency ‘building blocks’ , with the same overall distribution , and only varied in the specific order of tone-pips . The inter-trial interval was jittered between 1400 and 1800 ms . Unbeknown to participants , a few different REG patterns ( different for each participant ) reoccurred identically several times within the session ( RANREGr condition ) . Reoccurrences happened three times per block ( every ~3 min ) , and 9–15 times per session , depending on the number of blocks in the specific experiment . Note that the RAN part preceding each REGr was always novel . Reoccurrences were distributed within each block such that they occurred at the beginning ( first third ) , middle and end of each block . Two control conditions were also included within each block: sequences of tones of a fixed frequency ( CONT ) , and sequences with a step change in frequency partway through the trial ( STEP ) . The STEP condition served as a measure of individuals’ reaction time to simple acoustic changes . The RT to STEP was subtracted from the RT to RANREG sequences to obtain a lower bound measure of computation time required to detect the transition . Participants were instructed to respond , by pressing a keyboard button , as soon as possible after detecting a RANREG transition . Feedback about response accuracy and speed was delivered at the end of each trial . Since RT is a key performance measure in these experiments , it was important to motivate the participants to respond as quickly as possible . The feedback was given based on our previous work ( Barascud et al . , 2016 ) , and consisted of a green circle if the response fell within 2200 ms from the regularity onset in the RANREG conditions , or within 300 ms from the change of tone in the STEP condition . For slower RTs , an orange circle ( between 2200 and 2600 ms in the RANREG conditions , and between 300 and 600 ms in the STEP condition ) or a red circle were displayed . It was explained to participants that they should strive to obtain as much ‘green’ or ‘orange’ feedback as possible . The experimental session was delivered in ~8 min blocks , separated by brief breaks . Stimuli were presented with PsychToolBox in MATLAB ( 9 . 2 . 0 , R2017a ) in an acoustically shielded room and at a comfortable listening level ( self-adjusted by each listener ) . We initially ran a pilot experiment ( N = 20 , 16 females , age 23 . 5 ± 2 . 95 years ) which consisted of five consecutive blocks as in Exp . 1A . The effect size for the main effect of condition ( RANREG vs . RANREGr ) was ηp2 = . 48 and ηp2 = . 79 after the first 3 and 5 blocks respectively . Using ηp2 = 0 . 48 for a prospective power calculation ( beta = 0 . 8; alpha = 0 . 05 ) yielded a required sample size of N = 13 . We decided to increase our sample size up to N = 20 to account for possible drop outs due to low accuracy . The research ethics committee of University College London approved the experiment [Project ID Number]: 1490/009 , and written informed consent was obtained from each participant . The transition detection task was performed in three sessions: five blocks on day one ( ‘day1’ ) , one block after 24 hr ( ‘24 hr’ ) and another block after 7 weeks ( ‘7 w’ ) . Each block consisted of 60 stimuli ( ~8 min duration; 3 RANREGr x three reoccurrences per block , 18 RANREG , 27 RAN , 3 STEP , and 3 CONT ) . Feedback about the response accuracy and speed was delivered after each trial . Before starting , a short training block of 12 trials ( with the same conditions as in the main experiment , but no RANREGr ) was performed to acquaint participants with the task . The familiarity task was performed at the end of each session ( day1 , 24 hr , 7 w ) . In these tests the three REGr patterns were randomly intermixed with 18 novel REG sequences . Participants were informed that a ‘handful’ of patterns reoccurred during the just completed session and asked to indicate , by means of a button press , if each presented pattern sounded ‘familiar’ . Participants . Twenty paid individuals ( ten females; average age , 24 . 4 ± 3 . 03 years ) took part in the experiment . Because of poor accuracy ( d’ < 2 after the first block ) , one participant was excluded from the analysis . We were able to test only 14 participants after 7 weeks ( eight females; average age , 24 . 7 ± 3 . 02 years ) . No participant reported hearing difficulties . Participants performed the transition detection task for six consecutive blocks consisting of the same set of stimuli described for Exp . 1A . In the 5th block , each REGr was time reversed . Participants . Twenty paid individuals ( 13 females; average age = 24 . 25 ± 3 . 58 years ) took part in the experiment . No participant reported hearing difficulties . The stimulus set in the initial four blocks contained RANREG and RANREGr stimuli , as before , except they contained only two repeating cycles after the transition . To understand whether immediate repetition is necessary for memory to be formed two further conditions were used: PATinRAN stimuli contained two identical 20 tone patterns embedded amongst random tones ( mean separation of 1 . 7 s; drawn randomly from a range . 5–2 . 9; the second appearance always occurred at the end of the sequence as shown in Figure 3A ) . Similar to REGr , three different PAT were designated as reoccurring across trials ( different for each participant; three reoccurrences per block ) . The embedding RAN sequence and the spacing between the two PAT patterns were randomly set for each reoccurrence . Overall each block contained 82 stimuli ( 36 RAN , 9 RANREG , 9 RANREGr , 9 PATinRAN , 9 PATinRANr , 5 STEP , 5 CONT ) , with ISI between 2 . 4 and 2 . 8 s . Reoccurrences of RANREGr and PATinRANr occurred approximately every 3 . 6 min . Participants were informed of the presence of PATinRAN and RANREG stimuli ( but were naïve about RANREGr and PATinRANr ) and were instructed to indicate , by button press , if they detected the presence of a repeating pattern within the just-heard sequence . Feedback was provided at the end of each trial as in the above experiments , except that in the PATinRAN conditions we delivered a green circle if the response fell within 1200 ms from the second cycle onset , a red circle if the response was slower that 1600 ms , and an orange one if it fell in between . It was explained to participants that they should be fast but prioritise accuracy , given the generally difficult level of the task . In order to quantify any memory effects , in the 5th block ( ‘test’ block ) each of PATinRANr sequences were replaced by sequences with the two cycles set adjacently . We will refer to this condition as RANREGr* . The test block contained 36 RAN , 18 RANREG , 9 RANREGr , 9 RANREGr* , 5 STEP , 5 CONT Stimuli were about 5 . 45 ms long ( ~109 tones ) . Participants . Given the task complexity and expectation for a reduced SNR , we increased the number of participants , a-priori , by 50% relative to the previous experiment . Thirty paid individuals ( twenty females; average age , 24 . 26 ± 3 . 8 years ) took part in the experiment . No participants reported hearing difficulties . This experiment consisted of two days of testing . On the first day participants performed a transition detection task as in Exp . 1A , but two different sets of reoccurring patterns ( REGr1 and REGr2; three different patterns each ) were presented . REGr1 was presented over the first three blocks , and REGr2 over the subsequent three blocks . On day 2 ( after 24 hr ) , participants returned to the lab to perform two test blocks for the two sets of reoccurring regularities , REGr1 and REGr2 ( order counterbalanced across participants ) . Participants . We initially ran 20 participants ( one excluded from analysis ) , but decided to run an additional 10 participants ( +two excluded ) , to increase the SNR for the memory effects observed for the RANREGr1 and RANREGr2 conditions on day two . The results with N = 19 yielded qualitatively similar results ( see Figure 6—figure supplement 1B ) . Thirty-two paid individuals ( twenty females; average age , 24 . 5 ± 3 . 8 years ) took part in the experiment . No participant reported hearing difficulties . Because of poor accuracy ( d’ < 2 after the first block ) , three participants were excluded from the analysis . Participants performed the detection task through four consecutive blocks of 82 stimuli each . The stimulus set included the same conditions as described for Exp . 1A , but with six , instead of three , REGr sequences , each presented three times within a block ( 6 RANREGr x three reoccurrences per block , 18 RANREG , 36 RAN , 5 STEP , and 5 CONT ) . In block 4 , REGr were phase shifted ( see examples in Figure 7A ) . To ensure uniform sampling of possible phase shifts , for each REGr in block 4 , each of the three intra-block presentations was subject to pattern phase shift of 2 to 7 , 8 to 13 , or 14 to 19 tones from the onset of the original pattern . The phase shift was determined independently for each REGr and each intra-block presentation . Stimulus duration was 6 . 5 s , and the transition time was between 3 and 3 . 5 s from the sequence onset . Different REGr patterns reoccurred sparsely ( every ~3 . 4 min ) across trials and blocks . Participants . Twenty paid individuals ( fourteen females; average age , 23 . 5 ± 3 . 2 years ) took part in the experiment . No participant reported hearing difficulties . The experiment consisted of four blocks . The stimulus structure was as in Exp . 1A , except that for the first three blocks participants were instructed to respond to STEP changes only . They received no explanation about the regularity structure of the stimuli , and performed no practice . On the fourth block , they were instructed to detect RANREG transitions in addition to STEP transitions . Each block contained 72 stimuli ( 3 RANREGr x three reoccurrences per block , 18 RANREG , 27 RAN , 9 STEP , and 9 CONT; ISI between 900 and 1300 ) ; the number of STEP and CONT trials was increased relative to that in Experiment 1A due to the task change . As in Exp . 1A , participants performed the familiarity task at the end of the session . Participants . Nineteen paid individuals ( 14 females; average age , 23 . 4 ± 3 . 1 years ) took part in the experiment . No participant reported hearing difficulties . One participant was excluded from the analysis because of poor accuracy ( d’ < 1 ) . In the transition detection task , two indexes of performance were computed: sensitivity ( d’ ) and reaction time ( RT ) . For each participant and each block , d’ was quantified over trials ( collapsed over RANREG and RANREGr ) to give a general measure of sensitivity to the presence of regularities . Responses to RANREG and RANREGr , which occurred after the nominal transition were considered hits; Responses to RAN trials were considered false alarms . Participants who showed d’ < 2 after the first block of the transition detection task were excluded from the analysis . Because Exp . five had only one ‘active’ block and no previous training , we adopted a more lenient exclusion criterion of d’ < 1 . Note that d’ was not available in Exp . 2 because of the intermixed nature of the presentation of RANREG and PATinRAN stimuli . To quantify performance , we therefore focus on hit rates and false alarms . For the purpose of participant exclusion , we computed an overall d’ ( collapsing across conditions ) and set the threshold at d’ < 1 . 5 . Only RTs of correct trials ( hits ) were analysed . In all experiments , RT was defined as the time difference between the onset of the regular pattern ( ‘nominal transition’ in Figure 1 ) and the participant’s button press . However , Exp . 2 contained conditions with non-contiguous pattern presentations . RT was therefore computed from the onset of the second cycle ( as indicated in Figure 3A ) . Across all experiments , RTs which occurred before the transition to the regularity ( see Figure 1; ~1 . 3% of the trials ) were considered to indicate a false positive and excluded from the analysis . To control for individual latency of motor response to a simple acoustic change , RTs were then ‘baselined’ by subtracting the individual mean RT to the STEP transition . Moreover , for each participant and block , the RTs beyond 2 SD from the mean were discarded . To quantify the formation of a memory trace over REGr presentations , RT were averaged to yield a mean RT per condition per subject per block . Therefore , RT to RANREGr were based on nine trials ( 3 REGr x three presentations per block ) . However , to evaluate the immediate presence of a memory trace following certain experimental manipulations ( e . g . in Exp . 2 and 5 ) or when re-testing after 24 hr or 7 weeks ( as in Exp . 1A ) we also analysed RT for each intra-block presentation ( the first , second and third intra-block instance of a REGr pattern; see Figure 2—figure supplement 1; Figure 2—figure supplement 3; Figure 6—figure supplement 1A; Figure 7—figure supplement 1; Appendix 1—figure 1D; Appendix 1—figure 2D-J; Appendix 1—figure 3D ) . To calculate the ‘RT advantage’ for each intra-block presentation , mean RTs of 1st , 2nd or 3rd intra-block presentation ( averaged across the different REGr ) were subtracted from the mean RTs of novel REG which occurred at the beginning ( first third ) , middle or end of each block . Performance was statistically tested with linear analyses of variance ( ANOVA ) implemented in the R environment ( version 0 . 99 . 320 ) using the ‘ezANOVA’ function ( Michael Lawrence , 2016 ) . The analysis of d’ modelled the repeated measures factor block ( 1: N blocks ) . The analysis on RTs modelled the repeated measures factors: condition ( RANREG / RANREGr ) , block ( 1: N blocks ) , and their interaction term . P-values were Greenhouse-Geisser adjusted when sphericity assumptions were violated . Post hoc t-tests were used to test for differences in performance between conditions across blocks or experiments . A Bonferroni correction was applied by multiplying p values by the number of comparisons ( resulting values were capped at 1 . 0 ) . As a benchmark ( see Figure 9D ) across which to compare the effect of various manipulations on the RT advantage ( i . e . , Figure 8D , Appendix 1—figure 2C-G ) , we pooled data from several experiments to obtain a distribution of RT advantage values after each block: Pooled data-block1 , Pooled data-block2 , Pooled data-block3 were formed by pooling block 1 , 2 or 3 data , respectively , from Experiments 1A , 1B , 3 , 4 , S1 , S3 , and pilot experiment identical to Exp . 1 ( total N = 147 ) . Pooled data-block4 was formed by pooling block four data from Experiments 1A , 1B , S1 , S3 and the pilot ( total N = 98 ) , and Pooled data-block5 by pooling block 5 from Experiments 1A , S1 and the pilot ( total N = 58 ) . To obtain distributions of expected RT advantage values , data in each set were subjected to bootstrap resampling ( 1000 iterations ) where , on each iteration , N random participants ( N = number of participants in the experiment under examination ) were drawn from the full pool , and their mean RT advantage ( RANREG- RANREGr ) was computed . This procedure yielded a distribution to which the actual data from the experiment under examination were compared . The p values provided ( i . e . , Figure 8D , Appendix 1—figure 2C-G ) reflect the probability of the measured RT advantage ( red dots in the relevant figures ) relative to the benchmark distribution . The familiarity measurement required participants to categorize the presented patterns into ‘familiar’ ( REGr ) or ‘new’ ( REG ) . Each REGr was presented once only , to avoid learning during the testing session and hence the ‘familiar’ category included only three items ( six in Exp . S1 , see Appendix1 ) . These were presented among a larger set of foils ( 18 in Exp . 1A and Exp . 5 , 36 in Exp . S1 ) . Due to the small number of REGr , standard signal detection approaches are not useable . Instead , we computed the MCC score , which is a measure of the quality of a binary classification , applicable even when classes are of different sizes ( Boughorbel et al . , 2017; Powers , 2007 ) . The coefficient ranges between 1 ( perfect classification ) to −1 ( total misclassification ) and is calculated using the following formula: MCC=TP×TN-FP×FNTP+FPTP+FNTN+FPTN+FN2 . Where TP = number of true positives; TN = number of true negatives; FP = number of false positives; FN = number of false negatives . The MCC scores obtained for each participant in Exp . 1A are shown in Figure 2—figure supplement 2 . Prediction by Partial Matching ( PPM ) is a Markov modelling technique ( Cleary and Witten , 1984 ) that models statistical structure within symbolic sequences by tabulating occurrences of n-grams within a training dataset . PPM is a variable-order Markov model , meaning that it generates predictions by combining n-gram models of different orders; here we use a model combination technique called ‘interpolated smoothing’ ( Bunton , 1996; Bunton , 1997; see also Pearce and Wiggins , 2004; Harrison et al . , 2020; for more details ) . This approach combines the advantages of both the structural specificity afforded by high-order n-gram predictions and the statistical reliability afforded by low-order n-gram predictions . The PPM models used in prior cognitive research Barascud et al . , 2016; Cheung et al . , 2019; Gold et al . , 2019 have a ‘perfect’ memory , in that historic n-gram observations are preserved with the same fidelity as recent events , and are weighted the same in prediction generation . Noting that human memory exhibits clear capacity limitations and recency effects , Harrison et al . , 2020 modified PPM to incorporate a customizable decay kernel , whereby historic n-gram observations are down-weighted as a function of the time elapsed and the consequent n-grams observed since the initial observation . Modelling reaction-time data from a RANREG paradigm similar to Barascud et al . , 2016 , Harrison et al . concluded in favour of a capacity-limited high-fidelity echoic memory buffer followed by a lower-fidelity short-term memory phase with exponential decay . We likewise use an echoic-memory phase and a short-term memory phase in the present work , but add a slower-decaying long-term memory phase in order to capture the long-term learning observed in the present experiment . The modelling aimed to reproduce behavioural performance qualitatively rather quantitatively . Many simplifications are made including that inter-sequence intervals , and breaks between experimental blocks are modelled at a fixed rate of 1 s . We explored various parameter settings for the model , and retained the configuration that best reproduced the observed behavioural patterns in Experiments 1A , 2 , and S2A ( Figure 5 , and Appendix 1—figure 2K ) , which represent the key manipulations of memory duration . The resulting parameters are listed in Table 1; the decay kernel is plotted in Figure 4A . Further implementation details are described in Harrison et al . , 2020 . The model outputs a conditional probability estimate for each tone in each sequence experienced throughout an experiment , which we convert to information content ( the negative log probability in base 2 ) . An implementation of this model is freely available in our open-source R package ‘ppm’ ( https://github . com/pmcharrison/ppm; Harrison , 2020 ) . To identify changes in the information content profile corresponding to the RANDREG transition on a given trial , we use the nonparametric changepoint detection algorithm of Ross et al . , 2011 , which sequentially applies the Mann-Whitney test to identify changes in a time series’ location while controlling for Type I error . Here , the target Type I error rate was set to 1 in 10000 tones . Note that , for simplicity , the change point detection algorithm is free of memory constraints . Human listeners likely use a rougher ( less detailed ) statistical representation for transition detection . | Patterns of sound – such as the noise of footsteps approaching or a person speaking – often provide valuable information . To recognize these patterns , our memory must hold each part of the sound sequence long enough to perceive how they fit together . This ability is necessary in many situations: from discriminating between random noises in the woods to understanding language and appreciating music . Memory traces left by each sound are crucial for discovering new patterns and recognizing patterns we have previously encountered . However , it remained unclear whether sounds that reoccur sporadically can stick in our memory , and under what conditions this happens . To answer this question , Bianco et al . conducted a series of experiments where human volunteers listened to rapid sequences of 20 random tones interspersed with repeated patterns . Participants were asked to press a button as soon as they detected a repeating pattern . Most of the patterns were new but some reoccurred every three minutes or so unbeknownst to the listener . Bianco et al . found that participants became progressively faster at recognizing a repeated pattern each time it reoccurred , gradually forming an enduring memory which lasted at least seven weeks after the initial training . The volunteers did not recognize these retained patterns in other tests suggesting they were unaware of these memories . This suggests that as well as remembering meaningful sounds , like the melody of a song , people can also unknowingly memorize the complex pattern of arbitrary sounds , including ones they rarely encounter . These findings provide new insights into how humans discover and recognize sound patterns which could help treat diseases associated with impaired memory and hearing . More studies are needed to understand what exactly happens in the brain as these memories of sound patterns are created , and whether this also happens for other senses and in other species . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2020 | Long-term implicit memory for sequential auditory patterns in humans |
Haploinsufficiency of the melanocortin-4 receptor , the most common monogenetic obesity syndrome in humans , is associated with a reduction in autonomic tone , bradycardia , and incidence of obesity-associated hypertension . Thus , it has been assumed that melanocortin obesity syndrome may be protective with respect to obesity-associated cardiovascular disease . We show here that absence of the melanocortin-4 receptor ( MC4R ) in mice causes dilated cardiomyopathy , characterized by reduced contractility and increased left ventricular diameter . This cardiomyopathy is independent of obesity as weight matched diet induced obese mice do not display systolic dysfunction . Mc4r cardiomyopathy is characterized by ultrastructural changes in mitochondrial morphology and cardiomyocyte disorganization . Remarkably , testing of myocardial tissue from Mc4r−/− mice exhibited increased ADP stimulated respiratory capacity . However , this increase in respiration correlates with increased reactive oxygen species production – a canonical mediator of tissue damage . Together this study identifies MC4R deletion as a novel and potentially clinically important cause of heart failure .
During the last 30 years , obesity has become a leading cause of morbidity and mortality ( Ogden et al . , 2014; Christakis and Fowler , 2007 ) . Health risks associated with obesity include type 2 diabetes , hypertension , and coronary artery disease . Obesity is also an independent risk factor associated with the development of heart failure ( Kenchaiah et al . , 2002 ) ( Levitan et al . , 2009 ) . Haploinsufficiency of the Mc4r is the most common monogenetic obesity syndrome in man ( Farooqi et al . , 2003 ) and is responsible for 0 . 5–2 . 5% of all early onset morbid obesity ( Stutzmann et al . , 2008 ) making it an important consideration in personalized obesity care . The prevalence of the Mc4r obesity syndrome is a result of its dominant inheritance pattern , and penetrance of approximately 70% ( Tarnow et al . , 2008 ) ( Biebermann et al . , 2003; Tao , 2005 ) . Mc4r is expressed most heavily in the central nervous system where it plays a critical role in energy balance ( Cone , 2005 ) . Mc4r expressing neurons within the hypothalamus receive orexigenic and anorexigenic inputs from arcuate nucleus POMC and AgRP projections ( Balthasar et al . , 2004; Balthasar et al . , 2005 ) and act to maintain energy homeostasis through modulation of both food intake ( Fan et al . , 1997; Huszar et al . , 1997 ) and energy expenditure ( Ste Marie et al . , 2000 ) . A reduction of MC4R signaling , through either genetic or pharmacological means , results in hyperphagia , bradycardia ( Wang et al . , 2017 ) , and reduced blood pressure . Clinical studies have found that heterozygous Mc4r mutations confer protection from obesity-associated hypertension through reduced sympathetic tone ( Sweeney , 2010; Greenfield et al . , 2009; Sayk et al . , 2010 ) . Mice and humans with MC4R mutations also experience hyperinsulinemia that exceeds their degree of adiposity due to the role MC4R in the suppression of insulin release ( Fan et al . , 2000; Mansour et al . , 2010 ) . Furthermore , patients with Mc4r heterozygosity have reduced growth hormone suppression in response to obesity when compared to patients matched with standardized BMI ( Martinelli et al . , 2011 ) . Therefore , while the effects of Mc4r deletion on peripheral vascular resistance may be cardioprotective , other aspects of the Mc4r obesity syndrome , such as hyperinsulinemia , incomplete growth hormone suppression , and altered autonomic tone are potentially cardiotoxic . Since no group to our knowledge has directly examined the effects of MC4R deletion on myocardial function , we chose to examine how Mc4r deletion affects myocardial function in vivo , and further characterized its effects on myocardial energy metabolism ex vivo .
In order to characterize the effects of MC4R on heart function , age- and sex-matched Mc4r−/− , Mc4r+/- and WT mice were serially assessed using echocardiography . An age dependent cardiomyopathy was observed in male Mc4r−/− animals ( Figure 1A–B ) , which included cardiac dilatation and reduced contractility . At 26 weeks of age , a significant reduction in fractional shortening ( FS ) , a measure of myocardial contractility ( Figure 1C ) as well as ejection fraction ( Figure 1D ) could be observed . A significant increase in left ventricular diameter ( LVIDd ) was also observed in Mc4r−/− mice at 26 weeks of age , likely as a compensatory response to reduced contractility ( Figure 1E ) . As has been shown previously ( Stepp et al . , 2013 ) , a reduction in heart rate was also observed in the Mc4r−/− mice at later time points ( Figure 1F ) . No significant change in heart wall thickness ( LVPWd ) was observed indicating the absence of pathological hypertrophy or atrophy ( Figure 1G ) . Female Mc4r−/− mice were also examined with echocardiography . Mc4r−/− female mice displayed a similar phenotype to their male counterparts at 26 weeks ( Figure 1H–I ) . This included a similar reduction of FS ( Figure 1J ) and EF ( Figure 1K ) , an increase in LVIDd ( Figure 1L ) a reduction in heart rate ( Figure 1M ) and no significant reduction in LVPWd at 26 weeks ( Figure 1N ) . ECG rhythm strips of male Mc4r−/− mice also revealed a bradycardic arrhythmia with dropped p waves at the 30 week time point ( Figure 2 ) . Given these observations in male and female mice , we next sought to confirm that the cardiac deficits seen in the Mc4r−/− mouse were specific to the loss of Mc4r and not due to an anomalous background mutation . In order to accomplish this , the Mc4r loxTB mouse , a distinct Mc4r knockout model ( Balthasar et al . , 2005 ) was also examined by echocardiography . LoxTB Mc4r knockout animals displayed similar signs of myocardial dysfunction ( Figure 3A–B ) including reduced fractional shortening and ejection fraction with a trend towards increased LVIDd ( Figure 3C–E ) but did not have a significantly reduced heart rate , or heart wall thinning ( Figure 3F–G ) . Together , these results demonstrate that Mc4r deletion leads to a progressive cardiomyopathy . Both pre-clinical and epidemiological data support a causal role of hyperglycemia and insulin resistance in the development of heart failure ( Bugger and Abel , 2014 ) . Similarly , investigators have observed variable degrees of cardiomyopathy in mouse models of diet induced obesity ( DIO ) ( Battiprolu et al . , 2012; Brainard et al . , 2013; Calligaris et al . , 2013; Heydemann , 2016; Raher et al . , 2008; Wang et al . , 2017; Christoffersen et al . , 2003; Semeniuk et al . , 2002 ) . The Mc4r is not expressed in adult mouse myocardial tissue ( Figure 4A–C ) . Therefore , it is plausible that certain effects of the cardiomyopathy in Mc4r−/− mice are due directly to insulin resistance and/or negative metabolic effects of obesity . In order to address this possibility , a cohort of age matched diet induced obesity mice was generated and evaluated by echocardiography . Age matched male C57BL/6J wild type mice were placed on a 60% HFD for 35 weeks until they reached the same body weight as the Mc4r−/− animals ( Figure 5A ) . Despite their similar body weights , DIO animals and Mc4r−/− animals had distinct body compositions , as previously described . When compared to DIO mice , Mc4r−/− mice accumulate more lean mass ( Figure 5B–C ) while DIO animals preferentially accumulate fat mass ( Figure 5D–E ) . Despite these differences in body composition , Mc4r−/− and DIO animals had similar glucose tolerance in a glucose tolerance test when dosed in proportion to their lean mass ( Figure 5F–G ) . Furthermore , there was no difference in myocardial insulin sensitivity between DIO and Mc4r−/− mice ( Figure 5H ) Thus , this cohort was appropriately matched for body weight and glucose tolerance . Mc4r−/− animals exhibited reduced cardiac contractility , while DIO and WT controls did not ( Figure 5I–J ) . Mc4r knockouts had significantly lower FS and EF ( Figure 5K–L ) and significantly higher LVIDd ( Figure 5M ) while DIO and WT controls did not . In this cohort , Mc4r−/− males also displayed a trend towards a reduction in HR and LVPWd when compared to the DIOs ( Figure 5N–O ) however this difference did not reach statistical significance . These data demonstrate that the cardiomyopathy seen in Mc4r−/− is caused by the loss of MC4R function , rather than a secondary effect of the obesity caused by loss of the MC4R . Mc4r−/− myocardium was examined to determine if the tissue displayed signs of heart failure , such as interstitial fibrosis or lipid deposition . Mc4r−/− hearts appeared grossly larger ( Figure 6A ) and displayed a significantly higher cardiac wet weight when compared to WT controls at 30 weeks of age ( Figure 6B ) . At this time point there was no difference in the heart weight to lean body mass ( HW/LBM ) ratio ( Figure 6C ) , likely due to the generalized increase in lean mass seen in Mc4r−/− mice . However , the heart weight of Mc4r−/− mice was significantly higher than that of age matched DIO animals ( Figure 6D ) . Histological analysis of Mc4r−/− tissue by H&E stain was unremarkable and did not reveal signs of an overt pathological insult such as anoxia , inflammatory infiltration or fibrosis ( Figure 6E ) . Examination of H&E sections at 500 μm intervals further revealed that Mc4r−/− mice did not displayed myocyte hypertrophy ( Figure 6F ) . Similarly , lipid deposition was not observed by Oil Red O staining ( Figure 6G ) . However , transmission electron microscopy ( TEM ) showed ultrastructural signs of heart failure . TEM imaging of myocardium in Mc4r−/− mice revealed mitochondrial pleomorphy and cardiomyocyte dropout ( Figure 6H–I ) . Based on these findings , we hypothesized that Mc4r−/− animals were experiencing mitochondrial dysfunction . In order to determine if Mc4r−/− mice had alterations in mitochondrial number , tissue samples were examined for mitochondrial DNA content using qPCR . With this assay , there was no difference in mitochondrial DNA content relative to genomic DNA content in male Mc4r−/− myocardium ( Figure 7A ) . Accordingly , there was no difference in the abundance of electron transport chain ( ETC ) complex I-IV proteins or ATP synthase by western blot ( Figure 7B–C ) . Based on these results and the abnormal mitochondrial morphology on TEM , we then sought to characterize Mc4r−/− myocardial mitochondrial function in situ using high-resolution respirometry of saponin permeabilized left ventricle fibers . Using the protocol described in Figure 7D , no difference in the O2 consumption between genotypes ( Figure 7E ) was observed upon exposure to glutamate and malate in the absence of ADP ( complex I substrates , state IV respiration ) . Similarly , addition of palmitoylcarnitine in the absence of ADP ( Fatty Acid Substrate , state IV respiration ) did not result in any difference in O2 consumption . When ADP was added to this reaction ( Complex I substrates , state III respiration ) there was a trend towards Mc4r−/− myocardium consuming more O2 than control tissue . The increase in state III respiration became more pronounced in the presence of medium chain fatty acid substrates ( L-Octanoylcarnitine ) , as well as in the presence of succinate ( complex II substrate , state III mediated respiration ) . This increase in ETC capacity persisted when the ionophore CCCP was titrated into the reaction . Importantly , the fold increase upon ADP exposure was not different between samples indicating a reliable tissue preparation ( Figure 7F ) . This demonstrates that 30-week-old Mc4r−/− myocardium displayed a 2-fold increase in total respiratory capacity ( Figure 7G ) without a subsequent increase in tissue ATP content ( Figure 7H ) . These findings are in contrast to what is generally seen in dilated cardiomyopathy and right heart failure ( Talati et al . , 2016 ) but more closely mimics what has been observed in hypertrophic cardiomyopathy ( Rosca et al . , 2013 ) . In order to see if the mitochondrial phenotype precedes heart failure as well as the Mc4r−/− obesity phenotype , the O2 consuming capacity of young lean Mc4r−/− myocardium was then analyzed using high-resolution respirometry . Similar to the older cohort , young lean Mc4r−/− animals display increased ADP dependent respiration and ETC capacity without any changes in ADP independent respiration ( Figure 8A ) . Similar to the older animals , there was no difference in ADP fold change between Mc4r−/− and WT cardiac tissue ( Figure 8B ) . Overall , lean Mc4r−/− myocardial tissue displayed a 1 . 5 increase in O2 consuming capacity compared to WT controls ( Figure 8C ) . In order to determine if this change is specific to heart tissue , permeabilized red gastrocnemius muscle fibers were then examined with respirometry . Unlike the myocardial tissue , skeletal muscle O2 consumption capacity in lean Mc4r−/− animals was not different from controls with respect to both ADP dependent and independent respiration ( Figure 8D–F ) . Furthermore , no significant difference in ETC complex proteins was detected in muscle fibers ( Figure 9A–B ) . Thus , the observed increase in myocardial oxygen consumption in Mc4r−/− mice is specific to the myocardium , independent of obesity , and precedes the development of heart failure . Based on these observations , we then sought to understand how an increase in O2 consumption without an increase in ATP content might be contributing to the observed defect in myocardial contractility . One possible explanation is the excessive production of reactive oxygen species ( ROS ) . Formed as a byproduct of normal aerobic metabolism , ROS play both physiological and pathophysiological roles throughout the body . However , excessive ROS production is known to contribute to heart failure through irreversible modifications of cellular lipids , proteins and DNA ( Giordano , 2005 ) . In order to study ROS levels in Mc4r−/− myocardium , tissue lysates were examined using the 2’ , 7’ –dichlorofluorescein diacetate ( DCFDA ) oxidation assay . When compared to WT tissue , an increase in oxidized 2’ , 7’ –dichlorofluorescein ( DCF ) was observed in both young Mc4r−/− ( Figure 10A ) and old Mc4r−/− ( Figure 10B ) myocardium . Together with the respirometry data ( Figure 7D–F , Figure 8A–C ) and ATP measurements ( Figure 7G ) this finding suggests that increased O2 consumption in myocytes of Mc4r−/− mice does not lead to increased ATP production but rather the production of ROS . Since ROS can be both physiological and pathophysiological , pathological intermediates and end products of ROS were then examined . While no significant increase in malondialdehyde , an intermediate of lipid peroxidation , was found ( Figure 10C ) , a significant increase in 4-hydroxynonenal protein adducts was observed ( Figure 10D ) . 4-HNE adducts result from irreversible lipid-protein covalent bonds caused by excessive ROS , and have been shown to be a causative mechanism for ROS mediated tissue damage ( Mali and Palaniyandi , 2014 ) . After identifying ROS as a link between increased O2 consumption and tissue damage , we next sought to understand how loss of MC4R leads to this pathological insult . MC4R is primarily expressed in the central and peripheral nervous system . Previous studies have been unable to detect Mc4r expression in mouse myocardium . As described above , qRT-PCR was used to characterize the expression pattern of Mc4r . Similar to previous studies , a nearly 250 fold enrichment of Mc4r mRNA was observed in WT hypothalamus and brain stem versus hypothalamus or brainstem from Mc4r−/− tissue ( Figure 4 ) . However , no significant fold change in Mc4r mRNA was observed in the left ventricle between WT and Mc4r−/− mice ( Figure 4B ) . Furthermore , we did not detect myocardial GFP expression in Mc4r-Sapp animals ( Liu et al . , 2003 ) in either the atrial or ventricular tissue ( Figure 4C ) . Based on these results , loss of Mc4r appears to cause cardiomyopathy through an indirect mechanism . In order to determine a signaling pathway that facilitates this indirect mechanism , Mc4r−/− myocardium of 30 week old mice was compared to age matched control myocardium using RNA-seq . Similar to the qRT-PCR and western blot data , Mc4r expression in the myocardium was undetectable . However , 247 transcripts that were significantly different from WT controls ( Figure 11A ) were identified . Differential expression of select genes was confirmed using qRT-PCR including increased expression of Myl7 and reduced expression of Ppargc1a ( Figure 11B ) . In order to understand the significance of these gene changes , gene set enrichment analysis ( GSEA ) was performed to identify a known stimulus that promotes a similar transcriptional change . GSEA revealed that Mc4r−/− gene changes in the myocardium resemble that seen following doxorubicin treatment — a known inducer of ROS and heart failure ( Figure 11C ) . Pathway ontology analysis of all significantly different transcripts was then used to identify a potential mechanism responsible for heart failure . This analysis revealed a contribution 45 known signaling pathways and protein classes ( Figure 11D ) including growth factor signaling , GPCR signaling , cytoskeleton regulation , glycolysis and inflammation . Thus , while the pathway changes associated with Mc4r deficiency appear to affect multiple nodes of cardiac tissue function , these changes appear to induce a tissue phenotype that is consistent with oxidative stress . The data described above suggest that MC4R may somehow play a protective role in preventing cardiac ROS stress . In order to develop a model system for testing this , we decided to study anthracycline-induced cardiotoxicity in a condition in which MC4R signaling could still be modulated . Mc4r+/-animals and their WT siblings were injected with a sub-cardiotoxic dose of DOX ( 2 × 5 mg/kg IP injections ) ( Figure 12A ) . Body weight measurements reveal that DOX injected Mc4r+/-mice lost nearly twice as much weight as their DOX injected WT controls ( Figure 12B ) , suggesting that even partial reduction of MC4R signaling may increase sensitivity to cardiotoxic insults .
In this report , we describe the development of cardiomyopathy in Mc4r−/− mice . In this mouse model of obesity , we observed a progressive decline in contractility as well as an increase in cardiac chamber size . This decline in cardiac function is found in Mc4r−/− animals regardless of animal sex or knockout strategy but is absent in weight matched , diet induced obese mice . Using transmission electron micrographs , we observed grossly disorganized myofibers and pleomorphic mitochondria in Mc4r−/− myocardium . Subsequent functional testing of myocardium revealed a nearly 2-fold increase in O2 consuming capacity in 30-week-old Mc4r−/− myocardium as well as a 2-fold increase in ROS . Studies in young lean Mc4r−/− myocardium revealed similar findings with a 1 . 5-fold increase in O2 consuming capacity and a 1 . 5-fold increase in ROS . We did not detect any change in skeletal muscle O2 consuming capacity indicating that this change is specific to the myocardium . The underlying pathology appears to be due to an indirect mechanism as we were unable to detect Mc4r expression in atrial or ventricular tissue ( Figure 4A–C ) . Subsequent in silico analysis of RNAseq revealed gene set similarities in MC4R−/− myocardium to those seen in doxorubicin treated myocardial cells ( Figure 11 ) . Based on these findings , Mc4r+/-and WT mice were treated with a low dose doxorubicin treatment to see if a partial reduction of MC4R signaling might cause an increased sensitivity to a cardiotoxic challenge . These studies revealed that Mc4r+/-animals were more sensitive to doxorubicin than their WT controls ( Figure 12 ) , as indicated by an increased cachexigenic response . This was quite surprising , because inhibition of MC4R signaling is well documented to reduce the response to a variety of cachexigenic challenges ( Steinman and DeBoer , 2013 ) . This finding provides a potential model system for further studies of the role of MC4R signaling in cardiomyopathy , in which the level of MC4R signaling can be modulated . We are currently testing cardiac function in Mc4r+/- mice following doxorubicin treatment . Homozygous loss of MC4R function , and homozygous loss of proopiomelanocortin gene function , a preprohormone precursor of the endogenous MC4R agonist have both been reported in patients with early onset obesity ( Farooqi et al . , 2003; Krude et al . , 1998 ) . Based on our findings , these patients , though rare , should be followed for cardiomyopathy . While most of these studies were conducted on Mc4r−/− mice , it is important to note that a trend towards reduced contractility was seen in Mc4r+/-mice , as is demonstrated for fractional shortening and left ventricular internal dimension ( Figure 1C–E ) . Heterozygous hypomorphic/null alleles of the Mc4r are common , appearing in up to 1/1500 individuals ( Wang et al . , 2017 ) . Since most phenotypes characterized in Mc4r+/-mice have translated to Mc4r+/-patients ( Greenfield et al . , 2009; van der Klaauw et al . , 2016 ) , patients with heterozygous Mc4r mutations should also be followed as our studies suggest they may have an increased risk for the development of heart failure . Patients with dominant negative Mc4r mutations should also be identified and followed as they will more closely reflect the Mc4r−/− syndrome ( Tarnow et al . , 2008 ) ( Biebermann et al . , 2003; Tao , 2005 ) . Our finding may thus also have clinical relevance in identifying the etiological factor in certain patients with obesity associated cardiomyopathy . Additionally , our GSEA and doxorubicin treatment findings suggest that common mutations in the MC4R may be an important risk factor for increased sensitivity to cachexia , or possibly even cardiomyopathies induced by cardiotoxic drugs , such as doxorubicin ( Singal and Iliskovic , 1998 ) . Additional research will need to be conducted to determine the mechanism by which Mc4r deletion causes cardiomyopathy . However , the data shown here provides some important clues . First , since we were unable to observe Mc4r expression in adult mouse cardiac tissue , MC4R is likely acting indirectly . One possibility would be an early role for MC4R in cardiomyocyte development; developmentally restricted expression of the Mc4r has been observed in E14 through E18 rat heart , for example ( Mountjoy et al . , 2003 ) . Since we observed mitochondrial defects by respirometry in young lean Mc4r−/− mice ( Figure 8A–C ) not yet exhibiting significant cardiac dysfunction , it is possible that defective development and/or regulation of cardiomyocyte function contributes to the defective cardiac function seen by 26 weeks of age . Absent a role for early developmental expression in mouse cardiomyocytes , the prominent expression of the Mc4r in the central and peripheral nervous system implicates an autonomic mechanism . However , endocrinological mechanisms such as the hyperinsulinemia or abnormal obesity associated growth factor suppression may also need to be considered . Collectively , these studies characterize a novel form of heart failure with clinical importance and raise the need for further examination of how Mc4r deletion affects myocardial function in humans .
All mouse experiments with approval from the Vanderbilt animal care and use committee . All mouse lines were maintained on a C57B6/J background and bred using a het-het mating strategy . Mice were maintained on a 12 hr light-dark cycle and housed at 25°C . Unless other wise noted , mice were fed a chow diet ( Lab Diet; St . Louis , MO; S-5LOD - 13 . 5 kcal% fat , 32 . 98 kcal% Protein , 56 . 7 kcal% Carbohydrate ) following weaning at 3 weeks of age . For high fat diet studies ( Research Diets; New Brunswick , NJ; D12492 - 60 kcal% fat , 20 kcal% Protein , 20 kcal% Carbohydrate ) , food was administered starting at 4 weeks of age and continued throughout the study . For all post mortem studies , mice were deeply anesthetized with 5 mg/kg tribromoethanol and then sacrificed by decapitation . Tissues were rapidly collected and snap frozen in liquid nitrogen . All mouse lines were maintained on a C57BL/6J background with yearly backcrosses to wild type C57 mice ( Jackson Laboratory; Sacramento , California - Jax Stock No: 000664 ) . Mc4r−/− mice ( For Details see: Huszar et al . 199719 ) loxTB Mc4r mice ( Jax Stock No: 006414 ) Mc4r-tau-Sapphire ( Jax Stock No: 008323 ) Echocardiography was performed using the VEVO®2100 digital ultrasound system ( Visual Sonics; Toronto , Ontario ) . Studies were performed using the MS400 18–38 MHz transducer . Mice were placed on a heated platform in the supine position and given oxygen throughout the procedure . Fur was then shaved and the transducer oriented to obtain a parasternal long axis image in Brightness-Mode ( B-Mode ) . After obtaining this image , the transducer was rotated to obtain a short axis image at the level of the mid-papillary muscle . Simultaneous one lead ECGs were preformed on the heated platform . Once in position , Motion-Mode ( M-Mode ) images were taken for further processing . M-mode images were then processed as previously described using the Visual Sonics Software ver2 . 2 . LVIDd , %FS and HR measurements were made in a blinded manner using the LV trace function while LVPWd was obtained using the ruler function . After sacrifice , samples were rapidly placed in ice-cold MiR05 buffer ( 0 . 5 mM EGTA , 3 mM MgCl2 6 H2O , 60 mM Lactobionic acid , 20 mM Taurine , 10 mM KH2PO4 , 20 mM HEPES , 110 mM D-Sucrose , 1 g/L Essentially Fatty Acid Free BSA ) with an additional 2 mM EGTA to chelate extracellular calcium . Samples were permeabilized as previously described ( Talati et al . , 2016 ) . Briefly , samples were separated into fiber bundles using forceps under a dissecting scope . Samples were then moved into a MiR05 +2 mM EGTA +50 μg/mL saponin solution and incubated for 30 min on ice . Samples were then washed twice by incubation in MiR05 buffer with 2 mM EGTA for 15 min . Samples were then blotted dry and weighed so that a 2–4 mg sample was obtained . This sample was then placed in a pre-equilibrated chamber of an O2k Oxygraph ( Oroboros Instruments; Innsbruck , Austria ) that contained MiR05 buffer without additional 2 mM EGTA . Chambers were then closed and O2 consumption rate was measured in response to sequential additions of 10 mM glutamate and 2 mM malate , 0 . 05 mM Palmitoylcarnitine , 2 . 5 mM ADP , 0 . 25 mM Octanoylcarnitine , 10 mM Succinate and 500 nM CCCP . Samples were fixed with 2 . 5% glutaraldehyde in a 0 . 1M sodium cacodylate buffer . Following fixation , the samples were washed in 0 . 1M sodium cacodylate buffer . After washing , the samples were postfixed with 1% osmium tetroxide in 0 . 1M cacodylate buffer and then further washed in a 0 . 1M cacodylate buffer . The samples were then dehydrated through a graded series of ethanols and infiltrated with epoxy resin . The samples were oriented and the epoxy resin cured in flat-embed molds . Thick sections of the embedded tissue were cut , stained with 1% Toluidine blue , and reviewed to determine location . When the correct structure was identified in thick sections , a region of interest was selected for thin sections . Thin sections were stained with 2% aqueous uranyl acetate and Reynolds' lead citrate . Samples were viewed using a FEI Tencai T-12 electron microscope ( FEI Tencai; Hillsboro , Oregon ) operating at 100 keV . Mouse genotypes were blinded during imaging to avoid sampling bias . Myofiber images were taken from three random EM grids per sample . A trained pathologist subsequently analyzed images in a blinded manner . RNA was isolated using TRIzol® ( Thermo Fisher Scientific; Waltham , Massachusetts ) . 1 mL of Trizol was added to 10 mg tissue samples from the posterior wall of the left ventricle . Samples were homogenized using the TissueLyser II ( Qiagen; Venlo , Netherlands ) at max speed for 3 min . Samples were then centrifuged for 10 min at 4°C at 10 , 000 g to eliminate debris . After a 5 min incubation at room temperature , 200 μL of chloroform was added and the samples were shaken vigorously . After an additional 3 min incubation at room temperature , samples were centrifuged at 4°C and 12 , 000 g to separate phase layers . The upper phase was then mixed with 100% EtOH and added directly onto RNeasy ( Qiagen ) columns . Columns were used as instructed by the manufacturer and included an on column DNase ( Quiagen ) . Total RNA quality was assessed using the 2100 Bioanalyzer ( Agilent; Santa Clara , California ) . 200 ng of DNase-treated total RNA with a RNA integrity number greater than seven was used to generate polyA-enriched mRNA libraries using KAPA Stranded mRNA sample kits with indexed adaptors ( Roche; Basel , Switzerland ) . Library quality was assessed using the 2100 Bioanalyzer ( Agilent ) and libraries were quantitated using KAPA Library Quantification Kits ( Roche ) . Pooled libraries were subjected to 75 bp paired-end sequencing according to the manufacturer’s protocol ( HiSeq3000; Illumina; San Diego , California ) . Bcl2fastq2 Conversion Software ( Illumina; San Diego , California ) was used to generate de-multiplexed Fastq files . Read quality was checked using FastQCv0 . 11 . 5 ( Babraham Institute , Cambridge , UK ) . Fastq data files were then imported into Galaxy ( Afgan et al . , 2016 ) and converted to a fastsanger file . A quality control was run and reads were then mapped to the mm10 mouse reference genome with Tophat v2 . 1 . 0 ( Trapnell et al . , 2012 ) . Reads were then assembled into transcripts using Cufflinks v2 . 2 . 1 and merged using Cuffmerge v2 . 2 . 1 . RMPK values were then quantified using CuffQuant v2 . 2 . 1 . Cuffnorm v2 . 2 . 1 . 1 was then used to normalize counts for transcript length . Differential expression of transcripts was determined using the Cuffdiff v2 . 2 . 1 . 3 program and filtered for significance q > 0 . 05 . Once differential gene expression was determined , Gene Set Enrichment Analysis ( GSEA; Broad Institute; Cambridge , Massachusetts ) was used for pathway analysis and to determine enriched gene sets as previously described ( Subramanian et al . , 2005 ) ( Mootha et al . , 2003 ) . Pathway analysis was conducted using PANTHER Pathway analysis on significantly different transcripts identified by RNA-seq ( Mi and Thomas , 2009 ) . cDNA was generated with iScriptTM ( BioRad; Hercules , California ) according to the manufacturer’s instructions . qRT-PCR was performed using POWER Syber master mix ( Thermo Fisher Scientific ) with the following primers: mMc4r F CccggacggaggatgctatmMc4r R TCGCCACGATCACTAGAATGTmPpargc1a F GCCGTGACCACTGACAACGAGGCmPpargc1a R GCCTCCTGAGGGGGAGGGGTGCmMyl4 F CGGACTCCAACGGGAGAGATmMyl4 R GCTCCTTGTTGCGGGAGATmH19 F GTACCCACCTGTCGTCCmH19 R GTCCACGAGACCAATGACTGmCytb F GTCCACGAGACCAATGACTGmCytb R ACTGAGAAGCCCCCTCAAAT All PCR was performed on using standard cycling conditions on a Quantstudio 12 k Flex Real Time PCR system ( Thermo Fisher Scientific ) . Data was analyzed using the ΔΔCT method . All statistics were performed by comparing ΔCT values between groups and plotted as Fold Change ± SEM ( 2ΔΔCT ) . Tissue samples were snap frozen in Liquid N2 and then transferred to −80°C . Tissue was lysed in Radio Immune Assay Buffer ( 50 mM Tris , 150 mM NaCl , 1% Triton X-100 , 0 . 5% Na-Deoxycholate , 0 . 1% SDS , cOmpleteTM EDTA-Free Protease Inhibitor ( Roche ) Phos Stop ( Roche ) using a tissue homogenizer ( Pro-Scientific 200; Oxford , Connecticut ) . Following lysis , protein samples were kept on ice for 30 min , spun at max speed for 30 min and then quantified by Bradford assay ( BioRad; Hercules , California ) . Sample protein content was adjusted to 2 mg/mL and then diluted with 2x Laemmli buffer with 2% BME ( 1% final ) . Sampled were then run on a Mini-Protean gel , transferred to PVDF membranes , blocked for 1 hr in TBS-T 5%BSA solution and incubated with respective antibodies overnight ( Total OX-Phos Rodent WB Cocktail , 1:1000; Abcam; Cambridge , United Kingdom ) ( GAPDH , 1:5000; Cell Signaling; Boston , Massachusetts ) ( GFP , 1:1000; Abcam ) . After washes in TBST , membranes were incubated with 2O antibodies conjugated ( 1:10 , 000; Promega; Madison , Wisconsin ) with HRP for one hour . After this incubation , membranes were again washed with TBST , exposed to ECL and imaged ( ChemiDoc MP; BioRad; Hercules , California ) . Glucose tolerance testing was performed as previously described . 1 week prior to GTT , body composition was obtained . Mice were then habituated to handling for three consecutive sessions . Following habituation , mice were fasted for 4 hr from 2pm-6pm . A basal glucose reading was obtained and mice were then injected with a 2 mg/kg lean mass dose of glucose in PBS . Glucose readings were then obtained at 15 , 30 , 60 , and 120 min following injection . Mouse body composition including lean and fat mass was obtained by NMR ( mq10 Minispec; Bruker; Billericia , Massachusetts ) All ROS assays were conducted according to the manufacturer’s directions . For ROS ( OxiSelectTMIn Vitro ROS/RNS Assay Kit; Cell Biolabs , Inc . ; San Diego , California ) assays were conducted on tissue lysates from frozen tissue samples . Samples were weighted ( 10 mg/sample ) , homogenized in PBS and centrifuged ( 10 , 000 g , 5 min , 4°C ) . Following sample preparation , assays were conducted according to the manufacturer’s directions . MDA content was obtained using MDA lipid peroxidation assay ( Abcam ) according to the manufacturer’s directions . For 4-HNE , the OxiSelectTM 4-HNE Assay Kit ( Cell Biolabs , Inc . ) was used . Samples were homogenized in RIPA buffer with inhibitors and analyzed according to the manufacturer’s directions . Fresh tissue was isolated from posterior left ventricular wall , weighted and placed into a 2N perchloric acid solution , homogenized and then incubated on ice for 30 min . Samples were then centrifuged ( 13 , 000 g for 2 min at 4°C ) and 100 μL of supernatant was added to 500 μL of assay buffer . Samples were then neutralized with 2M KOH , vortexed and centrifuged ( 13 , 000 g for 15 min at 4°C ) to remove PCA . The resulting supernatant was then used in the Fluorometric ATP Assay Kit ( Abcam ) according to the manufacturer’s instructions . Doxorubicin ( Cayman Chemicals #15007 ) was solubilized in isotonic saline at a concentration of 1 mg/ml . 28 week old Mc4r+/-and WT mice received two intraperitoneal injections of 5 mg/kg doxorubicin or vehicle separated by 7 days ( cumulative dose of 10 mg/kg ) . Body weights were measured prior to injection , and at times indicated . Sample size estimation for echocardiography studies was conducted using the power equation ( α <0 . 05 , β = 0 . 1 , Δμ = 25% σ = 5 ) . Remaining sample sizes were estimated based on previous publications . All statistical tests were conducted on the GraphPad Prism 6 software ( Scientific Software; La Jolla , California ) . Data is presented as mean ±standard error of the mean . All data with p<0 . 05 was considered statistically significant . Statistical nomenclature: * OR ^=p < 0 . 05; **p<0 . 01; ***p<0 . 001; ****p<0 . 0001 . | Mutations in the gene that encodes a protein called the melanocortin-4 receptor are the most common genetic cause of early onset obesity in children . These mutations occur in about 1 in 1 , 500 people . The melanocortin-4 receptor is mostly found in the brain where it helps to balance how much a person eats with how many calories they burn . A mutation in just one of the two copies of the gene a person gets from their parents is enough to cause severe obesity . Mice that have been genetically engineered to lack this gene develop all the same symptoms as humans with the mutation . These symptoms include early onset obesity , a slower than normal heart rate , and reduced activity in the nerves that communicate with many body tissues including the gut . Patients with this syndrome are less likely to develop obesity-linked high blood pressure , which could be considered protective from some of the ill effects of excess weight . As a result , studying the animal model of the syndrome may help scientists better understand why mutations in the gene for the melanocortin-4 receptor cause obesity and how to better care for people with these mutations . Now , Litt et al . show that , contrary to expectations , mice lacking the gene for the melanocortin-4 receptor have a higher risk of heart failure than normal mice . An ultrasound scanner showed that the left side of the heart in the mice without the melanocortin-4 receptor becomes progressively larger and weaker . This reduces the heart’s ability to pump blood . Additionally , Litt et al . showed that the energy-producing structures within cells , called mitochondria , are defective in the heart cells of these mice . These defects cause the mitochondria to work harder and produce more harmful byproducts . The mitochondria in the animal’s muscles , however , appear normal . Further experiments showed that the genes active in the hearts of the mice lacking melanocortin-4 receptors are similar to genes active in heart cells treated with doxorubicin , a cancer drug that is toxic to the heart . This drug is known to cause heart failure in some people . The experiments suggest that physicians should watch for signs of heart failure in people who have mutations that affect their melanocortin-4 receptors . Mice with one good copy of the gene did not have signs of heart failure , but they appeared more sensitive to the toxic affects of doxorubicin . These findings suggest that clinical studies are needed to determine if there are potential heart problems or drug sensitivities in patients with mutations that affect the melanocortin-4 receptors . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"medicine"
] | 2017 | Loss of the melanocortin-4 receptor in mice causes dilated cardiomyopathy |
The ESCRT machinery along with the AAA+ ATPase Vps4 drive membrane scission for trafficking into multivesicular bodies in the endocytic pathway and for the topologically related processes of viral budding and cytokinesis , but how they accomplish this remains unclear . Using deep-etch electron microscopy , we find that endogenous ESCRT-III filaments stabilized by depleting cells of Vps4 create uniform membrane-deforming conical spirals which are assemblies of specific ESCRT-III heteropolymers . To explore functional roles for ESCRT-III filaments , we examine HIV-1 Gag-mediated budding of virus-like particles and find that depleting Vps4 traps ESCRT-III filaments around nascent Gag assemblies . Interpolating between the observed structures suggests a new role for Vps4 in separating ESCRT-III from Gag or other cargo to allow centripetal growth of a neck constricting ESCRT-III spiral .
Vesicle formation requires machinery to identify cargo , deform the membrane and release a vesicle . How cytoplasmic proteins cooperate to form vesicles that mediate transport between organelles is well defined . Less well understood is how cytoplasmic proteins drive the topologically reverse process of creating intralumenal vesicles ( ILVs ) in multivesicular bodies ( MVBs ) . Studies over the past decade have shown that a set of over 30 proteins are responsible for ILV formation and are also required for the topologically similar processes of viral budding and cytokinesis ( Hurley , 2010; Hanson and Cashikar , 2012; Sundquist and Krausslich , 2012; Henne et al . , 2013; McCullough et al . , 2013 ) . Most are components of ESCRT ( endosomal sorting complex required for transport ) complexes . The multisubunit ESCRT-0 , -I , and -II complexes are thought to recognize cargo and initiate vesicle formation , while ESCRT-III proteins and the AAA+ ATPase Vps4 that acts on them promote membrane remodeling and scission both on MVBs and the plasma membrane . A subject of much current interest is how , exactly , ESCRT-III and Vps4 drive these events . Seven ESCRT-III proteins in yeast and 12 in humans are structurally similar to each other but are not redundant for pathway function ( Morita et al . , 2010 , 2011; Peel et al . , 2011 ) . ESCRT-III proteins cycle between an inactive or closed monomeric state in the cytoplasm and an activated or open state in which they polymerize into filaments on the membrane ( Babst et al . , 2002; Lin et al . , 2005; Zamborlini et al . , 2006; Shim et al . , 2007; Hanson et al . , 2008 ) . ESCRT-II initiates ESCRT-III assembly on the MVB by binding to Vps20/CHMP6 ( Babst et al . , 2002; Teis et al . , 2010 ) while virally encoded sequence motifs bind ESCRT-I and/or Alix which in turn promote ESCRT-III assembly during viral budding ( von Schwedler et al . , 2003; Strack et al . , 2003 ) . Once nucleated , ESCRT-III polymers grow by recruiting other subunits ( Babst et al . , 2002; Shim et al . , 2007; Teis et al . , 2008; Saksena et al . , 2009 ) . Reconstituting ILV formation on giant unilamellar vesicles established that ESCRT-III proteins alone can promote vesicle release ( Wollert et al . , 2009; Wollert and Hurley , 2010 ) although studies in yeast suggest that they normally do this in cooperation with Vps4 ( Nickerson et al . , 2010; Adell et al . , 2014 ) . In addition to Vps20/CHMP6 as a nucleator , specific roles attributed to individual ESCRT-III proteins include cargo confinement by Snf7/CHMP4 ( Teis et al . , 2008 ) , polymer capping by Vps24/CHMP3 ( Saksena et al . , 2009 ) , and Vps4 recruitment by Vps2/CHMP2 and Did2/CHMP1 ( Lata et al . , 2008; Saksena et al . , 2009; Davies et al . , 2010; Adell et al . , 2014 ) . ESCRT-III polymers are remodeled and disassembled by the AAA ATPase Vps4 before ( Nickerson et al . , 2010; Baumgartel et al . , 2011; Jouvenet et al . , 2011; Adell et al . , 2014 ) and/or after ( Wollert et al . , 2009; Wollert and Hurley , 2010 ) membrane scission , returning individual subunits to their closed state in the cytoplasm . Structural studies demonstrate that ESCRT-III proteins each consist of a ∼7 nm helical hairpin ( α1–α2 ) stabilized and regulated by four or more short helices ( α3–α6 ) ( Muziol et al . , 2006; Bajorek et al . , 2009; Xiao et al . , 2009; Martinelli et al . , 2012 ) . α5–α6 and surrounding sequences are responsible for autoinhibition and thereby regulate membrane binding and polymer assembly ( Zamborlini et al . , 2006; Shim et al . , 2007; Bajorek et al . , 2009 ) . In their active or open state , ESCRT-III proteins polymerize in vitro and in vivo using a number of different protein–protein interfaces ( Muziol et al . , 2006; Ghazi-Tabatabai et al . , 2008; Bajorek et al . , 2009; Xiao et al . , 2009; Morita et al . , 2011 ) . Snf7/CHMP4 is the most abundant ESCRT-III protein , and forms membrane-associated polymers in vitro and in transfected cells ( Hanson et al . , 2008; Teis et al . , 2008; Fyfe et al . , 2011; Henne et al . , 2012 ) . Electron microscopy ( EM ) shows that Snf7/CHMP4 polymers are curved filaments <5 nm in diameter ( Ghazi-Tabatabai et al . , 2008; Hanson et al . , 2008; Pires et al . , 2009; Henne et al . , 2012 ) . Other ESCRT-III proteins shown to assemble in vitro into a variety of mostly tubular structures include CHMP2A and CHMP3 ( Lata et al . , 2008; Bajorek et al . , 2009; Dobro et al . , 2013; Effantin et al . , 2013 ) , CHMP1B ( Bajorek et al . , 2009; Dobro et al . , 2013 ) , CHMP2B ( Bodon et al . , 2011 ) , and Ist1 ( Bajorek et al . , 2009; Dobro et al . , 2013 ) . Finally , ∼17 nm filaments on the membrane in intercellular bridges before cytokinesis disappear when CHMP2A is missing , suggesting that they too might represent a form of ESCRT-III polymer ( Guizetti et al . , 2011 ) . These many views of ESCRT-III have led to proposals for how it and Vps4 contribute to membrane scission during ILV formation , viral budding , and cytokinesis ( Fabbro et al . , 2005; Lenz et al . , 2009; Saksena et al . , 2009; Wollert et al . , 2009; Elia et al . , 2012; Henne et al . , 2013; McCullough et al . , 2013 ) . However , much remains to be learned about the assembly and functional roles of different types of ESCRT-III polymers in their own right and in conjunction with Vps4 . For insight into ESCRT-III polymer structure in a physiological setting , we examined endogenous machinery in cultured mammalian cells by deep-etch EM . We find that ESCRT-III heteropolymers stabilized by depleting Vps4A & B are membrane-attached filaments that frequently spiral to delineate and fill circular domains ∼110 nm in diameter . Many of these spirals induce conical deformations directed away from the cytoplasm . Filaments built from transfected proteins show that coassembly of two ESCRT-III proteins ( in this case Snf7/CHMP4A and CHMP2A ) is required to create membrane deforming ESCRT-III spirals . To define the relationship between ESCRT-III filaments and cargo , we studied HIV-1 Gag-driven virus-like-particle ( VLP ) assembly . Notably , in cells depleted of Vps4 , we find that Gag assemblies are often encircled by ESCRT-III filaments . This suggests a previously unappreciated role for Vps4 in remodeling ESCRT-III around cargo-containing membrane domains in addition to its more canonical role in recycling ESCRT-III subunits .
Endogenous ESCRT-III proteins including CHMP4A and CHMP2B are diffusely localized in cells with little steady-state concentration on endosomes despite their known role in lumenal vesicle formation ( Figure 1A , left panels ) . This is not surprising given measurements in live cells showing that both they and Vps4 are present for a few minutes or less at their site of action during viral particle release ( Baumgartel et al . , 2011; Jouvenet et al . , 2011 ) or cytokinetic abscission ( Elia et al . , 2012 ) . To visualize ESCRT-III , we therefore set out to increase the lifetime of membrane associated polymers taking advantage of the known role for Vps4 in mediating disassembly and recycling of ESCRT-III ( Babst et al . , 1998; Lin et al . , 2005 ) . Depleting Vps4A & B , the two Vps4 proteins present in human cells , by RNAi ( Figure 1B ) caused ESCRT-III proteins to redistribute onto puncta localized throughout the cell ( Figure 1A , middle panels ) . Sedimentation of detergent solubilized cell extracts further demonstrated that a significant fraction of ESCRT-III proteins accumulated in the detergent insoluble fraction characteristic of polymerized ESCRT-III ( Figure 1C; Shim et al . , 2007 ) . 10 . 7554/eLife . 02184 . 003Figure 1 . Effects of depleting Vps4 on ESCRT-III . ( A ) Localization of indicated ESCRT-III protein in HeLa cells untreated or treated with Vps4A & Vps4B specific siRNA for ∼60 hr . Left and middle panels show maximum intensity projections from confocal z-series through the cells . Right panel shows immunostaining of unroofed plasma membranes from HEK293T cells treated with Vps4A & Vps4B siRNA . Scale bars represent 10 μm . ( B ) Representative immunoblots comparing lysates from cells untreated or treated with siRNA targeting Vps4A & Vps4B . ( C ) Immunoblots of detergent soluble ( left ) and insoluble ( right ) material from control or Vps4A & Vps4B siRNA-treated HeLa cells . DOI: http://dx . doi . org/10 . 7554/eLife . 02184 . 003 Deep-etch EM is particularly well suited to examining the morphology of the plasma membrane ( Heuser , 2000 ) , and although many ESCRT-III puncta were on endosome like structures throughout the cytoplasm , a subset appeared at the edge of the cell and remained associated with unroofed plasma membranes ( Figure 1A , right panels ) . Their presence could reflect a role for ESCRT-III in creating outwardly-directed vesicles known as ectosomes that bud from the plasma membrane ( Stein and Luzio , 1991; Nabhan et al . , 2012; Choudhuri et al . , 2014; Jimenez et al . , 2014 ) . Alternatively , ESCRT-III puncta could be transferred to the plasma membrane by MVB exocytosis ( Thery , 2011 ) . Finally , ESCRT-III may have some other , potentially structural , role on the plasma membrane . Importantly , the accessibility of the plasma membrane to deep-etch EM provided us with a starting point for studying the structure of cellular ESCRT-III . Deep-etch EM of plasma membranes from cultured cells depleted of Vps4A & B revealed unique filamentous assemblies intermingled among otherwise typical cytoskeletal structures , clathrin lattices , and caveolae ( Figure 2A ) . The new filaments were attached to the membrane and most clearly recognized in discrete spirals that often deformed the membrane into conical protrusions . Because these filaments resemble previously described ESCRT-III polymers , we used antibodies recognizing several ESCRT-III proteins and found that gold particles marking each of the proteins examined ( CHMP6 , CHMP2B , and CHMP1B ) bound on or near filaments and filament spirals but not elsewhere on membranes from cells depleted of Vps4A & B ( Figure 2B ) or on membranes from control HeLa cells ( not shown ) . Fortuitous immunodecoration inside broken but not fully unroofed Vps4-depleted cells showed that eversions containing ESCRT-III were also present along the cell's top surface ( Figure 2—figure supplement 1 ) . The apparent distribution of different ESCRT-III proteins on the filament spirals was similar ( Figure 2B and not shown ) , although generally predominant labeling near the perimeter may reflect limited epitope accessibility toward the center of the spirals . This immunodecoration argues against unique positioning of particular ESCRT-III proteins along the filaments , although higher resolution studies and additional antibodies will be required to determine exactly where different ESCRT-III proteins are present in each spiral . 10 . 7554/eLife . 02184 . 004Figure 2 . ESCRT-III filaments form conical spirals on the plasma membrane of cells depleted of Vps4A & Vps4B . ( A ) Survey view of the cytoplasmic surface of the plasma membrane from a HeLa cell unroofed ∼60 hr after transfection with siRNA targeting Vps4A & Vps4B . Pseudocoloring shows clathrin ( orange ) , caveolae ( green ) , and ESCRT-III spirals ( yellow ) . Similar filament spirals were seen in siRNA treated HEK293T , U2OS , and MCF-7 cells ( not shown ) . ( B ) Immunodecoration of ESCRT-III proteins in spirals on Vps4-depleted HeLa cell plasma membranes . Antibodies recognizing CHMP6 ( left ) , CHMP2B ( middle ) , and CHMP1B ( right ) detected with 18 nm gold that appears white in these contrast reversed EM images . Use view glasses for 3D structure in both panels ( left eye = red ) . Scale bars represent 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 02184 . 00410 . 7554/eLife . 02184 . 005Figure 2—figure supplement 1 . Immunodecoration of ESCRT-III proteins beneath protrusions from the cell surface . Fortuitous breaks in HeLa cells depleted of Vps4A & B subjected to immunogold labeling allowed antibody staining inside broken cells . ( A ) Immunodecoration of CHMP1B , ( B ) immunodecoration of CHMP2B . 18 nm gold particles appear white in these contrast reversed deep-etch EM images . Insets show enlarged images ( blue edges ) with gold particles marked yellow . Use view glasses for 3D structure . Scale bars represent 100 nm for main panels and 50 nm for insets . DOI: http://dx . doi . org/10 . 7554/eLife . 02184 . 005 The most striking and unexpected characteristic of these stabilized ESCRT-III filaments was their frequent organization into spirals ( Figure 3A , B ) . Other patterns including partial or complete rings and meandering filaments were also seen but were less common . The overall diameter of the spiral assemblies was 108 ± 30 nm ( Figure 3C ) , with an average of four 360° turns in each spiral . The shape of the spirals on the membrane varied , ranging from clearly conical to flat . The total length of the filament in each spiral was 660 ± 350 nm . While Vps4-driven remodeling may normally keep filaments from reaching this length ( Teis et al . , 2008 ) , these images set an upper limit on the size and shape of endogenous ESCRT-III polymer that is smaller and more uniform than that of the extensive polymers seen in vitro ( Lata et al . , 2008; Bajorek et al . , 2009; Henne et al . , 2012 ) or in cells overexpressing specific components of ESCRT-III ( Hanson et al . , 2008; Bodon et al . , 2011 ) . ESCRT-III assembly must therefore be controlled by something other than the availability of subunits . One possibility is that centripetal growth of a membrane-attached filament determines the diameter of the final assembly . Interestingly , ESCRT-III spirals surround and cover an area of membrane that corresponds well with that needed to generate a vesicle ∼50 nm in diameter—the typical size of ILVs in mammalian cells ( Murk et al . , 2003 ) —suggesting that an initial ESCRT-III circle might define the content of an incipient vesicle . 10 . 7554/eLife . 02184 . 006Figure 3 . Structural characteristics of endogenous ESCRT-III spirals . ( A ) Survey views of filaments and spirals on plasma membranes of HeLa cells depleted of Vps4A & Vps4B show range of shapes , filament diameter , and direction of spiraling . Examples of abrupt changes in filament diameter are highlighted in color ( thicker filaments in magenta and thinner filaments in green ) in right panel . Use view glasses for 3D structure ( left eye = red ) . Scale bar represents 100 nm . ( B ) Views of individual spirals with direction of spiral from perimeter towards center as shown . Each box corresponds to 185 nm . ( C ) Outer diameter of spirals defined as conical or flat based on appearance in 3D ( n = 184 , 61% conical 108 ± 29 nm; 39% flat 102 ± 20 nm ) . ( D ) Distribution of filament widths measured at three points per spiral . DOI: http://dx . doi . org/10 . 7554/eLife . 02184 . 006 For insight into how ESCRT-III filaments grow on the membrane , we asked if they preferentially spiral in one direction or the other . Surprisingly , we found both left- and right-handed spirals even in the same field of view ( Figure 3 ) . This was true both for obviously conical and flat spirals . Even without knowing precisely how subunits are arranged within the filaments , it is clear that changes along the spiral have the potential to affect positioning of functionally important motifs in the ESCRT-III proteins . For example , changes in the recently described N-terminal membrane insertion motif ( Buchkovich et al . , 2013 ) could contribute to membrane deformation . The apparent diameter of the finest filaments in ESCRT-III spirals was ∼4 nm ( Figure 3D ) . Because individual ESCRT-III proteins contain a ∼7 nm helical hairpin that is 3–4 nm wide ( Muziol et al . , 2006; Bajorek et al . , 2009; Xiao et al . , 2009; Martinelli et al . , 2012 ) , their assembly into filaments must involve interactions along an approximately longitudinal axis as previously envisioned for filaments and tubes assembled in vitro ( Lata et al . , 2008; Henne et al . , 2012; Dobro et al . , 2013; Effantin et al . , 2013 ) . Filament width varied from 4 to as much as 15 nm in both conical and flat spirals ( Figure 3D ) , excluding differences in accessibility to platinum deposition as a primary cause of the differences . Instead , differences in platinum decoration are likely to reveal variations in molecular structure ( Bachmann et al . , 1985 ) . Wider filaments—sometimes with a split down their middle suggesting the presence of parallel substrands—were most prevalent at the perimeter while thin filaments occurred throughout but were almost always present near the spirals' center . Filaments sometimes abruptly changed width ( see examples highlighted in Figure 3A ) further supporting the presence of subfilaments within wider structures . Given the functional requirement for two ESCRT-III nucleating sites in ESCRT-II ( Teis et al . , 2010 ) or Alix ( Pires et al . , 2009 ) , one possibility is that parallel filaments create the outer turns of a spiral but give way to single filaments with the potential for tighter curvature near the spiral's center . The ability of specific ESCRT-III proteins to cap filament assembly ( Teis et al . , 2008 ) could promote transition from wide to narrow filaments by ending one of two ( or more ) parallel filaments . Understanding the functional significance and control of differences in filament content and shape will be an important question for the future . We next wondered what it is about endogenous ESCRT-III filaments that promotes their assembly into stereotyped membrane-deforming spirals . While deep-etch EM does not have the resolution to define the arrangement of individual subunits , it allows us to compare filament shape and organization to that of filaments assembled in transfected cells . We previously found that overexpressed FLAG-tagged CHMP4 proteins create extensive networks of interconnected curved but flat filaments ( Hanson et al . , 2008 ) and others have seen that CHMP4B and yeast Snf7 behave similarly in vitro ( Pires et al . , 2009; Henne et al . , 2012 ) . To rule out any effects of the acidic FLAG-tag at the N-terminus of CHMP4A in our earlier study ( Hanson et al . , 2008 ) , we examined filaments formed by untagged CHMP4A ( Figure 4A ) . This was important because the FLAG-tag might specifically interfere with a recently described N-terminal membrane insertion motif in CHMP4-family proteins ( Buchkovich et al . , 2013 ) . As before , CHMP4A filaments circle and spiral along the membrane to form an anastomosing network but do not couple this to changes in membrane shape ( Figure 4A ) . CHMP4A filaments also remain flat when built from a constitutively ‘open’ CHMP4 ( α1–α5 ) truncation mutant ( Shim et al . , 2007; Figure 4B ) or a K52E mutant that enhances Snf7 assembly in yeast ( Henne et al . , 2012 ) ( not shown ) . These consistently flat networks indicate that interface ( s ) needed to deform the membrane are either not present or not exposed in CHMP4 homopolymers even when intramolecular autoinhibitory contacts are released . 10 . 7554/eLife . 02184 . 007Figure 4 . ESCRT-III polymer structure in transfected cells . Anaglyphs of plasma membranes from COS-7 cells expressing ( A ) untagged CHMP4A , ( B ) CHMP4A ( α1–α5 ) , and ( C–E ) CHMP4A ( α1–α5 ) and full-length CHMP2A together . ( C ) Cytoplasmic surface of the plasma membrane from an unroofed cell with high magnification shown in the inset , ( D ) corresponding top of a whole cell , and ( E ) top of a whole cell extracted with Triton X-100 and saponin after fixation to expose the underlying protein scaffold . Use view glasses for 3D structure ( left eye = red ) . Scale bars represent 100 nm except in C where the scale bar on the survey view corresponds to 500 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 02184 . 007 Given that ESCRT-III function in cellular events requires more than one ESCRT-III protein ( Babst et al . , 2002; Morita et al . , 2010 , 2011 ) , we hypothesized that coassembly of CHMP4 with additional ESCRT-III proteins might be required to create filaments with the shape of the endogenous filaments above . To test this idea , we examined the effect of coexpressing CHMP4A with CHMP2A based on the fact that the ESCRT-III function in a wide variety of pathways and organisms requires members of both the CHMP4 and CHMP2 subfamilies of proteins ( Babst et al . , 2002; Morita et al . , 2010 , 2011 ) . Previous biochemical analyses showed that full-length CHMP4A and CHMP2A do not efficiently coassemble while heteropolymers form readily when either protein is activated by deleting its autoinhibitory domain ( Shim et al . , 2007 ) . Strikingly , membranes from cells co-expressing ‘open’ CHMP4A ( α1–α5 ) and full-length CHMP2A were studded with eversions lined by spiraling filaments ( Figure 4C ) . Corresponding eversions were apparent protruding from the tops of whole cells ( Figure 4D ) ; underlying filaments were revealed by delipidating fixed samples with detergent ( Figure 4E ) . Similar membrane-deforming filaments developed when we coexpressed full-length CHMP4A with ‘open’ CHMP2A ( α1–α5 ) ( not shown ) , demonstrating that the shape changes do not depend on C-terminal autoinhibitory sequences in either protein . The fact that ESCRT-III heteropolymers differ from the sum of their parts supports a model in which subunits such as CHMP2A coassemble with CHMP4 to change the structure of the filament and its effects on membrane shape . Similar changes in filament morphology were previously seen with purified proteins in vitro ( Henne et al . , 2012 ) suggesting that distinctive heteropolymer shape may be a general feature of ESCRT-III biology . An important question for the future will be to define how recruitment and assembly of appropriate ESCRT-III subunits is regulated to create the requisite polymer shape . A challenge in thinking about the function of ESCRT-III spirals trapped by depleting Vps4 ( Figures 2 and 3 ) is that they are typically not in contact with nascent or released vesicles . If spirals represent ‘scars’ left behind after vesicle budding , they have separated from the vesicles they helped to create . An alternative possibility is that spirals represent the preferred arrangement of ESCRT-III when it assembles spontaneously as a ‘blank’ in the absence of cargo ( and Vps4 ) . Efforts to see ESCRT-III spirals on the surface of endosomes where their function in vesicle biogenesis is better defined were complicated by poor retention and accessibility of these organelles to deep-etch EM as well as the development of stacked class E compartments similar to those seen in Saccharomyces cerevisiae ( Nickerson et al . , 2006 , and not shown ) . We therefore turned to the best-characterized ESCRT-dependent event at the plasma membrane , Human Immunodeficiency Virus ( HIV-1 ) budding . HIV-1 assembly is driven by polymerization of the virally encoded Gag polyprotein , which recruits cellular ESCRT proteins to facilitate virion release from the plasma membrane and can be expressed alone to produce virus-like particles ( VLPs ) ( Gheysen et al . , 1989; Karacostas et al . , 1989; Sundquist and Krausslich , 2012 ) . Current thinking based on electron tomography of immature virions ( Wright et al . , 2007; Carlson et al . , 2008; Briggs et al . , 2009 ) and bud sites ( Carlson et al . , 2008 ) is that ESCRTs and especially ESCRT-III play important roles both in completing the viral sphere ( that is only 2/3 covered by polymerized Gag ) and in severing its connection to the cell . ESCRT-III and Vps4 are transiently recruited to Gag assemblies to mediate release ( Jouvenet et al . , 2011 ) . This machinery is typically thought to act on the cytoplasmic surface of the plasma membrane to constrict the vesicle neck and release a viral particle ( Sundquist and Krausslich , 2012 ) , although a recent study using fluorescently labeled proteins and superresolution imaging raised the possibility of similar constriction from within the viral particle ( Van Engelenburg et al . , 2014 ) . Using deep-etch EM allows us to capture snapshots of this process while assessing the relationship between ESCRT-III and HIV-1 Gag as a readily recognizable cargo . HEK293T cells transiently expressing HIV-1 Gag ( Figure 5 ) or Gag-GFP ( Figure 5—figure supplement 1 ) produce abundant VLPs that are readily apparent by deep-etch EM both on and around cells as well as beneath unroofed plasma membranes ( Figure 5A–C ) . Release of VLPs was corroborated by fluorescence microscopy of cells expressing Gag-GFP ( Figure 5—figure supplement 1 ) and by isolation and immunoblotting of VLPs ( not shown ) . Unroofed plasma membranes display unique circular and semi-spherical protein assemblies ranging in size up to the diameter of VLPs that appear to be nascent Gag assemblies ( Figure 5D ) . In order to ascertain that these in fact contain Gag , we immunodecorated unroofed cells with an antibody specific to the membrane-proximal matrix ( MA ) domain of Gag ( Figure 5E–H ) . Gold particles were numerous around putative Gag assemblies on unroofed plasma membranes ( Figure 5E , E′ ) and around VLPs ( Figure 5F , F′ ) when samples were delipidated by detergent extraction after fixation . When membranes were intact , immunodecoration of Gag assemblies was limited to their perimeter ( Figure 5G , G′ ) and was abolished in released VLPs ( Figure 5H ) as expected . By deep-etch EM , Gag-GFP assemblies were less uniform in size and shape than those containing Gag ( Figure 5—figure supplement 1C ) , consistent with the irregular distribution of Gag-GFP seen by thin section EM ( Pornillos et al . , 2003 ) and with the decreased Gag content of VLPs containing Gag fused to similarly sized fluorescent proteins ( Gunzenhauser et al . , 2012 ) . Notably there was no evidence by direct viewing or immunolabeling ( not shown ) to indicate the presence of ESCRT-III on or near any of these Gag assemblies . This is not surprising given live cell studies showing that ESCRT-III and Vps4 are only transiently recruited after Gag assembly is essentially complete ( Baumgartel et al . , 2011; Jouvenet et al . , 2011 ) . 10 . 7554/eLife . 02184 . 008Figure 5 . Deep-etch EM of HIV-1 VLP budding . ( A ) Low magnification view of an unroofed HIV-1 Gag-transfected HEK293T cell surrounded by VLPs . ( B ) Top view of whole cell budding VLPs . ( C ) View of unroofed plasma membrane showing bumps corresponding to VLPs trapped underneath the membrane . ( D ) Views of unroofed plasma membrane showing Gag assemblies exposed on the cytoplasmic surface of the plasma membrane . ( E–H ) Immunodecoration of Gag on detergent extracted plasma membranes ( E and E′ ) , detergent extracted VLPs ( F and F′ ) , intact unroofed plasma membranes ( G and G′ ) and intact VLPs ( H ) . ( E′ , F′ and G′ are same as E , F and G but show gold in yellow ) . Use view glasses for 3D structure ( left eye = red ) . Scale bars represent ( A ) 500 nm , ( B–H ) 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 02184 . 00810 . 7554/eLife . 02184 . 009Figure 5—figure supplement 1 . VLP formation by HIV-1 Gag-GFP . ( A ) Fluorescence microscopy of a live Gag-GFP expressing HEK293T cell showing released VLPs captured around the single cell . Scale bar represents 10 μm . ( B ) Low magnification view of an unroofed Gag-GFP transfected HEK293T cell surrounded by VLPs . Use view glasses for 3D structure ( left eye = red ) . Scale bar represents 500 nm . ( C–E ) High magnification views of developing Gag assemblies on the plasma membrane of unroofed cells . Small Gag-GFP assemblies ( C ) resemble those formed by untagged Gag , while larger assemblies ( E ) are discontinuous and do not form solid spherical assemblies . Scale bars in C–E represent 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 02184 . 009 To explore the role of ESCRT-III filaments in VLP biogenesis , we therefore once again depleted cells of Vps4 to stabilize ESCRT-III in its assembled state . As expected , expressing a dominant negative mutant of Vps4A ( Vps4A E228Q ) or silencing Vps4 as above increased the amount of Gag-GFP ( Figure 6—figure supplement 1A ) or Gag ( Figure 6—figure supplement 1B ) on the plasma membrane and decreased release of VLPs as detected by particle analysis methods ( data not shown ) . Strikingly , in cells lacking Vps4 Gag assemblies on the plasma membrane were now often surrounded by a single filament not seen in control cells ( Figure 6 ) . The encircling filaments appeared similar to the ESCRT-III filaments examined above , each ranging from 4 . 5–12 . 8 nm ( average 8 . 6 ± 1 . 6 nm , n = 96 ) in width with the majority wide enough to contain more than one ∼4 nm substrand . Immunolabeling confirmed that ESCRT-III proteins localized to the perimeter of Gag assemblies coincident with the filaments ( Figure 6D ) . Interestingly , the encircling filaments surrounded Gag assemblies of various sizes ranging from ∼60–150 nm suggesting that the threshold amount of Gag needed to activate ESCRT-III assembly may be decreased in the setting of reduced Vps4 leading to premature nucleation of the ESCRT-III ring . Alternatively , these smaller Gag assemblies may indicate a role for ESCRT-III filaments in confining Gag ( or other cargo proteins ) in a domain that normally expands upon remodeling by Vps4 . More deeply invaginated Gag assemblies were sometimes surrounded by what appeared to be tightly spiraled ESCRT-III filaments , and immunolabeling confirmed that ESCRT-III was present on the cytoplasmic rim and surrounding surface of these invaginating VLPs ( Figure 6D , right panel and not shown ) . These spirals may correspond to the darkly stained ring previously noted in thin section EM analysis of HIV-1 budding from cells depleted of CHMP2A ( and therefore unable to engage Vps4 ) ( Morita et al . , 2011 ) and seem likely to represent trapped intermediates in the biogenesis of VLPs . 10 . 7554/eLife . 02184 . 010Figure 6 . ESCRT-III filaments surround Gag assemblies in cells depleted of Vps4 . ( A ) HEK293T cells treated with Vps4A & Vps4B siRNA accumulate unique filament-encircled Gag assemblies . Inset is pseudo-colored to show the central Gag assembly ( blue ) , surrounding filament ( red ) , and perpendicular ‘struts’ between them ( green ) . Each field also shows the occasionally seen subplasmalemmal VLP bump to provide a sense of scale . ( B ) Individual views of Gag assemblies , ( C ) Gag-GFP assemblies , and ( D ) Gag assemblies immunodecorated with indicated gold conjugated antibodies . Scale bars represent 100 nm . Each box in B–D is a 320 nm square . Use view glasses for 3D structure ( left eye = red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02184 . 01010 . 7554/eLife . 02184 . 011Figure 6—figure supplement 1 . VLP release is impaired by inactivating or depleting Vps4 . ( A ) Confocal image of HEK293T cells expressing Vps4B-E228Q or Vps4A & Vps4B siRNA and transfected with Gag-GFP . Note that Gag accumulates on the plasma membrane instead of in released VLPs ( compare with Figure 5—figure supplement 1A ) . ( B ) Deep-etch EM of an unbroken Vps4-depleted cell expressing Gag showing unbudded VLPs accumulating on the cell surface . DOI: http://dx . doi . org/10 . 7554/eLife . 02184 . 011 How do Gag assemblies recruit ESCRT-III filaments ? We noticed that ESCRT-III filaments ( in cells depleted of Vps4 ) were frequently connected to the central Gag assembly by perpendicular ‘struts’ ∼14 nm long ( Figure 6 , see green highlighting in inset ) . These struts resemble previous negative stain images of purified Alix bound perpendicularly to Snf7 filaments ( Pires et al . , 2009 ) although based on size they could also correspond to ESCRT-I ( Boura et al . , 2011 ) , an ESCRT-I-ESCRT-II supercomplex ( Boura et al . , 2012 ) , or something else altogether . The C-terminal p6 domain of Gag contains a PTAP motif ( essential for interaction with the ESCRT-I protein Tsg101 [Garrus et al . , 2001] ) and a LYPXnL motif ( involved in interaction with ALIX [Strack et al . , 2003] ) and has been shown to be essential for budding of Gag VLPs as well as HIV-1 . To determine whether these motifs are required for ESCRT-III recruitment , we deleted the p6 domain from Gag ( referred to as GagΔp6 ) . In cells expressing GagΔp6 , very few VLPs were released and instead large numbers of arrested Gag-containing buds accumulated on the surface ( Figure 7 and not shown ) . Importantly , this was true regardless of whether Vps4 was present or not , confirming that the p6 domain is required for VLP release . On unroofed plasma membranes Gag assemblies similar to those formed by full-length Gag were observed but were not surrounded by ESCRT-III rings ( Figure 7 ) , demonstrating that p6 sequences are required to recruit ESCRT-III . Furthermore , no struts were apparent , consistent with the hypothesis that these correspond to ESCRT-I and/or Alix . We were , however , unable to confirm the identity of the struts because available antibodies were unsuitable . Detailed mutagenesis coupled with antibodies appropriate for immunodecoration on these samples will be needed to conclusively identify these structures . 10 . 7554/eLife . 02184 . 012Figure 7 . Deep-etch EM of Gag budding without its ESCRT-recruiting p6 domain . ( A ) Top surface of HEK293T cell expressing GagΔp6 showing blocked VLP budding . ( B ) Top surface of Vps4-depleted HEK293T cell expressing GagΔp6 showing similar blocked VLP budding . ( C ) Unroofed plasma membrane corresponding to A . ( D ) Unroofed plasma membrane corresponding to B . Note the invaginated Gag assemblies with no surrounding ESCRT-III ring ( C and D ) . C′ and D′ are identical to C and D but with Gag assemblies colored yellow for clarity . Scale bar represents 100 nm . Use view glasses for 3D structure ( left eye = red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02184 . 012 The data presented here provide the first views of endogenous ESCRT-III filaments forming the rings and spirals frequently drawn in models of ESCRT function but not previously seen on cellular membranes . Earlier studies of various ESCRT-III polymers in vitro and in transfected cells inspired thinking about how different filamentous and tubular structures might function in vesicle biogenesis and membrane scission . Ideas ranged from purse-string like constriction of a ring ( Saksena et al . , 2009 ) to polymer-driven membrane buckling ( Lenz et al . , 2009 ) to the more prevalent dome-based membrane scission model ( Fabrikant et al . , 2009 ) and its variants ( Boura et al . , 2012; Elia et al . , 2012; Henne et al . , 2012 ) . We propose a modified model that builds on many of these previous ideas , interpolating between the rings of ESCRT-III we see around HIV-1 Gag assemblies and the completed conical spirals that accumulate on the membrane ( Figure 8 ) . In this model , a first role for ESCRT-III is to encircle and thereby delineate cargo and membrane destined for inclusion in a vesicle or VLP ( Figure 8A ) . The encircling filament develops when nucleating factors activated by cargo form struts around a central domain and in turn recruit ESCRT-III . Either because of intrinsic filament curvature or because of multiple contacts with cargo , ESCRT-III filaments grow to surround the cargo ( Figure 8A ) . Because encircled Gag assemblies uniquely accumulate when Vps4 is missing ( Figure 6 ) , we propose that an important and previously unappreciated role for Vps4 is to release connection ( s ) between the ESCRT-III filament and cargo , perhaps displacing the struts while also opening the ring . The ESCRT-III filament can then grow to adopt its preferred shape as a conical spiral ( Figure 8B ) potentially driving neck constriction and vesicle release as shown ( Figure 8C ) . Finally ( and not shown ) , Vps4 is also responsible for disassembling ESCRT-III filaments to recycle their subunits during or after vesicle release . 10 . 7554/eLife . 02184 . 013Figure 8 . Speculative model describing ESCRT-III and Vps4 function in vesicle biogenesis and release . ( A ) Cargo - loosely defined to include either HIV-1 Gag or material destined for incorporation into ILVs–is concentrated in the circular domain shown in pink . After reaching some threshold , cargo recruits and/or activates factors to initiate ESCRT-III assembly . These are represented here by green ‘struts’ perpendicular to the cargo perimeter . Once nucleated , the ESCRT-III filament extends to surround and confine cargo . In the absence of Vps4 , this intermediate accumulates . ( B ) When present , we propose a new role for Vps4 in which it is engaged to break connection ( s ) between ESCRT-III and cargo , thereby allowing the ESCRT-III spiral to grow into its preferred spiral shape . ESCRT-III recruits new membrane into the neck as it grows , shown by the addition of blue membrane to the budding vesicle . ( C ) A fully assembled ESCRT-III spiral narrows the membrane neck , ultimately driving vesicle release . DOI: http://dx . doi . org/10 . 7554/eLife . 02184 . 013 While this model remains speculative because it is based on static images of ESCRT-III polymers stabilized by depleting Vps4 , it both supports important aspects of earlier models describing ESCRT-III driven membrane remodeling ( Fabrikant et al . , 2009; Wollert et al . , 2009; Elia et al . , 2012; Henne et al . , 2012 ) and provides explanations for a few previously enigmatic observations . Interestingly , there are similarities between the different forms of ESCRT-III seen here and structures previously described by EM of purified yeast proteins ( Henne et al . , 2012 ) . Specifically , Henne et al . found that combining ESCRT-II as a nucleating factor with ESCRT-III proteins ( Vps20 and Snf7 or a combination of four core ESCRT-III proteins ) generated rings of ESCRT-III ∼50–70 nm in diameter , appropriate for surrounding a domain needed to create 25–35 nm ILVs characteristic of yeast MVBs . In our images , the ∼110 nm average outer diameter ( Figure 3C ) of ESCRT-III spirals would in an analogous manner support formation of the ∼50 nm vesicles typical of mammalian ILVs and extracellular vesicles . Henne et al . also found that combining ESCRT-III proteins with each other in the absence of ESCRT-II created spring-like three-dimensional coils . Not clear was whether ( and how ) rings might convert into spiraling filaments . We suggest that Vps4 may have a general role in converting a cargo confining ring to a vesicle-generating spiral , likely by separating ESCRT-III from nucleating factors seen as struts in our model . The idea that Vps4 plays specific and essential roles both before and after membrane scission is supported by a number of observations . Fluorescently tagged Vps4 is recruited to growing viral particles before they are released ( Baumgartel et al . , 2011; Jouvenet et al . , 2011 ) , Vps4 appears at two stages prior to cellular abscission ( Elia et al . , 2011 ) , ESCRT-III and Vps4 are recruited simultaneously to the centroid of dividing Crenarchaea ( Samson et al . , 2011 ) , and perturbing Vps4 function and/or regulation changes ILV size and/or release in yeast ( Nickerson et al . , 2010; Wemmer et al . , 2011; Adell et al . , 2014 ) . A prediction of our model is that interfering with Vps4 activity—whether by depleting it or expressing a dominant negative mutant—will trap ESCRT-III around the base of incipient vesicular structures and in particular around the base of emerging Gag assemblies . This trapped ESCRT-III will remain attached to Gag inside any VLPs that are ultimately released . This in turn will increase ESCRT-III recovered in whatever VLPs are released , something that we ( data not shown ) and others ( Van Engelenburg et al . , 2014 ) have in fact observed . While we favor the model shown in Figure 8 , one can also envision other ways in which the remodeling of ESCRT-III rings and spirals could contribute to vesicle formation and membrane scission . Future coupling of the ability to see membrane-bound ESCRT assemblies using deep-etch EM with techniques capable of resolving and potentially controlling protein and membrane dynamics will be important in gaining new insight into these processes . Defining the detailed temporal relationship between Vps4 activity and ESCRT-III assembly during virus or vesicle formation will be particularly important . Additionally , while we have taken advantage of the fact that ESCRT polymers are present on the plasma membrane to facilitate deep-etch EM imaging , future studies of their function at the plasma membrane are clearly warranted . Overall , the work described here establishes that endogenous ESCRT-III takes the form of filaments that form circles and conical spirals on the membrane . We propose that these rings and spirals , while longer-lived than normal , represent functional states of a general ESCRT-III membrane scission machine .
HeLa , HEK293T , U2OS , and COS-7 cells originally derived from ATCC were grown in DME ( Invitrogen , Grand Island , NY ) containing 10% fetal bovine serum ( Atlanta Biologics , Atlanta GA ) and 2 mM L-glutamine . pCMV55 encoding HIV-1 Gag was kindly provided by Dr Lee Ratner ( Washington University , St . Louis MO ) and used as previously described ( Shim et al . , 2007 ) . The p6 domain of Gag was deleted by introducing a stop codon to remove 52 amino acids from the C-terminus and introduce an XhoI site using the primer AAAAAACTCGAGTTAAAAATTCCCTGGCCTTCCCTTG . Full length and p6 deleted Gag were cloned into pcDNA4TO ( Invitrogen ) . pGag-EGFP was obtained through the NIH AIDS Reagent Program , Division of AIDS , NIAID , NIH ( Cat#11468 ) from Dr Marilyn Resh ( Hermida-Matsumoto and Resh , 2000 ) . Plasmids encoding ESCRT-III proteins have been previously described , including untagged human CHMP4A ( also referred to as hSnf7-1 ) in pcDNA3 ( Lin et al . , 2005 ) , CHMP4A ( α1–α5 ) in pcDNA3 . 1FLAG ( Shim et al . , 2007 ) , and CHMP2A full length and α1–α5 truncation in pcDNA3 . 1FLAG ( Shim et al . , 2007 ) . Antibodies used include rabbit polyclonals against Gag ( MA-specific; kind gift from Dr Lee Ratner , Washington University School of Medicine , St . Louis MO ) , CHMP6 ( PA5-21831; ThermoScientific , Waltham MA ) , CHMP4A ( Lin et al . , 2005 ) , CHMP4B ( Shiels et al . , 2007 ) , CHMP2A ( sc67227; Santa Cruz , Dallas , TX ) , CHMP2B ( ab33174; AbCam , Cambridge , England ) , CHMP1B ( 14639-1-AP; Protein Tech Group , Chicago , IL ) , VPS4B ( Lin et al . , 2005 ) and mouse monoclonals against tubulin ( Sigma , St . Louis MO ) and α-SNAP ( SySy , Gottingen , Germany ) . siRNA duplexes targeting VPS4A ( CCGAGAAGCUGAAGGAUUAdTdT ) and VPS4B ( CCAAAGAAGCACUGAAAGAdTdT ) have been previously described ( Kieffer et al . , 2008 ) and were from ThermoScientific . Detergent solubility assays to monitor ESCRT-III polymer assembly after Vps4 depletion were as previously described ( Shim et al . , 2007 ) . HeLa and HEK293T cells plated on glass coverslips were stained as previously described ( Shim et al . , 2007 ) . Confocal imaging was performed on an Olympus FV500 or FV1200 microscope using a 60x 1 . 4 NA objective . Maximum intensity projections were prepared using ImageJ ( version 1 . 47u ) . Brightness and contrast were adjusted as necessary with Adobe Photoshop ( Adobe Systems , San Jose , CA ) and composite figures were prepared in Adobe Illustrator . Cells plated at ∼70% confluence were transfected with 15 nM each of siRNA duplexes targeting VPS4A and VPS4B using Dharmafect#1 according to manufacturer guidelines . Cells were trypsinized ∼24 hr later and replated onto 12 mm poly-L- or poly-D-lysine coated BioCoat coverslips ( BD Biosciences , East Rutherford NJ ) . Plasma membranes were prepared the following day , typically 40–50 hr after initial siRNA transfection . This was done as previously described ( Hanson et al . , 2008 ) . Briefly , coverslips were washed in 30 mM Hepes , pH 7 . 4 , 100 mM NaCl , 2 mM CaCl2 and then dipped into an intracellular buffer ( 30 mM Hepes , pH 7 . 2 , 70 mM KCl , 5 mM MgCl2 , and 3 mM EGTA ) and subjected to a brief pulse of ultrasound before transfer into the same buffer containing fixative ( 2% glutaraldehyde or 2% PFA if immunostaining was planned ) . The area of coverslip with the highest yield of plasma membranes was identified by phase contrast microscopy and trimmed with a diamond knife to ∼3 × 3 mm . Transfection of Gag or Gag-GFP encoding plasmids was performed using Lipofectamine 2000 ( Life Technologies , Grand Island , NY ) according to the manufacturer's instructions . When combined with siRNA transfections , plasmid transfections were carried out 24 hr after initial introduction of siRNA . ESCRT-III encoding plasmids were transfected as previously described ( Hanson et al . , 2008 ) again using Lipofectamine 2000 ( Life Technologies ) . To extract fixed samples with detergent ( Figures 4E , 5E , F ) , fixed coverslips were incubated for 2 hr in buffer containing 1% Triton X-100 and 0 . 1% saponin . Antibody staining was performed as previously described ( Hanson et al . , 2008 ) using 18 nm gold-conjugated goat anti-rabbit or anti-mouse ( Jackson Immunoresearch , West Grove , PA ) antibodies . Samples were prepared essentially as described ( Hanson et al . , 2008 ) except that platinum ( ∼2 nm , as before ) was evaporated onto samples from 17–18° above the horizontal . Replicas were viewed on a JEOL 1400 transmission electron microscope at two different tilt angles ( either ±5 or ±10° ) and images were captured using an AMT camera . Digital image pairs were made into anaglyphs by converting one each to red and blue/green , layering them on top of each other using the screen blending mode in Adobe Photoshop , and finally aligning them to each other using the auto-align function . Composite figures were prepared using Adobe Illustrator . Digital measurements were made using ImageJ ( v . 1 . 47u ) . | Cells contain compartments called organelles that are enclosed within membranes similar to the plasma membrane that surrounds the cell itself . Cells police the proteins on their membranes and move old or damaged proteins into a type of organelle called an endosome . This involves the membrane folding in on itself to form a multivesicular body . The multivesicular bodies deliver their contents to organelles called lysosomes where the old proteins are destroyed . Although it is known that over 30 proteins are involved in the formation of multivesicular bodies , many aspects of how they operate are not well understood . Moreover , disruptions to this process contribute to a large number of diseases including forms of cancer and neurodegeneration . Importantly , the same proteins are hijacked by viruses such as HIV to help them escape from the cells they have infected . Most of the proteins involved in forming multivesicular bodies are part of the ESCRT ( Endosomal Sorting Complex Required for Transport ) system of proteins . A special set of these proteins—ESCRT-III—is thought to cut the membrane to release vesicles and viruses , as well as helping the membrane to deform . Previously , researchers have been unsure how the ESCRT-III complex works because it has a short lifespan and is too small to see on cellular membranes using standard techniques . Now Cashikar , Shim et al . have used a technique called deep-etch electron microscopy in combination with gene knockdown strategies to reveal the structure of the ESCRT-III complex inside cells . A protein called Vps4 is known to recycle ESCRT-III complexes , so Cashikar , Shim et al . studied cells in which the levels of Vps4 had been depleted in order to increase the lifespan of ESCRT-III complexes . In these cells filaments made of ESCRT-III complexes tended to form conical spirals that matched the size and shape of the vesicles and viruses ESCRT-III is thought to produce . ESCRT-III filaments also accumulated as rings around the molecules destined for incorporation into a vesicle or virus . This indicated a new role for Vps4: it separates ESCRT-III from the contents of the vesicle , leaving it free to form a spiral that drives release of the vesicle or virus from the cell . The next challenge will be to test the predictions of this model using techniques that can capture individual vesicle formation events in real time . Understanding the function of ESCRT-III in greater detail may suggest ways to manipulate this pathway to limit the replication of viruses or the degradation of membrane proteins . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2014 | Structure of cellular ESCRT-III spirals and their relationship to HIV budding |
The COP9-Signalosome ( CSN ) regulates cullin–RING ubiquitin ligase ( CRL ) activity and assembly by cleaving Nedd8 from cullins . Free CSN is autoinhibited , and it remains unclear how it becomes activated . We combine structural and kinetic analyses to identify mechanisms that contribute to CSN activation and Nedd8 deconjugation . Both CSN and neddylated substrate undergo large conformational changes upon binding , with important roles played by the N-terminal domains of Csn2 and Csn4 and the RING domain of Rbx1 in enabling formation of a high affinity , fully active complex . The RING domain is crucial for deneddylation , and works in part through conformational changes involving insert-2 of Csn6 . Nedd8 deconjugation and re-engagement of the active site zinc by the autoinhibitory Csn5 glutamate-104 diminish affinity for Cul1/Rbx1 by ~100-fold , resulting in its rapid ejection from the active site . Together , these mechanisms enable a dynamic deneddylation-disassembly cycle that promotes rapid remodeling of the cellular CRL network .
Cullin–RING ubiquitin ligases comprise one of the largest families of regulatory enzymes in eukaryotic cells ( Deshaies and Joazeiro , 2009 ) . With as many as 240 different enzyme complexes , these E3s control a broad array of biological processes ( Skaar et al . , 2013 ) . CRLs comprise seven distinct cullin–RING cores , each of which interacts with its own dedicated set of adaptor–substrate receptor complexes . Although ubiquitination by CRL enzymes is often regulated by covalent modifications of the substrate that stimulate binding to the substrate receptor , the CRL enzymes themselves are also subject to regulation . A key mechanism that controls the activity of all known CRLs is the conjugation of the ubiquitin-like protein Nedd8 to a conserved lysine residue in the cullin subunit ( e . g . K720 in human Cul1 ) ( Enchev et al . , 2015 ) . The available structural and biochemical data indicate that Nedd8 conjugation ( neddylation ) stabilizes a profound conformational change in the C-terminal domain of the cullin . It loosens the interaction of the WHB domain with the RING subunit , allowing both of them to sample a greater conformational space ( Duda et al . , 2008 ) , thereby enhancing the ability of the RING domain to promote ubiquitin transfer to substrate ( Duda et al . , 2008; Saha and Deshaies , 2008; Yamoah et al . , 2008 ) . In addition to direct effects on ubiquitin ligase activity , Nedd8 also protects Skp1/Cul1/F-box ( SCF ) complexes from the substrate receptor exchange factor ( SREF ) Cand1 ( Pierce et al . , 2013; Schmidt et al . , 2009; Wu et al . , 2013; Zemla et al . , 2013 ) . Cand1 binds unmodified SCF complexes and promotes rapid dissociation of the F-box protein ( FBP ) /Skp1 substrate receptor–adaptor module from the Cul1/Rbx1 core . Cand1 can subsequently be dissociated from Cul1 by a different FBP/Skp1 complex , and as a result Cand1 functions as an SREF that accelerates the rate at which Cul1/Rbx1 comes to equilibrium with different FBP/Skp1 substrate receptor–adaptor complexes ( Pierce et al . , 2013 ) . Importantly , the SREF activity of Cand1 is tightly restricted by Nedd8 . Cand1 is not able to bind stably to Cul1 and promote dissociation of FBP/Skp1 when Cul1 is conjugated to Nedd8 ( Liu et al . , 2002; Pierce et al . , 2013 ) . These observations underscore the importance of neddylation not only for controlling the enzymatic activity of CRLs , but also potentially for controlling the repertoire of assembled CRLs . The key role of Nedd8 in CRL biology highlights the importance of the enzymatic pathways that attach and remove Nedd8 ( Enchev et al . , 2015 ) . Of particular significance is the rate of Nedd8 deconjugation , because it serves as the gateway for the exchange cycle; once Nedd8 is removed , a CRL complex is susceptible to the potent SREF activity of Cand1 , and its substrate receptor can be exchanged ( Pierce et al . , 2013 ) . Deconjugation of Nedd8 is mediated by the COP9-signalosome ( CSN ) , which is an eight-subunit Nedd8 isopeptidase ( Lyapina et al . , 2001 ) . The enzymatic activity of CSN resides in its Csn5 subunit , which contains a metalloprotease active site referred to as the ‘JAMM’ domain ( Cope et al . , 2002 ) . The JAMM domain has the general structure E76-Xn-H138-X-H140-X10-D151 ( the subscripts refer to the sequence position of these residues in human Csn5 ) , wherein the H and D residues coordinate a zinc ion . The fourth zinc-coordination site is occupied by a water molecule that that also forms a hydrogen bond to E76 ( Ambroggio et al . , 2004; Sato et al . , 2008; Tran et al . , 2003 ) . Deneddylation of CRLs by CSN is rapid but can be regulated by CRL substrates ( Emberley et al . , 2012; Enchev et al . , 2012; Fischer et al . , 2011 ) . Structural analysis suggests that a CRL ubiquitination substrate bound to a substrate receptor sterically prevents concurrent binding of CSN ( Enchev et al . , 2012; Fischer et al . , 2011 ) . This suggests a model wherein a CRL complex has a higher probability of being conjugated to Nedd8 ( and therefore of being shielded from Cand1 ) as long as it is bound to substrate . Upon dissociation of substrate , a race ensues between binding of either a new substrate or CSN . If CSN wins , Nedd8 is removed , paving the way for Cand1 to initiate substrate receptor exchange . Recently , a crystal structure of free CSN was determined ( Lingaraju et al . , 2014 ) . A major insight to emerge from the structure was the unexpected finding that Csn5 was present in an autoinhibited state , wherein a glutamate ( Csn5-E104 ) within the ‘insert-1’ ( INS1 ) sequence common to JAMM family members ( Sato et al . , 2008 ) forms a fourth ligand to the zinc , displacing the catalytic Csn5-E76-bound water molecule and shifting Csn5-E76 . Csn5-E104 is found in all Csn5 orthologs , but not in other JAMM proteins , suggesting that this mode of regulation is conserved but unique to CSN . Comparison of the structure of free CSN to the structure of a catalytically-dead mutant CSN bound to Nedd8-conjugated SCFSkp2 determined by negative stain electron microscopy ( Enchev et al . , 2012 ) implied that binding of substrate to CSN may induce several conformational changes in the latter , including movement of the N-terminal domains ( NTD ) of Csn2 and Csn4 towards the cullin . The latter movement , in turn , might be further propagated to the Csn5/6 module ( Lingaraju et al . , 2014 ) . Moreover , it is reasonable to expect that during catalysis INS1 moves out of the active site and Csn5-E76 adopts a position similar to that observed in a crystallographic structure of Csn5 in isolation ( Echalier et al . , 2013 ) . Interestingly , if Csn5-E104 is mutated to an alanine , CSN more rapidly cleaves the simple model substrate ubiquitin-rhodamine ( Lingaraju et al . , 2014 ) . This was interpreted to mean that the primary reason for the autoinhibited state is to keep CSN off until it binds a physiologic substrate , which would prevent spurious cleavage of non-cullin Nedd8 conjugates and possibly even ubiquitin conjugates . However , the full extent of the conformational changes required to form an activated complex between CSN and its neddylated substrate , as well as the detailed molecular basis for these changes remain to be established . Therefore , at present , the mechanism of how CSN is switched on and off and the significance of this switching behavior remain unknown .
To gain detailed insights into the molecular determinants underlying activation of CSN , we performed cryo electron microscopy ( cryo EM ) and single particle analysis of CSN5H138A ( we use the nomenclature CSN#x where # refers to subunit number and x to the specific mutation ) in complex with neddylated SCFSkp2/Cks1 ( the sample is described in Enchev et al . , ( 2012 ) ( Figure 1A , Figure 1—figure supplement 1 , Figure 1—figure supplement 2A ) . The Csn5-H138A mutant lacks one of the JAMM ligands that coordinate the catalytic zinc . This mutant forms a normal CSN complex that has been extensively characterized ( Enchev et al . , 2012 ) . We used ~75000 single molecular images for the final three-dimensional reconstruction and the structure was refined to a nominal resolution of 7 . 2 Å , according to the ‘gold standard’ criterion of a Fourier shell correlation ( FSC ) of 0 . 143 ( Rosenthal and Henderson , 2003; Scheres and Chen , 2012 ) ( Figure 1—figure supplement 1C; Figure 1—figure supplement 2A ) . However , some regions in the density map were better defined than others ( see below ) . To avoid over-interpretation , for the subsequent analysis we low-pass filtered the map to 8 . 5 Å , according to the more stringent criterion of an FSC of 0 . 5 . 10 . 7554/eLife . 12102 . 003Figure 1 . Cryo-electron microscopy of a CSN-SCF complex . ( A ) Molecular model of CSN5H138A-SCF-N8Skp2/Cks1 docked into the cryo-electron density map ( gray mesh ) . ( B ) Close-up view of the model , showing the observed conformations of Csn2 , Csn4 , Rbx1 , Csn5/6 and WHB-Nedd8 and ( C ) a cartoon representation of the differences between the apo CSN and substrate-bound state . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 00310 . 7554/eLife . 12102 . 004Figure 1—figure supplement 1 . Cryo-electron microscopy and single particle analysis of a CSN5H138A-N8-SCFSkp2/Cks1 complex , part 1 . ( A ) A representative cryo-electron micrograph of a CSN5H138A-N8-SCFSkp2/Cks1 complex with some single molecular views indicated by white circles ( left ) and a power spectrum indicating Thon rings reaching 6 Å ( right ) . Scale bar is 200 Å . ( B ) Representative two-dimensional class averages from the curated dataset , used for the subsequent analysis . Scale bar is as in ( A ) . ( C ) Surface views of the final , post-processed cryo-electron map . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 00410 . 7554/eLife . 12102 . 005Figure 1—figure supplement 2 . Cryo-electron microscopy and single particle analysis of a CSN5H138A-N8-SCFSkp2/Cks1 complex , part 2 . ( A ) Resolution estimate according to the FSC criteria of 0 . 143 and 0 . 5 . ( B ) Fit of the PCI-domain containing CSN subunits in the cryo-electron density map . Csn1 , 3 , 7 , and 8 match the density very well but the N-terminal domains of Csn2 and Csn4 do not , but their winged-helix domains fit well . The horseshoe arrangement of the six winged-helix domains is indicated with a dotted black line . ( C ) All the C-terminal helices of the CSN subunits match well the electron density map . ( D ) Fit of SCF in the electron density map . ( E ) Same view as in Figure 1B but prior to flexible docking of the N-terminal domains of Csn2 and Csn4 , the MPN domains of Csn5&6 , the WHB domain of Cul1 , and Nedd8 . ( F ) Movement of the N-terminus of Csn2 from its crystallographically-determined position ( left ) into the EM density map ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 00510 . 7554/eLife . 12102 . 006Figure 1—figure supplement 3 . Cryo-electron microscopy and single particle analysis of a CSN5H138A-N8-SCFSkp2/Cks1 complex , part 3 . ( A ) Movement of the N-terminus of Csn4 from its crystallographically-determined position ( left ) into the EM density map ( right ) . The two N-terminal helical repeats of Csn4 , red arrow , are in close proximity to the WHA domain of Cul1 ( green circle ) . ( B ) Localization of the RING domain of Rbx1 . The unfilled density that is indicated by a black ellipse in the right-hand panel of Figure 1—figure supplement 3A accommodates Rbx1 ( shown in red ) . The helices of Csn4 in close proximity to the RING domain of Rbx1 are indicated by a black arrow . ( C ) Re-localization of Csn5/6 . Comparing the left and right panels , Csn5/6 move leftward to occupy unfilled density . The tan and green circles below Csn5/6 indicate densities that are occupied by Nedd8 and the WHB domain , as depicted in ( D ) . ( E , F ) Deneddylation assays with ( E ) wild type Cul1-N8/Rbx1 and indicated CSN variants and ( F ) wild type CSN and mutant Cul1 variants . Note that all Cul1 constructs have an uncleaved C-terminal sortase tag , which is the reason for slower deneddylation of wild type Cul1-N8/Rbx1 by wild type CSN relative to the kinetics reported elsewhere in this work . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 006 The cryo EM structure reported here , alongside the available crystal structure of CSN ( Lingaraju et al . , 2014 ) , enabled us to visualize a broad array of conformational changes that take place upon complex formation in both CSN and neddylated Cul1/Rbx1 , well beyond what was possible with the prior lower resolution model based on negative stain EM ( Figure 1 ) . Specifically , this allowed us to describe movements of the N-terminal domains of Csn2 and Csn4 , the MPN domains of Csn5 and Csn6 . Moreover , in contrast to our previous work , we could locate the RING domain of Rbx1 , as well as Nedd8 and the winged-helix B ( WHB ) domain of Cul1 relative to Csn5 . Nevertheless , the present resolution precludes the determination of the exact orientations of the latter domains but notably , the relative positions of the RING , WHB and Nedd8 reported here have not been reported in any structural model of a cullin , and strongly suggest that both the enzyme and substrate undergo significant conformational rearrangements to enable catalysis . To obtain the model shown in Figure 1 , we initially docked the crystal structure of CSN ( Lingaraju et al . , 2014 ) and a model of Cul1-Nedd8/Rbx1/Skp1/Skp2/Cks1 ( Enchev et al . , 2012 ) as rigid bodies into the electron density map ( Figure 1—figure supplement 2B–E ) . We observed very good matches between the respective map segments and the atomic coordinates for the scaffold subunits Csn1 , Csn3 , Csn7 and Csn8 , the winged-helix domains of Csn2 and Csn4 ( Figure 1—figure supplement 2B ) , and the helical bundle formed by the C-termini of all eight CSN subunits ( Figure 1—figure supplement 2C ) as well as the expected recovery of secondary structure at this resolution . Similarly , there was a very good overlap between the coordinates of Cul1 ( with the exception of helix29 and the WHB domain , see below ) and Skp1 and the corresponding electron density segments ( Figure 1—figure supplement 2D ) . However , the local resolution was lower without recovery of secondary structure in the N-terminal domain of Cul1 . Moreover , the density of the substrate receptor Skp2/Cks1 was poorly defined ( Figure 1—figure supplement 2D ) , indicating a potential flexibility in this region . Since the presence of Skp1/Skp2 had modest effects on the affinity and deneddylation activity ( see below ) , we did not interpret this observation further . In contrast to the large segments of CSN that were unaltered upon binding substrate , there was nearly no overlap between the EM density map and the N-terminal portions of Csn2 and Csn4 , as well as the MPN-domains of Csn5 and Csn6 , the RING domain of Rbx1 , the WHB domain of Cul1 , and Nedd8 ( Figure 1—figure supplement 2E ) . We thus docked these domains individually ( Figure 1B , Video 1 ) . A Csn2 N-terminal fragment encompassing the portion between its crystallographically resolved N-terminus ( amino acid 30 ) through to a flexible loop at amino acid 180 , was docked as a rigid body ( Figure 1—figure supplement 2F ) , positioning it close to the four-helical bundle and helix 24 of Cul1 ( Zheng et al . , 2002 ) . An N-terminal fragment of Csn4 , spanning amino acids 1 to 295 , which ends in a previously reported hinge loop ( Lingaraju et al . , 2014 ) , was also docked independently as a rigid body ( Figure 1—figure supplement 3A ) . The resulting conformation of Csn4 resembles a crystal form of Csn4 observed in isolation ( Lingaraju et al . , 2014 ) . The two N-terminal helical repeat motifs of Csn4 make contacts with the winged-helix A domain of Cul1 ( Figure 1B and Figure 1—figure supplement 3A , right hand panel , red arrow and green circle ) . Moreover , these positions of Csn2 and Csn4 delineated a density in the map , which could accommodate the RING domain of Rbx1 ( Figure 1B and Figure 1—figure supplement 3A , right hand panel , black ellipse ) , with the RING proximal to two conserved helices between amino acids 160 and 197 of Csn4 ( Figure 1B and Figure 1—figure supplement 3B , black arrow ) and a loop in Csn2 located between residues 289 and 306 . The exact orientation of the RING domain awaits a structure at higher resolution . 10 . 7554/eLife . 12102 . 007Video 1 . Morphing CSN and Cul1-N8/Rbx1 conformational changes , occurring upon binding . Color code as in Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 007 To improve the fit of Csn5 and Csn6 , we moved their MPN domains as rigid bodies into the neighboring map segment of similar shape and dimensions ( Figure 1—figure supplement 3C ) . The local resolution in this region was lower , presumably due to higher flexibility around the catalytic site . Importantly , after docking Csn5 , we observed two empty neighboring densities ( Figure 1—figure supplement 3C , right hand panel , circles ) , which accommodated the two yet undocked protein components – Nedd8 and the WHB domain of Cul1 ( Figure 1—figure supplement 3D ) . The docking of the latter was enabled by allowing helix 29 and the WHB , amino acids 690 to the C-terminus of Cul1 , to move as a rigid body towards Csn5 . However , we did not observe an electron density around helix 29 of Cul1 , consistent with a structural flexibility in this region . This model places the neddylated WHB domain in close binding proximity to the RING domain , as well as both INS1 and INS2 of Csn5 . The hydrophobic patch of Nedd8 is facing INS1 , and not the WHB domain , as has been reported for the isolated neddylated C-terminal domain of Cul5 ( Duda et al . , 2008 ) . Similar to the RING domain , we cannot be fully certain about the exact orientations of Nedd8 and the WHB domain at the present resolution . Nevertheless , to further substantiate this docking , we mutagenized conserved charged residues in the INS2 domain of Csn5 as well as the WHB domain , and as expected all of these constructs showed reduced catalytic activity in deneddylation assays ( Figure 1—figure supplement 3E , F ) . We sought orthogonal experimental validation for the molecular docking of the individual subunits and domains in the electron density map by performing cross-linking coupled to mass spectrometric analysis of the cross-linked peptides ( Leitner et al . , 2014 ) following the procedure described in Birol et al . , ( 2014 ) ( Supplementary files 1–6 ) . For the cross-linker used in this study ( disuccinimidylsuberate H12/D12 ) , the maximum predicted distance between two cross-linked lysine residues is generally accepted to be below ~30 Å ( Politis et al . , 2014 ) . As shown in Supplementary file 1 , out of the 39 high-confidence inter-subunit cross-links detected within the CSN5H138A–N8-SCFSkp2/Cks1 complex at a false discovery rate ( FDR ) of 5 percent , the great majority was within regions of modeled atomic structure and only six links exhibited a distance larger than 30 Å when mapped onto our model . However , all of these larger-distance links are connected to the flexibly positioned Skp2 density . Moreover , we further performed similar cross-linking experiments on a number of different CSN-CRL complexes , varying the substrate receptor , the cullin and the neddylation state ( Supplementary files 2–6 ) . All results were consistent with the architecture proposed here for CSN5H138A-N8-SCFSkp2/Cks1 . Intriguingly , when taking into account cross-links with an FDR of up to 0 . 25 ( Supplementary files 2 and 3 ) , we found two cross-links that support proximity of K290 in Csn4 and K89 in the RING domain ( Supplementary file 2 ) , as well as K32 in Csn4-NTD and K587 in Cul1 , which is in the immediate vicinity of the WHA domain of Cul1 ( Supplementary file 3 ) , as suggested by our EM reconstruction . To understand how the structure of CSN and the CSN–SCF complex relates to substrate binding and the mechanism of deneddylation , we sought to develop quantitative binding assays to measure interaction of CSN with its substrates and products . To this end , the environmentally-sensitive dye dansyl was conjugated to the C-terminus of Cul1 using ‘sortagging’ ( Theile et al . , 2013 ) to generate dansylated Cul1/Rbx1 ( Cul1d/Rbx1 ) ( Figure 2—figure supplement 1A ) . Cul1d/Rbx1 exhibited normal E3 activity ( Figure 2—figure supplement 1B ) and bound CSN with an affinity similar to Cul1/Rbx1 based on their IC50 values for competitive inhibition of a deneddylation reaction ( Figure 2—figure supplement 1C; Emberley et al . , 2012 ) . When Cul1d/Rbx1 was incubated with CSN ( all CSN preparations used in this work are shown in Figure 2—figure supplement 1D ) , we observed an increase in dansyl fluorescence ( Figure 2A ) . This signal was due to specific binding because it was chased upon addition of excess unlabeled Cul1/Rbx1 ( Figure 2A , titration shown in Figure 2—figure supplement 2A ) or Cand1 ( Figure 2—figure supplement 2B ) , which competes for substrate deneddylation by CSN ( Emberley et al . , 2012; Enchev et al . , 2012 ) . Thus , we concluded that the increase in dansyl fluorescence accurately reported on the interaction of CSN with Cul1d/Rbx1 . Using this assay we determined that CSN bound Cul1d/Rbx1 with a Kd of 310 nM ( Figure 2B ) . Cul1d/Rbx1 binding to CSN was only modestly affected by the addition of free Nedd8 ( Figure 2—figure supplement 2C ) or assembly with Skp2/Skp1 ( Figure 2—figure supplement 2D ) or Fbxw7/Skp1 ( Figure 2—figure supplement 2E ) . 10 . 7554/eLife . 12102 . 008Figure 2 . Development and validation of a binding assay for CSN–Cul1/Rbx1 interaction . ( A ) Equilibrium binding of CSN to Cul1d/Rbx1 and competition by unlabeled Cul1/Rbx1 . The indicated proteins were mixed and allowed to equilibrate prior to determination of dansyl fluorescence in a fluorometer . Note that Cul1/Rbx1 blocks the fluorescence enhancement caused by CSN . CSN , Cul1d/Rbx1 , and Cul1/Rbx1 were used at 350 , 30 , and 4000 nM , respectively . ( B ) Equilibrium binding of CSN to Cul1d/Rbx1 . Cul1d/Rbx1 ( 30 nM ) was mixed with increasing concentrations of CSN and the proteins were allowed to equilibrate prior to determining the change in dansyl fluorescence in triplicate samples . Error bars represent standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 00810 . 7554/eLife . 12102 . 009Figure 2—figure supplement 1 . Supporting data for development and validation of CSN–Cul1d/Rbx1 binding assay , part 1 . ( A ) Dansylation of Cul1/Rbx1 constructs . Upper panel: dansylation of bacterially expressed and purified Cul1/Rbx1 . Lower panel: dansylation of Cul1/Rbx1 expressed and purified from insect cells . For details , see Materials and Methods . ( B ) Ubiquitination of 32P-labeled β-catenin substrate peptide by dansylated SCFβ-TrCP was monitored as described ( Saha and Deshaies , 2008 ) . The kcat measured here ( 0 . 048 min-1 ) compares favorably with that previously determined for wild type unmodified SCF ( 0 . 054 min-1 ) ( Saha and Deshaies , 2008 ) . ( C ) IC50 study of the inhibitory effects of unlabeled ( red ) or dansylated ( black ) product . Cul1/Rbx1 and Cul1d/Rbx1 were separately titrated into a deneddylation reactions containing 50 nM Cul1-[32P]N8/Rbx1 substrate and 0 . 5 nM CSN , and the resulting reaction rate was measured . ( D ) CSN preparations used in this study . 600 ng of each sample were fractionated by SDS-PAGE and stained with SYPRO Ruby . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 00910 . 7554/eLife . 12102 . 010Figure 2—figure supplement 2 . Supporting data for development and validation of CSN–Cul1d/Rbx1 binding assay , part 2 . ( A ) IC50 for competitive inhibition of CSN-Cul1d/Rbx1 complex formation by unlabeled Cul1/Rbx1 ( ~390 nM ) agrees with the Kd measured for binding of Cul1d/Rbx1 to CSN ( 310 nM ) . ( B ) Equilibrium binding of 100 nM CSN to 50 nM Cul1d/Rbx1 and competition by 500 nM Cand1 . The indicated proteins were mixed and allowed to equilibrate prior to determination of dansyl fluorescence . ( C–E ) Free Nedd8 and F-box box proteins do not appreciably change affinity of Cul1d/Rbx1 for CSN . Same as Figure 2C , except that either 5 µM free Nedd8 ( C ) , 100 nM Skp2/Skp1 ( D ) or 100 nM Fbxw7/Skp1 ( E ) was included in the binding reaction . All binding and activity measurements reported in this legend were carried out in triplicate and error bars represent standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 010 We next sought to measure binding of neddylated Cul1d/Rbx1 ( Cul1d-N8/Rbx1 ) to CSN but it was not possible because the substrate was rapidly deneddylated . To circumvent this problem , we performed binding assays with the extensively characterized inactive mutant CSN5H138A ( assay confirming loss of activity is shown in Figure 3—figure supplement 1A ) . Remarkably , CSN5H138A bound Cul1d-N8/Rbx1 ~200-fold more tightly than CSN bound Cul1d/Rbx1 ( Kd 1 . 6 nM vs . 310 nM; Figures 3A–B ) . Note that the estimated Kd falls well below the fixed concentration of Cul1d-N8/Rbx1 used in the assay . This introduces greater uncertainty into our estimate but nevertheless we can conclude with confidence that the binding of substrate to CSN5H138A is very tight ( ≤5 nM; see Materials and Methods for further discussion of this matter ) . As reported above for CSN binding to product , addition of Skp2/Skp1 or Fbxw7/Skp1 had comparatively minor effects on affinity ( Figure 3—figure supplement 1B , C ) . Thus , for the sake of simplicity , we used Cul1d/Rbx1 heterodimer for the remaining binding experiments . 10 . 7554/eLife . 12102 . 011Figure 3 . Quantitative determination of enzyme–substrate binding affinities for wild type and mutant proteins . ( A ) Tight binding of CSN5H138A to substrate . Cul1d-N8/Rbx1 and CSN5H138A were mixed and allowed to equilibrate prior to determining the change in dansyl fluorescence . ( B ) Summary of Kd measurements for the indicated CSN complexes tested against unmodified Cul1d/Rbx1 , Nedd8-conjugated Cul1d-N8/Rbx1 or Cul1d-N8/Rbx1∆RING ligand . Boxes shaded in gray indicate combinations that could not be analyzed due to deneddylation during the binding reaction . For some complexes that bound weakly it was not feasible to titrate to saturation and so a lower boundary for Kd is indicated . N . D . , not determined . * , due to the configuration of our assay , extremely low Kd values cannot be reliably determined . ( C ) Summary of kon and koff measurements for the indicated CSN complexes tested against Cul1d/Rbx1 or Cul1d-N8/Rbx1 . Each reported koff is the mean of at least 8 replicates . For comparison , koff values calculated from kon and Kd measurements are also shown . For cases where kon was not measured ( marked with asterisks ) , we assumed a value that was the average ( 1 . 83 x 107 M-1 s-1 ) of the three measured kon values . Boxes shaded in gray indicate combinations that could not be analyzed due to deneddylation upon complex formation . N . D . , not determined . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 01110 . 7554/eLife . 12102 . 012Figure 3—figure supplement 1 . Supporting experiments and titration curves for binding data in Figure 3B–C , part 1 . ( A ) CSN5H138A is inactive and is a dominant-negative inhibitor of deneddylation . CSN , CSN5H138A , and substrate were used at 2 nM , 100 nM , and 75 nM , respectively . For reactions containing CSN and CSN5H138A , mutant enzyme was preincubated with substrate for 30 sec prior to initiating time-course by adding CSN . ( B-E ) : The indicated proteins were mixed and allowed to equilibrate prior to determining the change in dansyl fluorescence . ( B ) CSN5H138A and dansylated , Nedd8-conjugated SCFSkp2 . ( C ) CSN5H138A and dansylated , Nedd8-conjugated SCFFbxw7 . Note that addition of Fbxw7–Skp1 greatly increased the variability in the measurement for unknown reasons . ( D ) CSN5H138A ( first prep ) and Cul1d/Rbx1 , ( E ) CSN5H138A ( second prep ) and Cul1d/Rbx1 , ( F-I ) : The indicated CSN complexes were preincubated with Cul1d/Rbx1 for 10 min , followed by addition of unlabeled Cul1/Rbx1 chase and measurement of the decay in dansyl fluorescence over time . Final protein concentrations are listed for each experiment . ( F ) CSN ( 2000 nM ) , Cul1d/Rbx1 ( 200 nM ) , and Cul1/Rbx1 ( 3000 nM ) , ( G ) CSN5E104A ( 600 nM ) , Cul1d/Rbx1 ( 200 nM ) , and Cul1/Rbx1 ( 3000 nM ) , ( H ) CSN5E76A , 5H138A ( 400 nM ) , Cul1d/Rbx1 ( 200 nM ) , and Cul1/Rbx1 ( 3000 nM ) , ( I ) CSN5E76A , 5H138A ( 200 nM ) , Cul1d-N8/Rbx1 ( 100 nM ) , and Cul1/Rbx1 ( 1500 nM ) , ( J–K ) : The indicated proteins were mixed and allowed to equilibrate prior to determining the change in dansyl fluorescence . ( J ) CSN5E76A , 5H138A and Cul1d/Rbx1 , ( K ) CSN5H138A or CSN2∆N , 5H138A and Cul1d-N8/Rbx1 . All measurements in panels B–E and J–K were carried out in triplicate and error bars represent standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 01210 . 7554/eLife . 12102 . 013Figure 3—figure supplement 2 . Supporting experiments and titration curves for binding data in Figure 3B-C , part 2 . ( A–H ) : The indicated proteins were mixed and allowed to equilibrate prior to determining the change in dansyl fluorescence . ( A ) CSN4∆N , 5H138A and Cul1d/Rbx1 , ( B ) CSN4∆N , 5H138A and Cul1d-N8/Rbx1 , ( C ) CSN5E76A or CSN5E76A , 5H138A and Cul1d-N8/Rbx1∆RING , ( D ) CSN5E104A and Cul1d/Rbx1 , ( E ) CSN5E76A and Cul1d/Rbx1 , ( F ) CSN5H138A , 6∆INS2 and Cul1d/Rbx1 , ( G ) CSN5H138A , 6∆INS2 and Cul1d-N8/Rbx1∆RING , ( H ) CSN5T103I and Cul1d/Rbx1 . All measurements in panels A–D and F–H were carried out in triplicate and error bars represent standard deviation . The measurement in panel E was performed in duplicates but the experiment was repeated on three independent occasions , obtaining similar results . Several of these results were independently confirmed in Zurich and Pasadena including panels Figure 3—figure supplement 1J , and panels Figure 3—figure supplement 2B , D , E and F . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 013 The strikingly high affinity we observed for binding of CSN5H138A to Cul1d-N8/Rbx1 led us to question whether it was mainly due to Nedd8 or whether the H138A mutation might also enhance affinity . To this end , we measured binding of CSN5H138A to Cul1d/Rbx1 and observed an unexpectedly low Kd of ~10 nM ( Figure 3B , Figure 3—figure supplement 1D ) , which was confirmed with an independent preparation of CSN5H138A ( Figure 3—figure supplement 1E ) . Thus , neddylation improved affinity of Cul1d/Rbx1 for CSN5H138A by ~6-fold , whereas the Csn5-H138A mutation improved affinity for Cul1d/Rbx1 by ~30-fold . The high affinity binding of CSN5H138A to substrate was supported by an orthogonal competition experiment in which 100 nM CSN5H138A completely blocked deneddylation of 75 nM Cul1-N8/Rbx1 ( Figure 3—figure supplement 1A ) . We considered the possibility that the Csn5-H138A mutation might enable formation of an aberrant , super-tight enzyme:substrate ( [ES] ) complex that does not normally form between the wild type proteins . However , as will be described later on , this hypothesis was rejected based on kinetic arguments . We next sought to determine whether the large differences we observed in Kd values were due to differences in kon or koff . Remarkably , despite a 200-fold difference in Kd for CSN5H138A binding to substrate compared to CSN binding to product , the kon values for formation of these complexes were nearly identical ( 2 . 0 x 107 M-1 sec-1 for CSN–product and 2 . 2 x 107 M-1 sec-1 for CSN5H138A–substrate; Figure 3C ) . This suggested that the difference in affinity was driven by a large difference in koff . To test this hypothesis , we directly measured koff values for select [ES] and enzyme-product complexes by pre-forming the complex and then adding excess unlabeled Cul1/Rbx1 chase and monitoring the reduction in dansyl fluorescence over time ( Figure 3C and Figure 3—figure supplement 1F–I; for this and a subsequent experiment in Figure 4B , we used CSN5E76A/ 5H138A in one of the assays instead of CSN5H138A; the double mutant behaved like CSN5H138A in that it bound Cul1d/Rbx1 with the same affinity as shown in Figure 3—figure supplement 1J ) . Consistent with the predictions from the Kd and kon values , substrate dissociated very slowly from CSN5E76A , 5H138A , whereas product dissociated ~65-fold faster from CSN . This suggests that as substrate is deneddylated to product , its affinity for CSN is strongly reduced and its koff speeds up . 10 . 7554/eLife . 12102 . 014Figure 4 . The N-terminal domains of Csn2 and Csn4 and the RING domain of Rbx1 play key roles in substrate binding and deneddylation . ( A ) Generation of Cul1-N8/Rbx1∆RING . Top: a TEV protease site was engineered between the N-terminal β-strand and the RING domain of Rbx1 as indicated . Only the first 50 amino acids of Rbx1 are shown . Bottom: Purified protein was subjected to the indicated treatments ( see Materials and Methods for details ) and reactions were fractionated by SDS-PAGE and stained with Coomassie Blue . ( B ) Deletion of the Csn4-NTD and Rbx1-RING domains independently reduce affinity of CSN for substrate . The indicated proteins were mixed and allowed to equilibrate prior to determining the change in dansyl fluorescence in triplicate samples . Error bars represent standard deviation . Kd values measured in this experiment are also reported in Figure 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 01410 . 7554/eLife . 12102 . 015Figure 4—figure supplement 1 . Biochemical characterization of Cul1/Rbx1TEV∆RING proteins . ( A ) Ubiquitination assay using the indicated Cul1-N8/Rbx1 variants ( 500 nM each ) as an E3 . Each reaction contained , in addition , 100 nM Ube1 , 1000 nM Cdc34b , 750 nM Skp1/Fbxw7 and 4000 nM CyclinE phosphopeptide , labeled with FAM . The samples were incubated at 25 °C for the indicated time points , analyzed by SDS PAGE and visualized by excitation at 473 nm . ( B ) Overlay of Superdex 200 size exclusion profiles of purified Cul1/Rbx1 variants isolated from insect cells . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 015 Armed with assays to measure binding and deneddylation of substrate , we next sought to test the implications that emerged from our structural analysis of the CSN5H138A–N8-SCFSkp2/Cks1 complex . First , we investigated the roles of the NTDs of Csn2 and Csn4 , both of which , upon binding substrate , underwent conformational changes and made contact with Cul1 and the RING domain of Rbx1 ( Figures 1B–C , Figure 1—figure supplement 2F and 3A ) ( Lingaraju et al . , 2014 ) . To measure the effect of these mutations on binding to Cul1d-N8/Rbx1 , we combined them with Csn5-H138A to prevent deneddylation . Deletion of the first 269 amino acids of Csn2 , observed to interact with Cul1 but not the RING domain of Rbx1 , caused a massive loss in binding to substrate ( Kd > 1300 nM; Figure 3B , Figure 3—figure supplement 1K ) . Thus , the contact we observed between Csn2-NTD and N8-SCFSkp2/Cks1 was critical to formation of the [ES] complex . By contrast , deletion of the first 297 amino acids NTD of Csn4 ( 4∆N ) , a portion which was observed to form interfaces with both Cul1 and the RING domain of Rbx1 , had a relatively modest effect; CSN4∆N , 5H138A bound Cul1d/Rbx1 and Cul1d-N8/Rbx1 with Kd values of > 750 nM and 20 nM , respectively ( Figure 3B , Figure 3—figure supplement 2A , B ) . In addition to the motions of the Csn2 and Csn4 NTDs , our structural analysis revealed formation of substantial interfaces between CSN and the RING domain of Rbx1 . To test the role of the RING domain in complex formation , we generated both Cul1/Rbx1 and Cul1d/Rbx1 in which the RING domain can be deleted by introducing a TEV protease cleavage site ( Dougherty et al . , 1989 ) after residue 37 of Rbx1 to generate Cul1 ( or Cul1d ) /Rbx1TEV ( Figure 4A ) . This was essential , because it would not be possible to conjugate Nedd8 to Cul1/Rbx1 expressed as a mutant lacking the RING domain . After conjugating Nedd8 to the purified complex , we treated it with TEV protease to remove the RING domain , yielding Cul1 ( or Cul1d ) -N8/Rbx1∆RING ( Figure 4A ) . The truncated Cul1/Rbx1∆RING was inactive in an ubiquitylation assay ( Figure 4—figure supplement 1A ) but behaved as a monodisperse sample with the expected hydrodynamic radius upon size exclusion chromatography ( Figure 4—figure supplement 1B ) . Notably , Cul1d-N8/Rbx1∆RING bound CSN5E76A , 5H138A and CSN5E76A with affinities ( 12 nM and 13 nM respectively; Figure 3B , Figure 3—figure supplement 2C ) similar to that observed for binding of wild type Cul1d-N8/Rbx1 to CSN4∆N , 5H138A . Given the similar effects of the Csn4-∆NTD and Rbx1-∆RING mutations on complex formation , we next tested whether their effects arose from loss of the interface that forms between these domains ( Figure 1—figure supplement 3B ) . However , double mutant analysis suggested that the Csn4-∆N and Rbx1-∆RING mutations had largely independent effects on binding ( Figure 4B ) . The overall picture that emerged from these studies in light of the structural data is that the interaction of Csn2-NTD with neddylated substrate makes a large contribution to binding energy , with modest enhancements independently provided by the Csn4-NTD and Rbx1-RING domains . The striking difference in the Kd for CSN5H138A binding to substrate compared to CSN binding to product suggested that a conformational rearrangement of the [ES] complex occurs upon cleavage of the isopeptide bond , resulting in a large increase in the product koff , thereby preventing the enzyme from becoming product-inhibited . However , we were puzzled by the relatively minor impact of Nedd8 on the affinity of Cul1d/Rbx1 for CSN5H138A; whereas substrate bound with Kd of 1 . 6 nM , product binding was only ~6-fold weaker ( Figure 3B ) . Why , then , did CSN bind so much less tightly to product ? We reasoned that a key difference between CSN5H138A and CSN is the absence of the active site zinc from CSN5H138A , which prevents formation of a stable apo-CSN complex in which E104 of the INS1 domain of Csn5 is bound to the active site zinc . If this conjecture is correct , it makes the prediction that CSN5E104A , which should also be unable to form stable apo-CSN , should likewise exhibit high affinity for product . This was confirmed: CSN5E104 bound Cul1d/Rbx1 with a Kd of 26 nM ( Figure 3B , Figure 3—figure supplement 2D ) . Furthermore , measurement of koff values revealed that product dissociated from CSN5H138A and CSN5E104A about eightfold more slowly than it dissociated from CSN ( Figure 3C ) . Based on these observations , we propose the ‘E-vict’ hypothesis , which is described in more detail in the Discussion . The essence of this hypothesis is that , following cleavage of the isopeptide bond and dissociation of Nedd8 , INS1 of Csn5 engages the active site zinc . This accelerates the rate of dissociation of deneddylated Cul1/Rbx1 , thereby preventing CSN from becoming clogged with product . We note that Csn5-E76 also contributes to the operation of this mechanism , because CSN5E76A bound tightly to product ( Figure 3—figure supplement 2E ) . We speculate that engagement of the active site zinc by Csn5-E104 forces Csn5-E76 into a configuration that promotes egress of product . Further insights into the exact sequence of events that accelerates product dissociation await high-resolution structures of CSN bound to Cul1/Rbx1 in various states . We next sought to address the effects of the enzyme and substrate mutations described in the preceding sections on the deneddylation reaction . We previously showed that CSN2∆N has severely reduced catalytic activity ( Enchev et al . , 2012 ) , which is consistent with the binding data reported here . CSN4∆N exhibited a 20-fold defect in substrate cleavage ( Figure 5A , Figure 5—figure supplement 1A ) . Meanwhile , the kcat for cleavage of Cul1-N8/Rbx1∆RING by CSN was reduced by a staggering ~18 , 000-fold relative to wild type substrate ( Figures 5A–B ) . Given that the neddylated ∆RING substrate bound to CSN with only modestly reduced affinity , we surmised that the principal defect of this mutant might be its failure either to induce the activating conformational change in CSN , and/or to position accurately the isopeptide bond in the active site . Although we do not have the tools to address the latter point , we queried the former by examining the Csn6-∆INS2 mutation , which partially mimics the effect of substrate binding in that it destabilizes the autoinhibited state ( Lingaraju et al . , 2014 ) . The Csn6-∆INS2 mutation slightly weakened binding to wild type product ( Figure 3B , Figure 3—figure supplement 2F ) but completely suppressed the modest binding defect of the neddylated ∆RING substrate ( Figure 3B , Figure 3—figure supplement 2G ) and promoted an ~8-fold increase in its deneddylation rate ( Figure 5—figure supplement 1B ) . This partial suppression effected by Csn6-∆INS2 suggests that the RING domain contributes to the constellation of conformational changes in CSN that occur upon substrate binding . Note that the CSN6∆INS2 enzyme nevertheless exhibited a >1000-fold defect towards the Cul1-N8/Rbx1∆RING substrate , strongly indicating further functions of the RING domain , which may include a potential role in substrate positioning as well . 10 . 7554/eLife . 12102 . 016Figure 5 . The N-terminal domains of Csn2 and Csn4 and the RING domain of Rbx1 are important for CSN-mediated deneddylation . ( A ) Summary of kinetic parameters for the indicated CSN mutants in multi- or single-turnover deneddylation reactions with Cul1-N8/Rbx1 or Cul1-N8/Rbx1∆RING substrate . Note that there may be modest discrepancies between these kcat values and koff values due to differences in assay configurations as described in Materials and Methods . The ∆RING substrate used here and in panel B contains the sortase sequence at the C-terminus of Cul1 that was used for generation of dansylated Cul1 . Control experiments revealed that this tag , with or without dansylation , reduced kcat by ~4-fold . In addition the wild type control for the ∆RING reaction exhibited kcat of 2 . 6 s-1 . The rates shown have been correspondingly adjusted to normalize them to other rates reported here . * , This rate is estimated from Figure 5—figure supplement 1B . ( B ) Kinetic analysis of deneddylation of Cul1-N8/Rbx1∆RING by CSN . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 01610 . 7554/eLife . 12102 . 017Figure 5—figure supplement 1 . Kinetic analysis of deneddylation . ( A ) Deneddylation reactions were carried out in triplicate with CSN4∆N at varying concentrations of Cul1-[32P]N8/Rbx1 substrate and quantified to generate the curve shown . Estimates of kcat and KM are indicated . ( B ) Deneddylation assays of Cul1-N8/Rbx1∆RING ( 100 nM ) , incubated with CSN ( 200 nM , upper panel ) or CSN6∆INS2 ( 200 nM , lower panel ) . Samples were taken at the indicated time points , and visualized by SDS PAGE and Sypro Ruby staining . Note that the ∆RING substrate contained an unreacted Sortase tag at the C-terminus of Cul1 that reduced kcat by ~4-fold . ( C ) Multi-turnover deneddylation reactions were carried out with CSN or CSN5E104A and Cul1-[32P]N8/Rbx1 . Substrate was assayed at 1 and 1 . 3 µM to confirm that saturation was achieved . ( D ) Single-turnover deneddylation reactions were carried out with CSN on Cul1-[32P]N8/Rbx1 +/- Skp1/Skp2 , and with CSN5E104A on Cul1-[32P]N8/Rbx1 . ( E ) Same as panel D except that CSN6∆INS2 was also evaluated . ( F ) Multi-turnover deneddylation reactions were carried out in triplicate with CSN5T103I at varying concentrations of Cul1-[32P]N8/Rbx1 substrate and quantified to generate the curve shown . Estimates of kcat and KM are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 017 A noteworthy feature of the deneddylation reactions carried out with CSN4∆N enzyme or ∆RING substrate is that although kcat was reduced in both cases , KM was also reduced ( Figure 5A ) . Whereas these results imply that deletion of the Csn4-NTD or Rbx1-RING improved affinity of the [ES] complex , our direct binding measurements indicated this was not the case . To understand this apparent paradox , it is essential to consider the kinetic behavior of CSN-mediated deneddylation . The formal definition of KM for a deneddylation reaction ( Equation 1 ) , as stipulated by Briggs and Haldane ( Briggs and Haldane , 1925 ) , is: KM = ( koff + kcat ) /kon . . In the special case of Michaelis-Menten kinetics , which is based on the assumption that koff is much larger than kcat the expression simplifies to koff/kon , or Kd . However , kcat for CSN ( ~1 . 1 sec-1 ) is actually much faster than koff measured for dissociation of substrate from the CSN5E76A , 5H138A mutant ( 0 . 017 sec-1 ) . The implication of this is that almost every binding event between CSN and substrate results in catalysis , and KM ( 200 nM; Figure 5A and ( Emberley et al . , 2012 ) is much larger than Kd ( 1 . 6 nM , Figure 3B ) . But , if kcat is reduced by mutation , the Briggs-Haldane equation predicts that KM should approach Kd . Indeed , this is exactly what we see for reactions that exhibit reduced kcat , including reactions with mutant CSN4∆N enzyme or mutant ∆RING substrate ( Figure 5A ) . In the slowest reaction ( cleavage of Cul1/Rbx1∆RING by CSN ) the KM ( 5 nM ) is in the same range as the Kd with which this substrate bound to CSN5E76A , H138A ( 12 nM; Figure 3B ) , and approaches the Kd measured for binding of substrate to CSN5H138A ( 1 . 6 nM ) . This provides strong support for our proposal that the CSN5H138A–Cul1d/Rbx1 complex is representative of the affinity that develops during normal catalysis . ( 1 ) [CSN]+[Cul1-N8/Rbx1]⇄koffkon [CSN/Cul1-N8/Rbx1] →kcat [CSN/Cul1 | N8/Rbx1] To understand the significance of the E-vict mechansim to CSN function in vitro and in cells , we measured the kcat for CSN5E104A and observed that it is 2 . 5-fold slower than for CSN ( Figure 5A , Figure 5—figure supplement 1C–E ) . This was unexpected , because it was reported that the Csn5-E104A mutation enhances the catalytic activity of CSN towards an unnatural substrate ( Lingaraju et al . , 2014 ) . Interestingly , a similar reduced rate was observed in both single- and multi-turnover reactions , indicating that under our specific reactions conditions , the activating conformational changes/chemical step are affected at least as much as product dissociation . This may not be the case in cells , where substrate receptors and other factors may further stabilize product binding . To test if Csn5-E104 contributes to CSN function in vivo , we generated a partial knockout of Csn5 in HEK293T cells using CRISPR/Cas9 ( Shalem et al . , 2014 ) . This cell line expressed severely reduced levels of Csn5 and consequently displayed hyper-accumulation of Nedd8-conjugated endogenous Cul1 ( Figure 6A ) , but retained sufficient protein to survive . We introduced either an empty retrovirus or retroviruses coding for Flag-tagged wild type or mutant Csn5 proteins into these cells , and then monitored the Cul1 neddylation status by immunoblotting . In contrast to wild type FlagCsn5 , cells expressing FlagCsn5-E104A , H138A or E76A did not regain a normal pattern of Cul1 neddylation ( Figure 6A ) . The same was observed for Cul2 , Cul3 , Cul4A , and Cul5 ( Figure 6—figure supplement 1A ) . Consistent with reduced CSN activity , as revealed by increased cullin neddylation , Skp2 levels were reduced in cells expressing mutant Csn5 proteins ( Figure 6A ) ( Cope and Deshaies , 2006; Wee et al . , 2005 ) . To test whether mutations in the catalytic site of Csn5 resulted in increased affinity for Cul1 , we immunoprecipitated wild type and mutant FlagCsn5 proteins and probed for co-precipitation of endogenous Cul1 . In addition to the mutants described above , we surveyed a much broader panel of catalytic site substitutions to determine whether the results observed in our in vitro experiments were specific to the mutations employed or were a general consequence of disrupting the active site . As shown in Figure 6B , the results were concordant with what was observed in vitro . On the one hand , FlagCsn5–H138A retrieved high levels of Cul1-N8 . The same was true for FlagCsn5 carrying mutations in other core residues of the JAMM domain ( e . g . H140 and D151 ) ( Cope et al . , 2002 ) . On the other hand , FlagCsn5-E104A retrieved high levels of unmodified Cul1 . We propose that this arises from its ability to bind and deneddylate substrate ( albeit at a reduced rate ) , but then remain tightly bound to the product due to loss of the E-vict mechanism . 10 . 7554/eLife . 12102 . 018Figure 6 . Functional analysis of Csn5 active site and INS1 mutants in biochemical and cellular assays . ( A ) Csn5-E104 is important for CSN function in cells . CSN5 alleles in HEK293T cells were partially knocked out ( KO ) by CRISPR/Cas9 to yield a major decrease in Csn5 that was nonetheless compatible with viability . Wild type and the indicated Flag-tagged CSN5 mutants were reintroduced by transduction of recombinant retroviruses that co-expressed GFP . Lysates of transduced cells were separated by SDS-PAGE and blotted with antibodies to the indicated proteins . CSN5 long refers to a long exposure of the Csn5 blot , captured to reveal residual Csn5 in the knock-out cells . # refers to transduced FlagCsn5 and * refers to endogenous Csn5 . ( B ) Any mutation of a core JAMM domain residue in Csn5 results in enhanced binding to Cul1 . Same as ( A ) , except additional Csn5 mutants were tested and the cell lysates were immunoprecipitated with anti-Flag and the immunoprecipitates were blotted for the indicated proteins . ( C ) Csn5-T103 is important for CSN function in cells . Same as ( A ) except that the Csn5-T103I mutant was analyzed in parallel with Csn5-E104A and wild type . ( D ) SILAC mass spectrometry of endogenous proteins bound to FlagCsn5-E104A or FlagCsn5-T103I , relative to wild type FlagCsn5 . Cells expressing mutant and wild type FlagCsn5 proteins were grown in light and heavy medium , respectively . L:H ratios >1 indicate higher recovery of the listed protein from cells expressing mutant FlagCsn5 , whereas ratios <1 indicate higher recovery from cells expressing wild type FlagCsn5 . Gray bars: FlagCsn5-E104A; black bars: FlagCsn5-T103I . Error bars represent the 95% confidence interval as calculated by limma ( Smyth , 2005 ) . Each protein was quantified in at least two of the four biological replicates and error bars represent standard deviations . Ratios indicated by * differed significantly from 1 . 0 ( p<0 . 05 ) . For CSN , only Csn5 is shown; the remainder is shown in Figure 6—figure supplement 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 01810 . 7554/eLife . 12102 . 019Figure 6—figure supplement 1 . Analysis of Csn5 mutations in cells . ( A ) Same as Figure 6A except that samples were immunoblotted for different cullins . ( B ) SILAC data for CSN subunits from pull-down analysis shown in Figure 6D . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 019 The unexpected reduction in activity observed for Csn5-E104A both in vitro and in cells ( Figures 5A , 6A ) suggested that the adjacent residue , T103 , may also be important for deneddylation . A T103I mutation in Drosophila melanogaster impedes proper interaction of photoreceptor neurons with lamina glial cells in the developing brain . If this mutant also causes a loss of CSN deneddylase activity , it would explain the recessive nature of this mutation in flies ( Suh et al . , 2002 ) . Indeed , like Csn5-E104A , Csn5-T103I did not restore a normal Cul1 neddylation pattern when expressed in Csn5-depleted cells ( Figure 6C ) and exhibited low deneddylase activity in vitro ( Figures 5A , Figure 5—figure supplement 1F ) . In contrast to CSN5E104A , however , CSN5T103I bound Cul1d/Rbx1 product with low affinity , both in vitro ( Figure 3B , Figure 3—figure supplement 2H ) and in cells ( Figure 6B ) . Therefore , although CSN5E104A and CSN5T103I both have diminished catalytic activity , their defects appear to have distinct molecular bases . To further explore the divergent effects of Csn5-E104A and Csn5-T103I mutations on Cul1 binding , HEK293T stably expressing wild type or mutant versions of FlagCsn5 were grown in ‘heavy’ SILAC medium ( wild type ) , or ‘light’ SILAC medium ( mutants ) . Each mutant lysate was individually mixed with wild type lysate , and then subjected to immunoprecipitation and SILAC mass spectrometry . Whereas all CSN subunits exhibited light:heavy ratios of ~1 ( Figure 6—figure supplement 1B ) , the FlagCsn5-E104A pull-down showed elevated levels of all cullins compared to wild type , whereas FlagCsn5-T103I pulled down cullins at levels equal to or less than wild type FlagCsn5 ( Figure 6D ) .
Figure 7 displays a model that incorporates published data and data presented in this manuscript . Panels A-C shows a schematic view of the structural transitions that occur upon substrate binding , and collectively contribute to efficient catalysis , whereas panel D provides the rate constants for the deneddylation cycle . We tentatively propose the following sequence of events . Free CSN exists in an inactive state in which E104 of Csn5-INS1 forms a fourth ligand to zinc ( Figure 7A ) ( Lingaraju et al . , 2014 ) . In this state the NTDs of both Csn2 and Csn4 are in “open” conformations relative to the cullin substrate , and the MPN domains of Csn5/Csn6 are in a distal position relative to it . Substrate binds this state rapidly ( Figure 7B ) , likely driven by electrostatic interactions between Cul1 and Csn2-NTD . This would account for the similar , extremely fast kon values that we measured for different combinations of Cul1d/Rbx1 and CSN . Binding of CSN to neddylated substrate results in a series of conformational changes in both complexes ( Figure 7B ) . These include ( i ) the translocation of the N-terminal helical repeats of Csn2 towards the CTD of Cul1 , ( ii ) the movement of the NTD of Csn4 towards the RING domain of Rbx1 and the WHA domain of Cul1 , ( iii ) the translocation of the MPN domains of Csn5–Csn6 towards the neddylated WHB domain of Cul1 , ( iv ) movement of the WHB domain towards Csn5 , ( v ) the opening of the interface between Nedd8 and the WHB domain , and ( vi ) the formation of a new interface between Csn5 and Nedd8 probably involving the hydrophobic patch of Nedd8 and neighboring residues , as well as a tenuous interface between the WHB and the Rbx1 RING domain . Furthermore , although not structurally resolved in the present study , movements of Csn5-E76 and E104 towards and away from the zinc atom ( vii ) , respectively , probably similar to the conformation reported in ( Echalier et al . , 2013 ) , must occur to enable catalysis . Finally , a series of other unresolved movements are likely to be germane including ( viii ) positioning of the extended C-terminus of Nedd8 , and the corresponding portion of the WHB domain for catalysis as well as contacts between the INS1 and INS2 domains of Csn5 and the WHB domain of Cul1 . 10 . 7554/eLife . 12102 . 020Figure 7 . Structural and kinetic models for CSN activation and the CSN enzyme cycle . ( A-C ) Proposed conformational changes that precede substrate cleavage . ( D ) Kinetic model for the deneddylation cycle . The asterisk denotes the activated form of CSN . Numbers in red , black , green , and blue represent koff ( sec-1 ) , kon ( M-1 sec-1 ) , kcat ( sec-1 ) , and conformational change ( sec-1 ) rates , respectively . For rates specified as >1 , the actual rate has not been measured but it is inferred to be >1 sec-1 because the overall rate for multiturnover catalysis is at least 1 . 1 sec-1 and thus all sub-steps must be at least this fast . The koff of SCF from CSN varied depending upon whether the rate was measured directly or inferred from Kd and kon ( see Figure 3C ) . The arrow connecting CSN and N8-SCF•CSN* combines two separate steps: binding of N8-SCF to CSN , and activation of CSN to CSN* . DOI: http://dx . doi . org/10 . 7554/eLife . 12102 . 020 To probe the significance of the conformational changes summarized above , we generated and analyzed mutant enzymes . Deletion of Csn2-NTD virtually eliminated substrate binding ( Figure 3—figure supplement 1K ) , suggesting that movement of this domain ( motion i ) enables a high affinity interaction between CSN and neddylated CRLs . Meanwhile a mutant lacking Csn4-NTD , CSN4∆N , 5H138A , bound Cul1-N8/Rbx1 ~10-fold less tightly than CSN5H138A , albeit still with a relatively high affinity ( 20 nM , Figure 3B ) . A similar effect on binding affinity was seen with a substrate lacking the RING domain of Rbx1 ( Figure 3B ) . Even though the RING and Csn4-NTD domains are adjacent in the enzyme-substrate [ES] complex ( Figure 1 ) , double mutant analysis suggests that they make substantially independent contributions to binding energy ( Figure 4B ) . Interestingly , enzyme assays revealed a much greater effect of deleting the RING than deleting the Csn4-NTD , suggesting that the RING domain makes a profound contribution to catalysis in a manner that does not depend on its proximity to Csn4-NTD . We do not know the extent to which the reduced catalytic rates for these mutants arise from defects in enzyme activation versus substrate positioning , but we note that cleavage of ∆RING substrate was accelerated by ~8-fold upon deletion of Csn6-INS2 , suggesting that at least a small part of the problem with this substrate is that it failed to properly trigger activating conformational changes in CSN . In addition to the movements of individual domains , formation of the [ES] complex is accompanied by wholesale translocation of the Csn5 and Csn6 subunits . We suggest that this motion contributes primarily a kcat effect , because deletion of Csn6-INS2 , which is proposed to facilitate this motion , enhanced kcat but had no noteworthy impact on binding to substrate ( Figure 3B ) , and removal of Csn5 from the complex did not substantially affect CSN assembly with substrate ( Enchev et al . , 2012 ) . We cannot conclude much about the other motions enumerated above , but we note that mutations that are predicted to reside near the interface of the Csn5-INS2 and Cul1-WHB domains cause significant reductions in substrate deneddylation ( Figure 1—figure supplement 3E , F ) . In addition , reorientation of Nedd8 away from Cul1-WHB and towards Csn5 as predicted here is consistent with the prior observation that the hydrophobic patch of Nedd8 recruits UBXD7 to neddylated CRLs ( den Besten et al . , 2012 ) . Presumably , the conformational changes that occur during the activation process are connected in some manner . Interestingly , CSN5E104A and CSN6∆INS2 both cleave ubiquitin-rhodamine at 0 . 04 sec-1 ( which is ~6-fold faster than wild type CSN ) , but CSN5E104A , 6∆INS2 is yet fivefold faster ( 0 . 2 sec-1 ) than either single mutant ( Lingaraju et al . , 2014 ) . The activities of the single and double mutants imply that the Csn6-∆INS2 mutation must destabilize binding of Csn5-E104 to the catalytic zinc , but only in a small fraction ( ≤20% ) of complexes . Meanwhile , movements at the Csn4/6 interface must do more to the active site than simply disrupt the interaction of Csn5-E104 with the catalytic zinc , implying the existence of at least two inputs to CSN activation . Resolving how binding of substrate is connected to enzyme activation awaits high-resolution structural analyses of the enzyme and substrate in various states . Upon formation of an [ES] complex , the conformational changes that occur in both CSN and substrate culminate in cleavage of the isopeptide bond that links Nedd8 to cullin . Although we don’t know the microscopic rate constants for the various conformational changes and bond cleavage , all evidence points to the former being slower than the latter , which can occur with k ≥ 6 . 3 sec-1 , based on the kcat for cleavage of N8-CRL4ADDB2 by CSN6∆INS2 ( Lingaraju et al . , 2014 ) . The actual cleavage may be even faster because this measurement was made under multi-turnover conditions , in which case product dissociation may have been rate limiting . Regardless , the sum total rate of the activating conformational motions plus isopeptide bond cleavage reported here ( ~1 s-1 ) is considerably faster than substrate dissociation from CSN5H138A ( ~0 . 017 s-1 ) , indicating that CSN conforms to Briggs-Haldane kinetics and essentially every [ES] complex that forms proceeds to cleavage , the physiological implications of which are considered in the next section . Cleavage of the isopeptide bond initiates a series of events leading to product release . Removal of Nedd8 increases dissociation of Cul1/Rbx1 by ~7–10 fold . We propose that dissociation of the cleaved Nedd8 also removes an impediment to Csn5-INS1 , which can now bind the catalytic site zinc via E104 to return CSN to its apo state . This engagement , which we refer to as the ‘E-vict’ mechanism , is a critical step in what is likely to be a series of conformational rearrangements that include repositioning of Csn5-E76 . Collectively , these movements reduce the affinity of CSN for product and accelerate its rate of dissociation by an additional order of magnitude . The removal of Nedd8 and E-vict together bring about an ~100-fold loss in affinity of Cul1/Rbx1 for CSN . The slow dissociation of product from CSN mutants that were unable to undergo E-vict ( 0 . 12–0 . 16 s-1; Figure 3C ) suggests that this mechanism is important for maintaining physiological rates of CRL deneddylation . This is further supported by the observation that Csn5-E104A , but not wild type Csn5 , co-precipitates substantial amounts of deneddylated Cul1 from cells ( Figure 6B ) . Slow clearance of product could explain , in part , the failure of this mutant to complement a Csn5 deficiency ( Figure 6A ) . The E-vict mechanism presents an elegant solution to a fundamental challenge facing enzymes: how to achieve high specificity without compromising rapid turnover . We note that the product koff for Cul1d/Rbx1 ( 1 . 1 s-1 ) is similar to the kcat we measured for both single- and multi-turnover reactions . This suggests that depending on the exact structure of the neddylated CRL substrate , the rate-limiting step may vary from one deneddylation reaction to another . Regardless , our biochemical and cell-based data suggest that if the E-vict mechanism did not exist , product dissociation would become the Achilles heel of deneddylation reactions . The kinetic parameters reported here coupled with quantitative measurements of protein concentrations by selected reaction monitoring mass spectrometry ( [Bennett et al . , 2010] and JR and RJD , unpublished data ) allow a preliminary estimate of the steady-state distribution of CSN in cells . The total cullin concentration in the 293T cell line used in this work is ~2200 nM . Meanwhile , the CSN concentration is ~450 nM . Although the total amount of Nedd8-conjugated cullins was not measured , immunoblot data suggest that ~1000 nM is a reasonable estimate . The Kd reported here for the [ES] complex ( ~2 nM ) , thus predicts potentially near-complete saturation of the cellular CSN pool with neddylated cullins . This implies that formation of new [ES] complexes is limited by the slowest step in the catalytic cycle , i . e . either the conformational rearrangements or product dissociation . In vitro , CSN follows Briggs-Haldane kinetics and cleaves Nedd8 off nearly every neddylated CRL that it binds . Because CSN is not in equilibrium with its substrates in our simplified in vitro system , it cannot rely on differences in substrate Kd to achieve specificity . Thus , differences in koff on the order of ≤10-fold , which might occur with different cullins or substrate adaptors , would be predicted to have minimal effects on catalytic efficiency provided that kon remains roughly the same , as was observed for different configurations of substrate and product in this study . Importantly , this parameter can potentially be profoundly altered by ubiquitylation substrates , E2 enzyme , or other in vivo binding partners of Nedd8-conjugated CRLs , which compete with CSN ( Emberley et al . , 2012; Enchev et al . , 2012; Fischer et al . , 2011 ) and thus should reduce its apparent kon . It is also conceivable that binding partners might alter the partitioning of the CSN–N8-CRL complex either by increasing koff and/or reducing kcat , such that N8-CRL bound to an ubiquitylation substrate dissociates prior to completion of the conformational rearrangements that culminate in its deneddylation . Based on measurements reported here , it is likely that CSN complexes in cells are constantly undergoing catalysis , dissociating rapidly from product , and rebinding other CRLs on the time-scale of a few seconds or less . Consistent with this picture , addition of a Nedd8 conjugation inhibitor to cells leads to nearly complete disappearance of neddylated cullins within 5 min , and this does not account for the time it takes the drug to equilibrate across the membrane and deplete the cellular pool of Nedd8~Ubc12 thioesters ( Soucy et al . , 2009 ) . The dynamic properties of CSN measured here reveal a CRL network of extreme plasticity that can be reconfigured in minutes to respond to changing regulatory inputs . Although quantitative studies of CRL network dynamics remain in their infancy , it is evident that the tools are at hand to begin to understand how these remarkable enzymes function and are regulated within cells .
All eight wild type CSN subunits were cloned into a single pFBDM baculovirus transfer MultiBac vector ( Berger et al . , 2004 ) . His6-Csn5 was inserted into the first multiple cloning site ( MCS1 ) of a pFBDM vector using NheI/XmaI and Csn1 was put into MCS2 of the same vector with BssHII/NotI . Similarly , Csn2 was inserted into a second pFBDM vector using BssHII/NotI and StrepII2x-Csn3 , containing an N-terminal PreScission-cleavable StrepII2x-tag , using NheI/XmaI . From this plasmid the Csn2/StrepII2x-Csn3 gene cassette was excised out with AvrII/PmeI and inserted into pFBDMCsn1/His6Csn5 , whose multiplication module had been linearized with BstZ17I and SpeI , yielding pFBDMCsn1/His6-Csn5/Csn2/StrepII2x-Csn3 . A pFBDMCsn4/Csn7b vector was generated using BssHII/NotI to insert Csn4 and NheI/XmaI for Csn7b , and the resultant gene cassette was inserted into linearized pFBDMCsn1/His6-Csn5/Csn2/StrepII2x-Csn3 , resulting in pFBDMCsn1/His6-Csn5/Csn2/StrepII2x-Csn3/Csn4/Csn7b . Finally , a pFBDMCsn6/Csn8 vector was generated using BssHII/NotI for Csn6 and NheI/XmaI for Csn8 insertion . Once again the resultant gene cassette was inserted into linearized pFBDMCsn1/His6-Csn5/Csn2/StrepII2x-Csn3/Csn4/Csn7b , yielding the full wild type CSN vector pFBDMCsn1/His6-Csn5/Csn2/StrepII2x-Csn3/Csn4/Csn7b/Csn6/Csn8 . A similar cloning strategy was applied for the generation of CSN5E76A , CSN5E76A , H138A , CSN5E212R , D213R and CSN4∆N1-297 , except that site-directed mutageneses were performed on pFBDMCsn1/His6Csn5 and pFBDMCsn4/Csn7b respectively . CSN5E104A and CSN5T103I were generated with the same general approach , except that that site-directed mutagenesis and sequence validation were performed on a pCRIITOPO plasmid ( Invitrogen ) containing StrepII2x-Csn5 . Those mutants were then ligated into a MCS1 linearized pFBDMCsn1 plasmid . For the production of CSN6∆Ins2 we used co-expression from two separate viruses . To this end we applied site-directed mutagenesis on the pFBDMCsn6/Csn8 vector to delete amino acids 174–179 in Csn6 , generating pFBDMCsn6∆Ins2/Csn8 . The gene cassette of the latter was excised out using AvrII/PmeI and inserted into BstZ17I/SpeI linearized pFBDMCsn4/Csn7b , yielding pFBDMCsn4/Csn7b/Csn6∆Ins2/Csn8 . The resultant bacmid was used together with a bacmid generated from pFBDMCsn1/His6-Csn5/Csn2/StrepII2x-Csn3 in order to generate two baculoviruses , which were used for co-infection to generate CSN6∆Ins2 . An analogous strategy was applied to generate CSN4∆N/6∆Ins2 , CSN5H138A/6∆Ins2 and CSN5H138A/4∆N . The TEV site in Rbx1 as well as mutations in the WHB domain of Cul1 were obtained by site-directed mutagenesis on the pFBDM-Cul1/Rbx1 vector described in ( Enchev et al . , 2010 ) , which further contained a C-terminal sortase tag described in the next section . Cloning of Cul3/Rbx1 used in the crosslinking/mass spectrometry experiments , Nedd8-pro-peptide-StrepII2x and StrepII2x-Den1 are described in Orthwein et al . , ( 2015 ) . Recombinant bacmid and virus generation as well as protein expression proceeded as described in ( Enchev et al . , 2012 ) . All genes were validated by sequencing as wild type or mutant . CSN and its mutant forms were purified as described in Enchev et al . ( 2012 ) . Nedd8-activating and conjugating enzymes were purified as described in Emberley et al . ( 2012 ) and Enchev et al . ( 2012 ) . Fluorescently-labeled Cul1 substrates were conjugated with untagged Nedd8 . Cul1-sortase was designed with GGGGSLPETGGHHHHHH inserted after the final amino acid of Cul1 into the pGEX vector described in Emberley et al . ( 2012 ) . All sortase reactions were done at 30 °C overnight with 30 μM Cul1/Rbx1 , 50 μM Sortase and 250 μM GGGGK-dansyl in 50 mM Tris pH 7 . 6 , 150 mM NaCl and 10 mM CaCl2 and purified by size exclusion chromatography to yield Cul1d/Rbx1 . Cul1d/Rbx1 was neddylated and purified as in Emberley et al . ( 2012 ) to yield Cul1d-N8/Rbx1 . Cand1 and Sortase were purified as described in Pierce et al . ( 2013 ) . Production of Cul1/Rbx1 and Cul3/Rbx1 baculovirus constructs used for electron microscopy and crosslinking mass spectrometry , bacterial split-and-co-express Cul1/Rbx1∆RING , Nedd8 with native N- and C-termini , used for electron microscopy and crosslinking mass spectrometry and for the experiments involving Cul1/Rbx1TEV , Den1 as well as the respective preparative neddylation were performed as described in Enchev et al . ( 2012 ) and ( Orthwein et al . , 2015 ) . Den1 was used in 1:50 ratio for 10 min at 25°C to remove poly-neddylation . Cul1/Rbx1 complexes with mutations in the WHB domain of Cul1 ( Figure 1—figure supplement 1E , F ) and Cul1/Rbx1TEV∆RING were purified from High Five insect cells as described in Enchev et al . ( 2010 ) . Dansylation of Cul1/Rbx1 variants expressed in insect cells was performed for 8 to 12 hr at 30°C while spinning at 5000 g , and purified by passing the dansylation reaction through a 5 ml HisTrap FF column ( GE Healthcare ) in 50 mM Tris-HCl , pH 7 . 6 , 400 mM NaCl , 20 mM imidazole . The Cul1d/Rbx1-containing flow through was concentrated , neddylated ( if required ) , and further purified over a Superdex 200 size exclusion column ( GE Healthcare ) equilibrated with 15 mM HEPES , pH 7 . 6 , 150 mM NaCl , 2 mM DTT , 2% ( v/v ) glycerol . Neddylation of Cul1/Rbx1TEV∆RING was performed at 25 °C for 12–14 hr in 50 mM Tris-HCl , pH 7 . 6 , 100 mM NaCl , 2 . 5 mM MgCl2 , 150 μM ATP , spinning at 2000 g , and was followed by 30 min incubation with 1:50 ( w/w ) Den1 to remove poly-neddylation . The reaction was purified over a Strep-Tactin Superflow Cartridge ( QIAGEN ) , and eluted in 15 mM HEPES , pH 7 . 6 , 250 mM NaCl , 2 mM DTT , 2% ( v/v ) glycerol , 2 . 5 mM d-desthiobiotin . RING cleavage was performed for 12–14 hr at 25°C , spinning at 2000 g , in the presence of 100 mM EDTA , pH 8 and 1:1 ( w/w ) TEV . Dansylation proceeded as described above . All deneddylation assays were performed in a buffer containing 25 mM Tris-HCl , pH 7 . 5 , 100 mM NaCl , 25 mM trehalose , 1 mM DTT , 1% ( v/v ) glycerol , 0 . 01% ( v/v ) Triton X-100 and 0 . 1 mg/ml ovalbumin or BSA . Radioactive deneddylation reactions with bacterially expressed substrates were done as described ( Emberley et al . , 2012 ) . Radioactive deneddylation reactions with substrates expressed in insect cells were performed at 24 °C with 0 . 5 nM CSN ( Figure 2—figure supplement 1C ) or 2 nM CSN ( Figure 5B ) . All remaining radioactive deneddylation reactions were performed with bacterially expressed Cul1-N8/Rbx1 substrates ( 50 nM ) and 2 nM CSN unless otherwise noted . Single-turnover reactions were done with 25 nM Cul1 substrates and 1 μM CSN on a Kintek RQF-3 Rapid Quench Flow at 24°C . Single-turnover data were fit to one phase decay function: Y= ( Y0 - EP ) *exp ( -kcat*X ) + EP ( where EP corresponds to reaction end point value ) , to determine the kcat . Deneddylation assays in Figure 1—figure supplement 1E , F were performed with 800 nM substrate and 20 nM enzyme and visualized by Coomassie stain . Depending on the exact protein preparations used and the laboratory , we observed rates for the wild type reaction ranging from 1 . 1–2 . 6 sec-1 . All assays were performed in a buffer containing 30 mM Tris pH 7 . 6 , 100 mM NaCl , 0 . 25 mg/ml ovalbumin or BSA and 0 . 5 mM DTT with 30 nM dansyl-labeled Cul1/Rbx1 and titrated concentrations of CSN . The mixtures were allowed to reach equilibrium by incubation at room temperature for ~10 min prior to measurements . Equilibrium binding assays using Cul1/Rbx1 variants expressed in insect cells ( Figure 2 , Figure 2—figure supplement 2A , Figure 3—figure supplement 2C , G , Figure 4B ) were read at 530 nm on a CLARIOstar plate reader ( BMG Labtech ) in 384-well plates ( Corning , low flange , black , flat bottom ) , 90 ul per well , while binding assays using bacterially expressed Cul1/Rbx1 variants were performed on a Fluorolog-3 ( Jobin Yvon ) ( all other binding data figures ) . Binding assay with Cul1d-N8/Rbx1 ( substrate ) and CSN5E76A were allowed to equilibrate for only 45 s , because although this mutant exhibited an ~300-fold decrease in activity ( data not shown ) the residual activity was high enough to cause substantial deneddylation in a 10 min incubation . It should be noted that several of the Kd values reported for CSN binding to Cul1d-N8/Rbx1 or Cul1d/Rbx1 are below the concentration of the dansylated ligand ( 30 nM ) . While this is generally not the preferred approach , we found that 30 nM was the lowest concentration that consistently yielded highly reproducible results . The estimated Kd is very sensitive to the density of data points at the inflection point of the curve , and thus these estimates can be more prone to error . Nevertheless , different investigators in Zurich and Pasadena have consistently obtained an estimate of 1 . 6–5 nM for binding of CSN5H138A to Cul1d-N8/Rbx1 and of 9–13 nM for binding to CSN5H138A to Cul1d/Rbx1 , using different protein preparations . To estimate Kd , all data points were fitted to a quadratic equation , Y = Y0+ ( Ymax-Y0 ) * ( Kd+A+X-sqrt ( ( Kd+A+X ) ^2-4A*X ) ) /2*A where A equals concentration of labeled protein , using Prism ( Graph Pad ) . On-rate and off-rate measurements were performed on a Kintek Stopped-flow SF-2004 by exciting at 340 nM and collecting emissions through a 520 +/- 20 nm filter . For off-rate measurements , the concentrations of proteins used in each reaction are provided in the legend of Figure 3—figure supplement 1F-I . Off rate data were fit to one phase decay function: Y= ( Y0 - EP ) *exp ( -koff*X ) + EP ( where EP corresponds to reaction end point value ) . Whereas Kd , on-rate , and off-rate measurements with different configurations of Cul1 or different CSN mutants are directly comparable , off-rate measurements are not directly comparable to kcat measurements and may differ from expectation by a few fold because different buffers were used , the Cul1/Rbx1 preparations were from different sources ( bacterial for kcat , baculoviral for koff ) , and the Cul1/Rbx1 preparations carried different labels ( dansylated Cul1 for koff , [32P]-Nedd8 for kcat . Cells were grown in Lonza DMEM containing 10% FBS ( Invitrogen ) . Transient transfections were done with FugeneHD per the manufacturers instructions ( Roche ) . Flag-tagged CSN5 coding sequences were cloned into a modified MSCV-IRES-GFP vector ( containing a pBabe multiple cloning site ) via BamHI and EcoRI . Lenti-CRISPR constructs were made as described ( Shalem et al . , 2014 ) using the targeting sequences 5’- CACCGCTCGGCGATGGCGGCGTCC - 3’ and 3’ - AAACGGACGCCGCCATCGCCGAGC - 5’ . Lenti- and retroviruses were produced in 293T cells and the supernatant subsequently used for transduction to establish stable cell lines . For Western Blot analysis cells were directly lysed in 2X SDS sample buffer . Lysates were sonicated for 15 s at 10% of maximum amplitude using a Branson Digital Sonifier and boiled for 10 min at 100°C . SILAC labeling was in Invitrogen DMEM containing 10% FBS and 13C615N2-lysine and 13C6-arginine from Cambridge Isotope Laboratory . For immunoprecipitations , cells were lysed in Pierce Lysis Buffer containing cOmplete Protease Inhibitor Cocktail ( Roche ) and lysates were sonicated for 10 s at 10% of maximum amplitude using a Branson Digital Sonifier . After a 5 min clearing at 18 , 000 x g at 4°C , proteins were immunoprecipitated with M2 Flag agarose beads ( Sigma ) for 30 min and prepared for mass spectrometry as described in Pierce et al . ( 2013 ) . Samples were analyzed using an EASY-nLC 1000 coupled to an Orbitrap Fusion and analyzed by MaxQuant ( v 1 . 5 . 0 . 30 ) . Digested peptides ( 250 ng ) were loaded onto a 26-cm analytical HPLC column ( 75 μm ID ) packed in-house with ReproSil-Pur C18AQ 1 . 9 μm resin ( 120 Å pore size , Dr . Maisch , Ammerbuch , Germany ) . After loading , the peptides were separated with a 120 min gradient at a flow rate of 350 nL/min at 50 °C ( column heater ) using the following gradient: 2–6% solvent B ( 7 . 5 min ) , 6–25% B ( 82 . 5 min ) , 25–40% B ( 30min ) , 40–100% B ( 1 min ) , and 100% B ( 9 min ) where solvent A was 97 . 8% H2O , 2% ACN , and 0 . 2% formic acid ) and solvent B was 19 . 8% H2O , 80% ACN , and 0 . 2% formic acid . The Orbitrap Fusion was operated in data-dependent acquisition ( DDA ) mode to automatically switch between a full scan ( m/z=350–1500 ) in the Orbitrap at 120 , 000 resolving power and a tandem mass spectrometry scan of Higher energy Collisional Dissociation ( HCD ) fragmentation detected in ion trap ( using TopSpeed ) . AGC target of the Orbitrap and ion trap was 400 , 000 and 10 , 000 respectively . Thermo RAW files were searched with MaxQuant ( v 1 . 5 . 3 . 8 ) ( Cox and Mann , 2008; Cox et al . , 2011 ) . Spectra were searched against human UniProt entries ( 91 , 647 sequences ) and a contaminant database ( 245 sequences ) . In addition , spectra were searched against a decoy database ( generated by reversing the target sequences ) to estimate false discovery rates . Trypsin was specified as the digestion enzyme with up to two missed cleavages allowed . Variable modifications included oxidation of methionine and protein N-terminal acetylation . Carboxyamidomethylation of cysteine was specified as a fixed modification . SILAC was specified as the quantitation method with Arg6 and Lys8 specified as the heavy labeled amino acids . Precursor mass tolerance was less than 4 . 5 ppm after recalibration and fragment mass tolerance was 0 . 5 Da . False discovery rates at the peptide and protein levels were less than 1% as estimated by the decoy database search . Ratios were calculated for proteins quantified in at least two of the four biological replicates . 95% confidence intervals and adjusted p-values were calculated using the R package limma ( Ritchie et al . , 2015 ) Chemical cross-linking of purified complexes was performed using DSS H12/D12 ( Creative Molecules ) as cross-linking agent and as previously described ( Birol et al . , 2014 ) . Subsequent MS analysis and cross-link assignment and detection were carried out essentially as described ( Leitner et al . , 2014 ) on an Orbitrap Elite ( Thermo Scientific ) using the xQuest/xProphet software pipeline . Proteins were separated by SDS-PAGE gel electrophoresis and transferred to a nitrocellulose membrane by wet blot . Primary antibodies used for detection were: anti-CSN5 mouse monoclonal Santa Cruz Biotechnology sc-393725 , anti-Cul1 mouse monoclonal Santa Cruz Biotechnology sc-17775 , anti-Cul2 rabbit polyclonal Thermo Scientific #51–1800 , anti-Cul3 rabbit polyclonal Cell Signaling #2769 , anti-Cul4A rabbit polyclonal Cell Signaling #2699 , anti-Cul5 rabbit polyclonal Bethyl Laboratories A302-173A , anti-β-actin mouse monoclonal Sigma A5316 , anti-GFP mouse monoclonal Clontech #632381 . CSN5H138A-SCF-Nedd8Skp2/Cks1 samples for cryo-electron microscopy were generated by pre-incubating the purified components as described in Enchev et al . ( 2012 ) and ran over a Superose 6 increase 3 . 2/300 column ( GE Healthcare ) at 4 °C , eluting 50 µl fractions in 15 mM HEPES , pH 7 . 6 , 100 mM NaCl , 0 . 5 mM DTT . The sample was kept on ice and its homogeneity and mono-dispersity from the peak elution was immediately confirmed by visualization in negative stain . For cryo EM preparation , the sample was diluted to 0 . 1 mg/ml and 2 µl were applied to Quantifoil grids ( R1 . 2/1 . 3 Cu 400 mesh ) , freshly coated with an extra layer of thin carbon and glow-discharged for 2 min at 50 mA and 0 . 2 mbar vacuum . The grids were manually blotted to produce a thin sample film and plunge-frozen into liquid ethane . Data were collected automatically using EPU software in low dose mode on a Titan Krios transmission electron microscope , equipped with a Falcon II direct electron detector ( FEI ) , and operated at 300 kV , an applied nominal defocus from -2 . 5 to - 5 . 0 µm in steps of 0 . 25 µm , and 80 , 460-fold magnification , resulting in a pixel size of 1 . 74 Å on the sample scale . Images were collected as seven separate frames with a total dose of 25 e-/Å2 . CTF-estimation and subsequent correction were performed using RELION ( Scheres , 2012 ) and CTFFIND3 ( Mindell and Grigorieff , 2003 ) . All micrographs were initially visually inspected and only those with appropriate ice thickness as well as Thon rings in their power spectra showing regularity and extending to 6 Å or beyond were used for subsequent analysis . In order to generate 2D references for automated particle selection , ~4 , 000 single particles were manually picked and subjected to 2D classification in RELION . Six well-defined 2D class averages were selected , low-pass filtered to 35 Å to prevent reference bias , and used as references . Approximately 150 , 000 single particles were automatically selected and subjected to reference-free 2D and 3D classification , in order to de-select the particles , which resulted in poorly defined or noisy averages . Approximately half of these single particles resulted in a well-defined 3D class average , which resembled the previously published negative stain EM map of the same complex ( Enchev et al . , 2012 ) . This dataset was subject to 3D auto-refinement in RELION , using a version low-pass filtered to 50 Åas an initial reference . The converged map was further post-processed in RELION , using MTF-correction , FSC-weighting and a soft spherical mask with a 5-pixel fall-off . Csn7b was modeled using Csn7a as a template on the Phyre2 server ( Kelley et al . , 2015 ) and the modeled coordinates were aligned to Csn7a in PDB ID 4D10 ( Lingaraju et al . , 2014 ) , effectively generating a CSN atomic model for the Csn7b-containing complex . Model visualization , molecular docking , distance measurements and morph movie generation were performed with UCSF Chimera ( Pettersen et al . , 2004 ) . The cryo electron microscopy density map of CSNCsn5H138A-SCF-Nedd8Skp2/Cks1 is deposited in the Electron Microscopy Data Bank under accession code EMD-3401 . | Just like you might clear out the old food in your refrigerator to make room for new groceries , cells constantly break down existing proteins to provide space for new ones . The enzymes that generally carry out the first step of this breakdown process are called ubiquitin ligases and human cells make hundreds of different ones . These ubiquitin ligases are not always active and a large group of them can be switched off by a group of proteins known as the COP9-Signalosome ( or CSN for short ) . To achieve this , CSN recognizes and cuts off a structure called Nedd8 from these ubiquitin ligases . However , CSN itself remains inactive until it finds and binds to ubiquitin ligases that have Nedd8 attached . Mosadeghi et al . have now used biophysical techniques to study how purified CSN binds to ubiquitin ligases , removes Nedd8 and then releases the inactivated enzymes . The experiments provided a clearer picture of what the CSN looks like when it binds its targets and revealed which parts of the proteins are involved in the interaction . Furthermore , the data showed that , immediately after Nedd8 is removed from the ubiquitin ligase , CSN quickly switches back into an “off” position that allows it to release the now inactive ubiquitin ligase . This helps to explain how CSN can remove Nedd8 from many ubiquitin ligase molecules in a short period of time . Mosadeghi et al . also confirmed these findings in human cells with various versions of CSN that have different levels of activity . A future challenge is to understand exactly how the newly revealed mechanisms actually play out in cells . Also , some components of CSN are present in abnormally large amounts in cancer cells and therefore this knowledge may eventually lead to new ideas about how to treat cancer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2016 | Structural and kinetic analysis of the COP9-Signalosome activation and the cullin-RING ubiquitin ligase deneddylation cycle |
Some remarkable animal species require an opposite-sex partner for their sexual development but discard the partner’s genome before gamete formation , generating hemi-clonal progeny in a process called hybridogenesis . Here , we discovered a similar phenomenon , termed pseudosexual reproduction , in a basidiomycete human fungal pathogen , Cryptococcus neoformans , where exclusive uniparental inheritance of nuclear genetic material was observed during bisexual reproduction . Analysis of strains expressing fluorescent reporter proteins revealed instances where only one of the parental nuclei was present in the terminal sporulating basidium . Whole-genome sequencing revealed that the nuclear genome of the progeny was identical with one or the other parental genome . Pseudosexual reproduction was also detected in natural isolate crosses where it resulted in mainly MATα progeny , a bias observed in Cryptococcus ecological distribution as well . The mitochondria in these progeny were inherited from the MATa parent , resulting in nuclear-mitochondrial genome exchange . The meiotic recombinase Dmc1 was found to be critical for pseudosexual reproduction . These findings reveal a novel , and potentially ecologically significant , mode of eukaryotic microbial reproduction that shares features with hybridogenesis in animals .
Most multicellular organisms in nature undergo ( bi ) sexual reproduction involving two partners of the opposite sex to produce progeny . In most cases , following the fusion of the two haploid gametes , the diploid zygote receives one copy of the genetic material from each parent . To produce these haploid gametes , a diploid germ cell of the organism undergoes meiosis , which involves recombination between the two parental genomes , generating recombinant product . Recombination confers benefits by bringing together beneficial mutations and segregating away deleterious ones ( Dimijian , 2005; Meirmans , 2009 ) . In contrast , some organisms undergo variant forms of sexual reproduction , including parthenogenesis , gynogenesis , androgenesis , and hybridogenesis , and in doing so , produce clonal or hemi-clonal progeny ( Avise , 2015; Neaves and Baumann , 2011 ) . In parthenogenesis , a female produces clonal progeny from its eggs without any contribution from a male partner ( Avise , 2015; Horandl , 2009 ) . Gynogenesis and androgenesis occur when the fusion of an egg with a sperm induces cell division to produce clonal female or male zygotes , respectively ( Lehtonen et al . , 2013 ) . During hybridogenesis , an egg from one species fuses with the sperm from another species to generate a hybrid diploid zygote ( Lavanchy and Schwander , 2019 ) . However , one of the parental genomes is excluded during development , in a process termed genome exclusion that occurs before gametogenesis . The remaining parental genome undergoes replication followed by meiosis to produce an egg or a sperm . The sperm or egg then fuses with an opposite-sex gamete to generate a hemiclonal progeny . Because only one parent contributes genetic material to the progeny , but both parents are physically required , this phenomenon has been termed sexual parasitism ( Lehtonen et al . , 2013; Umphrey , 2006 ) . While most of the reported cases of hybridogenesis are from female populations , recent reports suggest that it may also occur in male populations of some species ( Doležálková et al . , 2016; Schwander and Oldroyd , 2016 ) . Currently , hybridogenesis has only been observed in the animal kingdom in some species of frogs , fishes , and snakes . Plants also exhibit parthenogenesis ( aka apomixis ) , along with gynogenesis and androgenesis ( Lehtonen et al . , 2013; Mirzaghaderi and Hörandl , 2016 ) . Unlike animals , most fungi do not have sex chromosomes; instead , cell-type identity is defined by the mating-type ( MAT ) locus ( Heitman , 2015; Heitman et al . , 2013 ) . While many fungi are heterothallic , with opposite mating types in different individuals , and undergo sexual reproduction involving two partners of compatible mating types , other fungi are homothallic , with opposite mating types residing within the same organism , and can undergo sexual production during solo culture in the absence of a mating partner . One class of homothallic fungi undergoes unisexual reproduction , during which cells of a single mating type undergo sexual reproduction to produce clonal progeny , similar to parthenogenesis ( Heitman , 2015; Lee et al . , 2010 ) . Gynogenesis and hybridogenesis have not been identified in the fungal kingdom thus far . Cryptococcus neoformans is a basidiomycete human fungal pathogen that exists as either one of two mating types , MATa or MATα ( Sun et al . , 2019a ) . During sexual reproduction , two haploid yeast cells of opposite mating types interact and undergo cell-cell fusion ( Kwon-Chung , 1975; Kwon-Chung , 1976; Sun et al . , 2019b ) . The resulting dikaryotic zygote then undergoes a morphological transition and develops into hyphae whose termini mature to form basidia . In the basidium , the two parental nuclei fuse ( karyogamy ) , and the resulting diploid nucleus undergoes meiosis to produce four daughter nuclei ( Idnurm , 2010; Kwon-Chung , 1976; Sun et al . , 2019b; Zhao et al . , 2019 ) . These four haploid nuclei repeatedly divide via mitosis and bud from the surface of the basidium to produce four long spore chains . Interestingly , in addition to this canonical heterothallic sexual reproduction , a closely related species , C . deneoformans can undergo unisexual reproduction ( Lin et al . , 2005; Roth et al . , 2018; Sun et al . , 2014 ) . In a previous study , we generated a genome-shuffled strain of C . neoformans , VYD135α , by using the CRISPR-Cas9 system targeting centromeric transposons in the lab strain H99α . This led to multiple centromere-mediated chromosome arm exchanges in strain VYD135α when compared to the parental strain H99α , without any detectable changes in gene content between the two genomes ( Yadav et al . , 2020 ) . In addition , strain VYD135α exhibits severe sporulation defects when mated with strain KN99a ( which is congenic with strain H99α but has the opposite mating type ) , likely due to the extensive chromosomal rearrangements introduced into the VYD135α strain . In this study , we show that the genome-shuffled strain VYD135α can in fact produce spores in crosses with MATa C . neoformans strains after prolonged incubation . Analysis of these spores reveals that the products from each individual basidium contain genetic material derived from only one of the two parents . Whole-genome sequencing of the progeny revealed an absence of recombination between the two parental genomes . The mitochondria in these progeny were found to always be inherited from the MATa parent , consistent with known mitochondrial uniparental inheritance ( UPI ) patterns in C . neoformans ( Sun et al . , 2020a ) . Using strains with differentially fluorescently labeled nuclei , we discovered that in a few hyphal branches as well as in basidia , only one of the two parental nuclei was present and produced spores , leading to uniparental nuclear inheritance . We also observed the occurrence of such uniparental nuclear inheritance in wild-type and natural isolate crosses . Furthermore , we found that the meiotic recombinase Dmc1 plays a central role during this unusual mode of reproduction of C . neoformans . Overall , this mode of sexual reproduction of C . neoformans exhibits striking parallels with hybridogenesis in animals .
Previously , we generated a strain ( VYD135α ) with eight centromere-mediated chromosome translocations compared to the wild-type parental isolate H99α ( Yadav et al . , 2020 ) . Co-incubation of the wild-type strain KN99a with the genome-shuffled strain VYD135α resulted in hyphal development and basidia production , but no spores were observed during a standard 2-week incubation . However , when sporulation was assessed at later time points in the VYD135α×KN99a cross , we observed a limited number of sporulating basidia ( 16/1201=1 . 3% ) after 5 weeks compared to a much greater level of sporulation in the wild-type H99α×KN99a cross ( 524/599=88% ) ( Figure 1A–D ) . None of these strains exhibited any filamentation on their own even after 5 weeks of incubation , indicating that the sporulation events were not a result of unisexual reproduction ( Figure 1A–B ) . To analyze this delayed sporulation process in detail , spores from individual basidia were dissected and germinated to yield viable F1 progeny . As expected , genotyping of the mating-type locus in the H99α×KN99a progeny revealed that both MATa and MATα progeny were produced from each basidium ( Figure 1E and G , Table 1 ) . In contrast , the same analysis for VYD135α×KN99a revealed that all germinating progeny from each individual basidium possessed either only the MATα or the MATa allele ( Figure 1E and G , Table 1 ) . Polymerase chain reaction ( PCR ) assays also revealed that the mitochondria in all of these progeny were inherited from the MATa parent , in accord with known UPI ( Figure 1F–G ) . These results suggest the inheritance of only one of the parental nuclei in the VYD135α×KN99a F1 progeny . The presence of mitochondria from only the MATa parent in MATα progeny further confirmed that these progeny were the products of fusion between the parent strains and were not the products of unisexual reproduction . Next , we tested whether the uniparental inheritance detected at the MAT locus also applied to the entire nuclear genome . To address this , we established a fluorescence-based assay in which the nuclei of strains H99α and VYD135α were labeled with GFP-H4 , whereas the KN99a nucleus was marked with mCherry-H4 . In a wild-type cross ( H99α×KN99a ) , the nuclei in the hyphae as well as in the spores were yellow to orange because both nuclei were in a common cytoplasm and thus incorporated both the GFP-tagged and the mCherry-tagged histone H4 proteins ( Figure 2—figure supplement 1A and B ) . We hypothesized that in the cases of uniparental nuclear inheritance , only one of the nuclei would reach the terminal basidium and would thus harbor only one fluorescent nuclear color signal ( Figure 2—figure supplement 1A ) . After establishing this fluorescent tagging system using the wild-type strains H99α×KN99a , shuffled-strain VYD135α×KN99a crosses with fluorescently labeled strains were examined . In the wild-type cross , most of the basidia formed robust spore chains with both fluorescent colors observed in them , while a small population ( ~1% ) of basidia exhibited spore chains with only one color , representing uniparental nuclear inheritance ( Figure 2A and Figure 2—figure supplement 2A ) . In contrast , the majority of the basidium population in the shuffled-strain VYD135α×KN99a cross did not exhibit sporulation , and the two parental nuclei appeared fused but undivided ( Figure 2B and Figure 2—figure supplement 2B ) . A few basidia ( ~1% ) bore spore chains with only one fluorescent color , marking uniparental nuclear inheritance events . While the basidia with uniparental nuclear inheritance in the H99α×KN99a cross were a small fraction ( ~1% ) of sporulating basidia , the uniparental basidia accounted for all of the sporulating basidia in the VYD135α×KN99a cross . Taken together , these results show that the uniparental nuclear inheritance leads to the generation of clonal progeny but requires mating , the cell-cell fusion between parents of two opposite mating types . Thus , this process defies the main purpose of sexual reproduction , which is to produce recombinant progeny from two parents . Based on these observations , we define the process of uniparental nuclear inheritance during sporulation in C . neoformans as pseudosexual reproduction ( and it is referred to as such hereafter ) . The progeny obtained via this process will be referred to as the uniparental progeny because they inherit a nuclear genome derived from only one of the two parents . After establishing the pseudosexual reproduction of lab strains , we sought to determine whether such events also occur with natural isolates . For this purpose , we selected two wild-type natural isolates , Bt63a and IUM96-2828a ( referred to as IUM96a hereafter ) ( Desjardins et al . , 2017; Keller et al . , 2003; Litvintseva et al . , 2003 ) . IUM96a belongs to the same lineage as H99α/KN99a ( VNI ) and exhibits approximately 0 . 1% genome divergence from the H99α reference genome . Bt63a belongs to a different lineage of the C . neoformans species ( VNBI ) and exhibits ~0 . 5% genetic divergence from the H99α/KN99a genome . Both the Bt63a and the IUM96a genomes exhibit one reciprocal chromosome translocation with H99α , and as a result , share a total of 10 chromosome-level changes with the genome-shuffled strain VYD135α ( Figure 3A ) . None of these strains are self-filamentous even after prolonged incubation on mating media but both cross efficiently with H99α and VYD135α ( Figure 3—figure supplement 1A ) . The H99α×Bt63a strains crossed rapidly ( within a week ) producing robust sporulation from most of the basidia observed . The VYD135α×Bt63a cross underwent a low frequency of sporulation ( 12 spore-producing basidia/840 basidia=1 . 4% ) in 2–3 weeks ( Figure 3—figure supplement 1B ) . Dissection of spores from the H99α×Bt63a cross revealed a low germination frequency ( average of 25% ) with two of the basidia showing no spore germination at all ( Supplementary file 1a ) . This result is consistent with previous results , and the low germination frequency could be explained by the genetic divergence between the two strains ( Morrow et al . , 2012 ) . Genotyping of germinated spores from the H99α×Bt63a cross revealed both MATa and MATα progeny from individual basidia , with almost 75% of the meiotic events generating progeny that were heterozygous for the MAT locus ( Figure 3—figure supplement 1C and Supplementary file 1a ) . For the VYD135α×Bt63a cross , spores from 15/20 basidia germinated and displayed a higher germination frequency than the H99α×Bt63a cross ( Supplementary file 1a ) . Interestingly , all germinated progeny harbored only the MATα mating type , whereas the mitochondria were in all cases inherited from the MATa parent ( Figure 3—figure supplement 1C ) . These results suggest that pseudosexual reproduction also occurs with Bt63a and accounts for the high germination frequency of progeny from the VYD135α×Bt63a cross . The occurrence of pseudosexual reproduction was also identified using the fluorescence-based assay with crosses between the GFP-H4 tagged VDY135α and mCherry-H4 tagged Bt63a strains ( Figure 3—figure supplement 2 ) . Crosses with strain IUM96a also revealed a low level of sporulation ( 19/842=2 . 3% ) with VYD135α but a high sporulation frequency with H99α ( 91% ) ( Figure 3—figure supplement 1D ) . Analysis of progeny from crosses involving IUM96a revealed a similar pattern to what was observed with crosses involving KN99a . The progeny from H99α×IUM96a exhibited variable basidium-specific germination frequencies and inherited both MATa and MATα in each basidium , whereas VYD135α×IUM96a progeny from each basidium inherited exclusively either MATa or MATα ( Figure 3—figure supplement 1E , and Supplementary file 1b ) . Interestingly , we observed co-incident uniparental MAT inheritance and a high germination frequency in progeny of basidia 7 , 8 , and 9 from the H99α×IUM96a cross as well ( Figure 3—figure supplement 1E , and Supplementary file 1b ) . Taken together , these results suggest that this unusual mode of sexual reproduction occurs with multiple natural isolates . We further propose that pseudosexual reproduction occurs in nature in parallel with canonical sexual reproduction . As mentioned previously , H99α ( as well as the H99α-derived strain VYD135α ) and Bt63a have approximately 0 . 5% genetic divergence . The occurrence of pseudosexual reproduction in the VYD135α×Bt63a cross allowed us to test if the two parental genomes recombine with each other during development . We subjected progeny from crosses VYD135α×Bt63a and H99α×Bt63a to whole-genome sequencing . As expected , for the H99α×Bt63a cross , both parents contributed to the nuclear composition of their progeny , and there was clear evidence of meiotic recombination as determined by variant analysis ( Figure 3B ) . However , when the VYD135α×Bt63a progeny were similarly analyzed , the nuclear genome in each of the progeny was found to be inherited exclusively from only the VYD135α parent ( Figure 3C and Figure 3—figure supplement 3 ) , and the progeny exhibited sequence differences across the entire Bt63a genome . In contrast , the mitochondrial genome was inherited exclusively from the Bt63a parent ( Figure 3D and Figure 3—figure supplement 4 ) , in accord with the PCR assay results discussed above . In addition , the whole-genome sequencing data also revealed that while most of the H99α×Bt63a progeny exhibited aneuploidy , the genome-shuffled strain VYD135α×Bt63a progeny were euploid ( Figure 3—figure supplement 5A and B ) , and based on flow cytometry analysis , these uniparental progeny were haploid ( Figure 3—figure supplement 5C ) . The progeny from crosses involving IUM96a as the MATa partner were also sequenced . Similar to the Bt63a analysis , the H99α×IUM96a progeny exhibited signs of meiotic recombination , whereas the VYD135α ×IUM96a progeny did not ( Figure 3—figure supplement 6 ) . Congruent with the mating-type analysis , the progeny in each of the basidia exclusively inherited nuclear genetic material from only one of the two parents . Furthermore , the H99α×IUM96a progeny were found to be aneuploid for some chromosomes , while the VYD135α×IUM96a progeny were completely euploid ( Figure 3—figure supplement 7 ) . We also sequenced four progeny from basidium 7 from the H99α×IUM96a cross , which were suspected to be uniparental progeny based on mating-type PCRs . This analysis showed that all four progeny harbored only H99α nuclear DNA and had no contribution from the IUM96a nuclear genome , further supporting the conclusion that pseudosexual reproduction occurs in wild-type crosses ( Figure 3—figure supplement 6A ) . Similar to other progeny , the mitochondria in these progeny were inherited from the MATa parent ( Figure 3—figure supplement 1E , and Supplementary file 1b ) . Combined , these results affirm the occurrence of a novel mode of sexual reproduction in C . neoformans , which is initiated by the fusion of two strains of opposite mating types , but whose progeny inherit DNA exclusively from one parent . Fluorescence microscopy revealed that only one of the two parental nuclei undergoes meiosis and produces spores in approximately 1% of the total basidia population . Based on this finding , we hypothesized that the basidia with only one parental nucleus might arise due to nuclear segregation events during hyphal branching . To gain further insight into this process , the nuclear distribution pattern along the sporulating hyphae was studied . As expected , imaging of long hyphae in the wild-type cross revealed the presence of pairs of nuclei with both fluorescent markers along the length of the majority of hyphae ( Figure 4A ) . In contrast , tracking of hyphae from basidia with spore chains in the genome-shuffled strain VYD135α×KN99a cross revealed hyphal branches with only one parental nucleus , which were preceded by a hyphum with both parental nuclei ( Figure 4B , Figure 4—figure supplement 1A and B ) . Unfortunately , a majority of the hyphae ( >30 independent hyphae ) we tracked were embedded into the agar , and most of these could not be tracked to the point of branching . For some others , we were able to image the hyphal branching point where two nuclei separate from each other but were then either broken or did not have mature basidia on them ( Figure 4—figure supplement 1B ) . In total , we observed seven events of nuclear loss at hyphal branching in independent experiments and were able to track two of them to observe sporulation or basidia formation at the tip . We also observed long hyphae with only one parental nucleus in the VYD135α×Bt63a cross as well , suggesting the mechanism might be similar between strains . These results suggest that hyphal branching may facilitate the separation of one parental nucleus from the main hyphae harboring both parental nuclei . While this is the most plausible explanation based on our results , we cannot rule out other possible mechanisms , such as a role for clamp cells , leading to nuclear separation during hyphal growth . As a result , one of the parental genomes is excluded at a step before diploidization and meiosis , similar to the process of genome exclusion observed in hybridogenesis . We hypothesize that nuclear segregation can be followed by endoreplication occurring in these hyphal branches or in the basidium to produce a diploid nucleus that then ultimately undergoes meiosis and produces uniparental progeny , which will be explored in future studies . Because the genomes of the uniparental progeny did not show evidence of meiotic recombination between the two parents , we tested whether pseudosexual reproduction involves meiosis . In addition , we sought to test our hypothesis that pseudosexual reproduction involves endoreplication that is followed by meiosis . We therefore tested whether Dmc1 , a key component of the meiotic machinery , is required for pseudosexual reproduction . The meiotic recombinase gene DMC1 was deleted in congenic strains H99α , VYD135α , and KN99a , and the resulting mutants were subjected to crossing . A previous report documented that dmc1Δ bilateral crosses ( both the parents are mutant for DMC1 ) display significantly reduced , but not completely abolished , sporulation in Cryptococcus ( Lin et al . , 2005 ) . We observed a similar phenotype with the H99α dmc1Δ×KN99a dmc1Δ cross . While most of the basidia were devoid of spore chains , a small percentage ( 21/760=2 . 7% ) of the population bypassed the requirement for Dmc1 and produced spores ( Figure 5A and Figure 5—figure supplement 1A ) . When dissected , the germination frequency for these spores was found to be very low ( ~22% on average ) with spores from many basidia not germinating at all ( Supplementary file 1c ) . Furthermore , MAT-specific PCRs revealed that some of the progeny were aneuploid or diploid . For VYD135α dmc1Δ×KN99a dmc1Δ , many fewer basidia ( ~0 . 1% ) produced spore chains as compared to ~1% sporulation in VYD135α×KN99a ( Figure 5A , B and Figure 5—figure supplement 1B ) . dmc1 mutant unilateral crosses ( one of the two parents is mutant and the other one is wild-type ) sporulated at a frequency of 0 . 4% suggesting that only one of the parental strains was producing spores ( Figure 5B ) . When a few sporulating basidia from the VYD135α dmc1Δ×KN99a dmc1Δ bilateral cross were dissected , two different populations of basidia emerged , one with no spore germination , and the other with a high spore germination frequency and uniparental MAT inheritance ( Supplementary file 1c ) . We hypothesized that the basidia with a high spore germination frequency represent those that have escaped the normal requirement for Dmc1 . Overall , the DMC1 deletion led to a 20-fold reduction in viable sporulation in the VYD135α×KN99a cross , observed as a ten fold decrease from the number of sporulation events in the bilateral cross and a further two fold reduction in the number of basidia producing viable spores . To further support these findings , DMC1 was deleted in mCherry-H4 tagged KN99a and crossed with GFP-H4 tagged VYD135α . We hypothesized that GFP-H4 tagged VYD135α would produce spore chains in this cross because it harbors DMC1 , whereas mCherry-H4 tagged KN99a dmc1Δ would fail to do so . Indeed , all 11 observed basidia with only the GFP-H4 fluorescence signal were found to produce spores , but only 2 out of 19 mCherry-H4 containing basidia exhibited sporulation ( Figure 5—figure supplement 2 ) . These results combined with the spore dissection findings show that Dmc1 is critical for pseudosexual reproduction . While these results provide concrete evidence for meiosis as a part of pseudosexual reproduction , they also suggest the occurrence of a preceding endoreplication event . However , further studies will need to be conducted to validate and confirm endoreplication or alternate mechanisms to achieve the ploidy necessary for a classical meiosis event .
Hybridogenesis and parthenogenesis are mechanisms that allow some organisms to overcome some hurdles of sexual reproduction and produce hemiclonal or clonal progeny ( Avise , 2015; Horandl , 2009; Lavanchy and Schwander , 2019 ) . However , harmful mutations are not filtered in these processes , making them disadvantageous during evolution and thus restricting the occurrence of these processes to a limited number of animal species ( Lavanchy and Schwander , 2019 ) . In this study , we discovered and characterized the occurrence of a phenomenon in fungi that resembles hybridogenesis and termed it pseudosexual reproduction ( Figure 6—figure supplement 1 ) . Fungi are known to exhibit asexual , ( bi ) sexual , unisexual , and parasexual reproduction , and can switch between these reproductive modes depending on environmental conditions ( Heitman , 2015; Heitman et al . , 2013 ) . The discovery of pseudosexual reproduction further diversifies known reproductive modes in fungi , suggesting the presence of sexual parasitism in this kingdom . Hybridogenesis in animals occurs between two different species . The result of hybridogenesis is the production of gametes that are clones of one of the parents , which then fuse with an opposite-sex gamete of the second species , generating hemiclonal offspring . In our study , we observed a similar phenomenon where only one parent contributes to spores , the counterpart of mammalian gametes . However , we observed this phenomenon occurring between different strains of the same species , C . neoformans . It is important to note that these strains vary significantly from each other in terms of genetic divergence and in one case by chromosome rearrangements to the extent that they could be considered different species . This suggests that hybridogenesis in animals and pseudosexual reproduction in fungi are similar to each other . Hybridogenesis requires the exclusion of one of the parents , which is followed by endoreplication of the other parent’s genome and meiosis . The whole-genome sequence of the progeny in our study revealed the complete absence of one parent’s genome , suggesting manifestations of genome exclusion during hyphal growth . The mechanism by which the retained parental genome increases its ploidy before meiosis remains to be further investigated in C . neoformans . Endoreplication is known to occur in the sister species C . deneoformans during unisexual reproduction , and we think that this is the most likely route via which ploidy is increased during pseudosexual reproduction . The mechanism and time of genome exclusion during hybridogenesis in animals are not entirely understood , except for a few insights from diploid fishes of the genus Poeciliopsis and water frogs , Pelophylax esculentus . Studies using Poeciliopsis fishes showed that haploid paternal genome exclusion takes place during the onset of meiosis via the formation of a unipolar spindle , and thus , only the haploid set of maternal chromosomes is retained ( Cimino , 1972a; Cimino , 1972b ) . On the other hand , studies involving P . esculentus revealed that genome exclusion occurs during mitotic division , before meiosis , which is followed by endoreplication of the other parental genome ( Heppich et al . , 1982; Tunner and Heppich-Tunner , 1991; Tunner and Heppich , 1981 ) . A recent study , however , proposed that genome exclusion in P . esculentus could also take place during early meiotic phases ( Doležálková et al . , 2016 ) . Using fluorescence microscopy , we examined the steps of nuclear exclusion in C . neoformans and found that it occurs during mitotic hyphal growth and not during meiosis . We also observed that genome exclusion could happen with either of the two parents in C . neoformans , similar to what has also been reported for water frogs . However , for most other species , genome exclusion was found to occur with the male genome only , leaving behind the female genome for meiosis ( Cimino , 1972a; Holsbeek and Jooris , 2010; Lavanchy and Schwander , 2019; Umphrey , 2006; Uzzell et al . , 1976; Vinogradov et al . , 1991 ) . Multiple studies have shown the formation of meiotic synaptonemal complexes during hybridogenesis , clearly establishing the presence of meiosis during this process ( Dedukh et al . , 2019; Dedukh et al . , 2020; Nabais et al . , 2012 ) . Our results showed that the meiotic recombinase Dmc1 is required for pseudosexual reproduction , suggesting the presence of meiosis , whereas there is no direct evidence for the role of a meiotic recombinase in hybridogenetic animals . Taken together , these results indicate that the mechanism might be at least partially conserved across distantly related species . Future studies will shed more light on this , and if established , the amenability of C . neoformans to genetic manipulation will aid in deciphering some of the unanswered questions related to hybridogenesis in animals . The occurrence of pseudosexual reproduction might also have significant implications for C . neoformans biology . Most ( >95% ) of Cryptococcus natural isolates belong to only one mating type , α ( Zhao et al . , 2019 ) . While the reason behind this distribution is unknown , one explanation could be the presence of unisexual reproduction in the sister species C . deneoformans and C . gattii ( Fraser et al . , 2005; Lin et al . , 2005; Phadke et al . , 2014 ) . The presence of pseudosexual reproduction in C . neoformans might help explain the mating-type distribution pattern for this species . In this report , one of the MATa natural isolates , Bt63a , did not contribute to pseudosexual reproduction and the other isolate , IUM96a , produced uniparental progeny in only one basidium , while the rest of the basidia produced MATα progeny . We hypothesized that MATa isolates may be defective in this process due to either a variation in their genomes or some other as yet undefined sporulation factor . As a result , pseudosexual reproduction could lead to the generation of predominantly α progeny in nature , reducing the MATa population and thus favoring the expansion of the α mating-type population . However , it is still possible that the preferential inheritance of the nuclear genome from one of the two parents is decided by genetic elements located in regions other than MAT , and whether the uniparental nuclear inheritance is mating-type specific remains to be elucidated . Furthermore , the occurrence of pseudosexual reproduction in other pathogenic species such as C . deneoformans and non-pathogenic species such as C . amylolentus will be investigated in future studies . Attempts to identify the occurrence of pseudosexual reproduction between species where hybrids are known to occur , C . neoformans and C . deneoformans hybrids , will also be made . These studies will help establish the scope of pseudosexual reproduction in Cryptococcus species and could be extended to other basidiomycetes . We propose that pseudosexual reproduction can occur between any two opposite mating-type strains as long as each of them is capable of undergoing cell-cell fusion and at least one of them can sporulate . We speculate that pseudosexual reproduction might play a key role in C . neoformans survival during unfavorable conditions . In conditions where two mating partners are fully compatible , pseudosexual reproduction will be mostly hidden and might not be important ( Figure 6 , top panel ) . However , when the two mating partners are partially incompatible or completely incompatible due to high genetic divergence or karyotypic variation , pseudosexual reproduction will be important ( Figure 6 , left , right , and bottom panels ) . For example , most of the basidia in H99α and Bt63a cross largely produce aneuploid and/or inviable progeny leading to unsuccessful sexual reproduction . However , a small yet significant proportion of the basidia generate clonal progeny that are viable and fit via pseudosexual reproduction . We hypothesized that these progeny will have a better chance of survival and find a suitable mating partner in the environment whereas , the unfit recombinant progeny might fail to do so . In nature , this might allow a new genotype/karyotype to not only survive but also expand and will prove advantageous . If a new genotype/karyotype had only the option of undergoing sexual reproduction , it might not survive , restricting the evolution of a new strain . Overall , this mode of pseudosexual reproduction might act as an escape path from genomic incompatibilities between two related isolates and allow them to produce spores for dispersal . One of the key differences between pseudosexual reproduction and unisexual reproduction observed in the Cryptococcus species complex is the inheritance of mitochondrial DNA . While both unisexual and pseudosexual reproduction result in clonal progeny with respect to the nuclear genome , the mitochondria in pseudosexual reproduction are almost exclusively inherited from the MATa parent ( Figure 6—figure supplement 1 ) . This results in the exchange of mitochondrial DNA in the progeny that inherit the MATα nuclear genome , resembling the nuclear-mitochondrial exchange observed during cytoduction in Saccharomyces cerevisiae . During cytoduction , mutants defective in nuclear fusion produce haploid progeny with nuclear genome from one parent , but a mixture of both parents cytoplasm resulting in the inheritance of one parental mitochondrial genome with the other parent’s nuclear genome ( Conde and Fink , 1976; Lancashire and Mattoon , 1979; Zakharov and Yarovoy , 1977 ) . This process was used to study mitochondrial genetics with respect to the transfer of drug-resistance genes and other mitochondrial mutations . Similar to cytoduction , pseudosexual reproduction could be employed to study mitochondrial genetics , such as functional analysis of mitochondrial encoded drug resistance , and cytoplasmic inheritance of factors such as prions in C . neoformans . The fungal kingdom is one of the more diverse kingdoms with approximately 3 million species ( Sun et al . , 2020b ) . The finding of hybridogenesis-like pseudosexual reproduction hints toward unexplored biology in this kingdom that might provide crucial clues for understanding the evolution of sex . Fungi have also been the basis of studies focused on understanding the evolution of meiosis , and the presence of genome reduction , as well as the parasexual cycle in fungi , have led to the proposal that meiosis evolved from mitosis ( Hurst and Nurse , 1991; Wilkins and Holliday , 2009 ) . Pseudosexual reproduction may be a part of an evolutionary process wherein genome exclusion followed by endoreplication and meiosis was an ancestral form of reproduction that preceded the evolution of sexual reproduction . Evidence supporting such a hypothesis can be observed in organisms undergoing facultative sex or facultative parthenogenesis ( Booth et al . , 2012; Fields et al . , 2015; Hodač et al . , 2019; Hojsgaard and Horandl , 2015 ) . The presence of these organisms also suggests that a combination of both sexual and clonal modes of reproduction might prove to be evolutionarily advantageous .
C . neoformans wild-type strains H99α and KN99a served as the wild-type isogenic parental lineages for the experiments ( Nielsen et al . , 2003; Perfect et al . , 1993 ) , in addition to MATa strains Bt63a and IUM96-2828a ( Keller et al . , 2003; Litvintseva et al . , 2003 ) . Strains were grown in YPD media for all experiments at 30°C unless stated otherwise . G418 and/or NAT were added at a final concentration of 200 and 100 µg/ml , respectively , for the selection of transformants . MS media was used for all the mating assays , which were performed as described previously ( Sun et al . , 2019b ) . Basidia-specific spore dissections were performed after 2–5 weeks of mating , and the spore germination frequency was scored after 5 days of dissection . All strains and primers used in this study are listed in Supplementary file 1d and Supplementary file 1e , respectively . Mating type ( MAT ) and mitochondrial genotyping for all the progeny were conducted using PCR assays . Genomic DNA was prepared using the MasterPure Yeast DNA Purification Kit from Lucigen . To determine the MAT , the STE20 allele present within the MAT locus was detected because it differs in length between the two different mating types . Primers specific to both MATa and MATα ( JOHE50979-50982 in Supplementary file 1e ) were mixed in the same PCR mix , and the identification was made based on the length of the amplicon ( Figure 1E–G ) . For the mitochondrial genotyping , the COX1 allele present in the mitochondrial DNA was probed to distinguish between H99α/VYD135α and KN99a/IUM96a . For the differentiation between Bt63a and H99α/VYD135α , the COB1 allele was used because COX1 in Bt63a is identical to H99α/VYD135α . The difference for both COX1 and COB1 is the presence or absence of an intron and results in significantly different size products between MATα and MATa parents ( Figure 1 and Figure 3—figure supplement 1 ) . The primers used for these assays ( JOHE51004-51007 ) are mentioned in Supplementary file 1e . Genomic DNA for whole-genome sequencing was prepared using the CTAB-based lysis method , as described previously ( Yadav et al . , 2020 ) . Briefly , 50 ml of an overnight culture was pelleted , frozen at −80°C , and subjected to lyophilization . The lyophilized cell pellet was broken into a fine powder , mixed with lysis buffer , and the mix was incubated at 65°C for an hour with intermittent shaking . The mix was then cooled on ice , and the supernatant was transferred into a fresh tube , and an equal volume of chloroform ( ~15 ml ) was added and mixed . The mix was centrifuged at 3200 rpm for 10 min , and the supernatant was transferred to a fresh tube . An equal volume of isopropanol ( ~18–20 ml ) was added into the supernatant and mixed gently . This mix was incubated at −20°C for an hour and centrifuged at 3200 rpm for 10 min . The supernatant was discarded , and the DNA pellet was washed with 70% ethanol . The pellet was air-dried and dissolved in 1 ml of RNase containing 1× TE buffer and incubated at 37°C for 45 min . The DNA was again chloroform purified and precipitated using isopropanol , followed by ethanol washing , air drying , and finally dissolved in 200 µl 1× TE buffer . The DNA quality was estimated with NanoDrop , whereas DNA quantity was estimated with Qubit . Illumina sequencing of the strains was performed at the Duke sequencing facility core ( https://genome . duke . edu/ ) , using Novaseq 6000 as 150 paired-end sequencing . The Illumina reads , thus obtained , were mapped to the respective genome assembly ( H99α , VYD135α , Bt63a , or IUM96a ) using Geneious ( RRID:SCR_010519 ) default mapper to estimate ploidy . The resulting BAM file was converted to a . tdf file , which was then visualized through IGV to estimate the ploidy based on read coverage for each chromosome . For SNP calling and score for recombination in the progeny , Illumina sequencing data for each progeny was mapped to parental strain genome assemblies individually using the Geneious default mapper with three iterations . The mapped BAM files were used to perform variant calling using Geneious with 0 . 8 variant frequency parameter and at least 90× coverage for each variant . The variants thus called were exported as VCF files and imported into IGV for visualization purposes . H99α , Bt63a , IUM96a , and VYD135α Illumina reads were used as controls for SNP calling analysis . To obtain high-molecular-weight DNA for Bt63a genome PacBio and IUM96a genome Nanopore sequencing , DNA was prepared as described above . The size estimation of DNA was carried out by electrophoresis of DNA samples using PFGE . For this purpose , the PFGE was carried out at 6 V/cm at a switching frequency of 1–6 s for 16 hr at 14°C . Samples with most of the DNA ≥100 kb or larger were selected for sequencing . For PacBio sequencing , the DNA sample was submitted to the Duke sequencing facility core . Nanopore sequencing was performed in our lab using a MinION device on an R9 . 4 . 1 flow cell . After sequencing , reads were assembled to obtain a Bt63a genome assembly via Canu ( RRID:SCR_015880 ) using PacBio reads >2 kb followed by five rounds of pilon polishing ( RRID:SCR_014731 ) . For IUM96a , one round of nanopolish was also performed before pilon polishing . Once completed , the chromosomes were numbered based on their synteny with the H99α genome . For chromosomes involved in translocation ( Chr 3 and Chr 11 ) , the chromosome numbering was defined by the presence of the respective syntenic centromere from H99 . Centromere locations were mapped based on BLASTn analysis with H99α centromere flanking genes . Synteny comparisons between the genomes were performed with SyMAP v4 . 2 using default parameters ( Soderlund et al . , 2011 ) ( http://www . agcol . arizona . edu/software/symap/ ) . The comparison block maps were exported as . svg files and were then processed using Adobe Illustrator ( RRID:SCR_010279 ) and Adobe Photoshop ( RRID:SCR_014199 ) for representation purposes . The H99α genome was used as the reference for comparison purposes for plotting VYD135α , Bt63a , and IUM96a genomes . The centromere and telomere locations were manually added during the figure processing . GFP and mCherry tagging of histone H4 were performed by integrating respective constructs at the safe haven locus ( Arras et al . , 2015 ) . GFP-H4 tagging was done using the previously described construct , pVY3 ( Yadav and Sanyal , 2018 ) . For mCherry-H4 tagging , the GFP-containing fragment in pVY3 was excised using SacI and BamHI and was replaced with mCherry sequence PCR amplified from the plasmid pLKB25 ( Kozubowski and Heitman , 2010 ) . The constructs were then linearized using XmnI and transformed into desired strains using CRISPR transformation , as described previously ( Fan and Lin , 2018 ) . The transformants were screened by PCR , and correct integrants were obtained and verified using fluorescent microscopy . To observe the fluorescence signals in the hyphae and basidia , a 2- to 3-week-old mating patch was cut out of the plate and directly inverted onto a coverslip in a glass-bottom dish . The dish was then used to observe filaments under a DeltaVision microscope available at the Duke University Light Microscopy Core Facility ( https://microscopy . duke . edu/dv ) . The images were captured at 60× magnification with 2×2 bin size and z-sections of either 1 or 0 . 4 µm each . GFP and mCherry signals were captured using the GFP and mCherry filters in the Live-Cell filter set . The images were processed using Fiji-ImageJ ( https://imagej . net/Fiji ) ( RRID:SCR_002285 ) and exported as tiff files as individual maximum projected images . The final figure was then assembled using Adobe Photoshop software for quality purposes . To visualize hyphal growth and sporulation defects during mating assays , the mating plates were directly observed under a Nikon Eclipse E400 microscope . Hyphal growth and basidia images were captured using the top-mounted Nikon DXM1200F camera on the microscope . The images were processed using Fiji-ImageJ and assembled in Adobe Photoshop software . For crosses involving wild-type H99α , VYD135α , KN99a , Bt63a , and IUM96a , approximately 1000 total basidia were counted after 4 weeks of mating , and the sporulation frequency was calculated . For crosses involving VYD135 dmc1Δ strain , three mating spots were setup independently . From each mating spot periphery , six images were captured after 3–4 weeks of mating . Basidia ( both sporulating and non-sporulating ) in each of these spots were counted manually after some processing of images using ImageJ . The sporulation frequency was determined by dividing the sporulating basidia by the total number of basidia for each spot . Each mating spot was considered as an independent experiment and at least 3000 basidia were counted from each mating spot . Flow cytometry analysis was performed as described previously ( Fu and Heitman , 2017 ) . Cells were grown on YPD medium for 2 days at 30°C , harvested , and washed with 1× phosphate-buffered saline buffer followed by fixation in 70% ethanol at 4°C overnight . Next , cells were washed once with 1 ml of NS buffer ( 10 mM Tris-HCl , pH=7 . 2 , 250 mM sucrose , 1 mM EDTA , pH=8 . 0 , 1 mM MgCl2 , 0 . 1 mM CaCl2 , 0 . 1 mM ZnCl2 , 0 . 4 mM phenylmethylsulfonyl fluoride , and 7 mM β-mercaptoethanol ) , and finally resuspended in 180 μl NS buffer containing 20 μl 10 mg/ml RNase and 5 μl 0 . 5 mg/ml propidium iodide ( PI ) at 37°C for 3–4 hr . Then , 50 μl stained cells were diluted in 2 ml of 50 mM Tris-HCl , pH=8 . 0 , transferred to FACS compatible tube , and submitted for analysis at the Duke Cancer Institute Flow Cytometry Shared Resource . For each sample , 10 , 000 cells were analyzed on the FL1 channel on the Becton-Dickinson FACScan . Wild-type H99α and previously generated AI187 were used as haploid and diploid controls , respectively , in these experiments . Data analysis was performed using the FlowJo software ( RRID:SCR_008520 ) . | Sexual reproduction enables organisms to recombine their genes to generate progeny that have higher levels of evolutionary fitness . This process requires reproductive cells – like the sperm and egg – to fuse together and mix their two genomes , resulting in offspring that are genetically distinct from their parents . In a disease-causing fungus called Cryptococcus neoformans , sexual reproduction occurs when two compatible mating types ( MATa and MATα ) merge together to form long branched filaments called hyphae . Cells in the hyphae contain two nuclei – one from each parent – which fuse in specialized cells at the end of the branches called basidia . The fused nucleus is then divided into four daughter nuclei , which generate spores that can develop into new organisms . In nature , the mating types of C . neoformans exhibit a peculiar distribution where MATα represents 95% or more of the population . However , it is not clear how this fungus successfully reproduces with such an unusually skewed distribution of mating types . To investigate this further , Yadav et al . tracked the reproductive cycle of C . neoformans applying genetic techniques , fluorescence microscopy , and whole-genome sequencing . This revealed that during hyphal branching some cells lose the nucleus of one of the two mating types . As a result , the nuclei of the generated spores only contain genetic information from one parent . Yadav et al . named this process pseudosexual reproduction as it defies the central benefit of sex , which is to produce offspring with a new combination of genetic information . Further experiments showed that this unconventional mode of reproduction can be conducted by fungi isolated from both environmental samples and clinical patient samples . This suggests that pseudosexual reproduction is a widespread and conserved process that may provide significant evolutionary benefits . C . neoformans represents a flexible and adaptable model organism to explore the impact and evolutionary advantages of sex . Further studies of the unique reproductive strategies employed by this fungus may improve the understanding of similar processes in other eukaryotes , including animals and plants . This research may also have important implications for understanding and controlling the growth of other disease-causing microbes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"genetics",
"and",
"genomics"
] | 2021 | Uniparental nuclear inheritance following bisexual mating in fungi |
Neurons have complex electrophysiological properties , however , it is often difficult to determine which properties are the most relevant to neuronal function . By combining current-clamp measurements of electrophysiological properties with multi-variate analysis ( hierarchical clustering , principal component analysis ) , we were able to characterize the postnatal development of substantia nigra dopaminergic neurons' electrical phenotype in an unbiased manner , such that subtle changes in phenotype could be analyzed . We show that the intrinsic electrical phenotype of these neurons follows a non-linear trajectory reaching maturity by postnatal day 14 , with two developmental transitions occurring between postnatal days 3–5 and 9–11 . This approach also predicted which parameters play a critical role in phenotypic variation , enabling us to determine ( using pharmacology , dynamic-clamp ) that changes in the leak , sodium and calcium-activated potassium currents are central to these two developmental transitions . This analysis enables an unbiased definition of neuronal type/phenotype that is applicable to a range of research questions .
The morphology and assortment of voltage-dependent and voltage-independent conductances displayed by one particular neuronal type provide it with specific passive and active properties ( Johnston and Wu , 1995; Hille , 2001 ) . In turn , passive and active properties define the way the neuron produces or processes information , that is , its electrical phenotype . An important question we are facing as neurophysiologists is how to characterize the electrical phenotype of a neuronal type in the most unbiased manner , such that we can then study subtle variations in this phenotype under varying conditions ( development , perturbations , disease ) . In the current study , we propose an approach based on the combination of current-clamp measurements and multi-dimensional analysis of the measured electrophysiological properties , which allowed us to characterize and precisely quantify developmental changes in the electrical phenotype of dopaminergic neurons of the substantia nigra pars compacta . In vivo , mature SNc dopaminergic neurons can display regular tonic , irregular or bursting activity depending on the behavioral context ( Grace and Bunney , 1984a , 1984b; Tepper et al . , 1990; Kitai et al . , 1999; Grace et al . , 2007 ) . These variations in activity patterns are associated with a modulation of dopamine release such that , for instance , immediate early gene activation is triggered only by bursting patterns of activity ( Kitai et al . , 1999 ) . From a biophysical point of view , regular tonic activity seems to rely mainly on SNc dopaminergic neuron intrinsic conductances while both irregular and bursting patterns necessitate the release of various neurotransmitters ( glutamate , acetylcholine ) by SNc synaptic inputs ( Kitai et al . , 1999; Grace et al . , 2007 ) . In vitro however , mature SNc dopaminergic neurons mainly display a regular tonic ( also called pacemaker ) firing behavior ( Grace and Onn , 1989; Liss et al . , 2001; Seutin et al . , 2001; Puopolo et al . , 2007; Guzman et al . , 2009; Putzier et al . , 2009b; Tateno and Robinson , 2011; Amendola et al . , 2012 ) . The irregular pattern of spontaneous activity is observed in immature SNc DA neurons in vitro ( postnatal days 9–16 , P9–P16 ) and seems to be due to transient spontaneous activation of T-type calcium channels ( Seutin et al . , 2000; Cui et al . , 2004 ) and subsequent activation of calcium-activated potassium channels ( SK ) ( Seutin et al . , 1998; Wolfart and Roeper , 2002; Cui et al . , 2004 ) . Finally , the spontaneous bursting pattern of activity is usually absent from mature SNc DA neurons in vitro , but can be observed in juvenile neurons ( P15–P21 ) and depends on the activation of NMDA receptors ( Mereu et al . , 1997 ) , or can be promoted by pharmacological blockade of T-type calcium channels and SK potassium channels ( Wolfart and Roeper , 2002 ) . Although these different studies suggest that the activity pattern of SNc dopaminergic neurons changes during the first three postnatal weeks , involving modifications in both intrinsic and synaptic input properties , our knowledge of the precise timecourse of their electrophysiological development is still fragmented . In the current study , we have performed a detailed analysis of the development of the intrinsic properties of SNc dopaminergic neurons over the first four postnatal weeks ( from P2 to P29 ) . We measured 16 electrophysiological parameters , exhaustively characterizing the passive and active properties of SNc dopaminergic neurons . Applying multi-variate statistical analyses ( agglomerative hierarchical clustering and principal component analysis ) to the measured electrophysiological parameters revealed that the acquisition of mature regular pacemaking involves biphasic changes occurring mainly during the first two postnatal weeks: intrinsic electrophysiological maturity is essentially reached by the end of the second postnatal week . In addition , this type of analysis allowed us to determine the participation of specific ion currents ( leak , sodium and calcium-activated potassium currents ) to changes in the global electrical phenotype . This study provides the first comprehensive analysis of postnatal development of the intrinsic properties of SNc dopaminergic neurons and demonstrates the utility of high-dimensional electrophysiological characterization associated with multi-variate analysis methods to precisely define quantitative changes in electrical phenotype and to investigate the underlying biophysical mechanisms .
We first analyzed the spontaneous activity patterns displayed by SNc dopaminergic neurons , using two simple measures capturing the general features of activity: the averaged InterSpike Interval ( ISIavg ) , and the coefficient of variation of the ISI ( CVISI ) , which is proportional to the irregularity of firing ( Figure 2A ) . While ISIavg was found to be stable from P2 to P29 ( Figure 2B , C ) , CVISI strongly decreased over the first two postnatal weeks , reaching a stable value by P10–11 ( Figure 2B , D , Table 1 , Figure 3 ) . The decrease in CVISI was correlated with changes in firing pattern , with high CVISI values associated with a bursting pattern , intermediate values associated with irregular tonic firing and low values with regular tonic firing ( Figure 2B ) . In fact , CVISI threshold values of 20% and 80% were found to reliably separate these three types of firing patterns ( Figure 2D , E ) . Using these thresholds , the proportions of high CVISI ( bursting cells ) , medium CVISI ( irregular cells ) , and low CVISI ( regular cells ) were calculated for each developmental stage ( Figure 2E ) : all neurons were found to be bursters at P2–3 , most neurons were irregular between P5 and P11 , while pacemaker neurons became predominant after P12 and were the only type of neurons encountered after P21 ( Figure 2E ) . 10 . 7554/eLife . 04059 . 004Figure 2 . Postnatal evolution of spontaneous activity patterns in substantia nigra pars compacta dopaminergic neurons . ( A ) , typical voltage recording from a regular pacemaker dopaminergic neuron depicting the parameters extracted to characterize spontaneous activity patterns . Interspike intervals ( ISI1 , ISI2…ISIi…ISIn ) were averaged to calculate the ISIavg while the coefficient of variation of the n ISIs ( CVISI ) was computed from the standard deviation ( SD ) of the n ISIs and ISIavg . ( B ) , voltage traces showing the bursting ( high CVISI , left , red ) , irregular ( moderate CVISI , center , green ) and regular ( low CVISI , right , blue ) patterns of spontaneous activity observed during the first four postnatal weeks . ( C ) , box and whisker plot representing ISIavg vs postnatal age . ( D ) , box and whisker plot representing CVISI vs postnatal age . Two CVISI threshold values ( 20 and 80% , horizontal dotted lines ) separated three classes of activity patterns: low CV ( CVISI < 20% , black circle ) , medium CV ( 20% < CVISI < 80% , light gray triangle ) , high CV ( CVISI > 80% , dark gray square ) . The dark gray square , light gray triangle and black circle on the right indicate the symbols used to represent the three different CV classes ( high CV , medium CV and low CV , respectively ) in panel E . ( E ) , line and scatter plot representing the evolution of the percentages of high CV ( dark gray square ) , medium CV ( light gray triangle ) and low CV ( black circle ) activity patterns vs postnatal age . Scale bars: A , vertical 10 mV , horizontal 500 ms; B , vertical 20 mV , horizontal 2 s . Horizontal dotted lines in A and B indicate −60 mV . For all box and whisker plots , boxes represent the median , first and third quartile , error bars correspond to 10 and 90% , the thick line corresponds to the mean , and all outliers are represented . Colored boxes and symbols in C , D and E match the age and type of firing of the colored traces presented in B ( red for P2 , bursting; green for P6 , irregular; blue for regular , P16 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04059 . 00410 . 7554/eLife . 04059 . 005Figure 2—figure supplement 1 . Lack of effect of the recording condition on the activity pattern of substantia nigra pars compacta dopaminergic neurons . ( A ) , voltage recordings showing that regular ( left ) , irregular ( center ) and bursting ( right ) patterns of activity are observed independent of the concentration of EGTA ( top black traces , 10 mM; bottom red traces , 0 . 5 mM ) contained in the patch recording solution . ( B ) , box and whisker plot showing that the CVISI is not significantly different between the 10 mM and the 0 . 5 mM EGTA conditions ( unpaired t test , p = 0 . 201 ) . ( C ) , the amplitude of the after hyperpolarization ( AHP ) is not modified by the EGTA concentration . Left , average voltage traces showing the amplitude of the AHP following the AP in 10 mM ( black trace , n = 100 ) and in 0 . 5 mM EGTA ( red trace , n = 28 ) . Right , box and whisker plot showing that the AHP amplitude is not significantly different between the two recording conditions ( unpaired t test , p = 0 . 129 ) . ( D ) , voltage recordings obtained from three dopaminergic neurons showing that the three types of activity patterns are observed both in cell-attached ( top , green traces ) and in whole-cell configuration ( bottom , black traces ) . ( E ) , the firing rate is not significantly modified when going whole-cell . Left , box and whisker plot based on 33 P19 neurons showing that the ISIavg is not significantly modified by the whole-cell compared to the cell-attached configuration ( p = 0 . 898 ) . Right , scatter plot representing the ISIavg recorded in cell-attached configuration vs the ISIavg recorded in whole-cell configuration for the 33 cells represented in the box and whisker plot . The line represents the identity line . Scale bars: A , vertical 20 mV , horizontal 2 s; C , vertical 20 mV , horizontal 25 ms; D , vertical 20 mV , horizontal 2 s . Dotted lines in A , C and D ( lower traces ) indicate −60 mV . For all box and whisker plots , boxes represent the median , first and third quartile , and error bars correspond to the min and max of the distribution . DOI: http://dx . doi . org/10 . 7554/eLife . 04059 . 00510 . 7554/eLife . 04059 . 006Table 1 . Descriptive statistics of the 16 electrophysiological parameters measured on substantia nigra pars compacta dopaminergic neurons across postnatal developmentDOI: http://dx . doi . org/10 . 7554/eLife . 04059 . 006AgeISIavg ( s ) CVISI ( % ) τm ( ms ) Rin ( MOhm ) Cm ( pF ) Sag ( mV ) Rebound ( ms ) AP threshold ( mV ) MeanSDNP2–P31 . 060 . 849130 . 540 . 19200 . 38 . 051642420512934528 . 24 . 094651346−43 . 93 . 39P51 . 361 . 601557 . 622 . 61586 . 913 . 71164216311146521129 . 45 . 71433918812−40 . 34 . 515P60 . 780 . 561851 . 543 . 01870 . 323 . 91861217018118321830 . 27 . 21831612718−43 . 14 . 319P70 . 970 . 933638 . 428 . 03689 . 324 . 42565017426143382531 . 25 . 2252498919−39 . 84 . 035P8–P91 . 671 . 951838 . 023 . 21890 . 938 . 421576268212001412129 . 66 . 21740121215−42 . 84 . 320P10–P110 . 680 . 411221 . 18 . 61280 . 628 . 610327210102811091033 . 34 . 21033016510−41 . 74 . 211P12–P131 . 040 . 671519 . 615 . 11587 . 932 . 717369166172972261732 . 43 . 71637834116−45 . 43 . 718P140 . 640 . 481612 . 88 . 31692 . 344 . 01446019214205681435 . 62 . 9104894489−44 . 22 . 316P15–P160 . 880 . 382213 . 36 . 62277 . 828 . 51932511719245671927 . 96 . 53334020730−44 . 43 . 430P170 . 820 . 40169 . 95 . 61683 . 525 . 720323109222831602028 . 17 . 43537527026−44 . 63 . 528P180 . 630 . 30187 . 64 . 01889 . 024 . 21834011818280861829 . 54 . 22243322220−41 . 63 . 021P190 . 710 . 32249 . 66 . 12489 . 928 . 130352103302701063028 . 66 . 23845627832−44 . 24 . 133P200 . 770 . 42219 . 34 . 82188 . 541 . 82237511623231662228 . 58 . 32457249322−44 . 34 . 424P21–P230 . 720 . 32197 . 54 . 51993 . 148 . 221319115232931282127 . 56 . 62338825422−43 . 94 . 024P28–P290 . 920 . 541212 . 14 . 712148 . 342 . 6104728810316831034 . 14 . 51373637510−42 . 83 . 012AgeAP amplitude ( mV ) AP half-width ( ms ) AP rise slope ( mV/ms ) AP decay slope ( mV/ms ) AHP ( mV ) Gain start ( Hz/100pA ) Gain end ( Hz/100pA ) SFA indexP2–P359 . 59 . 692 . 380 . 62947 . 621 . 79−19 . 65 . 2914 . 14 . 9932 . 15 . 96207 . 861 . 760 . 536P547 . 17 . 4152 . 990 . 951524 . 510 . 815−16 . 94 . 61520 . 14 . 51517 . 7578 . 82 . 172 . 070 . 717P649 . 09 . 2192 . 990 . 711925 . 119 . 219−18 . 06 . 61921 . 03 . 71922 . 76 . 51011 . 73 . 7101 . 970 . 2710P746 . 37 . 8352 . 920 . 643523 . 88 . 635−17 . 64 . 13522 . 63 . 73517 . 54 . 7138 . 832 . 4132 . 080 . 7113P8–P954 . 47 . 4202 . 740 . 822032 . 814 . 820−20 . 25 . 92020 . 94 . 22018 . 14 . 31310 . 43 . 3131 . 800 . 3113P10–P1150 . 611 . 6112 . 510 . 831136 . 321 . 211−21 . 47 . 51120 . 74 . 911114 . 51011 . 55 . 3101 . 010 . 3010P12–P1362 . 16 . 3181 . 860 . 271854 . 214 . 518−29 . 74 . 21823 . 06 . 01813 . 251511 . 64 . 5151 . 290 . 8015P1461 . 07 . 1161 . 910 . 541654 . 918 . 416−30 . 98 . 51624 . 15 . 51612 . 24109 . 82 . 6101 . 300 . 5010P15–P1660 . 16 . 5301 . 630 . 493056 . 218 . 330−39 . 711 . 53026 . 14 . 0309 . 283 . 2269 . 284 . 2261 . 110 . 4426P1762 . 18 . 6281 . 610 . 372858 . 216 . 228−38 . 88 . 82828 . 15 . 0287 . 132 . 1148 . 282 . 9140 . 920 . 3114P1855 . 54 . 7211 . 450 . 212150 . 615 . 621−41 . 17 . 32128 . 43 . 2216 . 972127 . 592 . 9120 . 990 . 3512P1958 . 28 . 8331 . 550 . 383354 . 321 . 133−41 . 411 . 43327 . 44 . 6338 . 322 . 6199 . 32 . 6190 . 930 . 3119P2061 . 68 . 8241 . 470 . 312459 . 321 . 224−42 . 68 . 82428 . 05 . 0248 . 152 . 4169 . 213160 . 950 . 3316P21–P2362 . 68 . 2241 . 500 . 372461 . 521 . 024−44 . 111 . 42429 . 14 . 3246 . 792 . 12471 . 8241 . 010 . 3224P28–P2962 . 45 . 3121 . 490 . 291263 . 014 . 312−43 . 28 . 71228 . 66 . 2126 . 992 . 3106 . 821 . 4101 . 070 . 4310Abbreviations: ISIavg , averaged interspike interval; CVISI , coefficient of variation of the interspike interval; τm , membrane time constant; Rin , input resistance; Cm , membrane capacitance; Rebound , rebound delay; AP , action potential; AHP , after hyperpolarization; SFA , spike frequency adaptation . 10 . 7554/eLife . 04059 . 007Figure 3 . Statistical stacking table summarizing the statistical differences in 16 electrophysiological parameters across 15 developmental stages . Each major cell corresponding to the comparison between two developmental stages is subdivided in 16 sub-cells corresponding to the 16 electrophysiological parameters as depicted in the inset . The color of each sub-cell indicates the level of significance of the statistical comparison ( white , non-significant; light gray , p < 0 . 05; dark gray , p < 0 . 01; black , p < 0 . 001 ) . Statistical comparisons were performed for each electrophysiological parameter separately using a one-way ANOVA with post-hoc Tukey correction for multiple comparisons . Abbreviations: ISIavg , averaged interspike interval; Rin , input resistance; Cm , membrane capacitance; τm , membrane time constant; Reb , rebound delay; AHP , afterhyperpolarization; CVISI , coefficient of variation of the interspike interval; Thre , AP threshold; Ampl , AP amplitude; HW , AP half-width; Rise , AP rise slope; Decay , AP decay slope; GS , gain start; GE , gain end; SFA , spike frequency adaptation . DOI: http://dx . doi . org/10 . 7554/eLife . 04059 . 007 Several studies have suggested that bursting and irregular firing patterns are controlled in part by the calcium-activated potassium channels ( SK1-3 ) responsible for the medium afterhyperpolarization ( AHP ) following APs ( Seutin et al . , 1998; Wolfart et al . , 2001; Wolfart and Roeper , 2002; Cui et al . , 2004; Johnson and Wu , 2004; Vandecasteele et al . , 2011 ) . In SNc dopaminergic neurons , it was demonstrated that SK channels determine the regularity of spontaneous activity ( Seutin et al . , 1998; Wolfart et al . , 2001; Cui et al . , 2004; Vandecasteele et al . , 2011 ) , and that AHP developmental increase correlates with the disappearance of irregular firing ( Vandecasteele et al . , 2011 ) . Since our recordings were performed using 10 mM EGTA in the intracellular solution , we wondered whether the resulting calcium chelation could inhibit SK channel activation and thus modify spontaneous activity patterns ( favoring irregular or bursting firing patterns ) . We therefore performed two types of control experiments . In the first series of experiments , we reduced the intracellular EGTA concentration to 0 . 5 mM ( Figure 2—figure supplement 1A–C ) . In spite of the reduced calcium chelation , all firing patterns observed in 10 mM EGTA were also present in 0 . 5 mM EGTA ( Figure 2—figure supplement 1A ) . Moreover , the average CVISI values were not statistically different between both EGTA concentrations ( p = 0 . 201 , n = 80 for 10 mM , n = 34 for 0 . 5 mM , unpaired t test; Figure 2—figure supplement 1B ) when measured in neurons of the same developmental stage ( P18–P22 ) . Since the AHP can be influenced by calcium chelation and is involved in defining CVISI , we also verified that AHP amplitude was not significantly modified by changes in EGTA concentration ( Figure 2—figure supplement 1C ) : the mean AHP amplitude was not statistically different between both EGTA concentrations in P18–22 neurons ( p = 0 . 129 , n = 100 for 10 mM , n = 28 for 0 . 5 mM , unpaired t test; Figure 2—figure supplement 1C ) . As 0 . 5 mM EGTA could still be considered as a calcium chelation condition that could significantly modify firing ( Grace and Bunney , 1984b ) , we also verified that the three types of firing observed in whole-cell recordings in 0 . 5 mM EGTA or 10 mM EGTA were also present in cell-attached recordings . As demonstrated in Figure 2—figure supplement 1D , the firing patterns observed in 0 . 5 mM EGTA whole-cell recordings were found to be qualitatively similar to the firing patterns observed in cell-attached recordings in the same neurons before going whole-cell: regular , irregular , and bursting patterns were observed in cell-attached recordings . To ensure that whole-cell recorded firing patterns were also quantitatively similar to cell-attached recorded firing patterns , we compared the frequency of regular tonic firing in a subset of 33 neurons ( P16–P23 ) that were recorded both in cell-attached ( before breaking into the neuron ) and in whole-cell 10 mM EGTA ( Figure 2—figure supplement 1E ) . This analysis revealed that ISIavg was not altered by the whole-cell recording configuration ( p = 0 . 898 , n = 33 , paired t test; Figure 2—figure supplement 1E ) . Therefore , the activity pattern and activity level are not significantly altered by the recording conditions , and the developmental transition between the different types of firing patterns cannot be explained by the recording conditions . We then analyzed the passive membrane properties of SNc dopaminergic neurons , including the input resistance ( Rin ) , the membrane capacitance ( Cm ) , and the membrane time constant ( τm = Rin × Cm ) . Rin , Cm , and τm values were extracted from mono-exponential fits of the voltage responses to small hyperpolarizing current pulses applied around resting membrane potential ( Figure 4A , B ) . Except for P2–3 where Rin values were particularly high ( Figure 4C , Table 1 ) , Rin and Cm displayed almost symmetrical changes over the first four postnatal weeks: Rin decreased while Cm increased ( Figure 3 , Figure 4C , D , Table 1 ) , such that τm remained fairly constant over the same timeframe ( Figure 3 , Figure 4E , Table 1 ) . These data suggest that , after P2–3 , membrane surface area increases but that leak conductances' density stays fairly constant , such that the measured input conductance ( 1/Rin ) scales with Cm ( since Cm scales with membrane surface area ) . 10 . 7554/eLife . 04059 . 008Figure 4 . Postnatal evolution of passive properties in substantia nigra pars compacta dopaminergic neurons . ( A ) , parameters extracted to characterize the passive properties of dopaminergic neurons . The input resistance ( Rin ) , membrane capacitance ( Cm ) and the membrane time constant ( Taum ) were calculated based on the voltage response ( top , black trace ) of the neuron to a small step of hyperpolarizing current ( bottom , gray trace ) . ( B ) , voltage traces ( black ) obtained in response to current steps ( gray ) of different amplitudes in P3 ( top ) , P6 ( middle ) and P21 ( bottom ) dopaminergic neurons . ( C ) , box and whisker plot representing Rin vs postnatal age . ( D ) , box and whisker plot representing Cm vs postnatal age . ( E ) , box and whisker plot representing Taum vs postnatal age . Scale bars: A and B , vertical 5 mV , horizontal 250 ms . Dotted lines in A indicate the voltage values used to calculate ΔV . For all box and whisker plots , boxes represent the median , first and third quartile , error bars correspond to 10 and 90% , the thick line corresponds to the mean , and all outliers are represented . Colored boxes in C , D and E correspond to the age of the colored traces presented in B ( red for P3 , green for P6 , blue for P21 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04059 . 008 SNc dopaminergic neurons are characterized by a specific response to hyperpolarizing current pulses consisting of a large sag during the hyperpolarization due to the activation of IH , and a biphasic post-inhibitory rebound involving both IH and the transient potassium current IA ( Amendola et al . , 2012 ) . We analyzed the response to hyperpolarization by measuring both sag amplitude and rebound delay ( Figure 5A ) . In spite of the increase in amplitude of IH that has been described during postnatal development in SNc dopaminergic neurons ( Washio et al . , 1999 ) , we found that both sag amplitude and rebound delay remained constant during the first four postnatal weeks ( Figure 3 , Figure 5B , C , Table 1 ) , suggesting that the density of IA and IH remained fairly stable during this developmental timeframe . 10 . 7554/eLife . 04059 . 009Figure 5 . Postnatal evolution of sag and rebound delay in substantia nigra pars compacta dopaminergic neurons . ( A ) , sag amplitude and delay were extracted from the voltage response ( top , black trace ) of the neuron to a large hyperpolarizing current step ( bottom , gray trace ) . ( B ) , voltage recordings showing the typical sag and rebound delay of P3 ( top , red ) , P6 ( middle , green ) and P21 ( bottom , blue ) dopaminergic neurons in response to hyperpolarizing current steps ( gray traces ) . ( C ) , top , box and whisker plot representing sag amplitude vs postnatal age . Bottom , box and whisker plot representing rebound delay vs postnatal age . Scale bars: A and B , vertical 20 mV , horizontal 150 ms . Horizontal dotted line in A indicates the voltage peak of the hyperpolarizing response . Vertical dotted line indicates the end of the current pulse used to calculate the delay . For all box and whisker plots , boxes represent the median , first and third quartile , error bars correspond to 10 and 90% , the thick line corresponds to the mean , and all outliers are represented . Colored boxes in C correspond to the age of the colored traces presented in B ( red for P3 , green for P6 , blue for P21 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04059 . 009 SNc dopaminergic neurons are also characterized by a slow action potential ( AP ) ( Grace , 1990 ) that relies on the activation of multiple currents including calcium , transient sodium , and A-type , delayed rectifier and calcium-activated potassium currents ( Bean , 2007; Puopolo et al . , 2007 ) . In order to describe the developmental evolution of AP shape , we measured six properties defining the different phases of the AP recorded during spontaneous activity ( Figure 6A ) : AP threshold , AP rise slope , AP decay slope , AP half-width , AP amplitude , and AHP amplitude . Except for AP threshold , which was found to not vary significantly during the first four postnatal weeks ( Figure 6C , Table 1 , Figure 3 ) , most AP features were found to present significant modifications over the same developmental timeframe ( Figure 3 , Figure 6B–H , Table 1 ) . Some of these properties presented monophasic changes , such as AP decay slope , which decreased over time ( Figure 6B , F , Table 1 ) , or AHP amplitude , which increased over time ( Figure 6B , H , Table 1 ) . Surprisingly though , AP amplitude , rise slope and half-width displayed biphasic changes during the same developmental timeframe , with the first change between P2–3 and P5 and a second opposite change between P8–9 and P13 ( Figure 6B , D , E , G , Table 1 ) : AP amplitude and AP rise slope first decreased and then increased , while AP half-width presented an increase in value followed by a decrease . These changes are clearly visible on the recordings of APs from P3 , P6 , and P21 animals ( Figure 6B ) . Therefore , the different AP properties measured exhibit distinct evolutions , which suggest heterogeneous changes in expression or gating properties of the different types of ion channels involved in AP shape definition . 10 . 7554/eLife . 04059 . 010Figure 6 . Postnatal evolution of action potential properties in substantia nigra pars compacta dopaminergic neurons . ( A ) , voltage recording depicting the parameters extracted to characterize the action potential ( AP ) in dopaminergic neurons . Six parameters were extracted: rise slope , decay slope , half-width , amplitude , threshold and AHP amplitude . ( B ) , typical APs ( top traces ) recorded in P3 ( left , red ) , P6 ( center , green ) and P21 ( right , blue ) dopaminergic neurons with the corresponding phase plots representing the first time derivative of voltage vs voltage ( bottom ) . ( C ) , box and whisker plot representing AP threshold vs postnatal age . ( D ) , box and whisker plot representing AP amplitude vs postnatal age . ( E ) , box and whisker plot representing AP rise slope vs postnatal age . ( F ) , box and whisker plot representing AP decay slope vs postnatal age . ( G ) , box and whisker plot representing AP half-width vs postnatal age . ( H ) , box and whisker plot representing AHP amplitude vs postnatal age . Scale bars: A , vertical 20 mV , horizontal 2 ms; B , vertical 20 mV , horizontal 5 ms . For all box and whisker plots , boxes represent the median , first and third quartile , error bars correspond to 10 and 90% , the thick line corresponds to the mean , and all outliers are represented . Colored boxes in C–H correspond to the age of the colored traces presented in B ( red = P3 , green = P6 , blue = P21 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04059 . 010 In order to determine changes in membrane excitability , we measured the frequency/current response of SNc dopaminergic neurons in response to incremental depolarizing current pulses ( Figure 7A ) . Since SNc dopaminergic neurons have been shown to display spike frequency adaptation ( SFA ) during sustained depolarization ( Vandecasteele et al . , 2011 ) , we measured the gain of the spiking response ( in Hz/100 pA ) both at the start of the depolarizing pulse ( gain at start or GS , average frequency of the first three APs ) and at the end of the depolarizing pulse ( gain at end or GE , average frequency of the last three APs ) . An SFA index was then extracted by calculating the GS/GE ratio . While GS was found to gradually decrease over the first three postnatal weeks ( Figure 3 , Figure 7C , Table 1 ) , GE decreased mainly between P2–3 and P5 and remained stable afterwards ( Figure 3 , Figure 7D , Table 1 ) . As a consequence of these different developmental timecourses , SFA index was found to be stable between P2–3 and P8–9 and then to drop to a steady value ( Figure 3 , Figure 7E , Table 1 ) . 10 . 7554/eLife . 04059 . 011Figure 7 . Postnatal evolution of membrane excitability in substantia nigra pars compacta dopaminergic neurons . ( A ) , voltage traces depicting the parameters extracted to characterize membrane excitability . The firing frequency of the neuron was measured in response to 1 s current pulses of increasing amplitude ( from 50 pA to 300 pA for the neuron shown in this panel ) , and the starting and ending AP frequencies were calculated for each pulse , based on the first three and last three APs , respectively ( left and center panels ) . The gain at the start ( GS ) and at the end ( GE ) of the pulse were then extracted from the linear regression of the AP frequency vs current plot ( right panel ) . The spike frequency adaptation ( SFA ) index was computed as the ratio GS/GE . ( B ) , responses of P3 ( left , red ) , P8 ( center , green ) and P21 ( right , blue ) neurons to 100 pA depolarizing pulses . Voltage recordings ( bottom traces ) and corresponding timecourses of instantaneous spike frequency ( top plots ) are shown for the three developmental stages . SFA index values calculated as presented in A are given for each neuron . ( C ) , box and whisker plot representing GS vs postnatal age . ( D ) , box and whisker plot representing GE vs postnatal age . ( E ) , box and whisker plot representing SFA index vs postnatal age . Scale bars: A , vertical 20 mV , horizontal 250 ms . For all box and whisker plots , boxes represent the median , first and third quartile , error bars correspond to 10 and 90% , the thick line corresponds to the mean , and all outliers are represented . Colored boxes in C , D and E correspond to the age of the colored traces presented in B ( red for P3 , green for P8 , blue for P21 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04059 . 011 So far , we have presented a uni-variate analysis of 16 electrophysiological properties encompassing the general electrophysiological behavior of SNc dopaminergic neurons . While this analysis is very useful to determine the developmental evolution of each distinct parameter , it fails to provide a global vision of the key moments where most significant changes in electrophysiological behavior occur: some properties such as AHP amplitude , GS or CVISI display a gradual decrease over the first four postnatal weeks; some properties , such as AP amplitude , rise slope or half-width , present biphasic changes; and other properties do not present significant changes during the same developmental timeframe , such as AP threshold , sag amplitude , or rebound delay . In order to get a more global picture of the development of electrophysiological behavior , we started by representing the significant differences in each firing property between the 14 developmental stages analyzed in a ‘statistical stacking table’ ( Figure 3 ) . In this table , each major cell ( plain black line surrounding ) corresponds to one developmental stage , and the 16 minor cells constituting each major cell correspond to the electrophysiological properties presented in Figures 2 , 4 , 5 , 6 and 7 . The color coding of each minor cell then matches the level of statistical significance obtained with a one-way ANOVA comparison with post-hoc tukey correction for multiple comparisons performed between all developmental stages for each electrophysiological parameter: black corresponds to p < 0 . 001 , dark gray to p < 0 . 01 , and light gray to p < 0 . 05 . Although this table only represents independent uni-variate analyses of the changes in each firing property , the stacking of the statistical differences provides a first visual impression of where important transitions in electrophysiological behavior occur: a first transition occurs between P2–3 and P5 while a second transition occurs between P8–9 and P12–13 ( Figure 3 ) . Therefore , the second transition was less clearly defined , spanning several of the developmental stages analyzed ( a small number of statistical differences are still present between P10–P11 or P12–P13 and later stages ) . Outside of these transition phases , most electrophysiological parameters remain constant ( between P5 and P8–9 , then between P14 and P28 ) . Although the stacking of uni-variate analyses gives a good indication of when major transitions are occurring , it still fails to easily and rapidly identify which properties most significantly change during the two transitions . The two major transitions identified in Figure 3 supposedly separate three different developmental classes of electrophysiological behaviors associated with significant changes in the values of the intrinsic properties analyzed so far . In order to confirm this hypothesis and determine the key/important changes in properties involved , we performed agglomerative hierarchical clustering ( AHC ) analysis . We chose to include only the 8 electrophysiological parameters that can be extracted from the recordings of spontaneous activity ( ISIavg , CVISI , AP threshold , AP amplitude , AP rise slope , AP decay slope , AP half-width , and AHP amplitude ) and excluded the parameters that require current injection to be measured ( Rin , Cm , τm , sag amplitude , rebound delay , GS , GE , SFA index ) , the goal being to determine whether phenotype can be accurately defined with minimal electrophysiological manipulation ( no current injection ) . Using AHC with an automatic setting of the dissimilarity threshold ( XLSTAT software ) , the 263 recorded neurons were separated into three classes of heterogeneous sizes ( Figure 8A ) . The observation of the recordings corresponding to the central objects of each class reveals that bursting , irregular and regular tonic neurons were clustered in separate classes ( class 1 , 2 and 3 , respectively; Figure 8B ) . Unsurprisingly , CVISI was one of the parameters presenting the most significant differences between the central objects ( Figure 8B ) or the centroids ( Figure 8C ) of each class . Nonetheless , other parameters such as AP half-width , AP rise slope , AP decay slope , and AHP amplitude also presented notable differences between the three classes ( Figure 8B , C ) . Although age was not included as one of the variables in the AHC analysis , the age of the animals corresponding to each recorded neuron was known such that we were able to determine the age of each observation ( Figure 8A ) , of the central objects ( see Figure 8B ) but also the mean age for each class ( Figure 8C ) . Consistent with the developmental timecourse of firing pattern depicted in Figure 2E , the mean ages were P3 . 5 , P7 . 44 , and P17 . 32 for class 1 , 2 , and 3 , respectively ( Figure 8C ) . Therefore , AHC analysis based solely on spontaneous activity-related electrophysiological parameters clearly discriminates between the three developmental stages ( or electrophysiological behaviors ) observed between P2 and P28 . 10 . 7554/eLife . 04059 . 012Figure 8 . Agglomerative hierarchical clustering analysis of electrophysiological classes of substantia nigra pars compacta dopaminergic neurons during postnatal development . ( A ) , dendrogram representing the agglomerative hierarchical clustering ( AHC ) of the 263 P2–P29 recorded neurons into three classes based on the 8 electrophysiological parameters ( a to h ) listed in panel C . The graded gray heat map represents the value of each electrophysiological parameter for each neuron normalized to the mean of the parameter for the whole population ( scale bar at the bottom of the graph ) . The age of each neuron is plotted on the right and gives an indication of the relationship between the 3 electrophysiological classes and postnatal age . ( B ) , voltage recordings corresponding to the central objects of each class showing the differences in spontaneous activity patterns and AP shape associated with each class . ( C ) , table presenting the values of the 8 electrophysiological parameters for each class centroid . As an indication , the average age of each class was calculated and is given in the last row in gray ( mean ± SD ) . Scale bars: B , vertical 20 mV , horizontal 2 s and 5 ms for the left and right column traces , respectively . The gray and black dotted lines in panel B indicate −60 mV and the AP threshold , respectively . Colors in A and B indicate the three classes ( red for class 1 , green for class 2 , blue for class 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04059 . 012 Based on the parameter values of the centroids of each class , it is clear that most parameters do not follow a similar trajectory during development ( CVISI decreases monophasically while AP rise slope and AP amplitude present biphasic changes ) . Thus , we sought an analysis method that would provide a precise visualization of the developmental trajectory of the electrophysiological behavior of SNc dopaminergic neurons . By reducing a high-dimensional parameter space into two or three principal components that account for most of the variance observed in a sample , PCA allows coordinated changes in parameter values to be visualized . We ran PCA based on the covariance matrix ( n ) of the same observations/variables table used for the AHC analysis . The first two principal components ( PC1 and PC2 ) accounted for ∼94% of the variance of the variables ( Figure 9A ) , and therefore were chosen for the two-dimensional representation of the developmental evolution of the 8 electrophysiological parameters during postnatal development . The contribution of each electrophysiological parameter to PC1 and PC2 is represented in Figure 9A: while CVISI and AHP mainly contribute to PC1 , other parameters such as AP amplitude , AP rise slope , AP decay slope , and AP half-width contribute to a similar extent to PC1 and PC2 . Finally , ISIavg and AP threshold show minor contributions to both PCs . 10 . 7554/eLife . 04059 . 013Figure 9 . Principal component analysis of the electrophysiological behavior of substantia nigra pars compacta dopaminergic neurons during postnatal development . ( A ) , polar plot representing the respective contribution of each of the 8 electrophysiological parameters to the two principal components retained from the PCA ( PC1 and PC2 ) . ( B ) , scatter plot representing the factor loadings of the 263 P2–P29 neurons in the PC2 vs PC1 space . The points are color-coded as a function of the AHC class: red circles for class 1 , green squares for class 2 and blue triangles for class 3 . Light color symbols correspond to individual neurons while bright color symbols represent the averaged factor loadings for each class ( error bars represent the SD ) . ( C1 ) , scatter plot representing the factor loadings of the 263 P2–P29 neurons ( gray circles ) and the averaged factor loadings for each developmental stage ( colored circles ) . Error bars represent the standard error of the mean . ( C2 ) , expanded version of the scatter plot presented in C1 centered on the later developmental stages ( P12–P13 to P28–P29 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04059 . 013 Since the AHC and the PCA were run on the same observations/variables table , we first sought to determine whether PCA and AHC gave consistent results . To do so , we plotted the PC1 and PC2 factor loadings of the three classes identified by the AHC . Consistent with the AHC results , the three classes were found to cluster in distinct regions of the PC1/PC2 space ( Figure 9B ) . Since AHC classes were found to be associated with specific developmental stages ( Figure 8C ) and clustered in separate regions of the PC1/PC2 space , we wondered whether PCA could give more specific insights into the developmental trajectory of the electrophysiological properties of SNc dopaminergic neurons . Although age was not included as one of the parameters of the PCA , the mean PC1 and PC2 values and the associated standard deviations were calculated for each developmental stage analyzed , and could be represented in the PC1/PC2 space ( Figure 9C ) : the distribution of values demonstrated that the developmental trajectory of the electrophysiological properties of SNc dopaminergic neurons is composed of two distinct phases , a first phase where PC1 and PC2 simultaneously decrease ( between P2–3 and P5–11 ) and a second phase where PC1 keeps on decreasing while PC2 increases ( between P11 and P15 ) ( Figure 9C1 ) . Interestingly , PCA also reveals that the electrophysiological behavior of SNc dopaminergic neurons in vitro reaches a steady state by P14–15 , and that no significant change is observed after this stage , although some variability in the electrophysiological properties might still be present ( Figure 9C2 ) . Therefore , PCA demonstrates that the electrophysiological development of SNc dopaminergic neurons is composed of two phases that involve changes in different sets of electrophysiological properties , most likely indicating that different ion currents evolve with distinct developmental timecourses . Although averaging the data for each developmental stage was necessary to analyze the developmental trajectory ( Figure 9C ) , another important aspect to investigate is the variability in electrical phenotype observed for a given developmental stage ( Marder and Goaillard , 2006 ) . In particular , since the mean electrical phenotype reaches a steady-state by P14–P15 , we tried to determine whether this stabilization was also associated with changes in the variability of the phenotype . Since PCA involves a normalization step , standard deviations of the PCs can be used as an index of variability ( using the coefficient of variation is meaningless in this case ) : the error bars representing the standard deviations of the PCs in Figure 9B already suggest that variability in phenotype is higher for immature classes ( 1 and 2 ) than for the mature class ( 3 ) . In order to quantify the variability for each developmental stage , we extracted the values of the standard deviations of PC1 and PC2 ( SD PC1 and SD PC2 , respectively ) for each developmental stage presented in Figure 9C , and considered that the overall variability in phenotype could be illustrated by the changes in the surface area of the ellipse defined by these two values ( A = π × ( SD PC1 ) × ( SD PC2 ) ; Figure 10 ) . This analysis showed that phenotype is more variable at early than late developmental stages . Consistent with this analysis , and since PC1 accounts for ∼65% of the variance of our observations ( vs ∼28% for PC2 ) , the standard deviation of PC1 was found to be significantly and negatively correlated with age ( r = 0 . 802 , p < 0 . 001 , n = 15 ) . In summary , the multi-variate analyses reveal that the electrical phenotype of SNc dopaminergic neurons follows a progressively narrowing biphasic developmental trajectory . 10 . 7554/eLife . 04059 . 014Figure 10 . Developmental timecourse of electrical phenotype variability . Bubble plot representing the variability in electrical phenotype for each developmental stage in Figure 9C . The size of each ellipse is determined by the values of the standard deviations for PC1 ( horizontal semi-axis , a ) and PC2 ( vertical semi-axis , b ) , as indicated in the inset . The Y-coordinate of each ellipse ( variability ranking value ) is then determined by the surface area A of each ellipse , which is proportional to phenotype variability as defined in the PCA space . Each ellipse is positioned along the X-axis according to the developmental stage it represents ( gray drop line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04059 . 014 The question that arises next is whether a few specific biophysical mechanisms may explain the main developmental transitions in electrical phenotype . Although numerous voltage-dependent ion channels are involved in defining the electrical phenotype of SNc dopaminergic neurons and may be involved in these transitions , we focused on the role of three conductances: the apamin-sensitive calcium-activated potassium current , the TTX-sensitive sodium current , and the leak current . First , we investigated the transition in electrical phenotype occurring between P2–P3 and P5–P7 , which is characterized by a sudden decrease in Rin , CVISI , AP amplitude , and AP rise slope ( Table 1 , Figure 3 ) . A steep increase in the amplitude of the AHP is also observed between these two stages ( Table 1; Figure 6H ) . Since the apamin-sensitive calcium-activated potassium current responsible for the AHP has been linked to firing regularity ( Ping and Shepard , 1996; Wolfart et al . , 2001 ) , and a sudden decrease in Rin might explain changes in the properties of the AP ( especially its amplitude ) , we tested whether the manipulation of these two parameters could reproduce the phenotypic transition observed between P2–P3 and P5–P7 . Unfortunately , it is not possible to accurately simulate the increase in apamin-sensitive AHP occurring between P2–P3 and P5–P7 using dynamic-clamp , because of the calcium-dependence of this conductance . On the other hand , apamin very specifically blocks these conductances and simulates a decrease in AHP . Therefore , P7–P8 neurons were recorded , and apamin application ( 100 nM , to block 75–80% of the AHP; Wolfart et al . , 2001 ) and injection of a dynamic clamp-generated negative leak conductance ( average −0 . 55 nS , to simulate the increase in input resistance ) were used in an attempt to reverse their electrophysiological phenotype to the P2–P3 stage . As expected , apamin application induced a decrease in AHP ( Figure 11A , Figure 11—figure supplement 1 ) associated with an increase in CVISI , corresponding to a switch from regular to irregular firing ( Figure 11A ) . Artificially increasing Rin by the injection of a leak negative conductance amplified the apamin effect , resulting in the appearance of burst firing ( Figure 11A ) , consistent with the results obtained by Paladini et al . ( 1999 ) . However , artificially subtracting leak before apamin application failed to induce a significant change in phenotype ( CVISI = 30% vs 27% before and after leak subtraction , respectively , n = 2 ) . In order to determine whether these manipulations were able to precisely mimic the developmental transition , the PCA factor loadings corresponding to the neurons in control , apamin , and apamin—leak conditions were plotted in the PCA space ( Figure 11B ) . This analysis demonstrated that apamin application and leak subtraction induced an electrophysiological regression of P8 neurons to a P2–P3 phenotype ( Figure 11B , E ) , each isolated manipulation accounting for roughly 50% of the change in electrophysiological phenotype . The analysis of the changes in the individual electrophysiological properties also supported this conclusion ( Figure 11—figure supplement 1 ) . 10 . 7554/eLife . 04059 . 015Figure 11 . Reproducing the developmental transitions in electrical phenotype using pharmacology and dynamic clamp . ( A ) , current-clamp recordings ( top traces ) obtained from a P8 neuron in control condition ( left , green ) , during apamin application ( center , orange ) , and during apamin application + negative leak injection ( right , red ) . The middle traces ( gray ) correspond to the dynamic clamp-injected current while the lower traces show the changes in shape of the AHP in the different conditions . ( B ) , average factor loadings obtained for 10 P8 neurons recorded in the three conditions presented in A and plotted in the PCA space presented in Figure 9B . The background gray points correspond to the three classes identified using AHC , and are plotted as a reference . ( C ) , current-clamp recordings ( top traces ) of a P14 neuron recorded in control condition ( left , blue ) and in the presence of apamin and TTX ( right , green ) . The lower traces show the changes in the AP induced by apamin + TTX . ( D ) , average factor loadings obtained for 12 P14–16 neurons recorded in the two conditions presented in C and plotted in the PCA space presented in Figure 9 . The background gray points correspond to the three classes identified using AHC , and are plotted as a reference . ( E ) , summary of the changes in phenotype induced by the manipulations presented in A–B and C–D . The total vectors corresponding to apamin − leak ( P8 neurons ) and apamin + TTX ( P14–16 neurons ) were locked on the averages of the classes 2 and 3 of the AHC ( corresponding to average ages of P7 and P17 , respectively ) . Scale bars: A , vertical 20 mV ( top and bottom traces ) and 10 pA ( middle traces ) , horizontal 2 s ( top and middle traces ) and 50 ms ( bottom traces ) . ( C ) , vertical 20 mV , horizontal 1 s and 5 ms for the top and bottom traces , respectively . The black dotted lines in panel A and C indicate −60 mV . DOI: http://dx . doi . org/10 . 7554/eLife . 04059 . 01510 . 7554/eLife . 04059 . 016Figure 11—figure supplement 1 . Effect of dynamic-clamp and pharmacological manipulations on 8 individual electrophysiological properties . Scatter plots representing the changes in the 8 electrophysiological properties ( ISIavg , CVISI , AHP amplitude , AP threshold , AP amplitude , AP half-width , AP rise slope , AP decay slope ) corresponding to the changes in PCA factor loadings presented in Figure 11B , D . Paired statistical tests ( paired t test or Wilcoxon signed rank test ) were performed between the initial ( P14–P16 ctrl , P8 ctrl ) and the final conditions ( P14–P16 + apamin + TTX , P8 + apamin − leak , respectively ) and the corresponding p values are shown . Significant statistical differences appear in black while non-significant differences appear in gray . Data are presented as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 04059 . 016 We then wondered whether we could also reverse the electrical phenotype of P15 neurons to the P7 stage . The transition between P7 and P15 is mainly characterized by an increase in AHP and AP amplitude ( and an increase in the AP rise and decay slopes ) , but no dramatic change in Rin ( Table 1; Figure 3 , Figure 5 , Figure 6 ) . In order to reverse the electrophysiological phenotype of dopaminergic neurons from P14–16 to P7 , we tested the effect of reducing AHP amplitude ( using 5 nM apamin to block 20–25% of the AHP; Wolfart et al . , 2001 ) and reducing the sodium current involved in the action potential ( using 20 nM TTX ) . Both applications were performed simultaneously . As shown in Figure 11C ( see also Figure 11—figure supplement 1 ) , the apamin/TTX application induced an increase in CVISI and a decrease in AP amplitude ( and rise and decay slopes ) . Again , all electrophysiological parameters involved in the PCA were measured in these conditions ( Figure 11—figure supplement 1 ) , and we could plot the PCA factor loadings corresponding to the control and apamin/TTX-treated P14–P16 neurons in the PCA space ( Figure 11D ) . This analysis demonstrated that the apamin/TTX application induced a phenotype switch quantitatively similar ( and opposite in sign ) to the one observed between P7 and P15 ( Figure 11D , E ) .
So far , two main types of spontaneously occurring intrinsic patterns of activity have been described for SNc dopaminergic neurons in acute slices: a ‘mature’ regular tonic activity ( or pacemaking ) in animals above P15 ( Grace and Onn , 1989; Liss et al . , 2001; Seutin et al . , 2001; Puopolo et al . , 2007; Guzman et al . , 2009; Putzier et al . , 2009b; Tateno and Robinson , 2011; Amendola et al . , 2012 ) and an ‘immature’ irregular pattern of activity in juvenile animals ( between P6 and P15 ) ( Seutin et al . , 1998; Cui et al . , 2004 ) . Our results extend these observations by showing that an additional intrinsic pattern of activity , a bursting pattern , can be observed in acute slices from neonatal rats ( before P8 ) . As demonstrated in Figure 2 , SNc DA neurons switch from bursting to irregular and regular activity over the first three postnatal weeks , these transitions corresponding to a non-linear monophasic drop in the coefficient of variation ( inversely proportional to activity regularity ) of the interspike interval . While bursting activity had been previously observed in vitro in response to glutamatergic receptor agonists ( Mereu et al . , 1997; Johnson and Wu , 2004 ) or pharmacological blockade of small-conductance calcium-activated potassium ( SK ) currents ( Wolfart and Roeper , 2002; Johnson and Wu , 2004 ) , we demonstrate that SNc dopaminergic neurons are purely intrinsic bursters at very early stages of development , independent of the recording condition ( cell-attached , whole-cell , EGTA concentration ) . Interestingly , a very recent study using calcium imaging demonstrated that a significant proportion of SNc dopaminergic neurons intrinsically generate calcium spikes or calcium plateaus between E16 and P0 , while this calcium activity strongly decreases by P7 ( Ferrari et al . , 2012 ) . Since bursting behavior usually relies on the activation of calcium conductances ( Destexhe and Sejnowski , 2003 ) , including L-type calcium channels in SNc dopaminergic neurons ( Johnson and Wu , 2004 ) , our observation that most SNc dopaminergic neurons at P2–P3 are intrinsic bursters corroborates the findings of Ferrari et al . , ( 2012 ) . To date , however , the physiological significance of this intrinsic bursting is unknown . In several systems including the retina and the spinal cord ( Moody and Bosma , 2005 ) , spontaneous waves of activity early in development have been shown to participate in network development/synaptic refinement . Calcium has a central role in this process , and the patterns of activity observed in these systems ( synchronized bursts of action potentials ) promote intracellular calcium rises ( Moody and Bosma , 2005 ) . While midbrain dopaminergic neurons already project to the striatum at embryonic stage E16 and release dopamine at E18 ( Hu et al . , 2004; Ferrari et al . , 2012 ) , the refinement of their projections continues at least until P0 ( Hu et al . , 2004 ) , and it is likely that this pruning involves calcium-dependent mechanisms . The intrinsic bursting observed in SNc dopaminergic neurons at early postnatal stages may promote this process . From a biophysical point of view , regular tonic activity relies on many different voltage-dependent intrinsic conductances , including transient sodium currents ( Puopolo et al . , 2007; Guzman et al . , 2009; Drion et al . , 2011; Tucker et al . , 2012 ) , high and low voltage-activated calcium currents ( Chan et al . , 2007; Puopolo et al . , 2007; Guzman et al . , 2009; Putzier et al . , 2009a ) , transient potassium currents ( Liss et al . , 2001; Hahn et al . , 2003; Putzier et al . , 2009b; Amendola et al . , 2012 ) , and hyperpolarization-activated currents ( Seutin et al . , 2001; Neuhoff et al . , 2002; Chan et al . , 2007; Tateno and Robinson , 2011 ) , while the irregular pattern of activity observed in juvenile animals seems to be strongly dependent on the sporadic activation of the apamin-sensitive SK currents via the spontaneous activation of T-type calcium channels ( Seutin et al . , 1998; Cui et al . , 2004 ) . In fact , SK currents have been demonstrated to control the regularity of spontaneous activity in SNc dopaminergic neurons from both juvenile and mature animals in vitro ( Ping and Shepard , 1996; Wolfart et al . , 2001 ) . Moreover , SK current activation seems to be essential to the regular tonic pattern of activity as blockade of SK currents , under specific experimental conditions , induces a switch to a bursting pattern of activity , both in vitro ( Wolfart and Roeper , 2002; Johnson and Wu , 2004 ) and in vivo ( Waroux et al . , 2005 ) . To summarize these observations , although irregular juvenile activity seems to involve the activation of SK conductances ( via T-type calcium currents ) , the ‘mature’ regular tonic pattern of activity also relies on the strong activation of these same conductances . Consistent with these numerous indications of the role of SK currents in the definition of the type of activity generated by SNc dopaminergic neurons , the switch in activity pattern we observed over the first two postnatal weeks goes along with a developmental increase in the amplitude of the SK-mediated AHP ( Figure 6H ) . In conclusion , our results and the published literature suggest that the gradual switch between bursting , irregular and regular intrinsic activity is at least partly due to a developmental increase in the amplitude of SK currents , or at least of the SK-mediated AHP . These results are also consistent with a recent study that demonstrated that spike frequency adaptation ( SFA ) in SNc dopaminergic neurons depends on SK currents , and that SFA decreases between P6 and P21 ( Vandecasteele et al . , 2011 ) , an observation also made in the current study ( Figure 7 ) . Altogether , our results suggest that the developmental increase in the SK-mediated AHP over the first three postnatal weeks is essential to the maturation of SNc dopaminergic neuron spontaneous and evoked firing patterns . The pharmacology and dynamic-clamp experiments presented in Figure 11 confirm the central role of SK currents in defining the electrical phenotype of SNc dopaminergic neurons . Adjusting AHP amplitude to levels corresponding to specific developmental stages using apamin ( see Figure 11 and Figure 11—figure supplement 1 ) strongly contributes to reversing the overall electrical phenotype of SNc dopaminergic neurons to these stages . The electrophysiological measurements of passive properties ( input resistance , membrane capacitance , membrane time constant ) we obtained in the current study extend previous results on the morphological changes of SNc dopaminergic neurons during postnatal development: SNc dopaminergic neurons were shown to reach morphological maturity by P14 , since neither soma size ( Tepper et al . , 1994 ) nor cell area ( Park et al . , 2000 ) change significantly from P14 to adulthood . Consistent with these observations , our measurements show that membrane capacitance is fairly stable after P10–P11 . Moreover , our data seem to indicate that membrane capacitance is also fairly stable , at lower values , between P2–P3 and P8–P9 . Input resistance , however , shows a different developmental profile , with a very sudden drop between P2–P3 and P5 , and then a gradual decrease until P10–P11 . Therefore , the detailed sampling of early developmental stages seems to indicate that the morphology-related passive properties of SNc dopaminergic neurons reach maturity by the middle of the second postnatal week ( P10–P11 ) , extending previous morphological measurements ( Tepper et al . , 1994; Park et al . , 2000 ) . Surprisingly , the measurements of sag ampitude and rebound delay showed that these two properties are fairly stable across the first four postnatal weeks , although IH , which strongly contributes to these features ( Washio et al . , 1999; Franz et al . , 2000; Neuhoff et al . , 2002; Amendola et al . , 2012 ) , has been reported to increase over the first two postnatal weeks ( Washio et al . , 1999 ) . Our current interpretation is that the drop in input resistance observed during the same timeframe counteracts the reported increase in IH amplitude ( Washio et al . , 1999 ) , such that sag amplitude and rebound delay remain constant . One of the most surprising findings of the current study is that some of the properties of the AP show a clear biphasic trajectory over the first two postnatal weeks . In particular , AP amplitude and AP rise slope decrease from P2 to P5–P7 , then increase to reach a steady-state by P11–P15 . So far , previously published studies on SNc dopaminergic neurons have described more classical changes in these parameters over the same time range , that is , a monophasic increase in amplitude associated with an increase in AP rise slope ( or the subsequent decrease in AP half-width ) ( Washio et al . , 1999; Ferrari et al . , 2012 ) , although fewer developmental stages and smaller samples were analyzed compared to the current study . Nevertheless , previous studies dedicated to mouse cerebral cortex and hippocampus development described the same type of monophasic increase in amplitude and acceleration of the AP ( McCormick and Prince , 1987; Spigelman et al . , 1992; Picken Bahrey and Moody , 2003; Williams and Sutherland , 2004 ) . One of these studies demonstrated that it was due to an increase in the amplitude of the transient sodium current over the first two postnatal weeks ( Picken Bahrey and Moody , 2003 ) , consistent with the role of this current in defining the rise kinetics and the amplitude of the AP ( Bean , 2007 ) . Several factors could explain the biphasic timecourse observed in the current study . First , the analysis of passive properties demonstrated that the electrotonic size of SNc dopaminergic neurons suddenly increases between P2–P3 and P5 ( Figure 4 ) , mainly due to a drop in input resistance . Such a drop in input resistance , in the absence of concomitant changes in sodium channel number , would cause a sudden decrease in the ability of sodium currents to depolarize the neuron , and could explain the drop in AP amplitude and AP rise slope . Other morphological changes could also explain these biphasic changes in AP properties . For instance , it is known that the axon initial segment ( AIS ) , the main site of AP initiation , is located at highly variable distances from the soma ( between 0 and 250 µm ) in mature SNc dopaminergic neurons ( Hausser et al . , 1995 ) . Recent studies have demonstrated that mild changes in the location of the AIS can significantly influence the response of the neuron to sensory inputs ( Kuba et al . , 2006 , 2010 ) or current injection ( Grubb and Burrone , 2010a , 2010b ) . Along the same line , and since P2–P3 SNc dopaminergic neurons have a significantly smaller size than at later stages , one could hypothesize that these changes are associated with significant displacements of the AIS that would alter AP properties . However , this hypothesis would need further investigation . When a phenotype cannot be captured by a single characteristic and is better described by many distinct properties , as is the case for the electrophysiological phenotype of SNc dopaminergic neurons , analyzing all parameters in parallel using univariate analysis fails to provide a satisfying picture of the phenotypic variations , especially if the different parameters do not co-vary . While multivariate analyses such as PCA are designed to solve that problem by compressing the number of meaningful parameters ( see next section ) , we propose an intermediate solution which consists of a visual stacking of the results of the uni-variate analyses performed on each parameter . Dimensional stacking entails visualizing a high-dimensional data set in a two-dimensional space and has been recently applied to analyze the electrical behavior of databases of complex realistic neuron models ( Taylor et al . , 2006 ) . In the current study , we performed ANOVA on 16 electrophysiological parameters across 14 developmental stages , thus , to compress the results into two visible dimensions , we ( i ) designed a two entry-table comprising 105 cells corresponding to all stage-to-stage comparisons ( P2–P3 vs P5 , P2–P3 vs P6 , etc ) ( ii ) divided each of these 105 cells into 16 sub-cells corresponding to the 16 electrophysiological parameters analyzed at each developmental stage , and ( iii ) color-coded the sub-cells as a function of the statistical significance of the ANOVA test performed ( non significant , p < 0 . 05 , p < 0 . 01 , p < 0 . 001 ) . As a consequence , although Figure 3 strictly represents the results of separate univariate analyses , it also provides a general view of the changes in the electrical phenotype of dopaminergic neurons , since each cell summarizes the differences in 16 electrophysiological parameters between two specific developmental stages: while the evolution of 1 electrophysiological parameter can be tracked through development by looking at one specific sub-cell , looking at the mosaic pattern of the large cells gives an indication of the general change in electrophysiological phenotype . This type of stacked representation is likely to be useful to anyone analyzing changes in a large number of parameters across a large number of states , and is increasingly used when analyzing the behavior of databases of models ( Taylor et al . , 2006; Gutierrez et al . , 2013 ) . In our case , the statistical stacking revealed two main transitions in electrical phenotype occurring between P2–P3 and P5 and between P8–P9 and P12–P13 , respectively . So far , multivariate analyses , and in particular clustering analysis , have been mainly used in neurophysiology to classify and distinguish mature neuronal populations ( Ascoli et al . , 2008; Karagiannis et al . , 2009; McGarry et al . , 2010; Laramee et al . , 2013; Simonnet et al . , 2013 ) . For instance , research dedicated to the classification of cortical interneuron subpopulations ( for review see Ascoli et al . , 2008; Defelipe et al . , 2013 ) has led to the creation of an international consortium ( the Petilla Interneuron Nomenclature Group ) . This type of classification often relies on the use of hierarchical clustering analysis or similar techniques applied to a combination of morphological ( axon and dendrite length and shape… ) , molecular ( expression of specific genes ) , and electrophysiological ( frequency of spiking , frequency adaptation… ) criteria ( Karagiannis et al . , 2009; Battaglia et al . , 2013 ) . Principal component analysis has also been used to analyze the diversity of cortical interneurons ( McGarry et al . , 2010 ) or the anatomical differences of cortical principal neurons between different cortical areas ( Laramee et al . , 2013 ) . In the current study , we show that AHC and PCA can also be efficiently applied to the analysis of developmental changes in electrical behavior in a given population of neurons . In particular , the biphasic developmental trajectory of the electrical behavior of SNc dopaminergic neurons over the first four postnatal weeks is captured by both types of analysis: AHC identifies three separate classes of neurons ( P3 , P7 , and P17 , respectively ) while PCA identifies two main transitions ( one between P3 and P5 , the other between P9 and P12 ) . We also show that both types of analysis give consistent results , as the three AHC classes cluster in separate regions of the PCA space ( Figure 9B ) . PCA reveals a very clear biphasic developmental trajectory , with ( i ) a concerted decrease in PC1 and PC2 between P2–P3 and P5 and ( ii ) a decrease in PC1 coupled with an increase in PC2 between P8–P9 and P12–P13 . This biphasic trajectory may indicate that the different voltage-dependent ion channels involved in the spontaneous firing pattern of SNc dopaminergic neurons reach their mature expression pattern with heterogeneous timecourses . Moreover , PCA also clearly reveals the stationarity of the electrical phenotype after P12–P13 ( Figure 9C2 ) . Thus , the current study demonstrates that AHC and PCA can be used to precisely analyze the developmental trajectory of a specific population of neurons , and could be applied in the future to examine variations in neuronal electrophysiological phenotypes ( eg . , in disease states ) and to criteria other than electrophysiological parameters . Since each principal component is defined by the specific relative contributions of 8 electrophysiological parameters , PCA gives insight into which parameters may play a critical role in phenotype variation , and therefore , offers the possibility to investigate the participation of specific biophysical mechanisms in phenotypic variations . With appropriate methods ( for instance specific toxins and dynamic-clamp ) , it is then possible to modify the activity of specific ion channels and measure their precise effect on phenotype ( visualized in the PCA space ) . Using that type of approach , we were able to determine that the two developmental transitions can be explained by distinct combinations of changes in the activity of leak , sodium , and calcium-activated potassium currents . While the increase in calcium-activated potassium current is involved in both transitions , the first transition ( P3 to P5 ) involves a dramatic concomittant increase in leak conductance whereas the second transition ( P9 to P11 ) may rely on a concomittant increase in fast sodium conductance . Although we cannot conclude that these biophysical mechanisms are the only ones involved in these developmental transitions , our experiments demonstrate that manipulating these parameters quantitatively reproduces the changes in electrical phenotype observed between these stages ( see Figure 11E ) . The multi-variate analyses we performed not only revealed that the electrical phenotype of SNc dopaminergic neurons was following a biphasic trajectory , it also revealed that the variability in phenotype was decreasing across development ( Figure 10 ) . This observation is particularly interesting and consistent with the idea of canalization of development formulated by the English embryologist Conrad Hal Waddington ( Waddington , 1942 , 1957; Debat and David , 2001 ) . In this theory , Waddington postulated the following , referring in particular to organ development: ‘In a canalised system of the kind we have been considering , trajectories starting from any point within a certain volume will converge to a single end point which is the corresponding steady-state…’ ( Waddington , 1957 ) . This precisely describes what is occurring with SNc dopaminergic neuron's phenotype as defined in the PCA space: the variability in phenotype ( and the corresponding region of the PCA space ) is much larger at early developmental stages ( P2–P5 ) than at late developmental stages where all values cluster in a single point of space ( Figures 9C and 10 ) . This suggests that neurons may be very heterogeneous early in development and then use flexible developmental trajectories to reach a highly ‘attractive’ mature phenotype . What does this mean from a biophysical point of view ? Previous studies have shown that mature neurons generate highly similar patterns of activity using variable solutions of underlying conductances , commonly displaying twofold to fourfold ranges of expression/amplitude of ion channels/currents ( Marder and Goaillard , 2006; Schulz et al . , 2006 , 2007; Goaillard et al . , 2009; Taylor et al . , 2009; Amendola et al . , 2012 ) . For instance , mature SNc dopaminergic neurons can generate regular tonic activity patterns of similar frequencies while displaying substantial differences in the amplitude and gating properties of IA and IH ( Amendola et al . , 2012 ) . These findings may imply that immature neurons , showing heterogeneous phenotypes , display even larger variability in their underlying conductances . The developmental profiles of ion channel expression rapidly change over the first two postnatal weeks , as has been shown in many systems ( Moody and Bosma , 2005 ) , including the SNc dopaminergic neurons ( Washio et al . , 1999; Dufour et al . , 2014 ) . Slight differences in the developmental timecourses of neurons ( visible only at early developmental stages ) would certainly exacerbate the cell-to-cell variability in expression of ion channels already observed in mature neurons , and would reveal even larger ranges of expression of ion channels in immature neurons . To define what a neuron's electrical phenotype is , one would need to know what the function of a given neuronal type is . Although certain neuronal types such as motorneurons or sensory neurons may have clearly identifiable functions , determining the information ‘coded’ by other neuronal types ( such as the dopaminergic neurons analyzed in the current study ) may be much more challenging and debatable . As long as we do not know what the function of the neuron is , we cannot determine which of its electrophysiological properties are the most critical to its function ( AP shape , firing frequency , average membrane potential ? ) . How can we then define a neuron's electrical phenotype ? In the current study , we proposed one defensible approach , which consists of measuring as many electrophysiological properties as possible in the same individual neuron , and considering that the electrical phenotype of the neuron is defined in this high-dimensional space of parameters . The use of multi-variate analysis such as AHC and PCA then reduces the number of meaningful dimensions in order to obtain quantitative visualizations of phenotypic variations . This type of analysis is likely to become increasingly important as we move towards better defining neuronal types and phenotypes in an unbiased manner and understanding how these neuronal characteristics develop and are altered by perturbation and disease .
Acute slices were prepared from P2–P29 Wistar rats of either sex . All experiments were performed according to the European and institutional guidelines for the care and use of laboratory animals ( Council Directive 86/609/EEC and French National Research Council ) . Rats were anesthetized with halothane ( Nicholas Piramal India ) and decapitated . The brain was immersed briefly in oxygenated ice-cold low-calcium artificial cerebrospinal fluid ( aCSF ) containing the following ( in mM ) : 125 NaCl , 25 NaHCO3 , 2 . 5 KCl , 1 . 25 NaH2PO4 , 0 . 5 CaCl2 , 4 MgCl2 , 25 glucose , pH 7 . 4 , oxygenated with 95% O2/5% CO2 gas . The cortices were removed and then coronal midbrain slices ( 250 μm ) were cut on a vibratome ( Leica VT 1200S ) in oxygenated ice-cold low calcium aCSF . Following 30–45 min incubation in 32°C oxygenated low calcium aCSF the slices were incubated for at least 30 min in oxygenated aCSF ( 125 NaCl , 25 NaHCO3 , 2 . 5 KCl , 1 . 25 NaH2PO4 , 2 CaCl2 , 2 MgCl2 and 25 glucose , pH 7 . 4 , oxygenated with 95% O2/5% CO2 gas ) at room temperature prior to electrophysiological recordings . Picrotoxin ( 100 µM , Sigma Aldrich , St . Louis , MO ) and Kynurenate ( 2 mM , Sigma Aldrich ) were bath-applied via continuous perfusion in aCSF to block inhibitory and excitatory synaptic activity , respectively . All recordings ( 315 cells from 120 rats ) were performed on midbrain slices continuously superfused with oxygenated aCSF at 30–32°C . Picrotoxin and Kynurenate were systematically present for all recordings to prevent contamination of the intrinsic activity by spontaneous glutamatergic and GABAergic synaptic activity . Patch pipettes ( 1 . 8–4 . 5MΩ ) were pulled from borosilicate glass ( GC150TF-10 , Harvard Apparatus , Holliston , MA ) on a DMZ Universal Puller ( Zeitz Instruments , Germany ) . Patch solution contained in mM: 20 KCl , 10 HEPES , 10 EGTA , 2 MgCl2 , 2 Na-ATP and 120 K-gluconate , pH 7 . 4 , 290–300 mOsm; except for 40 cells where a patch solution of same composition but containing only 0 . 5 mM EGTA was used ( see Figure 2—figure supplement 1 ) . Neurobiotin ( 0 . 05%; Vector Labs , Burlingame , CA ) was included in the intracellular solution to allow identification of dopaminergic neurons using post-hoc tyrosine-hydroxylase immunolabeling ( Amendola et al . , 2012 ) ( Figure 1 ) . Whole-cell recordings were made from SNc dopaminergic neurons visualized using infrared differential interference contrast videomicroscopy ( QImaging Retiga camera; Olympus BX51WI microscope ) and identified as previously described ( Amendola et al . , 2012 ) ( Figure 1 ) . Whole-cell current-clamp recordings with an uncompensated series resistance <10 MΩ were included in the study , and the bridge was compensated . Capacitive currents and liquid junction potential ( +13 . 2 mV ) were compensated online and offset potentials were measured after removing the pipette from the neuron . Recordings with offset values above 1 mV were discarded from the analysis . Recordings were acquired at 10 kHz or 20 kHz ( action potential measurements ) and were filtered with a low-pass filter ( Bessel characteristic 2 . 9 kHz cutoff frequency ) . The interspike interval ( ISI ) and ISI coefficient of variation ( CVISI ) were calculated from a minimum of 40 s of stable current-clamp recording ( with no injected current ) within the first 5 min of obtaining the whole-cell configuration . Action potentials ( APs ) generated during this period of spontaneous activity were then averaged and the AP threshold , peak amplitude , rise slope , decay slope , afterhyperpolarization ( AHP ) amplitude ( AP threshold to trough of the AHP ) , and the duration of the AP at half of the maximal height of the AP ( AP half-width ) were measured . AP threshold was defined as the membrane voltage value reached when the voltage first derivative rises to 5% of its peak amplitude . Rise and decay slopes correspond to the average slopes measured between 10 and 90% of AP amplitude during the rise and decay of the AP , respectively . Hyperpolarizing 1 s current pulses were injected to elicit a hyperpolarization-induced sag with an average voltage value of −85 mV at the end of the hyperpolarizing current step in each cell ( Amendola et al . , 2012 ) . To measure the passive membrane properties ( τm , Rin , Cm ) , 0 . 5–1 s negative current pulses inducing small hyperpolarizations ( <10 mV ) were used , ensuring that no hyperpolarization-activated voltage-gated current ( IH ) contaminated the voltage signal . Incremental 1 s depolarizing pulses were injected to elicit trains of APs , and the frequency vs current curve was generated for each cell . Since SNc dopaminergic neurons display adaptation or facilitation of firing frequency during sustained current pulses ( Vandecasteele et al . , 2011 ) , we determined the average gain value ( Hz/100 pA ) both for the first three and the last three APs generated during the pulse ( see Figure 7 ) . Therefore , gain at the start ( GS ) and gain at the end ( GE ) values were determined and the spike frequency adaptation index ( SFA index ) was defined as the ratio GS/GE . Dopaminergic neurons are unable to sustain firing for strong current injections , a phenomenon known as depolarization block ( Tucker et al . , 2012 ) . This results in saturating frequencies of firing when injecting increasing depolarizing pulses of current . Our analysis of gain was focused on the ‘non-saturating’ firing behavior of dopaminergic neurons , that is , on the responses to current pulses inducing sustained firing of dopaminergic neurons . Thus , gain values could be extracted by linear regression of the frequency/current curve ( see Figure 7 ) . Dynamic-clamp experiments were performed using the SM-2 software developed by Hugh Robinson ( Robinson , 2008 ) ( Cambridge Conductance , UK ) , which runs on a scriptable digital-signal-processing ( DSP ) -based system for dynamic conductance injection . Conductance definition for the leak subtraction was compiled and downloaded from the PC to a P-25M DSP board ( Innovative Integration , Simi Valley , CA ) , which executes the conductance injection with a sampling rate of 40 KHz over a 2 V range with a resolution of 0 . 1 mV . The leak conductance was defined as having a reversal potential of −65 mV , and the absolute conductance value was adjusted in each neuron in order to reach an apparent input resistance of 1600 MOhm corresponding to the input resistance measured in P2–P3 neurons . Data were acquired with a HEKA EPC 10/USB patch-clamp amplifier ( HEKA Elektronik , Germany ) and patchmaster software ( HEKA Elektronik ) . Analysis was conducted using FitMaster v2x30 ( HEKA Elektronik ) and Igor Pro ( version 6 . 0 , WaveMetrics ) . Statistical analysis ( performed according to data distribution ) included: linear regression , unpaired t test , Mann Whitney , paired t test , one-way ANOVA and Fisher exact test for proportions , with a p value <0 . 05 being considered statistically significant . Statistics were performed utilizing SigmaPlot 10 . 0 ( Jandel Scientific , UK ) and Prism 6 ( GraphPad Software , Inc . , La Jolla , CA ) . Unless otherwise stated , data are presented as box and whisker plots , with the box representing the median value , the 25th and 75th percentiles , and the whiskers representing the 10th and 90th percentiles . In box and whisker plots , all outliers are represented . Agglomerative Hierarchical Clustering Analysis ( AHC ) and Principal Component Analysis ( PCA ) were performed using XLSTAT 2011 ( Addinsoft , France ) . AHC dissimilarity level was calculated based on euclidian distance , and agglomeration was performed using Ward's method . The dissimilarity threshold used for defining the number of classes of objects was automatically set by the XLSTAT software . PCA was based on the covariance matrix ( n ) of the observations/variables table . Only the PC1 and PC2 ( accounting for 93 . 76% of the total variance ) were kept , and the corresponding factor loadings ( F1 and F2 ) for each individual recording were analyzed and plotted . Experiments where the effects of TTX , apamin and leak negative conductance injection on firing were tested ( Figure 11 ) were added as supplementary data in the PCA but were not included in the calculation of the principal components . Acute slices containing Neurobiotin tracer-filled cells were fixed 30 min in 4% paraformaldehyde at 4°C and immunolabelled with anti-tyrosine hydroxylase ( Millipore , 1:9000 or Abcam , 1:2000 ) and Streptavidin Alexa Fluor 594 ( Invitrogen; 1:12 , 000; 1 . 66 ng/ml ) and donkey anti-sheep Alexa Fluor 488 ( Invitrogen; 1:1000 2 µg/ml ) as previously described ( Wolfart et al . , 2001; Amendola et al . , 2012 ) . All immunolabelling was viewed on a Leica TCS-SP2 confocal microscope ( Leica Microsystems , Wetzlar , Germany ) , and images were captured using Leica LAS-AF software . Figures were prepared using Sigma Plot , GraphPad Prism 5 , Igor Pro , Adobe Photoshop and Adobe Illustrator ( CS4 , Adobe Systems Inc . , San Jose , CA , U . S . A . ) and ImageJ ( MacBiophotonics , McMaster University , ON , Canada ) , with brightness and contrast adjustments performed consistently across the image to enhance clarity . | The brain contains hundreds of types of neurons , which differ in size and shape , and also in the chemicals that they use to communicate with each other . However , one thing all neurons have in common is that they all carry electrical signals that depend on the flow of ions through specialized channels in the membranes that surround each neuron . Nevertheless , the number and identity of these channels also vary markedly from one type of neuron to the next . This biophysical diversity underlies a variety of complex patterns of electrical activity observed in different types of neurons . This complexity often makes it difficult to pinpoint which of the myriad features of a neuron are most important for determining its function , and which ones are most affected by processes such as aging or disease . To address this problem , Dufour et al . have devised an approach for characterizing the electrical properties of neurons in a systematic manner , in order to generate a ‘phenotypic’ profile for individual types of neurons . Neurons from a region of the brain called the substantia nigra pars compacta were chosen for the study . These neurons are known for the fact that they are among the first to degenerate in Parkinson's disease . Using electrodes to record from slices of rat brain , a total of 16 electrical properties were measured for each neuron , including how often each cell ‘fired’ and how variable its firing pattern was . The experiments were repeated using brain slices from rats aged between 2 and 29 days , and the data from more than 300 neurons were then analyzed using statistical techniques designed to identify groups of features that change together over time . The analysis revealed that the cells could be grouped into three developmental stages , separated by two transitions: one occurring around days 3–5 and another around days 9–11 . This was confirmed by experiments showing that cells could be made to revert back to an earlier stage by applying chemicals and electrical currents to reverse the changes that had occurred during development . While the current study has provided insights into the postnatal development of one particular class of neurons in the substantia nigra , the approach could in principle be applied to any type of neuron . The detailed profile that is obtained will make it easier to identify subtle changes in neurons in response to development , aging , or disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2014 | Non-linear developmental trajectory of electrical phenotype in rat substantia nigra pars compacta dopaminergic neurons |
The timely transition from neural progenitor to post-mitotic neuron requires down-regulation and loss of the neuronal transcriptional repressor , REST . Here , we have used mice containing a gene trap in the Rest gene , eliminating transcription from all coding exons , to remove REST prematurely from neural progenitors . We find that catastrophic DNA damage occurs during S-phase of the cell cycle , with long-term consequences including abnormal chromosome separation , apoptosis , and smaller brains . Persistent effects are evident by latent appearance of proneural glioblastoma in adult mice deleted additionally for the tumor suppressor p53 protein ( p53 ) . A previous line of mice deleted for REST in progenitors by conventional gene targeting does not exhibit these phenotypes , likely due to a remaining C-terminal peptide that still binds chromatin and recruits co-repressors . Our results suggest that REST-mediated chromatin remodeling is required in neural progenitors for proper S-phase dynamics , as part of its well-established role in repressing neuronal genes until terminal differentiation .
The transcriptional repressor REST ( also called NRSF; Schoenherr and Anderson , 1995 ) binds to thousands of coding and non-coding genes that , as an ensemble , are required for the terminally differentiated neuronal phenotype ( Bruce et al . , 2004; Conaco et al . , 2006; Otto et al . , 2007; Mortazavi et al . , 2006 ) . In situ hybridization analysis shows a striking reciprocal pattern of REST expression to genes expressed within the developing nervous system , consistent with its role as a repressor ( Schoenherr and Anderson , 1995; Chong et al . , 1995 ) . In differentiating cells in culture , REST is down-regulated during the transition to a mature neuron ( Ballas et al . , 2005 ) and overexpression of REST by in utero electroporation leads to delay in neuronal maturation ( Mandel et al . , 2011 ) . These results suggest that REST serves as a timer of terminal neuronal differentiation . Counter to this model , in vivo loss-of-function analyses in mice have not shown evidence for precocious differentiation or de-repression of genes responsible for the mature neuronal phenotype in the embryonic nervous system . A global germline knockout ( KO ) of REST is early embryonic lethal , but no obvious morphological defects in neural tube formation were noted in that study ( Chen et al . , 1998 ) . Later , in embryogenesis , brain-specific loss of REST , by conventional Cre lox technology , also lacks an obvious nervous system phenotype ( Aoki et al . , 2012 ) , while conditional loss of REST from adult neural progenitors shows only transient and subtle precocious neuronal differentiation ( Gao et al . , 2011 ) . Despite these results , a recent study points to a role for REST in human neurogenesis and microcephaly through regulation of REST by a factor , ZNF335 , mutated in patients with a severe form of microcephaly ( Yang et al . , 2012 ) . Additionally , microcephaly results from dysregulation of a REST/BAF170/Pax6 repressor complex during neurogenesis ( Tuoc et al . , 2013 ) . Thus , the role of REST in embryonic neurogenesis remains an open question . To re-examine the role of REST during embryonic neurogenesis , we use mice containing a conditional gene trap ( GT ) cassette in an intron of the endogenous Rest gene that terminates transcription upstream from the initiator codon . Using this line , we generate mice with a REST deficiency in nestin-positive neural progenitors , prior to the time when REST is dismissed normally from chromatin during neurogenesis . We examined these mice for embryonic and adult brain phenotypes and found DNA damage , apoptosis , and smaller brain size as prominent defects . The DNA damage persisted and caused glioblastoma ( GBM ) in mice also lacking the tumor suppressor , p53 . We also characterized REST binding properties and embryonic phenotypes in a conventional brain-specific Rest KO line ( Gao et al . , 2011 ) , targeting Rest exon 2 , which we show still expresses a C-terminal REST peptide , for comparison with our Rest GT mice . Our results indicate that REST is required to protect genomic integrity , supervised by S phase surveillance , and that this function is key for regulating proper timing of terminal neuronal differentiation .
We exploited a mouse line carrying a GT in the Rest intron ( RestGT ) between non-coding exon 1a–c and the first coding exon , exon 2 ( Figure 1A ) . The GT cassette contains a splice acceptor site upstream of a promoter-less β-galactosidase and neomycin gene fusion ( β-geo ) and a polyadenylation sequence ( Schnutgen et al . , 2005 ) . Thus , β-geo expression in this line is under the control of endogenous Rest regulatory elements . The GT was confirmed in the Rest locus by Southern blot analysis ( Figure 1B ) and we verified the insertion of a single GT in the genome by additional Southern blot and DNA sequence analysis ( data not shown ) . RestGT/GT mice , like Rest KO mice generated by germ-line deletions of Rest exons 2 and 4 ( Chen et al . , 1998; Aoki et al . , 2012 ) , are growth-arrested ( Figure 1C ) and die between E9 . 5 and 11 . 5 , validating the RestGTallele as a model for loss-of-function . 10 . 7554/eLife . 09584 . 003Figure 1 . A GT in the Rest gene ( RestGT ) causes REST deficiency and embryonic lethality . ( A ) Schematic showing the GT in the first Rest intron . Red and green boxes indicate alternative 5’ untranslated exons ( 1a–c ) and first coding exon ( 2 ) , respectively . The GT cassette contains an SA site , a reporter gene encoding a β-galactosidase neomycin fusion gene ( β-geo ) , and a pA sequence . Arrowheads depict target sites for Flpe and Cre recombinases . Dashed line indicates probe location for Southern blot in B . ( B ) Southern blot of genomic DNA from indicated genotypes . ( C ) E10 . 5 wild type ( Rest+/+ ) and mutant ( RestGT/GT ) embryos . ( D ) Western blot analysis of REST protein in E10 . 5 embryos . Ecad ( E-cadherin ) , loading control . ( E ) qRT-PCR analyses of Rest mRNA levels , normalized to 18S RNA , in E9 . 5 embryos , n=6 mice/genotype . Means and SD are shown . ( F ) qRT-PCR analysis , normalized to 18S RNA , for Rest targets , n=3 mice/genotype . Means and SD are shown . Syp and Snap25 values in Rest+/+ are 4 . 3×10-4 ± 2 . 0×10−4 and 4 . 1×10−4 ± 2 . 0×10−4 , correspondingly . Statistical significance was determined by ANOVA with Tukey posthoc ( E ) and by unpaired t-test with Welch correction . ( F ) ( G ) Whole mount X-gal staining of E11 . 5 embryo . ( H ) Left , in situ hybridization analysis for Rest transcripts in E12 . 5 embryo . Arrowhead indicates region magnified in adjacent image . Counterstain ( pink ) is nuclear fast red . Right panel , Immuno-labeling showing location of TuJ1+ neurons used to determine PP and VZ boundaries . ( I ) Immuno-labeling of cortical section from E13 . 5 embryo using indicated antibodies and DAPI stain for nuclei . * , p<0 . 05 , ** , p<0 . 01 , *** , p<0 . 001 . ANOVA , analysis of variance; GT , gene trap; mRNA , messenger RNA; PP , preplate; qRT-PCR , quantitative real-time polymerase chain reaction; SA splice acceptor; SD , standard deviations; VZ , ventricular zone . DOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 003 In RestGT/GT mice , we also observed loss of both REST protein ( Figure 1D ) and messenger RNA ( mRNA; Figure 1E ) ( Rest+/+ , 0 . 95 , standard deviation [SD] , 0 . 27; RestGT/+ , 0 . 47 , SD 0 . 18; RestGT/GT , 0 . 7×10−2 , SD 0 . 003 ) . This was accompanied by the predicted reciprocal up-regulation of select REST target genes ( Figure 1F ) . The β-gal activity programmed from the GT correlated tightly with the pattern of endogenous Rest mRNA . For example , β-gal activity ( Figure 1G ) and endogenous Rest expression ( Figure 1H , left panel ) were both detected in non-neural tissues outside the developing nervous system . In the embryonic brain , endogenous Rest mRNA was confined largely to neural progenitors in the ventricular zone ( VZ ) and mostly absent from preplate cells that were populated with TuJ1+ neurons ( Figure 1H ) . Correspondingly , REST protein expression was confined predominantly to SOX2+ neural progenitors ( Figure 1I ) . We were not able to detect REST protein in the subventricular zone ( SVZ ) occupied by the more committed TBR2+ basal progenitors ( not shown ) , indicating that the down-regulation of REST occurred most robustly prior to the generation of mature neurons and the transition to basal progenitors . Early embryonic lethality , coincident with the onset of neurogenesis , precluded analysis of REST function in RestGT/GT neural progenitors . Therefore , we used a two-step breeding scheme to remove REST specifically from neural progenitors . In the first step , RestGT mice ( Figure 1A ) were crossed to mice expressing the Flpe recombinase transgene ( Dymecki et al . , 2000 ) . This resulted in inversion of the GT cassette ( GTinv ) to restore normal splicing of Rest exons 1a–c to exon 2 ( Figure 2A , top ) . In the second step , RestGTi/+ mice , heterozygous for the inverted allele , were bred to mice expressing a Nestin Cre recombinase transgene . This resulted in re-inversion of the GTinv cassette to create a mutant Rest allele in which exons 1a–c were spliced to the β-geo gene instead of exon 2 ( Figure 2A , bottom ) , terminating transcription upstream of remaining Rest sequences . All genotypes were confirmed by DNA sequence analysis . 10 . 7554/eLife . 09584 . 004Figure 2 . A conditional RestGT allele results in REST-deficiency in Nestin+ progenitors . ( A ) KO strategy . Top , mice bearing an inverted GT cassette ( RestGTinv ) , resulting in normal splicing , were generated by mating RestGT mice ( Figure 1A ) to mice containing the Flpe transgene . Bottom , conditional mutants , resulting in splicing of exon 1 to the GT cassette ( Cre+ , RestGTi ) , were generated by mating RestGTi mice ( top ) to mice bearing the Nestin Cre transgene . It is noteworthy that β-geo expression was still under the control of Rest regulatory elements . ( B ) Whole-mount X-gal staining of E11 . 5 RestGTi/+ ( left ) and Cre+ , RestGTi/+ ( right ) embryos . ( C ) X-gal staining of a coronal section of E13 . 5 cortex from Cre+ , RestGTi/+ mice . ( D ) Western blot of nuclear extracts from neuroepithelia of E13 . 5 mice . HDAC , histone deacetylase 2 , loading control . ( E ) qRT-PCR analysis of Rest transcripts , normalized to 18S RNA , in E13 . 5 brain ( n=6 mice/genotype ) and NPCs grown as neurospheres for 5 days ( n=3 mice/genotype ) . ( F ) Quantitative chromatin immunoprecipitation analysis of REST enrichment at RE1 sites in the glycine receptor ( Glra; +275 bp from TSS ) and Snap25 genes ( +867 bp from TSS ) ( n=4 mice/genotype ) . Amplicons were designed within 100 bp of RE1 binding sites . Snap25 CDS and Myf5 transcriptional start site lack RE1 sites . Means and SD are shown in ( E ) and ( F ) . Statistical significance was determined by Mann–Whitney test in E and by unpaired t-test with Welch correction in F . ( G ) Immuno-labeling of E13 . 5 telencephalon using indicated antibodies . Boxes indicate regions of higher magnification in images at right . White arrowheads indicate DAPI+ cells that express both proteins . * , p<0 . 05 , ** , p<0 . 01 , *** , p<0 . 001 . CDS , coding sequence; CP , cortical plate; GT , gene trap; KO , knockout; MZ , marginal zone; NPC , neural progenitor cells; qRT-PCR , quantitative real-time polymerase chain reaction; SD , standard deviation; VZ , Ventricular zone . DOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 004 We examined β-gal activity in the developing brains of Cre+ , RestGTi /GTi and RestGTi/GTi ( controls hereafter ) mice . Consistent with the global GT , β-gal activity matched expression of the endogenous Rest gene . Specifically , β-gal+ cells in the Cre+ , RestGTi/GTi mice were confined to neurogenic areas of E11 . 5 embryos ( Figure 2B ) . Within the cortical VZ at E13 . 5 , β-gal activity ( Figure 2C ) and endogenous Rest mRNA ( data not shown ) were detected primarily in the apical region , which is occupied by progenitors ( Götz and Huttner , 2005 ) . Interestingly , we also detected β-gal+ cells in some presumably mature cells in the marginal zone ( MZ ) of the cortical plate ( CP ) indicating Rest promoter activity in a subset of neurons ( Figure 2C ) . Ninety-five percent of the Cre+ , RestGTi /GTi mice survived into adulthood , but had significantly reduced REST protein and mRNA levels in neural progenitors and E13 . 5 brains compared with controls ( Figure 2D and E ) . Chromatin immunoprecipitation ( ChIP ) analysis of E13 . 5 brains from Cre+ , RestGTi /GTi mice indicated ~four-fold reduction in REST occupancy at consensus RE1 sites within 1kb of the Glycine receptor and Snap25 transcriptional start sites , in the first exon and intron , respectively ( Glycine receptor: RestGTi /GTi , 0 . 64 , SD 0 . 13; Cre+ , RestGTi /GTi , 0 . 16 , SD 0 . 07 ) , Snap 25: RestGTi /GTi , 0 . 32 , SD 0 . 04; Cre+ , RestGTi /GT , 0 . 07 , SD 0 . 04 ) ( Figure 2F ) . There was no significant change in REST occupancy in the Snap25 coding sequence or the myf5 promoter region that lacked RE1 binding sites ( Figure 2F ) . The very low levels of glycine receptor mRNA in control mice was not within the linear range of detection . Levels of this mRNA in the mutant , normalized to 18S RNA , were consistently within the linear range , but still very low ( 0 . 13 , SD 0 . 08 ) . For SNAP25 , we observed an ~2-fold increase in mutant mRNA compared with control levels ( RestGTi/GTi: 0 . 09 , SD 0 . 06; Cre+ , RestGTi/GTi: 0 . 2 , SD 0 . 05; p <0 . 05 , unpaired t-test , n=7–8 , E13 . 5 brain ) . It is possible that the low mRNA levels for both genes reflect the absence of transcriptional activators at this time . In control mice , SOX2 progenitors of the VZ stained positively for REST , but the majority of TuJ1+ neurons were not immuno-positive ( Figure 1I and 2G ) , consistent with the known down-regulation of REST during neurogenesis ( Ballas et al . , 2005 ) . Interestingly , however , we could detect REST protein in a small number of TuJ1+ neurons in the outmost MZ of the CP ( Figure 2G , right images ) , corroborating β-gal staining ( Figure 2C ) , suggesting either that REST is re-expressed at later stages of differentiation or that REST expression has not yet been extinguished completely in these neurons . Whether REST is bound to the RE1 sequence in the TuJ1 gene at this stage cannot be determined given the small number of labeled neurons . Because a previous study indicated that loss of REST was associated with microcephaly ( Yang et al . , 2012 ) , we asked whether this was also true for our RestGT mice . Indeed , significantly smaller brains were evident at birth ( data not shown ) and in postnatal ( P45 ) Cre+ , RestGTi /GTi mice when compared with control mice of two different control genotypes , RestGTi/GTi and Nestin Cre ( Figure 3A and B left panel ) . The brain size in the mutant was reduced to 71% of the brain size of RestGTi/GTi mice ( RestGTi/GTi , 0 . 46 , 95% confidence interval [CI] 0 . 45–0 . 46; Cre+ , RestGTi/GTi 0 . 33 , CI 0 . 32–0 . 34 ) , similar to the difference from Nestin Cre mice ( Nestin Cre+ , Rest+/+: 0 . 43 , CI 0 . 42 to 0 . 44 ) . A third potential control line is Cre+ , RestGTi/+ mice . However , we measured a significant brain size reduction Cre+ , RestGTi/+ mice ( 0 . 36 , CI 0 . 35–0 . 37 , p value <0 . 001 ) compared to RestGTi/GTi ( reduction to 78% ) or Cre+ , Rest+/+ ( reduction to 83% ) mice , likely due to the combination of the slightly reduced brain size of the Cre recombinase background and reduced REST levels ( Figure 3B and 1E ) . The hypomorphic effect is consistent with previous studies , indicating that gene expression levels are sensitive to small changes in REST levels ( Ballas et al . , 2005; Ballas and Mandel , 2005 ) , so the RestGTi/GTi mice were used as controls in the remaining experiments . 10 . 7554/eLife . 09584 . 005Figure 3 . Reduced brain size , thinner cortex , and reduced numbers of upper layer neurons in Nestin Cre+ , RestGTi/GTi mice . ( A ) Representative P45 brains from control and Cre+ , RestGTi/GTi littermates . ( B ) Comparison of brain mass in P45 mice deleted for REST using Nestin ( left panel ) and hGFAP ( right panel ) promoter driven Cre recombinases . The numbers of mice analyzed for each genotype are shown . Statistical significance determined by ANOVA with Tukey posthoc ( left ) and Kruskal–Wallis ANOVA test with Dunn posthoc ( right ) . ( C ) Nissl-stained P45 coronal brain sections . Boxes denote higher magnification views in right hand panels . * , corpus collosum . Scale bar for right panel , 0 . 5mm . ( D ) Measurements of cortical thickness in P45 brain ( n=8–11 mice/genotype ) . Statistical significance determined by ANOVA with Tukey posthoc . ( E ) and ( F ) Representative immuno-labeling of P1 cortical sections with indicated antibodies revealing cortical layering in control and Cre+ , RestGTi/GTi mice . The presence of both low and high expressing CTIP2 cells in layers 6 and 5 , respectively , is noteworthy . ( G ) Quantification of ( E ) and ( F ) numbers over brackets denote cortical layers where the cells were counted . Cells were counted in 400 μm of cortical thickness ( n=8–10 mice/genotype ) . Note that only 200 μm images are shown in E , F . Low ( layer 6 ) and high ( layer 5 ) CTIP2-expressing cells were used to differentiate between Tbr1+ , CTIP2low+ cells of layer 5 and CTIPhigh+ cells in layer 5 . SATB2+ CTIP2- cells were counted above layer of CTIPhigh+ cells . ( H ) Mean cell densities from three areas ( 100×150 μm2 ) in each cortical section , n=5–8 mice/genotype . Statistical significance was determined by Mann–Whitney t-test for ( G ) and ( H ) . Means and 95% CI are shown in B , D , F and G . * , p<0 . 05; ** , p<0 . 01; *** , p<0 . 001 . ANOVA , analysis of variance; ns , non-significant . DOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 005 We also tested inversion of the RestGTi allele using mice expressing an hGFAP Cre recombinase transgene active at mid-neurogenesis , slightly later than Nestin Cre ( Zhuo et al . , 2001 ) . In these mice , there was a smaller reduction in brain mass to only 92% of control levels ( Figure 3B right panel ) , suggesting a critical temporal window for REST function at early-to-mid neurogenic stages . This matches in utero electroporation experiments revealing an enhanced migration phenotype of REST knockdown performed at E13 . 5 that is not observed at E14 . 5 ( Yang et al . , 2012; Fuentes et al . , 2012 ) . The reduction in brain mass in REST mutant mice correlated with reduced cortical thickness , both rostrally and caudally , and reduced corpus callosum thickness ( Figure 3C and D ) . To determine whether these phenotypes were associated with an imbalance of temporally distinct progenitor types , we counted neurons in the six cortical layers that are born at different times during neocortical development , with deep layer neurons preceding the birth of upper layer neurons . To this end , we performed dual immunostaining for transcription factor markers specific for adjacent layers ( Alcamo et al . , 2008; Britanova et al . , 2008; Arlotta et al . , 2005; Bedogni et al . , 2010 ) . There was a reduction to 56% of control numbers of SATB2+/CTIP2- upper layer 2–4 neurons ( RestGTi/GTi , 769 . 3 , CI 733 . 2 – 805 . 4; Cre+ , RestGTi/GTi: 428 . 4 , CI 397 . 2–459 . 5 ) , but no statistically significant differences between mutant and control in layer 5 and 6 neurons that were born earlier in neurogenesis ( Figure 3E , F and G ) . However , CTIP2 immuno-labeling , which molecularly defines layer 5 , showed expansion into layer 6 and decreased density in mutant brain ( Figure 3F and H ) , pointing to some disorganization due to the premature loss of REST . Despite this finding , the predominant feature of loss of REST during neurogenesis was significantly fewer postnatal neurons in the upper cortical layers born during mid-to-late neurogenesis . The small brain size and diminished numbers of neurons at birth could reflect increased depletion of progenitor cells and/or cell death . To address the former possibility , we first distinguished apical and basal progenitors by immuno-labeling with antibodies to PAX6 and TBR2 , respectively ( Figure 4A ) . Numbers of PAX6+TBR2− apical progenitors in Cre+ RestGTi/GTi mice were reduced to ~71% of control values ( RestGTi/GTi , 273 . 1 , CI 203 . 4–342 . 9; Cre+ , RestGTi/GTi: 123 . 7 , CI 110 . 8 . 2–136 . 5; Figure 4B ) . There were also reduced numbers of TBR2+ basal progenitors , which are progeny of the apical PAX6+ progenitors , at E13 . 5 ( RestGTi /GTi , 116 . 0 , CI 100 . 2–131 . 8; Cre+ , RestGTi/GTi: 81 . 9 , CI 62 . 2–101 . 5 ) ( Figure 4B ) . 10 . 7554/eLife . 09584 . 006Figure 4 . Depletion of apical progenitors and premature cell cycle exit in Cre+ , RestGTi/GTi mice . ( A ) Representative immuno-labeling distinguishing apical ( PAX6+/TBR2− ) and basal ( TBR2+ ) progenitors in E13 . 5 cortices . Area outlined by dotted lines is shown in images at right . ( B ) Quantification of A . Measurements from 100 μm cortical width extending from the VZ to pial surface , n=9 mice /genotype . Statistical significance was determined by Mann–Whitney t-test . Red triangles show quantification of the images shown in A . ( C ) Representative immuno-labeling for Ki67 and BrdU in the cortical section of E14 . 5 mouse brain pulsed with BrdU for 24 h . The area outlined by dotted lines is magnified in the adjacent images . ( D ) Top , histogram showing percent stained cells relative to the number of DAPI+ nuclei . Bottom , fraction of cells exiting the cell cycle defined as fraction of Ki67-BrdU+ cells in total BrdU+ cell population . Measurements were made in 200 μm cortical width areas extending from ventricular to pial surface , n=7 mice/genotype . Statistical significance was determined by Mann–Whitney t-test . Means and 95% CI are shown in B and D . * , p<0 . 05; ** , p<0 . 01; ***p<0 . 001 . IZ , intermediate zone; SVZ , subventricular zone . DOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 006 To determine whether the depletion of apical progenitors correlated with premature cell cycle exit , we pulsed E13 . 5 embryos with BrdU to label cells that were undergoing DNA synthesis . After 24 hr , we immunolabeled for Ki67 , a marker of cycling cells , and incorporated BrdU , and quantified the results relative to the number of DAPI+ nuclei . The Ki67 staining in Cre+ , RestGTi/GTi cells was reduced to 69% of the control values ( RestGTi/GTi , 72% , CI 70–73%; Cre+ , RestGTi/GTi , 50% , CI 38–40% ) , with no change in the percentage of BrdU+ cells ( RestGTi/GTi , 42% , CI 37–48%; Cre+ , RestGTi/GTi , 43% , CI 38–49% ) ( Figure 4C and D , top panel ) . To determine the fraction of cells that exited the cell cycle in a 24 hr period , we quantified the proportion of Ki67- cells in the total BrdU+ cell population . Our results indicated that between E13 . 5 and 14 . 5 , the progenitor pools were increasingly depleted , as ~50% more progenitors exited the cell cycle in Cre+ , RestGTi/GTi compared with controls ( RestGTi/GTi , 21% , CI 20–23%; Cre+ , RestGTi/GTi , 32% , CI 26–38% ) ( Figure 4D , lower panel ) . This indicates a decreasing progenitor pool available to generate the late born upper layer neurons ( Figure 3E and G ) . Microarray analysis of E12 . 5 brain from Cre+ , RestGTi/GTi and control mice did not show significant up-regulation ( >1 . 3 fold ) in canonical REST neuronal target genes ( Supplementary file 1 ) . However , our analysis did reveal significant up-regulation of microglial signature genes ( Hickman et al . , 2013; and several p53-regulated pro-apoptotic genes ( Ko and Prives , 1996; Levine , 1997; Budanov and Karin , 2008 ) in brain tissue and LeX+ sorted progenitors ( Supplementary file 1 and Figure 5A ) . To test for cell death , we stained E13 . 5 brain sections from the cortex and lateral ganglionic eminence ( LGE ) with antibody against the activated form of cleaved Caspase3 ( ClCasp3 ) , a member of the cysteine–aspartic acid proteases family , which is a critical mediator of apoptosis and required for chromatin condensation and DNA fragmentation ( Janicke , 1998 ) . Unlike in control cortices , apoptosis was prominent in cells in Cre+ , RestGTi/GTi mice , particularly at the border between the VZ/SVZ and CP ( Figure 5B ) . Interestingly , although cells in the VZ/SVZ border area are densely populated with TBR2+ basal progenitors , the apoptotic cells were not positive for TBR2+ ( Figure 5—figure supplement 1A ) . 10 . 7554/eLife . 09584 . 007Figure 5 . Progenitors and neurons in cortex of Cre+ , RestGTi/GTi mice undergo apoptosis that is rescued by deletion of p53 . ( A ) qRT-PCR analyses normalized to 18S RNA , of p53 pro-apoptotic target mRNAs ( p21 , Sestrin2 and CyclinG ) , progenitor mRNAs ( Sox2 ) and neuronal mRNAs ( Tbr1 ) in LeX+ purified progenitors isolated from E13 . 5 brain . RestGTi/GTi , n=4 mice , Cre+ , RestGTi/GTi , n=6 mice . ( B ) Representative immuno-labeling for apoptotic cells with clCasp3 and neuronal marker TuJ1 in coronal telencephalic sections from E13 . 5 mice . ( C ) Temporal profiles of apoptosis in cortex and LGE of Cre+ , RestGTi/GTi mice and cortex of RestGTi/GTi mice , in areas of 100 μm ventricular width extending from VZ to the pial surface ( n=6–9 mice/time point ) . Means and SDs are shown . ( D ) Percentage of MAP2+ and SOX2+ cells in all apoptotic cells in E13 . 5 Cortex of Cre+ , RestGTi/GTi mice in 100μm ventricular width . n=5 mice . ( E ) Quantification of apoptosis in E13 . 5 cortex and LGE ( n=5–9 mice/genotype/100μmVZ ) . Note: The same data for Cre+ , RestGTi/GTi cortex and LGE from 5C is re-plotted for comparison . ( F ) qRT-PCR analysis of mRNA levels , relative to 18S RNA , of p53 pro-apoptotic ( p21 , CyclinG , Sestrin2 , Perp ) , non-apoptotic ( Btg2 ) and non p53 ( Snap25 ) targets in E12 . 5 brain ( n=7–8 mice/genotype ) . ( G ) Measurements of brain mass in P45 mice . Numbers of mice are indicated in the histogram . Note: the same data from Cre+ , RestGTi/GTiand RestGTi/GTiin Figure 3B is re-plotted for comparison . Means and 95% CI are shown in A , D , E , F , G . Statistical significance was determined by Mann–Whitney t-test ( A ) , ANOVA test with Tukey posthoc ( E and G ) and Kruskal–Wallis ANOVA test with Dunn posthoc ( F ) . * , p<0 . 05 , ** , p<0 . 01 , *** , p<0 . 001 , ns , non-significant . ANOVA , analysis of variance; clCasp3 , cleaved caspase3; CI , confidence interval; LGE , lateral ganglionic eminence; mRNA , messenger RNA; qRT-PCR , quantitative real-time polymerase chain reaction; SD , standard deviationDOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 00710 . 7554/eLife . 09584 . 008Figure 5—figure supplement 1 . Apoptotic cells are TBR2− and primarily post-mitotic , MAP2+ cells . ( A ) and ( B ) Representative E13 . 5 cortical brain sections from Cre+ , RestGTi/GTi mice stained with the indicated antibodies . Insets on left are magnified in right hand images . ( A ) Filled arrow points to Sox2+ TBR2+ ClCasp3− cell and empty arrow points to Sox2− TBR2− ClCasp3+ cells . ( B ) Arrows point to MAP2+ClCasp3+Ki67− cells . DOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 00810 . 7554/eLife . 09584 . 009Figure 5—figure supplement 2 . Volcano plot of microarray analysis comparing transcriptome of E12 . 5 brains of control , RestGTi/GTi , Trp53fl/fl , and Cre+ , RestGTi/GTi , Trp53fl/fl mice ( Supplementary file 2 ) . The log2 transformed fold changes in gene expression on the x-axis are plotted against negative log10-transformed p-values of the t- test on the y-axis . Fold changes are determined for each probe as an average ratio of normalized value from Cre+ , RestGTi/GTi , Trp53fl/fl mice relative to normalized value in control , RestGTi/GTi , Trp53fl/fl mice . n=4 embryos/genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 00910 . 7554/eLife . 09584 . 010Figure 5—figure supplement 3 . qPCR analyses of DNA in the Rest GT genomic locus normalized to Gapdh DNA . A RestGT primer set amplifies β-geo sequences within the GT cassette ( Rest GT ) , while a Rest endogenous primer pair amplifies DNA sequences located 1760 bp upstream from the GT insertion site . n= 6 ( Cre+ , RestGTi/GTi , Trp53fl/fl ) , n=4 ( RestGTi/GTi , Trp53fl/fl ) . Means and 95% CI are shown . Statistical significance measured by Mann–Whitney t-test . CI , confidence interval; GT , gene trap; ns , non-significant; qPCR , quantative polymerase chain reaction . DOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 010 Cell death peaked at E14 . 5 in the cortex , the time of mid-neurogenesis for this brain region and the beginning of upper layer neuronal specification ( Caviness et al . , 2009 ) , and then declined to control levels by E15 . 5 ( Figure 5C ) . The peak of apoptosis was earlier in LGE , peaking at E13 . 5 ( Figure 5C ) . The number of clCasp3+ cells in RestGTi/GTi LGE was similar to the low numbers in the cortex ( not shown ) . The distinct peak times of apoptosis for the cortex and LGE , coincident with their staggered time courses of differentiation ( Batista‐Brito and Fishell , 2009; Greig et al . , 2013; Colasante and Sessa , 2010 ) indicate that apoptosis was linked to the timing of premature cell cycle exit of distinct progenitor populations in these brain regions , and not to a nonspecific effect on all progenitors or effects on earlier-stage , less-committed neural progenitors . The majority of clCasp3+ cells were also positive for MAP2 and negative for Ki67 ( Figure 5D and Figure 5—figure supplement 1B ) . Because MAP2 is not a direct REST target gene , this result suggests progression of a mature neuronal program , but the tight correlation between MAP2 expression and apoptosis prevents knowledge of the sequence of events . To confirm that the apoptosis was due to activation of pro-apoptotic p53 pathways indicated by the microarray analysis , we generated compound conditional mutant mice that lacked both the Rest and Trp53 genes ( Jonkers et al . , 2001 ) in Nestin Cre+ cells and their progeny ( Cre+ , RestGTi/GTi , Trp53fl/fl ) . At E13 . 5 , numbers of apopotic cells/100μm VZ width in the compound mutant were rescued to control levels ( RestGTi/GTi ) , in both the cortex ( Figure 5E ) ( RestGTi/GTi , 0 . 7 , CI 0–0 . 4; Cre+ , RestGTi/GTi , Trp53fl/fl: 2 . 9 , CI 1 . 5–4 . 4; Cre+ , RestGTi/GTi , 49 . 1 , CI 39 . 5 -–58 . 6 ) and the LGE ( RestGTi/GTi: 0 . 07 , CI 0 . 0–0 . 3; Cre+ , RestGTi/GTi , Trp53fl/fl , 0 . 53 , CI 0 . 29 -–0 . 78; Cre+ , RestGTi/GTl , 128 . 7 , CI 107 . 8–149 . 5 ) . The mRNA levels of elevated p53 pro-apoptotic transcriptional targets were also reduced to control levels ( Figure 5F ) . Brain size was restored to only 78% of control values ( Figure 5G ) , suggesting that progenitor pool depletion by both premature cell cycle exit and apoptosis contribute to this phenotype . It was possible that microarray analysis showing lack of induction of neuronal genes in Cre+ , RestGTi/GTi mice ( Supplementary file 1 ) was due to the elimination of many cells by apoptosis . Even after eliminating apoptosis , however , microarray analysis still did not show significant up-regulation ( >1 . 5-fold over control ) of REST-regulated mRNAs ( Supplementary file 2 and Figure 5—figure supplement 2 ) , suggesting that premature differentiation per se was not the sole underlying cause of the apoptosis . GT alleles carrying β-geo cassettes have been used extensively for both global and conditional gene inactivation in mice , with no reports of β-geo toxicity or chromosomal damage during recombination of the GT allele ( Budanov and Karin , 2008; Gossler et al . , 1989; Skarnes et al . , 1992; Zambrowicz et al . , 2003; Krechowec et al . , 2012; Mao et al . , 1999; Petkau et al . , 2012; Theroux et al . , 2007; Peralta et al . , 2014; Ishizawa et al . , 2011 ) . Similarly , gene recombination defects using Cre recombination technology to stochastically invert tandem copies of a reporter gene have also not been reported ( Livet et al . , 2007 ) . Nevertheless , to rule out this possibility for the Rest GT allele , we compared relative amounts of Rest genomic and GT cassette DNA isolated from E13 . 5 brains of RestGTi/GTi , Trp53 fl/fl and Cre+ , RestGTi/GTi , and Trp53 fl/fl mice . Using primers specific to these two regions for qPCR , we found no evidence for loss of DNA with inversion of the GT allele by Cre recombinase activity ( Figure 5—figure supplement 3 ) . Activation of the p53-pro-apoptotic pathway is often initiated due to genotoxic stress ( Ko and Prives , 1996; Levine , 1997; Budanov and Karin , 2008; Elledge and Zhou , 2000 ) . Therefore , we tested for DNA damage in the cortex of Cre+ , RestGTi/GTi mice by staining for phosphorylated H2AX histone ( γH2AX ) , an established marker for DNA damage ( Figure 6A ) , and the presence of DNA damage nuclear foci , revealed by co-staining of phosphorylated ataxia telangiectasia mutated ( pATM ) kinase and p53 binding protein 1 ( 53BP1 ) ( Meek and Anderson , 2009 ) ( Figure 6B ) . While we observed γH2AX+ cells and cells with DNA damage nuclear foci in the cortices of E13 . 5 Cre+ , RestGTi/GTi mice , we detected foci rarely in sections from controls . Western blot analysis verified the increased amounts of the pATM and γH2AX in the brains of E13 . 5 mutant mice ( Figure 6C ) . In addition to molecular evidence for activation of a DNA damage-signaling cascade , we also observed an increased incidence of fragmented DAPI+ micronuclei and abnormal chromosomal bridges at the apical edge of the VZ in E13 . 5 Cre+ , RestGTi/GTimice when compared with controls , indicating abnormal separation of sister chromatids and DNA breakage ( Figure 6D ) . Cells with DNA damage were eliminated by apoptosis , because loss of REST alone , in E13 . 5 cortex and LGE , resulted in fewer cells with DNA damage foci than in Cre+ , RestGTi/GTi , Trp53 fl/fl mice ( Figure 6E ) . 10 . 7554/eLife . 09584 . 011Figure 6 . DNA damage in neural progenitors is due to loss of REST and persists in progenitors with loss of REST and p53 . ( A ) Representative immuno-labeling of E13 . 5 cortex for phosphorylated histone protein H2AX ( γH2AX ) . ( B ) Representative immuno-labeling for pATM and p53 binding protein 1 ( 53BP1 ) from cortex of E14 . 5 RestGTi/GTi mice . The region between the dotted lines is shown at higher magnification in the right panels . Arrowhead indicates the nuclear foci magnified in inset . ( C ) Western blot analysis of E13 . 5 brain protein lysates . Lane 1 , RestGTi/GTi; Lane 2 , Cre+ , RestGTi/GTi . α-tub ( tubulin ) , loading control . ( D ) Left , Representative image of nuclei in cells in E13 . 5 ventricular surface . Red arrow points to abnormal bridge-like structures in mutant mice . Yellow arrow points to micronuclei . Right , quantification from E13 . 5 cortex . n=4 mice/genotype/100 mitoses . ( E ) Left , immuno-labeling of VZ cells in E13 . 5 cortex . Right , quantification from cortex and LGE dual labeled for 53BP1 and pATM , n=5 mice/genotype , 3–5 sections/mouse . ( F ) Left , schematic showing nuclear positions of neural progenitors in S phase labeled by 30 m pulse of BrdU . Right , representative images from E13 . 5 cortical section after 30 m BrdU pulse . Cells in inset are shown at higher magnification . Arrowheads indicate DNA damage foci in cells that incorporated BrdU in the S phase . ( G ) Quantification of 53BP1 data represented in F . n=6 mice , 3–4 sections/mouse . ( H ) Quantification of cells with 53BP1 foci in E13 . 5 Cre+ , RestGTi/GTi brain sections ( I ) Left , representative images of immunostained E13 . 5 cortical section . Cells in inset are shown at higher magnification . Right , quantification , n=6 mice , 3–4 sections/mouse/200μmVZ . Means and SD are shown . Statistical significance in D and G was determined by Mann-Whitney t-test and by ANOVA test with Tukey posthoc in E and H . *p<0 . 05; **p<0 . 01; ***p <0 . 001 . ANOVA , analysis of variance; pATM , phosphorylated ataxia telangiectasia mutated; SD , standard deviation; VZ , ventricular zone . DOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 011 DNA damage responses often occur in association with defects in DNA replication , culminating in cell death . To test for this association , we pulsed E13 . 5 Cre+ , RestGTi/GTi and control embryos for 30 min with BrdU , to label cells in S phase , and co-stained BrdU+ progenitors and neurons with antibodies to 53BP1 and clCasp3 for damage response foci and apoptosis , respectively ( Figure 6F–H ) . At this developmental stage , due to interkinetic nuclear migration , a majority of cells in S phase are found in the VZ close to the border of the CP ( Ueno et al . , 2006 ) ; Baye and Link , 2007 ) . There were significantly more cells with 53BP1 foci in the VZ and CP in Cre+ , RestGTi/GTi brain compared with control ( Figure 6G ) . Moreover , there were more cells with 53BP1 foci in the BrdU+ population in the mutant than in the BrdU− populations ( Figure 6H , left ) . As expected , the majority of cells with 53BP1+ foci were also positive for BrdU in brains of both genotypes ( Figure 6H , right ) . Analysis of 53BP1 foci+ cells in the clCasp3+ population in mutant mice revealed a time lag between S-phase damage and apoptosis , because only ~20% of cells labeled with clCasp3+ cells were also positive for BrdU ( Figure 6I ) . We did not perform similar counts in control cortex because of the sparse numbers of clCasp3+ cells ( 2 . 4 cells , CI 03–4 . 4/ 200μm VZ width , n=4 animals , 3–4 sections per mouse ) . Previous studies support the idea that REST is a tumor suppressor outside of the nervous system ( Wagoner et al . , 2010; Baye and Link , 2007; Wagoner et al . , 2010; Westbrook et al . , 2008; Gurrola-Diaz et al . , 2003; Kreisler et al . , 2010; Moss et al . , 2009 ) , but tumors generally reflect the loss of more than one tumor suppressor . To test whether the DNA damage due to loss of REST , in the absence of the additional tumor suppressor p53 , was persistent and would promote tumorigenesis , we maintained Cre+ , RestGTi/GTi , Trp53fl/fl mice into adulthood . We found 66% of 131 mice developed brain tumors at 9–11 months of age ( Figure 7A and B and Table 1 ) , and , of these , 48% were high-grade GBMs ( Louis et al . , 2007 ) , based on proliferation indices , presence of mitotic figures , neovascularization with proliferating endothelial cells , and presence of necrosis ( Figure 7B , Figure 7—figure supplement 1 ) . Interestingly , tumors often consisted of undifferentiated cells with small cytoplasm ( Figure 7A and B ) , characteristic of GBM with a primitive neuroectodermal-like component ( PNET ) ( Kouyialis et al . , 2005; Shingu et al . , 2005; Ohba et al . , 2008; Song et al . , 2011 ) . While common in the pediatric population , GBMs with PNET component are very rare in adults and are often thought to arise from clonal expansion of progenitor/tumor cells in vascular rich areas ( Perry et al . , 2009; Zindy et al . , 2007 ) . Anatomically , the GBMs we observed were often ( 36% ) intraventricular or periventricular ( Figure 7A and B ) , consistent with the location of progenitor/stem cells in the adult brain and REST expression in the neural stem cell ( NSC ) niche . In addition , the majority of tumors in the REST , p53-deficient mice were β-gal+ ( Figure 7A and B ) , and expressed neural progenitor/stem cell markers ( Figure 7—figure supplement 2 ) , suggesting a progenitor/NSC tumor origin and consistent with the PNET character of GBM . 10 . 7554/eLife . 09584 . 012Table 1 . Glioma incidence and grade according to Rest and Trp53 genotypesDOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 012Cre+ , Rest GTi/GTi , Trp53 +/+ Cre+ , RestGTi/GTi , Trp53fl/fl Cre+ , RestGTi/+ , Trp53fl/fl Cre+ , Rest+/+ , Trp53fl/fl Incidence0 ( 30 ) *66% ( 131 ) 53% ( 34 ) 19% ( 68 ) Grade I-–III-46%94%77%Grade IV ( GBM ) -54%6%23%* Number in parenthesis indicates number of animals used in analyses . 10 . 7554/eLife . 09584 . 013Figure 7 . Persistent DNA damage due to loss of REST , in the context of loss of p53 , promotes GBMs in adult . ( A ) and ( B ) . Left panels , X gal staining of sections from Cre+ , RestGTi/GTi Trp53fl/fl mice at 7 . 5 months ( A ) and 10 . 5 months ( B ) of age . Arrowheads point to intraventricular tumors . Black asterisks indicate secondary tumor formation with subpial spread ( B ) . Middle and right panels , H&E staining reveals abnormally increased cellular densities . Higher magnifications reveal atypical morphology of the densely packed cells , with little cytoplasm and multiple areas of new blood vessel formation ( red arrowheads ) . Red asterisks indicate pseudopallisading necrosis . ( C ) Left panel , section from 11-month-old p53-deficient mouse brain showing a rare GBM , higher magnification image on right . Red asterisk , necrotic area . ( D ) Section from 9 . 5-month-old Cre+ , RestGTi/GTi mouse brain . Boxed area is enlarged at right . Arrows point to EC and SCN . EC , ependymal cells; GBM , glioblastoma; H&E , hematoxylin and eosin; SCN , stem cell niche . DOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 01310 . 7554/eLife . 09584 . 014Figure 7—figure supplement 1 . Immunostaining analysis defining grade IV ( GBM ) gliomas according to WHO classification . ( A ) Representative immuno-labeling of a GBM in 9-month-old mouse brain from Cre+ , RestGTi/GTi , Trp53fl/fl mice , using antibodies to indicated proteins and DAPI to demarcate nuclei . Panels on the right are high magnification images in inset . Newly forming blood vessel ( asterisk ) entirely consisting of hypertrophic and hyperplastic endothelial cells ( PCNA+CD31+ ) . ( B ) Representative immuno-labeling of GBM from Cre+ , RestGTi/GTi , Trp53fl/fl 11-month-old mouse brain . PH3 is a mitotic marker . ( C ) Representative GBM section stained with H&E to reveal necrotic areas ( N ) . PCNA , proliferation marker , CD31 , endothelial cell marker . V , blood vessel . Arrows , PCNA+ cells within blood vessels . GBM , glioblastoma; PH3 , phospho-histone 3; H&E , hematoxylin and eosin . DOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 01410 . 7554/eLife . 09584 . 015Figure 7—figure supplement 2 . GBM from Cre+ , RestGTi/GTi , Trp53 fl/fl mice show high proliferation indices and expression of progenitor and NSC molecular markers . ( A–G ) Representative immuno-labeling of a GBM from Cre+ , RestGTi/GTi , Trp53fl/fl mouse using markers of astrocytes ( GFAP ) , proliferation ( Ki67 , PCNA ) , oligodendrocyte and oligodendrocyte precursors ( OLIG2 ) , neural progenitors and stem cells , ( SOX2 , RC2 , CD133 and NESTIN ) , immature neurons ( DCX , TuJ1 ) and mature neurons ( NeuN , Synaptophysin ) . Normal brain tissue adjacent to tumor is on the left of the dashed line . GBM , glioblastoma; NSC , neural stem cell . DOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 015 Similar to a previous report ( Zheng et al . , 2008 ) , gliomas in the Cre+ , Trp53fl/fl mice were much less frequent ( 13/68 ) and only three of the 13 could be characterized as GBM ( Table 1 ) . In addition , none of the tumors were intraventricular or showed spread along the hemispheres ( Figure 7C ) . No tumors formed in RestGTi/GTi mice and βgal+ cells were confined to ependymal and stem cells niche areas in SVZ ( Figure 7D ) . The lack of tumors in control mice indicated that DNA damage due to loss of REST was insufficient by itself to cause tumors and/or that most of the damaged cells were eliminated by p53-mediated apoptosis . Despite their heterogeneity , GBMs can be characterized based on their molecular expression profile into classical , mesenchymal , proneural , and neural subtypes correlated with certain mutations , chromosomal aberrations , severity and responses to therapy ( Verhaak et al . , 2010 ) . We performed qRT-PCR analysis for a subset of the signature genes on 17 Cre+ , RestGTi/GTi , Trp53fl/fltumors ( Verhaak et al . , 2010 ) ( Figure 8A ) . The analysis indicated consistently high expression of Olig2 , Erbb3 , Ng2 , Pdgfrα , and Nkx2 . 2 genes in GBM , independent of the manifestation of clinical symptoms or anatomical location of the tumor ( Figure 8A and data not shown ) . The genes represented a typical constellation of a proneural GBM subtype ( Verhaak et al . , 2010 ) and were not up-regulated in normal cortex of control mice . The Olig2 , Pdgfrα , Ebrb3 , Ng2 and Nkx2 . 2 signature was maintained in tumor cells propagated in culture from Cre+ , RestGTi/GTi , Trp53fl/flmice ( data not shown ) . The expression of these genes also correlated with an oligodendrocyte-specific molecular signature identifying proneural type of GBM ( Verhaak et al . , 2010 ) , and E13 . 5 OLIG2+ SOX2+ progenitors in brains from Cre+ , RestGTi/GTi , Trp53fl/fl mice exhibited DNA damage ( Figure 8B ) . 10 . 7554/eLife . 09584 . 016Figure 8 . RNA analysis points to proneural type tumors . ( A ) Heat map representing qRT-PCR analysis of mRNA levels , normalized to 18S , of selected genes enriched in human proneural , classical , mesenchymal , and neural subtypes of GBMs . The values are color coded and plotted on Log2 scale . ( B ) Representative immuno-labeling of LGE from E13 . 5 Cre+ , RestGTi/GTi , Trp53fl/fl mice showing 53BP1 foci in progenitor cells . Boxed area on the left image is enlarged on right images ( scale bar , 10μm ) . Arrow points to cell with foci . GBM , glioblastoma; LGE , lateral ganglionic eminence; LV , lateral ventricle; qRT-PCR , quantitative real-time polymerase chain reaction . DOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 016 Our findings on the Cre+ , RestGTi/GTi mice were not recapitulated when we tested a brain-specific Rest KO with excision of exon 2 ( Cre+ , Restfl/fl mice ) ( Gao et al . , 2011 ) . In this line , the remaining exons still have the potential to encode several zinc finger domains in the DNA-binding domain as well as the nuclear localization signal ( Shimojo , 2001 ) and C-terminal repressor domain ( Figure 9A ) ( Tapia-Ramirez et al . , 1997; Thiel et al . , 1998 ) . The average brain sizes of P45 Cre+ , Restfl/fl females ( 0 . 46 SD 0 . 02 , n=7 ) were not statistically different from the sizes of Restfl/fl littermate females ( 0 . 44 SD 0 . 02 , n=9 ) or control RestGTi/GTi mice ( 0 . 46 SD 0 . 01 , n=15 ) . A potential explanation was provided by analysis of RNA and REST protein levels in E13 . 5 brain or progenitors from the Cre+ , Restfl/fl mice . Exons 3 and 4 in the Cre+ , Restfl/fl mice were transcribed ( Figure 9B ) and encoded an ~130 kDa C terminal REST peptide ( hereafter RESTC ) ( Figure 9C ) that was not present in the Cre+ , RestGTi/GTi mice but remained in mouse embryonic stem cells ESCs ) created using a similar deletion strategy ( Jorgensen et al . , 2009 ) . The size of RESTC suggests that an initiator methionine was utilized in exon 3 . 10 . 7554/eLife . 09584 . 017Figure 9 . A C-terminal REST peptide , translated in conventional Rest KO mice , recruits co-repressors and protects against apoptosis . ( A ) Schematic of REST functional domains ( top ) and exons ( Ex , bottom ) . ( B ) qRT-PCR analysis , relative to 18S RNA , spanning indicated exons from E13 . 5 brain . ( C ) Western blot analysis of neurospheres ( NPC ) , E13 . 5 brain , and ESC lysates in indicated genotypes using antibody raised against the C-terminus of REST6 . Full length REST protein migrates at ~200kDa6; 130 kDa peptide represents peptide containing the RESTC . α-tubulin , loading control . ( D ) Representative immunostained images in cortical sections from indicated genotypes . ( E ) Quantification of apoptotic cells , n=5–6 mice/genotype , 6–10 sections/mouse . Mean values and 95% CI are shown . Statistical significance was determined using Kruskal ANOVA with Dunn posthoc , ( F ) Representative Western blot from ESC nuclear extracts , n=3 independent experiments . Chromatin extracts were performed with increasing NaCl concentrations; 0 ( lanes 1 , 5 ) ; 100 mM ( lanes 2 , 6 ) ; 200 mM ( lanes 3 , 7 ) ; 500 mM ( lanes 4 and 8 ) . Top two panels represent different exposures of the same Western blot . ( G ) Representative Western blot analyses of REST and RESTC immuno-complexes from ESC nuclear extracts . Extracts from ESC were immunoprecipitated with REST antibody and probed with antibodies indicated at right , n=3 independent experiments . *p< 0 . 05; **p<0 . 01; ***p< 0 . 001 . RD , repressor domain; gray rectangle , nuclear localization signal; purple box , zinc finger motif; red flag , initiator methionine . ANOVA , analysis of variance; ESC , embryonic stem cell; qRT-PCR; RESTC , C terminal peptide of REST . DOI: http://dx . doi . org/10 . 7554/eLife . 09584 . 017 Because we had shown that apoptosis contributes to smaller brain size in the Cre+ , RestGTi/GTi mice , the above results suggested that the normal brain size in Cre+ , Restfl/fl mice was due to RESTC activity that protected against apoptosis . To test this idea , we took advantage of Cre+ , RestGTi/+ mice that are hypomorphic with respect to brain size and apoptosis . These mice exhibit levels of apoptosis 50 . 1 and 37 . 4% lower than Cre+ , RestGTi/GTi mice depending on the brain region ( Cortex: Cre+ , RestGTi/GTi , 146 . 9 , CI 127 . 6–127 . 6; Cre+ , RestGTi/+ 74 . 69 , CI 69 . 6 79 . 8; LGE: Cre+ , Rest GTi/GTi , 203 . 9 , CI 169 . 1–238 . 6 , Cre+ , RestGTi/+ , 76 . 3 , CI 60 . 8–91 . 8 ) ( Figure 9D and 9E ) . This result is also reflected in fewer and less aggressive gliomas in the context of p53 deletion in Cre+ , RestGT/+ , Trp53fl/fl mice ( Table 1 ) . To test whether RESTC could block apoptosis in neural progenitors , we generated Cre+ mice heterozygous for the RestGTi and Restfl alleles ( Cre+ , RestGTi/fl ) , and compared clCasp3 labeling with that of Cre+ , RestGTi/GTi mice . The results indicated significantly less apoptosis in the Cre+ , RestGTi/fl mice in both cortex and LGE , with decreases to 24 . 8 and 9 . 1% of Cre+ RestGTi/GTi values , respectively ( Cortex: Cre+ , RestGTi/fl , 36 . 4 , CI 34 . 9–37 . 9; LGE: Cre+ , RestGTi/fl 18 . 54 , CI 6 . 3–30 . 8 ) ( Figure 9D and E ) . This result demonstrates that the remaining RESTC protein generated using Exon 2 deletion can still function in progenitors to prevent the DNA damage and apoptosis that results from complete loss of REST . Previous studies have demonstrated that both C- and N-terminal domains of REST exhibit repressor activity through recruitment of distinct co-repressor complexes ( Andres et al . , 1999; Ballas et al . , 2001; Grimes , 2000; Dingledine et al . , 1999; Naruse et al . , 1999; Roopra et al . , 2000 ) . Thus , the REST C-terminal peptide could prevent DNA damage and apoptosis by maintaining its known function and recruiting co-repressors directly to the chromatin . Indeed , RESTC was bound to chromatin prepared from ESCs , albeit not as effectively as full length REST ( Figure 9F ) . RESTC was also in immuno-complexes with the known REST co-repressors HDAC2 and CoREST ( Andres et al . , 1999; Ballas et al . , 2001 ) but not with the co-repressor SIN3A ( Figure 9G ) , which is recruited only by the N-terminus of REST ( Grimes , 2000;Dingledine et al . , 1999; Naruse et al . , 1999; Roopra et al . , 2000 ) .
Here , we show a new role for REST in protecting genomic integrity in cycling neural progenitors . Premature loss of REST results in prominent DNA damage during S phase , as well as in chromosomal abnormalities , resulting ultimately in apoptosis of neurons that normally mature during early to intermediate neurogenesis . These events are distinct from the robust up-regulation of neuronal genes that are predicted prima facie from premature loss of REST , and point to a mechanism for ensuring that proper terminal neuronal differentiation occurs only after exit from the cell cycle . The S phase of the cell cycle in neural progenitors is a critical decision point for proper terminal neuronal differentiation . REST is an ideal candidate for probing links between proliferation and terminal neuronal differentiation because it represses a large coterie of genes that are required for mature neuronal functions . As a repressor of these genes , REST is down-regulated dramatically during embryonic neurogenesis to allow elaboration of the terminally differentiated phenotype . However , REST is still associated with its chromatin modifiers on neuronal gene chromatin in the dividing progenitors ( Figure 2F [Ballas et al . , 2005] ) . The precise timing of release of REST from chromatin during the cell cycle has never been determined . Our results indicate that its normal removal must coincide with completion of cell division , because loss of REST and its co-repressors in cycling progenitors results in DNA damage , early cell cycle exit , and eventual apoptosis . In particular , our results suggest strongly that premature loss of REST causes DNA damage during S phase . We find significantly more mutant than control cells having a DNA damage response and a larger percentage of BrdU+ cells in the VZ with damage foci compared to BrdU− cells ( Figure 6G and H ) . Both DNA damage and premature cell cycle exit persist in the Cre+ , RestGTiGTi , Trp53fl/fl mice , indicating that these events can occur independently of activation of the apoptotic pathway . In most cases , apoptosis is a common outcome of DNA damage ( Houlihan and Feng , 2014 ) . Because a majority of the MAP2+ cells that accumulated at the VZ/SVZ boundary in the mutant were also positive for clCasp3 , we were unable to determine whether expression of MAP2 followed a normal differentiation pathway or an abnormal up regulation of MAP2 contributed to the apoptosis . Time lapse imaging of individual cells in the mutant may answer this question in the future . Even with apoptosis and smaller brain size , Cre+ , RestGTi/GTi mice lived to maturity . This situation contrasts with the widespread apoptosis and loss of entire brain structures that is often characteristic of loss of other chromatin complexes during embryogenesis . For example , brain KO of both HDACs 1 and 2 results in embryonic lethality with evidence of apoptosis throughout the brain ( Hagelkruys et al . , 2014; Montgomery et al . , 2009 ) . Similarly , brain deletion of TOPBP1 , which has a global role in maintaining genomic stability , causes widespread apoptosis and loss of entire brain structures ( Lee et al . , 2012 ) . The more robust phenotype of these factors is likely due to their inclusion in many different transcriptional complexes , whereas REST binding sites are restricted to a more limited set of sites in the genome . The brain phenotype of our mice was most similar to loss of the nuclear factor NDE1 reported recently to safeguard genomic integrity during S phase through interactions with the cohesin complex , implicated in replication fork fidelity ( Houlihan and Feng , 2014 ) . We did not identify NDE1 in any of our purified REST complexes . Furthermore , unlike the nearly complete brain size rescue in the NDE1 , p53 compound KO mice , the deletion of Trp53 in Rest GT mice , although it did block apoptosis , allowed only partial recovery to normal brain size , indicating residual effects of REST loss that did not occur with loss of NDE1 ( Houlihan and Feng , 2014 ) . These distinctions suggest different underlying mechanisms for their functions in neural progenitors . Interestingly , significant up-regulation of neuronal genes during development does occur outside the nervous system in REST-deficient mice ( Figure 1F and Aoki et al . , 2012; Paquette et al . , 2000 ) , indicating distinct mechanisms for S phase surveillance in non-neural and neural tissues . Premature loss of REST from neural progenitors , coupled with loss of the tumor suppressor p53 , leads to invasive GBM . The DNA damage due to loss of REST , in the absence of apoptosis with p53 deletion , persisted into adulthood and led to primarily proneural type GBM with PNET character ( Kouyialis et al . , 2005; Shingu et al . , 2005; Ohba et al . , 2008; Song et al . , 2011 ) . However , some tumors also expressed markers typical for mesenchymal and neuronal GBM types ( Figure 8A , Figure 7—figure supplement 2 ) , reflecting the fact that GBM are very heterogeneous , even when derived from single clones from patients ( Yung et al . , 1982; Wikstrand et al . , 1983 ) including heterogeneity in REST expression ( Wagoner and Roopra , 2012; Kamal et al . , 2012; Conti et al . , 2012 ) . While the Rest gene was transcriptionally inactive due to the GT , the Rest promoter was active , evidenced by β-gal staining , in the tumor cells in Cre+ , RestGTi/GTi , Trp53fl/fl mice ( Figure 7A and B ) , likely reflecting their progenitor/NSC character . Up-regulation of Rest mRNA and protein are a prominent feature in some human GBMs and medulloblastomas ( Wagoner and Roopra , 2012; Kamal et al . , 2012; Conti et al . , 2012; Su et al . , 2006; Majumder et al . , 2000 ) . Only a small percentage ( 23% , 3/13 gliomas ) of Cre+ , Trp53fl/fl mice developed GBM with PNET pathology , and these lacked subpial spread . Taken together , our results reinforce the idea , proposed previously ( Wagoner et al . , 2010; Westbrook et al . , 2008; Gurrola-Diaz et al . , 2003; Kreisler et al . , 2010; Moss et al . , 2009 ) for non-neuronal cells , that DNA damage due to loss of REST does not initiate tumors but rather promotes transformation and migration . A C-terminal repressor domain in REST protects genomic integrity in neural progenitors . Because REST recruits chromatin modifiers to repress its target genes , it seems reasonable to propose that DNA damage effects from premature loss of REST reflect a failure to properly re-establish correct chromatin modifications during the cell cycle . There are two well-established repressor domains in the N and C termini of REST ( Tapia-Ramirez et al . , 1997; Thiel et al . , 1998 ) . Each of these domains recruits histone deacetylases , as well as enzymes that methylate or demethylate histones ( Battaglioli et al . , 2002; Mulligan et al . , 2008; Lunyak , 2002; Lee et al . , 2005 ) . We showed that in a conventionally targeted REST KO mouse line targeting Rest exon2 , remaining C terminal exon ( s ) are translated into a protein that can bind neuronal chromatin and recruit the CoREST/HDAC complex . The presence of RESTC in the Rest GT heterozygote greatly reduced DNA damage and apoptosis , consistent with the idea that proper chromatin modifications are required to pass the S phase surveillance test . A study using a different Rest KO model , which targets exon 4 and leaves intact sequences coding for N-terminal repressor exons ( Aoki et al . , 2012 ) , did not report microcephaly or DNA damage . It is thus possible that the remaining N-terminal peptides are functional , consistent with the ability of both N- and C-terminal repression domains of REST to mediate repression function ( Tapia-Ramirez et al . , 1997 ) . It will be of interest to determine DNA damage and premature differentiation are separable events , and this may be resolved through more in-depth comparisons between Rest GT and conventional Rest KO mouse models . A new model for REST function during embryonic neurogenesis . We propose that REST repression protects the integrity of neuronal genes whose expression must be delayed until terminal differentiation . There is no evidence for direct interactions of REST with DNA replication machinery . Therefore , we suggest that REST protection occurs by maintaining proper chromatin modifications during S phase . This could be achieved simply by preventing premature expression of terminal neuronal genes , thereby preventing aberrant chromatin modifications and premature cell cycle exit , and/or by functions of components in the REST repressor complex dedicated to DNA damage control . Although the magnitude and number of mRNAs de-repressed in the REST mutant at E12 . 5 was very low ( Supplementary file 2 ) , it is possible that abnormally timed transition of their chromatin from repressor to activator marks could contribute to a DNA damage response . With respect to the REST complex , several associated factors have the potential to protect against DNA damage . Intriguingly , FACT ( FAcilitates Chromatin Transcription ) turned up recently as a member of the REST complex in ESCs ( Mcgann et al . , 2014 ) In mammalian cells and yeast , loss of FACT results in DNA double-strand breaks that interfere with DNA replication through the formation of R loops ( Herrera-Moyano et al . , 2014 ) . Further studies are required to test this new model . However , our current results indicate that a functional REST repressor complex is required for proper cell cycle transition during neurogenesis , and that normal loss of REST repression from neuronal gene chromatin is timed precisely to coincide with cell cycle exit and the ‘permission’ to terminally differentiate . | In the brain , cells called neurons connect to each other to form complex networks through which information is rapidly processed . These cells start to form in the developing brains of animal embryos when “neural” stem cells divide in a process called neurogenesis . For this process to proceed normally , particular genes in the stem cells have to be switched on or off at different times . This ensures that the protein products of the genes are only made when they are needed . Proteins called transcription factors can bind to DNA to activate or inactivate particular genes; for example , a transcription factor called REST inactivates thousands of genes that are needed by neurons . During neurogenesis , the production of REST normally declines , and some studies have shown that if the production of this protein is artificially increased , the formation of neurons is delayed . However , other studies suggest that REST may not play a major role in neurogenesis . Here , Nechiporuk et al . re-examine the role of REST in mice . The experiments used genetically modified mice in which the gene that encodes REST was prematurely switched off in neural stem cells . Compared with normal mice , these mutant mice had much smaller brains that contained fewer neurons because the stem cells stopped dividing earlier than normal . Unexpectedly , many genes that are normally switched off by REST , were not significantly changed , while genes that are not normally regulated by REST – such as the gene that encodes a protein called p53 – were active . It is known from previous work that p53 is expressed when cells are exposed to harmful conditions that can damage DNA . This helps to prevent cells from becoming cancerous . Nechiporuk et al . found that cells that lacked REST had higher levels of DNA damage than normal cells due to errors during the process of copying DNA before a cell divides . Furthermore , when both REST and p53 were absent , the neural stem cells became cancerous and formed tumors in the mice . Nechiporuk et al . ’s findings suggest that REST protects the DNA of genes that are needed for neurons to form and work properly . The new challenge is to understand where in the genome the damage is occurring . | [
"Abstract",
"Introduction",
"Results",
"Discussion"
] | [
"developmental",
"biology"
] | 2016 | The REST remodeling complex protects genomic integrity during embryonic neurogenesis |
The basidiomycete yeast Rhodosporidium toruloides ( also known as Rhodotorula toruloides ) accumulates high concentrations of lipids and carotenoids from diverse carbon sources . It has great potential as a model for the cellular biology of lipid droplets and for sustainable chemical production . We developed a method for high-throughput genetics ( RB-TDNAseq ) , using sequence-barcoded Agrobacterium tumefaciens T-DNA insertions . We identified 1 , 337 putative essential genes with low T-DNA insertion rates . We functionally profiled genes required for fatty acid catabolism and lipid accumulation , validating results with 35 targeted deletion strains . We identified a high-confidence set of 150 genes affecting lipid accumulation , including genes with predicted function in signaling cascades , gene expression , protein modification and vesicular trafficking , autophagy , amino acid synthesis and tRNA modification , and genes of unknown function . These results greatly advance our understanding of lipid metabolism in this oleaginous species and demonstrate a general approach for barcoded mutagenesis that should enable functional genomics in diverse fungi .
Rhodosporidium toruloides ( also known as Rhodotorula toruloides [Wang et al . , 2015] ) is a basidiomycete yeast ( subdivision Pucciniomycotina ) . Rhodotorula/Rhodosporidium species are widely distributed in the phyllosphere and diverse soils ( Rosa and Peter , 2006; Sláviková et al . , 2009; Butinar et al . , 2005; Pulschen et al . , 2015 ) . They accumulate high concentrations of carotenoid pigments ( Mata-Gómez et al . , 2014; Lee et al . , 2014 ) , giving their colonies a distinctive orange , red , or pink hue . When R . toruloides is cultured under nitrogen ( Zhu et al . , 2012 ) , sulfur ( Wu et al . , 2011 ) , or phosphorus ( Wu et al . , 2010 ) limitation , it can accumulate as much as 70% of cellular biomass as lipids ( Wiebe et al . , 2012 ) , primarily as triacylglycerides ( TAG ) . Eukaryotes accumulate neutral lipids in complex , dynamic organelles called lipid droplets . Lipid droplets emerge from the endoplasmic reticulum ( ER ) membrane as a core of TAG surrounded by sterol esters , a phospholipid monolayer derived from ER phospholipids , and a targeted ensemble of proteins mediating inter-organelle interaction , protein trafficking , cellular lipid trafficking and regulated carbon flux in and out of the lipid droplet ( Walther and Farese , 2012; Farese and Walther , 2009; Gao and Goodman , 2015 ) . Aberrant lipid droplet formation contributes to many human diseases ( Krahmer et al . , 2013a; Welte , 2015 ) and impacts cellular processes as diverse as autophagy ( Shpilka et al . , 2015 ) and mitosis ( Yang et al . , 2016 ) . The propensity of R . toruloides to form large lipid droplets under a variety of conditions makes it an attractive platform to study conserved aspects of the cellular biology of these important organelles across diverse eukaryotes . Rhodosporidium toruloides is also an attractive host for production of sustainable chemicals and fuels from low-cost lignocellulosic feedstocks . Wild isolates of R . toruloides can produce lipids and carotenoids from a wide variety of carbon sources including glucose ( Wiebe et al . , 2012 ) , xylose ( Wiebe et al . , 2012 ) , and acetate ( Huang et al . , 2016 ) , as well as complex biomass hydrolysates ( Fei et al . , 2016 ) . They are relatively tolerant to many forms of stress including osmotic stress ( Singh et al . , 2016 ) and growth-inhibiting compounds in biomass hydrolysates ( Hu et al . , 2009; Kitahara et al . , 2014 ) . Rhodosporidium toruloides has been engineered to produce lipid-derived bioproducts such as fatty alcohols ( Fillet et al . , 2015 ) and erucic acid ( Fillet et al . , 2017 ) from synthetic pathways . To enable more efficient production of terpene-derived and lipid-derived chemicals , it has also been engineered for enhanced carotenoid ( Lee et al . , 2016 ) and lipid ( Zhang et al . , 2016a ) production . These efforts , while promising , have for the most part employed strategies adapted from those demonstrated in evolutionarily distant species such as Saccharomyces cerevisiae and Yarrowia lipolytica . To truly tap the biosynthetic potential of R . toruloides , a better understanding of the unique aspects of its biosynthetic pathways , gene regulation and cellular biology will be required . Recently , transcriptomic and proteomic analysis of R . toruloides in nitrogen limited conditions ( Zhu et al . , 2012 ) identified over 2 , 000 genes with altered transcript abundance and over 500 genes with altered protein abundance during lipid accumulation . These genes included many enzymes involved in the TCA cycle , a putative PYC1/MDH2/Malic Enzyme NADPH conversion cycle ( Wynn et al . , 1999 ) , fatty acid synthesis , fatty acid beta-oxidation , nitrogen catabolite repression , assimilation and scavenging , autophagy , and protein turnover . Proteomics of isolated lipid droplets ( Zhu et al . , 2015 ) identified over 250 lipid droplet-associated proteins including fatty acid synthesis genes , several putative lipases , a homolog of the lipolysis-regulating protein perilipin ( Bickel et al . , 2009 ) , vesicle trafficking proteins such as Rab GTPases and SNARE proteins , as well as several mitochondrial and peroxisomal proteins . While these studies were unambiguous advances for the field , significant work remains to establish the genetic determinants of lipid accumulation in R . toruloides . Differential transcript or protein abundance under nitrogen limitation is suggestive of function in lipid accumulation , but transcriptional regulation and gene function are often poorly correlated in laboratory conditions ( Price et al . , 2013 ) . Similarly , sequestration in the lipid droplet may help regulate availability of some proteins for functions not necessarily related to lipid metabolism ( Cermelli et al . , 2006 ) . More direct functional data would help the R . toruloides community prioritize this extensive list of genes for more detailed study and identify additional genes not identifiable by proteomic and transcriptomic methods . Finally , these studies highlighted dozens of genes with no known function , and hundreds more with only limited functional predictions . A more functional approach may yield more insights into unique aspects of R . toruloides biology . Fitness analysis of gene deletion or disruption mutants within pooled populations is a flexible , powerful approach for elucidating gene function . In these experiments the relative growth rate of thousands of mutant strains are simultaneously measured by tracking the relative abundance of unique sequence identifiers for each mutant . These identifying sequences could be short sequence ‘barcodes’ inserted into targeted deletion mutants ( Giaever et al . , 2002 ) , or genomic DNA flanking random transposon insertions ( Sassetti et al . , 2001 ) . Early fitness experiments tracked strain abundance by hybridization of identifier sequences to DNA micro-arrays ( Giaever et al . , 2002; Sassetti et al . , 2001 ) . The advent of high-throughput sequencing and the development of broad host range transposons enabled more widespread use of fitness analysis in bacteria by direct sequencing of transposon insertion sites ( TnSeq ) ( Gawronski et al . , 2009; Langridge et al . , 2009 ) . The scalability and precision of TnSeq is improved when random sequence barcodes are added to each randomly integrated transposon ( RB-TnSeq ) ( Wetmore et al . , 2015 ) . Once insertions sites have been mapped , strain abundance can then be more accurately measured with a simple , consistent PCR amplification of the barcode sequences from known priming sites ( BarSeq ) . TnSeq and RB-TnSeq have been employed extensively in bacteria ( Kwon et al . , 2016 ) , and in a few eukaryotic species ( Michel et al . , 2017; Pettitt et al . , 2017 ) . Although some of the first barcoded fitness experiments were performed on mutant pools of S . cerevisiae ( Giaever et al . , 2002 ) and advances in TnSeq methods continue in that species ( Michel et al . , 2017 ) , to date relatively low transformation efficiencies and a lack of functional transposon systems has limited the application of TnSeq and RB-TnSeq in most fungal species . Random mutagenesis of fungi by the bacterium Agrobacterium tumefaciens is one route to overcome these technical barriers . Agrobacterium tumefaciens , an opportunistic plant pathogen , has evolved an efficient system to transfer virulence genes into eukaryotic cells ( Gelvin , 2003 ) . Once in the host cell , these transfer DNAs ( T-DNAs ) integrate randomly into the genome ( Bundock et al . , 2002 ) . Agrobacterium tumefaciens-mediated transformation ( ATMT ) has been used extensively in plants ( Gelvin , 2003 ) and to transform diverse fungi at high efficiency ( Bundock et al . , 2002; Michielse et al . , 2005; Walton et al . , 2005; Kunitake et al . , 2011; Sullivan et al . , 2002; Blaise et al . , 2007 ) . Recently , Esher et al . used ATMT followed by mutant selection and high-throughput sequencing to identify several mutants with altered cell wall biosynthesis in the basidiomycete yeast Cryptococcus neoformans ( Zhang et al . , 2016a ) . The methods they employed were only viable for characterization of a small pool of highly enriched mutants , but they demonstrated an effective paradigm to bring high-throughput functional genomics to diverse fungi . In this study , we demonstrate the construction of a randomly barcoded , random insertion library in R . toruloides by ATMT and its application for functional genomics ( RB-TDNAseq ) . We report a list of 1 , 337 genes , including 36 unique to basidiomycetes , that were recalcitrant to T-DNA insertion , the first full genome survey of putatively essential genes in a basidiomycete fungus . We use our barcoded mutant library to explore fatty acid catabolism in R . toruloides , demonstrating its utility in rapidly assessing mutant phenotypes . We show that mitochondrial beta-oxidation is important for fatty acid utilization in this species and that some members of its expanded complement of peroxisomal acyl-CoA dehydrogenases are necessary for growth on different fatty acids , suggesting substrate specificity or conditional optimality for each enzyme . We investigate perturbed lipid accumulation in the mutant pool by fractionation of the population by buoyancy and fluorescence activated cell sorting . We identify 150 genes with significant roles in lipid accumulation , notably genes involved in signaling cascades ( 28 genes ) , gene expression ( 15 genes ) , protein modification or trafficking ( 15 genes ) , ubiquitination or proteolysis ( nine genes ) , autophagy ( nine genes ) , and amino acid synthesis ( eight genes ) . We also find evidence that tRNA modification affects lipid accumulation in R . toruloides , identifying five genes with likely roles in thiolation of tRNA wobble residues . These results significantly advance our understanding of lipid metabolism in R toruloides; identify key biological processes that should be explored and optimized in any oleaginous yeast engineered for lipid production; support emerging evidence of deep connections between lipid droplet dynamics , vesicular trafficking , and protein sorting; and demonstrate a general approach for barcoded mutagenesis that should enable functional genomics in a wide variety of fungal species .
To enable functional genomics in R . toruloides IFO 0880 , we first improved the existing genome assembly and annotation ( Zhang et al . , 2016a ) using a combination of long-read PacBio sequencing for a more complete de novo assembly , a more comprehensive informatics approach for gene model predictions and functional annotation , and manual refinement of those models using evidence from mRNA sequencing ( Genbank accession LCTV02000000 ) , also available at the Mycocosm genome portal ( Grigoriev et al . , 2014 ) ( see Appendix 1 for details ) . Summary tables of gene IDs , predicted functions , and probable orthologs in other systems are included in Supplementary file 1 . For brevity , we will refer to R . toruloides genes by the common name for their Saccharomyces cerevisiae orthologs ( e . g . MET2 ) when such orthologous relationships are unambiguous . Otherwise , we will give the Mycocosm protein ID , e . g . RTO4_12154 and RTO4_14576 are both orthologs of GPD1 . Because no method existed for high-throughput genetics in R . toruloides , we adapted established protocols for mapping barcoded transposon insertions ( RB-TnSeq ) ( Wetmore et al . , 2015 ) , to mapping barcoded T-DNA insertions introduced with Agrobacterium tumefaciens-mediated transformation ( ATMT ) . We call this method RB-TDNAseq ( Figure 1A ) . In brief , we generated a diverse library of binary ATMT plasmids bearing nourseothricin resistance cassettes with ~10 million unique 20 base-pair sequence ‘barcodes’ by efficient Type IIS restriction enzyme cloning ( Engler et al . , 2008 ) , introduced the library into A . tumefaciens EHA105 by electroporation , then transformed R . toruloides with ATMT . Using a TnSeq-like protocol , we mapped the unique locations of 293 , 613 individual barcoded T-DNA insertions in the R . toruloides genome ( see Appendix 1 for details ) . Once insertion sites were associated with their barcodes , pooled fitness experiments were performed using a simple , scalable BarSeq protocol as previously described ( Wetmore et al . , 2015 ) . Insertions were sufficiently well dispersed to map at least one T-DNA in 93% of nuclear genes , despite some local and fine-scale biases in insertion rates ( see Appendix 1 for details ) . Insertion density in coding regions followed an approximately normal distribution ( as expected for random integration ) centered on nine inserts per thousand base pairs , except for a subpopulation of genes with fewer than two inserts/kb ( Figure 1B ) . These very low-insertion genes were highly enriched for orthologs of genes reported as essential in Aspergillus nidulans ( Arnaud et al . , 2012 ) , Cryptococcus neoformans ( Ianiri and Idnurm , 2015 ) , Saccharomyces cerevisiae ( Cherry et al . , 2012 ) , or Schizosaccharomyces pombe ( Wood et al . , 2012 ) , or for which only heterokaryons could be obtained in the Neurospora crassa deletion collection ( Colot et al . , 2006 ) . We therefore infer that the majority of these genes recalcitrant to T-DNA insertion are likely essential in our library construction conditions , or at least that mutants for these genes have severely compromised growth . Based on the above criterion , we identified 1 , 337 probable essential genes , which we report in Supplementary file 1 . This list includes over 400 genes not reported as essential in the above-mentioned model fungi and is enriched for genes with homologs implicated in mitochondrial respiratory chain I assembly and function , dynein complex , the Swr1 complex , and mRNA nonsense mediated decay . For a full list of GO term enrichments see Supplementary file 1 . This list also includes 36 genes unique to basidiomycetes . Before investigating more novel aspects of R . toruloides’ biology , we tested if RB-TDNAseq could be used to correctly identify gene function in well-conserved amino acid biosynthetic pathways . We cultured the mutant pool in defined medium ( DM ) , consisting of glucose and yeast nitrogen base without amino acids and in DM supplemented with ‘drop-out mix complete’ ( DOC ) , a mix of amino acids , adenine , uracil , p-aminobenzoic acid , and inositol . To establish if RB-TDNAseq could produce statistically robust results with minimal experimental replication , we recovered three independent starter cultures from frozen aliquots of the mutant pool and used each replicate to inoculate both supplemented and non-supplemented cultures . We grew these cultures for seven generations and measured fitness across the mutant pool with BarSeq . Secondary mutations are prevalent even in well-curated mutant collections ( Comyn et al . , 2017 ) and ATMT can introduce several types of confounding mutations ( see Appendix 1 for details ) . To mitigate the influence of such mutations on our analysis , we adapted the established methods and software of Wetmore et al . ( Wetmore et al . , 2015; Price et al . , 2016; Cole et al . , 2017; Sagawa et al . , 2017 ) for our BarSeq analysis . These algorithms compute a fitness score for each mutant strain as a log2 ratio of abundance in the experimental condition to abundance in a ‘Time 0’ sample from its seed culture . A composite fitness score ( F ) is then computed for each gene by combining multiple fitness scores from strains bearing insertions in that gene . A ‘moderated T-statistic’ calculated from the average and variance of strain fitness scores indicates the consistency of F across strains and experiments . See the Materials and methods section and ( Wetmore et al . , 2015 ) for more information on how these metrics are calculated . For more information on sequencing depth , behavior of T-statistics and detailed examples of how individual strain fitness scores contribute to F , see Appendix 1 . All fitness scores and T-statistics ( combined across biological replicates ) are available in Supplementary file 2 and online in a dynamic fitness browser , adapted from ( Price et al . , 2016 ) : http://fungalfit . genomics . lbl . gov/ . Different aliquots of the mutant pool have subtly different starting compositions and experience stochastic variations in the length of lag phase as they recover from frozen stocks . Subtle variations in Illumina library preparation and sequencing for samples processed at different times may add further batch-specific biases to count data . For these reasons , direct comparisons of BarSeq counts between conditions tested in different batches and seeded from different starter cultures are not advisable . Expressing the data as F and T relative to Time 0 reduces it to a more portable format , allowing for comparisons of mutant fitness across conditions not necessarily tested in the same experiment . Given F and T in two different conditions ( FC1 , TC1 and FC2 , TC2 ) , we calculate relative fitness FC1-C2 = FC1-FC2 and relative T-statistics TC1-C2 = ( FC1-FC2 ) /sqrt ( var ( FC1 ) +var ( FC2 ) ) . Fitness scores for 6 , 558 genes in cultures grown on DM and DM supplemented with DOC are shown in Figure 2A . Mutants for 28 genes had fitness scores suggesting auxotrophy: fitness defects in non-supplemented media ( FDM < −1 ) with consistently different scores in supplemented versus non-supplemented media ( TDM-DOC < −3 ) . When we grew the mutant pool in defined media with methionine or arginine supplementation ( Figure 2B ) , the 28 auxotrophic mutants partitioned into 11 mutants rescued by methionine , eight mutants rescued by arginine , seven mutants rescued by neither amino acid and two mutants rescued by both amino acids . All of the identified methionine and arginine auxotrophic mutants have orthologous genes for which mutants are auxotrophic for methionine/cysteine or arginine , respectively , in S . cerevisiae or A . nidulans . Alternatively , when we hierarchically clustered the fitness scores for genes with F < −1 and T < −3 versus Time 0 in any supplementation condition ( Figure 2C ) , the resulting clusters included twelve and nine mutants rescued by methionine and arginine respectively; this was a nearly complete recovery of genes with predicted functions in this pathway ( shown in Figure 2D–E with additional discussion in Appendix 1 ) . Based on these data , we chose |T| > 3 as a conservative threshold for consistent , reliable fitness scores in further BarSeq experiments . We next sought to understand how R . toruloides utilizes distinct fatty acids as growth substrates , as a window onto the complex lipid metabolism in this fungus . For this purpose , we used RB-TDNAseq to measure mutant fitness on three fatty acids as the sole carbon source: oleic acid ( the most abundant fatty acid in R . toruloides [Li et al . , 2007] ) ricinoleic acid ( a high-value fatty acid produced naturally in plants ( Dyer et al . , 2008 ) and synthetically in fungi [Holic et al . , 2012] ) , and methyl ricinoleic acid ( a ricinoleic acid derivative used in lactone production [Endrizzi et al . , 1996] ) . A total of 129 genes had consistently low fitness scores on one or more fatty acids including genes implicated in beta-oxidation of fatty acids , gluconeogenesis , mitochondrial amino acid metabolism , and several other aspects of cellular metabolism and gene regulation ( See Figure 3—figure supplement 1 and Appendix 1 for a clustering analysis of fitness scores for these genes and Supplemental file 2 for a complete list ) . We were particularly interested in beta-oxidation of fatty acids in the peroxisome and mitochondria , as these pathways are critical for lipid homeostasis ( Kohlwein et al . , 2013; Rambold et al . , 2015 ) , with major implications for both human health ( Houten et al . , 2016; Waterham et al . , 2016 ) and metabolic engineering in fungi ( Dulermo and Nicaud , 2011; Beopoulos et al . , 2014 ) . Fitness scores for R . toruloides genes homologous to enzymes with known roles in beta-oxidation of fatty acids are shown in Figure 3A . The localization for these enzymes is inferred mostly from observations in distantly related species , but orthologs of five enzymes localized to the predicted compartments in the basidiomycete yeast Ustilago maydis ( Camões et al . , 2015 ) adding some confidence to these predicted locations . Mutants for mitochondrial enzymes had the most consistent fitness scores across all three fatty acids , whereas mutants for the peroxisomal enzymes and peroxins had more variable fitness scores among fatty acids . Mutants for seven peroxisomal beta-oxidation enzymes and three peroxins had different fitness scores on oleic acid versus ricinoleic acid and methylricinoleic acid ( listed in Appendix 1 , full fitness scores in Supplementary file 2 ) , while 11 other predicted peroxisomal beta-oxidation enzymes had no consistent fitness scores at all . These results demonstrate how RB-TDNAseq can be used to rapidly identify condition-specific phenotypes among closely related members of a gene family . All together our data are consistent with a model of fatty acid beta-oxidation in R . toruloides in which diverse long-chain fatty acids are shortened in the peroxisome and a less structurally diverse set of short-chain fatty acids are oxidized to acetyl-CoA in the mitochondria ( Figure 3—figure supplement 2 ) . To validate our fitness data on fatty acids , we made targeted deletion mutants for several predicted peroxisomal and mitochondrial proteins by homologous recombination into a non-homologous end joining deficient YKU70∆ strain ( also known as KU70 ) ( Ninomiya et al . , 2004; Zhang et al . , 2016b ) . We grew these mutant strains on oleic or ricinoleic acid media and compared their growth to the parental YKU70∆ strain in mid-log phase . Relative growth for the deletion strain for each gene is compared to its fitness scores in the BarSeq experiment in Figure 3B and Figure 3C . The PEX7∆ mutant had similar fitness defects on both fatty acids , but mutants for RTO4_8673 ( similar to PEX11 ) and RTO4_14567 ( similar to H . sapiens ACAD11 ) , had stronger fitness defects on ricinoleic acid , and the mutant for acyl-CoA dehydrogenase RTO4_8963 had stronger fitness defects on oleic acid as predicted from fitness scores . Over a 96 hr time course , the RTO4_14567∆ mutant failed to grow at all on ricinoleic acid , whereas the RTO4_8963∆ mutant and the PEX11 homolog RTO4_8673∆ mutant had more subtle phenotypes , approaching the same final density of the YKU70∆ control strain after a longer growth phase ( Figure 3—figure supplement 3 ) . These data showed that BarSeq fitness scores were reliable predictors of significant growth defects for mutants in pure culture . To dissect the genetic basis of lipid accumulation in R . toruloides , we induced lipid accumulation by nitrogen limitation ( R . toruloides lipid droplets visualized in Figure 4A ) , and used two measures of cellular lipid content to fractionate the mutant pool ( Figure 4B and Appendix 1 ) . We used the neutral-lipid stain BODIPY 493/503 ( Bozaquel-Morais et al . , 2010 ) and fluorescence activated cell sorting ( FACS ) to enrich populations with larger/more or smaller/fewer lipid droplets ( Terashima et al . , 2015 ) . We also used buoyancy separation on sucrose gradients to enrich for populations with higher or lower total lipid content ( Eroglu and Melis , 2009; Kamisaka et al . , 2006; Liu et al . , 2015 ) . Because many mutations can affect cell buoyant density independent of lipid accumulation ( Novick et al . , 1980; Bryan et al . , 2010 ) , we also grew the mutant pool in rich media ( YPD ) and subjected it to sucrose gradient separation as a control for lipid-independent buoyancy phenotypes . For each pair of high and low lipid fractions , we then calculated an ‘enrichment score’ , E , and T-statistic for each gene . E is analogous to our fitness scores based on growth , except that it is the log2 ratio of abundance in the high lipid fraction to the low lipid fraction , whereas F is the log2 ratio of final to initial abundance . Hierarchical clusters of enrichment scores for 271 genes for which mutants have consistently altered lipid accumulation ( |E| > 1 and |T| > 3 ) are shown in Figure 5A . Enrichment scores and T-statistics for all 6 , 558 genes with sufficient BarSeq data are reported in Supplementary file 2 . To assess the reliability of these enrichment scores in predicting phenotypes for null mutants , we constructed 29 single gene deletion mutants by homologous recombination in a YKU70∆ strain of IFO 0880 and measured lipid accumulation by average BODIPY fluorescence for 10 , 000 cells from each strain using flow cytometry . Figure 5B and C show relative BODIPY signal for targeted deletion mutants versus the YKU70∆ parental strain ( see Appendix 1 for more information on normalization and power analysis ) . When enrichment scores from both assays were strongly positive ( LA1 ) , we found that 7 of 8 deletion mutants had the expected phenotype ( i . e . increased lipid accumulation ) . When only one assay yielded a strongly positive score ( clusters LA2 and LA3 ) , only 3 of 5 mutants had apparent increases in lipid content as measured by flow cytometry . Further , for the two mutants for genes in cluster LA3 with the greatest apparent increase in lipid content ( PMT4 and RTO4_10302 , similar to C . neoformans CMT1 ) that measurement was likely an artifact of incomplete cell separation . Both mutants formed long chains of cells ( see Figure 7—figure supplement 1 for microscopy images ) , which would be analyzed as a single cell by our FACS assay . Genes in clusters LA4 and LA5 had conflicting enrichment scores between the two assays . Of three targeted deletion strains for genes in these clusters , only one ( CCC1∆ ) had a statistically significant phenotype , with decreased lipid accumulation . When the FACS assay gave a strongly negative score and there was no strong contrary buoyancy score ( clusters LA6 , LA7 , and LA8 ) , 11 of 13 mutants had reduced lipid accumulation . These data confirm that both separation techniques are fundamentally sound , though in isolation each method has a significant rate of false positives . In combination , the two assays identified a large set of high-confidence candidate genes with important roles in lipid accumulation . We manually curated homology-based predicted functions for the 393 genes with consistent fitness or enrichment scores in this study ( Supplementary file 1 ) . An overview of predicted localizations and functions for genes we identified with roles in fatty acid utilization or lipid accumulation is shown in Figure 6 , with more detail for mutants with increased and decreased lipid accumulation in Tables 1 and 2 , respectively . Note that we have excluded genes for which only one enrichment technique indicated altered lipid accumulation from this analysis . Mutants with increased lipid accumulation ( cluster LA1 , 56 genes ) were most notably enriched for genes involved in signaling cascades , post-translational protein modification and trafficking , and in amino acid biosynthesis . Genes involved in signaling cascades included several homologs to G-proteins such as RAS1 and mammalian RAC1 and their effectors , as well as several kinases , indicating a complex signaling network regulating lipid accumulation . Genes involved in protein trafficking included P24 adapter proteins , suggesting they play an important role in delivering lipid-mobilizing genes to the lipid droplet or removing lipid biosynthesis genes from the endomembrane network . Mutants for several genes identified in our auxotrophy experiments also had increased lipid accumulation , most notably genes involved in sulfate assimilation for cysteine and methionine biosynthesis . Not all auxotrophic mutants had altered lipid accumulation , suggesting that arrested protein synthesis is not necessarily sufficient to increase lipid accumulation . Mutants with decreased lipid accumulation ( clusters LA6 , LA7 , and LA8 , 94 genes ) were most notably enriched for genes with roles in autophagy , protein phosphorylation , and tRNA-modifcation . Mutants in nine core components of autophagy were deficient for lipid accumulation , consistent with previous findings that chemical inhibition of autophagy reduced lipid accumulation in Y . lipolytica ( Qiao et al . , 2015 ) . Mutants in several proteases and ubiquitin ligases also had reduced lipid accumulation , highlighting the importance of efficient recycling of cellular materials to refactor the cell for high lipid accumulation . Mutants in at least nine protein kinases , three phosphatases or their binding partners had reduced lipid accumulation; likely these genes mediate nutrient sensing cascades that stimulated lipid accumulation . Several genes with likely roles in thiolation of tRNA wobble residues had lower lipid accumulation . Though these mutants also had apparent buoyancy phenotypes on YPD , two deletion strains ( NCS6∆ and NCS2∆ ) had reduced lipid content in pure culture ( Figure 5C ) . They may play a role in regulating global carbon metabolism ( Laxman et al . , 2013 ) . RTO4_16381 , a distant homolog of H . sapiens PLIN1 ( perilipin ) , was also necessary for high lipid accumulation , consistent with its homolog’s known roles in lipid body maintenance and regulation of triglyceride hydrolysis ( Bickel et al . , 2009 ) and previous observations that it localized to lipid droplets in R . toruloides ( Zhu et al . , 2015 ) . To further characterize the phenotypes of our lipid accumulation mutants , we performed differential interference contrast ( DIC ) and fluorescence microscopy . The mutants showed a variety of phenotypes with respect to both cellular and lipid droplet morphology . Eight examples are highlighted in Figure 7 . While wild type cells most commonly had two lipid droplets of similar size , several high lipid accumulation mutants had qualitatively more cells with three or more lipid droplets ( e . g . MET14∆ , Figure 7 ) ) or cells with a single dominant droplet ( e . g . RAC1∆ , Figure 7 ) . RAC1∆ also had qualitatively larger , more spherical cells . A KDELC-like∆ mutant with increased lipid accumulation also showed a defect in cell separation likely reflective of combined defects in lipid accumulation , secretion , and cell wall/septum formation . All strains had a wide cell-to-cell variation in lipid droplet size , consistent with high variance in BODIPY intensity measured by flow cytometry ( Figure 4—figure supplement 2A ) . Most low-lipid strains appeared morphologically similar to wild type with smaller lipid bodies ( Figure 7—figure supplement 1 ) . However , a BSCL2-like∆ ( seipin ) mutant showed an even larger variation in droplet size than wild type , consistent with observations in S . cerevisiae mutants for the homolog SEI1/FLD1 ( Fei et al . , 2008 ) and likely reflective of a conserved function in lipid droplet formation and efficient delivery of lipid biosynthetic proteins to the growing lipid droplet ( Wang et al . , 2016; Pagac et al . , 2016; Salo et al . , 2016 ) . Autophagy mutants ( ATG2∆ ) had the most uniformly small lipid droplets in elongated cells with enlarged vacuoles . Overall , the morphological phenotypes we observed in R . toruloides are similar to a number of previous microscopic screens for altered lipid accumulation in diverse eukaryotes ( Fei et al . , 2008; Szymanski et al . , 2007; Guo et al . , 2008; Zehmer et al . , 2009; Ashrafi et al . , 2003 ) .
We employed an established method , Agrobacterium tumefaciens-mediated transformation , to extend barcoded insertion library techniques ( Wetmore et al . , 2015 ) into a non-model basidiomycetous fungus . The efficiency of A . tumefaciens transformation in diverse fungal species ( Michielse et al . , 2005; Martínez-Cruz et al . , 2017; Wu et al . , 2016; Zhang et al . , 2015; Liu et al . , 2013; Zhang et al . , 2014; Li et al . , 2013; Han et al . , 2012; Muniz et al . , 2014; Rodrigues et al . , 2013; Celis et al . , 2017 ) will enable use of RB-TDNAseq in many fungal species with limited genetic tools . We used RB-TDNAseq to simultaneously track mutants in over 6 , 500 genes for altered lipid catabolism and neutral lipid accumulation using a simple , scalable BarSeq protocol . The phenotypes measured in our high-throughput experiments were consistent with those observed for single gene deletion strains , demonstrating the reliability of this approach . In some respects R . toruloides was an ideal species to develop these methods . The R . toruloides genome is relatively compact ( just over 20% of the sequence is predicted to be intergenic ) , and it grows as a haploid yeast . Effective BarSeq analysis on species with larger , less dense genomes will require greater sequence depth per sample . Typical fungal genomes are only modestly larger , though , around 35–45 Mb ( Mohanta and Bae , 2015 ) vs 20 Mb for R . toruloides . Sequencing limitations are thus already minimal and will only decrease in the foreseeable future . A greater challenge will be adapting this technology in fungi that grow mainly as diploids or in filamentous , multicellular , or multinucleate forms harboring genetically distinct nuclei . Many of those species also produce haploid , uninucleate spores for sexual reproduction , asexual dispersal , or both . RB-TDNAseq can be applied to study the germination of these spores and their growth into nascent , isogenic colonies prior to their fusion into more physiologically and genetically complex networks of mycelia and fruiting bodies . We found that genes recalcitrant to T-DNA insertion were highly enriched in orthologs for known essential genes , suggesting that most genes with very low insertion rates were likely essential in our mutagenesis conditions . Previous studies employing high-density transposon mutagenesis in fungi and bacteria have demonstrated the general utility of this approach ( Michel et al . , 2017; Le Breton et al . , 2015 ) . The high efficiency of A . tumefaciens-mediated transformation in diverse fungi should enable similar surveys in many poorly annotated fungi . We hope the provisional list of essential genes identified here will serve as a useful resource for genetics in R . toruloides and related species . In particular , orthologs to these genes may be potential targets for new antifungal strategies against basidiomycete pathogens , such as the closely related rusts of the Pucciniomycotina subphylum ( Singh et al . , 2015; Park et al . , 2015 ) and the more distantly related human pathogen Cryptococcus neoformans ( May et al . , 2016 ) . The presence of a probable mitochondrial fatty acid beta-oxidation pathway in R . toruloides has been noted previously ( Zhu et al . , 2012 ) . Our results confirm that this pathway is functional and essential for fatty acid utilization and add to mounting evidence that mitochondrial beta-oxidation is widespread in fungi ( Khan et al . , 2012 ) . In mammals , some branched long-chain fatty acids are shortened in the peroxisome , then transferred via the acylcarnitine shuttle to the mitochondria for complete oxidation ( Wanders et al . , 2015; Swigonová et al . , 2009 ) , while other long-chain fatty acids are metabolized solely in the mitochondria ( Chegary et al . , 2009 ) . Rhodosporidium toruloides has orthologs to the mammalian mitochondrial short , branched-chain and medium-chain acyl-CoA dehydrogenases ACADSB and ACADM , but not to the long-chain and very long-chain acyl-CoA dehydrogenases ACADL and ACADVL . Rhodosporidium toruloides also has several homologs to peroxisomal long chain acyl-CoA dehydrogenases ACAD10 and ACAD11 . In our experiments , both peroxisomal and mitochondrial beta-oxidation were necessary for robust growth on fatty acids and peroxisomal beta-oxidation enzymes had more variable fitness scores between different fatty acids . These observations are consistent with a model of beta-oxidation in which a large ensemble of peroxisomal enzymes shorten diverse long-chain fatty acids in the peroxisome and a smaller ensemble of enzymes metabolize short-chain fatty acids in the mitochondria . Our results demonstrate how a barcoded insertion library can accelerate discrimination of function between closely related members of a diversified gene family . Fitness assays on a much larger panel of substrates should yield further insights into the individual functions of R . toruloides’ diverse complement of peroxisomal enzymes and guide experimental design for their biochemical characterization . While pooled fitness experiments have been used extensively to identify novel gene function , work so far has primarily focused on growth-based phenotypes , with only limited exploration of other phenotypes ( Sliva et al . , 2016; Hassan et al . , 2016; Tyo et al . , 2009 ) . In this study we used two proven strategies for differentiating between cells with altered lipid accumulation , buoyant density centrifugation ( Eroglu and Melis , 2009; Kamisaka et al . , 2006; Liu et al . , 2015 ) and FACS ( Terashima et al . , 2015; Xie et al . , 2014 ) , and applied them to our barcoded mutant pool . Inconsistencies between the two assays and with respect to independent BODIPY staining of targeted deletion strains suggests significant false positive rates for each assay in isolation . When both assays were in agreement , however , 18 of 21 deletion mutants had the expected phenotype in independent experiments . This approach identified 150 high confidence candidate genes with strong impacts on lipid accumulation under nitrogen limitation . While this set is likely incomplete , it complements previous transcriptional and proteomic studies to establish critical genes and cellular processes supporting lipid accumulation that deserve more intensive study . As has been noted in previous functional screens ( Smith et al . , 2006 ) , there was limited overlap between genes for which mutants had a detectable lipid accumulation phenotype in our study and genes with altered protein abundance in R . toruloides during lipid accumulation ( Zhu et al . , 2012 ) ( 14 genes ) or genes that co-purified with R . toruloides lipid droplets ( five genes ) ( Zhu et al . , 2015 ) . The different ensembles of genes identified by each technique illustrate that these systems-level approaches complement each other . Proteomic , transcriptomic , mutagenic and over-expression surveys of lipid metabolism have been carried out in several model eukaryotic systems including S . cerevisiae ( Bozaquel-Morais et al . , 2010; Fei et al . , 2008; Szymanski et al . , 2007; Grillitsch et al . , 2011; Fei et al . , 2011; Ruggles et al . , 2014; Currie et al . , 2014; Bouchez et al . , 2015 ) , C . elegans ( Ashrafi et al . , 2003; Zhang et al . , 2010; Liu et al . , 2014; Lee et al . , 2014; Lapierre et al . , 2011 ) , D . melanogaster ( Cermelli et al . , 2006; Guo et al . , 2008; Beller et al . , 2006; Beller et al . , 2008; Krahmer et al . , 2013b ) , various mammalian cell lines ( Zehmer et al . , 2009; Nishino et al . , 2008; Tu et al . , 2009 ) , and Y . lipolytica ( Athenstaedt et al . , 2006; Pomraning et al . , 2017; Silverman et al . , 2016 ) ( see Supplementary file 5 for a summary of genes identified in 35 studies ) . These studies employed different analytical techniques and culture conditions , and identified many genes without clear orthologs across the different species used , making a granular meta-analysis extremely difficult . A few broad themes are apparent , however . Protein trafficking and organelle interaction are inextricably linked with lipid body formation , growth and mobilization . Membrane-bound G proteins in the endomembrane network have conserved roles regulating trafficking and cellular morphology in response to metabolic states . A complex network of signaling cascades , protein modifications and transcription factors mediate the transition to lipid accumulation or lipid mobilization . A major output of this regulation is amino acid metabolism . Lipid metabolism and autophagy are deeply linked in a complex manner . Our findings were consistent with these general themes , including some orthologs to genes identified in the studies above , but the importance of general functions was more conserved across species than the roles of specific orthologous gene sets . The genes and processes we identify here should be considered in any strategy to optimize lipid metabolism in R . toruloides specifically or oleaginous yeasts in general . Comparative study of these processes across diverse species in standardized conditions will likely be required to uncover which aspects are fundamental to lipid droplet accumulation , maintenance and variation , and which processes are integrated by specific regulatory circuits in a given organism . See Appendix 1 for a deeper discussion of the individual genes for which mutants had altered lipid accumulation in our experiments and how those observations relate to previous work . In this study , we identified 46 R . toruloides genes with no functional predictions ( Supplementary file 1 ) , but which had important functions in lipid metabolism as evidenced by reduced fitness when grown on fatty acids or altered lipid accumulation . These included nine genes with broad conservation across ascomycete and basidiomycete fungi and seven genes with conservation across several basidiomycete species . These genes are of particular interest for further study into their specific functions in lipid metabolism . Moreover , the mutant pool generated in this study should be an excellent tool to assign functions for uncharacterized R . toruloides genes . Cofitness analysis is a particularly powerful method for uncovering the function of novel genes in pathways and processes for which one or more well-characterized genes is also required ( Hillenmeyer et al . , 2010 ) . Closely interacting genes exhibit strongly correlated fitness scores across large panels of diverse conditions . Because the T-DNA insertions in the mutant pool are barcoded , fitness experiments are inherently scalable to a large number of conditions . Because the analytical methods we employed maximize portability and scalability across large compendiums of experiments ( Wetmore et al . , 2015 ) , individual experiments can be conducted at different times under specialized culture conditions , at different scales , and even by different laboratories , yet the data can be effectively compared , maximizing the power of cofitness analysis . We encourage the R . toruloides community and the broader fungal community to make use of this new resource and collaborate with us to maximize its potential . In conclusion , we believe that RB-TDNAseq holds great promise for rapid exploration of gene function in diverse fungi . Because ATMT has been demonstrated in numerous , diverse fungi , we expect this method will be portable to many non-model species . Because the fitness analysis is inherently scalable , it will enable rapid fitness analysis over large compendia of conditions . Cofitness analysis of such compendia will accelerate the annotation of new genomes and identify new classes of genes not abundant in established model fungi . In this study , we demonstrated the application of RB-TDNAseq to the study of lipid metabolism in an oleaginous yeast that has significant potential to become a new model system for both applied and fundamental applications . We identified a large set of genes from a wide array of subcellular functions and compartments that impact lipid catabolism and accumulation . These processes and genes must be considered and addressed in any metabolic engineering strategy to optimize lipid metabolism in R . toruloides and other oleaginous yeasts . Deeper understanding of the extreme cell-to-cell variation in lipid accumulation seen across eukaryotes will likely require deeper mechanistic understanding of these processes and their interaction with the lipid droplet . The principles learned from exploring lipid metabolism and storage across diverse eukaryotes will inform biotechnological innovations for the production of biofuels and bioproducts , as well as new therapies for metabolic disorders .
We used R . toruloides IFO 0880 ( also called NBRC 0880 , obtained from Biological Resource Center , NITE ( NBRC ) , Japan ) as the starting strain for all subsequent manipulations . We used Agrobacterium tumefaciens EHA 105 and plasmids derived from pGI2 ( Abbott et al . , 2013 ) for A . tumefaciens-mediated transformation ( ATMT ) of R . toruloides ( strain and plasmid kindly provided by Chris Rao , UIUC ) . The barcoded mutant pool was constructed by ATMT . We made all gene deletions in a non-homologous end-joining deficient YKU70∆ background ( Zhang et al . , 2016b ) by homologous recombination of a nourseothricin resistance cassette introduced by either ATMT or electroporation of a PCR product . For deletions made by ATMT we used flanking arms of ~1000–1500 bp for homologous recombination . We found that as few as 40 bp of flanking sequence were sufficient for homologous recombination of PCR products at many loci . All strains used in this study , and primers used for strain construction and verification are listed in Supplementary file 4 . For most experiments , we used optical density ( OD ) as measured by absorbance at 600 nm on a GENESYS 20 spectrophotometer ( Thermo Fisher Scientific , 4001–000 , Waltham , MA ) as a metric for growth and to control inoculation density . For IFO 0880 grown in rich media , 1 OD unit represents approximately 30 million cells/mL . Unless otherwise noted , cultures were grown at 30°C in 100 mL liquid media in 250 mL baffled flasks ( Kimble Chase , 25630250 , Vineland , New Jersey ) with 250 rpm shaking on a New Brunswick Innova 2300 platform shaker ( Eppendorf , M1191-0000 , Hauppauge , New York ) with constant illumination using a LUMAPRO 6W LED lamp ( Grainger , 33L570 , San Leandro , CA ) . We used yeast-peptone-dextrose ( YPD ) media ( BD Biosciences , BD242820 , San Jose , CA ) for general strain maintenance and rich media conditions . For auxotrophy experiments we used 0 . 67% w/v yeast nitrogen base ( YNB ) w/o amino acids ( BD Biosciences , BD291940 ) with 111 mM glucose ( Sigma-Aldrich , G7528 , St . Louis , MO ) as our defined media and supplemented with 75 mM L-methionine ( Sigma-Aldrich , M9625 ) , 75 mM L-arginine ( Sigma-Aldrich , A5006 ) , or 0 . 2% w/v drop-out mix complete ( DOC ) , which contains all 20 amino acids , adenine , uracil , p-aminobenzoic acid , and inositol ( US Biological , D9515 , Salem , MA ) . To test growth and fitness on oleic acid ( Sigma-Aldrich , O1008 and 364525 ) , ricinoleic acid ( Sigma-Aldrich , R7257 ) , and methyl ricinoleic acid ( Sigma-Aldrich , R8750 ) , we used this same defined media formulation with 1% fatty acid ( by volume ) instead of glucose . For lipid accumulation experiments , we pre-cultured strains for two generations in YPD ( OD 0 . 2 to OD 0 . 8 ) then washed them twice and resuspended them at OD 0 . 1 in low nitrogen medium; 0 . 17% w/v yeast nitrogen base ( YNB ) w/o amino acids or ammonium sulfate ( BD Biosciences , BD233520 ) , 166 mM D-glucose , 7 mM NH4Cl ( Thermo Fisher Scientific , S25168A ) , 25 mM KH2PO4 ( Thermo Fisher Scientific , P285-3 ) , and 25 mM Na2HPO4 ( Sigma-Aldrich , S0876 ) . This is the C:N 120 formulation from Nicaud et al . ( Nicaud et al . , 2014 ) . Unless otherwise specified , cultures were harvested for lipid quantification or fractionation after 40 hr of growth and lipid accumulation . In all experiments biological replicates refer to samples from independent cultures in the experimental condition . Biological replicates processed on the same day were usually inoculated from the same YPD pre-culture , except for BarSeq experiments . For BarSeq experiments we seeded independent starter cultures in YPD and collected a ‘Time 0’ reference sample after two generations . In downstream fitness or enrichment analysis , we explicitly paired each sample from an experimental condition with the Time 0 sample from the starter culture replicate from which it was seeded . To generate an improved genome assembly for IFO 0880 we prepared genomic DNA for PacBio RS II sequencing ( Pacific Biosciences , Menlo Park , CA ) . Genomic DNA was purified using a two-step protocol , first using glass bead lysis and phenol-chloroform extraction , as previously described ( Zhang et al . , 2016a ) , followed by a QIAGEN Genomic-tip 100/G method ( QIAGEN , 10243 , Germantown , MD ) . All QIAGEN buffers were obtained from a Genomic DNA Buffer Set ( QIAGEN , 19060 ) . Briefly , the dry genomic DNA pellet was first resuspended in G2 buffer supplemented with 200 µg/mL RNase A ( QIAGEN , 19101 ) and 13 . 5 mAU/ml Proteinase K ( QIAGEN , 19131 ) , incubated at 50°C for one hour , and then loaded on a Tip-100 column . After three washes with QC buffer and elution with QF buffer , the DNA was precipitated with isopropanol and removed by spooling using a glass Pasteur pipet . The genomic DNA was washed with 70% ethanol and after air-drying , resuspended in EB buffer ( pH 7 . 5 ) . DNA concentration was determined using a Qubit 3 . 0 fluorometer ( Thermo Fisher Scientific , Q33218 ) and submitted to University of Maryland Genomics Resource Center for library preparation and sequencing . A 10 kb insert , size selected ( BluePippin , Sage Science , Beverly , MA ) SMRTbell library was prepared and sequenced on a PacBio RS II platform using P4C2 chemistry and 10 SMRT cells . De novo assembly of 610 , 663 polymerase reads ( mean subread length of 5 , 193 bp ) was performed using SMRT Analysis version 2 . 3 . 0 . 140936 ( http://www . pacb . com/support/software-downloads/ ) and the RS_HGAP_Assembly . 3 protocol ( HGAP3 ) using default settings except for a genome size of 20 , 000 , 000 bp . The final assembly contained 30 polished contigs ( mean coverage of 131-fold ) with a total genome size of 20 , 810 , 536 bp . Paired-end Illumina data ( 17 , 817 , 326 PE100 reads , [Zhang et al . , 2016a] ) was used for error correction using Pilon version 1 . 13 ( https://github . com/broadinstitute/pilon ) . As expected , the most common type of correction ( 569 in total ) was insertion or deletion of a nucleotide in homopolymer regions . The final error corrected scaffolds were annotated by JGI and submitted to Genbank under the accession LCTV02000000 . Raw sequence data ( PacBio and Illumina ) has been deposited in the NCBI SRA ( SRP114401 and SRP058059 , respectively ) . To harvest RNA for improved gene model prediction , we inoculated R . toruloides into 50 mL cultures in M9 Minimal Salts Solution ( BD Biosciences , BD248510 ) , 2 mM MgSO4 ( Sigma-Aldrich , M7506 ) , 100 µM CaCl2 ( Sigma-Aldrich , C5670 ) , and Yeast Trace Elements Solution ( 88 µg/mL nitrilotriacetic acid , 175 µg/mL MgSO4 7H2O , 29 µg/mL MnSO4 H2O , 59 µg/mL NaCl , 4 µg/mL FeCl2 , 6 µg/mL CoSO4 , 6 µg/mL CaCl2 2H2O , 6 µg/mL ZnSO4 7H2O , 0 . 6 µg/mL CuSO4 5H2O , 0 . 6 µg/mL KAl ( SO4 ) 2 12H2O , 6 µg/mL H3BO3 , 0 . 6 µg/mL Na2MoO4 H2O ) , pH 7 . 0 , with 2% glucose ( Sigma-Aldrich , D9434 ) or 10 mM p-coumaric acid ( trans-4-hydroxycinnamic acid; Alfa Aesar , A15167 , Tewksbury , MA ) , and incubated overnight at 30°C with 200 rpm shaking . We harvested cultures at mid-log phase , centrifuged at 3 , 000 RCF for 10 min at room temperature , removed the supernatant and flash-froze the cell pellet in an ethanol/dry ice bath and stored at −80°C . We lyophilized pellets overnight in a FreeZone-12 freeze dry system ( Labconco , 7754030 , Kansas City , MO ) and extracted total RNA with a Maxwell RSC Plant RNA Kit ( Promega , AS1500 , Madison , WI ) using a Maxwell RSC instrument ( Promega , AS4500 ) . RNA was sequenced and mapped to the R . toruloides IFO 0880 genome at the Department of Energy Joint Genome Institute ( JGI ) in Walnut Creek , CA with in-house protocols . The improved genome assembly was annotated using the JGI Annotation pipeline ( Grigoriev et al . , 2014 ) . Owing to relatively small intergenic spacing in the R . toruloides genome , fused gene models were a common problem . We hand curated over 500 gene models by searching for homology to unrelated proteins at each end of the automated gene models and inspecting agreement with assembled transcripts from our RNAseq experiments . Briefly , for all protein models over 400 amino acids long , we used the N-terminal and C-terminal 30% of each sequence in separate BLAST queries ( NCBI BLAST-plus software 2 . 2 . 30 ) to a custom database of proteins from 22 other eukaryotic genomes ( see Orthology relationships , below ) . We then compared the significant alignments for each terminus of a given gene and scored them for disagreement in regards to the respective orthology groups to which each target sequence belonged with a custom Python script ( Coradetti , 2018a; copy archived at https://github . com/elifesciences-publications/fusedgenemodels ) . The top-scoring 500 gene models were manually inspected for uncharacteristically long introns and for predicted introns and exons not supported by RNAseq reads and modified as required using the Mycocosm genome browser . The current genome annotation is publicly available at the JGI Mycocosm web portal ( Grigoriev et al . , 2014 ) : http://genome . jgi . doe . gov/Rhoto_IFO0880_4 We predicted orthologous proteins for our R . toruloides gene models in H . sapiens , D . melanogaster , C . elegans , A . thaliana , C . reinhartii , S . cerevisiae , and 16 other fungi with the orthomcl software suite version 2 . 0 . 9 ( Li et al . , 2003 ) . See Supplementary file 1 for a full list of ortholog groups and details on the genomes used in this analysis . To efficiently construct a large and diverse mutant pool of barcoded mutants we first constructed a large library of barcoded vectors with an optimized Type IIS endonuclease cloning strategy ( Engler et al . , 2008 ) . We modified the ATMT vector pGI2 ( Abbott et al . , 2013 ) to act as a barcode receiving vector by first removing the two pGI2 SapI sites already present on the vector backbone through SapI restriction digestion , treatment with T4 DNA polymerase for blunt end formation and subsequent blunt end ligation . Next , we introduced two divergent SapI recognition sites just inside the right border of the T-DNA ( vector pDP11 ) as the integration site for random barcoding . We added the barcodes by synthesizing the oligonucleotide GATGTCCACGAGGTCTCTNNNNNNNNNNNNNNNNNNNNCGTACGCTGCAGGTCGAC and amplifying with primers TCACACAAGTTTGTACAAAAAAGCAGGCTGGAGCTCGGCTCTTCGCCCGATGTCCACGAGGTCTCT and CTCAACCACTTTGTACAAGAAAGCTGGGTGGATCCGCTCTTCAATTGTCGACCTGCAGCGTACG . We then combined 4 μg of vector and 140 ng of barcode fragments in a 50 µl reaction with 5 µl 10x T4 ligase buffer , 5 µl 10x NEB CutSmart buffer ( NEB , B7204S , Ipswich , MA ) , 2 . 5 µl T7 ligase ( NEB , M0318L ) , and 2 . 5 µl of SapI ( NEB , R0569S ) . We incubated the reaction at 37°C for 5 min , then 25 cycles of 37°C for 2 min and 20°C for 5 min , before denaturing the enzymes for 10 min at 65°C . Without cooling the product , we added 1 µl SapI and incubated for 30 min at 37°C to digest any uncut vector , then cooled to 10°C . We purified the barcoded plasmids using a Zymo DNA clean and concentrator kit ( Zymo Research , D4014 , Irvine , CA ) , eluting in 15 µl of elution buffer and pooled 10 barcoding reactions . We then transformed E . coli electrocompetent 10-beta cells ( NEB , C3019I ) according to the manufacturers specifications in 30 independent transformations . We estimated the diversity of the barcoded vector pool by performing barcode sequencing as described below , sequencing on an Illumina MiSeq system and estimating the true pool size by the relative proportion of barcodes with 1 or 2 counts . See the script Multicodes . pl from Wetmore et al . ( Wetmore et al . , 2015 ) for details . This yielded a barcoded pool estimated to consist of ~100 million clones . We transformed the barcoded vector pool into A . tumefaciens EHA 105 with a protocol adapted from established methods ( Mersereau et al . , 1990 ) . We diluted a stationary phase starter culture 1:100 in 500 ml Luria-Bertani broth ( BD Biosciences , BD244620 ) and cultured for 6 hr at 30°C . We pelleted cells at 3 , 000 RCF for 10 min at 4°C , washed pellets in ice-cold 1 mM HEPES ( Thermo Fisher Scientific , BP310 ) , pH 7 . 0 , then washed them in ice-cold 10% glycerol 1 mM HEPES , suspended cells in 5 ml ice-cold 10% glycerol 1 mM HEPES , and flash froze 50 µl aliquots in liquid nitrogen . To produce a large transformant pool of A . tumefaciens bearing millions of unique barcode sequences , we electroporated 5 ml of competent cells with 50 µg of plasmid DNA ( 50 µl per well ) in a HT100 96-well plate chamber ( BTX , 45-0400 , Holliston , MA ) with a 2 . 5 kV pulse , 400 ohm resistance and 25 µF capacitance from an ECM 630 wave generator ( BTX , 45-0051 ) . We recovered cells in LB for 2 hr at 30°C , and plated on LB agar with 50 µg/ml kanamycin ( Sigma-Aldrich , K4000 ) . Approximately 14 million transformation events were scraped and collected into a mixed pool for transformation of R . toruloides . We grew the barcoded A . tumefaciens pool to OD 1 in 50 mL YPD in a baffled flask at 30°C , then pelleted the cells and suspended in 10 mL induction medium ( 1 g/L NH4Cl , 300 mg/L MgSO4 7H2O , 150 mg/L KCl ( Thermo Fisher Scientific , P267-500 ) , 10 mg/L CaCl2 ( VWR , 0556 , Radnor , PA ) , 750 µg/L FeSO4 7H2O ( Thermo Fisher Scientific , AC423731000 ) , 144 mg/L K2HPO4 ( VWR , 0705 ) , 48 mg/L NaH2PO4 ( Thermo Fisher Scientific , BP329 ) , 2 g/L D-Glucose , 10 mg/L thiamine ( Sigma-Aldrich , T4625 ) , 20 mg/L acetosyringone ( Sigma-Aldrich , D134406 ) , and 3 . 9 g/L MES ( Sigma-Aldrich , 69892 ) , adjusted to pH 5 . 5 with KOH ) and incubated 24 hr at room temperature in culture tubes on a roller drum . We cultured R . toruloides in 10 mL YPD to OD 0 . 8 , then pelleted the cells and suspended in the induced A . tumefaciens culture for 5 min at room temperature . We filtered the mixed culture on a 0 . 45 µm membrane filter ( EMD Millipore , HAWP04700 , Bedford , MA ) then transferred the filter to induction media 2% agar ( BD Biosciences , BD214010 ) plates for incubation at 26°C for 4 days . We then washed the filters in YPD and plated on YPD 2% agar with 300 µg/ml cefotaxime ( Sigma-Aldrich , C7039 ) and 300 µg/ml carbenicillin ( Sigma-Aldrich , C1389 ) and incubated at 30°C for two days . We scraped these plates to collect transformed R . toruloides , recovered the mutant pool in YPD plus cefotaxime and carbenicillin for 24 hr , added glycerol to 15% by volume and stored at −80°C . We repeated this protocol 40 times to recover approximately 2 million transformation events . In some rounds of transformation , we also included 0 . 05% casamino acids ( BD Biosciences , BD223120 ) or 1% CD lipid concentrate ( Thermo Fisher Scientific , 11905–031 ) in the induction media plates to promote recovery of mutants with impaired amino acid or lipid biosynthesis . We then recovered each of these transformation subpools on YPD plus cefotaxime and carbenicillin 12 hr to clear residual A . tumefaciens and combined them into one master pool , divided it into 1 ml aliquots in YPD 15% glycerol and stored them at −80°C . Laboratories with an interest in experimenting with this mutant pool should contact the corresponding authors . To isolate high quality genomic DNA we harvested ~108 cells from a fresh YPD culture of the mutant pool , washed the pellet in water and suspended in 200 µl TSENT buffer ( 2% Triton X-100 ( Sigma-Aldrich , T8787 ) , 1% SDS ( Thermo Fisher Scientific , AM9820 ) , 1 mM EDTA ( Sigma-Aldrich , ED2SS ) , 100 mM NaCl ( Sigma-Aldrich , S5150 ) , 10 mM Tris-HCl , pH 8 . 0 ( Invitrogen , 15568–025 , Carlsbad CA ) ) . We then added the sample to 200 µl 25:24:1 phenol/chloroform/isoamyl alcohol ( Invitrogen , 15593–031 ) in screw-top tubes with glass beads ( Sigma-Aldrich , Z763748 ) on ice and vortexed for 10 min at 4°C . We added 200 µl TE buffer ( Thermo Fisher Scientific , AM9858 ) , centrifuged at 21 , 000 RCF for 20 min at 4°C , removed the aqueous phase to 1 mL 200 Proof ethanol ( Koptec , V1016 , King of Prussia , PA ) and centrifuged at 21 , 000 RCF for 20 min at 4°C to pellet DNA . DNA was dried and suspended in 200 µl TE , treated with 0 . 5 µl RNase A ( Qiagen , 19101 ) , then purified with a Genomic DNA Clean and Concentrator Kit ( Zymo Research , D4064 ) . We checked DNA quality on a 0 . 8% agarose E-Gel ( Thermo Fisher Scientific , G51808 ) and quantified with a Qubit 3 . 0 fluorometer using the dsDNA HS reagent ( Invitrogen , 1799096 ) . To sequence sites of genomic insertions we followed the TnSeq protocol of Wetmore et al . ( Wetmore et al . , 2015 ) , using their Nspacer_barseq_universal primer and P7_MOD_TS_index primers for final amplification ( Supplementary file 4 ) . Because we found a high proportion of non-specific products in our TnSeq mapping and highly variable recovery of the same insertions between technical replicates , we sequenced multiple replicates for each batch of ATMT mutants ( around 10 , 000–100 , 000 mutants per batch ) and used at least two annealing temperatures for the final PCR enrichment for each batch . In total , we sequenced about 900 million reads from 64 independent TnSeq libraries . A full summary of TnSeq libraries used to map the mutant pool is listed in Supplementary file 4 . Libraries were submitted for single-end 150 bp Illumina sequencing on a HiSeq 2500 platform at the UC Berkeley Vincent J . Coates Genomics Sequencing Laboratory , except for a subset of smaller runs on an Illumina MiSeq platform as indicated in Supplementary file 4 . Sequence data have been submitted to the NCBI Short Read Archive ( SRP116146 ) . We used a similar strategy as Wetmore et al . ( Wetmore et al . , 2015 ) to map the location of each barcoded T-DNA insertion , with minor alterations ( Coradetti , 2018b ) . MapTnSeq_trimmed . pl processes the TnSeq reads to identify the barcode sequence and is a modified version of MapTnSeq . pl ( Wetmore et al . , 2015 ) , with three minor alterations . We ignore the last 10 bases of the T-DNA sequence , as the length of T-DNA border sequence included in the final insertion is variable . We also allow for barcode sequences of 17–23 base pairs instead of exactly 20 . We relaxed this restriction because on manual inspection of our TnSeq data we found that approximately 10% of barcodes appeared to be slightly shorter or longer than 20 base pairs , likely a result of imperfect PAGE purification after oligonucleotide synthesis . We report all TnSeq reads in which sequence past the end of the expected T-DNA insert aligns with other regions of the T-DNA sequence , or with the outside vector as ‘past end’ reads . These are mappings of junctions between concatemeric T-DNA inserts and unprocessed T-DNA vectors , respectively . RandomPoolConcatemers . py is a custom script that associates barcode sequences mapped in MapTnSeq_trimmed . pl with genomic locations and then filters those barcodes for insertions at unique , unambiguous locations . First , for all barcodes sequenced , the number of reads mapping to any genomic location and the number of reads mapping to concatemeric junctions are tabulated . Any barcodes that only differ by a single base pair from a barcode with 100 times more reads are removed as likely sequencing errors and reported as ‘off by one’ barcodes . Any barcode for which there are more than seven times as many ‘past end’ reads as reads mapping to genomic locations as ‘past-end’ barcodes . The past-end barcodes are further characterized as ‘head-to-tail’ concatemers ( majority of Tnseq reads map to the left border T-DNA sequence ) , ‘head-to-head’ concatemers ( majority of the reads map to the right border T-DNA sequence ) , or ‘Run-on’ insertions ( majority of reads map to pGI2 outside the T-DNA sequence ) . Any barcodes for which the majority of TnSeq reads map ambiguously to the genome are removed and reported as ambiguous barcodes . Any barcodes for which 20% or more of the TnSeq reads map to a different location than the most commonly observed location are removed and reported as ‘multilocus’ barcodes . Finally , any barcodes mapped within 10 bases of a more abundant barcode for which there is a Levenshtein edit distance ( Levenshtein , 1966 ) less than five are removed as likely sequencing errors and reported as ‘off by two’ barcodes . The remaining unfiltered barcodes are reported as the mutant pool . InsertionLocationJGI . py is a custom script to match the genomic locations of barcodes in the mutant pool to the nearest gene in the current JGI R . toruloides gene catalog and report whether the insertion is in a 5-prime intergenic region , a 5-prime UTR , an exon , an intron , a 3-prime UTR , or a 3-prime intergenic region of that gene . InsertBias . py is a custom script to analyze potential biases in T-DNA insertion rates . The script tracks number of insertions versus scaffold length for all scaffolds in the genome , GC content in the local regions of insertion , and insertion rates in promoter regions , 5-prime untranslated mRNA , exons , introns , 3-prime untranslated mRNA , and terminator regions . To assess fine-scale biases in insertion locations , all locations in the genome are apportioned to one of the above feature types , then for each feature type , the same number of insertions as were observed for that feature type in the mutant pool are sampled at random ( without replacement ) from all the genomic locations assigned to that feature type . We isolated genomic DNA with a Fungal/Bacterial DNA MiniPrep kit ( Zymo Research , D6005 ) . We used Q5 high-fidelity polymerase with GC-enhancer ( NEB , M0491S ) to amplify unique barcode sequences flanked by specific priming sites , yielding a 185 bp Illumina-sequencing-ready product ( Figure 1—figure supplement 1 ) . We used BarSeq primers from Wetmore et al . ( de Hoon et al . , 2004 ) ( Supplementary file 4 ) , except we replaced primer P1 with a mix of primers with 2–4 random bases to improve nucleotide balance for optimal sequencing of low-diversity sequences ( Illumina , 2013 ) . We cleaned PCR products with a DNA clean and concentrator kit ( Zymo Research , D4014 ) . We quantified product yield with a Qubit 3 . 0 fluorometer system and mixed as appropriate for sequencing as multiplexed libraries . We sequenced libraries on an Illumina HiSeq 4000 system at the UC Berkeley Vincent J . Coates Genomics Sequencing Laboratory . If necessary , libraries were purified further with a Pippin Prep system ( Sage Biosciences ) before loading with 15% PhiX DNA as a phasing control for low diversity samples ( Illumina , 2013 ) . We sequenced each biological replicate to a depth of at least 20 million reads . We counted occurrences of T-DNA barcodes in each sample with the script MultiCodes_Variable_Length . pl , a modified version of MultiCodes . pl from Wetmore et al . ( Wetmore et al . , 2015 ) that allows for barcodes of 17–23 base pairs . For all BarSeq experiments , we thawed frozen aliquots of the mutant pool on ice and inoculated them into YPD at OD 0 . 2 . Cultures were recovered for about 12 hr until OD 600 was approximately 0 . 8 . Cultures were pelleted at 3 , 000 RCF for 5 min , washed twice in the appropriate media , and transferred to the condition of interest . Samples were taken from the YPD starter cultures ( Time 0 ) and after 5–7 doublings in the experimental condition . Average fitness scores and T-like statistics as metrics for consistency between individual insertion mutants in each gene were calculated with the scripts combineBarSeq . pl and FEBA . R from Wetmore et al . ( Wetmore et al . , 2015 ) . Briefly , for each biological replicate and condition , for any barcode with an average of at least three counts in Time 0 samples , a strain fitness score is calculated as Fstrain = log2 ( Ccondition +sqrt ( P ) ) – log2 ( CTime0 +1/sqrt ( P ) ) , where C is the raw counts for the barcode and P is a gene-specific ‘pseudocount’ added to reduce noise in fitness scores for low-count strains . These strain fitness scores are then normalized such that the median score is 0 to correct for coverage differences between the samples . The strain fitness scores are then assigned a weight proportional to the harmonic mean of counts at Time 0 and in the condition sample . For any one barcode , the weighting mean is capped at 20 reads , which has the effect of limiting the influence of generally more abundant outlier strains ( Wetmore et al . , 2015 ) . T is calculated as the gene fitness divided by the square root of the variance in strain fitness scores . This variance is estimated as the maximum value of a naïve estimate based on Poisson noise or the observed variance ( a weighted sum squares of differences in strain fitness versus gene fitness scores plus an estimate of global variance in gene fitness scores calculated by comparing fitness scores in the first and second half of every gene ) . See the methods subsection ‘BarSeq data analysis and calculation of gene fitness’ in the original publication by Wetmore et al . ( Wetmore et al . , 2015 ) for more detail on these algorithms . Wetmore et al . limited their analysis to genes with an average of at least 30 total counts at Time 0 , spread across three strains . Because the list of genes satisfying this requirement can change from experiment to experiment , we established a list of genes that met this requirement in any of our experiments and used that list for our analysis . As a result , a minority of genes ( 649 ) have fitness scores based on data from one or two barcodes . The number of barcodes used in fitness analysis of each gene is listed in all relevant tables in Supplementary file 2 . In general , genes with data from only one or two barcodes had smaller T-statistics and thus were filtered out in later analyses . Because Wetmore et al . ’s software does not consider biological replication between independent cultures , we then averaged fitness scores for each condition and combined T-statistics across replicates with the script AverageReplicates . py , treating them as true T-statistics . That is: Tcondition = Sum ( Treplicates ) /Sqrt ( Nreplicates ) . To assess consistency of differences in observed fitness between growth conditions we computed Tc1 – c2 = ( Fc1 – Fc2 ) /Sqrt ( ( Fc1/ Tc1 ) 2 + ( ( Fc2/ Tc2 ) 2 ) with the script ResultsSummary . py . We generated K-means clusters of fitness scores using Pearson correlation as the similarity metric using Cluster 3 . 0 ( de Hoon et al . , 2004 ) . For comparing enrichment in density and FACS separated fractions we computed F and T for each fraction versus the T0 control . The enrichment score E and T between fractions was then calculated as E = Fhigh lipid – Flow lipid and Thigh lipid – low lipid = ( Fhigh lipid – Flow lipid ) /Sqrt ( ( Fhigh lipid/ Thigh lipid ) 2 + ( ( Flow lipid/ Tlow lipid ) 2 ) with the script ResultsSummary . py . We generated hierarchical clusters of enrichment scores using Pearson correlation as the similarity metric and average linkage as the clustering method . All fitness data are available in Supplementary file 2 and the fitness browser ( http://fungalfit . genomics . lbl . gov/ ) . Custom Python scripts are available at ( Coradetti , 2018b; copy archived at https://github . com/elifesciences-publications/rb-tdnaseq ) . Sequence data have been submitted to the NCBI Short Read Archive ( SRP116193 ) We cultured R . toruloides overnight in 10 mL YPD on a roller drum to an OD 600 of 2 , then pelleted cells at 3 , 000 RCF for 5 min at 4°C in a benchtop centrifuge ( Eppendorf , 5810 R ) . Cells were kept at 4°C from this point . We transferred the pellets to 1 . 5 mL tubes and washed them four times with ice cold 0 . 75 M D-sorbitol ( Sigma-Aldrich , S1876 ) , centrifuging each wash 30 s at 8 , 000 RCF , 4°C ( Eppendorf , 5424 ) . After the final wash , we removed excess D-sorbitol and added 35 µl of cell pellet to 10 µl of fresh 0 . 75 M D-sorbitol and ~1 µg of PCR product in 5 µl water in a chilled 0 . 1 cm cuvette . We electroporated cells at 1 . 5 kV , 200 ohms and 25 µF with an ECM 630 ( BTX ) electroporation system . We then added 1 mL cold 1:1 mixture of YPD and 0 . 75 M D-sorbitol and transferred to 14 mL round bottom culture tubes for a 3 hr recovery culture at 30°C with shaking at 200 rpm on a platform shaker . We then pelleted the cultures at 8 , 000 RCF for 30 s , suspended in 200 µl YPD , and then plated on YPD with 100 µg/mL nourseothricin ( 5 . 005 . 000 , Werner Bioagents , Germany ) . We scored enrichment of gene ontology terms with a custom script that performs a hypergeometric test on the frequency of each term in the genome versus the frequency in given gene set ( script GOenrich . py , available at [Coradetti , 2018b] ) . We corrected for multiple hypothesis testing with the Benjamini-Hochberg correction ( Benjamini and Hochberg , 1995 ) . We extended the GO terms associated with R . toruloides genes in the current JGI annotation by collecting terms for orthologous genes in Arabidopsis thaliana , Aspergillus nidulans , Caenorhabditis elegans , Candida albicans , Homo sapiens , Mus musculus , and Saccharomyces cerevisiae , obtained from the Gene Ontology Consortium ( Ashburner et al . , 2000; Gene Ontology Consortium , 2015 ) . Cell lysis , extraction of total lipids , and conversion to fatty acid methyl esters ( FAMEs ) was based on a published protocol ( Browse et al . , 1986 ) . We cultured IFO 0880 , a selection of seven targeted deletion strains ( see Supplementary file 6 ) and one overexpression strain ( RT880-AD , [Zhang et al . , 2016a] ) in low nitrogen medium for 48 or 96 hr . We collected paired 5 mL samples from each in screw-top glass tubes ( Corning , 99502–10 , Corning , NY ) and 15 mL polyethylene tubes ( Corning , 352096 ) for lipid extraction and mass determination , respectively . We pelleted samples by centrifugation at 2 , 000 RCF for 20 min at 4°C , and washed once in water to remove salts and unused glucose . We then transferred the mass determination sample to a pre-tared 1 . 5 mL microcentrifuge tube . We froze both samples at −20°C overnight , then lyophilized them 48 hr in a FreeZone freeze dry system ( Labconco , 7754042 ) before weighing/extraction . We added 1 mL methanol spiked with 250 µg methyl tridecanoate to each sample to serve as an internal standard ( ISTD ) . We then resuspended lipid extraction samples ( usually about 10–20 mg ) by vortexing in 3 mL 3N methanolic HCl ( Sigma-Aldrich , 33050-U ) and 200 µl chloroform ( Sigma-Aldrich , 472476 ) and incubated at 80°C water bath for 1 hr . Cell lysis and conversion to FAMEs occurs during this incubation . To extract FAMEs we then added 2 mL hexane ( Sigma-Aldrich , 650552 ) and vortexed samples well before centrifugation at 3 , 000 RCF for 3 min . One µL of the hexane layer was injected in split mode ( 1:10 ) onto a SP-2330 capillary column ( 30 m x 0 . 25 mm x 0 . 2 µm , Sigma-Aldrich , 24019 ) . An Agilent 7890A gas chromatograph equipped with a flame ionization detector ( FID ) was used for analysis with the following settings: Injector temperature 250°C , carrier gas: helium at 1 mL/min , temperature program: 140°C , 3 min isocratic , 10 °C/min to 220°C , 40 °C/min to 240°C , 5 min isocratic . FAME concentrations were calculated by comparing the peak areas in the samples to the peak areas of ten commercially available high-purity standards ( C16:0 , C16:1 , C17:0 , C18:0 , C18:1 , C18:2 , C20:0 , C20:1 , C22:0 , C24:0 ) ( Sigma-Aldrich ) in known concentration relative to the internal standard , respectively . We inoculated deletion mutants and the YKU70∆ parental strain at OD 0 . 1 in low nitrogen medium and cultured for 40 hr . We fixed samples by adding 180 µl cell culture to 20 µl 37% formaldehyde ( Electron Microscopy Sciences , Hatfield , PA ) and incubating for 15 min at room temperature . We then diluted fixed cells 1:100 in 200 µl PBS ( from 10X concentrate , Thermo Fisher Scientific , 70011–44 ) with 0 . 5 M KI and 0 . 25 µg/mL BODIPY 493/503 ( Thermo Fisher Scientific , D-3922 ) , then incubated 30 min at room temperature . We quantified BODIPY signal for 10 , 000 cells per sample on a Guava HT easyCyte system ( EMD Millipore ) in the green channel ( excitation 488 nm , emission 525 nm ) using InCyte software ( EMD Millipore ) . Due to logistical constraints , samples were processed in batches of at most 30 cultures at a time . Each batch included three biological replicates of the YKU70∆ parental strain as an internal reference . Distribution of mutant strains into these batches was not explicitly randomized , but each batch included both strains expected to accumulate more lipid and strains expected to accumulate less lipid than the parent . Each mutant was processed in at least two different batches . We cultured the barcoded mutant pool in low nitrogen medium for 40 hr . We then diluted unfixed cells 1:100 in 10 ml PBS with 0 . 5 M KI and 0 . 25 µg/mL BODIPY 493/503 , then incubated 30 min at 30°C with shaking . We then sorted the population on a Sony SH800 cell sorter with a 70 µM fluidic chip , sorting in semi-purity mode . We first applied a gate for single cell events with forward scatter height within 15% of forward scatter area . We sorted a sample of 10 million cells with the scattering gate alone as a control population , to account for effects of growth , sorting , and collection that are independent of lipid accumulation . Then we collected the 10% of the size-filtered population with the highest and lowest signals in the FITC channel . We collected 10 million cells each for the high and low signal populations . We collected all sorted cells in YPD with 300 µg/ml cefotaxime ( Sigma-Aldrich , C7039 ) and 300 µg/ml carbenicillin ( Sigma-Aldrich , C1389 ) , then grew them to saturation in our standard culture conditions and pelleted 1 mL sample , and then stored at −20°C for BarSeq analysis . We prepared linear sucrose gradients with the method of Luthe et al . ( Luthe , 1983 ) . For example , to prepare a 65–35% sucrose gradient; we prepared four solutions of sucrose ( Sigma-Aldrich , G7528 ) at 65 , 55 , 45 , and 35 grams per 100 mL in PBS , then successively froze 10 mL layers of each concentration in a 50 mL conical tube ( Corning , 430829 ) on dry ice and stored the gradient at −20°C . We selected appropriate gradients to maximize the physical separation of the cell population by running trial experiments with wild type IFO 0880 cultures on a number of sucrose gradients . The gradients used in each experiment are described in Table 3 . Approximately 24 hr before performing density separation on cell population , the appropriate step gradient was moved to 4°C to thaw , yielding a linear gradient ( Luthe , 1983 ) . To perform the separation , we centrifuged 50 mL of culture at 6 , 000 RCF at 4°C for 20 min . We then suspended the pellet in 5 ml PBS at 4°C and carefully loaded it onto a sucrose gradient . We centrifuged the gradients for 1 hr at 5 , 000 RCF at 4°C with slow acceleration and no brake for deceleration in an Avanti J-26 XP centrifuge with a JS5 . 3 swinging bucket rotor ( Beckman Coulter , Brea , CA ) . To collect fractions , we pierced the bottom of each tube with the tip of a 16 gauge needle ( BD Biosciences , 305197 ) , to slowly drain the gradient from the bottom , at 1 drop every 1–5 s . We collected 2 mL fractions , estimated average fraction density by weighing a 100 µl sample and measured the distribution of the cell population across the sample by optical density . The appropriate fractions were then combined to sample the least buoyant ( highest density ) 5–10% , median buoyancy 30–50% , and most buoyant ( lowest density ) 5–10% of the population . For each biological replicate , we also collected a 1 mL sample from the culture before separation to monitor growth in the experimental condition . Cover slips were submerged in 0 . 1% v/v polylysine ( Sigma-Aldrich , P8920 ) for 15 min . Cover slips were removed from polylysine and blotted dry from the bottom of vertically-held slips . Slips were then washed several times with ddH2O and rapidly dried with compressed air . Directly prior to imaging , slips were visually inspected for streaks and dust and softly cleaned with lens paper . Cells were grown 40 hr in low nitrogen medium , 1 mL of culture was transferred to 2 mL microcentrifuge tubes with 1 mL of PBS , and tubes were mixed briefly by vortexing . Cells were pelleted at 9 , 000 RCF for 1 min in a microcentrifuge , and then resuspended in 100 µl of fluorescent staining solution ( PBS with 0 . 5 M KI and 0 . 25 µg/mL BODIPY 493/503 ) to visualize intracellular lipid droplets . Four µl of stained cells were pipetted up and down and transferred to the clean slides . Polylysine-coated cover slips were carefully placed on the 4 µl drop to ensure even spreading of liquid . Cells were observed on an Axio Observer microscope ( Carl Zeiss Microscopy , Thornwood , NY ) with a plan-apochromat 100x DIC objective ( Carl Zeiss Microscopy , 440782-9902-000 ) , ORCA-Flash 4 . 0 camera ( Hamamatsu , C11440-22CU , Japan ) , and ZenPro 2012 ( blue edition ) software . For BODIPY imaging cells were illuminated with an X-cite Series 120 arc-lamp ( EXFO Photonics Solutions , Canada ) and 38HE filter set , 450–490 excitation , 500–550 emission ( Carl Zeiss Microscopy , 489038-9901-000 ) . Zvi files were converted to 16 bit TIFF images and representative fields of view were cropped and channels merged using FIJI image processing software ( Schindelin et al . , 2012 ) . | The fungus Rhodosporidium toruloides can grow on substances extracted from plant matter that is inedible to humans such as corn stalks , wood pulp , and grasses . Under some growth conditions , the fungus can accumulate massive stores of hydrocarbon-rich fats and pigments . A community of scientists and engineers has begun genetically modifying R . toruloides to convert these naturally produced fats and pigments into fuels , chemicals and medicines . These could form sustainable replacements for products made from petroleum or harvested from threatened animal and plant species . Fungi , plants , animals and other eukaryotes store fat in specialized compartments called lipid droplets . The genes that control the metabolism – the production , use and storage – of fat in lipid bodies have been studied in certain eukaryotes , including species of yeast . However , R . toruloides is only distantly related to the most well-studied of these species . This means that we cannot be certain that a gene will play the same role in R . toruloides as in those species . To assemble the most comprehensive list possible of the genes in R . toruloides that affect the production , use , or storage of fat in lipid bodies , Coradetti , Pinel et al . constructed a population of hundreds of thousands of mutant fungal strains , each with its own unique DNA ‘barcode’ . The effects that mutations in over 6 , 000 genes had on growth and fat accumulation in these fungi were measured simultaneously in several experiments . This general approach is not new , but technical limitations had , until now , restricted its use in fungi to a few species . Coradetti , Pinel et al . identified hundreds of genes that affected the ability of R . toruloides to metabolise fat . Many of these genes were related to genes with known roles in fat metabolism in other eukaryotes . Other genes are involved in different cell processes , such as the recycling of waste products in the cell . Their identification adds weight to the view that the links between these cellular processes and fat metabolism are deep and widespread amongst eukaryotes . Finally , some of the genes identified by Coradetti , Pinel et al . are not closely related to any well-studied genes . Further study of these genes could help us to understand why R . toruloides can accumulate much larger amounts of fat than most other fungi . The methods developed by Coradetti , Pinel et al . should be possible to implement in many species of fungi . As a result these techniques may eventually contribute to the development of new treatments for human fungal diseases , the protection of important food crops , and a deeper understanding of the roles various fungi play in the broader ecosystem . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2018 | Functional genomics of lipid metabolism in the oleaginous yeast Rhodosporidium toruloides |
Disentangling the effect on genomic diversity of natural selection from that of demography is notoriously difficult , but necessary to properly reconstruct the history of species . Here , we use high-quality human genomic data to show that purifying selection at linked sites ( i . e . background selection , BGS ) and GC-biased gene conversion ( gBGC ) together affect as much as 95% of the variants of our genome . We find that the magnitude and relative importance of BGS and gBGC are largely determined by variation in recombination rate and base composition . Importantly , synonymous sites and non-transcribed regions are also affected , albeit to different degrees . Their use for demographic inference can lead to strong biases . However , by conditioning on genomic regions with recombination rates above 1 . 5 cM/Mb and mutation types ( C↔G , A↔T ) , we identify a set of SNPs that is mostly unaffected by BGS or gBGC , and that avoids these biases in the reconstruction of human history .
Human genomic diversity has evolved under diverse and complex constraints ( Auton et al . , 2015 ) , such as past demography , selection , mutations , or genomic rearrangements ( Lohmueller et al . , 2011; Schiffels and Durbin , 2014; Sudmant et al . , 2015; Mallick et al . , 2016 ) . However , the influence of these evolutionary forces and their interactions remain to be fully understood . For instance , it is yet unclear which fraction of the genome evolves under positive or purifying selection ( McVicker et al . , 2009; Rands et al . , 2014; Corbett-Detig et al . , 2015 ) . Such information is crucial to our understanding of what portion of the genome is evolving neutrally , and necessary to form a clear basis for demographic inference , the detection of selective events , or the inference of the distribution of fitness effects of new mutations . Genome-wide variation in recombination may strongly affect neutral variants ( Spencer et al . , 2006; Corbett-Detig et al . , 2015 ) , as selection will have more impact on linked polymorphism in regions of low recombination ( Charlesworth et al . , 1995 ) , whereas biased gene conversion , which can also mimic the effect of selection ( Galtier and Duret , 2007; Ratnakumar et al . , 2010 ) , will occur mostly in regions of high recombination ( Katzman et al . , 2011 ) . In humans , various measures of diversity are positively correlated with levels of recombination ( Nachman , 2001; Spencer et al . , 2006; Cai et al . , 2009; Lohmueller et al . , 2011 ) . While a direct mutagenic effect of recombination seems unlikely ( McVicker et al . , 2009; Schaibley et al . , 2013 ) except at CpG sites ( Arbeithuber et al . , 2015 ) , there is still some debate about whether the correlation between diversity and recombination is driven by recurrent selective sweeps ( hitchhiking of neutral and slightly deleterious mutations ) or background selection ( BGS; i . e . purifying selection against deleterious mutations at linked sites ) ( McVicker et al . , 2009; Stephan , 2010; Hernandez et al . , 2011; Lohmueller et al . , 2011 ) . The modeling of genomic diversity under selection in humans suggests that it can be explained entirely by BGS ( Lohmueller et al . , 2011 ) , whereas a combination of both BGS and positive selection seems to best explain genomic diversity in Drosophila ( Elyashiv et al . , 2016 ) . However , the correlation between diversity and recombination is generally relatively weak in humans for most tested statistics and seems restricted to genomic regions of relatively low-recombination rate ( <1 cM/Mb , ( Cai et al . , 2009; Lohmueller et al . , 2011 ) ) . Given the positive relationship between recombination and genetic variability , it has been proposed that the genomic regions most suitable for demographic inferences should be far away from genes and have high-recombination rates ( Lohmueller et al . , 2011 ) . However , regions of high recombination might be prone to GC-biased gene conversion ( gBGC ) , a process by which GC alleles in recombination tracts are preferentially transmitted in GC/AT heterozygotes ( Duret and Galtier , 2009 ) . This process thus increases the frequency of G and C derived alleles ( usually denoted as strong or S alleles , Lachance and Tishkoff , 2014 ) relative to A and T ( denoted as weak or W alleles ) , especially in recombination hotspots ( Spencer et al . , 2006; Glémin et al . , 2015 ) . By modifying allele frequencies in high-recombination regions , gBGC affects the site frequency spectrum ( SFS ) ( Lachance and Tishkoff , 2014; Glémin et al . , 2015 ) such that it becomes right-shifted for W-to-S ( WS ) mutations and left-shifted for S-to-W ( SW ) mutations . In addition , gBGC affects various classical statistics used to detect selection , and WS SNPs show larger levels of population differentiation than other SNPs ( Lachance and Tishkoff , 2014 ) . Overall , gBGC is believed to directly affect only 1% to 2% of the human genome , near recombination hotspots ( Glémin et al . , 2015 ) , but due to the transient nature of these hotspots , a larger fraction of the genome could have been affected in the long term . Here , we use two whole-genome human datasets to determine how and to what extent recombination and selective forces affect genome-wide diversity in humans . We examine the relationship between recombination rate and the average derived allele frequency per individual , as well as the SFS . After determining the parts of the genome that are least affected by BGS and gBGC , we examine the impact of these two processes on the SFS , and how they affect demographic inference based on the SFS .
For individuals belonging to five geographic regions , we studied the variability of DAFi¯ across the genome by computing it separately for SNPs that belong to different recombination classes and averaging it across individuals within each region ( Figure 1A ) . Local recombination rates around each SNP were obtained from the 1000G Yoruba recombination map ( Frazer et al . , 2007 ) ( see Materials and methods ) , but the use of alternative recombination maps leads to similar patterns ( Figure 1—figure supplement 3 ) . We find that the average intra-bin DAFi¯ increases almost log-linearly with the average recombination rate . The exception is for the lowest recombination class , most likely because low-recombination rates are difficult to estimate ( Kong et al . , 2010; Wegmann et al . , 2011 ) . We observe the same log-linear relationship in a set of 20 individuals chosen to represent five continents that were sequenced as part of the Simons Genome Diversity Project ( Mallick et al . , 2016 ) at higher coverage ( 31–60× ) than the 1000G individuals ( Figure 1—figure supplement 4A ) . The log-linear relationship between recombination rate and DAFi¯ is conserved among geographic regions ( Figure 1A ) and it is also observed at the level of single individuals ( Figure 1—figure supplement 5A ) , as expected from our theoretical derivations . Note that this very similar behavior among individuals and populations is not in line with a differential action of positive selection ( selective sweeps ) in different continents . Therefore , if adaptive events were involved in shaping allele frequencies and creating this relation , they should have occurred before the human lineage split into different continental groups . Since most variation in exonic regions has emerged in the last 10 , 000 years ( Fu et al . , 2013 ) , a pure adaptive explanation for this relation seems unlikely . As expected if purifying selection was removing deleterious variation predominantly in coding regions , we find a stronger effect of BGS in transcribed ( TR ) than in non-transcribed ( NTR ) regions , in the sense that DAFi¯ is more reduced in regions of low recombination in TR than in NTR regions ( Figure 1—figure supplement 6A ) . At the same distance from exons ( between ~0 . 001 and~0 . 1 cM , Figure 1—figure supplement 6B ) , DAFi¯ is slightly larger for NTR than for TR regions suggesting that BGS is stronger in TR regions . However , DAFi¯ converges to similar values in high-recombination regions , in line with the view that BGS is not acting in these regions . Interestingly , BGS is clearly acting in NTR regions even when we focus on NTR regions more than 50 kb away from any transcribed region ( Figure 1B ) . This result confirms that BGS is acting in NTRs ( Asthana et al . , 2007; Comeron , 2014; Rands et al . , 2014 ) , which could be either due to the presence of functional elements in these regions such as non-coding RNAs , histone marks , enhancers or insulators ( Kellis et al . , 2014; Bonev and Cavalli , 2016; Van Nostrand et al . , 2017 ) , or due to remote effects of exonic deleterious mutations on SNPs in NTR . However , since the influence of exonic regions on DAFi¯ is largely limited above 0 . 01 cM ( Figure 1—figure supplement 5B ) , we suspect that functionally constrained elements are widespread in NTRs . Conservation scores have also been used to assess a potential effect of selection on DAFi¯ . Sites associated to GERP RS scores between –2 and +2 are thought to be evolving neutrally in mammals ( Davydov et al . , 2010 ) , but we still find a positive log-linear relationship between DAFi¯ and recombination rate for those sites ( Figure 1—figure supplement 5C ) , suggesting that these sites are also influenced by BGS due to selection at linked sites . Note that we also find a positive relationship between DAFi¯ and recombination for more conserved sites that could be directly under negative selection ( Figure 1—figure supplement 5D–E ) suggesting that their diversity is also affected by BGS at neighbouring sites . These observations suggest that filtering by GERP score may not be sufficient to completely remove the effect of BGS . Since DAFi¯ patterns seem to be driven by BGS , we would expect that they are correlated with statistics that have been specifically developed to measure the extent of BGS in various regions of the genome , such as the B-statistic ( McVicker et al . , 2009 ) . Indeed , the B-statistic measures the relative reduction in genetic diversity due to BGS and it ranges from 0 in regions highly affected by BGS to 1 in regions unaffected by BGS . As expected , we find that DAFi¯ and the average B-statistic , computed both in the same 20 recombination rate bins defined in Figure 1A , are highly correlated ( Figure 1—figure supplement 5F ) . This result suggests that the average DAFi¯ and average B-statistic are affected by the same process , and thus that DAFi¯ provides information on the strength of background selection among a set of SNPs . Since the impact of BGS is mediated by recombination , BGS should have a minimal influence in regions of high recombination ( Hudson and Kaplan , 1995; Nordborg et al . , 1996 ) . However , it has been shown that GC biased gene conversion ( gBGC ) is acting in GC/AT heterozygotes in these regions , particularly in the vicinity of recombination hotspots ( Spencer et al . , 2006 ) , potentially increasing the frequency of G and C derived alleles ( usually denoted as strong or S alleles , see Lachance and Tishkoff , 2014 ) as compared to A and T ( denoted as weak or W alleles ) . We have thus examined the relationship between DAFi¯ and local recombination rate for three combinations of S and W alleles ( Figure 1C , Figure 1—figure supplement 6C ) . If the ancestral allele is W and the derived allele is S ( WS sites , Figure 1C , left ) , we see the same log-linear relation between DAFi¯ and recombination as if we consider all SNPs ( Figure 1A ) . However , at SW sites ( Figure 1C , center ) , DAFi¯ decreases for recombination rates above ~1 . 5 cM/Mb . This non-monotonic behavior at SW sites is consistent with gBGC favoring the transmission of G and C alleles , and thus decreasing the frequency of derived A and T alleles . Finally , for mutations not affected by gBGC ( WW and SS sites ) , DAFi¯ increases with local recombination rate until it reaches a plateau starting at ~1 . 5 cM/Mb , which suggests that the effect of BGS is absent or strongly reduced above this recombination threshold ( Figure 1C , right ) . This latter observation implies that the linear increase of DAFi¯ above 1 . 5 cM/Mb at WS sites ( Figure 1C , left ) is entirely due to gBGC . Note that the exact same pattern holds for SGDP populations ( Figure 1—figure supplement 4C ) . Moreover , if we analyze all possible types of substitutions separately , gBGC appears to affect the 12 types of SNP according to whether the SNP type belongs to the SW , WS , or WW +SS class ( Figure 1—figure supplement 7 . These results suggest that SNPs located in regions where recombination is higher than 1 . 5 cM/Mb are affected by gBGC and not by BGS ( Figure 1C , Figure 1—figure supplement 6C ) . Therefore , WW and SS sites with a recombination rate above 1 . 5 cM/Mb ( representing 2 . 88% and 2 . 94% of all SNPs for 1000G and SGDP datasets , respectively ) should be optimal for demographic inference , as they appear to evolve mainly neutrally . Since DAFi¯ increases with recombination rate ( Figure 1 ) , BGS does not simply amount to lowering the effective population size ( Charlesworth , 1994; Charlesworth et al . , 1995; Hudson and Kaplan , 1995 ) , as this simple rescaling would not modify allele frequencies . BGS thus affects the SFS ( Zeng and Charlesworth , 2011 ) in complex ways ( Nicolaisen and Desai , 2013 ) , and the comparison of sites that are differentially exposed to BGS allows us to better examine this influence . The SFS computed in ten 1000G populations for different recombination classes ( Figure 2A , Figure 2—figure supplement 1 ) shows distortions that are qualitatively similar in all populations , irrespective of differences in demographic history . As compared to the highest recombination class , the second-to-lowest recombination class ( which is potentially the one most strongly affected by BGS ) not only shows an excess of singletons , but also a deficit of intermediate and high frequency variants ( Figure 2A ) . Similar distortions are also observed in non-transcribed regions , and even ( but to a lower extent ) in regions at least 50 kb away from transcribed regions ( Figure 2—figure supplement 2 ) , in line with our results for DAFi¯ . To understand the respective effects of gBGC and BGS on the SFS , we computed the SFS for subsets of mutations differentially affected by gBGC in the Yoruba ( YRI ) and Japanese ( JPT ) 1000G populations ( Figure 2B ) . In line with previous work ( Lachance and Tishkoff , 2014 ) , we find that the difference between the SFSs of unbiased mutations ( WW + SS ) and biased mutations ( SW and WS ) increases with recombination rate . In particular , WS mutations show a deficit of low-frequency variants and an excess of intermediate- and high-frequency variants in regions of high recombination ( Figure 2B ) . As previously recognized ( Katzman et al . , 2011; Lachance and Tishkoff , 2014 ) , the excess of high-frequency variants at WS sites is not compensated by a corresponding deficit of high-frequency variants at SW sites , implying that gBGC could contribute to the increase of nearly fixed derived alleles that has previously been attributed to mislabelled ancestral states or positive selection ( Hernandez et al . , 2007 ) . To investigate the impact that the choice of SNPs may have on demographic inference , we estimated demographic parameters for the Yoruba and Japanese populations using three different SFSs ( Figure 3A ) : the synonymous SFS commonly used in exome resequencing studies; the SFS inferred on non-transcribed regions at least 50 kb away from coding regions ( NTR-50kb ) , and on our best-filtered dataset ( WW + SS sites in ≥1 . 5 cM/Mb regions ) , hereafter called the ‘neutral’ dataset . Note that this neutral SFS was computed over both TR and NTR regions since they show the same SFS ( Figure 2C ) . Interestingly , the SFS observed at synonymous sites differs markedly from that observed at neutral sites , as it comparatively shows a significant deficit of low-frequency variants and a large excess of high-frequency variants ( Figure 3A , Figure 3—figure supplement 1A ) . It appears that this latter excess is due to gBGC , as it disappears when one computes the SFS on synonymous sites not affected by gBGC ( Figure 3—figure supplement 1B ) . Using a simple demographic model of a focal population going through three successive bottlenecks and receiving some migrants from surrounding populations ( modelled as a ghost population for simplicity ) ( Figure 3C ) , we can fit almost perfectly the three SFSs ( Figure 3—figure supplement 2 ) . Yet , the inferred parameters differ considerably ( Supplementary file 3 - Table S3 ) . For the Yoruba population , the differences in demography are especially important in the old periods ( >100 ky , Figure 3B ) . With the neutral SFS , we nevertheless infer a more recent last bottleneck dated at the end of the Last Glacial Maximum ( LGM ) , a more pronounced and more recent admixture event from surrounding populations . The ancient demography is markedly different with a significantly more ancient second bottleneck and a significantly lower ancient population size inferred from both synonymous and NTR-50 kb SFS . The Japanese demography inferred from the three data sets shows more similarity over the last 600 ky but the demography inferred from the neutral data set suggests a stronger recent bottleneck ( pre LGM ) and no population expansion as compared to what is inferred from the synonymous SFS neutral data set . Our results thus clearly show that very different demographies can be inferred from neutral and non-neutral SFSs . However , even though BGS and gBGC affect the SFS of populations with distinct histories in a qualitatively similar way , they have different consequences on their reconstructed demography . It thus appears difficult to predict how demographic parameters will be biased when using non-neutral SFS . To confirm that our observed patterns were compatible with background selection , we ran individual-based forward simulations implementing BGS with SLiM v . 2 . 3 ( Haller and Messer , 2017 ) in populations having the demography estimated from neutral sites in the Japanese and the Yoruba populations ( see Supplementary file 3 - Table S3 ) . Overall , the simulated BGS patterns qualitatively match the observation very well ( Figure 4 , Figure 4—figure supplement 1 , and Figure 4—figure supplement 2 ) . As observed in real data ( Figure 1 ) , neutral sites simulated next to selected regions present a strong increase in DAFi¯ with recombination rate ( Figure 4A ) , and the SFS at neutral sites shows a considerable excess of singletons and a deficit of intermediate- and high-frequency variants for low-recombination rates ( Figure 4B ) , respectively . These results show that BGS can reproduce both the observed correlation between DAFi¯ and local recombination rates , and the observed distortions of the SFS in low-recombination regions .
Delineating the neutrally evolving part of the human genome remains a challenge , as variation in the intensity of recombination , mutation , and selection are increasingly recognised as having a strong effect on observable genomic diversity in humans ( Corbett-Detig et al . , 2015; Elyashiv et al . , 2016 ) and other organisms ( e . g . Elyashiv et al . , 2016; Ravinet et al . , 2017 ) . Here , we have shown that a surprisingly large proportion ( up to 95% ) of our genome might be affected by background selection ( BGS ) and/or GC-biased gene conversion ( gBGC ) . These two processes , which both depend on recombination , strongly affect observed measures of genetic diversity along the genome and can lead to biased demographic inference if not properly taken into account ( Figure 3 ) . We have interpreted the striking linear relationship observed between DAFi¯ and recombination rate ( Figure 1 ) as evidence for the pervasive effect of BGS but other processes could in principle lead to a similar relationship . For instance , a mutagenic effect of recombination could lead to an increased diversity in regions of high recombination ( Hellmann et al . , 2003 ) . The examination of extremely low-frequency mutations , which should be enriched for new mutations , did not reveal any association between recombination rate and the density of new mutations in a large human sample ( Schaibley et al . , 2013 ) , but a more recent study of de novo mutations suggested the existence of such a correlation ( Francioli et al . , 2015 ) . Alternatively , a correlation between mutation and recombination rates could occur if these rates were both affected by the same process , such as replication timing ( Stamatoyannopoulos et al . , 2009; Koren et al . , 2012 ) or transcription rate ( Gerton et al . , 2000; Park et al . , 2012 ) . However , a mere correlation between mutation and recombination rates cannot explain two key aspects of our observations . First , DAFi¯ plateau at high recombination rates once the effect of gBGC is removed ( Figure 1C ) , whereas it should continue increasing if only mutation-recombination correlation was driving the relationship between DAFi¯ and recombination . Second , we find a significant difference in the shape of SFS computed in regions of low and high recombination ( Figure 2A ) , even though mutation rate should have no effect on the shape of the SFS . To better investigate the effect of a possible mutation-recombination correlation , we have used the fact that DAFi¯ is correlated with the B-statistic ( Figure 1—figure supplement 5F ) , for which a simple model ( Hudson and Kaplan , 1995 ) predicts its value as a function of mutation and recombination rates . We find that the B-statistics inferred by McVicker et al . ( 2009 ) are significantly better fitted as a function of the recombination rate if we assume a log-log linear relationship between recombination and deleterious mutation rate than if we impose a constant mutation rate across the genome ( Figure 1—figure supplement 8 ) . Interestingly , under our log-log linear model , both the observed and predicted B-statistics reach a plateau value of ~0 . 9 above a recombination rate of ~1 . 5 cM/Mb . This pattern remains if we only consider subsets of SNPs ( e . g . WW + SS sites; Supplementary file 4 - Table S4 ) . Therefore , these results suggest that in addition to BGS and gBGC , some correlation between mutation and recombination rate is required to best explain our observed patterns . Moreover , given the relationship observed between B-statistics and DAFi¯ ( Figure 1—figure supplement 5F ) , the reduced effect of recombination on B above 1 . 5 cM/Mb should translate into a similar absence of change in DAFi¯ above the same threshold , thus explaining the plateau we see in Figure 1C above 1 . 5 cM/Mb . The occurrence of pervasive positive selection , either in the form of soft or hard sweeps ( Kern and Hahn , 2018 ) or of positive selection on polygenic traits ( Boyle et al . , 2017 ) in our genome could also lead to a correlation between genetic diversity and recombination , as the effect of selection on linked neutral sites should decrease with recombination . However , positive selection should lead to an increase of both low- and high-frequency variants in the SFS ( Fay et al . , 2000; Hernandez et al . , 2007; Huber et al . , 2016; Pavlidis and Alachiotis , 2017 ) , whereas we only observe an increase of low-frequency variants in low-recombination regions where the effect of selection should be strongest ( Figure 2A ) , which is the expected effect of BGS ( Figure 4 ) . The exact proportion of the genome that is influenced by selection is still the source of an intense debate ( Bernstein et al . , 2012; Rands et al . , 2014; Graur , 2017; Kern and Hahn , 2018 ) . Here , we show that up to 80–85% of the human genome is probably affected by background selection ( BGS ) , an effect that is not subtle ( Reed et al . , 2005 ) and that is visible from single individuals genomes ( Figure 1—figure supplement 5A ) . Even though our estimate of the fraction of the human genome influenced by BGS matches relatively well with that reported to be biochemically functional by the ENCODE consortium ( Bernstein et al . , 2012 ) , our results do not imply that 80–85% of the human genome is functional . They rather show that functional sites that are the direct target of purifying selection in both coding and non-coding regions ( potentially representing 8–15% of the genome , Rands et al . ( 2014 ) ; Graur , 2017 ) have an important but indirect influence on most of the genome . As expected , the effect of BGS is clearly mediated by local recombination rate , but it extends well beyond coding regions in humans ( Hernandez et al . , 2011 ) ( Figure 1 ) , and it is thus not restricted to species with a large effective size ( Corbett-Detig et al . , 2015 ) . Our results also show that the influence of gBGC is not restricted to recombination hotspots ( Spencer et al . , 2006; Glémin et al . , 2015 ) , but that it has also a strong footprint in regions with a recombination rate larger than 1 . 5 cM/Mb , but note that it could affect ( to a lesser degree ) regions with an even lower recombination rate ( see Figure 1—figure supplement 5D . These regions represent about 15 . 9% and 16 . 2% of the polymorphic positions for the 1000G and SGDP datasets , respectively . Taken together , BGS and gBGC thus affect more than 95% of the polymorphic sites in our genome , and we have identified only a small fraction of all genomic SNPs ( ~3% , Supplementary file 2 - Table S2 ) that can be considered as evolving neutrally . Interestingly , our neutral SNPs are found in both transcribed and non-transcribed-regions ( Figure 2C ) , and they are enriched close to telomeric regions ( Figure 1—figure supplement 10 ) , where BGS is predicted to be weaker ( Charlesworth , 2012 ) . Whereas SNPs included in our best-filtered set are evolving mostly neutrally , it does not imply that all other SNPs are influenced by BGS and gBGC . Indeed , our way of identifying selection and biased gene conversion is indirect and operates on arbitrarily defined recombination-rate categories . Thus , DAFi¯ cannot be used to identify the presence of selection at the SNP level or in small genomic regions , or inversely , the presence of neutral SNPs in low recombining segments between recombination hotspots . A more precise mapping of selected genomic segments could use information on the positions of known functional elements ( Siepel et al . , 2005; Kellis et al . , 2014; Rands et al . , 2014; Elkon and Agami , 2017 ) or B-statistics ( McVicker et al . , 2009; Elyashiv et al . , 2016 ) , which could also be used to evidence neutrally evolving regions in both low- and high-recombination regions . To investigate if and how DAFi¯ depends on potential co-variates within our neutral set of SNPs , we have examined its relationship with several statistics , such as B-statistics or the distance ( in map units ) to exons , as well as distances to conserved elements and to recombination hotspots . In our neutral set , we find virtually no relationship between DAFi¯ and recombination rate , with average DAFi¯ remaining close to its mean value of 0 . 146 ( Figure 1—figure supplement 9A ) , but we find a negative relation with the distance to recombination hotspots ( Figure 1—figure supplement 9B , a positive relationship with distance to conserved elements and with B-statistics ( Figure 1—figure supplement 9C–D ) , and a small positive correlation with distance to exons ( DAFi¯ varies from 0 . 145 to 0 . 15 , close to the average , Figure 1—figure supplement 9E ) . It thus seems that recombination hotspots still play a role in decoupling selected from neutral sites , and that sites furthest away from hotspots might still be slightly sensitive to BGS . Purifying selection in phastCons conserved elements ( Siepel et al . , 2005 ) is also exerting a strong negative pressure on derived allele frequencies , with average DAFi¯ below 0 . 14 at sites less than 0 . 0003 cM away from these elements ( which correponds approximatively to a distance of 200 bp if RR = 1 . 5 cM/Mb ) . Contrastingly , being further than 0 . 05 cM away from these conserved elements allows DAFi¯ to rise above 0 . 16 , an average value that is barely reached for sites with associated mean B values close to 1 . These results suggest that phastCons elements represent the covariate that has the strongest remaining influence on DAFi¯ within our neutral set . The SFS of each population is affected by BGS and gBGC ( Figure 2 , Figure 2—figure supplement 1 ) , and the demography inferred from neutrally evolving SNPs differs markedly from that based on synonymous sites or sites in non-transcribed regions ( Figure 3A ) . However , we show that BGS and gBGC can have different impacts on the inferred demography of the populations . For instance , we found that they lead to an underestimation of the age of a bottleneck and an overestimation of the magnitude of a demographic expansion in the Yoruba population , but we do not observe such strong biases in the Japanese population . It therefore appears difficult to predict the specific biases introduced by these evolutionary forces on demographic inference , except perhaps under simple evolutionary scenarios ( Ewing and Jensen , 2016 ) . We therefore suggest that future studies of demographic history should be based on a set of markers that is minimally influenced by these non-neutral forces . We have also computed the observed SFS for subsets of neutral SNPs with various values of the covariates mentioned above ( Figure 1—figure supplement 9 ) . SNPs in the 1st and 4th distance-quartiles to hotspot show similar SFS , with a slight excess of singletons and high-frequency variants for the sites furthest to hotspots ( Figure 2—figure supplement 3A . Even though conserved elements had the strongest influence on DAFi¯ , the SFSs computed at sites belonging to the 4th distance quartile and to all sites still look very similar , especially in the Japanese population , while sites in the 1st distance quartileshow an excess of singletons and a deficit of high-frequency variants ( Figure 2—figure supplement 3B . Exonic and non-exonic SFSs within our neutral set differ mainly by increased frequencies of singletons for exonic SNPs , yet the removal of exonic SNPs has no impact on the SFS ( Figure 2—figure supplement 3C ) . In conclusion , even though exonic SNPs and those located close ( ≤0 . 0003 cM ) to phastCons elements show different SFS shapes ( Figure 2—figure supplement 4 ) , their removal from our neutral set would have no major effect on the shape of the SFS , since they represent only a small fraction ( 2 . 2% and 16 . 9% respectively ) of the SNPs in our neutral set . It is interesting to compare our neutral set of SNPs to another previously defined set of neutral regions of the human genome that has been used as a reference for demographic inferences in a series of studies ( e . g . Gronau et al . , 2011; McManus et al . , 2015; King and Wakeley , 2016; Veeramah et al . , 2018 ) . Gronau et al . ( 2011 ) have identified a set of 37 , 574 potentially neutral regions of 1 kb in length with carefully chosen properties ( e . g . at least 1 kb away from exons and 100 bp away from phastCons elements , without CpG sites , separated by at least 50 kb , without recombination hotspots ) . The SFS computed on this alternative neutral set departs significantly from our neutral set , with a significant excess of singletons , a deficit of sites with intermediate allele frequencies , and an excess of nearly fixed variants , a pattern that can be explained by the action of both BGS and gBGC ( Figure 3—figure supplement 3A . Since a large B-statistic is also indicative of relaxed BGS , one could be tempted to use regions associated with B values larger than 0 . 9 as being potentially neutral . However , we see that its SFS also departs from that of our neutral set , with a small deficit of singleton and an excess of other frequency classes in Yoruba , and a slight excess of high-frequency variants in Japan ( Figure 3—figure supplement 3A ) . These differences in SFS shapes also lead to inferred demographies that are markedly different from that inferred from our own neutral set , and this especially for the Yoruba population ( Figure 3—figure supplement 3B ) . We suspect that the main discrepancy with our neutral set is the presence of gBGC in regions with B > 0 . 9 , such that filtering out SW and WS SNPs may result in a good alternative data set on which to perform demographic inferences Methods of demographic inference based on whole genomes ( e . g . Li and Durbin , 2009; Sheehan et al . , 2013; Schiffels and Durbin , 2014 ) should also be sensitive to BGS and gBGC , since they assume that heterozygosity levels within individuals is not driven by local recombination rates nor selection . In this respect , the history of human populations as well as that of other species might be more readily inferred from methods that can conveniently analyze restricted sets of neutrally evolving sites interspersed across the genome . Similarly , other types of inference using a biased neutral SFS as a reference could also be affected , such as inferences of the distributions of fitness effects ( DFE ) ( Keightley and Eyre-Walker , 2010; Kim et al . , 2017; Tataru et al . , 2017 ) , even though the magnitude of the effect remains to be investigated . In conclusion , we show that BGS and gBGC had a pervasive effect on most of our genome , but that we can conveniently define a set of sites ( representing about 3% of all polymorphic sites of both 1000G and SGDP datasets ) that should not be too influenced by these two evolutionary forces , even though some sites close to conserved elements could still be affected by BGS . Contrary to previously used sets of SNPs , these sites should lead to essentially unbiased demographic inferences and serve as a reference for future demographic reconstructions in humans . Due to its simplicity , our approach can be readily applied to any species for which a recombination map is available .
We analyzed two distinct whole genome datasets . The first one consisted of 100 individuals from ten 1000G populations ( Auton et al . , 2015 ) . For each 1000G population , we selected the ten individuals with the highest depth of coverage ( coverage >10× ) , such as to maximize the number of sites having no missing data . We also analyzed 20 individuals from panel C of the Simons Genome Diversity Project ( SGDP ) ( Mallick et al . , 2016 ) . These individuals were selected from ten SGDP populations that were geographically close to those analyzed for the 1000G project . Coverage was higher for the SGDP individual and ranged between 31 × and 64× ( see Supplementary file 1 - Table S1 for IDs and location of the 1000G and SGDP samples ) . We processed the 1000G and SGDP datasets identically . We removed all sites with any missing data and kept only diallelic sites from autosomal chromosomes . The ancestral state of each variant in these genomes was set to the chimpanzee reference genome ( panTro4 genome assembly ) to avoid any discrepancy between African and non-African populations . Only diallelic SNPs for which one of the variants observed in the 1000G or SGDP datasets corresponded to the chimpanzee ancestral state were kept for later analyses . In addition , we removed the CpG sites that present a peculiar mutation profile and are correlated with recombination rate ( Arbeithuber et al . , 2015 ) . We used the LD-based Yoruba-specific recombination map from the 1000 Genomes project ( Frazer et al . , 2007 ) to obtain the local recombination rate ( RR ) surrounding each SNP . We also estimated local RR by using three other maps: the LD-based CEU or JPT-specific recombination maps ( Frazer et al . , 2007 ) and the sex-averaged pedigree-inferred deCode map ( Kong et al . , 2010 ) . For each of these maps , we filtered out SNPs without RR information ( see Supplementary file 2 - Table S2 ) . We used the Yoruba-specific map to define hotspots as regions with RR >10 cM/Mb . Using Biomart ( http://grch37 . ensembl . org/biomart/martview/ ) , we assigned SNPs to transcribed ( TR ) and non-transcribed regions ( NTR ) . For each site , we inferred the distance to the closest exonic region in cM and in bp using the Ensembl exon positions ( ftp://ftp . ensembl . org/pub/grch37/release-90/gtf/homo_sapiens/Homo_sapiens . GRCh37 . 87 . gtf . gz ) . The B-statistic ( McVicker et al . , 2009 ) ( indicative of the strength of local background selection ) associated with each SNPs was retrieved from http://www . phrap . org/othersoftware . html and lifted over from the hg18 to the hg19 reference genome using the UCSC liftOver tool . Genomic Evolutionary Rate Profiling ( GERP ) rejection scores ( Davydov et al . , 2010 ) that quantify the level of evolutionary constraint acting on polymorphic sites and conserved elements identified using PhastCons on the primate subset of 46 vertebrates ( Siepel et al . , 2005 ) were downloaded from the UCSC platform ( Speir et al . , 2016 ) . The number of SNPs from the 1000G and SGDP datasets retained for each filter is reported in Supplementary file 2 - Table S2 . We finally retrieved 37 , 574 potentially neutral regions of 1 kb ( e . g . Gronau et al . , 2011; McManus et al . , 2015; King and Wakeley , 2016; Veeramah et al . , 2018 ) from http://compgen . cshl . edu/GPhoCS/data . php to make comparisons between our neutral set of SNPs to another possible sets . As gBGC favors strong ( abbreviated as S , and representing C and G bases ) compared to weak ( abbreviated as W , and representing A and T bases ) alleles , we defined three groups of SNPs according to the expected consequences of gBGC: ( 1 ) SNPs for which the derived state is favoured ( WS sites ) ; ( 2 ) SNPs for which the ancestral state is favoured ( SW sites ) , and ( 3 ) SNPs on which gBGC has no effect ( WW or SS sites ) . To quantify a local effect of selection and/or gBGC , we used the average derived allele frequency per individual ( DAFi¯ ) , where this average is computed over a set of sites found polymorphic in a collection of individuals . We show in the following that this statistic is ideally suited to evidence the potential effect of selection ( or mutation ) , as difference in the demography of the populations from which individuals are sampled should not translate into different values of this statistic among individuals . Start by considering a single non-recombining locus ( k ) with mutation rate uk , and for the sake of simplicity , let us consider just two individuals i and j , drawn from two different populations . Note that the same reasoning can be extended to an arbitrary number of individuals drawn from an arbitrary number of populations . Now , suppose that the two homologous alleles of these individuals have coalesced ti and tj generations ago , and that the most recent common ancestor of these four homologous alleles is tglobal . Now , the frequency of the derived allele in individual i at the k-th locus is simply given by ( 1 ) DAFik=nik2Stot , kwhere Stot , k is the total number of sites that are polymorphic at this k-th locus for this sample of two individuals , and nik is the number of derived alleles observed in individual i . Since nik is the number of heterozygous sites ( Hetik ) plus two times the number of homozygous derived sites ( Homik ) ( see Figure 1—figure supplement 1 ) , the expected value of nik can be expressed as a function of tglobal and the mutation rate uk as ( 2 ) E ( nik ) =E ( Hetik+2 Homik ) =2 uk tik+2 uk ( tglobal , k−tik ) =2 uk tglobal , k , which does not depend on tik , the coalescence times between homologous alleles in individuals 1 or 2 , as illustrated in Figure 1—figure supplement 1 . Therefore , E ( njk ) =E ( nik ) =2 uk tglobal , k , and ( 3 ) E ( DAFik ) = E ( nik ) /E ( 2Stot , k ) =tglobal , k/Ttot , k , ∀i , where Ttot , k is the total tree length at the k-th locus . Since the average derived allele frequency computed over an arbitrary number of unlinked loci m is obtained as the ratio of the total number of derived alleles over twice the total number of polymorphic sites , its expectation is then obtained as ( 4 ) E ( DAFi¯ ) =E ( ni ) E ( 2 Stot ) =∑kmE ( nik ) ∑kmE ( 2 Stot , k ) =∑kmuk tglobal , k∑kmukTtot , k , an equation that is valid irrespective of the number of individuals and populations sampled if one computes the number of derived alleles over all sites found polymorphic in the collection of individuals . If the mutation rate is uniform across loci , then equation ( 1 . 4 ) simplifies to ( 5 ) E ( DAFi¯ ) = t¯global/T¯tot , which only depends on the average global coalescence time of the total sample t¯global , and on the average tree length over all loci T¯tot , and not on the coalescence times in each population . Therefore , even though E ( DAFi¯ ) depends on the overall demography of the collection of individuals and on the composition of the samples , which both condition the global tMRCA and total tree lengths , the specific demographic histories of the sampled populations will not translate , in expectation , into different DAFi¯ among individuals examined for the same set of loci . Selection in some portion of the genome will affect tMRCAs , which should thus translate into differences in DAFi¯ computed for these regions . Differences in mutation rates across the genome might also affect DAFi¯ for some regions , but should not lead to individual differences , unless mutation rates are different in specific populations . For both SGDP and 1000G data sets , we ranked SNPs according to their associated recombination rate and binned them into 20 equal-sized classes of increasing recombination rates . We performed a similar binning for the different groups of SNPs we considered ( the three types of mutations , within a transcribed region or not , etc . ) or after ranking SNPs according to their distance to the nearest exon , to hotspots or to conserved elements . We then computed DAFi¯ for each bin b as DAFib¯=nib/ ( 2Stot , b ) . We estimated the unfolded SFS for ten 1000G population samples using different filters ( e . g . different recombination classes , different types of mutations ) . The SFS was then normalized ( Lapierre et al . , 2017 ) by dividing each entry by its expectation in a stationary population . To estimate if two SFSs are statistically different , we used a permutation approach . We first computed a distance between the two SFS as the sum of the squared difference in site frequencies over all SFS entries ( noted Dobs ) . We divided the SNPs into three categories: those shared by the two SFS ( if any ) , and those that were private to one of the SFS . We then randomly permuted sites among the two latest categories and re-evaluated the distance noted Dest . When one SFS was based on a subset of variants from another SFS , we subsampled sites from the largest dataset and re-evaluated Dest . We repeated the permutations or the resampling procedure 1000 times and estimated a p value as the frequency of Dobs ≥ Dest . For each filter ( e . g . per recombination class or per type of mutation ) , we identified sets of 100 adjacent SNPs along the genome and we sampled them with replacement such as to keep the same number of sites as in the non-bootstrapped set when computing statistics of interest ( DAFi¯ , SFS ) . We repeated the sampling 1000 times to obtain 1000 block-bootstrap sets of SNPs . 95% confidence intervals were computed by identifying the 2 . 5 and 97 . 5 quantiles of the resulting bootstrap distributions . We estimated the parameters of the demographic model shown in Figure 3C from the SFS of two 1000G populations ( Japan and Yoruba ) using the program fastsimcoal2 ( Excoffier et al . , 2013 ) ver 2 . 6 . We used the following command line options: . /fsc26 -t pop . tpl -n200000 -d -e pop . est -M -l25 -L50 -q −0 -C1 -c1 -B1 , where pop denotes either the Japan or the Yoruba population . We used the tpl and est setting files defined in Supplementary file SF1 . For each population , we performed 50 independent estimations and retrained the parameters that maximized the model likelihood . The confidence intervals of the parameters were estimated from 100 block-bootstrapped SFS obtained in a way similar to that described above . For each population , estimations were performed on each bootstrap dataset independently , using the maximum likelihood ( ML ) parameters values estimated above as initial values . Since we started parameter estimation close to the observed ML values , we only did five estimations per bootstrap and retained the parameters with maximum associated likelihood . A 99% confidence interval was then obtained for each parameter by estimating the 0 . 5% and 99 . 5% quantiles of its resulting bootstrap distribution . We performed individual-based simulations using the software SLiM v . 2 . 3 ( Haller and Messer , 2017 ) to check that BGS could reproduce observations . We simulated the demographic scenario inferred from the ‘neutral’ SFS ( i . e . from WW + SS sites with r ≥ 1 . 5 cM/Mb ) for the Japanese ( JPT ) and Yoruba ( YRI ) 1000G populations as described above ( Demographic inference ) . We simulated a linear genome of 50 Mb made up of 1000 regions of 5 kb . Each of these regions consisted of a 1 kb stretch experiencing purifying selection against deleterious mutations , followed by a 4 kb stretch with neutral mutations . We also simulated an alternative genomic architecture with 10 , 000 regions of 500 bp , each consisting of a 100 bp stretch under purifying selection , followed by a 400 bp stretch with neutral mutations . For computational efficiency , we scaled the inferred event times and population sizes by a factor of 0 . 1 and give below the rescaled values . We set the per-site mutation rate to 1 . 25 × 10–7 for deleterious and neutral mutations . The fitness contribution of all deleterious mutations was 1 – s in homozygous form and 1 – s/2 in heterozygous form . The fitness of individuals was computed multiplicatively across sites . We ran independent simulations for four recombination rates ( r = 10–9 , 10–8 , 10–7 , and 10–6 ) . For each demographic scenario and recombination rate , we simulated a scenario with background selection ( s = –0 . 1 ) and a neutral scenario ( s = 0 ) . For each parameter combination , we performed 100 independent replicates starting with a period of 4 × NANC generations , where NANC is the number of haploid genomes in the ancestral population ( Figure 4—figure supplement 1 ) . We set NANC = 4000 for both the Yoruba and Japanese simulation . At the end of each simulation , we output the full population and computed the number of derived alleles for each individual across a fixed number arbitrarily set to 40 , 000 SNPs , subsampled from all SNPs . These 40 , 000 SNPs were subsampled individually for each replicate simulation . The SFS of the population was subsampled to 10 individuals ( i . e . 20 haploid genomes ) following Nielsen et al . ( 2005 ) as pi , 20=k−1∑j=1k ( fji ) ( nj−fj20−i ) / ( nj20 ) , where k is the total number of SNPs in the dataset , and nj and fj are the number of haploid genomes in the full sample and the number of derived alleles in the full sample at the jth SNP , respectively ( see also Liu et al . , 2017 ) . We computed the SFS separately for each replicate simulation , and then calculated the mean and the 2 . 5 and 97 . 5 percentiles across these replicates for each entry pi , 20 . We normalized the SFS as described above ( subsection SFS ) . To model a potential correlation between mutation and recombination , we assumed that the per-base pair deleterious mutation rate ud depends on the local recombination rate r asudr=u0rb . This assumption implies a log-log linear relationship between mutation and recombination , with an intercept of logu0 and a slope of b . In the special case of b = 0 , mutation is independent of recombination . We then modified the approximate BGS model of Hudson and Kaplan ( 1995 ) by substituting ud ( r ) for the deleterious mutation rate . The reduction in the nucleotide diversity a at a focal site due to BGS is then predicted to be ( 6 ) B=ππ0≈exp ( −ud ( r ) r ) =exp ( −u0 rbr ) =exp ( −u0r ( b−1 ) ) where π0 is the baseline nucleotide diversity in the absence of BGS , and π is the effective nucleotide diversity with BGS . We fit this modified BGS model to the relationship between the B-statistic from McVicker et al . ( 2009 ) and the recombination rate associated with our polymorphic SNPs using the method of non-linear least squares as implemented in the nls function in R v 3 . 4 . 4 ( R core Team , 2018 ) . We then used the Akaike information criterion ( AIC , Akaike , 1974 ) to compare this extended BGS model to the original BGS model in which the mutation rate does not depend on the recombination rate ( b=0 ) . Note that McVicker et al . ( 2009 ) obtained their B-statistics by fitting a more complex BGS model to polymorphism and recombination data ( assuming no specific correlation between recombination and mutation ) . However , the model of Hudson and Kaplan ( 1995 ) used here is just a simplified version of that used by McVicker et al . ( 2009 ) . It assumes that neutral sites on which diversity is measured are in the middle of a region containing sites under negative selection , that recombination rates are uniform in the considered region , and that selection coefficients at deleterious sites are small relative to the total recombination rate in the region . These assumptions seem reasonable except for sites that are very close to recombination hotspots or close to telomeres , but we expect a qualitatively global agreement between these two models . An exact quantitative match is not required here , since our goal here is simply to assess whether a correlation between mutation and recombination rates needs to be invoked rather than to accurately estimate the parameters of the model ( u0 and b ) . | Human chromosomes are made up of DNA , which contains about 3 billion ‘letters’ that carry the instructions needed to build and maintain an individual . However , only about 10 percent of the human genome is made up of genes that code for proteins , or have a defined role in the body . The DNA sequence is largely the same in all people , but some modifications – or variants – occur about every hundred letters . These produce different versions of the same gene , which give us our unique features , such as the color of our hair or eyes . The frequencies of some genetic variants can change over time , which makes human populations diverge genetically and physically . This can happen through different mechanisms . Positive selection keeps variants that are beneficial in specific environments , while negative selection removes genetic changes that are detrimental , for example because they cause disease . Transmission bias favors one of the two variants from our two parents . Chance alters the frequencies of neutral variants , which are neither good nor bad for the individual . It is important to distinguish between these different scenarios , as they inform us about the forces that act on human evolution . For example , neutral variants tell us about the demography and migration patterns between populations . Variants under negative selection reveal which genetic areas are under pressure to stay the same because they are important for the organism to function correctly . Until now , it was unclear how we could best identify the variants affected by different evolutionary pressures , and how much of the genome was under negative selection . Pouyet , Aeschbacher et al . created a measure of genetic diversity that is only affected by selection or transmission bias . The results showed that negative selection influences as much as 85 percent of our genome , whereas transmission bias affects a majority of the rest of the genome . After removing these two biases , less than 5 percent of the human genome is found to evolve by chance . This suggests that while most of our genetic material is formed of non-functional sequences , the vast majority of it evolves indirectly under some type of selection . These findings define which parts of our genome evolves neutrally and can therefore be used to correctly reconstruct the past demography and migration events of humans around the world . The next step could be to reassess the history of human populations that was drawn using genomic data . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"genetics",
"and",
"genomics"
] | 2018 | Background selection and biased gene conversion affect more than 95% of the human genome and bias demographic inferences |
The comprehensive understanding of cellular signaling pathways remains a challenge due to multiple layers of regulation that may become evident only when the pathway is probed at different levels or critical nodes are eliminated . To discover regulatory mechanisms in canonical WNT signaling , we conducted a systematic forward genetic analysis through reporter-based screens in haploid human cells . Comparison of screens for negative , attenuating and positive regulators of WNT signaling , mediators of R-spondin-dependent signaling and suppressors of constitutive signaling induced by loss of the tumor suppressor adenomatous polyposis coli or casein kinase 1α uncovered new regulatory features at most levels of the pathway . These include a requirement for the transcription factor AP-4 , a role for the DAX domain of AXIN2 in controlling β-catenin transcriptional activity , a contribution of glycophosphatidylinositol anchor biosynthesis and glypicans to R-spondin-potentiated WNT signaling , and two different mechanisms that regulate signaling when distinct components of the β-catenin destruction complex are lost . The conceptual and methodological framework we describe should enable the comprehensive understanding of other signaling systems .
Cellular signaling systems have evolved complex circuitry involving multiple layers of regulation , making their comprehensive characterization a major challenge . Forward genetics in model organisms has been a general and unbiased way to identify new components in signaling pathways and to map their connectivity . However , since signaling pathways have often diverged between humans and these simpler model systems , their analysis in human cells becomes an important goal . Indeed , our ability to identify the best therapeutic strategy or to predict the effectiveness of drugs targeting specific proteins is often hampered by an incomplete understanding of signaling circuitry in human cells ( Lito et al . , 2013 ) . Recent methodological advances have enabled the interrogation of biological processes in human cells through powerful genome-wide screens that overcome many of the limitations associated with previous platforms ( Carette et al . , 2009; Gilbert et al . , 2014; Shalem et al . , 2014; Wang et al . , 2014 ) . Yet , inferring functional relationships in complex pathways from such screens remains a major obstacle that has only recently began to be addressed ( Bassik et al . , 2013; Blomen et al . , 2015; Parnas et al . , 2015; Wang et al . , 2015 ) . Genetics has long relied on the use of sensitized backgrounds , modifier screens and synthetic effects to uncover the myriad layers of regulation in signaling pathways . We reasoned that one way to discover both epistatic relationships on a genome scale and unique context-specific requirements would be through the quantitative comparison of genome-wide screens in which the pathway is activated by different ligands , and of suppressor screens following targeted disruption of critical nodes . We took advantage of two methodologies to conduct a systematic genetic analysis of WNT signaling in human cells: forward genetics in haploid cells using gene trap ( GT ) -based insertional mutagenesis ( Carette et al . , 2009 ) , and targeted genome engineering by clustered regularly-interspaced short palindromic repeats ( CRISPR ) /CRISPR-associated protein 9 ( Cas9 ) ( Cong et al . , 2013; Mali et al . , 2013 ) . The WNT pathway is a fundamental signaling system that plays central roles in embryonic development , regeneration and cancer ( reviewed in Hoppler and Moon , 2014 ) . During development , WNT signaling orchestrates transcriptional programs that regulate cell proliferation and survival , cell fate determination , and tissue patterning . In adults , WNT signaling is instrumental in defining stem cell niches in multiple organs , which maintain tissue homeostasis during routine turnover or following injury . Overactive WNT signaling can be oncogenic , driving both the initiation and maintenance of various types of cancer , most notably colorectal cancer ( CRC ) . While the pathway has been studied intensively ( we provide a snapshot in Figure 1A and refer readers to the legend for details ) , critical steps remain poorly understood even 34 years after the discovery of ‘int1’ , as mammalian WNT was initially called ( Nusse and Varmus , 1982 ) . The complex circuitry of the pathway may mask unknown regulatory mechanisms overlaid on the core module , making it an ideal system for an in-depth , methodical genetic dissection , extending a rich tradition of genetic studies ( Nüsslein-Volhard and Wieschaus , 1980 ) . Known pathway components would serve to benchmark any new discoveries , and new discoveries would likely have important therapeutic implications due to the pathway’s direct relevance to stem cell biology and cancer . 10 . 7554/eLife . 21459 . 003Figure 1 . Reporter-based , forward genetic screens in haploid human cells identify negative , attenuating and positive regulators of WNT signaling . ( A ) Schematic model of canonical WNT signaling , highlighting the main pathway components and regulatory events in the absence ( left panel ) and presence ( center panel ) of ligands , and other known regulators relevant to this work ( right panel ) . When the pathway is off , the transcriptional co-activator β-catenin ( CTNNB1 ) is constitutively targeted for proteasomal degradation by the destruction complex , composed of the scaffold proteins adenomatous polyposis coli ( APC ) and AXIN , and the kinases glycogen synthase kinase 3 ( GSK3 ) and casein kinase 1α ( CSNK1A1 ) . The T-cell-specific transcription factor ( TCF ) /lymphoid enhancer-binding factor ( LEF ) family of transcription factors , together with transducin like enhancer of split ( TLE ) and histone deacetylases ( HDAC ) , repress WNT target genes . Binding of WNT to its co-receptors frizzled ( FZD ) and low-density lipoprotein receptor-related protein 6 ( LRP6 ) leads to the assembly of a receptor complex that inactivates the destruction complex through a mechanism involving recruitment of AXIN by LRP6 and the adapter protein dishevelled ( DVL ) . Consequently , CTNNB1 accumulates in the cytoplasm , translocates to the nucleus and promotes WNT target gene transcription in cooperation with TCF/LEF and other co-activators such as CREB-binding protein ( CREBBP ) and B-cell CLL/lymphoma 9 protein ( BCL9 ) . R-spondins ( RSPOs ) are secreted proteins that potentiate the response of stem cells to WNT ligands by blocking the degradation of FZD and LRP6 receptors . RSPO binds to leucine-rich repeat-containing G-protein-coupled receptors ( LGRs ) and neutralizes two transmembrane E3 ubiquitin ligases , ZNRF3 and RNF43 , that clear WNT receptors from the cell surface . Other regulatory mechanisms include modulation of AXIN levels by the poly ADP-ribosylation-dependent E3 ubiquitin ligase RNF146 , and recruitment of DOT1L and MLLT10 , two proteins involved in histone H3 K79 methylation , to WNT target genes . ( B ) Schematic of WNT reporter-based forward genetic screens in haploid human cells using a GT-bearing retrovirus for mutagenesis , followed by phenotypic enrichment by FACS . LTR , long terminal repeats; SA , splice acceptor; pA , polyadenylation signal . ( C–F ) Circle plots depicting genes enriched for GT insertions in screens for negative ( C ) , attenuating ( D ) and positive ( E and F ) regulators of WNT signaling . Two independent screens for positive regulators were performed at low ( E ) and high ( F ) selection stringencies by sorting for cells with the lowest 10% and 2% WNT reporter fluorescence , respectively . The y-axis indicates the significance of GT insertion enrichment in the sorted vs . the control cells ( expressed in units of -log10FDR-corrected p-value ) and the x-axis indicates genes ( in random order ) for which GT insertions were mapped in the sorted cells . Genes with FDR-corrected p-value<0 . 01 are labeled and colored in light blue if they encode a known pathway component , or in pink if their product has not been previously implicated as a regulator of canonical WNT signaling . The diameter of each circle is proportional to the number of unique inactivating GT insertions mapped in the sorted cells , which is also indicated next to the gene name for the most significant hits with FDR-corrected p-values<10−4 . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 00310 . 7554/eLife . 21459 . 004Figure 1—figure supplement 1 . Characterization of HAP1-7TGP , a haploid human cell line harboring a WNT-responsive fluorescent reporter ( A–D ) , and depiction of FACS gates and phenotypic enrichment during various reporter-based forward genetic screens ( E–H ) . ( A ) Histogram depicting WNT reporter activity ( EGFP fluorescence of ~4500 cells in arbitrary units ( a . u . ) , plotted on a logarithmic x-axis ) for the parental haploid human cell line HAP1 and the WNT reporter cell line HAP1-7TGP . Where indicated , cells were treated with 50% WNT3A conditioned medium ( CM ) . Basal EGFP fluorescence in HAP1-7TGP cells was minimal and increased by approximately two orders of magnitude in response to WNT3A . ( B ) Dose response curve depicting fold-induction in WNT reporter ( average EGFP fluorescence from ~3500 cells ) following treatment with WNT3A , as a function of WNT3A CM concentration ( expressed as percentage of total medium ) . The WNT3A CM concentrations used in the screens for attenuating and positive regulators of WNT signaling ( Figure 1D–1F ) are indicated . EGFP fluorescence in HAP1-7TGP cells increased by up to 155-fold in response to saturating WNT3A . ( C ) Histogram depicting WNT reporter activity for ~10 , 000 HAP1-7TGP cells . Where indicated , cells were treated with 50% control CM from L-cells , 50% WNT3A CM or 150 ng/ml recombinant WNT3A . HAP1-7TGP cells responded to WNT3A CM and recombinant WNT3A , but not control CM . ( D ) Histogram depicting WNT reporter activity for ~5000 HAP1-7TGP cells . Where indicated , cells were treated with 50% WNT3A CM alone or together with 10 μM of the tankyrase inhibitor XAV-939 , which elevates the levels of AXIN , or with 10 μM of the GSK3 antagonist CHIR-99021 . WNT3A-induced reporter expression in HAP1-7TGP cells has the pharmacological hallmarks of being mediated by canonical WNT signaling . ( E ) Dot plot depicting WNT reporter activity ( EGFP fluorescence vs . pacific blue auto-fluorescence from ~17 , 500 cells , plotted on biexponential axes ) for HAP1-7TGP cells mutagenized with GT retrovirus and enriched during the WNT negative regulator screen ( Figure 1C ) . Unsorted cells and cells amplified following sort 1 using the indicated EGFP+ gate are depicted , and the percentage of cells within this gate for each population is indicated in parenthesis . Cells were analyzed for GT insertions following a second round of FACS sorting ( not shown ) using an equivalent gate . ( F ) Histogram depicting WNT reporter activity for ~7000 mutagenized HAP1-7TGP cells enriched during the WNT attenuating regulator screen ( Figure 1D ) . Where indicated , cells were treated with 12 . 5% WNT3A CM . Unsorted cells and cells amplified following sort 1 using the indicated highest 2% EGFP fluorescence gate are depicted . The percentage of cells within this gate for each population is indicated in parenthesis . Cells were analyzed for GT insertions following a second round of FACS sorting ( not shown ) using an equivalent gate . ( G ) Histogram depicting WNT reporter activity for ~12 , 500 mutagenized HAP1-7TGP cells enriched during the WNT positive regulator screen done at low stringency ( Figure 1E ) . Cells were treated with 50% WNT3A CM where indicated . Unsorted cells and cells amplified following sort 1 or sort 2 using the indicated lowest 10% EGFP fluorescence gate are depicted , and the percentage of cells within this gate for each population is indicated in parenthesis . ( H ) Histogram depicting WNT reporter activity for ~12 , 500 mutagenized HAP1-7TGP cells enriched during the WNT positive regulator screen done at high stringency ( Figure 1F ) . Cells were treated with 50% WNT3A CM where indicated . Unsorted cells and cells amplified following sort 2 using the indicated lowest 2% EGFP fuorescence gate are depicted , and the percentage of cells within this gate for each population is indicated in parenthesis . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 004 We initially probed the intact pathway through genome-wide , forward genetic screens for negative , attenuating and positive regulators to define the basic set of genes required for signaling in haploid human cells . We identified many of the known core pathway components and found a new requirement for the transcription factor AP-4 ( TFAP4 ) . Unexpectedly , these screens uncovered a dominant allele of AXIN2 that dissociated β-catenin ( CTNNB1 ) protein stabilization , considered the decisive event in WNT signaling , from its transcriptional activity . To find factors required for amplification of WNT responses by R-spondins ( RSPOs , Figure 1A ) , we devised a sensitized screen for RSPO-enhanced WNT signaling and uncovered a requirement for the glycophosphatidylinositol ( GPI ) anchor biosynthetic machinery and glypicans . Finally , we searched for mutations that could suppress constitutive signaling in cells with compromised function of the CTNNB1 destruction complex , recapitulating the most common defect in oncogenic WNT signaling . Suppressor screens in cells lacking adenomatous polyposis coli ( APC ) or casein kinase 1α ( CSNK1A1 ) , the two rate-limiting negative regulators of the pathway in haploid cells , revealed two distinct mechanisms that regulate CTNNB1 abundance and transcriptional activity , respectively . One mechanism was specific to cells lacking CSNK1A1 , but not APC , suggesting that different components of the destruction complex have different functions in WNT signaling beyond their common function controlling CTNNB1 protein abundance . Through a quantitative comparative analysis across seven screens , we confirmed epistatic relationships for known regulators and predicted them for new ones . The combined results of these screens provide a comprehensive resource for understanding the regulation of canonical WNT signaling .
A central goal of this project was to search for quantitative and context-specific regulators of WNT signaling in an unbiased and comprehensive manner . We adopted two design principles that exploited the flexibility of forward genetics in cultured human cells . First , as a means of phenotypic enrichment , we chose a fluorescence-based , quantitative transcriptional reporter of WNT signaling . Since WNT reporter fluorescence is a continuous readout , in contrast to digital readouts such as cell viability or the presence or absence of a phenotype , it enabled us to enrich for cells with enhanced or reduced signaling phenotypes by fluorescence activated cell sorting ( FACS ) with complete flexibility on the stringency of selection . Second , all screens were performed in a pooled format following genome-scale insertional mutagenesis using a GT-bearing retrovirus , which contains a strong splice acceptor site and can therefore disrupt genes when it integrates in either exons or introns . This mutagenesis method is untargeted , distinguishing it from approaches in which short hairpin RNAs or single guide RNAs ( sgRNAs ) are designed to perturb a pre-defined set of cistrons . We constructed and thoroughly characterized a clonal haploid human cell line , hereafter called HAP1-7TGP , in which expression of enhanced green fluorescent protein ( EGFP ) is driven by an established WNT-responsive element containing the seven TCF/LEF-binding sites , minimal promoter and 5’UTR of the SuperTOPflash reporter ( Fuerer and Nusse , 2010; Figure 1—figure supplement 1A–1D ) . While this construct has been used extensively to report on WNT responses , it may not mimic all endogenous regulatory sequences driving WNT target gene expression . In particular , the effects of proteins involved in modifying chromatin structure could differ between the reporter and endogenous target genes . Given these limitations , whenever possible we confirmed new regulatory mechanisms by measuring endogenous WNT target gene activity or assessing WNT-dependent phenotypes in model organisms . To obtain saturating mutational coverage of the genome , we started our screens with 120 million HAP1-7TGP cells ( or engineered derivatives thereof ) mutagenized with the GT retrovirus , ensuring that the mutant cell population as a whole contained multiple ( up to a few hundred ) independent lesions in every gene ( Figure 1B ) . The screens should therefore capture most genes involved in the phenotype being enriched for , except for genes required for the viability of haploid cells and genes with redundant function , since the probability of independent GT integrations disrupting redundant genes in the same haploid cell is vanishingly small . This limitation is inherent to all forward genetic screens that use random or untargeted mutagenesis . After sequential rounds of FACS-based phenotypic enrichment and growth ( HAP1-7TGP cells do not require WNT signaling for growth , enabling the propagation of cells with decreased or increased WNT signaling activity following phenotypic enrichment ) , we mapped retroviral integration sites at nucleotide resolution by deep sequencing an amplified library containing junctions between GTs and flanking genomic DNA ( Figure 1B; see Materials and methods ) . Sequence reads from the sorted cells were compared to those from control , unsorted cells to identify genes enriched for GT insertions in the sorted cell population . Disruption of these genes would be expected to cause the phenotype used as the basis for selection . We devised a genome-wide screen to identify the rate-limiting negative regulators of WNT signaling ( i . e . genes whose disruption leads to constitutive pathway activity ) in haploid human cells . We used FACS to sort mutagenized HAP1-7TGP cells with high WNT reporter activity in the absence of WNT ligand ( ‘EGFP+’ gate in Figure 1—figure supplement 1E ) . Following two rounds of sorting ( see Materials and methods ) , GT insertions in only three genes showed statistically significant ( false discovery rate ( FDR ) -corrected p-value<0 . 01 ) enrichment in the sorted cells: CSNK1A1 , APC and PYY ( Figure 1C and Supplementary file 1 ) . We mapped 144 independent GT insertions in CSNK1A1 and 29 in APC , showing that our mutagenesis had indeed targeted each gene multiple times . CSNK1A1 and APC are core components of the destruction complex that suppresses WNT signaling by promoting CTNNB1 degradation ( Figure 1A ) ; their identification as top hits reassured us that our screening strategy could identify important regulators of canonical WNT signaling . Genes encoding other known negative regulators of the pathway , such as GSK3A and GSK3B or AXIN1 and AXIN2 , presumably did not score as hits in this screen due to redundancy , as suggested by their expression profile in HAP1 cells ( Table 1 ) . We demonstrate later that AXIN1 and AXIN2 are indeed functionally redundant in HAP1 cells ( Figure 3—figure supplement 1B ) . The fact that APC , but not APC2 , was identified as a hit indicates that these two genes are not redundant in HAP1 cells , a conclusion that is supported by the relatively low expression level of APC2 ( Table 1 ) . 10 . 7554/eLife . 21459 . 005Table 1 . Relative gene expression level of selected WNT pathway regulators in HAP1 cells . RPKM values from duplicate RNAseq datasets generated as described in Materials and methods from two different passages of WT HAP1 cells are shown . Groups of paralogues and genes with similar functions are shaded in alternating colors to facilitate comparisons . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 005GeneRPKMReplicate 1Replicate 2AverageLGR4160 . 61174 . 69167 . 65LGR50 . 020 . 000 . 01LGR60 . 020 . 000 . 01ZNRF330 . 9033 . 3032 . 10RNF430 . 120 . 080 . 10LRP555 . 9264 . 3860 . 15LRP6109 . 51121 . 08115 . 30FZD119 . 5718 . 8519 . 21FZD219 . 5621 . 0920 . 33FZD348 . 0255 . 8251 . 92FZD419 . 6022 . 1020 . 85FZD531 . 8534 . 5233 . 19FZD633 . 5331 . 9532 . 74FZD713 . 8914 . 8914 . 39FZD84 . 024 . 484 . 25FZD93 . 662 . 803 . 23FZD1010 . 409 . 8610 . 13DVL173 . 9169 . 6171 . 76DVL251 . 7448 . 8050 . 27DVL388 . 8490 . 2589 . 54APC80 . 4785 . 2282 . 84APC22 . 943 . 693 . 32AXIN155 . 9754 . 0755 . 02AXIN210 . 0412 . 5111 . 27CSNK1A1111 . 18109 . 57110 . 37GSK3A75 . 9769 . 2172 . 59GSK3B62 . 7969 . 9866 . 39TCF7L223 . 8927 . 6925 . 79LEF112 . 3414 . 8013 . 57CTNNB1324 . 05308 . 53316 . 29CREBBP141 . 92165 . 58153 . 75PIGL4 . 074 . 514 . 29GPC4209 . 39229 . 86219 . 63GPC613 . 8814 . 9014 . 39TFAP436 . 9941 . 9439 . 46SERBP1721 . 65698 . 99710 . 32HUWE1631 . 58777 . 06704 . 32 To identify attenuating regulators of WNT signaling ( i . e . genes whose disruption amplifies cellular responses to WNT ligands ) , we stimulated mutagenized HAP1-7TGP cells with a sub-saturating dose of WNT3A ( 12 . 5% WNT3A conditioned medium ( CM ) , Figure 1—figure supplement 1B ) and sorted for cells with the highest 2% EGFP fluorescence ( ‘highest 2%’ gate in Figure 1—figure supplement 1F ) . Following two rounds of FACS sorting , three genes were significantly enriched for GT insertions ( Figure 1D and Supplementary file 1 ) . ZNRF3 was the top hit . Eliminating ZNRF3 or RNF43 , two transmembrane E3 ubiquitin ligases , has been shown to amplify WNT signaling by increasing FZD and LRP6 levels on the cell surface ( Hao et al . , 2012; Koo et al . , 2012; Figure 1A ) . Only ZNRF3 is expressed at significant levels in HAP1 cells ( Table 1 ) , explaining why it was a hit in this screen . The second most significant hit of this screen was TCF7L2 , encoding a TCF/LEF family transcription factor that can also function as an attenuating regulator of WNT target genes ( Tang et al . , 2008 ) . Loss of either ZNRF3 or TCF7L2 is predicted to potentiate signaling responses rather than making them WNT-independent , explaining why these genes did not score in the negative regulator screen ( Figure 1C and Supplementary file 1 ) . These findings highlight one of the advantages of using a reporter with a graded output: different regulatory layers in the pathway can be revealed by subtle alterations in selection conditions . In a screen for positive regulators of canonical WNT signaling ( i . e . genes whose disruption reduces signaling output ) , we stimulated mutagenized HAP1-7TGP cells with a near-saturating dose of WNT3A ( 50% WNT3A CM , Figure 1—figure supplement 1B ) and enriched for cells with the lowest 10% reporter fluorescence ( ‘lowest 10%’ gate in Figure 1—figure supplement 1G ) during two sequential rounds of FACS sorting and amplification . Thirty-three genes were significantly enriched for GT insertions in the sorted cells ( Figure 1E and Supplementary file 1 ) . These included genes encoding several known positive regulators of the pathway , such as the WNT co-receptor LRP6 , components of the WNT transcription complex including CTNNB1 , CREBBP and BCL9 , and components of a histone H3 K79 methyltransferase complex including DOT1L and MLLT10 ( Figure 1A ) . As expected , regulators with redundant expression profiles in HAP1 cells , such as FZDs or DVLs ( Table 1 ) , were not recovered in this screen . Increasing the stringency of selection by sorting for cells with the lowest 2% reporter fluorescence ( ‘lowest 2%’ gate in Figure 1—figure supplement 1H ) did not change the results of the screen significantly ( Figure 1F and Supplementary file 1 ) ; despite differences in the significance of GT insertion enrichment compared to the less stringent screen , the order of the top hits was generally maintained . Considering the multiple experimental steps involved , the results are remarkably reproducible . Henceforth , we refer to each of these two screens for positive regulators as the 'low stringency' ( Figure 1E ) and the 'high stringency' ( Figure 1F ) screen , respectively , and to both of them jointly as the ‘WNT screens’ . The second most significant hit in both WNT screens for positive regulators , following LRP6 , was TFAP4 ( Figure 1E and F , and Supplementary file 1 ) , a gene encoding a transcription factor not previously implicated in regulation of canonical WNT signaling . The fourth and third most significant hit in the low ( Figure 1E ) and high ( Figure 1F ) stringency WNT screens , respectively , was AXIN2 , encoding the CTNNB1 destruction complex scaffold AXIN2 . It was perplexing to find AXIN2 in a screen for positive regulators , since components of the destruction complex are negative regulators of the pathway , as illustrated by the presence of APC and CSNK1A1 in our initial screen ( Figure 1C ) . Experimental validation and analysis of both TFAP4 and AXIN2 follows in the two sections below . These results establish that reporter-based , genome-wide forward genetic screens in haploid human cells are an effective way to identify many non-redundant components of signaling pathways . The versatility afforded by the combination of a reporter with a continuous fluorescence readout and FACS as a means of enrichment enables identification of functionally distinct classes of genes including negative , attenuating and positive regulators . The second most significant hit in both screens for positive regulators of WNT signaling ( Figure 1E and F , and Supplementary file 1 ) , was the gene encoding the transcription factor TFAP4 , outranked only by the gene encoding the WNT co-receptor LRP6 . TFAP4 is a helix-loop-helix leucine zipper transcription factor and a target of MYC ( Jung and Hermeking , 2009 ) that has been implicated in epithelial-to-mesenchymal transformation and metastasis in CRC ( Jackstadt et al . , 2013; Shi et al . , 2014 ) . Despite multiple reports correlating TFAP4 expression and malignancy in gastrointestinal tumors ( Cao et al . , 2009; Liu et al . , 2012; Xinghua et al . , 2012 ) , TFAP4 has not been previously implicated as a regulator of canonical WNT signaling . We used CRISPR/Cas9 to generate two HAP1-7TGP cell lines , designated TFAP4CR-1 and TFAP4CR-2 , the first of which lacks TFAP4 and the second of which produces a truncated protein product that retains the leucine zipper motif ( see Materials and methods , Figure 2C and Supplementary file 2 ) . We note that these and all the other cell lines generated using CRISPR/Cas9 and used in this work were isolated without any phenotypic selection and were genotyped by sequencing the single allele of the disrupted gene ( see Materials and methods and Supplementary file 2 ) . TFAP4CR-1 and TFAP4CR-2 cells showed a substantial reduction in WNT3A-induced expression of endogenous AXIN2 , a target gene commonly used as a metric for pathway activity ( Figure 2A ) . The defect in target gene induction correlated with the severity of the two mutant alleles of TFAP4 . WNT3A-induced reporter activation in TFAP4CR-1 cells could be rescued by re-expression of TFAP4 ( Figure 2B ) . TFAP4 overexpression in WT HAP1-7TGP cells increased WNT3A-induced reporter signal by 2 . 6-fold but did not induce reporter activity in unstimulated cells ( Figure 2B ) , suggesting TFAP4 is a limiting factor for WNT signaling in these cells . The gain- and loss-of-function effects of TFAP4 demonstrate an important regulatory role in WNT signaling in human cells , consistent with its prominent position among the hits of the WNT screens—and indeed of several other screens described later in this work . 10 . 7554/eLife . 21459 . 006Figure 2 . The transcription factor TFAP4 regulates WNT signaling downstream of the CTNNB1 destruction complex ( A–D ) , and ectopic expression of TFAP4 , SERBP1 and HUWE1 in X . laevis embryos induces secondary body axis formation ( E–F ) . ( A ) AXIN2 mRNA ( average ± standard deviation ( SD ) AXIN2 mRNA normalized to HPRT1 mRNA , each measured in triplicate reactions ) , relative to untreated WT cells , for single WT HAP1-7TGP and TFAP4CR clonal cell lines ( see Materials and methods and Supplementary file 2 for descriptions of all CRISPR/Cas9-engineered and GT-containing clonal cell lines ) . Cells were treated with 50% WNT3A CM where indicated . ( B ) WNT reporter activity ( median ± standard error of the median ( SEM ) EGFP fluorescence from 1000 transfected cells ) , relative to untreated WT cells transfected with empty vector , for WT and TFAP4CR-1 cells transfected with pCS2+ empty vector or pCSDest-TFAP4 ( together with pmCherry as a co-transfection marker ) . Cells were treated with 50% WNT3A CM where indicated . ( C ) Immunoblot analysis of WT and TFAP4CR clonal cell lines treated with 50% WNT3A CM where indicated . CTNNB1 protein levels ( CTNNB1 intensity normalized to GAPDH intensity ) , relative to untreated WT cells , are shown below the blots . Molecular weight standards ( in kilodaltons ( kDa ) ) are indicated on the left and the identity of the protein measured in each blot is indicated on the right . ( D ) WNT reporter activity ( median ± SEM EGFP fluorescence from 10 , 000 WNT3A- or LiCl-treated cells , or from 2800 cells transfected with non-degradable ( ND , S33Y mutant ) CTNNB1 ) for WT and TFAP4CR-1 cells , depicted as percentage of WT . Cells were treated with 50% WNT3A CM or with 40 mM of the GSK3 inhibitor LiCl , or they were transfected with ND CTNNB1 and pmCherry as a co-transfection marker . ( E ) Four-cell stage X . laevis embryos were injected ventrally with 5 ng of mRNA encoding yellow fluorescent protein ( YFP ) , X . laevis Wnt8 , TFAP4 , SERBP1 or HUWE1 and grown to stage 34 . Dorsal ( top panel of each pair ) and lateral ( bottom panel of each pair ) views for groups of three embryos are shown . Scale bar = 1 mm . ( F ) Percentage of embryos with a secondary body axis . The total number of injected embryos is indicated below the group name . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 006 Because it is a transcription factor , TFAP4 is likely to function downstream of the destruction complex ( Figure 1A ) . Depletion and truncation of TFAP4 in TFAP4CR-1 and TFAP4CR-2 cells , respectively , did not affect WNT-dependent accumulation of CTNNB1 protein , a measure of destruction complex activity ( Figure 2C ) . We confirmed this conclusion by epistasis analysis , in which we activated signaling in WT HAP1-7TGP and TFAP4CR-1 cells at various levels of the pathway by 1 ) addition of WNT3A , which acts at the cell surface , 2 ) addition of the GSK3 inhibitor LiCl , which inactivates the destruction complex , or 3 ) transfection with a non-degradable ( ND ) , constitutively active CTNNB1 mutant ( S33Y ) , which activates the transcription complex directly ( Figure 2D ) . The response of TFAP4CR-1 cells was reduced in all cases when compared to WT HAP1-7TGP cells . Thus , TFAP4 must act together with or downstream of CTNNB1 . To test if TFAP4 can influence WNT signaling during development , we employed an established body axis duplication assay in Xenopus laevis embryos . Activation of WNT signaling in the dorsal side of the early X . laevis embryo is a critical event in the formation of the Spemann organizer , an important tissue-organizing center found in vertebrates ( Spemann , 1938 ) , and ectopic activation of WNT signaling in the ventral side leads to formation of a second body axis . Microinjection of mRNA encoding TFAP4 into X . laevis embryos caused the formation of a secondary body axis ( Figure 2E and F ) , demonstrating that TFAP4 can promote ectopic activation of WNT signaling during development . Future work will focus on defining the contexts in which TFAP4 regulates WNT transcriptional responses under physiological and pathological conditions , given that its site of action downstream of the CTNNB1 destruction complex could be favorable for therapeutic intervention in cancers where WNT signaling is activated by loss of APC or by mutations that stabilize CTNNB1 . AXIN genes encode the principal scaffold ( reviewed in Tacchelly-Benites et al . , 2013; Song et al . , 2014 ) and limiting component ( Lee et al . , 2003 ) of the CTNNB1 destruction complex . The two paralogues in mammals , AXIN1 and AXIN2 , are functionally redundant ( Chia and Costantini , 2005 ) , although their expression patterns are quite distinct: AXIN1 is expressed ubiquitously , while AXIN2 is expressed at low levels in the absence of WNT signals ( Jho et al . , 2002 ) . AXIN2 is also the key component of a negative feedback loop in the WNT pathway ( Lustig et al . , 2002 ) . As a universal and direct target gene of the pathway , its increased expression following stimulation with WNT can lead to elevated levels of the destruction complex and , consequently , reduced levels of CTNNB1 . Given this well-established negative regulatory role , the enrichment of GT insertions mapping to AXIN2 in HAP1-7TGP cells with reduced WNT reporter activity recovered during the WNT screens ( Figure 1E and F , and Supplementary file 1 ) presented us with a paradox . An important clue emerged from a careful inspection of the distribution of GT insertions mapping to AXIN2 in the sorted cells . In most hits from haploid genetic screens , exemplified by LRP6 ( Figure 3A ) , GT insertions cluster at the 5’ end of the gene because of the propensity of retroviral integration near transcriptional start sites and because such insertions are likely to generate null alleles ( Carette et al . , 2011a ) . Contrary to this general case , nearly all GT insertions in AXIN2 mapped to the opposite end of the gene in the last intron ( Figure 3A ) . These insertions are predicted to produce a truncated AXIN2 protein product lacking exon 11 ( Figure 3B ) , comprising half of the DAX domain , which has been implicated both in CTNNB1 destruction complex function and in interactions with the receptor complex at the plasma membrane ( reviewed in Tacchelly-Benites et al . , 2013; Song et al . , 2014 ) . 10 . 7554/eLife . 21459 . 007Figure 3 . The C-terminal DAX domain of AXIN2 controls CTNNB1 transcriptional activity . ( A ) GT insertions in LRP6 ( top histogram ) and AXIN2 ( bottom histogram ) mapped for the sorted cells from the WNT positive regulator , low stringency screen ( Figure 1E ) . The histograms depict the number of GT integrations in the sense ( blue ) or antisense ( red ) orientation , relative to the coding sequence of the gene , within consecutive 500 base pair ( bp ) intervals along the length of each gene . Due to the directionality of the splice acceptor in the GT , typically only sense GT insertions in introns disrupt the gene , whereas GT insertions in exons generally disrupt the gene regardless of orientation . RefSeq gene tracks for LRP6 and AXIN2 are shown beneath each histogram following the University of California , Santa Cruz ( UCSC ) genome browser display conventions: coding exons are represented by thick blocks , UTRs by thin blocks , and introns by horizontal lines connecting the blocks . Both genes are displayed with their 5’ ends to the left , and encompass chromosome 12 , bps 12267499–12116000 for LRP6 , and chromosome 17 , bps 65561999–65528500 for AXIN2 ( hg18 ) . ( B ) Schematic representation of the human AXIN2 protein drawn to scale in the horizontal dimension . Amino acid numbers are indicated below , and arrows show the sites at which truncations were made by CRISPR/Cas9-mediated genome editing in the indicated cell lines . Known domains , regions and motifs ( based on UniProt annotation ) are depicted in gray ( TB , tankyrase-binding motif ) . Exon 11 , eliminated by GT insertions found in cells sorted during the WNT screens for positive regulators , is delineated by a thinner white block . ( C ) Fold-induction in WNT reporter ( median EGFP fluorescence from 20 , 000 cells ) following treatment with 50% WNT3A CM or 10 μM of the GSK3 inhibitor CHIR-99021 . Each circle represents a unique clonal cell line ( determined by genotyping , Supplementary file 2 ) , and the average of three to four independent clones for each genotype is indicated by a horizontal line . For each treatment , percentage reporter activation relative to WT cells is also indicated above each group of circles to facilitate comparisons . Significance was determined by one-way ANOVA , and is indicated as **** ( p<0 . 0001 ) or ns ( not significant ) . ( D ) Fold-induction in soluble CTNNB1 protein ( average CTNNB1 intensity normalized to ACTIN intensity from duplicate immunoblots ) following treatment with 50% WNT3A CM or 10 μM CHIR-99021 . Each circle represents a unique clonal cell line , and the average of two independent clones for each genotype is indicated by a horizontal line . Significance was determined by unpaired t-test with Welch’s correction . Representative immunoblots used for quantification of CTNNB1 and ACTIN are shown in Figure 3—figure supplement 1C . ( E ) Nuclear CTNNB1 ( average nuclear fluorescence per unit area from three fields of view ) in single clonal cell lines of the indicated genotypes was quantified as described in Materials and methods . Cells were treated with 50% WNT3A CM where indicated . For each cell line , the fold-increase in CTNNB1 nuclear accumulation following treatment with WNT3A , expressed as percentage of WT , is also indicated above the bars to facilitate comparisons . For WNT3A-treated cells , differences in nuclear CTNNB1 between WT and mutant cells were not statistically significant as determined by one-way ANOVA . Examples of confocal sections used for quantification of nuclear CTNNB1 are shown in Figure 3—figure supplement 2B–2D . ( F ) WNT reporter activity ( median EGFP fluorescence from 10 , 000 cells ) , soluble CTNNB1 protein ( average CTNNB1 intensity normalized to ACTIN intensity from duplicate immunoblots ) , and nuclear CTNNB1 protein ( average nuclear fluorescence per unit area from 2 to 3 fields of view ) , depicted as percentage of WT , for cells treated with 50% WNT3A CM . Each circle represents a unique clonal cell line , and where applicable the average of two independent clones is indicated by a horizontal line . ( G ) Adult D . melanogaster wings expressing Axin-V5 ( top images ) or AxinΔC-V5 ( bottom images ) under the control of the c765-Gal4 driver . Loss of sensory bristles and tissue at the wing margin , indicative of impaired Wg signaling , is shown ( arrow ) in the higher magnification view of the delineated area . While loss of Notch signaling can also result in wing margin defects due to a requirement of Notch signaling for Wg expression at the dorso-ventral boundary in the wing imaginal disc ( Diaz-Benjumea and Cohen , 1995 ) , we ruled out that possibility by confirming intact Wg expression in the wing imaginal disc of flies expressing AxinΔC-V5 ( Figure 3—figure supplement 3B ) . Scale bars = 20 μm . ( H ) Percentage of flies with wing margin defects . 4 . 3% of flies expressing Axin-V5 exhibited loss of bristles at the wing margin , but no loss of wing tissue; 46 . 6% of flies expressing AxinΔC-V5 exhibited loss of bristles at the wing margin as well as loss of wing tissue . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 00710 . 7554/eLife . 21459 . 008Figure 3—figure supplement 1 . AXIN1 and AXIN2 are redundant in haploid human cells ( A–B ) , and CTNNB1 protein is stabilized normally in AXIN2∆C cells ( C ) . ( A ) Immunoblots of total AXIN1 , AXIN2 and GAPDH in various AXIN1 and/or AXIN2 deletion and/or truncation cell lines . GAPDH loading controls are shown below their respective blots . Single WT and AXIN1KO; AXIN2KO clonal cell lines , and two independent clonal cell lines for all other genotypes were analyzed . Genotypes are indicated above the blots . Cells were treated with 10 μM CHIR-99021 to increase AXIN2 expression , since under unstimulated conditions AXIN2 levels were almost undetectable . Both AXIN1 and AXIN2 are expressed in HAP1-7TGP cells . Disruption of AXIN1 led to a marked increase in full-length and truncated AXIN2 protein abundance . ( B ) WNT reporter activity ( median EGFP fluorescence from 10 , 000 cells ) , relative to untreated WT cells . Cells were treated with 50% WNT3A CM where indicated . Each circle represents a unique clonal cell line ( see Supplementary file 2 ) . Single WT and AXIN1KO; AXIN2KO cell lines were analyzed . For AXIN2KO and AXIN1KO cells , the average of two independent clonal cell lines is indicated by a horizontal line . The top graph shows an expanded view of the y-axis to clearly show low levels of reporter activity for untreated cells . AXIN2 and AXIN1 are functionally redundant in HAP1-7TGP cells . ( C ) Representative immunoblots of soluble CTNNB1 and ACTIN used for quantification in Figure 3D . Two WT and two AXIN2∆C clonal cell lines were analyzed . Cells were treated with 50% WNT3A CM ( ‘W’ ) or 10 μM CHIR-99021 ( ‘C’ ) where indicated . CTNNB1 protein was stabilized normally in AXIN2∆C cells following both treatments . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 00810 . 7554/eLife . 21459 . 009Figure 3—figure supplement 2 . Cells lacking the C-terminal DAX domain of AXIN2 exhibit normal CTNNB1 nuclear accumulation following WNT treatment . ( A ) Kinetics of CTNNB1 nuclear accumulation following treatment with 50% WNT3A CM . Nuclear CTNNB1 ( average nuclear fluorescence per unit area from three fields of view in a . u . ) was quantified as described in Materials and methods . In all subsequent experiments , nuclear CTNNB1 was measured following 4 hr of treatment with WNT3A . ( B–F ) Single confocal sections though the center of the nucleus for the indicated cell lines , stained with DAPI ( left images ) , immunostained for CTNNB1 ( center images ) and merged ( right images ) . All images for a given channel were acquired using identical microscopy settings , and brightness and contrast was adjusted equally for display purposes . Cells were treated with 50% WNT3A CM for 4 hr where indicated . Representative regions demarcated by dashed squares ware magnified 3x and are shown in the insets . Scale bar = 18 μm in the main images , and 6 μm in the insets . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 00910 . 7554/eLife . 21459 . 010Figure 3—figure supplement 3 . Expression of an Axin transgene encoding C-terminally truncated protein compromises Wg signaling in D . melanogaster . ( A ) Schematic representation of the protein products encoded by transgenes Axin-V5 and AxinΔC-V5 drawn to scale in the horizontal dimension . Amino acid numbers are indicated below and known domains ( based on UniProt annotation ) are depicted in gray . The C-terminal 41 amino acids truncated in AxinΔC-V5 correspond to those encoded by human exon 11 , eliminated by GT insertions found in cells sorted during the WNT screens for positive regulators ( see Figure 3A and B ) . ( B ) Confocal images of third larval instar wing discs expressing Axin-V5 ( top images ) or AxinΔC-V5 ( bottom images ) using the c765-Gal4 driver , stained with Senseless ( Sens , left images ) or Wingless ( Wg , right images ) antibodies . Scale bars = 10 μm . In the fly imaginal wing disc , Wg secreted by cells at the dorso-ventral boundary controls expression of Sens in adjacent sensory organ precursor cells . Expression of Axin∆C-V5 , but not WT Axin-V5 , resulted in loss of Sens expression ( arrows ) . Expression of Wg itself was unaffected by expression of Axin∆C-V5 , indicative of reduced Wg signaling rather than impaired morphogen secretion . ( C ) Confocal images of stage 8 , 9 and 11 embryos ( embryonic stage of development and hours after egg lay ( AEL ) are indicated above the images ) expressing Axin-V5 or AxinΔC-V5 ( as indicated to the left of the images ) using the mat-Gal4 driver , stained with Engrailed ( En , top four rows of images ) or Wg ( bottom two rows of images ) antibodies . For En staining , high-magnification views are also shown ( second and fourth rows of images ) . Anterior is to the left , dorsal on top . Scale bars = 25 μm . Wg signaling in segmental stripes is required for maintenance , but not initiation , of En expression . In embryos expressing Axin-V5 , En stripes two to three cells wide were observed at stages 8 , 9 and 11 ( dumbbell bars ) . In embryos expressing Axin∆C-V5 , initiation of En expression at stage eight was normal but decayed after stage 9 ( arrows ) , indicating defective Wg signaling . Expression of Wg itself in segmental stripes was unaffected by expression of Axin∆C-V5 , suggesting that the loss of En was due to reduced Wg signaling output as opposed to defective morphogen production . ( D ) Quantification of embryos displaying normal En stripes at stages 9 and 11 . The number of embryos quantified for each genotype and stage is indicated above each bar . ( E ) Immunoblot analysis of V5-fusion proteins and tubulin in lysates prepared from embryos ( collected at 0–2 hr AEL ) expressing Axin-V5 or Axin∆C-V5 . ( F ) Quantification of immunoblot analysis shown in E . Axin-fusion protein ( average ± SEM of V5 intensity from three biological replicates ) is depicted as percentage of WT . Deleting the C-terminal domain of Axin did not affect expression levels . ( G ) First instar larval cuticles from embryos expressing Axin-V5 ( top images ) or AxinΔC-V5 ( bottom images ) . The right image in each pair is a higher magnification view of the left image . Scale bars = 25 μm . The normal pattern of alternating naked cuticle and denticle belts , established by Wg-dependent specification of cell fate , was disrupted by expression of Axin∆C-V5 . ( H ) Embryonic hatch rate ( percentage of fertilized embryos that hatched within 24 hr AEL ) . The number of embryos quantified for each genotype is indicated above each bar . The embryonic hatch rate of embryos expressing Axin∆C-V5 was substantially reduced . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 010 The results of the WNT screens ( Figure 1E and F , and Supplementary file 1 ) suggested that deletion of half of the DAX domain of AXIN2 reduces responsiveness to WNT . The ease of genome editing in haploid cells enabled us to precisely assess the magnitude of signaling defects by comparing multiple independent WT and mutant HAP1-7TGP clonal cell lines containing specific mutations in the single endogenous AXIN2 allele , which minimized the likelihood of non-specific effects and accounted for interclonal variability . We used CRISPR/Cas9 to generate HAP1-7TGP cell lines lacking exon 11 of AXIN2 ( designated AXIN2∆C , Figure 3B and Supplementary file 2 ) . Since incomplete protein domains can remain unfolded , we also generated multiple cell lines lacking the complete DAX domain ( AXIN2∆DAX ) and , as a control , cell lines lacking AXIN2 entirely ( AXIN2KO ) ( Figure 3B and Supplementary file 2 ) . AXIN2KO clones exhibited no defect in WNT3A-dependent reporter induction ( Figure 3C ) due to redundancy with AXIN1 ( Figure 3—figure supplement 1A and B , and Table 1 ) . However , AXIN2∆C and AXIN2∆DAX clones exhibited , on average , a 65% and a 76% reduction in WNT3A-induced signaling , respectively , compared to WT HAP1-7TGP cells ( Figure 3C ) . All AXIN2 mutant cell lines responded normally to the GSK3 inhibitor CHIR-99021 ( Figure 3C ) , demonstrating that the reduction in WNT3A-induced signaling was not due to defects in downstream steps or to irrelevant factors affecting reporter fluorescence . The fact that C-terminally truncated AXIN2 reduced signaling induced by WNT3A explained why GT insertions generating this unusual AXIN2 allele were enriched in the WNT screens for positive regulators . Since HAP1-7TGP cells also express AXIN1 ( Figure 3—figure supplement 1A and Table 1 ) , the effect of C-terminally truncated AXIN2 must be dominant . We asked whether the reduction in WNT signaling caused by truncated AXIN2 was due to a defect in WNT3A-induced CTNNB1 stabilization . Surprisingly , WNT3A robustly promoted the stabilization of soluble CTNNB1 protein in AXIN2∆C cells ( Figure 3D and Figure 3—figure supplement 1C ) , revealing a disconnect between CTNNB1 protein abundance and transcriptional activity . CTNNB1 accumulation in response to CHIR-99021 was also normal in AXIN2∆C cells ( Figure 3D , and Figure 3—figure supplement 1C ) . The discrepancy between CTNNB1 stability and transcriptional activity was not caused by defective nuclear accumulation; in both AXIN2∆C and AXIN2∆DAX cells , accumulation of nuclear CTNNB1 following WNT treatment was largely normal ( Figure 3E and Figure 3—figure supplement 2 ) . We conclude from these experiments that deleting the DAX domain of AXIN2 dissociates CTNNB1 protein abundance from its transcriptional activity . These effects could reflect an autonomous property of the AXIN2 protein lacking the DAX domain , or a more complex interaction with the remaining WT AXIN1 . To rule out confounding effects due to AXIN1 , we disrupted the single allele of AXIN1 in individual AXIN2∆C and AXIN2∆DAX clonal cell lines to generate double-mutant cell lines ( AXIN1KO; AXIN2∆C and AXIN1KO; AXIN2∆DAX , respectively , Supplementary file 2 and Figure 3—figure supplement 1A ) . The only AXIN protein present in these cells is truncated AXIN2 lacking either half or the entire DAX domain . Truncated AXIN2 caused the same effects upon elimination of AXIN1—decreased WNT3A-induced reporter activity despite normal accumulation of soluble and nuclear CTNNB1 ( Figure 3F and Figure 3—figure supplement 2 ) . These findings provide further evidence that AXIN2 truncations disrupt CTNNB1-mediated transcription through a novel mechanism intrinsic to this allele . We tested the generality of our findings by introducing Axin transgenes into the fly Drosophila melanogaster , a model organism that has been used extensively for genetic studies of WNT signaling . We generated a transgene encoding an epitope-tagged ( V5 ) fusion of the single D . melanogaster Axin protein lacking the last 41 amino acids ( Axin∆C-V5 , Figure 3—figure supplement 3A ) . These amino acids correspond to those encoded by exon 11 of human AXIN2 , the exon disrupted by GT insertions in the WNT screens ( Figure 3A and B ) . Expression of Axin∆C-V5 impaired Wingless ( Wg , fly WNT ) signaling based on target gene expression and phenotypic readouts during both embryonic and larval development ( Figure 3G and H , and Figure 3—figure supplement 3B–H ) . The observed defects in Wg signaling were not due to decreased expression of Wg itself ( Figure 3—figure supplement 3B and C ) or to increased expression of Axin∆C-V5 protein ( Figure 3—figure supplement 3E and F ) . In control experiments , expression of WT Axin-V5 ( Figure 3—figure supplement 3A ) at physiological levels ( Wang et al . , 2016 ) using the same promoter as for Axin∆C-V5 did not disrupt Wg signaling ( Figure 3G and H , and Figure 3—figure supplement 3B–H ) . The fact that in flies , like in haploid human cells , expression of an Axin transgene lacking the C-terminal domain reduces Wg signaling even in the presence of endogenous Axin is consistent with a dominant effect that restrains CTNNB1 transcriptional activity . In summary , unbiased genome-wide screens for positive regulators of WNT signaling uncovered an unsuspected role for the C-terminal DAX domain of AXIN2 in controlling CTNNB1 transcriptional activity , since deletion of this domain led to a severely compromised transcriptional response despite normal accumulation of CTNNB1 protein . While the mechanistic basis of this process remains to be elucidated , our results cannot be explained by previously described functions of the AXIN DAX domain . The DAX domain has been implicated in AXIN polymerization , in interactions with DVL , and in mediating an intramolecular , auto-inhibitory interaction that allows the receptor complex to inactivate the destruction complex in a catalytic manner ( Fiedler et al . , 2011; Kim et al . , 2013 ) . These models predict that loss of the DAX domain would impair communication between the receptor complex and the destruction complex , leading to defective WNT-induced CTNNB1 stabilization , in contrast to our results ( Figure 3D and F ) . Thus , the disconnect between CTNNB1 abundance and transcriptional activity caused by deletion of the DAX domain demonstrates a new biochemical function for this domain . The discovery of this dominant allele of AXIN2 was made possible by the untargeted nature of GT-based insertional mutagenesis and thus would not have emerged from strict loss-of-function screens such as those mediated by RNA interference or CRISPR/Cas9 . Given that rare dominant alleles can provide mechanistic insights distinct from those afforded by null alleles , our findings justify the design of comprehensive ‘exome-wide’ sgRNA libraries for CRISPR/Cas9-based screens . RSPOs are stem cell growth factors that potentiate responses to WNT ligands by binding to LGR-family receptors and neutralizing the ZNRF3 and RNF43 E3 ubiquitin ligases to increase levels of WNT receptors on the cell surface ( reviewed in de Lau et al . , 2014; Figure 1A ) . Recurrent translocations in genes encoding RSPOs are found in some colorectal tumors and targeting the resulting fusion proteins blocks tumorigenesis ( Seshagiri et al . , 2012; Storm et al . , 2016 ) . Mutations in RNF43 that mimic the effect of stimulation with RSPO have also been reported in multiple cancers ( Giannakis et al . , 2014 ) . Regulators of RSPO-enhanced WNT signaling could therefore be important both in normal physiological and in pathological contexts . HAP1-7TGP cells were responsive to RSPO-mediated effects on WNT signaling . RSPO1 markedly amplified the reporter response to low concentrations of WNT3A CM but was completely inactive in the absence of WNT ( Figure 4—figure supplement 1A ) . We determined the concentration of WNT3A CM at which responsiveness to RSPO1 was maximal ( Figure 4—figure supplement 1B ) and used these conditions in a sensitized genome-wide screen for mediators of RSPO-enhanced WNT signaling . Notably , the concentration of WNT used in this screen , henceforth referred to as the ‘low WNT + RSPO screen , ’ was 49-fold lower than that used in the WNT screens for positive regulators ( Figure 1E and F ) . Following treatment with WNT3A CM plus RSPO1 , we isolated cells with the lowest 7% EGFP fluorescence ( ‘lowest 7%’ gate in Figure 4—figure supplement 1C ) . After four consecutive sorts , which resulted in a marked enrichment of cells with diminished responsiveness to WNT3A CM plus RSPO1 ( Figure 4—figure supplement 1C ) , we sequenced and mapped GT integrations ( Figure 4A and Supplementary file 1 ) . Reassuringly , the top hit was LGR4 , the gene encoding the RSPO1 receptor , confirming that the screen was sensitive to requirements for RSPO1-dependent signaling . Top hits of this screen included many genes encoding known WNT regulators also uncovered in the WNT screens for positive regulators ( Figure 1E and F , and Supplementary file 1 ) : LRP6 , AXIN2 , DOT1L , MLL , CTNNB1 , CREBBP , BCL9 and RNF146 . TFAP4 , encoding the required transcription factor we described earlier in this study , was also among the top hits ( Figure 4A and Supplementary file 1 ) . 10 . 7554/eLife . 21459 . 011Figure 4 . A comparative analysis of screens uncovers requirements for RSPO-potentiated signaling in response to low levels of WNT . ( A ) Circle plot depicting genes enriched for GT insertions in the low WNT + RSPO screen for regulators of RSPO-enhanced WNT signaling . See legend to Figure 1C–F for details . ( B ) Heat map comparing the two WNT positive regulator screens ( Figure 1E and F ) and the low WNT + RSPO screen ( Figure 4A ) . Genes enriched for GT insertions ( FDR-corrected p-value<10−4 ) in at least one of the three screens were clustered based on their IGTIOB score in each screen ( see Materials and methods and Supplementary file 3 ) . A group of genes preferentially enriched for GT insertions in the low WNT + RSPO screen is indicated , headlined by the RSPO receptor LGR4 . Genes selected for a detailed analysis are labeled in red . ( C ) Fold-induction in WNT reporter ( average EGFP fluorescence from 10 , 000 cells ) following treatment with 50% WNT3A CM , 12 . 5% WNT3A CM , 1 . 43% WNT3A CM + 20 ng/ml RSPO1 or 10 μM of the GSK3 inhibitor CHIR-99021 , expressed as percentage of the average for WT cells to facilitate comparisons . Each circle represents the fold-induction for a unique clonal cell line ( determined by genotyping , Supplementary file 2 ) , and where applicable , the average of two to four independent clones for each genotype is indicated by a horizontal line . Significance was determined by one-way ANOVA , and is indicated as **** ( p<0 . 0001 ) , * ( p<0 . 05 ) or ns ( not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 01110 . 7554/eLife . 21459 . 012Figure 4—figure supplement 1 . Comparative genetic screens uncover requirements for RSPO-potentiated signaling in response to low levels of WNT . ( A ) Dose-response curves depicting WNT reporter activity ( average EGFP fluorescence from 10 , 000 cells in a . u . ) following treatment with the indicated concentrations of WNT3A CM ( key on the right , expressed as percentage of total medium ) , as a function of RSPO1 concentration ( in ng/ml ) . RSPO1-mediated effects in HAP1-7TGP cells are completely dependent on WNT . ( B ) RSPO1-dependent WNT reporter activity ( average EGFP fluorescence from 10 , 000 cells following treatment with 5 ng/ml RSPO1 , divided by EGFP fluorescence in the absence of RSPO1 ) as a function of WNT3A CM concentration ( expressed as percentage of total medium ) . Treatment with 1 . 04% WNT3A CM , the concentration used in the low WNT + RSPO screen ( Figure 4A ) , resulted in negligible pathway activation on its own , but a robust 14 . 3-fold increase in WNT reporter fluorescence in conjunction with RSPO1 . ( C ) Histogram depicting WNT reporter activity from ~20 , 000 mutagenized HAP1-7TGP cells enriched during the low WNT + RSPO screen ( Figure 4A ) . Cells were treated with 1 . 04% WNT3A CM or 1 . 04% WNT3A CM + 10 ng/ml RSPO1 where indicated . Unsorted cells and cells amplified following sort four using the indicated lowest 7% EGFP fluorescence gate are depicted , and the percentage of cells within this gate for each population is indicated in parenthesis . ( D ) Schematic of the mammalian GPI biosynthesis pathway ( adapted with permission from Essentials of Glycobiology , second edition , Cold Spring Harbor Laboratory Press , Cold Spring Harbor , New York , © 2009 by The Consortium of Glycobiology Editors , La Jolla , California ) . Proteins involved in each enzymatic step are indicated next to the arrows , and the products of genes with an FDR-corrected p-value<0 . 05 for GT insertion enrichment in the low WNT + RSPO screen are labeled in red . Key at bottom right . ( E ) Dose-response curves depicting WNT reporter activity ( average EGFP fluorescence from 10 , 000 cells in a . u . ) following treatment with the indicated concentrations of WNT3A CM ( expressed as percentage of total medium ) in the absence of RSPO1 , for a single LGR4KO; LGR5KO; LGR6KO clonal cell line and two independent clonal cell lines for all other genotypes . PIGLKO and GPC4KO; GPC6KO cells manifested a pronounced reduction in signaling when stimulated with low doses of WNT3A alone . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 012 We distinguished genes selectively required for RSPO-enhanced WNT signaling through a comparative analysis of the low WNT + RSPO screen ( Figure 4A and Supplementary file 1 ) and both WNT screens for positive regulators conducted at a near-saturating dose of WNT ( Figure 1E and F , and Supplementary file 1 ) . For this analysis , we used two different measures of gene disruption caused by GT integrations ( see Materials and methods , Figure 4B and Supplementary file 3 ) . First , the FDR-corrected p-value , reflecting enrichment of GT integrations in the sorted vs . the unsorted cells from each screen , was used to set a stringent cutoff for inclusion of hits in the analysis . Second , an Intronic GT Insertion Orientation Bias ( IGTIOB ) score , reflecting enrichment of sense vs . antisense GT integrations ( relative to the coding sequence of the gene ) in introns only for the sorted cells from each screen , was used to compare hits between screens . The IGTIOB score relies on the fact that generally only sense GT insertions in introns should inactivate genes due to the directionality of the splice acceptor . Genes selectively required for RSPO-enhanced WNT signaling should show a pattern of GT enrichment similar to that of the gene encoding the RSPO receptor , LGR4 . Conversely , regulators required for WNT signaling under all treatment conditions should be equally enriched for GT integrations in all screens ( Figure 4B and Supplementary file 3 ) . The most striking outcome of this analysis was the identification of multiple genes encoding components of the GPI-anchor biosynthetic pathway that were enriched for GT insertions in the low WNT + RSPO screen but not the WNT screens ( Figure 4B and Supplementary file 3 ) . Fourteen genes in the GPI biosynthesis pathway had an FDR-corrected p-value<0 . 05 ( Supplementary file 1 and Figure 4—figure supplement 1D ) . Therefore , a GPI-anchored protein may be particularly important in mediating signaling triggered by a combination of RSPO and a low dose of WNT . The gene encoding the glypican GPC4 , a GPI-linked heparan sulfate proteoglycan ( HSPG ) , was also preferentially enriched for GT insertions in the low WNT + RSPO screen ( Figure 4B and Supplementary file 3 ) , with an FDR-corrected p-value=1 . 92x10−5 , more significant than those of established WNT signaling components such as CREBBP and BCL9 ( Figure 4A and Supplementary file 1 ) . Glypicans are important for concentrating extracellular ligands at the cell surface , and GPC4 has been proposed to bind and concentrate WNT3A and WNT5A in the vicinity of their receptors ( Sakane et al . , 2012 ) . However , since neither GPI biosynthesis nor glypican genes were significant hits in the WNT screens for positive regulators ( Figure 1E and F , and Supplementary file 1 ) , we hypothesized that these genes must play a crucial role under the conditions of the low WNT + RSPO screen , either by mediating RSPO responsiveness like LGR4 , or by selectively affecting reception of WNT3A at the very low concentration used in this screen . To distinguish between these two possibilities , we analyzed the signaling response to WNT3A alone or to a low concentration of WNT3A plus RSPO1 in clonal HAP1-7TGP cell lines in which we disrupted PIGL , a gene in the GPI biosynthesis pathway ( Figure 4—figure supplement 1D ) , or GPC4 ( designated PIGLKO and GPC4KO , respectively , Supplementary file 2 ) . The glypican GPC6 is redundant with GPC4 in certain contexts ( Allen et al . , 2012 ) , so we also generated HAP1-7TGP cell lines in which we disrupted GPC6 alone or in combination with GPC4 ( designated GPC6KO and GPC4KO; GPC6KO , respectively , Supplementary file 2 ) . As a control we generated a HAP1-7TGP cell line lacking all three RSPO receptors ( designated LGR4KO; LGR5KO; LGR6KO , Supplementary file 2 ) . As expected from the role of LGRs as exclusive mediators of responsiveness to RSPO but not to WNT , LGR4KO; LGR5KO; LGR6KO cells did not respond to RSPO1 in the presence of a low dose of WNT3A , but exhibited no signaling defects when stimulated with higher doses of WNT3A alone ( Figure 4C and Figure 4—figure supplement 1E ) . In contrast , PIGLKO and GPC4KO; GPC6KO cells manifested some reduction in signaling when stimulated with a near-saturating dose of WNT3A alone , but this reduction was more pronounced following treatment with lower doses of WNT3A alone or a low dose of WNT3A combined with RSPO1 ( Figure 4C and Figure 4—figure supplement 1E ) . GPC4KO cells were stimulated normally by a near-saturating dose of WNT3A and exhibited a smaller defect than GPC4KO; GPC6KO when stimulated with a lower dose of WNT3A alone or in combination with RSPO1 , while GPC6KO cells had no signaling defect at all ( Figure 4C ) . These results suggest that GPC4 and GPC6 are partially redundant in HAP1 cells , since they are both expressed albeit at very different levels ( Table 1 ) . In a control experiment , WNT signaling induced by the GSK3 inhibitor CHIR-99021 was largely unaffected in all mutant cell lines ( Figure 4C ) , demonstrating that there were no signaling defects downstream of the receptor complex ( Figure 1A ) . Taken together these results indicate that genes in the GPI biosynthesis pathway and GPC4/6 are required for signaling in response to low levels of WNT , and explain why they may have been more prominent hits in the low WNT + RSPO than in the WNT screens ( Figures 1E , F and 4A and Supplementary file 1 ) . Presently , we cannot confirm or discount an additional , direct contribution of GPC4/6 or another GPI-anchored protein to RSPO reception , as we have been unable to directly measure responses to RSPO alone in HAP1 cells . Yet , the presence in all RSPOs of a thrombospondin domain capable of binding heparin sulfate and mediating interactions with HSPGs such as glypicans ( Nam et al . , 2006; Ohkawara et al . , 2011 ) makes this an intriguing possibility . In summary , our comparative analysis shows that forward genetic screens in haploid human cells are exquisitely sensitive to both the identity and concentration of ligands used to initiate signaling . They can uncover ligand-specific receptors , such as LGR4 , or accessory factors that are rate-limiting for signaling only under specific regimes of ligand concentrations , such as GPI biosynthetic enzymes and glypicans . Of note , the low WNT + RSPO screen ( Figure 4A and Supplementary file 1 ) was sensitive enough to reveal redundant regulators , such as FZD5 and DVL3 ( Table 1 ) , that were not significant hits under the near-saturating WNT3A dose used in the WNT screens ( Figure 1E and F , and Supplementary file 1 ) . Given the potential of comparative screens to identify context-specific regulators , we searched for genes whose inactivation would suppress the pathological signaling that ensues when key negative regulators of the WNT pathway are lost . Since negative regulators such as APC are frequently mutated in cancer , suppressor mutations and the mechanisms through which the affected genes regulate signaling may reveal therapeutic targets . Our initial screen for rate-limiting negative regulators of WNT signaling ( Figure 1C and Supplementary file 1 ) suggested that disruption of the single allele of APC or CSNK1A1 in HAP1-7TGP cells should lead to constitutive activation of the pathway . We designed two screens to uncover suppressors of ligand-independent signaling induced by loss of APC or CSNK1A1 ( Figure 5A ) . We disrupted APC or CSNK1A1 in HAP1-7TGP cells using CRISPR/Cas9 and isolated two clonal cell lines designated APCKO-1 and CSNK1A1KO-1 , respectively ( Supplementary file 2 ) . Sequencing revealed frameshift mutations in the single allele of each gene . We confirmed by immunoblotting that the APC signal was reduced by >96 . 5% in APCKO-1 cells , and CSNK1A1 was undetectable in CSNK1A1KO-1 cells ( Figure 5—figure supplement 1A ) . The level of constitutive WNT reporter fluorescence in both the APCKO-1 and CSNK1A1KO-1 clones was higher than that induced by near-saturating WNT3A or by the GSK3 inhibitor CHIR-99021 in WT HAP1-7TGP cells ( Figure 5—figure supplement 1B ) . 10 . 7554/eLife . 21459 . 013Figure 5 . Suppressor screens in cells lacking APC or CSNK1A1 reveal genotype-specific signaling requirements . ( A ) Schematic of WNT reporter-based suppressor screens . APC or CSNK1A1 was disrupted by CRISPR/Cas9-mediated genome editing of the WNT reporter haploid cell line HAP1-7TGP . Individual clonal cell lines were isolated ( APCKO-1 and CSNK1A1KO-1 , Supplementary file 2 ) and mutagenized using GT retrovirus . Cells with reduced reporter activity were enriched by FACS to identify suppressor mutations . ( B–C ) Circle plots depicting genes enriched for GT insertions in suppressor screens in which constitutive WNT signaling was induced by loss of APC ( B ) or CSNK1A1 ( C ) . See legend to Figure 1C–1F for details . ( D ) Heat map comparing the WNT positive regulator , low stringency screen ( Figure 1E ) , and the APC and CSNK1A1 suppressor screens ( Figure 5B and C ) . Genes enriched for GT insertions ( FDR-corrected p-value<10−4 ) in at least one of the three screens were clustered based on their IGTIOB score in each screen ( see Materials and methods and Supplementary file 3 ) . Classes of genes preferentially enriched for GT insertions in various screens are indicated . Genes selected for a detailed analysis are labeled in red . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 01310 . 7554/eLife . 21459 . 014Figure 5—figure supplement 1 . Suppressor screens in cells lacking APC or CSNK1A1 reveal genotype-specific signaling requirements . ( A ) Immunoblot analysis of the APCKO-1 ( left blots ) and CSNK1A1KO-1 ( right blots ) cell lines used for suppressor screens . For the APC blot , Ponceau S staining was used to assess loading . For the CSNK1A1 blot , the appropriate lanes from the CSNK1A1 blot shown in Figure 6H and the corresponding GAPDH loading controls ( not pictured in Figure 6H ) were cropped and juxtaposed . ( B ) Histogram depicting WNT reporter activity for ~22 , 500 WT HAP1-7TGP , APCKO-1 , CSNK1A1KO-1 and APCKO-2 cells . Where indicated , WT cells were treated with 50% WNT3A CM or 10 μM of the GSK3 antagonist CHIR-99021 . ( C–D ) Histograms depicting WNT reporter activity from ~9500 mutagenized APCKO-1 cells ( C ) or ~20 , 000 mutagenized CSNK1A1KO-1 cells ( D ) enriched during the APC and CSNK1A1 suppressor screens , respectively ( Figure 5B and C ) . Unsorted cells and cells amplified following sort 1 or sort 2 using the indicated lowest 10% EGFP fluorescence gate are depicted , and the percentage of cells within this gate for each population is indicated in parenthesis . WT HAP1-7TGP cells are shown for reference . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 014 In two independent screens , henceforth referred to as the ‘APC suppressor screen’ and the ‘CSNK1A1 suppressor screen , ’ respectively , we mutagenized APCKO-1 and CSNK1A1KO-1 cells with GT retrovirus and enriched for cells with the lowest 10% WNT reporter fluorescence ( ‘lowest 10%’ gate in Figure 5—figure supplement 1C and D ) . Following two sequential rounds of sorting and amplification , 43% of APCKO-1 and 42% of CSNK1A1KO-1 cells were within this gate ( Figure 5—figure supplement 1C and D ) . We analyzed the cells sorted during each screen and their respective unsorted control populations for enrichment of GT insertions ( Figure 5B and C , and Supplementary file 1 ) . A three-way comparative analysis of the WNT positive regulator , low stingency screen in WT HAP1-7TGP cells ( Figure 1E ) , the APC suppressor screen and the CSNK1A1 suppressor screen ( all done at similar selection stringencies ) revealed expected similarities and differences based on established epistatic relationships , as well as a number of unexpected findings ( Figure 5D and Supplementary file 3 ) . The isogenic background of the cells in these three screens and the very high statistical significance of the top hits ( only genes with an FDR-corrected p-value<10−4 in at least one screen were included in this analysis ) enabled us to make meaningful predictions , some of which we confirmed experimentally . Several groups of genes were clearly discernible based on their GT insertion enrichment patterns across the three screens ( Figure 5D and Supplementary file 3 ) . As expected , genes encoding components of the pathway that function downstream of the destruction complex , including CTNNB1 and CREBBP , were enriched for GT insertions in all three screens ( Figure 5D and Supplementary file 3 ) . TFAP4 was also significantly enriched for GT insertions in all three screens ( Supplementary file 3 ) , and it acts downstream of the destruction complex , as we confirmed experimentally ( Figure 2 ) . However , TFAP4 had a low IGTIOB score in all screens ( Figure 5D and Supplementary file 3 ) because it represents a rare case of a gene that can be disrupted by both sense and antisense GT insertions in an intron ( Supplementary file 1 ) , a finding that will be described in detail elsewhere . Also as expected , genes encoding components of the pathway upstream of the destruction complex , such as LRP6 , were predominantly enriched for GT insertions in the WNT , but not the APC or CSNK1A1 suppressor screens ( Figure 5D and Supplementary file 3 ) . AXIN2 was also enriched for GT insertions in the WNT screen exclusively , suggesting that the mechanism responsible for reducing WNT responsiveness in cells containing AXIN2 truncations depends on other components of the CTNNB1 destruction complex . Hits enriched for GT insertions in both the APC and CSNK1A1 supprssor screens , but not the WNT screen , likely represent a class of genes capable of modulating WNT signaling in the absence of destruction complex activity . The most prominent hit in this category was SERBP1 ( Figure 5D and Supplementary file 3 ) , encoding an RNA binding protein that has not been previously implicated in WNT signaling . In the section that follows we explored how SERBP1 regulates WNT signaling in cells lacking destruction complex activity . Genes enriched for GT insertions predominantly in the APC suppressor screen included those encoding various RNA binding proteins , components of the mRNA nonsense-mediated decay pathway , and negative regulators of RNA polymerase ( Figure 5D and Supplementary file 3 ) , suggesting a connection between RNA metabolism and signaling in APCKO-1 cells . Surprisingly , there were a number of genes predominantly enriched for GT insertions in the CSNK1A1 suppressor screen ( Figure 5D and Supplementary file 3 ) . Given that the principal role of CSNK1A1 in WNT signaling is thought to be phosphorylation of CTNNB1 through the destruction complex , it was not obvious why these same genes were not enriched for GT insertions in the APC suppressor screen , where destruction complex activity was also disabled . The existence of this class of genes , apparently required for WNT signaling only in cells lacking CSNK1A1 , suggested a role for CSNK1A1 in WNT signaling independent of the destruction complex . The most prominent gene in this class encodes HUWE1 , an E3 ubiquitin ligase that has been proposed to downregulate WNT signaling by ubiquitinating DVL and preventing its multimerization ( de Groot et al . , 2014 ) . In contrast , the fact that mutations of HUWE1 caused a reduction in WNT reporter fluorescence during the CSNK1A1 suppressor screen suggests a positive regulatory role . Below , we describe the peculiar role of HUWE1 in mediating WNT signaling specifically in the context of CSNK1A1 loss . APC is a prototypical human tumor suppressor gene frequently lost in both sporadic and familial CRC . Importantly , reduction of WNT signaling through restoration of APC in a mouse model of CRC can reverse tumorigenesis even in the presence of mutations in other potent cancer genes such as TP53 and KRAS ( Dow et al . , 2015 ) . Hence , genes selectively required to sustain the high-level WNT signaling that ensues when APC or CSNK1A1 are lost , such as those suggested by our comparative analysis , may represent potential therapeutic targets . The second top hit of the APC suppressor screen , after CTNNB1 , was the gene encoding the mRNA binding protein SERBP1 ( Figure 5B and Supplementary file 1 ) , also known as PAI-RBP1 . SERBP1 was also a significant hit in the CSNK1A1 suppressor screen ( Figure 5C and Supplementary file 1 ) . SERBP1 was initially identified as an mRNA binding protein that interacts with the cyclic nucleotide-responsive sequence of the Type-1 plasminogen activator inhibitor mRNA and may play a role in regulation of mRNA stability ( Heaton et al . , 2001 ) . Yet , its cellular function remains largely unknown , and it has never been implicated in regulation of WNT signaling . To explore the consequences of disrupting SERBP1 in cells lacking APC we used an independently isolated HAP1-7TGP clonal cell line with a lesion in the APC locus introduced by a GT insertion ( APCKO-2 , see Materials and methods , Supplementary file 2 and Figure 6D ) . This ensured that any effects on WNT signaling were not specific to the CRISPR/Cas9-induced lesion in the APCKO-1 cells used for the APC suppressor screen . As expected , APCKO-2 cells had constitutive WNT reporter expression ( Figure 5—figure supplement 1B ) . We used CRISPR/Cas9 to generate multiple independent clonal cell lines derived from APCKO-2 cells that harbored additional inactivating mutations in SERBP1 , hereafter called APCKO-2; SERBP1KO cells ( Supplementary file 2 and Figure 6D ) . Disrupting SERBP1 in cells lacking APC caused a substantial reduction in constitutive WNT reporter fluorescence , endogenous AXIN2 mRNA and soluble CTNNB1 protein abundance ( Figure 6A–D , and Figure 6—figure supplement 1A and B ) . Disrupting SERBP1 in WT HAP1-7TGP cells ( SERBP1KO , Supplementary file 2 and Figure 6D ) did not affect basal or WNT3A-induced levels of AXIN2 mRNA or soluble CTNNB1 ( Figure 6B–D , and Figure 6—figure supplement 1A and B ) , explaining why SERBP1 was not enriched for GT insertions in the WNT screen for positive regulators of ligand-induced signaling ( Figure 5D and Supplementary file 3 ) . Microinjection of SERBP1 mRNA into X . laevis embryos resulted in duplication of the body axis , establishing SERBP1 as a bona fide positive regulator of WNT signaling in vertebrates ( Figure 2E and F ) . 10 . 7554/eLife . 21459 . 015Figure 6 . The mRNA binding protein SERBP1 controls CTNNB1 abundance in cells lacking APC ( A–D ) , and the E3 ubiquitin ligase HUWE1 regulates WNT signaling in the absence of CSNK1A1 ( E–I ) . ( A , E , I ) WNT reporter activity ( median EGFP fluorescence from 5000 ( A ) , 20 , 000 ( E ) or 2000 ( I ) cells ) for the indicated single- and double-mutant cell lines . Each circle represents a unique clonal cell line and the average of 10 ( A ) , ≥12 ( E ) or ≥19 ( I ) independent clones for each genotype is indicated by a horizontal line . The average percentage reporter activity relative to single-mutant cell lines is also indicated above each group of circles . Significance was determined by unpaired t-test with Welch’s correction and is indicated as **** ( p<0 . 0001 ) or ns ( not significant ) . ( B , F ) AXIN2 mRNA ( average ± SD of AXIN2 mRNA normalized to HPRT1 mRNA , each measured in triplicate reactions ) , relative to untreated WT cells , for a single clonal cell line of each indicated genotype . Cells were treated with 50% WNT3A CM where indicated . The same cell lines analyzed in B were also analyzed in C and D; the same cell lines analyzed in F were also analyzed in G and H . Analysis of additional independent clonal cell lines is presented in Figure 6—figure supplement 1 . ( C , G ) Soluble CTNNB1 protein ( average ± SD of CTNNB1 intensity normalized to GAPDH intensity from duplicate immunoblots ) , relative to untreated WT cells . Cells were treated with 50% WNT3A CM where indicated . ( D , H ) Representative immunoblots of the indicated clonal cell lines . The CTNNB1 and corresponding GAPDH blots depicted in D and H were used for quantification in C and G , respectively . Genotypes and treatments are indicated above the blots . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 01510 . 7554/eLife . 21459 . 016Figure 6—figure supplement 1 . The mRNA binding protein SERBP1 controls CTNNB1 abundance in cells lacking APC ( A–C ) , and the E3 ubiquitin ligase HUWE1 regulates WNT signaling in the absence of CSNK1A1 ( D–G ) . ( A ) AXIN2 mRNA ( average AXIN2 mRNA normalized to HPRT1 mRNA , each measured in triplicate reactions ) for SERBP1WTand SERBP1KO genotypes in APCWTand APCKO-2 genetic backgrounds , expressed as percentage of the average for SERBP1WT cells to facilitate comparisons . Cells were treated with 50% WNT3A CM where indicated . Each circle represents a unique clonal cell line and the average of two to four independent clones for each genotype is indicated by a horizontal line . The cell lines used in A were also analyzed in B and C . In all panels of this figure , significance was determined by unpaired t-test with Welch’s correction and is indicated as *** ( p<0 . 001 ) , ** ( p<0 . 01 ) , * ( p<0 . 05 ) or ns ( not significant ) . ( B ) CTNNB1 protein ( average CTNNB1 intensity normalized to ACTIN intensity from duplicate blots ) expressed as percentage of the average for SERBP1WT cells . Cells were treated with 50% WNT3A CM where indicated . ( C ) CTNNB1 mRNA ( average CTNNB1 mRNA normalized to HPRT1 mRNA , each measured in triplicate reactions ) expressed as percentage of the average for SERBP1WT cells . ( D ) WNT reporter activity ( median EGFP fluorescence from ~10 , 000 cells in a . u . ) . Each circle represents a unique clonal cell line and the average of ≥8 independent clones for each genotype is indicated by a horizontal line . The percentage reporter activity relative to control cells is also indicated above each group of circles . ( E ) AXIN2 mRNA ( average AXIN2 mRNA normalized to HPRT1 mRNA , each measured in triplicate reactions ) for HUWE1WTand HUWE1KO genotypes in CSNK1A1WTand CSNK1A1KO-1 genetic backgrounds , expressed as percentage of the average for HUWE1WT cells to facilitate comparisons . Cells were treated with 50% WNT3A CM where indicated . Each circle represents a unique clonal cell line and the average of two to three independent clones for each genotype is indicated by a horizontal line . The cell lines used in E were also analyzed in F and G . ( F ) CTNNB1 protein ( average CTNNB1 intensity normalized to ACTIN intensity from duplicate blots ) expressed as percentage of the average for HUWE1WT cells . Cells were treated with 50% WNT3A CM where indicated . ( G ) WNT reporter activity ( median EGFP fluorescence from 20 , 000 cells ) for WT and HUWE1KO cells ( in a CSNK1A1WT genetic background ) following treatment with 10 μM of the GSK3 inhibitor CHIR-99021 . The percentage of reporter activity relative to WT is also indicated above the circles . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 016 In the absence of destruction complex activity , SERBP1 could conceivably reduce CTNNB1 abundance by decreasing transcript or protein levels . No significant changes in CTNNB1 mRNA levels were detected when SERBP1 was disrupted in either WT or mutant APC genetic backgrounds ( Figure 6—figure supplement 1C ) , suggesting instead a reduction in CTNNB1 protein . Thus , SERBP1 can regulate CTNNB1 abundance in cells lacking APC . This mechanism , independent of destruction complex activity , could be particularly well suited for therapeutic interference in tumors where APC function is lost . The third most significant hit of the CSNK1A1 suppressor screen , following CTNNB1 and CREBBP , was HUWE1 ( Figure 5C and Supplementary file 1 ) . HUWE1 ( also known as MULE , LASU1 and UREB1 ) encodes a 480–482 kDa E3 ubiquitin ligase containing a C-terminal HECT domain with diverse cellular functions ( Bernassola et al . , 2008 ) . Previous work has implicated HUWE1 as a negative regulator of WNT signaling ( de Groot et al . , 2014 ) . However , the results of our screens suggested that in the absence of CSNK1A1 , HUWE1 is a positive regulator of WNT signaling . To test the role of HUWE1 in this context , we further engineered the CSNK1A1KO-1 cells used for the CSNK1A1 suppressor screen , as well as an independently derived cell line with a different lesion in CSNK1A1 ( CSNK1A1KO-2 , Supplementary file 2 ) . In each parental cell line , we used two different sgRNAs to disrupt HUWE1 and isolated multiple double-mutant clonal cell lines ( designated CSNK1A1KO-1; HUWE1KO and CSNK1A1KO-2; HUWE1KO , Supplementary file 2 and Figure 6H ) . Disruption of HUWE1 resulted in a consistent 82–90% reduction of constitutive WNT reporter fluorescence and an 80–85% reduction of endogenous AXIN2 mRNA ( Figure 6E and F , and Figure 6—figure supplement 1D and E ) . In addition , microinjection of HUWE1 mRNA into X . laevis embryos resulted in duplication of the body axis at low frequency ( Figure 2E and F ) , supporting a more general role as a positive regulator . In contrast to the 80–90% reduction in both WNT reporter fluorescence and target gene expression ( Figure 6E and F , and Figure 6—figure supplement 1D and E ) , depleting HUWE1 in CSNK1A1KO-1; HUWE1KO cells reduced soluble CTNNB1 levels by only 20–32% ( Figure 6G and H , and Figure 6—figure supplement 1F ) . These results show that in cells lacking CSNK1A1 , HUWE1 has a minor influence on CTNNB1 abundance , but that its predominant role in WNT signaling is distinct from the regulation of CTNNB1 protein levels . Additionally , HUWE1 was not a significant hit in the WNT screens or the APC suppressor screen ( Figure 5D and Supplementary file 3 ) , suggesting that its role is specific to cells lacking CSNK1A1 . Indeed , disruption of HUWE1 in WT HAP1-7TGP cells ( HUWE1KO , Supplementary file 2 and Figure 6H ) did not cause significant changes in WNT3A-induced AXIN2 mRNA or CTNNB1 protein abundance ( Figure 6F–H , and Figure 6—figure supplement 1E and F ) . To directly test whether HUWE1 disruption reduces WNT signaling in cells lacking CSNK1A1 but not other destruction complex components , we disrupted HUWE1 in APCKO-1 cells ( APCKO-1; HUWE1KO , Supplementary file 2 ) , and found no measurable defect in WNT reporter florescence ( Figure 6I ) . Signaling driven by inhibition of GSK3 was also unaffected by the loss of HUWE1 in HUWE1KO cells ( Figure 6—figure supplement 1G ) . In summary , the drastic defect in signaling caused by loss of HUWE1 in cells lacking CSNK1A1 and its ability to promote formation of a secondary body axis when expressed ectopically in X . laevis embryos demonstrate a positive role for HUWE1 in WNT signaling . These effects are largely independent of changes is CTNNB1 protein abundance and are not observed when other components of the destruction complex are inactivated . From these results we conclude that CSNK1A1 regulates WNT signaling by an additional mechanism distinct from its established role in CTNNB1 turnover , and that this mechanism is mediated by HUWE1 .
A systematic genetic analysis in human cells revealed new regulatory features at most levels of the WNT pathway , from signal reception to transcriptional activation ( Figure 7 ) . Based on a comparative analysis of seven genome-wide screens , we confirmed known epistatic connections and assigned new ones ( Figure 7A ) . Even for some of the known WNT components , our analysis suggested unexpected regulatory mechanisms . 10 . 7554/eLife . 21459 . 017Figure 7 . A comparative analysis of seven genome-wide screens revealed epistatic connections and regulatory mechanisms in WNT signaling . ( A ) Summary of known regulators , and new regulators or regulators mediating new mechanisms in WNT signaling validated in this study . An ‘X’ denotes that the gene was enriched for GT insertions in the sorted cells from the indicated genetic screen ( FDR-corrected p-value<0 . 05 ) . Known regulators are grouped into functional modules and arranged according to previously described epistatic relationships . Epistatic relationships for new regulators or regulators mediating new mechanisms are inferred based on their patterns across screens . The screens in which cells were sorted for increased WNT reporter fluorescence are labeled in red , and those in which cells were sorted for reduced WNT reporter fluorescence are labeled in green . For the ‘WNT positive regulator’ column , hits from the WNT screens done at both low and high stringency ( Figure 1E and F ) were considered together . For AXIN2 , the asterisk indicates that GT insertions mapped in the sorted cells generate a dominant allele that encodes a truncated protein product . ( B ) Model of WNT/CTNNB1 signaling , highlighting in red new regulatory mechanisms uncovered and validated in this study . Red arrows represent genetic ( rather than biochemical ) interactions . The various proposed mechanisms are discussed throughout the Results and Discussion sections . DOI: http://dx . doi . org/10 . 7554/eLife . 21459 . 017 First , as predicted by their enrichment in the WNT screens for positive regulators ( Figure 7A ) , atypical GT insertions in AXIN2 caused an unexpected decrease in WNT signaling ( Figure 3 ) . These results are explained by the observation that in cells lacking the DAX domain of AXIN2 , CTNNB1 is appropriately stabilized and localized to the nucleus following WNT stimulation , but remains inactive ( Figure 3 ) . Second , genes encoding components of the GPI anchor biosynthetic machinery , such as PIGL , and the glypican GPC4 were predominantly enriched for GT insertions in the low WNT + RSPO screen ( Figure 7A ) , and we demonstrated that they indeed play a critical role in mediating signaling especially under the low WNT conditions in which RSPOs exert their strongest effect ( Figure 4 ) . Third , the enrichment of GT insertions in HUWE1 only in the CSNK1A1 suppressor screen ( Figure 7A ) revealed a unique signaling condition created by disruption of CSNK1A1 , but not other destruction complex components such as APC or GSK3 . Only in this very specific context WNT signaling was dependent on HUWE1 ( Figure 6 ) . This positive regulatory function of HUWE1 is evidently different from the negative feedback regulation described previously ( de Groot et al . , 2014 ) . The presence of mutations in known regulators in the expected screens demonstrates the predictive power of our approach , which enabled us to infer the site of action of newly identified pathway components . The transcription factor TFAP4 would be predicted to act downstream of the CTNNB1 destruction complex based on its disruption in all screens for positive regulators of signaling ( Figure 7A ) , as our experimental results confirmed ( Figure 2 ) . The selective disruption of SERBP1 in only the APC and CSNK1A1 suppressor screens ( Figure 7A ) suggests a regulatory role on signaling independent of destruction complex activity , which we demonstrated experimentally ( Figure 6 ) . An important conclusion from our studies is that WNT signaling can be regulated by processes other than control of CTNNB1 protein abundance by the destruction complex . We demonstrate two distinct instances in which CTNNB1 transcriptional activity can be dissociated from protein levels , one caused by truncation of the AXIN2 DAX domain and the other caused by depletion of HUWE1 in cells lacking CSNK1A1 . It will be interesting to explore if these phenomena can be exploited for therapeutic purposes in tumors driven by inappropriate stabilization of CTNNB1 . We also provide evidence that the destruction complex does not have a unitary function in controlling CTNNB1 protein abundance , since disrupting distinct components produces different outcomes . Supressor screens in cells lacking APC or CSNK1A1 revealed mutations in substantially different sets of genes ( Figure 5 ) , and while SERBP1 controls CTNNB1 abundance in cells lacking APC , the effects of HUWE1 in cells lacking CSNK1A1 are largely independent of changes in CTNNB1 levels ( Figure 6 ) . From these studies a more elaborate picture of the core WNT signaling cascade emerges , with additional regulation superimposed on the core module ( Figure 7B ) . Further studies will be required to elucidate the mechanisms that mediate each of these new layers of regulation and to identify the physiological or pathological contexts in which they act . Yet , the comparative analysis of seven unbiased genome-wide screens and the characterization of hits through a quantitative assessment of CRISPR/Cas9-engineered clonal cell lines provided many insights into this complex developmental signaling pathway . The conceptual and methodological framework described in this work should enable the comprehensive understanding of other signaling systems .
Reagents were obtained from the following companies: Thermo Fisher Scientific , Waltham , MA; Sigma-Aldrich , St . Louis , MO; Bio-Rad , Hercules , CA; Cell Biolabs , San Diego , CA; Clontech , Mountain View , CA; Promega , Madison , WI; GE Healthcare Life Sciences , Logan , UT; GE Dharmacon , Lafayette , CO; Addgene , Cambridge , MA; BD Biosciences , San Jose , CA; Abcam , Cambridge , MA; EMD Millipore , Billerica , MA; Bethyl Laboratories , Montgomery , TX; Santa Cruz Biotechnology , Dallas , TX; R and D Systems , Minneapolis , MN; Cell Signaling Technology , Danvers , MA; Li-Cor , Lincoln , NE; Jackson ImmunoResearch Laboratories , West Grove , PA; Developmental Studies Hybridoma Bank at the University of Iowa ( DSHB ) , Iowa City , IA; American Type Culture Collection ( ATCC ) , Manassas , VA; Atlanta Biologicals , Flowery Branch , GA; Pall Corporation , Fribourg , Switzerland; Selleckchem , Houston , TX; Roche , Mannheim , Germany; QIAGEN Sciences , Hilden , Germany; New England Biolabs ( NEB ) , Ipswich , MA; Epicentre , Madison , WI; Corning , Corning , NY; Bloomington Drosophila Stock Center at Indiana University ( BDSC ) , Bloomington , IN . 7TGP was a gift from Roel Nusse ( Addgene plasmid # 24305 ) . pX330-U6-Chimeric_BB-CBh-hSpCas9 ( pX330 ) was a gift from Feng Zhang ( Addgene plasmid # 42230 ) . pCSDest was a gift from Nathan Lawson ( Addgene plasmid # 22423 ) . pCS2+ , pCS2-YFP , pCS2-β-cat-S33Y , pCS2-xWnt8 , VSV-G and Δ8 . 9 were a gift from Henry Ho . The following plasmids were purchased: pAdVAntage ( Promega Cat . # E1711 ) , pCMV-VSV-G ( Cell Biolabs Cat . # RV-110 ) , pCMV-Gag-Pol ( Cell Biolabs Cat . # RV-111 ) , pENTR-D-TOPO ( Thermo Fisher Scientific Cat . # K240020 ) , pENTR2B ( Thermo Fisher Scientific Cat . # A10463 ) . To generate pCSDest-TFAP4 , human TFAP4 was amplified by PCR from MGC Human Sequence-verified cDNA ( Clone ID 4181538 , GE Dharmacon Cat . # MHS6278-202757542 ) using primers hTFAP4-FOR ( 5’-CACCATGGAGTATTTCATGGTGCCCA-3’ ) and hTFAP4-REV ( 5’- TCAGGGAAGCTCCCCGT-3’ ) , designed to add a directional TOPO cloning sequence at the 5’ end . The resulting PCR product was cloned directionally into pENTR-D-TOPO . Individual clones were screened by digestion with NotI and PstI for the presence of the insert in the correct orientation , and one clone was sequenced completely and subcloned into pCSDest using the Gateway LR Clonase II Enzyme mix ( Thermo Fisher Scientific Cat . # 11791100 ) . To generate pCSDest-1D4-SERBP1 , human SERBP1 was amplified by PCR from MGC Human Sequence-verified cDNA ( Clone ID 4477452 , GE Dharmacon Cat . # MHS6278-202758125 ) using primers FNNterminal ( 5’-TTTTGAATTCGCCACCATGACCGAGACCAGCCAGGTGGCCCCTGCAGGCGGCCGGCCACCTGGGCACTTACAGGAAGG-3’ ) , designed to add an N-terminal Kozak sequence , a 1D4 tag and a single glycine linker flanked by EcoR1 and FseI restriction sites , and RNNterminal ( 5’-TTTTCTCGAGGGCGCGCCTTAAGCCAGAGCTGGGAATG-3’ ) , designed to add tandem AscI and XhoI restriction sites after the stop codon . The product was digested with EcoRI and XhoI , and subcloned into pENTR2B digested at the same sites . One clone was sequenced completely and subcloned into pCSDest using the Gateway LR Clonase II Enzyme mix . To generate pCSDest-HUWE1 , codon optimized HUWE1 in pDONR221 , a gift from Sarah J . Luchansky ( Developmental and Molecular Pathways , Novartis Institutes for Biomedical Research , Cambridge , MA ) , was subcloned into pCSDest using the Gateway LR Clonase II Enzyme mix . pCherry-C1 was created by replacing DsRed with mCherry RFP ( Shaner et al . , 2004 ) in pDsRed-monomer-C1 ( Clontech Cat . # 632466 ) . An mCherry RFP PCR fragment was obtained using primers 5’-GATCGCTAGCACCATGGTGAGCAAGGGCGAGGAGGATAAC-3’ and 5’-GATCCTCGAGATCTCTTGTACAGCTCGTCCATGCCGCC-3’ . The PCR product was digested with NheI and XhoI and ligated into pDsRed-monomer-C1 digested with the same enzymes . pGT-mCherry and pGT+1mCherry retroviral gene trap ( GT ) vectors ( diagrammed schematically in Figure 1B ) , containing an inactivated 3’LTR , a strong adenoviral ( Ad40 ) splice acceptor site , mCherry RFP in two different reading frames following the splice acceptor site , and the SV40 polyadenylation signal , were created by replacing GFP with mCherry RFP in pGT-GFP ( Carette et al . , 2009 ) . To create pGT-mCherry , the following primers were used to generate a PCR product containing the splice acceptor site followed by mCherry: 5’-GATCATCGATGCGCAGGCGCAATCTTCGCATTTCTTTTTTCCAGATGGTGAGCAAGGGCGAGG-3’ and 5’-GATCGGATCCTTACTTGTACAGCTCGTCCATG-3’ . To create pGT+1mCherry , the following primers were used: 5’-GATCATCGATGCGCAGGCGCAATCTTCGCATTTCTTTTTTCCAGGATGGTGAGCAAGGGCGAGG-3’ and 5’-GATCGGATCCTTACTTGTACAGCTCGTCCATG-3’ . The PCR products were digested with ClaI and BamHI and cloned into pGT-GFP digested with the same enzymes . All constructs were confirmed by sequencing . L Wnt-3A ( ATCC Cat . # CRL-2647 ) , L cells ( ATCC Cat . # CRL-2648 ) , 293T ( ATCC Cat . # CRL-3216 ) and 293FT cells ( Thermo Fisher Scientific Cat . # R70007 ) were grown at 37°C and 5% CO2 in complete growth medium ( CGM ) 1: Dulbecco's Modified Eagles Medium ( DMEM ) with High Glucose , without L-Glutamine and Sodium Pyruvate ( GE Healthcare Life Sicences Cat . # SH30081 . 01 ) ; 1X GlutaMAX-I ( Thermo Fisher Scientific Cat . # 35050079 ) ; 1X MEM Non-Essential Amino Acids ( Thermo Fisher Scientific Cat . # 11140050 ) ; 1 mM Sodium Pyruvate ( Thermo Fisher Scientific Cat . # 11360070 ) ; 40 Units/ml Penicillin , 40 µg/ml Streptomycin ( Thermo Fisher Scientific Cat . # 15140122 ) ; 10% Fetal Bovine Serum ( FBS ) ( Atlanta Biologicals Cat . # S11150 ) . HAP1 haploid human cells ( kindly provided by Thijn Brummelkamp , now available from Horizon Discovery , Cambridge , United Kingdom ) were derived and characterized as described previously ( Carette et al . , 2011b ) . Throughout the course of experiments , the ploidy of HAP1 cells and derivatives thereof was routinely tested by DNA content analysis of propidium iodide ( PI ) -stained nuclei , as described below . Genetically modified clonal derivatives were confirmed by sequencing of target loci and in some cases immunoblotting , as described below ( see also Supplementary file 2 ) . HAP1 cells and derivatives thereof were grown at 37°C and 5% CO2 in CGM 2: Iscove's Modified Dulbecco's Medium ( IMDM ) with L-glutamine , with HEPES , without Alpha-Thioglycerol ( GE Healthcare Life Sciences Cat . # SH30228 . 01 ) ; 1X GlutaMAX-I; 40 Units/ml Penicillin , 40 µg/ml Streptomycin; 10% FBS . L Wnt-3A cells or L cells were seeded in 15 cm tissue culture-treated dishes at a density of 1 . 5 × 106 cells per dish and grown in 25 ml of CGM 1 . After 3 days , the medium was refreshed , and after an additional 3 days , WNT3A or L cell conditioned medium ( used as a control ) was collected , filtered through a 0 . 2 µm polyethersulfone ( PES ) membrane filter , aliquoted , flash-frozen in liquid nitrogen and stored at −80°C . The medium was thawed immediately before use , and leftover medium was used only after one additional freeze-thaw cycle . Lentivirus containing the 7TGP WNT reporter construct ( Fuerer and Nusse 2010 ) was produced in 293T cells . ~24 hr before transfection , 293T cells were seeded in a 10 cm dish at a density of 4 × 106 per dish and grown in 10 ml of CGM 1 . When the cells were nearly confluent , the medium was replaced with CGM 1 without antibiotics . To prepare a calcium phosphate transfection mix , 8 µg of 7TGP plasmid , 4 µg of VSV-G , 4 µg of Δ8 . 9 and 0 . 5 µg of pCS2-YFP ( used as a co-transfection marker ) in 450 µl of sterile water were mixed with 50 µl of 2 . 5 M CaCl2 . The DNA/CaCl2 solution was added to 500 µl of 2X HBS ( 42 mM HEPES pH 7 . 04 , 274 mM NaCl , 10 mM KCl , 15 mM dextrose , 1 . 4 mM Na2HPO4 . 7H2O ) and mixed by bubbling air through the solution . Following a 20 min incubation at RT , the transfection mix was added drop-wise to the dish of cells and mixed by gentle agitation . The medium was refreshed 16 hr post-transfection , at which time the efficiency of transfection was assessed by microscopic inspection of YFP fluorescence . Lentivirus-containing medium was collected 36 , 48 and 60 hr post-transfection ( 30 ml total ) , filtered through 0 . 45 µm filters ( Acrodisc syringe filters with Supor membrane , Pall Corporation ) and concentrated by ultracentrifugation for 1 . 5 hr at 23 , 000 rpm in a Sorvall Surespin 630 rotor . The supernatants were discarded , and the pellets overlaid by a total of 200 µl of sterile phosphate buffered saline ( PBS ) supplemented with 1 mg/ml bovine serum albumin ( BSA ) . Following 12 hr of incubation at 4°C , the pellets were resuspended , aliquoted , flash-frozen in liquid nitrogen , and stored at −80° . Approximately 48 hr before transduction , HAP1 cells were seeded in 10 cm dishes at a density of 1 . 5 × 106 per dish and grown in 10 ml of CGM 2 . Cells were transduced with 145 µl of freshly thawed lentivirus diluted in 10 ml of CGM 2 supplemented with 4 µg/ml polybrene . ~48 hr post-transduction cells were treated with 1 µg/ml puromycin to select for those with a stably integrated 7TGP cassette . After selection was complete ( assessed by death of >99% of control , untransduced HAP1 cells ) , surviving cells were treated for 16–24 hr with 50% WNT3A CM in CGM 2 , and single WNT3A-responsive cells exhibiting the highest ~50% EGFP fluorescence were sorted by FACS into 96-well plates containing 200 μl of CGM 2 per well . After 14 days of undisturbed growth , individual clones were amplified and haploid HAP1-7TGP clonal cell lines were identified by DNA content analysis of PI-stained nuclei ( Nicoletti et al . , 1991 ) : 1 . 5 × 106 cells were incubated with 300 µl of hypotonic fluorochrome solution ( 50 µg/ml PI , 0 . 1% sodium citrate , 0 . 1% Triton X-100 ) for >15 min at RT and PI fluorescence was measured by FACS in a BD LSRFortessa cell analyzer ( BD Biosciences ) using a 561 laser and 600 LP , 610/20 BP filters . A haploid cell line with low basal reporter activity and a high dynamic range of EGFP fluorescence in response to WNT3A ( Figure 1—figure supplement 1A ) was expanded and used in all subsequent studies . To measure WNT reporter activity in HAP1-7TGP cells or derivatives thereof , ~24 hr before treatment cells were seeded in 24-well plates at a density of 8 × 104 per well and grown in 0 . 5 ml of CGM 2 . Cells were treated for 16–24 hr with the indicated concentrations of WNT3A CM , L cell CM , recombinant mouse WNT3A ( R and D Systems Cat . # 1324-WN ) recombinant human RSPO1 ( R and D Systems Cat . # 4645-RS ) , LiCl , CHIR-99021 ( CT99021 ) ( Selleckchem Cat . # S2924 ) or XAV-939 ( Selleckchem Cat . # S1180 ) diluted in CGM 2 . Cells were washed with 0 . 5 ml PBS , harvested in 150 μl of 0 . 05% Trypsin-EDTA ( 0 . 05% ) ( Thermo Fisher Scientific Cat . # 25300054 ) , resuspended in 450 μl of CGM 2 , and EGFP fluorescence was measured immediately by FACS on a BD LSRFortessa cell analyzer ( BD Biosciences ) using a 488 laser and 505 LP , 530/30 BP filters , or on a BD Accuri RUO Special Order System ( BD Biosciences ) . Typically , fluorescence data for 5000–20 , 000 singlet-gated cells was collected and , unless indicated otherwise , the median EGFP fluorescence ± standard error of the median ( SEM = 1 . 253 σ/√n , where σ = standard deviation and n = sample size ) was used to represent the data . The FDR-corrected p-value for enrichment of GT insertions in the sorted compared to the control cells depends on many experimental variables affecting both datasets , making it an equivocal metric to directly compare hits between screens . We therefore developed an additional scoring metric that depends only on the dataset for the sorted cells . It relies on the fact that generally only sense GT insertions in introns should inactivate genes due to the directionality of the splice acceptor in the GT , and it captures the relative abundance of intronic GT insertions in the sense and antisense orientations , as well as the overall number of intronic insertions . We defined the Intronic GT Insertion Orientation Bias ( IGTIOB ) score as log2 ( S/A ) x ln ( S x A ) , where ‘S’ and ‘A’ equal one plus the number of unique sense or antisense intronic GT insertions , respectively , in a given gene . High , positive IGTIOB scores generally indicate genes whose disruption promotes the phenotype enriched for during the screen . To compare hits between screens , we first generated a list of genes including only those with a stringent GT enrichment FDR-corrected p-value<10−4 in at least one of the screens being considered Supplementary file 3 . We then used the IGTIOB scores of those genes in all the relevant screens to build a heat map ( Figures 4B and 5D ) . Genes were clustered using the absolute value/city block setting and complete linkage method ( without normalization ) in the hierarchical clustering tool of the Partek Genomics Suite software , and the data range was adjusted to encompass only non-negative IGTIOB scores ( i . e . negative values were displayed as 0 ) . Groups of genes preferentially enriched for GT insertions in one or more screens ( as determined by visual inspection ) are indicated next to the heat maps . Oligonucleotides encoding single guide RNAs ( sgRNAs ) ( Supplementary file 4 ) were selected from a published library ( Shalem et al . , 2014 ) , or designed using either of two online CRISPR design tools ( Hsu et al . , 2013; Doench et al . , 2014 ) and cloned into pX330 according to a published protocol ( original version of ‘Target Sequence Cloning Protocol’ from http://www . genome-engineering . org/crispr/wp-content/uploads/2014/05/CRISPR-Reagent-Description-Rev20140509 . pdf; Cong et al . , 2013 ) . Clonal HAP1-7TGP cell lines were established by transient transfection with pX330 containing the sgRNA followed by single cell sorting as follows . A transfection mix was prepared by diluting 450 ng of pX330 and 50 ng of pmCherry ( used as a co-transfection marker for FACS sorting ) in 48 µl Opti-MEM I , adding 2 µl of X-tremeGENE HP and incubating for 20 min at RT . HAP1-7TGP cells or derivatives thereof were reverse-transfected in a well of a 24-well plate by overlaying 0 . 5 ml of CGM 2 ( without antibiotics ) containing 6 × 105 cells over the 50 µl of transfection mix . Cells were passaged to a 10 cm dish ~24 hr post-transfection , using 150 µl of Trypsin-EDTA ( 0 . 25% ) ( Thermo Fisher Scientific Cat . # 25200056 ) to detach them ( reverse-transfection of HAP1 cells caused unusually high adherence , hence the higher trypsin concentration ) . Four to five days post-transfection , single transfected ( mCherry-positive ) cells were sorted into 96-well plates containing 200 µl of CGM 2 per well and grown undisturbed for 16 to 17 days . Single colonies were passaged to 24-well plates , and a small number of cells was reserved for genotyping . For genotyping , genomic DNA was extracted by adding 4 volumes of QuickExtract DNA Extraction Solution ( Epicentre Cat . # QE09050 ) to the cells . Extracts were incubated 10 min at 65°C , 3 min at 98°C , and 5 µl were used as input for PCR amplification of the sgRNA target site in 15 µl reactions containing 1X LongAmp Taq reaction buffer , 300 µM of each dNTP , 400 nM of each of the flanking primers indicated in Supplementary file 4 ( most of them designed using the Primer-BLAST online tool from the NCBI ) and 0 . 1 units/µl of LongAmp Taq DNA polymerase ( NEB Cat . # M0323L ) . The presence of desired mutations was initially assessed by analysis of the PCR products on a 1% agarose gel ( i . e . the absence of an amplicon or a shift in its size was deemed indicative of a lesion ) and was confirmed by sequencing the amplicons using the primers indicated in Supplementary file 4 . Given that most engineered cell lines remained haploid , sequencing results were usually unequivocal . A summary of the sequencing results for all clonal cell lines used in the study is presented in Supplementary file 2 , and for selected clonal cell lines immunoblot analysis of the protein products is presented in Figures 2C , 6D and H , Figure 3—figure supplement 1A and Figure 5—figure supplement 1A . Whenever possible , multiple independent mutant cells lines , often generated using two different sgRNAs ( see Supplementary file 2 ) , were expanded and used for further characterization . For some of the comparisons between WT and mutant cells , multiple individual cell lines confirmed by sequencing to be WT at the sgRNA target site were also expanded and used as controls . To generate double and triple mutant cell lines , a single clonal cell line with the first desired mutation was used in a subsequent round of transfection with pX330 containing the second and , if applicable , third sgRNAs . Alternatively , WT HAP1-7TGP cells were directly transfected with a combination of pX330 constructs targeting two genes simultaneously . To facilitate screening of mutant clones by PCR when targeting two genes simultaneously , we sometimes targeted one of them at two different sites within the same exon or on adjacent exons and amplified genomic sequence encompassing both target sites . Mutant clones were readily identified by the reduced size of the resulting amplicon , and the precise lesion was confirmed by sequencing . As an alternative to CRISPR/Cas9-mediated genome editing , a mutant cell line containing a GT insertion in APC was isolated as follows . Mutagenized HAP1-7TGP cells enriched during the WNT-negative regulator screen ( Figure 1C ) were used to isolate the APCKO-2 clone . Following the screen , the same FACS gate used during the screen was used to sort single cells into 96-well plates . Colonies were harvested after 16 days , 1/10th of each clone was passaged for continued growth , and the remainder of the cells from adjacent pairs of rows in each plate were pooled . A portion of all the cells from pairs of plates were further pooled so as to obtain ‘plate-pair’ pools , and ‘row-pair’ sub-pools . Cells from plate-pair pools were harvested by centrifugation , genomic DNA was prepared using the QIAamp DNA mini kit and each pool was probed for clones containing GTs in APC using a nested PCR strategy . A genomic region of APC enriched for GT insertions was amplified by PCR using a forward primer complementary to a unique sequence in the GT ( pGT-Puro4: 5’-TCTCCAAATCTCGGTGGAAC-3’ ) and a reverse primer complementary to a unique genomic sequence adjacent to the GT-enriched region in APC ( APC_GT: 5’-TGCTACAATGAGCTGTTAAAATGG-3’ ) . 400 ng of genomic DNA was used as input for PCR amplification in 25 µl reactions containing 1X LongAmp Taq reaction buffer , 300 µM of each dNTP , 400 nM of each primer and 0 . 1 units/µl of LongAmp Taq DNA polymerase . For each plate-pair pool , the presence of clones containing a GT insertion was evident as discrete bands when the PCR products were analyzed on a 1% agarose gel . Once positive plate-pair pools were identified , genomic DNA from the corresponding row-pair sub-pools was prepared using QuickExtract DNA Extraction Solution and the PCR procedure was repeated to identify row pairs containing clones with GT insertions . Finally , the individual clonal cell lines that had been passaged were harvested from the 96-well plates containing GT-positive row pairs , and GT-containing clones were identified using the same procedure . To map the precise genomic location of the GT insertion in APCKO-2 ( see Supplementary file 2 ) , the final PCR product obtained from an individual GT-positive clone was sequenced . Rescue of TFAP4CR-1 cells and overexpression of TFAP4 in WT HAP1-7TGP cells ( Figure 2B ) was done by transient transfection . TFAP4CR-1and HAP1-7TGP were reverse-transfected in a well of a 24-well plate by overlaying 0 . 5 ml of CGM 2 ( without antibiotics ) containing 6 × 105 cells over 50 µl of a transfection mix containing 450 ng of pCSDest-TFAP4 or pCS2+ ( vector control ) and 50 ng of pmCherry ( as a co-transfection marker for FACS analysis ) in Opti-MEM I , and 2 µl of X-tremeGENE HP . Cells were harvested using 150 µl of Trypsin-EDTA ( 0 . 25% ) and passaged to a 6 cm dish ~24 hr post-transfection . ~48 hr post-transfection cells were treated for 16–24 hr with 50% WNT3A CM where indicated , and the WNT reporter ( EGFP ) fluorescence of mCherry-positive , singlet-gated cells was measured by FACS . Epistasis analysis in WT HAP1-7TGP and TFAP4CR-1 cells ( Figure 2D ) was done following treatment with WNT3A , the GSK3 inhibitor LiCl , or following transient transfection with dominant negative CTNNB1 ( S33Y mutant ) as follows . WT HAP1-7TGP and TFAP4CR-1 cells were treated for 16–24 hr with 50% WNT3A CM or with 40 mM LiCl in CGM 2 where indicated , and the WNT reporter fluorescence of singlet-gated cells was measured by FACS . Alternatively , WT HAP1-7TGP and TFAP4CR-1 cells were reverse transfected with pCS2+ ( vector control ) or pCS2-β-cat-S33Y and pmCherry ( as a co-transfection marker for FACS analysis ) as described in the previous section . Cells were passaged ~24 hr post-transfection , and ~48 hr post-transfection the WNT reporter ( EGFP ) fluorescence of mCherry-positive , singlet-gated cells was measured by FACS . Approximately 24 hr before treatment , cells were seeded in 24-well plates at a density of 2 × 105 per well and grown in 0 . 5 ml of CGM 2 . Cells were treated for 12 hr with 50% WNT3A CM in CGM 2 where indicated . The medium was removed , and cells were harvested without washing in 800 µl of TRIzol Reagent ( Thermo Fisher Scientific Cat . # 15596018 ) . Extracts were processed according to the manufacturer’s protocol taking the appropriate precautions to avoid contamination with nucleases , and total RNA was resuspended in 40 µl of DEPC-treated water ( Thermo Fisher Scientific Cat . # AM9920 ) . To synthesize cDNA , 250 ng of RNA were diluted in 8 µl DEPC-treated water , 2 µl of 5X iScript Reverse Transcrition Supermix for RT-qPCR ( Bio-Rad Cat . # 170–8841 ) were added , and the reaction was incubated 5 min at 25°C , 30 min at 42°C and 5 min at 85°C . cDNA was diluted 1:100 in water , and 5 µl were mixed with 5 µl of iTaq Universal SYBR Green Supermix ( Bio-Rad Cat . # 172–5121 ) containing 400 nM each of forward and reverse primer ( hAXIN2-RT-PCR-1-FOR: 5’-GTCCAGCAAAACTCTGAGGG-3’ , hAXIN2-RT-PCR-1-REV: 5’-CTGGTGCAAAGACATAGCCA-3’; hCTNNB1-RT-PCR-1-FOR: 5’-AAAGCGGCTGTTAGTCACTGG-3’ , hCTNNB1-RT-PCR-1-REV: 5’-CGAGTCATTGCATACTGTCCAT-3’; hHPRT1-RT-PCR-1-FOR: 5’-TGCTGAGGATTTGGAAAGGG-3’ , hHPRT1-RT-PCR-1-REV: 5’-ACAGAGGGCTACAATGTGATG-3’ ) . Triplicate reactions for each cDNA and primer pair were prepared in a MicroAmp Optical 384-well Reaction Plate ( Thermo Fisher Scientific Cat . # 4309849 ) , sealed with MicroAmp Optical Adhesive Film ( Thermo Fisher Scientific Cat . # 4311971 ) and run using standard parameters in an ABI 7900 Fast Real-Time PCR system ( Applied Biosystems ) controlled by the Sequence Detection Systems software version 2 . 4 . 1 provided by the manufacturer . The formula 2-∆Ct was used to calculate the average relative abundance of AXIN2 ( Figures 2A , 6B and F and Figure 6—figure supplement 1A and E ) or CTNNB1 ( Figure 6—figure supplement 1C ) mRNA normalized to HPRT1 mRNA , and fold-changes in mRNA abundance were calculated as the quotient between the experimental and reference samples , with appropriate error propagation of the respective standard deviations ( SD ) . mRNA was synthesized using the mMESSAGE mMACHINE SP6 Kit ( Thermo Fisher Scientific Cat . # AM1340 ) from a pCSDest-based construct containing the CDS encoding the protein of interest . One microgram of plasmid DNA was linearized and mRNA was synthesized according to the protocol provided by the manufacturer . The mRNA was treated with TURBO DNase and purified using the RNeasy Mini Kit ( QIAGEN Cat . # 74104 ) , following the ‘RNA cleanup’ protocol . The purified product was analyzed on a 1% agarose gel . X . laevis eggs were fertilized , de-jellied with L-cysteine , and equilibrated in Marc’s Modified Ringers ( MMR: 0 . 1 M NaCl , 2 mM KCl , 1 mM MgSO4 , 2 mM CaCl2 , 5 mM HEPES pH 7 . 8 , 0 . 1 mM EDTA ) containing 2% Ficoll . One ventral blastomere of four-cell stage embryos was injected with 3 nl of a 1 . 67 ng/nl mRNA solution ( 5 ng total ) using an MPPI-2 Pressure Injector ( Applied Scientific Instrumentation ) . Embryos were incubated at RT in 1/3 MMR until stage 34 and scored for body axis duplication ( Figure 2F ) on a Zeiss Stemi 2000-C stereo microscope . Embryos were fixed overnight in MEMFA buffer ( 100 mM MOPS , 2 mM EGTA , 1 mM MgSO4 , 3 . 7% formaldehyde ) transferred to 100% glycerol and images ( Figure 2E ) were taken with an Olympus DP72 camera at 2 . 5X magnification . Gene expression levels in WT HAP1 cells were determined from previously published data ( Dubey et al . , 2016 ) . Briefly , the raw RNAseq data for WT HAP1 cells ( NCBI Gene Expression Omnibus ( GEO ) Series accession number GSE75515 ) was aligned , quantified and analyzed using Partek Flow software , version 4 . 0 ( Partek Inc . ) . Reads were aligned to human reference genome build hg19 using the STAR Align and Quantify pipeline in Partek Flow . RPKM ( Reads Per Kilobase of transcript per Million mapped reads ) normalization was used to obtain the values reported in Table 1 . The normalized data was imported into Partek Genomics Suite 6 . 6 software ( Partek Inc . ) to visualize the quantification of normalized reads for selected genes , and Table 1 was assembled in Excel ( Microsoft ) . Illustrations were prepared using PowerPoint ( Microsoft ) and Illustrator CS6 ( Adobe ) . Circle plots depicting the hits from each screen , GT insertion histograms , dose response graphs , tables and supplementary files were prepared using Excel ( Microsoft ) . Bar and circle graphs were prepared using Prism 6 ( GraphPad Software ) and statistical analysis was performed using the same software ( details of statistical tests used are given in the figure legends ) . Heat maps were generated using Partek Genomics Suite 6 . 6 software ( Partek Inc . ) and finished in Illustrator CS6 . FACS histograms and dot plots were generated using FlowJo ( FlowJo , LLC ) and finished in Illustrator CS6 . Pictures of immunoblots and model organisms were only adjusted for contrast and brightness when necessary for clarity using Photoshop CS6 ( Adobe ) , and were arranged in Illustrator CS6 . | When an embryo is developing , its cells must communicate with one another to coordinate the processes that shape the body’s tissues and organs . Cells often communicate by releasing signaling molecules that engage with proteins called receptors on the surface of other cells . This triggers a series of events that sends the signal along a “pathway” of biochemical reactions inside the receiving cell and leads to the activation of genes . One such signaling pathway is triggered by the WNT proteins and is used extensively in all animals . The WNT pathway instructs cells to grow and divide , establishes the identity of specific cell types and maintains populations of stem cells that can regenerate tissues in adulthood as well . The WNT pathway must be carefully regulated because various types of cancer can develop if the pathway becomes too active . Some signaling pathways are well conserved between different animals . Many genetic studies into the WNT pathway have focused on animals that are easier to work with in the laboratory , like worms or flies . However , there may be differences in the way these pathways are regulated between these model animals and humans . Therefore , to understand how the WNT pathway operates in humans , it was important to study it in human cells too . Lebensohn et al . have now carried out a series of genetic screens in human cells that contain only one copy of each gene instead of the usual two . These cells – referred to as haploid cells – are ideal for genetic studies because only a single copy of a gene has to be disrupted in order to analyze the consequences of that gene’s loss . The screens searched for genes that regulate WNT signaling: those that keep the pathway “off” in the absence of WNT and those that turn the pathway “on” in response to WNT . By comparing the outcomes of these screens , Lebensohn et al . identified previously unknown regulators and uncovered new roles for known regulators of the WNT pathway . For instance , a regulator called TFAP4 , which had not previously been linked to the pathway , was shown to activate WNT signaling . In another case , enzymes that make molecules called glycophosphatidylinositol anchors , and cell-surface proteins that are modified with those anchors , were found to amplify WNT signaling . Lebensohn et al . also identified genes that were needed to sustain the uncontrolled WNT signaling in cells that carried cancer-causing mutations in this pathway . Further studies could now explore if drugs can target these genes , or the molecules encoded by them , to treat cancers in which the WNT pathway is excessively activated . Other studies could also use the same methods to explore more signaling pathways and gain new insights into important biological processes in human cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"genetics",
"and",
"genomics"
] | 2016 | Comparative genetic screens in human cells reveal new regulatory mechanisms in WNT signaling |
Hypothalamic oxytocinergic magnocellular neurons have a fascinating ability to release peptide from both their axon terminals and from their dendrites . Existing data indicates that the relationship between somatic activity and dendritic release is not constant , but the mechanisms through which this relationship can be modulated are not completely understood . Here , we use a combination of electrical and optical recording techniques to quantify activity-induced calcium influx in proximal vs . distal dendrites of oxytocinergic magnocellular neurons located in the paraventricular nucleus of the hypothalamus ( OT-MCNs ) . Results reveal that the dendrites of OT-MCNs are weak conductors of somatic voltage changes; however , activity-induced dendritic calcium influx can be robustly regulated by both osmosensitive and non-osmosensitive ion channels located along the dendritic membrane . Overall , this study reveals that dendritic conductivity is a dynamic and endogenously regulated feature of OT-MCNs that is likely to have substantial functional impact on central oxytocin release .
Oxytocin ( OT ) is a nine amino acid peptide synthesized almost exclusively in hypothalamic neurons of the supraoptic and paraventricular nucleus ( SON , PVN ) . Despite the highly localized nature of OT synthesizing neurons , OT receptors ( OTRs ) are distributed widely throughout the CNS . Significant evidence indicates an important modulatory role for central oxytocinergic signaling in a wide variety of processes including fear conditioning , stress responding , anxiety-related behaviors , maternal behavior , sexual behavior , and social recognition ( e . g . see Bosch et al . , 2005; Knobloch et al . , 2012; Jurek et al . , 2015; Veening et al . , 2015; Neumann and Slattery , 2016; Lin et al . , 2018; Tan et al . , 2019; Valtcheva and Froemke , 2019; Winter and Jurek , 2019 ) . Numerous studies have demonstrated effects of OTR agonists and antagonists delivered intracerebrally or intraventricularly on social , stress , and anxiety-related behaviors , while conversely , animal models with genetic disruptions in OT signaling systems exhibit a range of social and behavioral deficits ( e . g . see Jin et al . , 2007; Young , 2007; Lee et al . , 2008; Higashida et al . , 2010; Pobbe et al . , 2012; Kent et al . , 2013; Morales-Rivera et al . , 2014; Peters et al . , 2014; Burkett et al . , 2016; Caldwell et al . , 2017; Lee et al . , 2018 ) . Collectively , these types of data implicate the central OT signaling system as a promising therapeutic target for a variety of conditions impacting mental health . However , effective therapeutic delivery of exogenous OTR agonists into the CNS of humans remains difficult ( Ermisch et al . , 1985; McEwen , 2004; Veening and Olivier , 2013; Leng and Ludwig , 2016; Quintana and Woolley , 2016; De Cagna et al . , 2019 ) , and thus a more detailed mechanistic understanding of how the brain naturally regulates release of endogenous OT may facilitate development of new approaches for therapeutic manipulation of central OT signaling . The question of exactly how , and from where , endogenous OT is released to act on both hypothalamic and extrahypothalamic OTRs in the CNS is a complex one . It is complicated by the fact that there are both magnocellular and parvocellular hypothalamic OT synthesizing neurons ( OT-MCNs , OT-PCNs ) , and by the fact that OT-MCNs can release peptide both from their axon terminals and from their dendrites . The current study focuses on dendritic physiology of PVN OT-MCNs because ( 1 ) OT-MCNs substantially outnumber OT-PCNs ( Althammer and Grinevich , 2017 ) , ( 2 ) most axons of OT-MCNs project through the median eminence to the posterior pituitary where activity ( action potential ) dependent release into the vasculature increases peripheral ( and not central ) OT concentration ( Guzek , 1987; Robinson et al . , 1989; Falke , 1991 ) , ( 3 ) a large portion of OT available for release into the CNS exists in dendritic rather than axonal vesicles that are subject to activity , and calcium , dependent exocytosis ( Pow and Morris , 1989; Ludwig et al . , 2002; Ludwig and Leng , 2006 ) , ( 4 ) such exocytosis supports functionally important peptide mediated paracrine signaling within the hypothalamus ( Son et al . , 2013; Smith et al . , 2015; Pati et al . , 2020 ) , and ( 5 ) likely also drives volume transmission to increase functional activation of OTRs in a variety of extrahypothalamic cortical and limbic areas ( Veening et al . , 2010; Fuxe et al . , 2012; Ludwig and Stern , 2015; Brown et al . , 2020 ) . Indeed , this mechanism seems likely to work in concert with limited/targeted release from centrally projecting axon collaterals of OT neurons , as has been effectively demonstrated in several extrahypothalamic areas to date ( Knobloch et al . , 2012; Eliava et al . , 2016; Oettl et al . , 2016 ) . In the current study , we use a combination of electrophysiological and subcellular optical recording techniques to evaluate dendritic physiology of OT-MCNs . Somatic activity was induced with a reproducible train of action potential-like voltage pulses delivered to the soma , and calcium influx induced by this somatic activity was quantified using high frequency two-photon line scans across the proximal and distal dendrite . The results reveal that the dendrites of OT-MCNs are weak conductors of somatic voltage changes . We further report that acute hyperosmotic challenge preferentially reduces activity-induced calcium influx in distal vs . proximal OT-MCN dendrites , while acute hypoosmotic challenge increases it . Extensive control experiments indicate these effects are likely mediated by modulation of a previously unidentified osmosensitive channel expressed along the dendritic membrane . Finally , we report that that activity-induced calcium influx in the distal dendrites of OT-MCNs is also preferentially and robustly inhibited , absent any change in osmolarity , by activation of dendritic GABAA receptors . Collectively , these results significantly increase our understanding of the mechanisms through which OT-MCNs are likely to dynamically regulate the relationship between somatic activity and calcium-dependent dendritic release of OT into the CNS .
All experiments for this study were performed in an OT-reporter mouse line designed to selectively expresses a red fluorescent protein ( tdTomato ) in oxytocinergic neurons ( Figure 1A , see Materials and methods ) . In order to validate specificity and selectivity of this reporter in the PVN , we used immunohistochemical techniques to evaluate co-expression of tdTomato and neurophysin 1 ( NP1 , an OT carrier protein found in oxytocinergic neurons , Figure 1B ) . Overall , we found that a strong majority of tdTomato expressing neurons in the PVN were immunoreactive for NP1 ( 92 . 8 ± 0 . 8% across all animals tested , 93 . 2 ± 1 . 0% in males , and 92 . 4 ± 0 . 5% in females , n = four animals total , two male , two female , 2838 total tdTomato-positive PVN neurons evaluated , 1352 from males , and 1486 from females ) . Based on these data , we used a combination of IR-DIC and epifluorescence microscopy ( see Figure 1C and Materials and methods ) to efficiently target PVN OT neurons for whole-cell patch clamp recording . Once patched , PVN OT neurons were categorized as either magnocellular or parvocellular based on the presence or absence ( respectively ) of a transient outwardly rectifying potassium current , IA ( Luther and Tasker , 2000 ) . This current was readily apparent as an outward current in voltage clamp observed in the first 100 msec after stepping from −70 mV to −50 mV ( Figure 1D–E ) . It also contributed to a clear delay to first action potential observed in current clamp in response to a suprathreshold current injection ( Figure 1F–G ) . By contrast , OT-PCNs not only lacked IA , but often also displayed a small inward current after stepping from −70 mV to −50 mV in voltage clamp ( Figure 1D , bottom trace ) . This current often produced a low threshold spike that was apparent in current clamp ( Figure 1F , bottom trace ) , and contributed to the short delay to first spike observed in OT-PCNs ( Figure 1G ) . We noted that OT-MCNs had significantly higher somatic input resistance than OT-PCNs ( 1836 . 87 ± 107 . 84 MΩ vs . 854 . 81 ± 73 . 92 MΩ , respectively , n=62 , 22 , t = 7 . 5 , p<0 . 0001 , unpaired two-sample t-test ) , while whole-cell capacitance at −70 mV did not significantly differ ( Cm: 31 . 1 ± 1 . 4 pF vs . 29 . 82 ± 1 . 8 pF , n=62 , 22 , t = 0 . 5 , p = 0 . 621 , unpaired two-sample t-test ) . Importantly , none of the core intrinsic electrophysiological features of OT-MCNs reported here varied by sex ( See Figure 1—figure supplement 1 and legend ) . In order to directly evaluate the osmosensitivity of PVN OT-MCNs in vitro , we used whole-cell current clamp recordings , in combination with bath application of mannitol ( MT , an inert osmolyte ) , to evaluate the effect of acute hyperosmotic challenge on firing frequency . We found that bath application of 30 mOsm MT for 5 min ( in the presence of glutamate and GABA receptor antagonists , see Materials and methods ) increased basal firing rate of OT-MCNs by 3 . 37 ± 0 . 55 Hz ( n = 9 , t = 6 . 2 , p = 0 . 0002 , Figure 2A–B . ) A smaller but longer lasting hyperosmotic stimulus ( 15 mOsm MT for 10 mins ) increased basal firing rate by 1 . 39 ± 0 . 55 Hz ( n=9 , t = 2 . 6 , p = 0 . 03 , Figure 2B–C ) , and was chosen as a standard hyperosmotic stimulus throughout the rest of the study . This effect of MT is cell type specific , as 30 mOsm MT failed to produce a similar effect in PVN OT-PCNs ( ΔFreq: 0 . 7 ± 0 . 3 Hz , t = 2 . 49 , p = 0 . 06 , t = 3 . 58 , p = 0 . 005 vs . effect of 30 mOsm MT observed in OT-MCNs , Figure 2C ) . Although additional synaptic input from osmosensitive circumventricular organs might further enhance OT-MCN activity in vivo during times of increased osmolarity , the present results demonstrate PVT OT-MCNs are directly osmosensitive and highlight that a primary effect of acute hyperosmotic challenge , when observed somatically , is excitatory . It is clear from prior literature that dendritic release of OT from MCNs is calcium dependent , and it is often presumed that somatic action potentials lead to dendritic calcium influx in a way that promotes dendritic release of peptide ( Di Scala-Guenot et al . , 1987; Fisher and Bourque , 1996; Tobin et al . , 2012 ) . That said , the relationship between somatic activity and dendritic release of OT is likely not constant ( see Discussion ) , and key aspects of dendritic physiology potentially impacting this relationship have not been directly examined before . For these reasons , we used a combination of electrophysiological and optical techniques to quantify activity-induced calcium influx along the length of OT-MCN dendrites as observed in response to a two-second train of action potential like voltage steps delivered to the soma at 20 Hz ( Figure 3A , calcium indicator delivered via the patch pipette , see Materials and methods ) . This approach revealed that activity-induced calcium influx drops rapidly with increasing distance from the soma ( main effect of distance , F1 . 6 , , 6 . 4 = 15 . 0 , p = 0 . 005 , one-way repeated measures ANOVA , Figure 3B , Figure 3C , blue trace ) , suggesting that under basal conditions OT-MCN dendrites lack the ability to actively propagate somatic voltage changes to distal locations , and likely also have a relatively low membrane resistance . However , similar data might be expected if calcium indicators had failed to perfuse into the distal dendrites , or if distal dendrites expressed substantially fewer voltage-gated calcium channels . To evaluate both possibilities simultaneously , we repeated the experiment using a cesium gluconate internal solution ( see Materials and methods ) to block cesium-sensitive leak channels and thereby increase dendritic membrane resistance . Notably , in these conditions , we observed an overall increase in activity-induced calcium influx ( main effect of internal , F1 , 9 = 30 . 55 , p = 3 . 7 x 10−4 , two-way repeated measures ANOVA , Figure 3C , red trace ) , with post-hoc tests indicating significantly increased response at distances > 60 µm from the soma . These results indicate that functional calcium indictor was present in the distal dendrites , and that voltage-gated calcium channels are still robustly expressed at distal locations . As such , they also substantially reinforce the conclusion that loss of activity-induced calcium influx at increasing distance from the soma , as observed when using a more physiological K-gluconate-based internal solution , was produced by loss of current through open channels along the dendritic membrane . Next , we repeated the experiment for a third time , using a K-gluconate-based internal solution , to evaluate activity-induced calcium influx along the length of OT-MCN axons . Axons were distinguished from dendrites based primarily on their smaller initial diameter as observed with 2P epifluorescence microscopy ( e . g . Figure 4A ) . However , consistent with prior reports ( Hatton , 1990; Stern and Armstrong , 1998 ) , we also noted that axons often ( but not always ) branch off a primary dendrite very close to the soma . The results of this experiment indicated that OT-MCN axons , unlike dendrites , reliably and actively propagate somatic voltage changes in cells filled with potassium gluconate ( Figure 3C , green trace ) . Specifically , a two-way repeated measures ANOVA revealed a main effect of structure on activity-induced calcium influx ( axon vs . dendrite , both recorded using a K-gluconate-based internal solution , F1 , 4 = 9 . 17 , p = 0 . 04 ) , while post-hoc tests revealed that significantly less loss of response was observed in axons compared to dendrites at distances > 90 µm from the soma . Interestingly , these data also highlight expression of voltage-gated calcium channels along the length of OT-MCN axons . Finally , we tested the effect of removing external calcium on activity-induced calcium influx as observed in OT-MCN dendrites . Specifically , we found that a 15 min exposure to calcium-free ACSF ( see Materials and methods ) reduced activity-induced calcium influx by 80 . 3 ± 3 . 5% when measured in the proximal dendrite ( ~25 µm from the soma , n=5 , t = −22 . 9 , p = 2 . 15 x 10−5 , one-sample t-test ) , and by 67 . 9 ± 3 . 3% when measured in the distal dendrite ( ~125 µm form the soma , n=6 , t = −20 . 75 , p = 4 . 81 x 10−6 , one-sample t-test ) . Both effects rapidly recover when extracellular calcium is restored . These results , presented in Figure 3—figure supplement 1 , indicate that external calcium influx is the primary driver of the calcium signal observed in OT-MCN dendrites after somatic stimulation . We next tested the hypothesis that acute hyperosmotic challenge ( as produced by bath application of 15 mOsm MT ) directly modulates the relationship between somatic activity and activity-induced dendritic calcium influx in OT-MCNs . Toward that end , we used a technical approach similar to that employed in Figure 3; however , instead of measuring activity-induced influx at multiple locations under control conditions , we picked just two dendritic locations , proximal and distal to the soma , and repeatedly measured activity-induced calcium influx before and after acute hyperosmotic challenge . Proximal dendritic locations were again defined as being ~25 µm from the soma , while distal ones were located at ~125 µm from the soma ( See Figure 4A , blue and red dashed lines , respectively ) . In order to generate activity-induced calcium influx at these locations the soma was stimulated with the same 2 s 20 Hz train of action potential like voltage steps as used in Figure 3 , 2P line scan data were collected from each dendritic location and analyzed in an identical manner , and experiments were again performed in the continuous presence of bath applied antagonists for glutamate and GABA receptors ( see previous Results section and Materials and methods ) . We found that acute hyperosmotic challenge had a small inhibitory effect on activity-induced calcium influx in the proximal dendrites of OT-MCNs ( −14 . 3 ± 3 . 37% , n = 13 , t = −4 . 2 , p = 0 . 001 , one-sample t-test ) , but had a much larger inhibitory effect in the distal dendrites ( −40 . 6 ± 4 . 23% , n = 20 , t = 4 . 44 , p = 1 . 1 x 10−4 vs . proximal , Figure 4B , Figure 4C , left panel , see Figure 4—figure supplement 1 for data separated by sex ) . Notably , these changes were not associated with any significant/detectable impact on somatic membrane resistance ( baseline: 1355 ± 109 . 8 MΩ , after MT: 1256 ± 109 . 3 MΩ , n=38 , t = 1 . 38 , p = 0 . 18 , paired two-sample t-test ) , or on the voltage clamp current applied to the soma during the train ( 94 . 9 ± 5 . 21% of baseline after MT , n=11 , t = −1 . 0 , p = 0 . 4 , one-sample t-test ) . Considered together , these observations suggest a site of action along the dendritic membrane . Although glutamate and GABA receptors were antagonized during the experiments described above , it was possible that increased osmolarity might promote action potential dependent release of other modulators which then act locally on OT-MCN dendrites to reduce activity-induced calcium influx . In order to test this possibility , we repeated the experiment above with 1 µM TTX in the bath ( in addition to glutamate and GABA receptor antagonists ) . However , under these conditions , we found that acute hyperosmotic challenge continued to robustly and preferentially inhibit activity-induced calcium influx in the distal vs . proximal dendrites ( by −55 . 8 ± 6 . 53 vs . −10 . 5 ± 11 . 0% , distal , proximal , n=9 , nine respectively , t = 4 . 51 , p = 0 . 002 , Figure 4C , right panel ) , suggesting activity-induced release of other modulators is not required . Data from this same experiment further revealed that the ratio of distal to proximal calcium influx within individual OT-MCNs in response to somatic stimulation did not change with bath application of TTX ( n=5 , ratio = 0 . 52 ± . 1 and 0 . 40 ± 0 . 1 before and after bath application of TTX respectively , t = 1 . 43 , p = 0 . 23 , paired two-sample t-test ) . This result further indicates that TTX-sensitive voltage-gated sodium channels play no apparent role in supporting propagation of voltage in OT-MCN dendrites . Next , in order to eliminate any unexpected impact of the extra time required to change bath conditions in this experiment , we measured activity-induced calcium influx exclusively in the distal dendrites after 15 and 25 min of recording under constant bath conditions and noted that there was no significant change when measured at either time point ( −2 . 0 ± 3 . 8% , 4 . 6 ± 5 . 3% , at 15 and 25 min , respectively , n=4 , 4; t = −0 . 52 , 0 . 86; p = 0 . 64 , 0 . 45 , one-sample t-tests , Figure 4D ) , or when measured across time points ( n=4 , 4 , t = −1 . 28 , p = 0 . 29 , paired two-sample t-test ) . Similarly , we asked whether acute hyperosmotic challenge caused any change in basal calcium levels prior to somatic stimulation , and whether any such changes would impact subsequent measurement of activity-induced calcium influx . We found that on average , basal calcium levels changed by ≤ 6 . 1% in response to acute hyperosmotic challenge , and that these small changes were not correlated with observed effects on activity-induced calcium influx ( Figure 4—figure supplement 2 ) . Finally , we also evaluated the effect of acute hyperosmotic challenge on activity-induced calcium influx in OT-MCNs in current clamp , using a train of suprathreshold 250 μsec current pulses , delivered to the soma at 20 Hz for 2 s . We noted that within each individual OT-MCN , action potential shape and amplitude observed during this stimulus closely matched that observed during spontaneous firing ( e . g . see Figure 4—figure supplement 3A ) . Further , the effects of acute hyperosmotic challenge on activity-induced calcium influx as observed in both proximal and distal dendrites were comparable to those observed in voltage clamp ( proximal: −13 . 3 ± 3 . 61% , distal: −38 . 3 ± 3 . 92% , n=7 , 6 , t = 4 . 68 , p = 0 . 0007 , Figure 4—figure supplement 3B–C ) . Collectively , these data indicate that despite having an excitatory effect on action potential frequency as observed in the soma ( Figure 2 ) , acute hyperosmotic challenge also preferentially reduces activity-induced calcium influx as observed in the distal vs . proximal dendrites of OT-MCNs . To further test the hypothesis that acute hyperosmotic challenge decreases membrane resistance , and thus conductivity , in OT-MCN dendrites , we performed an additional control experiment . Specifically , we initiated whole-cell patch clamp recordings from OT-MCNs using the same internal and external solutions as was used above , except now without glutamate receptor antagonists in the bath . We then stimulated them alternately ( every 15 s ) with two distinct stimuli ( one delivered to the soma and one to the distal dendrites ) . The somatic stimulus was the same 2 s 20 Hz train of action potential like voltage steps used in earlier experiments , while the dendritic stimulus was focal application of exogenous glutamate ( accomplished using a picospritzer , see Materials and methods ) . For each stimulus , responses were measured in both the soma ( as a whole-cell current ) and in a distal dendrite ( as ΔF/F generated with a 2P line scan ) . Collectively , this experimental design ( Figure 5A-B ) provides insight , in each individual OT-MCN tested , on how acute hyperosmotic challenge effects current propagating along the dendrite in both directions , either from the soma toward the distal dendrites , or from the distal dendrites toward the soma . Acute hyperosmotic challenge reduced somatic current observed in response to exogenous glutamate delivered to the distal dendrite by 62 . 5 ± 10 . 1% ( n = 5 , t = −6 . 17 , p = 0 . 003 , Figure 5C ) , but had no effect on somatic current observed in response to a train of AP like voltage pulses delivered to the soma ( Δ current = −2 . 4 ± 4 . 53% , n = 10 , t = −0 . 5 , p = 0 . 612 , Figure 5C ) . Conversely , acute hyperosmotic challenge reduced dendritic calcium influx produced by delivering a train of voltage pulses to the soma ( by −33 . 13 ± 7 . 5% , n = 10 , t = −4 . 4 , p = 0 . 002 , Figure 5D ) , but had no effect on dendritic calcium influx observed in response to locally applied glutamate ( Δ Peak Δ F/F: 1 . 94 ± 8 . 7% of baseline , n = 10 , t = 0 . 22 , p = 0 . 828 , Figure 5D ) . These results demonstrate that acute hyperosmotic challenge consistently and selectively inhibits whichever response is measured distal from the stimulus that produced it , irrespective of whether that response is measured using electrical or optical techniques . These results , in combination with other data presented above , effectively rule out the idea that the observed effects of acute hyperosmotic challenge depend on direct modulation of voltage-gated calcium channels , calcium induced calcium release , or on other aspects of calcium homeostasis in the dendrites . As such , they are consistent with the hypothesis that acute hyperosmotic challenge is directly modulating dendritic membrane resistance , and thus voltage propagation , in OT-MCNs . Finally , we attempted to directly measure changes in dendritic membrane resistance during an acute hyperosmotic challenge using dual patch clamp recordings from the soma and distal dendrites of a single OT-MCN . This approach has been used successfully to study the physiology of large dendrites in rat cortical pyramidal neurons and in cerebellar Purkinje cells; however , existing literature highlights that dendritic patching in smaller multipolar neurons is only feasible at proximal locations ( Davie et al . , 2006 ) . Indeed , we found that the very small diameter and variable orientation of mouse OT-MCN distal dendrites , combined with tissue movement associated with changes in osmotic pressure , prohibited direct measurement of dendritic membrane resistance in this manner . Thus , we conclude that the combined electrical / optical approaches used in this study effectively and reliably reveal novel aspects of dendritic physiology in OT-MCNs that are not readily accessible to direct electrical recording . Next , to eliminate any possible generalized or nonspecific effects of acute hyperosmotic challenge , we designed experiments to test the hypothesis that the effects are both compartment specific and cell type specific . Compartment specificity was evaluated using techniques identical to those employed for Figure 4 , except that we compared activity-induced calcium influx in the distal dendrite to that observed in the distal axon . We found that acute hyperosmotic challenge again effectively inhibited activity-induced calcium influx in distal dendrites ( Δ Peak Δ F/F: −44 . 0 ± 8 . 0% , n = 5 , t = −5 . 5 , p = 0 . 005 ) , but produced a much smaller effect in the distal axon ( Δ Peak Δ F/F: −10 . 0 ± 2 . 3% , n = 5 , t = 3 . 4 , p = 0 . 027 vs . distal dendrite , Figure 6A , E ) . This result demonstrates significant compartment specificity within individual OT-MCNs . In order to evaluate cell type specificity , we used identical approaches to measure activity-induced calcium influx in proximal vs . distal dendrites in PVN OT-PCNs ( identified as described in Figure 1 ) , and in CA1 pyramidal cells . In OT-PCNs , there was no significant effect of acute hyperosmotic challenge in proximal or distal dendrites ( Δ Peak Δ F/F: −7 . 7 ± 4 . 3% , n = 5 , t = −1 . 8 , p = 0 . 147; 6 . 6 ± 3 . 4% , n = 5 , t = 1 . 96 , p = 0 . 12 , respectively , Figure 6B , F ) . In CA1 pyramidal cells , we noted a weak inhibitory effect in proximal dendrites , but no effect in distal dendrites ( proximal: −6 . 4 ± 2 . 4% , n=6 , t = −2 . 7 , p = 0 . 045; distal: −3 . 80 ± 3 . 5% , n = 6 , t = −1 . 1 , p = 0 . 33 , Figure 6C , G ) . These results effectively demonstrate cell type specificity . Next , we reasoned that if acute hyperosmotic challenge inhibits calcium influx in the distal dendrites of OT-MCNs by opening a distinct dendritic osmosensitive ion channel , and if that channel is not completely closed in control conditions , then an acute hypoosmotic stimulus should have opposite effects . Indeed , we found that acute reduction of osmolarity by 30 mOsm ( achieved by diluting the bath solution with water ) effectively increased activity-induced calcium influx in the distal dendrites of OT-MCNs ( by 32 ± 10 . 7% of baseline , n = 10 , t = 3 . 0 , p = 0 . 014 , one-sample t-test ) , while having no effect in the proximal dendrites ( −2 . 03 ± 3 . 18% of baseline , n = 10 , t = −0 . 6 , p = 0 . 536 , one-sample t-test , Figure 6D , H ) . As with effects of acute hyperosmotic challenge , these changes occurred absent any significant effect on somatic membrane resistance or voltage clamp current observed during stimulation ( n=10 , t = 1 . 3 , p = 0 . 21; n=10 , t = −1 . 84 , p = 0 . 1 , respectively , one-sample t-tests ) . Note that the hypoosmotic stimulus used here will dilute extracellular calcium by ~10% ( from ~2 . 4 mM to ~2 . 16 mM ) , which is expected to decrease driving force on calcium by < 2 mV . Collectively , these results demonstrate that the effects of acute osmotic challenge on activity-induced calcium influx , as observed in the distal dendrites of OT-MCNs , are compartment specific , cell type specific , and bidirectional . The observation that the somatic response to activation of glutamate receptors on the distal dendrites of OT-MCNs is significantly inhibited by acute hyperosmotic challenge , in combination with other results above , suggests that changes in dendritic membrane resistance are likely to impact integration of synaptic inputs , as well as activity-induced calcium influx . In order to test this hypothesis directly we used minimal stimulation techniques ( see Materials and methods ) to evoke glutamate release from one or few axons that make synaptic contact with either the proximal or distal dendrites of an OT-MCN ( Figure 7A ) . After identifying a clear evoked excitatory postsynaptic current ( eEPSC ) , we bath applied 15 mM MT as in prior experiments . We found that acute hyperosmotic challenge reliably and reversibly reduced eEPSC amplitude as evoked by a minimal stimulator placed near the distal dendrite ( by 63 . 4 ± 8 . 7% , n = 8 , t = −7 . 3 , p = 1 . 6 x 10−4 , one-sample t-test Figure 7A , C ) . As in earlier experiments , this result was not associated with a significant change in somatic membrane resistance ( n=8 , t = −0 . 14 , p = 0 . 90 , paired two-sample t-test ) . If , as expected , the effect is instead produced primarily by a drop in dendritic membrane resistance , then the same acute osmotic stimulus should have less of an inhibitory effect on EPSCs generated using a minimal stimulator placed near the proximal dendrite . Indeed , consistent with this hypothesis , we found that 15 mM MT reduced proximally evoked EPSC amplitude by 30 ± 9 . 6% ( n=6 , t = −3 . 16 , p = 0 . 03 , Figure 7A , B ) . Consistent with our hypothesis , this effect is significantly smaller than observed when using a stimulator placed near the distal dendrite ( Figure 7D , n=6 , 8 , proximal , distal , respectively , t = 2 . 55 , p = 0 . 03 ) . In order to confirm that eEPSCs involved in these experiments were glutamatergic , a subset of MT-sensitive responses ( n=3 ) were challenged with bath applied glutamate receptor antagonists after recovery , and were effectively eliminated ( not illustrated ) . The observation that acute hyperosmotic challenge preferentially inhibits activity-induced calcium influx in the distal vs . proximal dendrites of OT-MCNs suggests that there are functional osmosensitive ion channels expressed along the length of the dendritic membrane . By contrast , the observation that the same osmostic stimulus produces an increase in spontaneous firing frequency of OT-MCNs as measured in current clamp , suggests that functional somatic osmosensors also exist . Because the somatic effect on spontaneous firing frequency requires depolarization , but the dendritic effect on activity-induced calcium influx does not significantly alter basal calcium levels , we hypothesized that somatic and dendritic osmosensors are likely to be molecularly distinct . Significant prior work has indicated that vasopressinergic MCNs are osmosensitive and has further revealed that transient receptor potential subfamily V member 1 ( TRPV1 ) channels contribute to their osmosensitivity ( Sharif Naeini et al . , 2006; Bourque , 2008; Moriya et al . , 2015; Prager-Khoutorsky and Bourque , 2015 ) . Therefore , we evaluated the effect of ruthenium red ( RR ) , a generic antagonist of transient receptor potential ( TRP ) ion channels , on both somatic and dendritic effects of acute hyperosmotic challenge observed in OT-MCNs . We found that pre-treatment with 10 µM RR blocked the effect of acute hyperosmotic challenge on basal firing rate as observed in current clamp ( Δ action potential frequency: 0 . 28 ± 0 . 21 , n=7 , t = 0 . 5 , p = 0 . 61 , t = 4 . 0 , p = 0 . 001 vs . response to same stimulus absent RR , Figure 8A–B ) . Conversely , RR did not alter the effect of acute hyperosmotic challenge on activity-induced calcium influx as observed in either the proximal or distal dendrites ( Figure 8C–D , see text of legend for further details ) . Similarly , RR also did not block the inhibitory effect of acute hyperosmotic challenge observed on distally evoked EPSCs ( Δ EPSC amplitude: −71 . 5 ± 6 . 0% , n=6 , t = −11 . 9 , p<0 . 001 , t = 0 . 72 , p = 0 . 49 vs . effect observed absent RR , See Figure 8—figure supplement 1 ) . Collectively , these results reinforce the hypothesis that PVN OT-MCNs express distinct osmosensors in somatic vs . dendritic compartments and suggest that the somatic but not dendritic osmosensor may be a member of the TRPV family ( see Discussion ) . Interestingly , we also noted that the effect of MT on activity-induced calcium influx as observed in the distal dendrites of OT-MCNs is blocked in cells filled with a cesium-gluconate internal solution , which suggests that the dendritic osmosensor may be cesium sensitive ( See Figure 8—figure supplement 2 for additional details ) . Next , we sought to determine whether dendritic membrane resistance in OT-MCNs is an aspect of dendritic physiology that can be actively manipulated to modify the relationship between somatic activity and dendritic calcium influx , even under conditions that do not involve changes in osmolarity . Toward that end , we performed experiments similar to those presented in Figure 4 , except we replaced the hyperosmotic stimulus with bath application of 400 nM muscimol ( a GABAA receptor agonist ) . Unlike acute hyperosmotic challenge , bath application of muscimol significantly reduced somatic membrane resistance ( from 1178 ± 166 . 6 MΩ to 226 ± 43 . 5 MΩ , n = 6 , t = 4 . 9 , p = 0 . 004 , paired two-sample t-test ) and produced a tonic inhibitory current ( of 15 . 4 ± 5 . 42 pA , n = 6 , t = 2 . 8 , p = 0 . 036 , one-sample t-test ) apparent in cells voltage clamped at −70 mV , consistent with tonic activation of somatic GABAA receptors . However , importantly , like acute hyperosmotic challenge , muscimol preferentially inhibited activity-induced calcium influx in the distal vs . proximal dendrites ( −75 . 8 ± 2 . 7% vs . −31 . 6 ± 6 . 4% , respectively , n=6 , t = 5 . 6 , p = 0 . 003 , Figure 9 ) . This finding is consistent with opening of GABA receptors along the dendritic membrane ( see also Pirker et al . , 2000; Park et al . , 2006 ) , and notably , highlights that dendritic membrane resistance is likely to be under constant and dynamic regulation in OT-MCNs even absent changes in osmolarity .
This study uses a combination of electrical and optical recording techniques to examine activity-induced calcium influx in the dendrites of PVN OT-MCNs in OT-tdTomato reporter mice . Somatic activity was induced with a well-controlled train of action potential like stimuli delivered to the soma , while activity-induced calcium influx was measured in OT-MCN dendrites using quantitative two-photon microscopy . We demonstrate that PVN OT-MCN dendrites are weak conductors of somatic voltage changes in basal conditions in both male and virgin female mice , and importantly , we also find that activity-induced calcium influx in the dendrites is subject to robust modulation by a diverse set of stimuli acting on distinct types of ion channels in the dendritic membrane . The primary stimulus used in the current study is an acute increase in osmotic pressure . Hyperosmotic stimuli are well-recognized for an ability to increase activity of hypothalamic MCNs , and in so doing , for promoting significant increases in both peripheral and central OT concentration ( Mason , 1980; Bourque and Renaud , 1984; Ludwig et al . , 1994; Ludwig , 1998; Tasker et al . , 2020 ) . Increases in peripheral OT concentration are understood to depend on axonal release in the neurohypophysis ( which delivers OT directly into the vasculature ) , while increases in central concentration are likely to depend heavily on dendritic release into the extracellular space , the subarachnoid space , and/or the third ventricle ( Bourque , 1991; Ludwig and Leng , 2006; Veening et al . , 2010 ) . A curious feature of the neurohormonal response to hyperosmotic challenge is that increases in peripheral OT concentration are both rapid and frequency dependent ( with concentration closely tracking MCN activity ) , while increases in central OT concentration are temporally separated from peak MCN firing , often by over an hour ( Ludwig et al . , 1994; Ludwig , 1998 ) . This is somewhat counter intuitive because like axon terminals , OT-MCN dendrites also contain many large dense core vesicles loaded with oxytocin , and these dendritic vesicles are also subject to both activity and calcium-dependent release ( Mason et al . , 1986; Pow and Morris , 1989; Ludwig et al . , 1995; Wang et al . , 1995 ) . Notably , other types of peripheral stimuli ( e . g . suckling ) can drive more concurrent increases in peripheral and central OT concentration suggesting more synchronous release of OT from both axon terminals and dendrites ( Neumann et al . , 1993 ) , while some locally synthesized signaling molecules ( e . g . α-melanocyte stimulating hormone ) have been demonstrated to promote dendritic release of OT while actively inhibiting both firing rate and release from axon terminals in the neurohypophysis ( Neumann et al . , 1993; Sabatier et al . , 2003; Sabatier , 2006 ) . Collectively , these types of data effectively highlight that although at least loosely coupled , the relationship between somatic activity and dendritic release of peptide in hypothalamic MCNs is likely subject to modulation . The most well-established basis for understanding this flexibility invokes a model of conditional priming , whereby specific endogenous modulators , acting directly on the dendrites , can prime dendritic vesicles to be more available for activity-induced release ( Morris and Ludwig , 2004; Ludwig and Leng , 2006 ) . However , in the specific case of hyperosmotic challenge , maximizing calcium-dependent dendritic priming prior to hyperosmotic challenge was found to increase the amount of dendritic release of OT , remarkably , without altering its time course ( Ludwig et al . , 2002 ) . This striking result suggests that there must be some aspect of OT-MCN physiology that is activated by hyperosmotic challenge , that is separate from conditional priming , and that is capable of rapidly yet transiently reducing the probability of activity-induced exocytosis of dendritic OT . In that regard , a key finding of the current study is that an acute hyperosmotic stimulus delivered in vitro decreases activity-induced calcium influx in OT-MCN dendrites , while an acute hypoosmotic stimulus increases it . In each case , we noted minimal effect of changing bath conditions on somatic input resistance , or on somatic current observed during the stimulus , and further noted that changes in osmolarity had a more robust effect in distal vs . proximal dendrites . We noted no similar inhibitory effect of acute hyperosmotic stimuli on activity-induced calcium influx in the distal axons of OT-MCNs , indicating compartment specificity , or in the distal dendrites of OT-PCNs or CA1 pyramidal cells , demonstrating cell-type specificity . Further , in experiments that involved both somatic and dendritic stimulation techniques , we demonstrated that hyperosmotic challenge selectively inhibits responses measured distal from the stimulus , irrespective of whether those measurements are made using electrical or optical techniques . Collectively , these data indicate for the first time that osmolarity modulates activity-induced calcium influx in OT-MCN dendrites by acting on osmosensitive ion channels expressed along the dendritic membrane . Based on these data , it is reasonable to postulate that this mechanism may reduce activity-induced dendritic release of OT during times of high somatic activity as induced by acute hyperosmotic challenge . As such , it may be interesting for future studies to evaluate whether low levels of dendritic calcium influx produced by somatic activity when dendritic osmosensitive channels are open , or concurrent action of other dendritic modulators , helps promote priming of dendritic vesicles in a way that contributes to enhanced dendritic release once basal osmolarity is restored . Another aspect of this study worth specifically highlighting is the novel implication that OT-MCNs express molecularly distinct osmosensitive channels in their soma vs . dendrites . This conclusion is supported by the observation that a non-selective TRP ion channel receptor antagonist , RR , blocked the effect of acute hyperosmotic challenge on action potential firing frequency without altering effects on activity-induced calcium influx as observed in either the proximal or distal dendrites , or the effects on evoked EPSCs produced by a minimal stimulator placed near the distal dendrite . RR is an antagonist of TRPM6 , TRPA1 , TRPC3 , and of all members of the TRPV family ( TRPV1-6 , Clapham , 2007; Lichtenegger and Groschner , 2014 ) . That said , to our knowledge TRPV channels are the only RR sensitive channels that have potential for osmosensitivity and that are strongly expressed in the hypothalamus ( Sharif Naeini et al . , 2006; Sladek and Johnson , 2013; Prager-Khoutorsky and Bourque , 2015; Zaelzer et al . , 2015; Shenton and Pyner , 2018 ) , suggesting that the somatic osmosensor described here may be a member of the TRPV family . Additional potential effects of RR , for example on ryanodine receptors or mitochondrial calcium transporters , are not expected to be a factor in current experiments since RR is membrane impermeant ( Luft , 1971a; Luft , 1971b; Garcha and Hughes , 1994 ) , and we only applied it extracellularly . Additional evidence for the hypothesis that OT-MCNs have distinct somatic and dendritic osmosensors comes from the observation that intracellular cesium blocked the effects of acute hyperosmotic challenge on activity-induced calcium influx in the distal dendrites even through TRPV receptors are cesium permeant ( Caterina et al . , 1997; Puopolo et al . , 2013 ) . Thus , overall , we hypothesize that the somatic osmosensor described here is a member of the TRPV family , while the dendritic osmosensor is , or is coupled to , a cesium sensitive and potassium permeant channel . While some prior literature is generally consistent with these hypotheses ( Liu et al . , 2005; Sharif Naeini et al . , 2006; Zhang et al . , 2009; Prager-Khoutorsky and Bourque , 2015 ) , very few studies have examined these questions with respect to OT-MCNs in particular . Next , it is interesting to highlight that all novel aspects of OT-MCN dendritic physiology revealed here , as well as most other core intrinsic features of OT-MCNs observed , were identical in male and virgin female mice , suggesting that they have a fundamental and sex independent role in regulation of oxytocinergic signaling . That said , it seems plausible that aspects of OT-MCN physiology likely relevant to central OT signaling , such as the dendritic conductivity , could be regulated in a context and sex specific way . As such , it may be interesting to determine whether activity-induced calcium influx in the distal dendrites of OT-MCNs is altered in pregnant or lactating females . Indeed , concurrent increases in both peripheral and central release of OT have been reported in response to suckling ( Neumann et al . , 1993 ) . Other aspects of this study make two additional important points . First , we demonstrate directly that the effects of acute hyperosmotic challenge on OT-MCN dendrites are not limited to modulation of activity-induced calcium influx , but also robustly inhibit the somatic response to endogenous synaptic inputs arriving at the distal dendrites . This finding , in combination with other observed effects , suggests that activation of dendritic osmosensors not only transiently reduces the probability of activity-induced dendritic release of OT into the CNS , but also simultaneously reduces the impact of descending central inputs forming dendritic synapses on MCN firing rate . These changes are expected to effectively but transiently prioritize activity-induced increases in peripheral OT concentration . Second , we report that an ability to modulate activity-induced calcium influx by acting on dendritic ion channels can also be produced by stimuli that do not alter osmolarity . Specifically , we note that bath application of a GABAA receptor agonist not only decreases somatic membrane resistance in OT-MCNs but also preferentially inhibits activity-induced calcium influx in the distal vs . proximal dendrites . The later finding is consistent with prior reports that multiple types of GABAergic receptor subunits , including the typically extrasynaptic δ subunit , are expressed on MCN dendrites and/or in magnocellular regions of the PVN ( Fenelon et al . , 1995; Pirker et al . , 2000; Belelli et al . , 2009 ) , and is interesting to consider along with the fact tonic GABAergic currents have been directly observed in MCNs ( Park et al . , 2006 ) . It is also important in the context of this study because it highlights that endogenous regulation of dendritic conductivity may represent an important mechanism for regulating central OT concentration even in situations where osmolarity remains constant . For example , it is intriguing to note that OT-MCNs receive significantly more frequent bursts of GABAergic IPSCs during lactation ( Popescu et al . , 2019 ) , and that lactation has also been associated with a significant positive shift in the reversal potential of GABA-receptor-mediated currents ( Lee et al . , 2015 ) . Overall , we expect that it will be important for future studies to evaluate the ability of additional endogenous modulators to impact conductivity of OT-MCN dendrites , and to identify specific contexts in which such modulators are active . Finally , as noted in more detail in the Introduction , central OT signaling systems are a promising therapeutic target for a variety of conditions impacting mental health , and yet the best available current strategy for therapeutically modulating activation of central OTRs in humans involves intranasal delivery of an exogenous agonist that has low permeability to the blood brain barrier ( Evans et al . , 2014; Leng and Ludwig , 2016; Quintana and Woolley , 2016; Quintana et al . , 2018 ) . In our view , a better understanding of OT-MCN physiology , particularly as it relates to release of endogenous OT in the CNS , may ultimately lead to improved therapeutic options that are better able to mimic natural concentration , kinetic , and context-dependent aspects of central OT signaling . In the current study , measures of activity-induced calcium influx in OT-MCN dendrites reveals a novel aspect of dendritic physiology likely to underlie the flexible relationship between somatic activity and dendritic release of OT . Future studies may develop new methods for spatially precise quantification of dendritic exocytosis and/or for detection of quantal amounts of OT release very close to the dendritic membrane , which would in turn promote a more direct evaluation of the relationship between dendritic calcium influx and dendritic exocytosis .
All experiments in this study were performed using 1- to 3-month-old OT-reporter mice which expresses red fluorescent protein ( tdTomato ) in oxytocinergic neurons . These mice were generated by crossing OT-IRES-Cre knock-in mice which express Cre targeted to the Oxt locus ( Jackson Labs Stock #024234 ) with Ai14 mice that express tdTomato following a loxP-flanked stop cassette in the Rosa26 locus under control of the CAG promoter ( Jackson Labs Stock #007914 ) . This same strategy has been used in several previous studies to facilitate identification of oxytocinergic neurons ( Clipperton-Allen et al . , 2016; Xiao et al . , 2017 ) . Throughout the course of this study , experiments were performed on approximately equal numbers of male and virgin female mice . Results of all core experiments , including immunohistochemical evaluation of OT-reporter animals , analysis of intrinsic properties of OT-MCNs , and the evaluation of the effects of acute osmotic challenge on OT-MCNs , observed both somatically and in the distal dendrites , were carefully compared across sex . Because we noted no major sex-based differences between males and virgin females in any of these experiments , data in all primary figures are combined across sex . See main text of results section , Figure 1—figure supplement 1 , and Figure 4—figure supplement 1 for further details . All animals were group-housed on a 12 hr light/dark cycle , and all animal procedures were approved by the University of Florida Institutional Animal Care and Use Committee ( IACUC ) . We used immunohistochemical techniques to quantify co-expression of tdTomato and oxytocin-neurophysin 1 ( NP1 ) . Two male and two female mice were anesthetized with sodium pentobarbital ( 1 . 56 mg/g i . p . ) and transcardially perfused with 0 . 15 M NaCl followed by 4% paraformaldehyde . Brains were extracted and post-fixed for 4 hr in 4% paraformaldehyde , then transferred to a sucrose solution ( 30% sucrose in PBS ) and stored at 4°C for at least 24 hr . Brains were then sectioned at 30 microns and stored at −20°C in cryoprotective solution ( 1 L of 0 . 1 M PBS supplemented with 20 g PVP-40 , 600 mL ethylene glycol , and 600 g sucrose ) . Unless otherwise noted , immunohistochemistry was conducted at room temperature using free-floating sections in 12-well plates ( 3 mL/well ) on an orbital shaker . Following five rinses ( 5 min each ) with 50 mM potassium PBS ( KPBS ) , tissue was incubated in blocking solution ( KPBS with 2% normal donkey serum and 0 . 2% Triton X-100 ) for 1 hr . Subsequently , sections were incubated in a primary antibody against NP1 overnight at 4°C ( mouse monoclonal , PS-38 , Dr . H . Gainer , National Institute of Health , 1:400 ) ( Ben-Barak et al . , 1985; de Kloet et al . , 2016 ) . Following five rinses ( 5 min each ) in KPBS , sections were incubated in the secondary antibody ( Alexa 647 Donkey anti-mouse , Jackson Immuno 715-605-150 , 1:500 ) for 2 hr . Following five rinses ( 5 min each ) with KPBS , slices were mounted on glass slides , air dried overnight , and coverslipped with polyvinyl alcohol mounting medium . Images were captured and analyzed using a Nikon C2+ scanning confocal microscope and NIS-Elements AR 5 . 02 software . Identical imaging parameters were used for the acquisition and analysis all images . Mice received an IP injection of ketamine ( 0 . 1 mL of 100 mg/mL , in sterile physiological saline ) and were euthanized using a small animal guillotine . Brains were extracted and coronal sections 200 µm thick were made using a Leica VT1000 S vibratome . During this procedure , slices were submerged in ice-cold , sucrose-laden , artificial cerebrospinal fluid ( ACSF ) containing ( in mM ) : 87 NaCl , 2 . 5 KCl , 1 . 25 NaH2PO4 , 7 MgCl2 , 10 dextrose , 0 . 5 CaCl2 , 75 sucrose , and 25 NaHCO3 . After sectioning was complete , brain slices containing the PVN were transferred to an incubator filled with a low-calcium high-magnesium ACSF designed to improve slice viability during the incubation period . That ACSF contained ( in mM ) : 124 NaCl , 2 . 5 KCl , 1 . 23 NaH2PO4 , 10 dextrose , 1 CaCl2 , 3 MgSO4 , and 25 NaHCO3 . Both solutions were continuously saturated with 95% O2/5% CO2 and had a pH of ~7 . 3 . After 30 min of incubation at 37°C , slices were passively equilibrated to room temperature for an additional 30 min ( minimum ) prior to use . In preparation for in vitro electrophysiological experiments , slices were transferred from the slice incubator to a low turbulence perfusion chamber ( JG-23W/HP , Warner Instruments ) where they were continuously perfused a rate of 2 mL/min with ACSF containing ( in mM ) 126 NaCl , 11 dextrose , 1 . 5 MgSO4 , 3 KCl , 1 . 2 NaH2PO4 , 2 . 4 CaCl2 , and 25 NaHCO3 . This solution was continuously oxygenated with 95% O2 and 5% CO2 , had a pH of ~7 . 3 , and was maintained at 28°C . Where noted , calcium-free ACSF was identical , except that 2 . 4 mM CaCl2 was replaced with 3 . 6 mM NaCl to maintain osmolarity . tdTomato positive PVN neurons were identified using an Olympus BX51WI stereomicroscope that supported both infrared differential interference contrast ( IR-DIC ) and conventional epifluorescence microscopy . Both IR-DIC and epifluorescence images were acquired through an Olympus 40X water immersion objective ( LUMPFL40XWI/IR-2 , Olympus ) using 12-bit IR CCD camera ( QICAM Fast 1394 ) controlled by Fiji software ( Schindelin et al . , 2012 ) . An X-Cite Series 120Q ( Lumen Dynamics ) light source coupled with an XF406 filter set ( Omega Optical ) was used for conventional epifluorescence imaging . Patch pipettes were prepared using a Flaming/Brown pipette puller ( Sutter Instruments , P-97 ) . Borosilicate glass capillaries ( 1 . 5 mm/0 . 8 mm ) were pulled to produce patch pipettes with an open tip resistance of 4–6 MΩ when filled with an intracellular solution that contained ( in mM ) : 1 MgCl2 , 1 EGTA , 10 HEPES , 125 K-gluconate , 10 phosphocreatine , 2 Na2-ATP , 0 . 25 Na-GTP , adjusted to pH 7 . 25 and 295 mOsm . This solution was used for all experiments that involved whole-cell recording but not simultaneous calcium imaging . For experiments that required concurrent calcium imaging ( see next section ) , EGTA was omitted and replaced with 0 . 3 mM Fluo-5F ( ThermoFisher F14221 ) and 0 . 03 mM Alexa 594 ( ThermoFisher A10438 ) . Where noted , K-gluconate in this solution was replaced with equimolar Cs-gluconate . Whole-cell recordings were made using a Multiclamp 700B amplifier , Digidata 1440A digitizer , and Clampex 10 . 7 software ( Molecular Devices ) . All whole-cell data were sampled at 20 kHz and low-pass filtered at 2 kHz . When necessary , synaptic responses were generated using a small tipped bipolar stimulator pulled from theta glass ( ~1–1 . 5 µm inner diameter ) , connected to a constant current stimulus isolator ( 0 . 2 msec pulse duration ) . Simulator placement and stimulation intensity were adjusted until an evoked response ≥ 50 pA could be reliably generated . On-line analysis of whole-cell patch clamp data was performed with custom software using Python 3 . 6 and the pyABF module . Off-line analysis was performed using custom software written by CJF in OriginC ( OriginLab Corporation , Northampton , MA ) . Access resistance , input resistance and whole-cell capacitance were continuously monitored in voltage clamp using data generated with brief ( 200 msec ) hyperpolarizing steps from −70 mV to −100 mV . Cells were excluded from analysis if they did not survive until experiments were completed , or if they suffered a sudden change in patch quality as indicated by large or sudden changes in either holding current or access resistance during the course of recording . All experiments were performed in the continuous presence of bath applied antagonists for kainate/AMPA receptors , NMDA receptors , GABAA receptors , and GABAB receptors ( 20 µM DNQX , 40 µM AP5 , 100 µM picrotoxin ( PTX ) , and 10 µM CGP-55845 ( CGP ) , respectively ) , except for those that explicitly involved measuring evoked excitatory synaptic currents ( EPSCs ) , or effects of an exogenous glutamate or GABA receptor agonist . Some experiments transiently delivered tetrodotoxin ( TTX , 1 µM ) , muscimol ( 400 nM ) , or ruthenium red ( RR , 10 µM ) using a syringe pump in-line with the bath perfusion system . In experiments that involved changing osmolarity of the bath , mannitol ( MT , in ACSF ) or purified deionized water were also delivered using a syringe pump . Local application of glutamate was achieved using a Parker Picospritzer III ( model R374-01C , 5–10 ms pulse duration , 10–20 psi ) , connected to a glass pipette , identical to those used for whole-cell recording , loaded with 100 mM glutamate solubilized in ACSF . All drugs used during in vitro experiments were obtained from TOCRIS except PTX and glutamate which were obtained from Sigma-Aldrich . Although stock solutions for PTX and CGP were dissolved in DMSO , total bath concentration of DMSO never exceeded ~0 . 1% and remained stable throughout experiments . Subcellular two-photon calcium imaging in OT-MCN axons and dendrites was accomplished using an Ultima laser scanner ( Bruker Scientific , Billerica , MA ) , powered by a Mira Ultrafast Ti:sapphire laser and a Verdi 5W pump ( both from Coherent , Inc Santa Clara , CA ) . The emission wavelength of the Mira was set to 810 nm to simultaneously excite the calcium sensitive indicator , Fluo-5F , and the calcium insensitive indicator , Alexa Fluor 594 ( see above for detailed composition of internal solution ) . All cells were permitted to rest for 15 min following establishment of whole-cell configuration to ensure robust diffusion of fluorophore into dendrites before proceeding with calcium imaging experiments . Laser power reaching the slice during two-photon imaging was adjusted ( using a model 350–80 pockels cell , Conoptics , Danbury , DT ) to a level where the basal ( unstimulated ) signal on the green ( Fluo-5F ) channel was visible , but this value was not precisely measured . A 40x water immersion objective ( LUMPFL40XWI/IR-2 , Olympus ) was used for all imaging experiments . Somatic activity was induced in cells voltage-clamped at −70 mV using a train of brief action potential like voltage steps ( to +50 mV for five msec , delivered at 20 Hz for 2 s ) . Activity-induced calcium influx was measured in dendrites or axons using two-photon line scans run across the structure of interest at a known distance from the soma . Line scans were acquired at 85 Hz , beginning 2 s before somatic stimulation and continuing for a total of 11 . 9 s . Where mentioned in the text , the voltage clamp current applied to the soma during the train was quantified over time by measuring the area under the curve of the last pulse . Emissions from the calcium sensitive indicator ( Fluo-5F , green ) and the calcium insensitive indicator ( Alexa Fluor 594 , red ) were separated at 575 nm with a dichroic mirror , and simultaneously measured by two separate photomultiplier tubes ( PMTs , Hamamatsu R3896 SEL ) . Raw data from the PMTs was used to calculate the fluorescence ratio of calcium-sensitive to calcium insensitive fluorescence ( F/F ) over time . This curve was then baseline subtracted using the mean F/F as observed in a 1 . 2 s baseline period that occurred immediately before the stimulus , smoothed using a 0 . 5 s Gaussian-weighted moving window function , and reported as ΔF/F . Peak ΔF/F is the peak of the ΔF/F curve as observed in response to the somatic stimulus , while Δ Peak ΔF/F is used to describe the change in peak ΔF/F resulting from a change in bath conditions , expressed as a percentage of the baseline response . Dendrites were distinguished from axons by their larger diameter and by their enhanced passive fill with somatically delivered Alexa Fluor 594 . On-line acquisition of line scan data was performed using Prairie View 5 . 4 software ( Bruker Scientific , Billerica , MA ) . Off-line analysis of line scan data was performed using custom software written in C# and OriginC by SWH and CJF . Within-cell changes in individual parameters produced by an experimental procedure were evaluated with a two-tailed 1-sample Student’s t-test ( null hypothesis mean = 1 or mean = 0 , for normalized vs . baseline subtracted data , respectively ) . When comparing a parameter of interest across two independent groups of cells , or across two different dendritic locations , a two-tailed two-sample Student’s t-test was used ( null hypothesis: Group one mean = Group two mean ) . Welch’s correction was applied in cases where population variance was significantly different between samples . To determine whether there were sex-based differences in the intrinsic properties of OT-MCNs , or in the effect of acute hyperosmotic challenge as observed in proximal vs . distal OT-MCN dendrites , a two-way ANOVA was used with either sex and cell-type , or sex and dendritic location , as factors , respectively . For evaluating whether distance from the soma impacted activity-induced calcium influx within a single-cell type and internal solution , and when > two distances were measured per cell , a one-way repeated measures ANOVA was used , with activity-induced calcium influx as the repeated measure and distance from the soma as the factor . For comparing activity-induced dendritic calcium influx observed at > two distances form the soma across groups of OT-MCNs recorded with different internal solutions , or across different compartments of individual OT-MCNs , a two-way repeated measures ANOVA was used , with activity-induced calcium influx as the repeated measure , distance from the soma as one factor , and either internal solution or cellular compartment as the other . For all ANOVAs that involved repeated measures , Mauchly’s test of sphericity was used , and Greenhouse-Geisser corrections were applied if necessary . In cases where significant interaction between factors was noted , Holm-Sidak post-hoc tests were used to test for mean differences in measured values ( e . g . of activity-induced calcium influx ) , across one factor ( e . g . internal solution ) , at specific levels of the other ( e . g . distance from the soma ) . In all cases , p-values ≤ alpha level of 0 . 05 were considered statistically significant . Error bars on all plots represent the SEM . | Oxytocin is often referred to as a ‘love hormone’ because it can be released during activities such as hugging , snuggling , or sex . Reality , of course , can be a bit more complicated . In the brain , oxytocin can have powerful and diverse effects on mood , stress , anxiety , and social interactions . In the body it helps regulate fluid balance , promotes contractions during childbirth , and stimulates the letdown of milk during breastfeeding . Much of the oxytocin produced in both humans and rodents comes from oxytocin-synthetizing magnocellular neurons located in an area of the brain called the hypothalamus . These very specialized neurons have separate , but overlapping , mechanisms for releasing oxytocin into the brain and into the rest of the body . This means that while certain signals cause the neurons to release oxytocin into the body and the brain at the same time , others can cause them to release the hormone preferentially into the body or the brain . Sheng et al . wanted to better understand how these different release mechanisms work , and , in particular , to learn more about how release of oxytocin into the brain is regulated . This is important , because when oxytocin is given as a medicine , much of it fails to reach the brain . A lot of the oxytocin that acts in the brain is released from a specific part of the oxytocin-synthesizing magnocellular neurons called the dendrites . When these neurons are stimulated , calcium enters the dendrites , triggering the release of oxytocin directly into the brain . Sheng et al . used electrical and optical tools on brain tissue extracted from mice to measure how different signals change the amount of calcium that enters the dendrites of oxytocin-synthesizing magnocellular neurons in response to a consistent stimulus . The results showed that increasing the osmolarity , the amount of water-soluble particles that cannot spontaneously cross the cell membrane , in the liquid surrounding the neurons reduced the amount of calcium that flowed into the dendrites during stimulation . Meanwhile , decreasing osmolarity had the opposite effect . Sheng et al . also found that the influx of calcium induced by stimulating the neurons can be strongly regulated by activating receptors in the dendrites that detect a common molecule in the brain called GABA . This occurs even absent a change in osmolarity . These results shed light on some of the physiological processes that control the release of oxytocin into the brain . Understanding these processes is a necessary step towards developing new drugs intended to regulate levels of oxytocin in the brain . Such drugs could be useful in the treatment of several types of mental health disorders . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2021 | Dendritic osmosensors modulate activity-induced calcium influx in oxytocinergic magnocellular neurons of the mouse PVN |
The occipital cortex of early blind individuals ( EB ) activates during speech processing , challenging the notion of a hard-wired neurobiology of language . But , at what stage of speech processing do occipital regions participate in EB ? Here we demonstrate that parieto-occipital regions in EB enhance their synchronization to acoustic fluctuations in human speech in the theta-range ( corresponding to syllabic rate ) , irrespective of speech intelligibility . Crucially , enhanced synchronization to the intelligibility of speech was selectively observed in primary visual cortex in EB , suggesting that this region is at the interface between speech perception and comprehension . Moreover , EB showed overall enhanced functional connectivity between temporal and occipital cortices that are sensitive to speech intelligibility and altered directionality when compared to the sighted group . These findings suggest that the occipital cortex of the blind adopts an architecture that allows the tracking of speech material , and therefore does not fully abstract from the reorganized sensory inputs it receives .
The human cortex comprises a number of specialized units , functionally tuned to specific types of information . How this functional architecture emerges , persists , and develops throughout a person’s life are among the most challenging and exciting questions in neuroscience research . Although there is little debate that both genetic and environmental influences affect brain development , it is currently not known how these two factors shape the functional architecture of the cortex . A key topic in this debate is the organization of the human language system . Language is commonly thought to engage a well-known network of regions around the lateral sulcus . The consistency of this functional mapping across individuals and its presence early in development are remarkable , and often used to argue that the neurobiological organization of the human language system is the result of innate constraints ( Dehaene-Lambertz et al . , 2006; Berwick et al . , 2013 ) . Does the existence of a highly consistent set of regions for language acquisition and processing imply that this network is ‘hardwired’ and immutable to experience ? Strong nativist theories for linguistic innateness leave little room for plasticity due to experience ( Bates , 1999 ) , suggesting that we should conceive ‘the growth of language as analogous to the development of a bodily organ’ ( Chomsky , 1976 , p . 11 ) . However , studies in infants born with extensive damage to cortical regions that are typically involved in language processing may develop normal language abilities , thereby demonstrating that the language network is subject to reorganization ( Bates , 2005 ) . Perhaps the most intriguing demonstrations to show that the neurobiology of language is susceptible to change due to experience come from studies showing functional selectivity to language in primary and secondary ‘visual’ areas in congenitally blind individuals ( Röder et al . , 2002; Burton , 2003; Amedi et al . , 2004; Bedny et al . , 2011; Arnaud et al . , 2013 ) . Such reorganization of the language network is particularly fascinating because it arises in the absence of injury to the core language network ( Bates , 2005; Atilgan et al . , 2017 ) . However , the level at which the occipital cortex is involved in speech representation in the early blind ( EB ) , remains poorly understood . Speech comprehension requires that the brain extracts meaning from the acoustic features of sounds ( de Heer et al . , 2017 ) . Although several neuroimaging studies have yielded valuable insights about the processing of speech in EB adults ( Arnaud et al . , 2013; Bedny et al . , 2011; Büchel , 2003; Lane et al . , 2015; Röder et al . , 2002 ) and infants ( Bedny et al . , 2015 ) , these methods do not adequately capture the fast and continuous nature of speech processing . Because speech unfolds over time , understanding spoken language relies on the ability to track the incoming acoustic signal in near real-time ( Peelle and Davis , 2012 ) . Indeed , speech is a fluctuating acoustic signal that rhythmically excites neuronal populations in the brain ( Poeppel et al . , 2008; Gross et al . , 2013; Peelle et al . , 2013 ) . Several studies have demonstrated that neuronal populations in auditory areas entrain to the acoustic fluctuations that are present in human speech around the syllabic rate ( Luo and Poeppel , 2007; Kayser et al . , 2009; Szymanski et al . , 2011; Zoefel and VanRullen , 2015 ) . It has therefore been suggested that entrainment reflects a key mechanism underlying hearing by facilitating the parsing of individual syllables through adjusting the sensory gain relative to fluctuations in the acoustic energy ( Giraud and Poeppel , 2012; Peelle and Davis , 2012; Ding and Simon , 2014 ) . Crucially , because some regions that track the specific acoustic rhythm of speech are sensitive to speech intelligibility , neural synchronization is not only driven by changes in the acoustic cue of the auditory stimuli , but also reflects cortical encoding and processing of the auditory signal ( Peelle et al . , 2013; Ding and Simon , 2014 ) . Speech tracking is therefore an invaluable tool to probe regions that interface speech perception and comprehension ( Poeppel et al . , 2008; Gross et al . , 2013; Peelle et al . , 2013 ) . Does the occipital cortex of EB people synchronize to speech rhythm ? Is this putative synchronization of neural activity to speech influenced by comprehension ? Addressing these questions would provide novel insights into the functional organization of speech processing rooted in the occipital cortex of EB people . In the current study , we investigated whether neuronal populations in blind occipital cortex synchronize to rhythmic dynamics of speech , by relating the amplitude fluctuations in speech to electromagnetic dynamics recorded from the participant’s brain . To this end , we quantified the local brain activity and directed connectivity in a group of early blind ( EB; n = 17 ) and sighted individuals ( SI; n = 16 ) using magnetoencephalography ( MEG ) while participants listened to short narrations from audiobooks . If the occipital cortex of the blind entrains to speech rhythms , this will support the idea that this region processes low-level acoustic features relevant for understanding language . We further tested whether the putative synchronization of occipital responses in EB people benefits from linguistic information or only relates to acoustic information . To separate linguistic and acoustic processes , we relied on a noise-vocoding manipulation . This method spectrally distorted the speech signal in order to impair intelligibility gradually , but systematically preserves the slow amplitude fluctuations responsible for speech rhythm ( Shannon et al . , 1995; Peelle and Davis , 2012 ) . Furthermore , to go beyond differences in the local encoding of speech rhythm in the occipital cortex , we also investigated whether the connectivity between occipital and temporal regions sensitive to speech comprehension is altered in early blindness .
Participants listened to either natural speech segments ( nat-condition ) or to altered vocoded versions of these segments ( see Materials and methods for details ) . In the 8-channel vocoded condition , the voice of the speaker is highly distorted but the intelligibility is unperturbed . By contrast , in the 1-channel vocoded condition , the speech is entirely unintelligible . After listening to each speech segment , participants were provided with a short statement about the segment , and asked to indicate whether the statement was true or false . Behavioral performance on these comprehension statements was analyzed using linear mixed-effects models with maximum likelihood estimation . This method is a linear regression that takes into account dependencies in the data , as present in repeated measures designs . Blindness and intelligibility were included as fixed effects , while subject was modeled as a random effect . Intelligibility was nested in subjects . Intelligibility had a significant effect on story comprehension ( χ ( 2 ) =110 . 7 , p<0 . 001 ) . The effect of blindness and the interaction between intelligibility and blindness were non-significant ( χ ( 1 ) =1 . 14 , p=0 . 286 and χ ( 2 ) =0 . 24 , p=0 . 889 ) . Orthogonal contrasts demonstrated that speech comprehension was stronger in the nat and 8-channel condition versus the 1-channel condition ( b = 0 . 78 , t ( 62 ) =14 . 43 , p<0 . 001 , r = 0 . 88 ) . There was no difference between the nat and the 8-channel condition ( b = 0 . 02 , t ( 62 ) =1 . 39 , p=0 . 17 ) . Thus , speech intelligibility was reduced in the 1-channel , but not the 8-channel vocoded condition ( Figure 1D ) . The lack of effect for the factor blindness suggests that there is no evidence for a potential difference in comprehension , attention or motivation between groups . Descriptive statistics of the group , and condition means are depicted in Table 1 . The overall coherence spectrum , which highlights the relationship between the amplitude envelope of speech and the signal recorded from the brain , was maximal over temporal sensors between 6 Hz and 7 Hz ( Figure 1E ) . This first analysis was performed on the combined dataset , and hence is not prone to bias or circularity for subsequent analyses , targeting group differences . The peak in the current study is slightly higher than those reported in previous studies ( Gross et al . , 2013 ) . A likely explanation for this shift is the difference in syllabic rate between English ( ~6 . 2 Hz ) , used in previous studies , and Italian ( ~7 Hz ) ( Pellegrino et al . , 2011 ) . The syllabic rate is the main carrier of amplitude fluctuations in speech , and thus is most prone to reset oscillatory activity . To capture the temporal scale of cerebro-acoustic coherence effects ( 6–7 Hz ) optimally , and to achieve a robust source estimate , source reconstruction was performed on two separate frequency windows ( 6 ± 2 , and 7 ± 2 Hz ) for each subject and condition . The two source images were averaged for subsequent analysis , yielding a single image representing the frequency range 4–9 Hz . By combining the two frequency bands , we acquire a source estimate that emphasizes the center of our frequency band of interest ( 6–7 Hz ) and tapers off towards the edges . This source estimate optimally represents the coherence spectrum observed in sensor space . The choice of the center frequency was also restricted by the length of the time window used for the analysis . That is , with a 1 s time window and a resulting 1 Hz frequency resolution , a non-integer center frequency at , for example , 6 . 5 Hz was not feasible . The source-reconstructed coherence at the frequency-range of interest confirmed that cerebro-acoustic coherence was strongest across groups and conditions in bilateral temporal lobes , including primary auditory cortex . ( Figure 1F ) . To test whether envelope tracking in the current study is modulated by intelligibility , we compared the coherence maps for the intelligible ( nat ) versus non-intelligible ( 1-channel ) condition , in all groups combined ( i . e . , EB and SI ) , with dependent-samples permutation t-tests in SPM . The resulting statistical map ( Figure 1G , p<0 . 05 , FWE-corrected ) revealed a cluster in right superior and middle temporal cortex ( STG , MTG ) , where synchronization to the envelope of speech was stronger when participants had access to the content of the story . In other words , the effect of intelligibility on the sensory response ( coherence with the speech envelope ) suggests an interface between acoustic speech processing and comprehension in STG that is present in both groups . Whether EB individuals recruit additional neural substrate for tracking the envelope of speech was tested using independent-samples permutation t-tests in SPM , contrasting coherence maps between EB and SI for all three conditions combined . The statistical maps ( Figure 1H , p<0 . 05 , FWE-corrected ) revealed enhanced coherence in EB versus SI in parieto-occipital cortex along the medial wall , centered on bilateral Precuneus , branching more extensively into the right hemisphere ( see Table 2 ) . This main effect of group highlights that neuronal populations in blind parieto-occipital cortex show enhanced synchronization to the acoustic speech signal . That is , blind participants show a stronger sensitivity to the sensory properties of the external speech stimulus at this level . Finally , to investigate whether and where envelope tracking is sensitive to top-down predictions during intelligible speech comprehension , we subtracted the unintelligible ( 1-channel ) from the intelligible ( nat ) condition , and computed independent-samples t-tests between groups . As highlighted in the introduction , we were particularly interested in the role played by the Calcarine sulcus ( V1 ) in processing semantic attributes of speech ( Burton et al . , 2002; Röder et al . , 2002; Amedi et al . , 2003 ) , and therefore we restricted the statistical analysis to the area around the primary visual cortex . The search volume was constructed from four 10 mm spheres around coordinates in bilateral Calcarine sulcus ( [−7 , –81 , −3]; [−6 , –85 , 4]; [12 , -87 , 0; 10 , –84 , 6] ) . The coordinates were extracted from a previous study on speech comprehension in EB ( Burton et al . , 2002 ) . The resulting mask also include regions highlighted in similar studies by other groups ( Röder et al . , 2002; Bedny et al . , 2011 ) . A significant effect of the interaction between blindness and intelligibility [ ( natEB – 1-channelEB ) – ( natSI – 1-channelSI ) ] was observed in right calcarine sulcus . To explore the spatial specificity of the effect , we show the whole-brain statistical map for the interaction between intelligibility and blindness at a more liberal threshold ( p<0 . 005 , uncorrected ) in Figure 1I . This shows that intelligible speech selectively engages the area around the right calcarine sulcus in the blind versus the sighted . Specifically , the region corresponding to right ‘visual’ cortex showed enhanced sensitivity to intelligible speech in EB versus SI ( p<0 . 05 , FWE-corrected ) . However , the analysis contrasting the two intelligible conditions ( nat and 8-channel ) did not yield a significant effect in the EB , suggesting that the low-level degrading of the stimulus alone does not drive the effect , but rather the intelligibility of the speech segment . Follow-up post hoc comparisons between the two groups revealed that coherence was stronger during the intelligible condition for EB versus SI ( t[30] = 3 . 09 , p=0 . 004 ) , but not during the unintelligible condition ( t[30] = –1 . 08 , p=0 . 29 ) . A group difference between EB and SI across conditions was not observed in the calcarine region ( t[30] = 1 . 1 , p=0 . 28 ) . The lack of an overall group effect in calcarine sulcus suggests that there is not a simple enhanced sensory response to speech in the blind . This was different to the response we observed in parietal cortex , in which synchronization was stronger in the blind than in the sighted . Rather , the response in calcarine sulcus only differed between the two groups when speech was intelligible . While overall coherence with the speech envelope was equally high in EB and SI , the blind population showed significantly higher coherence in the intelligible speech condition than did the sighted , who showed higher cerebro-acoustic coherence in the unintelligible condition ( 1-Chan ) . The latter is reminiscent of the fact that the more adverse the listening condition ( low signal-to-noise ratio or audiovisual incongruence ) , the more the visual cortex is entrained to the visual speech signal of actual acoustic speech when presented together with varying levels of acoustic noise ( Park et al . , 2016; Giordano et al . , 2017 ) . Moreover , Giordano et al . ( 2017 ) showed an increase of directed connectivity between superior frontal regions and visual cortex under the most challenging ( acoustic noise and uninformative visual cues ) conditions , again suggesting a link between the reorganization observed in the occipital cortex of blind individuals and typical multisensory pathways involving the occipital cortex in audio-visual speech processing ( Kayser et al . , 2008 ) . Visual deprivation since birth , however , triggers a functional reorganization of the calcarine region that can then dynamically interact with the intelligibility of the speech signal . For illustration purposes , cerebro-acoustic coherence from functional peak locations for intelligibility ( across groups ) in STG , blindness in the parieto-occipital cortex ( Figure 1G; POC ) , and the interaction are represented as boxplots in Figure 1J . To further investigate whether cerebro-acoustic peaks in CS and STG which are sensitive to intelligible speech in EB are indicative of a more general re-organization of the network , we conducted functional connectivity analysis . Statistical analysis of the connectivity estimates was performed on the mean phase-locking value in the theta ( 4–8 Hz ) range . Using linear mixed-effects models , blindness ( EB , SI ) , intelligibility ( nat , 1-channel ) , and the interaction between blindness and intelligibility were added to the model in a hierarchical fashion . Blindness and intelligibility were modeled as fixed effects , while subject was a random effect . Intelligibility was nested in subject . A main effect was observed only for blindness ( χ[1] = 4 . 32 , p=0 . 038 ) and was caused by greater connectivity for EB versus SI ( Figure 2A–B ) . The main effects of intelligibility and of the interaction between intelligibility and blindness were non-significant ( p=0 . 241 and p=0 . 716 , respectively ) . Subsequently , linear mixed-effects models were applied to test for the effects of blindness and intelligibility on the directional information flow ( PSI ) between CS and STG . The fixed and random effects structure was the same as that described in the previous analysis . Here , only the main effect of blindness was significant ( χ[1] = 4 . 54 , p=0 . 033 ) , indicating that the directional information flow differs between groups . The effect of intelligibility and the interaction between intelligibility and blindness were both non-significant ( p=0 . 51 and p=0 . 377 , respectively ) . Follow-up post-hoc one-sample t-tests on the phase slope estimates for each group individually revealed a significant direction bias from CS to STG for SI ( t[14] = –2 . 22 , p=0 . 044 ) . No directional bias was found for EB ( t[16] = 0 . 74 , p=0 . 47 ) . As depicted in Figure 2C–D , these results suggest that CS projects predominantly to STG in SI , whereas in EB , this interaction is more balanced and trending in the opposite direction .
Although language is not the only cognitive process that selectively activates the occipital cortex of EB people , it is arguably one of the most puzzling . The reason is that reorganization of other processes , such as auditory motion perception and tactile object recognition , appears to follow the topography of the functionally equivalent visual processes in the sighted brain ( Ricciardi et al . , 2007a; Amedi et al . , 2010; Dormal et al . , 2016 ) . For example , the hMT+/V5 complex , which is typically involved in visual motion in the sighted , selectively processes auditory ( Poirier et al . , 2006; Dormal et al . , 2016; Jiang et al . , 2016 ) or tactile ( Ricciardi et al . , 2007b ) motion in blind people . However , in the case of language , such recruitment is striking in light of the cognitive and evolutionary differences between vision and language ( Bedny et al . , 2011 ) . This led to the proposal that , at birth , human cortical areas are cognitively pluripotent: capable of assuming a broad range of unrelated cognitive functions ( Bedny , 2017 ) . However , this argument resides on the presupposition that language has no computational relation with vision . But does this proposition tally with what we know about the relationship between the visual system and the classical language network ? Rhythmic information in speech has a well-known language-related surrogate in the visual domain: lip movements ( Lewkowicz and Hansen-Tift , 2012; Park et al . , 2016 ) . Indeed , both acoustic and visual speech signals exhibit rhythmic temporal patterns at prosodic and syllabic rates ( Chandrasekaran et al . , 2009; Schwartz and Savariaux , 2014; Giordano et al . , 2017 ) . The perception of lip kinematics that are naturally linked to amplitude fluctuations in speech serves as an important vehicle for the everyday use of language ( Kuhl and Meltzoff , 1982; Weikum et al . , 2007; Lewkowicz and Hansen-Tift , 2012 ) and helps language understanding , particularly in noisy conditions ( Ross et al . , 2007 ) . Indeed , reading lips in the absence of any sound activates both primary and association auditory regions overlapping with regions that are active during the actual perception of spoken words ( Calvert et al . , 1997 ) . The synchronicity between auditory and visual speech entrains rhythmic activity in the observer’s primary auditory and visual regions , and facilitates perception by aligning neural excitability with acoustic or visual speech features ( Schroeder et al . , 2008; Schroeder and Lakatos , 2009; Giraud and Poeppel , 2012; Mesgarani and Chang , 2012; Peelle and Davis , 2012; van Wassenhove , 2013; Zion Golumbic et al . , 2013b; Park et al . , 2016; Giordano et al . , 2017 ) . These results strongly suggest that both the auditory and the visual components of speech are processed together at the earliest level possible in neural circuitry , based on the shared slow temporal modulations ( around 2–7 Hz range ) present across modalities ( Chandrasekaran et al . , 2009 ) . Corroborating this idea , it has been demonstrated that neuronal populations in visual cortex follow the temporal dynamics of lip movements in sighted individuals , similar to the way in which temporal regions follow the acoustic and visual fluctuations of speech ( Luo et al . , 2010; Zion Golumbic et al . , 2013a; Park et al . , 2016 ) . Similar to temporal cortex , occipital cortex in the sighted also shows enhanced lip tracking when attention is directed to speech content . This result highlights the fact that a basic oscillatory architecture for tracking the dynamic aspects of ( visual- ) speech in occipital cortex exists even in sighted individuals . Importantly , audiovisual integration of the temporal dynamics of speech has been suggested to play a key role when learning speech early in life: young infants detect , match , and integrate the auditory and visual temporal coherence of speech ( Kuhl and Meltzoff , 1982; Rosenblum et al . , 1997; Lewkowicz , 2000 , 2010; Brookes et al . , 2001; Patterson and Werker , 2003; Lewkowicz and Ghazanfar , 2006; Kushnerenko et al . , 2008; Bristow et al . , 2009; Pons et al . , 2009; Vouloumanos et al . , 2009; Lewkowicz et al . , 2010; Nath et al . , 2011 ) . For instance , young infants between 4 and 6 months of age can detect their native language from lip movements only ( Weikum et al . , 2007 ) . Around the same period , children detect synchrony between lip movements and speech sounds , and distribute more attention towards the mouth than towards the eyes ( Lewkowicz and Hansen-Tift , 2012 ) . Linking what they hear to the lip movements may provide young infants with a stepping-stone towards language production ( Lewkowicz and Hansen-Tift , 2012 ) . Moreover , infants aged 10 weeks already exhibit a McGurk effect , again highlighting the early multisensory nature of speech perception ( Rosenblum et al . , 1997 ) . Taken together , these results suggest that an audio-visual link between observing lip movements and hearing speech sounds is present at very early developmental stages , potentially from birth , which helps infants acquire their first language . In line with these prior studies , the current results may support the biased connectivity hypothesis of cross-modal reorganization ( Reich et al . , 2011; Hannagan et al . , 2015; Striem-Amit et al . , 2015 ) . Indeed , it has been argued that reorganization in blind occipital cortex may be constrained by functional pathways to other sensory and cognitive systems that are also present in sighted individuals ( Elman et al . , 1996; Hannagan et al . , 2015 ) . This hypothesis may explain the overlap in functional specialization between blind and sighted individuals ( Collignon et al . , 2011; Dormal et al . , 2016; He et al . , 2013; Jiang et al . , 2016; Peelen et al . , 2013; Pietrini et al . , 2004; Weeks et al . , 2000; Poirier et al . , 2006 ) . In our experiment , sensitivity to acoustic dynamics of intelligible speech in blind occipital cortex could arise from pre-existing occipito-temporal pathways connecting the auditory and visual system that are particularly important for the early developmental stages of language acquisition . In fact , the reorganization of this potentially predisposed pathway to process language content would explain how language-selective response may appear in blind children as young as 3 years old ( Bedny et al . , 2015 ) . Previous studies have suggested that language processing in the occipital cortex arises through top-down projections from frontal regions typically associated with the classical language network ( Bedny et al . , 2011; Deen et al . , 2015 ) , and that the representational content is symbolic and abstract rather than sensory ( Bedny , 2017 ) . Our results contrast with this view by showing that neuronal populations in ( peri- ) calcarine cortex align to the temporal dynamics of intelligible speech , and are functionally connected to areas sensitive to auditory information in temporal cortex . In sighted individuals , regions of the temporal lobe including STS are sensitive to acoustic features of speech , whereas higher-level regions such as anterior temporal cortex and left inferior frontal gyrus are relatively insensitive to these features and therefore do not entrain to the syllabic rate of speech ( Davis and Johnsrude , 2003; Hickok and Poeppel , 2007; see confirmation in Figure 1F–G ) . This suggests that occipital areas respond , at least partially , to speech at a much lower ( sensory ) level than previously thought in EB brains , which may be caused by the reorganization of existing multisensory pathways connecting the ‘auditory’ and ‘visual’ centers in the brain . Functional dependencies between sensory systems exist between the earliest stages of sensory processing in both human ( Ghazanfar and Schroeder , 2006; Kayser et al . , 2008; Schroeder and Lakatos , 2009; Murray et al . , 2016 ) and nonhuman primates ( Falchier et al . , 2002; Lakatos et al . , 2007; Schroeder and Lakatos , 2009 ) . Several neuroimaging studies have demonstrated enhanced functional connectivity in sighted individuals between auditory and visual cortices under multisensory conditions ( see Murray et al . , 2016 for a recent review ) , including multisensory speech ( Giordano et al . , 2017 ) . Moreover , neuroimaging studies have shown that hearing people consistently activate left temporal regions during silent speech-reading ( Calvert et al . , 1997; MacSweeney et al . , 2000 , 2001 ) . We therefore postulate that brain networks that are typically dedicated to the integration of audio-visual speech signal , might be reorganized in the absence of visual inputs and might lead to an extension of speech tracking in the occipital cortex ( Collignon et al . , 2009 ) . Although an experience-dependent mechanism related to EB affects the strength and directionality of the connectivity between the occipital and temporal regions , the presence of intrinsic connectivity between these regions — which can also be observed in sighted individuals ( e . g . for multisensory integration ) — may constrain the expression of the plasticity observed in our task . Building on this connectivity bias , the occipital pole may extend its sensitivity to the intelligibility of speech , a computation this region is obviously not originally dedicated to . A number of recent studies have suggested that visual deprivation reinforces the functional connections between the occipital cortex and auditory regions typically classified as the language network ( Hasson et al . , 2016; Schepers et al . , 2012 ) . Previous studies using dynamic causal modelling support the idea that auditory information reaches the reorganized occipital cortex of the blind through direct temporo-occipital connection , rather than using subcortical ( Klinge et al . , 2010 ) or top-down pathways ( Collignon et al . , 2013 ) . In support of these studies , we observed that the overall magnitude of functional connectivity between occipital and temporal cortex is higher in blind people than in sighted people during natural speech comprehension . Moreover , directional connectivity analysis revealed that the interaction between the two cortices is also qualitatively different: sighted individuals show a strong feed-forward drive towards temporal cortex , whereas blind individuals show a more balanced information flow , and a trend in the reverse direction . These results highlight one possible pathway by which the speech signal is enhanced in the occipital cortex of EB individuals . However , this does not mean that the changes in connectivity between blind and sighted individuals are limited to this specific network ( e . g . see Kayser et al . , 2015; Park et al . , 2015 ) . We observed that neuronal populations in right superior temporal cortex synchronize to the temporal dynamics of intelligible , but not non-intelligible , speech in both EB and SI groups . Why does speech intelligibility modulate temporal regions of the right , but not the left , hemisphere ? According to an influential model of speech comprehension – the asymmetric sampling in time model ( AST; Giraud and Poeppel , 2012; Hickok and Poeppel , 2007; Poeppel , 2003 ) – there is a division of labour between the left- and right auditory cortices ( Poeppel , 2003; Boemio et al . , 2005; Hickok and Poeppel , 2007 ) , with the left auditory cortex being more sensitive to high-frequency information ( +20 Hz ) , whereas the right temporal cortex is more sensitive to low-frequency information ( ~6 Hz ) such as syllable sampling and prosody ( Belin et al . , 1998; Poeppel , 2003; Boemio et al . , 2005; Obleser et al . , 2008; Giraud and Poeppel , 2012 ) . Several studies have shown that the right hemisphere is specifically involved in the representation of connected speech ( Bourguignon et al . , 2013; Fonteneau et al . , 2015; Horowitz-Kraus et al . , 2015; Alexandrou et al . , 2017 ) , whereas other studies have directly demonstrated the prevalence of speech-to-brain entrainment while listening to sentences or stories in delta and theta bands in the right hemisphere more than in the left hemisphere ( Luo and Poeppel , 2007; Abrams et al . , 2008; Gross et al . , 2013; Giordano et al . , 2017 ) . The present study therefore replicates these results by showing enhanced phase coupling between the right hemisphere and the speech envelope at the syllabic rate ( low-frequency phase of speech envelope ) , consistent with the AST model . An interesting observation in the current study is that right hemispheric sensitivity to intelligible speech in temporal areas coincides with the enhanced right hemispheric sensitivity to intelligible speech in the occipital cortex of blind individuals . Having more cortical tissue devoted to sentence processing and understanding could potentially support enhanced sentence comprehension . Previous studies have indeed demonstrated that , as compared to sighted individuals , blind people have enhanced speech discrimination in noisy environments ( Niemeyer and Starlinger , 1981 ) , as well as the capability to understand speech displayed at a much faster rate ( sighted listeners at rates of 9–14 syllables/s andblind listeners at rates of 17–22 syllables/s; Moos and Trouvain , 2007 ) . Crucially , listening to intelligible ultra-fast speech ( as compared to reverse speech ) has been shown to cause enhanced activity in the right primary visual cortex in early and late blind individuals when compared to sighted controls , and activity in this region correlates with individual ultra-fast speech perception skills [Dietrich et al . , 2013] ) . These results raise the interesting possibility that the engagement of right V1 in the analysis of the speech envelope , as demonstrated in our study , may support the enhanced encoding of early supra-segmental aspects of the speech signal , supporting the ability to understand an ultra-fast speech signal . However , our behavioral task ( speech comprehension ) did not allow us to directly assess this link between the reorganized occipital cortex and speech comprehension , as by design performance was almost at ceiling in the nat and 8-chan condition but at chance in the 1-chan condition ( see Table 1 ) . However , it is important to note that linking brain activity and behavior is not a trivial issue . Many studies have not found a direct link between crossmodal reorganization and non-visual processing . More generally , as a behavioral outcome is the end product of a complex interactive process between several brain regions , linking the role of one region in isolation ( e . g . the reorganized region in the occipital cortex of blind people ) to behavior ( e . g . performance ) in a multifaced task is not straightforward . An absence of a direct relation between behavior ( e . g . speech processing ) and crossmodal plasticity could be explained by the fact that this complex process is supported by additional networks other than the reorganized ones . Interestingly , recent studies have shown that early visual deprivation triggers a game of ‘balance’ between the brain systems typically dedicated to a specific process and the reorganized occipital cortex ( Dormal et al . , 2016 ) . It has indeed been proposed that early visual deprivation triggers a redeployment mechanism that would reallocate part of the processing typically implemented in the preserved networks ( i . e . the temporal or frontal cortices for speech processing ) to the occipital cortex deprived of its most salient input ( vision ) . Two recent studies using multivoxel pattern analysis ( MVPA ) showed that the ability to decode the different auditory motion stimuli was enhanced in hMT+ ( a region typically involved in visual motion in sighted individuals ) of early blind individuals , whereas an enhanced decoding accuracy was observed in the planum temporale in the sighted group ( Jiang et al . , 2016; Dormal et al . , 2016 ) . Moreover , Bedny et al . ( 2015 ) reported an enhanced activation of occipital cortex and a simultaneous deactivation of prefrontal regions during a linguistic task in blind children . The authors suggested that the increased involvement of the occipital cortex might decrease the pressure on the prefrontal areas to specialize in language processing . Interestingly , a transcranial magnetic stimulation ( TMS ) study reported a reduced disruptive effect of TMS applied over the left inferior prefrontal cortex during linguistic tasks for both blind and sighted individuals , whereas TMS applied over the occipital cortex caused more disruption in early blind as compared to sighted people ( Amedi et al . , 2004 ) . Such studies support the idea that brain regions that are typically recruited for specific tasks in sighted individuals may become less essential in EB people if they concomitantly recruit occipital regions for the same task . An important open question for future research therefore concerns the relative behavioral contribution of occipital and perisylvian cortex to speech understanding . We demonstrate that the primary ‘visual’ cortex synchronizes to the temporal dynamics of intelligible speech at the rate of syllable transitions in language ( ~6–7 Hz ) . Our results demonstrate that this neural population is involved in processing the sensory signal of speech , and therefore contrasts with the proposition that occipital involvement in speech processing is abstracted from its sensory input and purely reflects higher-level operations similar to those observed in prefrontal regions ( Bedny , 2017 ) . Blindness , due to the absence of organizing visual input , leaves the door open for sensory and functional colonization of occipital regions . This colonization might however not be stochastic , but could be constrained by modes of information processing that are natively anchored in specific brain regions and networks . Even if the exact processing mode is still to be unveiled by future research , we hypothesise that the mapping of language onto occipital cortex builds upon pre-existing oscillatory architecture typically linking auditory and visual speech rhythm ( Chandrasekaran et al . , 2009 ) . Our study therefore supports the view that the development of functional specialisation in the human cortex is the product of a dynamic interplay between genetic and environmental factors during development , rather than being predetermined at birth ( Elman et al . , 1996 ) .
Seventeen early blind ( 11 female; mean ± SD , 32 . 9 ± 10 . 19 years; range , 20–67 years ) and sixteen sighted individuals ( 10 female; mean ± SD , 32 . 2 ± 9 . 92 years; range , 20–53 years ) participated in the current study . There was no age difference between the blind and the sighted group ( t[30] = 0 . 19 , p=0 . 85 ) . All blind participants were either totally blind or severely visually impaired from birth; however , two participants reported residual visual perception before the age of 3 and one before the age of 4 , as well as one participant who lost their sight completely at age 10 . Causes of vision loss were damage or detached retina ( 10 ) , damage to the optic nerve ( 3 ) , infection of the eyes ( 1 ) , microphtalmia ( 2 ) , and hypoxia ( 1 ) . Although some participants reported residual light perception , none were able to use vision functionally . All participants were proficient braille readers and native speakers of Italian . None of them suffered from a known neurological or peripheral auditory disease . The data from one sighted individual were not included because of the discovery of a brain structural abnormality that was unknown to the experimenters at the time of the experiment . The project was approved by the local ethical committee at the University of Trento . In agreement with the Declaration of Helsinki , all participants provided written informed consent before participating in the study . Auditory stimuli were delivered into the magnetically shielded MEG room via stereo loudspeakers using a Panphonics Sound Shower two amplifier at a comfortable sound level , which was the same for all participants . Stimulus presentation was controlled via the Matlab ( RRID:SCR_001622 ) Psychophysics Toolbox 3 ( http://psychtoolbox . org; RRID:SCR_002881 ) running on a Dell Alienware Aurora PC under Windows 7 ( 64 bit ) . Both sighted and blind participants were blindfolded during the experiment , and the room was dimly lit to allow for visual monitoring of the participant via a video feed from the MEG room . Instructions were provided throughout the experiment using previous recordings from one of the experimenters ( FB ) . The stimuli consisted of 14 short segments ( ~1 min ) from popular audiobooks ( e . g . , Pippi Longstocking and Candide ) in Italian ( nat ) . Furthermore , channel-vocoding in the Praat software was used to produce two additional control conditions . First , the original sound file was band-pass filtered into 1 ( 1-channel ) or 8 ( 8-channel ) logarithmically spaced frequency bands . The envelope for each of these bands was computed , filled with Gaussian white noise , and the different signals were recombined into a single sound file . The resulting signal has an amplitude envelope close to the original , while the fine spectral detail was gradually distorted . Perceptually , the voice of the speaker is highly distorted in the 8-channel condition , but intelligibility is unperturbed . By contrast , in the 1-channel condition , speech is entirely unintelligible . In total , 42 sound files were presented in a pseudo-randomized fashion , distributed among seven blocks . Each block contained two sound files from each condition , and the same story was never used twice in the same block . Examples of the stimuli are provided in supplementary media content ( see Video 1 ) . To verify story comprehension , each speech segment was followed by a single-sentence statement about the story . Participants were instructed to listen to the story carefully , and to judge whether the statement at the end was true or false , using a nonmagnetic button box . Responses were provided with the index and middle finger of the right hand . MEG was recorded continuously from a 306 triple sensor ( 204 planar gradiometers; 102 magnetometers ) whole-head system ( Elekta Neuromag , Helsinki , Finland ) using a sampling rate of 1 kHz and an online band-bass filter between 0 . 1 and 300 Hz . The headshape of each individual participant was measured using a Polhemus FASTRAK 3D digitizer . Head position of the subject was recorded continuously using five localization coils ( forehead , mastoids ) . Data pre-processing was performed using the open-source Matlab toolbox Fieldtrip ( www . fieldtriptoolbox . org; RRID:SCR_004849 ) . First the continuous data were filtered ( high-pass Butterworth filter at 1 Hz , low-pass Butterworth filter at 170 Hz , and DFT filter at 50 , 100 , and 150 Hz to remove line-noise artefacts in the signal ) , and downsampled to 256 Hz . Next , the data were epoched into segments of 1 s for subsequent analysis . Rejection of trials containing artefacts and bad channels was performed using a semi-automatic procedure . First , outliers were rejected using a pre-screening based on the variance and range in each trial/channel . Then , algorithmically guided visual inspection of the raw data was performed to remove any remaining sources of noise . Amplitude envelopes of the stories were computed using the Chimera toolbox ( Chandrasekaran et al . , 2009 ) and custom code , following the procedure described by Gross and colleagues ( Gross et al . , 2013 ) . First , the sound files were band-pass filtered between 100 and 1000 Hz into nine frequency-bands , using a fourth order Butterworth filter . The filter was applied in forward and backward direction to avoid any phase shifts with respect to the original signal , and frequency-bands were spaced with equal width along the human basilar membrane . Subsequently , the analytic amplitude for each filtered segment was computed as the absolute of the Hilbert transform . Finally , the amplitude envelopes of all nine bands were summed and scaled to a maximum value of 1 . The resulting envelope was combined with the MEG data , and processed identically henceforth . To determine where , and at what temporal scale , neuronal populations follow the temporal dynamics of speech , we computed spectral coherence between the speech envelope and the MEG signal . Coherence is a statistic that quantifies the phase relationship between two signals , and can be used to relate oscillatory activity in the brain with a peripheral measure such as a speech signal ( Peelle et al . , 2013 ) . The first analysis was conducted in sensor space , across conditions and participants , to determine the temporal scale at which coherence between the speech envelope and the MEG signal is strongest . To this end , a Hanning taper was applied to the 1 s data segments , and Fourier transformation was used to compute the cross-spectral density between 1 and 30 Hz , with a step size of 1 Hz . To render the coherence values more normally distributed , a Fisher z-transform was applied by computing the inverse hyperbolic tangent ( atanh ) . Source-space analysis was centered on the coherence frequency-band of interest ( FOI ) identified in the sensor space analysis . Source reconstruction was performed using a frequency domain beamformer called Dynamic Imaging of Coherent Sources ( DICS ) ( Gross et al . , 2001; Liljeström et al . , 2005 ) . DICS was used to localize spectral coherence , observed at the sensors , on a three-dimensional grid ( 8 × 8 × 8 mm ) . The forward model was based on a realistic single-shell headmodel ( Nolte , 2003 ) for each individual . As structural scans could not be acquired for all participants , we approximated individual anatomy by warping a structural MNI template brain ( MNI , Montreal , Quebec , Canada; www . bic . mni . mcgill . ca/brainweb ) into individual headspace using the information from each participant’s headshape . Functional connectivity between occipital and temporal cortex was computed by extracting virtual sensor time-series at the locations of interest in CS and STG using time-domain beamforming . These virtual sensor time series were used to compute non-directional and directional connectivity metrics . The rationale behind focusing our analyses on STG and CS is to limit our hypothesis space by the results as they unfolded in our analytic steps . Indeed , we found that the right STG was the only region modulated by the intelligibility of speech in both groups , confirming previous results using similar methods ( Luo and Poeppel , 2007; Abrams et al . , 2008; Gross et al . , 2013; Giordano et al . , 2017 ) . In addition to STG , we found enhanced cerebro-acoustic coherence for intelligible speech in CS of EB individuals . We therefore decided to focus our connectivity analyses on these two ROIs since they were functionally defined from our first analytical step ( modulation of cerebro-acoustic coherence by speech intelligibility ) . The advantage of following this procedure is that connectivity analyses are based on nodes that we can functionally interpret and for which we have clear predictions . We successfully used the same hierarchical analytic structure in previous studies ( Collignon et al . , 2013 , 2015; Benetti et al . , 2017 ) , following guidelines on how to investigate directional connectivity in brain networks ( Stephan et al . , 2010 ) . Virtual sensor time-series at the two locations of interest were extracted using a time-domain vector-beamforming technique called linear constrained minimum variance ( LCMV ) beamforming ( Van Veen et al . , 1997 ) . First , average covariance matrices were computed for each participant to estimate common spatial filter coefficients . These filter coefficients were multiplied with the single-trial cleaned MEG data . To reduce the resulting three-dimensional time-series to one singular value , decomposition ( SVD ) was applied , resulting in a single time-series for each trial and participant . Functional connectivity between the two regions was computed at peak locations using the phase-locking value ( PLV ) ( Lachaux et al . , 1999 ) . First , single-trial virtual sensor time courses were converted to the frequency domain ( 0–50 Hz ) using Fourier transformation . The data were padded to 2 s and a Hanning taper was applied to reduce spectral leakage . Coherence was used as a proxy for functional connectivity . To disentangle phase consistency between regions from joint fluctuations in power , the spectra were normalized with respect to the amplitude , resulting in an estimate of the phase-locking ( Lachaux et al . , 1999 ) between regions . Connectivity estimates were normalized using a Fischer z-transform ( atanh ) , as in the analysis of cerebro-acoustic coherence . PLV is a symmetric proxy for connectivity and does not allow for inferences regarding the direction of information flow between two regions of interest . To test whether differences in functional connectivity between EB and SI are accompanied by changes in the directionality of the flow of information between the regions of interest , we computed the phase slope index ( PSI ) ( Nolte et al . , 2008 ) . The PSI deduces net directionality from the time delay between two time series ( x1 and x2 ) . Time series x1 is said to precede , and hence drive , time series x2 in a given frequency band if the phase difference between x1 and x2 increases with higher frequencies . Consequently , a negative phase slope reflects a net information flux in the reverse direction , that is , from x2 to x1 . Here , we computed the PSI using a bandwidth of ±5 Hz around the frequencies of interest . Following the recommendation by Nolte and colleagues ( Nolte et al . , 2008 ) , PSI estimates were normalized with the standard error , which was computed using the jackknife method . Statistical testing of the behavioral comprehension scores , as well as the connectivity estimates , was performed using linear mixed-effects models in R . Differences between conditions and groups in source space were evaluated using Statistical Parametric Mapping ( SPM12; RRID:SCR_007037 ) , and the Threshold-Free Cluster Enhancement ( TFCE ) toolboxes in Matlab . TFCE ( Smith and Nichols , 2009 ) computes new values for each voxel in a statistical map as a function of the original voxel value and the values of the surrounding voxels . By enhancing the statistical values in voxels with a high T-value that are also part of a local cluster , TFCE optimally combines the benefits of voxel-wise and cluster-based methods . TFCE was applied to the whole-brain contrasts . Final correction for multiple comparisons was applied using a maximum statistic based on 1000 permutations of group membership ( independent testing ) or conditions ( dependent testing ) . Here , we applied a variance smoothing of 15 mm FWHM to reduce the effects of high spatial frequency noise in the statistical maps . The smoothing kernel used in the current study is higher than that in comparable fMRI studies because of the inherent smoothness of the source-reconstructed MEG data . | Scientists once thought that certain parts of the brain were hard-wired to process information from specific senses or to perform specific tasks . For example , some had concluded that language processing is built into certain parts of the brain , because the way the brain responds to language is remarkably similar in different people even from very early on in life . Yet other studies with individuals who were born blind emphasize that experience also shapes the way the brain works . In people who are born blind , parts of the brain that typically interpret visual information in sighted people are often put to other uses . Now , van Ackeren et al . show that people who became blind early in life are able to repurpose parts of the brain that are more typically used for vision to understand spoken language instead . A technique called magnetoencephalography was used to map how different parts of the brain respond when both people with sight and those who are blind listen to recordings of someone talking . In some of the experiments , the speech was distorted , making it unintelligible . In both groups , areas of the brain known to process sound information showed patterns of activity that match the rhythms present in the speech . The group with blindness also showed similar activity in parts of the brain usually used to process visual information , and even more so when they were exposed to intelligible speech . The experiments show that brain efficiently reshapes to adapt to a world with no visual input . It may do this by making use of connections that already exist between the auditory and visual brain centers . For instance , very young children use these connections to link what they hear to the lip movements of adults . Future studies are needed to determine if individuals whose ability to see is restored would be able to process the visual information or if the adaptation of the visual processing parts of the brain to help understand speech would interfere with their sight . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Neuronal populations in the occipital cortex of the blind synchronize to the temporal dynamics of speech |
Increased mTORC1 signaling from TSC1/TSC2 inactivation is found in cancer and causes tuberous sclerosis complex ( TSC ) . The role of mesenchymal-derived cells in TSC tumorigenesis was investigated through disruption of Tsc2 in craniofacial and limb bud mesenchymal progenitors . Tsc2cKOPrrx1-cre mice had shortened lifespans and extensive hamartomas containing abnormal tortuous , dilated vessels prominent in the forelimbs . Abnormalities were blocked by the mTORC1 inhibitor sirolimus . A Tsc2/mTORC1 expression signature identified in Tsc2-deficient fibroblasts was also increased in bladder cancers with TSC1/TSC2 mutations in the TCGA database . Signature component Lgals3 encoding galectin-3 was increased in Tsc2-deficient cells and serum of Tsc2cKOPrrx1-cre mice . Galectin-3 was increased in TSC-related skin tumors , angiomyolipomas , and lymphangioleiomyomatosis with serum levels in patients with lymphangioleiomyomatosis correlating with impaired lung function and angiomyolipoma presence . Our results demonstrate Tsc2-deficient mesenchymal progenitors cause aberrant morphogenic signals , and identify an expression signature including Lgals3 relevant for human disease of TSC1/TSC2 inactivation and mTORC1 hyperactivity .
Mechanistic target of rapamycin complex 1 ( mTORC1 ) is a central regulator of cell growth and metabolism ( Laplante and Sabatini , 2012 ) . Activation of mTORC1 , caused by dysregulated upstream signaling through phosphoinositide 3-kinase ( PI3K ) , PTEN , AKT , and TSC1-TSC2 , is observed in cancers , hamartoma syndromes such as tuberous sclerosis complex ( TSC ) , and vascular anomalies ( Dibble and Cantley , 2015; Krymskaya and Goncharova , 2009; Nathan et al . , 2017 ) . The supposition that many of the pathological abnormalities in these conditions arise from increased signaling through mTORC1 is supported by response to treatment using mTOR inhibitors such as sirolimus and everolimus , particularly for tumors in TSC ( Bissler et al . , 2008; Krueger et al . , 2010; McCormack et al . , 2011; Taveira-DaSilva et al . , 2011 ) . However , clinical responses may be inadequate and require lifelong treatment ( Taveira-DaSilva and Moss , 2015 ) . There is a need for greater understanding of potential downstream effectors of mTORC1 that may represent new targets for treatment and/or markers of disease severity . TSC is a familial tumor syndrome characterized by highly vascular , hamartomatous tumors in multiple organs including the skin ( eg . facial angiofibromas ) , kidneys ( angiomyolipomas ) , and lungs ( lymphangioleiomyomas ) . Facial angiofibromas can be disfiguring and their highly vascular nature makes them prone to bleeding with minimal trauma ( Darling et al . , 2010 ) . Angiomyolipomas ( AMLs ) are a leading cause of death in patients with TSC due to hemorrhage and renal failure ( Franz et al . , 2010; Byard et al . , 2003 ) . They have large , tortuous , and thick-walled vessels that may lack elastin , making them prone to aneurysms and life-threatening hemorrhage ( Byard et al . , 2003; Bissler and Kingswood , 2004; Tweeddale et al . , 1955 ) . Lymphangioleiomyomatosis ( LAM ) involves a proliferation of abnormal smooth-muscle like cells that invade the axial lymphatics and lung to cause lymphangioleiomyomas and cystic lung disease , respectively ( Taveira-DaSilva and Moss , 2015 ) . LAM occurring in the absence of TSC , called sporadic LAM ( S-LAM ) , is also associated with the development of AMLs ( Taveira-DaSilva and Moss , 2015 ) . The lymphangiogenic factor VEGF-D is elevated in LAM , correlates with disease severity and response to treatment , and is associated with lymphatic involvement ( McCormack et al . , 2011; Glasgow et al . , 2009; Young et al . , 2013; Seyama et al . , 2006; Budde et al . , 2016; Malinowska et al . , 2013 ) . VEGF-D is also increased in a mouse model of LAM ( Goncharova et al . , 2012 ) . TSC and S-LAM are caused by inactivating mutations in either TSC1 or TSC2 ( Cheadle et al . , 2000 ) , genes that are also mutated in some cancers , particularly bladder carcinoma ( Sjödahl et al . , 2011; Pymar et al . , 2008; Guo et al . , 2013 ) . Proteins encoded by the TSC1 and TSC2 genes , TSC1 ( also known as hamartin ) and TSC2 ( aka tuberin ) , suppress mTORC1 signaling by forming a ternary complex with TBC1D7 that suppresses RHEB-mediated activation of signaling through mTORC1 by converting RHEB-GTP to RHEB-GDP ( Dibble and Cantley , 2015; Dibble et al . , 2012 ) . Loss of function of either TSC1 or TSC2 inhibits RHEB inactivation , leading to hyperactive mTORC1 signaling ( Dibble et al . , 2012; Huang and Manning , 2008 ) . mTORC1 incorporates signals from growth factor signaling , especially through the PI3K-AKT pathway , and acts as a sensor of cellular stress , levels of amino acids , energy , and oxygen to mediate its downstream effects ( Dibble and Cantley , 2015 ) . TSC1 or TSC2 loss of function and subsequent mTORC1 activation , which drives tumor formation and vascular changes , have been investigated using rodent models , exploiting a spontaneous mutation in Tsc2 in the Eker rat , or using targeted disruption of Tsc1 or Tsc2 in mice . In the Eker rat , renal tumors develop with 100% penetrance and these rats additionally develop pituitary adenomas , uterine leiomyomas , and splenic tumors ( Eker , 1954; Yeung et al . , 1994 ) . In mice , homozygous disruption of Tsc1 or Tsc2 is lethal during embryogenesis , and heterozygous Tsc1+/− and Tsc2+/− mice develop renal cystadenomas , liver hemangiomas , and infrequently , paw angiosarcomas ( Kobayashi et al . , 1999; Onda et al . , 1999; Kobayashi et al . , 2001; Kwiatkowski et al . , 2002 ) . Several models of Tsc1 deficiency have shown its role in the development of vascular abnormalities . Conditional disruption of Tsc1 in vascular smooth muscle cells resulted in mice with vascular smooth muscle hyperplasia and hypertension ( Malhowski et al . , 2011; Houssaini et al . , 2016 ) . Deletion of Tsc1 expression specifically in endothelial cells using Tie2-cre led to embryonic lethality with embryos displaying a disorganized vascular network with edema and hemorrhage ( Ma et al . , 2014 ) . By using an inducible Tie2-cre to disrupt Tsc1 in postnatal mice , cutaneous lymphangiosarcomas and Prox1-positive thin-walled vascular channels developed with an increase in VEGFA levels within cutaneous tumors ( Sun et al . , 2015 ) . Another model of Tsc1 conditional disruption using Darpp32-cre developed kidney cysts by 8 weeks of age and angiosarcomas within the digits visible by postnatal day 21 ( Leech et al . , 2015 ) . These models have demonstrated that Tsc1 deficiency in endothelial cells induces the formation of tumors by a mechanism involving mTORC1 , but additional models are needed to replicate the pathological vascular changes observed in larger vessels in TSC , particularly since analysis of human AMLs has demonstrated TSC2 loss and mTORC1 activation in vessel walls ( Karbowniczek et al . , 2003 ) . Our previous research has demonstrated that TSC skin lesions usually contain TSC2-deficient fibroblast-like cells with hyperactive mTORC1 signaling ( Li et al . , 2005 , 2008; Tyburczy et al . , 2014 ) . These fibroblast-like cells , upon incorporation into xenografts , resulted in skin with increased blood vessel size and number ( Li et al . , 2011 ) . As we previously determined that dermal but not epidermal TSC2 loss occurs in human TSC skin samples ( Li et al . , 2008 ) , we hypothesized that conditional disruption of mouse Tsc2 in mesenchymal cells including dermal cells of the skin would be sufficient to induce highly vascular skin tumor formation and produce a source of Tsc2-deficient cells that could be used to discover factors that contribute to TSC tumorigenesis or have potential as diagnostic or prognostic markers of disease . Here we report the generation of a mouse model with a Prrx1-cre transgene ( Logan et al . , 2002 ) to disrupt a conditional Tsc2 allele ( Hernandez et al . , 2007 ) in the lateral plate mesoderm , which contains cells that give rise to the limb bud and craniofacial mesenchyme . In addition to its activity in dermal fibroblasts within these regions , Prrx1-cre is regionally expressed within adipocytes , chondrocytes , and osteoblasts , but not blood cells or endothelial cells ( Logan et al . , 2002; Greenbaum et al . , 2013; Calo et al . , 2010 ) . Transcriptomic analysis of Tsc2-deficient neonatal dermal fibroblasts from these mice in the presence or absence of sirolimus was used to screen for TSC2-dependent genes that could be indicators for TSC2 deficiency and TSC . This expression signature of TSC2 loss was increased in human bladder cancer with TSC1 or TSC2 mutation . Gal-3 , a pro-angiogenic lectin , was increased in mouse and human samples with TSC2 deficiency and Gal-3 serum levels correlated with LAM severity and the presence of AMLs in a cohort of patients with LAM .
We generated mice with disruption of Tsc2 in mesenchymal progenitor cells by crossing mice with a conditional Tsc2 allele ( Hernandez et al . , 2007 ) ( Tsc2fl ) with mice carrying the cre recombinase transgene driven by a Prrx1 enhancer element ( Logan et al . , 2002 ) and with mice carrying an EYFP fluorescent reporter gene ( Srinivas et al . , 2001 ) to track Tsc2−/− ( KO ) cells . Mice containing homozygous Tsc2fl allele and heterozygous Prrx1-cre transgene ( herein , Tsc2cKOPrrx1-cre mice ) were born live and in expected ratios from crosses ( see Materials and methods section ) . EYFP expression was detected in the limbs , ventral skin and craniofacial regions ( Figure 1A ) . 10 . 7554/eLife . 23202 . 003Figure 1 . Characterization of Tsc2cKOPrrx1-cre mice . ( A ) In vivo imaging of EYFP fluorescence from neonates without ( left 2 pups ) or expressing Prrx1-cre ( right 2 pups ) . Fluorescence was observed in stomach from milk . ( B ) Genotype of neonatal dermal fibroblasts from Prrx1-cre expressing mice ( left panel ) and neonatal epidermis ( right panel ) . Labels represent genotype that was observed from tail DNA , which does not express Prrx1-cre or contain recombined ( Tsc2- ) alleles . ( C ) Western blot of protein from neonatal dermal fibroblasts isolated from ventral skin or limbs of Tsc2cKOPrrx1-cre ( KO ) or Tsc2fl/fl controls ( WT ) probed with indicated antibodies . Similar results seen with greater than 10 cell lines . ( D ) Flow cytometry of cultured EYFP-expressing neonatal limb dermal fibroblasts . Approximately 95% of cells expressed EYFP . ( E ) Histology of skin of Tsc2cKO mice demonstrating greater dermal thickness and cellularity than controls . Scale bar , 0 . 1 mm . Dermal thickness of dorsal forepaw skin in control and Tsc2cKO mice ages 2–5 months ( control n = 9 , Tsc2cKO n = 7 ) . Data presented as mean ± SD , **p<0 . 001 . ( F ) Kaplan-Meier survival analysis was used to determine the median survival of Tsc2cKO mice . The Tsc2cKO differed from control ( p<0 . 001 , log-rank test ) . Among Tsc2cKO mice , median survival for males was 28 weeks of age and for females 24 weeks , however this difference was not significant ( p=0 . 392 , log-rank test ) . Controls n = 28 ( all 28 were censored ) , M Tsc2cKO n = 22 ( 8 censored ) , F Tsc2cKO n = 24 ( 9 censored ) . The source data for this figure are in Figure 1—source data 1 , 2 and 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 00310 . 7554/eLife . 23202 . 004Figure 1—source data 1Source data for Figure 1E . Dermal thickness in control vs . Tsc2cKOPrrx1-cre mice . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 00410 . 7554/eLife . 23202 . 005Figure 1—source data 2 . Source data for Figure 1F . Survival data for Tsc2cKOPrrx1-cre mice . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 00510 . 7554/eLife . 23202 . 006Figure 1—source data 3 . Source data for Figure 1F . Serum chemistry and CBC analysis of Tsc2cKOPrrx1-cre mice . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 00610 . 7554/eLife . 23202 . 007Figure 1—figure supplement 1 . Facial and skeletal phenotype of Tsc2cKOPrrx1-cre ( cKO ) mice . ( A ) Facial abnormalities of cKO mice included bulbous snout and thickened eyelids . ( B ) Histology of whisker pad skin showing increased cellularity , vessels , and collagen . Scale bars , 0 . 1 mm . ( C ) CT image of WT and cKO mice showing skeletal abnormalities of the limbs and cranium . ( D ) Cranial bone thickness is increased more than 3-fold in Tsc2cKOPrrx1-cre mice . Measurement of skull thickness was done by extrapolating from cranial space from similar cross section MRI images near the center of cranial bone in 4–6 month-old adult WT ( n = 4 ) and cKO mice ( n = 8 ) , **p<0 . 001 . ( E ) X-ray of WT and Tsc2cKOPrrx1-cre mice showing shorter and thicker bones with greater bone mass . ( F ) H&E of WT and Tsc2cKOPrrx1-cre showing that increased bone acquisition has drastically increased the bone diameter and reduced the marrow space . Scale bars , 0 . 1 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 007 Analysis by PCR showed that nearly all limb skin fibroblasts , but not epidermal cells , contained the recombined Tsc2fl allele ( herein , Tsc2- allele ) ( Figure 1B ) , confirming mesenchymal specificity . Western blot analysis of fibroblasts from ventral and limb skin of Tsc2cKOPrrx1-cre mice ( KO ) showed nearly undetectable Tsc2 protein in KO as well as increased phosphorylation of S6 , indicating mTORC1 hyperactivation ( Figure 1C ) . Flow cytometry analysis of cultured KO neonatal leg skin fibroblasts indicated that approximately 95% of cells expressed EYFP ( Figure 1D ) , corresponding to the dramatically reduced level of Tsc2 protein observed in these cells . Postnatal Tsc2cKOPrrx1-cre mice had shorter and thicker extremities , thicker bones of the cranium and limbs and a bulbous snout ( Figure 1—figure supplement 1 ) . Histological examination of dorsal forepaw ( Figure 1E and Figure 1—source data 1 ) and whisker pad skin ( Figure 1—figure supplement 1b ) from adult Tsc2cKO mice revealed a thickened , hypercellular dermis . Survival was reduced in Tsc2cKOPrrx1-cre mice with median survival of approximately 24 weeks , and no significant gender difference ( Figure 1F and Figure 1—source data 2 ) . Blood counts and serum chemistry were performed on 17 week old Tsc2cKOPrrx1-cre mice to explain early death and revealed evidence of anemia , however serum markers of kidney and liver function were not different from controls ( Figure 1—source data 3 ) . A growth on the volar surface of forepaws of all Tsc2cKOPrrx1-cre mice was visible beginning at about 3 weeks of age ( Figure 2A ) . Serial magnetic resonance imaging ( MRI ) analysis , done at 4 , 8 and 12 weeks of age in Tsc2cKOPrrx1-cre mice ( n = 6 ) and controls ( WT ) of similar age , showed fluid-filled spaces and nodular masses in some kidneys as early as 4 weeks and in kidneys of 6 of 6 mice by 12 weeks ( Table 1 , Figure 2B green arrow ) . The spleen became enlarged with poorly defined internal structure/patterning in 3 of 6 mice by 8 weeks , and 5 of 6 mice by 12 weeks ( Table 1 and Figure 2B , blue arrow ) . A fluid-containing abnormality in the subcutaneous layer appeared in the shoulder region near the neck and axilla , which increased in size and frequency with age to become nearly universal by 12 weeks of age ( Table 1 and Figure 2B , pink arrows ) . Changes in the liver were less frequently detected by MRI , with dark speckling observed in 1/6 animals by 12 weeks of age . 10 . 7554/eLife . 23202 . 008Figure 2 . Gross and histopathology of Tsc2cKOPrrx1-cre mice ( Tsc2cKO ) . ( A ) Forepaw growth in adult Tsc2cKO . ( B ) Full-body MRI 2D images of 12 week control and serial-imaged Tsc2cKO mice . High signal intensity was detected in upper chest/shoulder ( pink arrows ) . Other abnormalities are seen in kidney ( small cysts , green arrow ) and spleen ( enlargement and irregular patterning , blue arrow ) . ( C ) Gross appearance of kidney and liver from 4 to 5 month old control ( upper panels ) and adult Tsc2cKO mice ( lower panels ) showing typical tumors . ( D ) Gross appearance of spleen and shoulder region from 4 to 5 month old control ( upper panels ) and adult Tsc2cKO mice ( lower panels ) . ( E ) Histologic appearance of shoulder and axillary region of Tsc2cKO mouse . ( Ei ) Hamartomatous region with abnormal vessels and abundant fat . ( Eii ) Abnormally large blood vessel with thick , fibrotic vessel wall and endothelial dysplasia . ( Eiii ) Lipomatous area contains adipose-like cells with variably-increased eosinophilic cytoplasm . ( Eiv ) Lymphatic dysplasia near lymph node . Immunohistochemical studies including anti-GFP ( F , G ) anti-CD31 ( H , I ) , anti-alpha-smooth muscle actin ( SMA ) ( J , K ) and anti-HMB-45 ( L , M ) of forelimb hamartomas in Tsc2cKO mice ( F , H , J , L ) and forelimb vessels of control mice ( G , I , K , M ) . IHC staining of controls and Tsc2cKO tissue was consistent in sections from at least n = 4 mice . Scale bars: 0 . 5 mm for E; 0 . 05 mm for Ei , 0 . 025 mm for Eii-Eiv , G and K; 0 . 05 mm for F , H , I J , L and M . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 00810 . 7554/eLife . 23202 . 009Figure 2—figure supplement 1 . IHC of kidney and liver tumors . A-D: Liver and kidney tumors in Tsc2cKOPrrx1-cre mice were similar in histologic appearance to published reports in Tsc1- and Tsc2- haploinsufficient mice . ( A ) H&E staining of 4 month Tsc2cKOPrrx1-cre kidney showing presence of cystademas developing in the tubular epithelium . ( B ) Anti-GFP staining showing the presence of EYFP-expressing Tsc2 KO cells in kidney cyst epithelium . ( C ) H&E staining of liver tumors in Tsc2cKO mice showed vascular channels filled with blood . ( D ) Anti-GFP staining showing the presence of EYFP-expressing Tsc2 KO cells in tumor region but not hepatocytes . Scale bars for A-D: 0 . 15 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 00910 . 7554/eLife . 23202 . 010Figure 2—figure supplement 2 . MRI angiography of Tsc2cKOPrrx1-cre mice . ( A–D ) Sagittal-view angiograms were created from reconstructed images with maximum signal intensity projections . ( A ) WT anterior region ( B ) Tsc2cKOPrrx1-cre anterior region shows numerous tortuous vessels in head and foreleg regions compared to WT . ( C ) WT posterior region . ( D ) Tsc2cKOPrrx1-cre posterior region contains numerous tortuous vessels . E-F: Transverse view of upper chest region . ( E ) WT upper chest and upper forelimbs . ( F ) Tsc2cKOPrrx1-creupper chest with abnormal vasculature including numerous disorganized vessels in the upper forelimb and cranial regions . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 01010 . 7554/eLife . 23202 . 011Figure 2—figure supplement 3 . Triglyceride and cholesterol analysis of fluid from axillary mass outside thorax . Photos show fluid with cloudy mixture of fluid before and after centrifugation in serum separator tube . Fluid samples in 3/3 mice had higher triglyceride levels ( 408 , 230 , and 1726 mg/dL ) compared to paired serum samples ( 136 , 74 , and 167mg/dL , respectively ) . Normal for serum = 41-258 mg/dL . Cholesterol levels in fluid of these three mice was lower in 3/3 mice ( 90 , 90 , and 95 mg/dL ) compared to serum ( 119 , 117 , and 170 mg/dL ) . High triglyceride levels compared to paired serum samples suggest a contribution of chyle in the fluid . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 01110 . 7554/eLife . 23202 . 012Figure 2—figure supplement 4 . Elastin staining of human and mouse blood vessels . ( A ) Normal-appearing vessel from LAM patient kidney ( left ) with arrows indicating elastic fiber . Dysplastic vessel ( right ) from AML region of the same kidney with arrows showing thin , fragmented elastic fibers . Scale bars , 0 . 100 mm . ( B ) Blood vessel from WT mouse ( left ) with typical stained elastic fiber ( arrow ) . Tumor from shoulder of TSc2cKOPrrx1-cre mouse ( right ) with dysplastic vessel with arrow indicating thin and fragmented elastic fibers . Elastin stained with Hart’s method using reagents purchased from Rowley Biochemical , Inc . , #F-379 . Scale bars , 0 . 050 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 01210 . 7554/eLife . 23202 . 013Figure 2—figure supplement 5 . Upper forelimb lymphatic hamartoma in Tsc2cKOPrrx1-cre mice . ( A ) H&E showing longitudinal forelimb containing a dysplastic and cystic axillary lymph node as part of a lymphatic hamartoma . Bar , 1 mm . ( B–E ) serial sections from region indicated in ‘A’ . Scale bars , 0 . 050 mm . ( B ) Region of dysplastic lymph node containing fibrosis and thin , dilated vessels with rounded endothelial cells . ( C ) Lymphatic marker LYVE1 expression within abnormal lymphatic vessels from lymph node . ( D ) Lymphatic marker VEGFR3 expression within abnormal lymphatic vessels from lymph node . ( E ) Anti-GFP staining to detect EYFP expression shows endothelial expression of EYFP in Tsc2cKO axillary lymph node . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 01310 . 7554/eLife . 23202 . 014Figure 2—figure supplement 6 . Histology of Tsc2cKO forepaw and spleen tumors . ( A ) H&E staining of forepaw showing extensive hamartoma in the volar region . Scale bar , 0 . 250 mm . ( Ai ) . From Figure 2—figure supplement 1A inset . Scale bar , 0 . 05 mm . ( B ) Anti-GFP immunostaining of EYFP expression of forepaw tumor . Scale bar , 0 . 05 mm . ( C ) Anti-GFP staining showing blood vessel from EYFP positive control mouse forepaw . Scale bar , 0 . 025 mm . ( D ) αSMA staining of hamartoma from cKO forepaw . Scale bar , 0 . 05 mm . ( E ) CD31 staining of hamartoma from cKO forepaw . ( F ) H&E staining of splenic hamartoma nodule . Scale bar , 0 . 250 mm . ( Fi ) : From Figure 2—figure supplement 1F inset . Scale bar , 0 . 05 mm . ( G ) Anti-GFP of splenic hamartoma . Scale bar , 0 . 05 mm . ( H ) Anti-GFP of splenic vessel from EYFP positive control mouse . Scale bar , 0 . 025 mm . ( I ) Anti-αSMA staining of splenic tumor . Scale bar , 0 . 05 mm . ( J ) Anti-CD31 staining . Scale bar , 0 . 05 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 01410 . 7554/eLife . 23202 . 015Table 1 . Numbers of mice with abnormalities in organs or upper chest ( UC ) based on serial MRI analysis of 6 Tsc2cKO mice at 4 , 8 , and 12 weeks of age to determine the onset of detectable hamartoma formation . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 015AgeKidneySpleenLiverUC 4 weeks1/60/60/63/6 8 weeks5/63/60/63/6 12 weeks6/65/61/65/6 Kidneys showed grossly visible cystic lesions ( Figure 2C ) that were microscopically apparent in 9 of 11 mice . EYFP expression , an indication of KO cells , was observed in cyst wall epithelial cells ( Figure 2—figure supplement 1B ) . Livers developed vascular growths in 8 of 11 Tsc2cKOPrrx1-cre mice ( Figure 2C ) , and contained EYFP expression in vessel intima ( Figure 2—figure supplement 1D ) . All adult Tsc2cKOPrrx1-cre mice developed enlarged spleens with grossly visible tumor nodules ( Figure 2D , left ) . Results of microscopic evaluation of 5 male and 6 female Tsc2cKOPrrx1-cre mice at an average age of 24 weeks is presented in Table 2 . 10 . 7554/eLife . 23202 . 016Table 2 . Gross and histopathological analysis of tumors present in 11 Tsc2cKO mice . Tumors were considered present by microscopic observation . The age range from mice in this group was 18–40 weeks with an average of 24 weeks . NL=normal , TA= tubular adenoma , VH= vascular hamartoma . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 016MouseKidneyLiverSpleenPaw 1NLVHVHNL 2TAVHVHVH 3TAVHVHVH 4TAVHVHVH 5TAVHVHVH 6NLNLVHVH 7TAVHVHVH 8TAVHVHVH 9TANLVHVH 10TANLVHVH 11TAVHVHVH totals9/118/1111/1110/11 By approximately 12 weeks of age , swelling in the upper thorax near the neck/shoulder was grossly visible . Post-mortem examination revealed cysts involving the proximal forelimbs and upper chest ( Figure 2D right ) . Angiography demonstrated large , tortuous vessels both in the anterior and posterior regions of Tsc2cKOPrrx1-cre mice ( Figure 2—figure supplement 2 ) . In addition the cystic masses also appeared to have a lymphatic component based on fluid collected from subcutaneous axillary masses outside the thorax . This fluid was pink and cloudy ( Figure 2—figure supplement 3 ) and contained higher triglyceride than paired serum samples from these mice , suggesting a contribution of chylous fluid ( Figure 2—figure supplement 3 ) . Histological examination of upper forelimbs showed vascular anomalies with tortuous , dilated blood and lymphatic vessels amid abnormal adipose tissue and skeletal muscle ( Figure 2E ) . Different size proliferations of blood vessels was observed including some containing hyalinized tunica media , smooth muscle hyperplasia , and smooth muscle surrounded by proliferations of smaller capillary-like vessels ( Figure 2Ei ) . Large , dilated vessels , many with thick walls , contained thin , fragmented elastic fibers reminiscent of those observed in human AML ( Figure 2Eii , Figure 2—figure supplement 4 ) . Adipose tissue was reduced in the subcutaneous space of Tsc2cKOPrrx1-cre mice . In contrast , hamartomas found in shoulder frequently contained abundant adipose tissue containing plump nuclei , increased cytoplasm , and either single or multiple fat vesicles ( Figure 2Eiii ) . Abnormal lymphatics contained tortuous lymph node sinuses and vessels lined with plump endothelial cells and surrounded by increased collagen ( Figure 2Eiv ) . Abnormal lymphatic vessels expressed typical lymphatic markers ( VEGFR3 , LYVE1 ) , and also expressed EYFP indicating loss of Tsc2 expression ( Figure 2—figure supplement 5 ) . To detect KO cells in sections of forelimbs , EYFP expression was examined using anti-GFP antibodies . EYFP expression was found throughout the hamartomatous tumors of upper forelimbs and frequently seen in vessels including endothelial cells ( Figure 2F ) . In contrast , control vessels from the upper forelimbs showed EYFP expression in connective tissue fibroblasts and mural cells of vessels , but EYFP staining was not observed in luminal cells ( Figure 2G ) . Within thickened vessels , staining using the blood vessel endothelial marker CD31 highlighted abnormal plump endothelial cells and slit-like spaces in the vessel walls ( Figure 2H ) , features not evident in control vessels ( Figure 2I ) . These abnormal vessel walls contained greater numbers of smooth muscle cells in disorganized arrangement ( Figure 2J ) than control vessels from the upper forelimb ( Figure 2K ) . Hamartomas also expressed HMB-45 , an immunohistochemical marker of LAM and AMLs , in the anomalous vessels of upper forelimbs ( Figure 2L ) but not control vessels ( Figure 2M ) . Tumors of the forepaws and spleen showed abnormal proliferations of blood vessels without accompanying alterations of fat and lymphatics as in the shoulder region . Vascular anomalies of the volar forepaw spanned the bone to the dermis ( Figure 2—figure supplement 6A ) with numerous abnormally thick , hyalinized vessels with smooth muscle hyperplasia and many smaller vessels ( Figure 6—figure supplement 6Ai ) . Grossly visible tumors were not observed in hind paws , possibly a reflection of the later Prrx1-cre expression observed in the developing hindlimbs compared to forelimbs ( Logan et al . , 2002 ) . As in the shoulder tumors , EYFP expression was observed throughout the tumor including perivascular and vessel intima ( Figure 2—figure supplement 6B ) . In contrast , similar age EYFP-expressing control mice ( Tsc2fl/+ , Prrx1-cre+/- ) displayed EYFP expression in forepaw fibroblasts and vessel mural cells , but not endothelium ( Figure 2—figure supplement 6C ) . Smooth muscle hyperplasia was confirmed by staining for SMA ( Figure 2—figure supplement 6D ) and endothelial cells in both larger vessels and proliferations of smaller slit-like vessels stained positively with CD31 ( Figure 2—figure supplement 6E ) . Splenic tumor morphology appeared as nodules forming from large dysplastic blood vessels surrounded by areas containing extensive proliferations of smaller vessels and fibrosis ( Figure 2—figure supplement 6F , Fi ) . EYFP expression was present in endothelial and perivascular cells , both of which were highly abundant within these tumors ( Figure 2—figure supplement 6G ) . Similar to the forepaw , endothelial cells from EYFP-expressing controls did not express EYFP ( Figure 2—figure supplement 6H ) . The large abnormal vessel walls were thickened with SMA-positive mural cells ( Figure 2—figure supplement 6I ) and surrounded by extensive proliferations of thin-walled smaller vessels with CD31-positive endothelial cells ( Figure 2—figure supplement 6J ) . To detect the presence of Tsc2-deficient cells , DNA from neonatal Tsc2cKOPrrx1-cre organs were tested using semi-quantitative 3-primer PCR . Compared to the expected recombination in the limbs , low levels of Tsc2- allele were present in kidney , liver and spleen of Tsc2cKOPrrx1-cre neonates , indicating the presence of KO cells before tumor formation ( Figure 3A upper panel ) . Tumors of adult mice showed increased Tsc2- allele , providing evidence of an expansion of KO cells in tumor-containing tissues ( Figure 3A , lower panel ) . To confirm that EYFP organ-staining represented the presence of KO cells , real-time PCR copy number assays using primers within the conditional region of the Tsc2fl allele ( exon 3 ) were used on isolated splenocytes from EYFP-fluorescing cells enriched by FACS ( Figure 3B and Figure 3—source data 1 ) . EYFP-positive cells had approximately 10% of the Tsc2 exon 3 copy number as cells from WT spleen . Both Tsc2 exon 6 ( outside of the conditional region ) and Tsc1 copy number were unchanged in EYFP-positive cells . Therefore , EYFP positivity in this model is an accurate representation of KO cells . To enrich for mesenchymal cells , isolated splenocytes were cultured on plastic , using fibroblast growth medium . After enzymatic dissociation , cultured splenocytes were stained with fluorescent antibodies against the cell surface marker Thy1 . 2 ( CD90 . 2 ) , a known antigen of mesenchymal lineages . By flow cytometry , we found that about half of cultured splenocytes stain for CD90 . 2 , including most EYFP expressing ( KO ) cells ( Figure 3C ) . Consistent with IHC results , cultured cells from dissociated paw tumors contained an endothelial fraction that expressed EYFP ( Figure 3D , middle right ) . These results demonstrate that anomalous vessels of the spleen and paw contain a combination of KO mesenchymal and endothelial cell components . However , in cells isolated from control EYFP-expressing forepaws , we observed very few EYFP-positive endothelial cells ( Figure 3D , lower right ) . This is also consistent with published results of Prrx1-cre expression absent in bone marrow endothelial cells ( Greenbaum et al . , 2013 ) . To test if the thickened dermis and tumors in Tsc2cKOPrrx1-cre mice responded to mTORC1 inhibition , sirolimus treatment was started in recently weaned 25 day-old mice , for 30 days with alternate-day IP injection ( 5 mg/kg ) . Mice were sacrificed at approximately 7 . 5 weeks of age , two days following the last injection . Forepaw thickness was measured with calipers throughout the treatment course . In both male and female Tsc2cKOPrrx1-cre mice , a significant decrease in thickness across the middle of the forepaw was observed as compared to forepaws from vehicle-treated mice ( Figure 3E and F and Figure 3—source data 2 ) . Dermal thickness of the forepaws was reduced 35% compared to vehicle-treated controls ( Figure 3G and H and Figure 3—source data 3 ) . Post-treatment analysis revealed forepaw vascular hamartomas in 8 of 8 vehicle-treated Tsc2cKOPrrx1-cre mice compared to 1 of 10 sirolimus-treated mice . Kidney tumors and spleen tumors also responded to sirolimus treatment ( Table 3 ) . 10 . 7554/eLife . 23202 . 017Figure 3 . Hamartomas of Tsc2cKOPrrx1-cre ( Tsc2cKO ) mice contain both mesenchymal and endothelial KO cells and are sirolimus-sensitive . ( A ) 3-primer PCR detection of Tsc2- , Tsc2fl and Tsc2+ alleles in neonatal organs ( upper panel ) and adult tumor tissue using genotyping primers ( lower panel ) . Similar results found in three separate neonatal and adult Tsc2cKO mice . ( B ) Copy numbers of Tsc2 exon 3 were reduced in DNA isolated from EYFP-expressing splenocytes enriched by FACS compared to DNA isolated from WT cells . Tsc2 exon 6 and Tsc1 copy number assays run as controls ( n = 3 from YFP+ spleen tumors ) . ( C ) The presence CD90 . 2 expression in EYFP-positive cultured splenocytes from cKO mice confirms Tsc2-deficient mesenchymal component containing Tsc2-deficient cells in spleen tumor . Similar results observed in two other cKO spleens . ( D ) Flow cytometry of dissociated and cultured forepaw tumor confirms that these tumors contain Tsc2-deficient CD31-positive cells . Upper: YFP-negative Tsc2cKO forepaw tumor cells . Middle: YFP-positive Tsc2cKO forepaw tumor cells . Lower: Dissociated and cultured EYFP-expressing cells from control forepaw . E , F: Reduction of tumor size by sirolimus . Sirolimus ( 5 mg/kg ) or vehicle was injected IP every other day in Tsc2cKO mice starting at postnatal day 25 for 30 days . Forepaw thickness was then measured weekly by calipers and mice were sacrificed at day 30 . ( E ) Histological sections of forepaws from 30 day treatment with vehicle ( left ) or sirolimus ( right ) . ( F ) Forepaw thickness in cKO measured during treatment ( day 24 *p=0 . 03 , day 30 *p<0 . 001 ) . G , H: Partial normalization of dermal thickness by sirolimus . ( G ) Histological image of dermis following 30 day sirolimus . Scale bar , 0 . 1 mm . ( H ) Measurement of dermal thickness following 30 day sirolimus ( vehicle n = 8 , sirolimus n = 10 mice , *p=0 . 003 ) . Error bars for F and H indicate ± S . D . The source data for this figure are in Figure 3—source datas 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 01710 . 7554/eLife . 23202 . 018Figure 3—source data 1 . Source data for Figure 3B . Measurement of Tsc2 gene copy number in splenic cells enriched for EYFP fluorescence . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 01810 . 7554/eLife . 23202 . 019Figure 3—source data 2 . Source data for Figure 3H . Dermal thickness in Tsc2cKO mice treated with sirolimus or vehicle . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 01910 . 7554/eLife . 23202 . 020Figure 3—source data 3 . Source data for Figure 3F . Paw thickness measurement in Tsc2cKO mice treated with sirolimus and vehicle . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 02010 . 7554/eLife . 23202 . 021Table 3 . Tsc2cKO mice were treated with either vehicle ( n = 8 ) or 5 mg/kg sirolimus ( n = 10 ) every other day for 30 days . Kidney , spleen , and paw were collected and analyzed by histological examination . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 021LocationVehicleSirolimusKidney4/80/10Spleen7/80/10Paw8/81/10 To identify novel Tsc2-dependent factors abnormally expressed in Tsc2-deficient dermal fibroblasts , RNA sequencing and gene expression data analysis were performed on the transcriptome of Tsc2−/− ( KO , n = 3 ) and Tsc2fl/fl non-cre expressing control neonatal mouse leg skin fibroblasts ( WT , n = 3 ) . In untreated cells , 1387 genes were overexpressed in KO compared to WT , while 437 were overexpressed in WT compared to KO ( false discovery rate <10% ) . These results are summarized as a heatmap in Figure 4A ( first 3 lanes vs middle 3 lanes , and see supplementary Figure 4—source data 1 for the full list ) , which demonstrates high reproducibility of these differentially expressed genes among the dermal fibroblasts lines . Gene ontology enrichment analysis ( see Materials and methods section ) revealed that the signature in KO included genes involved in glucose metabolism , the regulation of cell cycle , and HIF1α responses . 10 . 7554/eLife . 23202 . 022Figure 4 . Transcriptomic analysis of mouse neonatal dermal fibroblasts identified Tsc2-dependent and mTORC1-dependent signature genes including LGALS3 , whose mRNA is elevated in cancers with TSC1 or TSC2 mutations . ( A ) Heatmap of differentially regulated genes from n = 3 ( each sample represents one neonate from a different litter of pups ) WT , Tsc2−/− + vehicle , and Tsc2−/− + sirolimus-treated mouse neonatal dermal fibroblasts with FDR of <10% . Genes are centered to the median of wild type vehicle and Tsc2−/−vehicle . ( B ) Heatmap of selected genes from both Tsc2-dependent and mTORC1-dependent signature genes based on statistically over-represented gene ontology categories ( p<0 . 001 ) including response to hypoxia , regulation of cell death , regulation of cell cycle , and glycolytic processes . ( C ) Heatmap of 11 genes overexpressed in Tsc2−/− and decreased by sirolimus treatment that matched to GO categories ‘extracellular region’ and ‘signaling’ . ( D ) Lgals3 expression in WT , Tsc2−/− + vehicle , and Tsc2−/− + sirolimus-treated dermal fibroblasts . ( E ) Mouse Tsc2-dependent gene expression signature is increased in human bladder cancers with non-silent mutations in either TSC1 or TSC2 ( n = 43 ) compared with tumors containing WT TSC1 or TSC2 genes ( n = 348 , p=0 . 005 ) . ( F ) LGALS3 mRNA expression is elevated in human bladder cancers with non-silent mutations in TSC1 or TSC2 , p=0 . 015 . Boxplot horizontal lines mark 25th , 50th , and 75th percentiles , whiskers extend to the furthest point less than or equal to 1 . 5 times the interquartile range . The source data for this figure are in Figure 4—source data 1 , 2 , 3 , 4 and 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 02210 . 7554/eLife . 23202 . 023Figure 4—source data 1 . Source data for Figure 4A . Gene expression changes in KO ( Tsc2-/- ) fibroblasts versus WT ( wild-type ) fibroblasts with FDR < 10% . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 02310 . 7554/eLife . 23202 . 024Figure 4—source data 2 . Source data for Figure 4A . Gene expression changes in KO fibroblasts treated with sirolimus or vehicle . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 02410 . 7554/eLife . 23202 . 025Figure 4—source data 3 . Source data-Figure 4A . Gene expression changes in WT fibroblasts treated with sirolimus or vehicle . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 02510 . 7554/eLife . 23202 . 026Figure 4—source data 4 . Source data-Figure 4A . Differentially expressed genes corrected by sirolimus . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 02610 . 7554/eLife . 23202 . 027Figure 4—source data 5 . Source data 5-Figure 4E and F . Non-silent mutations in TSC1 or TSC2 found in bladder tumors ( The Cancer Genome Atlas ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 027 To confirm that elevated mTORC1 signaling had an expected role in Tsc2-deficient KO fibroblasts , KO and WT neonatal dermal fibroblasts were treated with 20 nM sirolimus for 24 hr . Sirolimus treatment resulted in 7282 underexpressed genes in KO fibroblasts and 2567 overexpressed genes in KO fibroblasts ( Figure 4A middle 3 lanes vs last 3 lanes , and Figure 4—source data 2 ) . The effect of sirolimus was less in WT cells with 2852 underexpressed genes and 223 overexpressed mRNAs ( Figure 4—source data 3 ) . Genes effected by sirolimus in WT fibroblasts were nearly a subset of the genes effected by sirolimus in KO fibroblasts ( overexpressed genes 87% in common; underexpressed genes 88% in common ) . Sirolimus corrected the effect of Tsc2-deficiency ( WT vs . KO ) for many mRNAs . Ninety-two percent ( 1275 of 1387 ) of genes overexpressed in KO compared to WT were also underexpressed after sirolimus treatment in KO . Likewise , 80% ( 349 of 437 ) of genes that were underexpressed in KO versus WT were also overexpressed after sirolimus treatment in KO ( Figure 4—source data 4 ) . These results indicate sirolimus had the expected effect of reversing much of the dysregulation caused by Tsc2 deficiency in these cells . To identify individual genes that may be of relevance to the diagnosis and/or treatment of TSC , we screened sirolimus-sensitive genes also overexpressed in KO cells and manually reviewed the sirolimus-sensitive genes known to mediate developmental programs and/or angiogenesis as potential mediators of TSC pathogenesis ( Figure 4B ) . Additionally , using PANTHER analysis ( Mi et al . , 2017 , 2013 ) , we screened for genes that were present in both of the ontology categories ‘extracellular region’ and ‘signaling’ , producing a list of 11 genes . ( Figure 4C ) . The only transcript that matched both lists in Figure 4B and C was Lgals3 , which codes for galectin-3 ( Gal-3 ) , a lectin with specificity for beta-galactoside moieties on glycoproteins and has been reported to play roles in angiogenesis and fibrosis ( Li et al . , 2014 ) . Lgals3 was greater in KO than WT fibroblasts ( p=0 . 007 ) and corrected by sirolimus ( Figure 4D ) . As TSC1 or TSC2 mutations occur in cancers , we sought to determine if the TSC2 loss-of-function expression signature was present in human bladder tumors since this tumor type often sustains inactivating TSC1 or TSC2 mutations . Utilizing published gene expression data of bladder tumors from The Cancer Genome Atlas ( TCGA ) ( Cancer Genome Atlas Research Network , 2014 ) , we calculated a signature score for 391 bladder tumors . Of these , 43 contained non-silent mutations in either TSC1 or TSC2 which included missense , nonsense , frame shift , splice site , in frame deletions , or 5’UTR mutations ( Figure 4—source data 5 ) . Signature scores were greater in tumors having non-silent TSC1 or TSC2 mutations versus other bladder tumors ( Figure 4E ) indicating that these cancers have an identifiable gene expression signature derived from inactivation of TSC1 or TSC2 . Interrogation of the TCGA cohort revealed that TSC1 and TSC2 mutant bladder tumors overexpressed LGALS3 versus other bladder tumors . ( Figure 4F ) , suggesting that Gal-3 may be an individual marker for bladder cancers containing inactivating mutations in TSC1 or TSC2 . In the forelimbs of WT mice , cells immunoreactive for Gal-3 were identified in larger vessels with staining seen in some skeletal muscle nuclei ( Figure 5A , left ) , whereas in the forelimb tumor of Tsc2cKOPrrx1-cre mice , areas of dense Gal-3 positivity were observed in both vascular and perivascular cells ( Figure 5A , right ) . In the skin , Gal-3-positive cells were observed in both WT and Tsc2cKOPrrx1-cre mice in the epidermis ( Figure 5B ) . Additionally , Tsc2cKOPrrx1-cre mice contained increased numbers of positively stained dermal fibroblasts ( Figure 5B , lower ) . In early passage KO fibroblasts , both intracellular Gal-3 levels ( Figure 5C ) and secreted Gal-3 ( Figure 5D and Figure 5—source data 1 ) were sharply increased , and 48 hr sirolimus [20 nM] treatment resulted in their partial normalization . Serum of adult Tsc2cKOPrrx1-cre mice had 67% higher Gal-3 levels ( p=0 . 015 ) than similar-age controls ( Figure 5E and Figure 5—source data 2 ) . In sirolimus-treated Tsc2cKOPrrx1-cre mice , Gal-3 serum levels were decreased by 25% ( p=0 . 036 ) ( Figure 5F and Figure 5—source data 2 ) . 10 . 7554/eLife . 23202 . 028Figure 5 . Increased production and secretion of Gal-3 in Tsc2cKOPrrx1-cre mice ( cKO ) , which is partially under the control of mTORC1 . ( A ) Gal-3 immunostaining of forelimb tissues of control and cKO . Hamartoma of cKO mice shows Gal-3 positive staining within vascular and perivascular cells . ( B ) Gal-3 immunostaining of forepaw dermis from control and cKO mice . ( C ) Western blot of neonatal WT and KO dermal fibroblasts untreated or treated with 20 nM sirolimus for 48 hr . Blots were probed with antibodies to Tsc2 , Gal-3 and actin . ( D ) Gal-3 secretion from 48 hr culture supernatants of WT and KO neonatal dermal fibroblasts treated as indicated . Gal-3 levels were measured by a mouse ELISA assay ( n = 3 cell lines for each group ) . *p<0 . 05 . ( E ) A significant increase ( p=0 . 004 ) in serum levels of Gal-3 from adult Tsc2cKOPrrx1-cre mice ( n = 11 ) compared to control mice ( n = 15 ) was observed . ( F ) Serum from 8 week-old , 30 day sirolimus-treated treated mice ( n = 8 mice ) showed a significant ( p=0 . 04 ) decrease in Gal-3 compared to controls ( n = 6 mice ) . Error bars for D-F indicate ± S . D . Scale bars in A and B are 0 . 1 mm . The source data for this figure are in Figure 5—source data 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 02810 . 7554/eLife . 23202 . 029Figure 5—source data 1 . Source data-Figure 5D . Gal-3 levels ( pg/mL ) in Tsc2 WT or KO limb dermal fibroblasts ± 48 hr treatment with sirolimus . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 02910 . 7554/eLife . 23202 . 030Figure 5—source data 2 . Source data-Figure 5E and F . Gal-3 levels in serum of Tsc2cKO and control mice ( 5E ) and Gal-3 serum levels in Tsc2cKO mice treated with sirolimus ( 5F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 030 Gal-3 immunostaining of normal-appearing control skin obtained from TSC patients showed positivity in the epidermis but very little in the dermis ( Figure 6A ) . In contrast , TSC skin tumors had abundant Gal-3 positive dermal fibroblasts ( Figure 6B ) . Gal-3 ELISA of supernatants from fibroblasts grown from TSC skin tumors released more Gal-3 than fibroblasts grown from normal-appearing skin ( Figure 6C and Figure 6—source data 1 ) . Western blot analysis of samples from four patients demonstrated higher intracellular protein levels of Gal-3 in TSC skin tumor fibroblasts than paired normal-appearing skin fibroblasts , although absolute levels of Gal-3 varied among patients ( Figure 6D ) . Gal-3 staining in tissue sections from lungs of patients with LAM showed Gal-3 expression in LAM nodules ( LAM , Figure 6E , Figure 6—figure supplement 1 ) . In renal angiomyolipomas ( AML ) , Gal-3 expression was observed in smooth muscle and adipose cells ( Figure 6F , Figure 6—figure supplement 2 ) . In patients with LAM not taking mTOR inhibitors , percent predicted one second forced expiratory volume ( %FEV1 ) negatively correlated with Gal-3 levels , ( Figure 6G and Figure 6—source data 2 and 3 ) . In serum from patients that were being treated with mTOR inhibitors , no correlation of %FEV1 was found with Gal-3 serum levels ( Figure 6—figure supplement 3 ) . Gal-3 levels were also analyzed with two-way factorial ANOVA in treatment naïve patients with or without a confirmed diagnosis of AML , grouped according to severe LAM ( %FEV1 <80 ) and mild LAM ( %FEV1 >80 ) . Gal-3 levels were higher in the mild LAM group with AML compared to no AML suggesting AMLs were an additional source of serum Gal-3 ( Figure 6H- and Figure 6—source data 2 and 3 ) . As it is known that BMI affects Gal-3 levels ( Weigert et al . , 2010 ) , we tested if our results could be explained by differences in BMI of the patients whose samples we tested . There was a significant positive correlation between BMI and galectin-3 in patients without mTOR inhibitor ( r = 0 . 315 , p=0 . 013 ) . After adjusting for BMI , the partial correlation coefficient between FEV1 and galectin-3 in patients without mTOR inhibitor was still significant ( partial correlation r = −0 . 366 , p=0 . 004 ) . No significant difference in Gal-3 levels was observed between patients with LAM not taking mTOR inhibitor ( 4649 ± 1980 pg/mL , n = 64 ) and normal subjects ( 5023 ± 1646 pg/mL , n = 25 ) . 10 . 7554/eLife . 23202 . 031Figure 6 . Gal-3 expression in TSC skin tumors , LAM nodules and AML of the kidney . ( A ) Gal-3 immunostaining of normal-appearing skin biopsy from TSC patient . ( B ) Gal-3 expression in TSC periungual fibroma skin tumor . ( C ) Gal-3 levels from culture supernatants of fibroblasts grown from TSC skin tumors . Normal-appearing skin from ear ( NL ) n = 11 patients; angiofibroma ( AF ) n = 7 patients , p=0 . 020 vs . NL; periungual fibroma ( PF ) n = 5 patients , p=0 . 028 vs . NL; fibrous cephalic plaque ( FCP ) n = 3 patients , p=0 . 026 vs . NL . For some patients , Gal-3 levels from multiple skin tumor cell lines were averaged . ( D ) Western blot showing correlation of TSC2 and Gal-3 levels . Paired skin samples from four representative patients are shown . NL = cultured normal skin fibroblasts and T = TSC skin tumor fibroblasts . ( E ) Gal-3 expression in LAM nodule representative of n = 4 LAM patient samples tested . Airway epithelium in lower left of panel is also positive . ( F ) Gal-3 expression in angiomyolipoma ( AML ) lesion representative of n = 3 AML patient samples tested . ( G ) Significant negative correlation of %FEV1 with Gal-3 serum levels in patients with proven LAM and not taking mTOR inhibitor ( r = −0 . 32 , p=0 . 010 , n = 64 with only one sample per individual used for analysis ) . N = no TSC ( sporadic LAM ) . Y = TSC-LAM . ( H ) Comparison of Gal-3 levels in LAM patients with and without AML . For patients with mild LAM ( %FEV1 > 80 ) , Gal-3 levels were higher in those with AML ( n = 33 ) than without ( n = 25 ) . *p=0 . 045 . There was no significant difference for Gal-3 levels with respect to AML status for patients with %FEV1 < 80 . For all panels , * represents p<0 . 05 . Error bars indicate ± S . D . Scale bars in A , B , E , and F are 0 . 05 mm . The source data for this figure are in Figure 6—source data 1 , 2 and 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 03110 . 7554/eLife . 23202 . 032Figure 6—source data 1 . Source data-Figure 6C . Gal-3 levels ( log2 ) in culture supernatants from TSC patient-derived skin tumor cell lines . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 03210 . 7554/eLife . 23202 . 033Figure 6—source data 2 . Source data-Figure 6G and H . Gal-3 levels in serum from LAM patients not taking mTOR inhibitor . AML status and %FEV1 are indicated . Each row is from a different patient . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 03310 . 7554/eLife . 23202 . 034Figure 6—source data 3 . %FEV1 and Gal-3 levels in serum from LAM patients taking mTOR inhibitor . Each row is a different patient . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 03410 . 7554/eLife . 23202 . 035Figure 6—figure supplement 1 . Galectin-3 expression in LAM nodule tumor cells , as well as lung epithelium . Serial sections of a LAM nodule are shown in A-D . ( A ) H&E of LAM nodule . ( B ) Anti-Gal-3 at higher magnification ( C ) Anti-αSMA showing positive smooth muscle LAM cells . ( D ) Anti-HMB45 showing scattered positive staining characteristic of LAM nodules . Scale bars: A-D = 0 . 025 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 03510 . 7554/eLife . 23202 . 036Figure 6—figure supplement 2 . Galectin-3 expression AML tumor cells , as well as normal kidney . Serial sections of a renal AML are shown in A-D . ( A ) H&E of AML . ( B ) Anti-Gal-3 at higher magnification . ( C ) Anti-αSMA showing positive smooth muscle AML cells . ( D ) Anti-HMB45 showing scattered positive staining characteristic of AML nodules . Scale bars: A-D = 0 . 025 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 03610 . 7554/eLife . 23202 . 037Figure 6—figure supplement 3 . No correlation of %FEV1 with Gal-3 serum levels in patients with proven LAM taking mTOR inhibitor . Gal-3 serum levels of patients receiving mTOR inhibitor treatment plotted against %FEV1 . Data show that for LAM patients ( n = 72 ) taking mTOR inhibitor , no correlation of Gal-3 with %FEV1 was found ( R=0 . 066 , p=0 . 580 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23202 . 037
Disruption of Tsc2 in mesenchymal progenitors caused extensive and remarkable vascular abnormalities , including dilated , thickened , and tortuous blood vessels in the limbs and neck along with dilated lymphatic vessels and large lymphatic cysts in the axillary and neck regions . In the forepaws , hamartomatous tumors with vascular anomalies formed that were comparable to angiosarcomas or lymphangiosarcomas reported in Tsc1 or Tsc2 genetic mouse models ( Onda et al . , 1999; Kwiatkowski et al . , 2002; Sun et al . , 2015; Leech et al . , 2015 ) , but these models lacked the large thickened tortuous arteries with smooth muscle dysplasia and variable fibrosis observed in the larger vessels in Tsc2cKOPrrx1-cre mice . The microscopic appearance of the shoulder tumors shared features of the vascular abnormalities observed in AMLs in patients with TSC , including large dysplastic vessels with smooth muscle hyperplasia staining positive for HMB45 and thin , fragmented elastic fibers . Additional abnormalities in these mice were similar to those observed in other models , such as liver hemangiomas , renal cystadenomas ( Kobayashi et al . , 1999; Onda et al . , 1999; Kobayashi et al . , 2001; Kwiatkowski et al . , 2002 ) and sclerotic bone ( Fang et al . , 2015a , Fang et al . , 2015b ) in mouse models , and spleen hemangiosarcomas in the Eker rat ( Yeung et al . , 1995; Kubo et al . , 1995 ) . The breadth , predictability and rapid formation of multiple manifestations of tissue dysplasia with high penetrance in the Tsc2cKOPrrx1-cre mice make this an attractive preclinical model for TSC rather than using different mouse models for each phenotype . These mice also provide a novel model system to investigate vascular pathologies of major significance in TSC , since AMLs are a source of life-threatening hemorrhage in TSC ( Byard et al . , 2003 ) , and aneurysms and other TSC-related vascular abnormalities can cause morbidity and mortality ( Salerno et al . , 2010 ) . EYFP reporter expression in the larger abnormal blood vessels in Tsc2cKOPrrx1-cre demonstrated that Tsc2 deletion occurs in cells of the vessel wall and perivascular cells as expected due to embryonic expression patterns of Prrx1-cre ( Logan et al . , 2002; Durland et al . , 2008 ) . In addition , EYFP expression was noted in many endothelial cells , an unexpected finding based on the lack of EYFP expression observed within endothelial cells of WT Prrx1-cre expressing tissues ( Figures 2 , Figure 2—figure supplement 6 , and Figure 3 ) and as reported elsewhere ( Greenbaum et al . , 2013 ) . One possible explanation is that EYFP-positive mural cells with loss of Tsc2 are progenitors for abnormal-appearing endothelial cells in these enlarged vessels , consistent with the presence of populations of vascular wall progenitor cells with potential for forming endothelial cells ( Psaltis and Simari , 2015 ) . Two reports of Tsc1 disruption in vascular smooth muscle did not , however , result in development of Tsc1 KO vascular endothelial cells ( Malhowski et al . , 2011; Houssaini et al . , 2016 ) . It is also possible that EYFP-positive endothelial cells result from expansion of a rare population of Tsc2-deficient endothelial cells . This explanation also fits with the EYFP-positive lymphatic endothelial cells comprising the abnormal lymphatics near lymph nodes . In any case , the dramatic vascular changes and their normalization by sirolimus highlights the importance of controlled mTORC1 signaling in the development and postnatal organ homeostasis of Prxx1-expressing mesenchymal-derived tissues . Fibroblasts grown from Tsc2cKOPrrx1-cre mice were used to identify a gene expression signature of Tsc2 gene inactivation , which included genes involved in glucose metabolism , cell cycle regulation , and HIF1α responses . These processes are consistent with known regulation of these processes by loss of TSC1 or TSC2 ( Laplante and Sabatini , 2013 ) . Enrichment for this expression signature was tested in cancer , focusing on bladder cancer since these show mutations in TSC1 and/or TSC2 in about 15% of cases ( Sjödahl et al . , 2011; Pymar et al . , 2008 ) . Using the TCGA database of cancers , the TSC2 loss-of-function expression signature , as well as levels of LGALS3 , were associated with bladder cancers harboring TSC1/TSC2 inactivating mutations . We propose that TSC2 loss imparts a common transcriptional expression signature including LGALS3 that could be considered for diagnosis and/or treatment options . Mutations in genes concurrently with TSC1/TSC2 , such as those frequently occurring in bladder cancers ( Guo et al . , 2013 ) , will likely define the best treatment course . Refinement of the signature based on effects of additional mutations , differing cell types and/or sirolimus sensitivity may improve the utility of this approach . We found that Gal-3 is elevated inTSC2 deficiency and mTORC1 activation , as Gal-3 levels were increased in human TSC skin tumors and TSC2-null skin tumor fibroblasts . Gal-3 serum levels negatively correlated with severity of LAM disease and positively correlated with the presence of AML in patients with mild LAM ( %FEV1 >80 ) . The usefulness of Gal-3 as a serum marker for LAM is unclear , as serum levels in LAM patients were not different from normal subjects . It is possible that the positive correlations of serum Gal-3 levels with disease severity in LAM patients is influenced by baseline differences in Gal-3 production . Polymorphisms in LGALS3 are known to impact Gal-3 serum levels ( Okada et al . , 2006; Hu et al . , 2011 ) , so future studies could test for LGALS3 polymorphisms as a marker for rates of disease progression . Gal-3 is a pleiotropic carbohydrate-binding protein that can be located intracellular or secreted , whose expression is HIF1α-inducible ( Greijer et al . , 2005 ) , is a known angiogenic factor ( Markowska et al . , 2010; Nangia-Makker et al . , 2000 ) , and frequently has altered expression in cancer ( Thijssen et al . , 2015 ) . Gal-3 overexpression is not restricted to bladder cancer or TSC since it is highly expressed in other cancers and various fibrotic tissues ( Li et al . , 2014; Liu and Rabinovich , 2005 ) and Gal-3 serum levels provides prognostic information for heart failure ( Yancy et al . , 2013 ) . It is not yet known whether additional TSC tumors express Gal-3 or whether Gal-3 levels are elevated in children with TSC . Our findings that sirolimus only partially reduced Gal-3 levels in Tsc2cKOprrx1-cre mice is consistent with the idea that Gal-3 may reflect residual amounts of Tsc2-deficient cells during treatment; although tumors were nearly eliminated by sirolimus in the mice , Tsc2-deficient mesenchymal cells were still present and maintained higher than normal serum Gal-3 levels . A consequence of increased Gal-3 in TSC and LAM may be the stimulation of tumor-promoting pathways . Indeed , Gal-3 is involved in stimulating angiogenesis , neoplastic transformation , resistance to apoptosis , and in metastasis ( Liu and Rabinovich , 2005 ) . Gal-3 levels in bladder cancer are associated with tumor proliferation , progression , and clinical outcome ( Zeinali et al . , 2015 ) . Determining the extent to which Gal-3 is related to the pathology or progression of TSC or LAM will be instructive regarding its potential as a new therapeutic target .
Serum and skin samples used for this study were from a cohort of 139 patients diagnosed with LAM based on a combination of clinical , histopathological , radiological , and serum VEGF-D criteria were used for this study . Patients were enrolled in protocols at the National Institutes of Health ( NIH ) Clinical Center ( protocol 95 H-0186; 96 H-0100; 00H0051 ) , which were approved by the National Heart , Lung , and Blood Institute Institutional Review Board and , written informed consent was obtained for each individual . Mice were housed at the USU animal facility and at the National Heart , Lung , and Blood Institute ( NHLBI ) . All animal studies were performed in adherence to protocols that were approved by the Uniformed Services University ( USU ) Institutional Animal Care and Use Committee and NHLBI Animal Care and Use Committee protocol ( under protocol H-0128 . ) Mice carrying the Tsc2-floxed allele ( Hernandez et al . , 2007 ) , were a gift from Dr . Michael Gambello . Tsc2cKOPrrx1-cre mice were generated by crosses consisting of Prrx1-cre+/- males and homozygous ( Tsc2fl/fl ) females . Male and female Tsc2cKOPrrx1-cre mice were subfertile and therefore were not used for breeding . Tsc2 floxed mice were crossed with ( Rosa26 ) Loxp-stop-Loxp-EYFP cre reporter mice ( Srinivas et al . , 2001 ) to track any cell that expressed or was derived from a Prrx1-cre-expressing cell . Mouse lines expressing the Prrx1-Cre ( Logan et al . , 2002 ) transgene and the EYFP cre reporter ( GT ( Rosa ) 26Sor ) transgene were purchased from The Jackson Laboratory . PCR was performed on DNA isolated from earpunch samples . Mice were genotyped for Tsc2 alleles using three primers in one PCR reaction: Fwd Tsc2 ( common ) : 5’-GCAGCAGGTCTGCAGTGAAT , Rev Tsc2 ( Tsc2fl , Tsc2+ ) : 5’-GCAGCAGGTCTGCAGTGAAT , Rev ( Tsc2- ) : 5’-CCTCCTGCATGGAGTTGAGT . Band sizes were Tsc2+ ( 390 bp ) , Tsc2fl ( 434 bp ) and Tsc2- ( 547 bp ) . For Prrx1-cre genotyping: Fwd Prrx1-cre: 5’-CTCCCTCCTCCTCTCTTGCT , Rev Prrx1-cre: 5’-CCATGAGTGAACGAACCTGGTCG . A band size of 761 bp was present for the transgene . For genotyping the EYFP Gt ( ROSA26 ) reporter: Fwd Gt ( ROSA26 ) Sor 5’-AAGACCGCGAAGAGTTTGTC , Rev Gt ( ROSA26 ) Sor: 5’-AAAGTCGCTCTGAGTTGTTAT . PCR product sizes were 320 bp for mutant ROSA26 locus and 600 bp for WT ROSA26 locus . To detect Tsc2 gene copy number , TaqMan real-time DNA copy number assays were used for Tsc2 intron 3-exon 3 and intron 5-exon six and Tsc1 exon 6-intron 6 ( ThermoFisher Scientific ) . Neonatal mouse dermal fibroblasts: Isolation was carried out essentially as described ( Lichti et al . , 2008 ) , except that skin from limbs was used instead of trunk skin . Each neonate was genotyped using PCR prior to cell isolation and genotyping . Cells were cultured in DMEM with 10% FBS and antibiotics . Human skin tumor fibroblasts: Biopsies used for cell culture were cut into pieces and placed into 35 mm culture dishes with enough DMEM with 10% FBS containing antibiotics to just cover the pieces . Adherent fibroblasts that migrated out were expanded and cryopreserved . Cell lines were all tested for tuberin ( TSC2 ) levels and pS6 levels under serum-starved conditions by Western blot . Cells displaying decreased tuberin and TSC2 activation were used for analysis of Gal-3 levels . Mouse and human cells were free from detectable mycoplasma , using the ATCC Universal Mycoplasma Detection Kit #30–1012K . Mouse Gal-3 ELISA assays were purchased from R&D Systems ( DY1197 ) and human Gal-3 ELISA was from eBioscience/Affymetrix ( BMS279/4 ) . Mouse cell culture supernatants were diluted 1:50 , while mouse serum was diluted 1:400 . For ELISA of human samples , culture supernatants were undiluted , and human serum diluted 1:10 . Gal-3 levels were calculated based on a standard curve of purified recombinant Gal-3 using ELISA analysis software ( http://www . elisaanalysis . com/ ) . 250 , 000 cells were seeded per well of 6-well tissue culture dishes . After 24 hr , the media were removed , cells washed 1X with PBS and media ( DMEM plus 1% FBS ) was added either containing 20 nM sirolimus or DMSO as a vehicle control . Media was changed after 24 hr and supernatants and cell lysates were collected after 48 hr incubation . Gal-3 levels were calculated from a standard curve and normalized to total cellular protein content . Mice were injected I . P . with sirolimus or vehicle 5 mg/kg every other day . A stock solution of sirolimus ( LC Laboratories ) was dissolved in 100% ethanol to a concentration of 50 mg/mL . For injection , sirolimus was suspended to a concentration of 0 . 5 mg/mL in a vehicle consisting of 5% Tween 80 ( Sigma ) and 5% PEG 400 ( Sigma ) . Footpad thickness was measured weekly using calipers . MRI was performed in a 7T , 16 cm horizontal Bruker MRI system ( Bruker , Billerica , MA ) with Bruker ParaVision 5 . 1 software . Mice were anesthetized with 2–3% isoflurane with ECG and respiratory detection ( SA Instruments , Stony Brook , NY ) . Mice were imaged in a 35 mm , m2m Imaging birdcage volume coil ( m2m Imaging , Cleveland , OH ) . Magnevist ( gadopentetate dimeglumine , Bayer HealthCare , Montville , NJ ) diluted 1:10 with sterile 0 . 9% saline , was administered IV at 0 . 1 to 0 . 3 mmol Gd /kg . ECG-gated 2D spin echo images of the chest and abdomen ( TR = 1000 ms , TE = 12 ms , 15–20 , 1 mm slices , 100–120 micron in-plane resolution ) and respiratory-gated 3D FISP images ( TR = 7 . 72 , TE = 3 . 35 , flip angle ( FA ) = 15 , approximately 100 x 100 × 450 micron resolution varying slightly with body size ) of the whole body were acquired . 2D MR angiography of the head , abdomen and hips were acquired for selected mice ( TR = 20 ms , TE = 4 . 2 ms , FA 90 , 86–96 slices , 0 . 3 mm slice thickness , 82–94 micron in plane resolution ) . Images were analyzed with ImageJ software . In vivo EYFP fluorescence in neonatal mice was detected using the Bruker In Vivo Xtreme imaging system ( Billerica , MA ) . Neonates were euthanized by carbon dioxide immediately prior to imaging . Blood was extracted from sacrificed mice by cardiac puncture and divided between tubes for serum ( BD Microtainer #365967 ) and blood ( Sarstedt 1 . 3 ml K3E ) . Fluid was extracted from sacrificed Tsc2cKOPrrx1-cre mice which showed visible swelling in the shoulder/axilla region using a 3 mL syringe with 20G needle . Fluid extracted was variably pink or reddish and cloudy . Both blood and extrapleural fluid were centrifuged in serum separator tube to remove red blood cell component . Serum chemistry and CBC analysis were performed at the NIH Diagnostic and Research Services Branch , Division of Veterinary Resources . Sections were deparaffinized in xylene , and rehydrated through graded alcohol series using distilled water . Sections were heated for antigen retrieval in boiling 0 . 01 M citrate buffer pH 6 . 0 for 10 min or treated with 0 . 1% pepsin ( for anti-HMB-45 only ) . After being washed in PBS , the tissue sections were incubated with 5% goat serum in PBS for 1 hr at room temp to block nonspecific-binding sites . Primary antibodies ( see Table S5 for details ) were diluted in blocking buffer applied to tissue sections overnight at 4°C in a moisture chamber . The following day , tissue sections were washed with PBS and incubated with biotinylated secondary antibody for 30 min at room temperature , then for 30 min in avidin-biotinylated complex ( Vectastain ABC kits , Vector Laboratories , Inc . ) after washing . Staining was visualized with Alkaline Phosphatase substrate ( Vector Laboratories ) for about 30 min . Antibodies used for immunohistochemistry were: anti-GFP ( Life Technologies , #A11122 ) 1:1000 , anti-galectin-3 ( Abcam , #ab53082 ) 1:200 , anti-alpha SMA ( Abcam , #ab5694 ) 1:200 , anti-CD31 ( Abcam , , #ab28364 ) 1:30 , anti-VEGFR-3 ( BD Biosciences , #552857 ) 1:30 , anti-melanosome , clone HMB-45 ( Dako/Agilent Technogies ) . The sections were washed thoroughly in tap water . Meyer’s haematoxylin served as a counterstain . Finally , the sections were mounted in permanent mounting medium ( Vector Laboratories ) . For Gal-3 IHC of TSC skin tumors 4 normal , 7 angiofibromas , 4 periungual fibromas , and 1 fibrous cephalic plaques were stained and analyzed . Most images were taken on a Nikon Eclipse Ti microscope with Nikon DS-Ri2 color CMOS camera . For analysis of morphology and measurements of skin thickness , H&E slides were converted into high resolution digital image files with a NanoZoomer Digital Pathology System ( Hamamatsu ) available in the USU Bioinstrumentation Center ( BIC ) . NDP . view2 software was used to open NanoZoomer files and perform digital measurements of skin thickness . RNA was extracted from Tsc2−/− ( KO ) ( n = 3 ) , KO treated with sirolimus ( n = 3 ) , Tsc2fl/fl ( WT ) ( n = 3 ) , and WT treated with sirolimus ( n=3 ) neonatal mouse dermal fibroblasts using an RNeasy Mini Kit ( Qiagen ) and on-column DNA digestion . Sequencing libraries were generated using the TruSeq Stranded mRNA Library Preparation Kit ( Illumina ) before assessing library size distribution using the Fragment Analyzer ( Advanced Analytical Technologies ) and quantity using the KAPA Library Quantification Kit for NGS ( Kapa Biosystems ) . Sequencing was conducted using a NextSeq 500 ( Illumina ) with paired-end reads at 75 bp length . Sequencing data were aligned to a mouse transcript models using STAR ( Dobin et al . , 2013 ) and expression was quantified using RSEM ( Li and Dewey , 2011 ) to obtain FPKM expression values . FPKM values were adjusted by adding 1 and applying log2 transformation . Differential expression between sample groups were calculated by two class SAM for wild type versus KO , and by paired two-class SAM for sirolimus treated versus vehicle ( Tusher et al . , 2001 ) . Differentially expressed genes were selected as those with FDR < 10% . Genes differentially expressed between KO and WT were referred to as the TSC2 expression signature . Differentially expressed candidate transcripts were queried for cellular component and biological process enrichment analysis using PANTHER Classification System ( Mi et al . , 2017 , 2013 ) and ConsensusPathDB ( Kamburov et al . , 2009 ) . Data are available through the Gene Expression Omnibus ( GEO ) under accession number GSE92589 . The results published here are based on data generated by the TCGA Research Network: http://cancergenome . nih . gov/ . Somatic mutation and gene expression quantification data of The Cancer Genome Atlas ( Cancer Genome Atlas Research Network , 2014 ) were downloaded from http://firebrowse . org/ ( n = 391 tumors ) . Expression data ( normalized RSEM values ) were adjusted by adding 1 , applying log2 transformation , and standardized to z-scores . Expression data were reduced to those human genes contained in the mouse Tsc2 expression signature . Expression of genes in the signature that were underexpressed in KO vs WT were multiplied by −1 so that all genes in the signature are in the same direction . For each human bladder tumor , a TSC2 expression score was defined as the mean of the resulting expression values , similar a published method for calculating expression scores in human tumors from a model system ( Wilkerson et al . , 2012 ) . Human tumors with a non-silent TSC1 or TSC2 mutation were considered TSC1/2 mutant and others as wild type . Cultured fibroblasts were lysed in 20 mM Tris , pH7 . 5 , 150 mM NaCl , 20 mM NaF , 2 . 5 mM Na4P2O7 × 10H2O , 1 mM β-glycerophosphate , 1% NP-40 , 1 mM benzamidine , 10 mM 4-nitrophenyl phosphate , 0 . 1 mM PMSF ( reagents from Sigma ) . Ten percent SDS-PAGE was run using 5 ug of total protein lysate . Proteins were transferred to Invitrolon PVDF ( Life Technologies ) . For blocking membranes , TBS/0 . 1% Tween 20/5% NF milk for 1 hr was used . Antibodies were diluted in blocking buffer and incubated overnight at 4°C . Antibodies which reacted with tuberin/Tsc2 , p-S6 ribosomal protein ( ser235/236 ) , and total S6 ribosomal protein were purchased from Cell Signaling ( #4308 , #2211 and 2217 , respectively ) . Monoclonal anti-β-actin was purchased from Sigma ( #A5441 ) . Paw tumors or approximately 0 . 3 g splenic tumor were excised , minced and digested for 2 hr with 0 . 35% collagenase type I ( Worthington Biochemical ) dissolved in DMEM containing 10% FBS . Cells were washed three times with PBS and cultured on gelatin-coated dishes in endothelial cell growth media ( Vasculife EnGs , LifeLife Technologies ) containing antibiotics and antifungals . CD31 and CD90 . 2 extracellular expression in cells derived from Tsc2cKO tumors were analyzed by flow cytometry . Cells cultured on plastic were harvested with Accutase ( Innovative Cell Technologies ) . Cells were washed with autoMACs Rinsing Solution containing 1% BSA . After washing , 2 . 5 × 105 cells were labeled with APC-labeled anti-CD31 antibodies ( Miltenyi Biotec Cat# 130-097-420 ) , or APC-anti-CD90 . 2 ( Miltenyi Biotec Cat# 130-091-790 ) . Cells were then analyzed on a BD LSRII flow cytometer . The enrichment of YFP positive splenic cells was done by fluorescent sorting on a BD FACSAria cell sorter . Means are presented as mean ± standard deviation unless otherwise indicated . Parametric or nonparametric statistical analysis was chosen based on visual assessment of the normality of distribution of the data . For comparison of means between two normally distributed groups , student’s t-test was used . For effect of sirolimus on footpad thickness , data were analyzed using a mixed model for repeated measures with group as a between-subjects factor , time as a within-subjects factor , and a first-order autoregressive structure for the within-subject correlation . Because the overall model was significant ( p<0 . 001 ) and included a significant group x time interaction ( p<0 . 001 ) , follow-up ANOVA models were performed to compare sirolimus vs . vehicle at each time point with a Bonferroni adjustment for multiple comparisons . For pairwise comparison of means from non-normally distributed groups ( Gal-3 secretion from patient-derived cell lines: tumor vs . normal ) ELISA values were log transformed and pairwise student’s t-test was used . Data were then back-transformed and plotted . For survival analysis , Kaplan Meier plots and log-rank tests were used . For correlation studies , Pearson correlation was used ( 2-tailed ) . The required sample size to detect a correlation of 0 . 3 or greater with 80% power and 5% significance , ( two sided ) is 85 . The actual sample size for this study was limited to the number of patients samples that could accrued over a 6 month time period . Samples from patients without mTOR inhibitor were analyzed using either two-way factorial ANOVA or with main effects of AML ( yes/no ) and %FEV1 ( <80 or >80 ) and their interaction . The simple main effect of AML was then estimated at each level of %FEV1 . A multivariate linear regression model was used to identify variables that were independently associated with Gal-3 . Independent variables included in this model were age , DLCO , lymphatic involvement , AML and %FEV1 . In the multivariate regression model , none of the variables tested ( age , DLCO , lymphatic involvement , AML and FEV1 ) showed a significant association with Gal-3 after adjusting for the other variables ( all P values >0 . 15 ) . | Tuberous sclerosis complex is a genetic condition that causes non-cancerous tumours with lots of blood vessels . It is caused by mutations that inactivate either of two genes known as TSC1 and TSC2 . A signalling molecule called mTOR also contributes to the disease , and drugs that block its activity provide some relief for patients . However , mTOR regulates a wide variety of molecules and so researchers are looking for which ones are responsible for the formation of the tumours . Mesenchymal cells produce bone , muscle and other structural tissues in the body . They also support the formation of blood vessels . Mice – which are often used as model animals in health research – also have mesenchymal cells and a gene that is very similar to the human TSC2 gene ( known as Tsc2 ) . Klover et al . hypothesized that disrupting the Tsc2 gene specifically in the mesenchymal cells of mice may mimic aspects of tuberous sclerosis complex in humans . The experiments show that disrupting Tsc2 in mesenchymal cells does indeed mimic features of the human disease; the mice had shorter lifespans and they developed many tumours with dilated and winding blood vessels . Treating the mice with a drug that inhibits mTOR caused the tumours to shrink . Further experiments show that the loss of Tsc2 alters the production of many proteins involved metabolism , cell growth and sensing the levels of oxygen . For example , mouse cells that lack Tsc2 produce more of a protein called galectin-3 , which appears to help blood vessels and tumours to grow in cancers . Klover et al . also studied tumours from patients with tuberous sclerosis complex and a lung disease that is caused by mutations in TSC2 ( called lymphangioleiomyomatosis ) . The experiments found that many tumours produce higher levels of galactin-3 than normal cells . Bladder cancers with mutations in TSC1 or TSC2 also had higher levels of galectin-3 , suggesting that other diseases linked with mutations in these genes may also result in increased production of galectin-3 . The findings of Klover et al . suggest that galectin-3 may be a useful marker to assess the severity of tuberous sclerosis complex , lymphangioleiomyomatosis and to detect cancers with mutations in TSC1 or TSC2 . The next step is to investigate whether galectin-3 alters blood vessels and tumour growth in these conditions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cancer",
"biology"
] | 2017 | Tsc2 disruption in mesenchymal progenitors results in tumors with vascular anomalies overexpressing Lgals3 |
The importance of mechanical activity in the regulation of muscle progenitors during chick development has not been investigated . We show that immobilization decreases NOTCH activity and mimics a NOTCH loss-of-function phenotype , a reduction in the number of muscle progenitors and increased differentiation . Ligand-induced NOTCH activation prevents the reduction of muscle progenitors and the increase of differentiation upon immobilization . Inhibition of NOTCH ligand activity in muscle fibers suffices to reduce the progenitor pool . Furthermore , immobilization reduces the activity of the transcriptional co-activator YAP and the expression of the NOTCH ligand JAG2 in muscle fibers . YAP forced-activity in muscle fibers prevents the decrease of JAG2 expression and the number of PAX7+ cells in immobilization conditions . Our results identify a novel mechanism acting downstream of muscle contraction , where YAP activates JAG2 expression in muscle fibers , which in turn regulates the pool of fetal muscle progenitors via NOTCH in a non-cell-autonomous manner .
Skeletal muscle development , growth and regeneration rely on muscle stem cells . A major goal of muscle research is to understand the signals that regulate the ability of stem cells to self-renew or differentiate . Skeletal muscle formation involves successive and overlapping phases of embryonic , fetal , perinatal and adult myogenesis . The paired homeobox transcription factors , PAX3 and PAX7 , define the pool of muscle stem cells during developmental , postnatal and regenerative myogenesis ( Gros et al . , 2005; Kassar-Duchossoy , 2005; Relaix et al . , 2005 ) . Fetal myogenesis depends on PAX7-expressing muscle progenitors and is associated with muscle growth ( Hutcheson et al . , 2009; Kassar-Duchossoy , 2005; Relaix et al . , 2005 ) . Muscle progenitors undergo myogenic differentiation program with the activation of the bHLH Myogenic Regulatory Factors ( MRFs ) , MYF5 , MRF4 , MYOD , MYOG ( Tajbakhsh , 2009 ) . By the end of fetal myogenesis , PAX7+ cells adopt a satellite cell position under the basal lamina of muscle fibers ( Biressi et al . , 2007; Bröhl et al . , 2012 ) . During development , mechanical forces generated by muscle contraction are essential for the correct establishment of the musculoskeletal system . Although the influence of the mechanical forces for cartilage , joint , and bone development has been previously addressed ( Nowlan et al . , 2010; Rolfe et al . , 2014; Shwartz et al . , 2013 ) , the consequence of muscle-induced mechanical load for the development of muscle itself is largely unknown . The NOTCH signaling pathway is a central regulator of skeletal muscle stem cells during embryonic , fetal and adult myogenesis [reviewed in Mourikis and Tajbakhsh ( 2014 ) ] . Activation of the NOTCH signaling pathway requires direct cell-cell contact between a signal-sending cell that expresses the NOTCH ligand and a signal-receiving cell that expresses the NOTCH receptor . Upon ligand activation , the intracellular domain of the NOTCH receptor is cleaved , translocates into the nucleus , associates with the transcription factor RBPJ and activates the transcription of the bHLH transcriptional repressor genes , HES and HEY [reviewed in Andersson et al . ( 2011 ) ] . In adult myogenesis , NOTCH is involved in satellite cell activation , proliferation and quiescence [reviewed in Mourikis and Tajbakhsh ( 2014 ) ] and the absence of NOTCH signaling in muscle stem cells results in satellite cell depletion due to premature differentiation ( Bjornson et al . , 2012 ) . In addition , during development , NOTCH has been described to activate embryonic myogenesis in somites ( Rios et al . , 2011 ) . During developmental myogenesis , active NOTCH signaling is associated with proliferating muscle progenitors , while NOTCH ligands are expressed in differentiated muscle cells ( Delfini et al . , 2000; Vasyutina et al . , 2007 ) . NOTCH loss-of-function experiments in mice induce a loss of the muscle progenitor pool due to premature muscle differentiation ( Bröhl et al . , 2012; Czajkowski et al . , 2014; Schuster-Gossler et al . , 2007; Vasyutina et al . , 2007 ) , whereas NOTCH activation represses muscle differentiation in chick and mouse embryos ( Delfini et al . , 2000; Hirsinger et al . , 2001; Mourikis et al . , 2012 ) . While studies have identified NOTCH target genes in fetal muscle progenitors ( Bröhl et al . , 2012; Mourikis et al . , 2012 ) , upstream regulators of NOTCH signaling during developmental myogenesis have not attracted attention . Similarly to NOTCH , the co-transcriptional activator YAP ( Yes-Associated Protein ) promotes satellite cell proliferation and inhibits muscle differentiation in culture ( Judson et al . , 2012; Watt et al . , 2010 ) . In addition to being a nuclear effector of the Hippo pathway , YAP has been identified as a sensor of mechanical activity and mediates cellular and transcriptional responses downstream of mechanical forces ( Aragona et al . , 2013; Dupont et al . , 2011; Porazinski et al . , 2015 ) . In addition to other transcription factors , YAP binds to TEAD DNA binding proteins ( Varelas , 2014 ) . YAP and TEAD1 have been shown to occupy 80% of the same promoters in human mammary epithelial cells ( Zhao et al . , 2008 ) , indicating that YAP/TEAD interaction could constitute the major molecular mechanism of YAP-mediated regulation of gene transcription . The TEAD transcription factors recognize and bind to MCAT elements ( CATTCC ) , which are enriched in regulatory regions of muscle-related genes [reviewed in Wackerhage et al . ( 2014 ) ] . In addition to being involved in muscle stem cell proliferation ( Judson et al . , 2012; Watt et al . , 2010 ) , YAP has been recently shown to be a critical regulator of skeletal muscle fiber size in adult mice ( Goodman et al . , 2015; Watt et al . , 2015 ) . A link between mechanical forces ( provided by muscle contraction ) and signaling pathways that regulate fetal myogenesis has not been established . In this study , we show the importance of mechanical forces in the regulation of the number of fetal muscle progenitors . We show that immobilization induces a NOTCH loss-of-function phenotype in muscles . We further provide evidence that , downstream of mechanical forces , YAP positively regulates the expression of the NOTCH ligand JAG2 in fibers , which maintains the pool of muscle progenitors by activating NOTCH signaling .
To study the effect of mechanical signals on muscle progenitors , we set up an unloading model during chick fetal myogenesis ( Figure 1A ) . We used decamethonium bromide ( DMB ) , which blocks muscle contraction and induces rigid paralysis that leads to immobilization ( Nowlan et al . , 2010 ) . Two days after the inhibition of muscle contraction , we observed a reduction in the overall muscle size ( Figure 1D , H ) , which is consistent with previous reports ( Crow and Stockdale , 1986; Hall and Herring , 1990 ) . In addition , we observed a decrease of 58 . 07% ( ±17 . 66 ) in the number of PAX7+ muscle progenitors in paralyzed compared to control muscles ( Figure 1B , D-F , H-J ) . Consistently , the relative expression levels of PAX7 and MYF5 genes were significantly decreased in DMB-treated limbs compared to control limbs , as early as 12 hr after DMB application and with a more prominent effect at 48 hr ( Figure 1C ) . In addition to the reduction of the muscle progenitor pool , we also observed an increase of myogenic differentiation assessed by an increase of MYOD and MYOG expression using RT-q-PCR or in situ hybridization in immobilized limbs compared to control limbs ( Figure 1C , G , K ) . The number of MYOD-expressing cells was also increased in paralyzed muscles compared to control muscles ( Figure 1M ) . In addition , muscle fibers were larger in limb muscles of DMB-treated fetuses compared to control muscles ( Figure 1E , I ) . The large muscle fibers were associated with several nuclei in paralyzed muscles , while control muscle fibers displayed only one nucleus on transverse sections ( Figure 1L ) . Injection of pancuronium bromide ( PB ) , a drug that exerts a flaccid skeletal muscle paralysis ( Nowlan et al . , 2010 ) led to a similar but less pronounced effect , i . e . a 28 . 92% ( ±7 . 27 ) reduction in the number of PAX7+ cells and a concomitant increase of muscle differentiation ( Figure 1—figure supplement 1 ) . DMB or PB treatments of chick fetal myoblast cultures did neither affect muscle progenitors nor their differentiation and did not change the expression levels of PAX7 , MYF5 , MYOD and MYOG ( Figure 1—figure supplement 2 ) , indicating that DMB and PB did not have any off-target effect on myogenic cells . We conclude that the inhibition of muscle contraction leading to rigid or flaccid paralysis reduces the pool of fetal muscle progenitors and increases their propensity to differentiate . 10 . 7554/eLife . 15593 . 003Figure 1 . Inhibition of muscle contraction decreases the number of limb fetal muscle progenitors . ( A ) Chick fetuses were treated with DMB at E7 . 5 and E8 . 5 , in order to block muscle contraction . ( B ) Number of PAX7+ cells in paralyzed and control muscles . PAX7+ cell number was counted per unit area in dorsal and ventral muscles of three DMB limbs and three control limbs . Results are shown as percentage of control . Error bars indicate standard deviations . The p-value was obtained using the Mann-Withney test . ( C ) RT-q-PCR analyses of muscle gene expression levels in limbs 12 hr , 24 hr and 48 hr after DMB treatment compared to control limbs . For each gene , the mRNA levels of control limbs were normalized to 1 . Graph shows means ± standard errors of the mean of 11 limbs . The p-values were calculated using the Wilcoxon test . Asterisks indicate the p-values *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . ( D–K ) Control limbs ( D–G ) ( N = 5 ) and limbs from DMB-treated embryos ( H–K ) ( N = 5 ) were transversely sectioned at the level of the zeugopod and analyzed for muscle progenitors and differentiated cells by immunohistochemistry using the PAX7 and MF20 antibodies , respectively ( D–F , H–J ) , or for MYOD expression by in situ hybridization followed by immunohistochemistry with MF20 antibody ( G , K ) ( N = 4 ) . Nuclei were labeled with Hoechst ( blue ) . Limb sections are dorsal to the top and posterior to the left . u , ulna , r , radius . ( L ) High magnifications of muscle fibers to show the grouped nuclei in fibers of paralyzed muscles compared to control muscles . ( M ) Number of Hoechst+ nuclei of MYOD-expressing cells versus all Hoechst+ nuclei . Results are shown as percentage of control . Error bars indicate standard deviations . The p-value was obtained using the Mann-Withney test . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 00310 . 7554/eLife . 15593 . 004Figure 1—figure supplement 1 . Muscle flaccid paralysis decreases the number of PAX7+ muscle progenitors and increases their differentiation . Chick embryos were treated with Pancuronium bromide ( PB ) at E7 . 5 and E8 . 5 to induce muscle flaccid paralysis . Embryos were processed 48 hr after PB exposure ( at E9 . 5 ) . Control ( A–D ) and PB-treated ( E–H ) limbs were transversely sectioned and analyzed by immunohistochemistry ( A–C , E–G ) ( N = 3 ) or by in situ hybridization followed by immunohistochemistry ( D , H ) ( N = 3 ) . Immobilized muscles visualized with MF20 were decreased in size compared to control muscles ( E versus A ) . PB-treated embryos displayed a diminution in the number of PAX7+ cells in limb muscles ( F , G ) compared to limb muscles of control embryos ( B , C ) . ( D , H ) MYOG expression was increased in immobilized ( H ) versus control ( D ) muscles . ( I ) Number of PAX7+ cells in paralyzed and control muscles . PAX7+ cell number was counted per unit area in dorsal and ventral muscles of three PB limbs and three control limbs . Results are shown as percentage of controls . Error bars indicate standard deviations . The p-value was calculated using the non-parametric Mann-Whitney test . ( J ) RT-q-PCR analyses of muscle gene expression levels in limbs 48 hr after PB treatment compared to control limbs . For each gene , the mRNA levels of control limbs were normalized to 1 . The p-values were calculated using the non-parametric Wilcoxon test . Error bars indicate standard errors of the mean of eleven samples . Asterisks indicate the p-values **p<0 . 01 . u , ulna , r , radius . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 00410 . 7554/eLife . 15593 . 005Figure 1—figure supplement 2 . DMB or PB treatment in primary cultures of chick fetal myoblasts did not change the expression of muscle genes . ( A–D ) Primary cultures of chick fetal myoblasts were treated with DMB or PB for 48 hr and fixed for immunochemistry . DMB-treated ( C ) or PB-treated ( D ) myoblasts displayed similar PAX7 and MF20 staining compared to control cultures ( A , B ) . ( E , F ) RT-q-PCR analyses of muscle gene expression levels in cultured fetal myoblasts 48 hr after DMB ( E ) ( N = 6 ) or PB ( F ) ( N = 6 ) exposure compared to control cultures ( N = 6 ) . For each gene , the mRNA levels of control cultures were normalized to 1 . Graphs show the means ± standard errors of the mean . The relative mRNA levels of PAX7 , MYF5 , MYOD and MYOG genes were not significantly changed in fetal myoblasts treated with DMB or PB compared to controls . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 005 The concomitant decrease of the muscle progenitor pool and increase of muscle differentiation following muscle paralysis was reminiscent of a NOTCH loss-of-function phenotype . In the murine system , loss of NOTCH signaling results in a reduction of the progenitor pool due to a precocious shift toward differentiation ( Bröhl et al . , 2012; Vasyutina et al . , 2007 ) . To determine whether NOTCH activity was modified upon immobilization , we examined the expression of components of the NOTCH signaling pathway in paralyzed muscles . During fetal myogenesis , the NOTCH ligand JAG2 was expressed in MF20+ differentiated muscle cells ( Delfini et al . 2000 ) , while HES5 a recognized transcriptional readout of NOTCH activity ( Andersson et al . , 2011 ) was excluded from differentiated muscle fibers ( Figure 2A , D ) . The NOTCH ligand DLL1 is expressed during chick and mouse embryonic myogenesis ( Vasyutina et al . 2007 , Delfini et al . , 2000 ) but is not detected by in situ hybridization in chick limb fetal muscles ( Delfini et al . , 2000 ) . JAG2 and HES5 were also expressed in blood vessels ( Figure 2B , E , arrowheads ) . In immobilized fetuses , the expression of the JAG2 and HES5 was decreased in paralyzed muscles ( Figure 2C , F , I ) compared to control muscles ( Figure 2B , E , H ) . We believe that the downregulation of JAG2 and HES5 expression in paralyzed muscles reflected a muscle-specific loss of gene expression , since JAG2 and HES5 expression was not affected in blood vessels ( Figure 2B , C , E , F , H , I , arrowheads ) . Moreover , blood vessels are surrounding fetal muscles in normal conditions ( Figure 2J , K ) , when muscle splitting is accomplished ( Tozer et al . , 2007 ) . Consistently , the JAG2 and HES5 mRNA levels were moderately but significantly downregulated in limbs of immobilized animals ( Figure 2G ) . We believe that the unchanged JAG2 and HES5 expression in blood vessels obscures the changes in JAG2/HES5 expression levels in muscles . HeyL is another transcriptional readout of NOTCH that responds to NOTCH activation in limb fetal muscle cells in mice ( Mourikis et al . , 2012; Bröhl et al . , 2012 ) . The relative mRNA levels of HEYL gene were also significantly downregulated in limbs of immobilized animals compared to controls ( Figure 2G ) . In summary , NOTCH activity was reduced in paralyzed muscles , as assessed by reduced expression of the NOTCH ligand JAG2 in muscle fibers and of the transcriptional readout of NOTCH activity , HES5 , in muscles . The downregulation of NOTCH signaling ( Figure 2 ) and the concomitant reduction of the muscle progenitor pool and increase of differentiation ( Figure 1 , Figure 1—figure supplement 1 ) observed in unloading conditions led us to conclude that immobilization mimics a NOTCH loss-of-function phenotype in muscles . 10 . 7554/eLife . 15593 . 006Figure 2 . Muscle contraction inhibition decreases NOTCH activity in limb muscles . ( A–F , H–I ) In situ hybridization to transverse limb sections at the level of the zeugopod of E9 . 5 chick fetuses treated ( C , F , I ) or not treated ( A , B , D , E , H ) with DMB , with JAG2 ( A–C ) or HES5 ( D–F , H , I ) probe followed by immunostaining with MF20 antibody ( A–D ) ( N = 4 ) . The expression of the NOTCH ligand JAG2 was lost in muscle fibers of paralyzed muscles ( C ) compared to JAG2 normal expression ( A , B ) . ( B , C , arrowheads ) JAG2 expression in blood vessels was not affected in DMB limbs . The expression of the transcriptional readout of NOTCH , HES5 was also decreased in paralyzed muscles ( F , I ) compared to control muscles ( D , E , H ) . ( G ) RT-q- PCR analyses of the expression levels of NOTCH signaling components in limbs of 12 hr and 48 hr DMB-treated fetuses . For each gene , the mRNA levels of control limbs were normalized to 1 . Graph shows means ± standard errors of the mean of eight limbs . The p-values were calculated using the Wilcoxon test . Asterisks indicate the p-values , *p<0 . 05 , **p<0 . 01 . ( J , K ) Transverse limb sections of E9 . 5 fetuses were immunostained with MEP21 and MF20 antibodies to visualize blood vessels and differentiated muscles , respectively ( N = 3 ) . Hoescht was used to visualize nuclei in blue . Limb sections are dorsal to the top and posterior to the left . u , ulna , r , radius . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 006 As the pool of muscle progenitors was decreased after muscle paralysis ( Figure 1 ) , we assessed proliferation and apoptosis of muscle progenitors by EdU incorporation and TUNEL analyses in immobilization conditions ( Figure 3 ) . EdU incorporation into PAX7+ cells was not altered , indicating that the proliferative rate of PAX7+ cells was not changed during immobilization compared to control conditions ( Figure 3A–G ) . Despite the reduced number of PAX7+ cells in paralyzed muscles ( Figure 1 ) , their proliferation capacities were unaffected ( Figure 3A–G ) . While apoptotic cells were very rare in control muscles ( Figure 3H , J , L , N ) , we observed an increase of apoptotic figures in fetal muscles of immobilized animals , 48 hr after DMB treatment ( Figure 3I , K , M , O ) . We found no increase of apoptosis 24 hr after DMB treatment ( data not shown ) . The apoptotic figures were not observed in PAX7+ cells , 48 hr after DMB treatments ( Figure 3H–K ) , nor in MF20+ cells or in Desmin+ cells ( Figure 3J–O ) . The absence of apoptotic figures in myogenic cells suggests that cells undergoing apoptosis upon immobilization are possibly muscle connective tissue cells . Thus , immobilization did neither affect the proliferative capacity nor apoptosis of PAX7+ cells , which was reminiscent of the unchanged proliferation and apoptosis rate of muscle progenitors in NOTCH loss-of-function in mice ( Vasuytina et al . , 2007 ) . 10 . 7554/eLife . 15593 . 007Figure 3 . The proliferation rate of muscle progenitors is not modified in paralyzed muscles . Limb muscles of control ( A–C ) and DMB-treated ( D–F ) fetuses were analyzed for cell proliferation by EdU incorporation . Control and paralyzed muscles displayed EdU+/PAX7+ cells showing proliferating muscle progenitors ( A–F , arrows ) . The percentage of EdU+/PAX7+ cells on PAX7+ cells was analyzed in three muscles of three DMB limbs and three control limbs . ( G ) The percentage of EdU+/PAX7+ cells on PAX7+ cells was similar in both control and paralyzed muscles . Error bars show standard deviations . Control ( H , J , L , N ) and paralyzed ( I , K , M , O ) muscles were analyzed for apoptosis . TUNEL staining was rarely visualized in control muscles ( H , J , L , N ) , while paralyzed muscles displayed an increase of apoptotic figures ( I , K , M , O ) , which were not located in PAX7+ cells ( I ) , in Myosin+ cells ( M ) or in Desmin+ cells ( O ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 007 The inhibition of muscle contraction mimicked a NOTCH loss-of-function phenotype in fetal muscles ( Figures 1–3 ) . To determine whether NOTCH signaling acts downstream of mechanical signals to regulate the muscle progenitor pool , we performed NOTCH rescue experiments in immobilization conditions . DLL1-induced NOTCH activation in chick embryos inhibits muscle differentiation ( Delfini et al . , 2000; Hirsinger et al . , 2001 ) . We first performed ligand-dependent NOTCH activation and analyzed the consequences for PAX7+ cells in normal conditions of muscle activity . In DLL1-activated NOTCH limbs , the pool of PAX7+ cells was maintained at E9 . 5 , despite inhibition of muscle differentiation ( Figure 4—figure supplement 1 ) , which is consistent with the maintenance of PAX7+ progenitors in NICD-expressing cells in mouse fetuses ( Mourikis et al . , 2012 ) . Thus , ligand-activated NOTCH maintained the number of chick fetal muscle progenitors over time and impaired their differentiation . We then forced NOTCH activity in limbs of immobilized fetuses and analyzed the consequences for muscle using the contralateral non-grafted limbs as control ( Figure 4 ) . Under immobilization conditions , the left limbs displayed a reduction in progenitor numbers and increased differentiation ( Figure 4A , C , E , G ) . In contrast , in DLL1-expressing regions of ( right ) limbs , the number of PAX7+ cells was increased by 215 . 81% ( ± 29 . 36 ) and myosin expression was decreased compared to control ( left ) limbs ( Figure 4A–I ) . Since immobilization leads to a 2 . 3-fold ( 58 . 07% ) reduction in the number of PAX7+ cells ( Figure 1B ) , we estimated that retroviral DELTA1-induced NOTCH rescued around 93% the number of PAX7+ cells in paralyzed muscles . We conclude that NOTCH activation prevents the decrease in the number of PAX7+ cells by preventing their inappropriate differentiation under immobilization conditions . 10 . 7554/eLife . 15593 . 008Figure 4 . Ligand-induced NOTCH activity prevents the diminution in the number of fetal muscle progenitors and the increase of muscle differentiation in immobilized fetuses . Transverse sections of contralateral ( A , C , E , G ) and DELTA1/RCAS grafted ( B , D , F , H ) limbs of DMB-treated fetuses were hybridized with the DLL1 probe ( A , B ) to visualize ectopic DLL1 expression in right grafted limbs ( B ) or immunostained with PAX7 and MF20 antibodies ( C–H ) ( N = 3 ) . ( I ) The quantification of PAX7+ cell number per unit area and muscle area was performed in DELTA1/RCAS-grafted and contralateral limbs of the same immobilized embryos . Result was presented as percentage of contralateral limbs , in immobilization conditions . Error bars represent standard deviations of six sections originating from three independent experimental embryos . The p-value was calculated using the Wilcoxon test . Limb sections are dorsal to the top and posterior to the left . u , ulna , r , radius . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 00810 . 7554/eLife . 15593 . 009Figure 4—figure supplement 1 . A continuous source of NOTCH ligand maintained the number of PAX7+ muscle progenitors despite the inhibition of muscle differentiation . ( A ) DELTA1/RCAS-producing cells were grafted into limb buds of E3 . 5 chick embryos . Grafted-embryos were fixed 6 days later at E9 . 5 ( N = 3 ) . DELTA1/RCAS-grafted ( B , D , F , H ) and contralateral ( C , E , G ) limbs were cut transversely and analyzed for DLL1 expression by in situ hybridization ( B ) or for muscle progenitors and differentiated cells by immunohistochemistry using PAX7 and MF20 antibodies ( C–H ) . Hoechst was used to visualize nuclei . Ectopic DLL1 expression visualized in dorsal muscle masses ( B ) maintained the pool of PAX7+ cells ( D , F ) compared to contralateral limbs ( C , E ) despite the decreased number of muscle fibers ( H versus G ) . ( I ) PAX7+ cell number was counted per unit area in DLL1-infected-muscles of three DELTA1/RCAS-grafted and in equivalent muscles in control limbs . Results are shown as percentage of controls . Error bars indicate standard deviations . Sections are dorsal to the top and posterior to the left . u , ulna , r , radius . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 009 NOTCH signaling depends on the direct contact between signal-sending cells that express the NOTCH ligands and signal-receiving cells that express NOTCH receptors and display active NOTCH ( Andersson et al . , 2011 ) . Since we observed a concomitant loss of the expression of the NOTCH ligand JAG2 in muscle fibers and reduced expression of NOTCH target genes in paralyzed muscles ( Figure 2 ) , it was unclear which cell type , between differentiated fibers and progenitors , first sensed muscle contraction . To test whether the loss of NOTCH ligand in muscle fibers would suffice to reduce the size of the muscle progenitor pool , we blocked NOTCH ligand function specifically in muscle fibers . For this , we overexpressed a dominant-negative form of DELTA1 ( DELTA1/DN ) ( Figure 5A ) , which prevents NOTCH ligand processing in signal-sending cells and therefore blocked NOTCH activation in signal-receiving cells ( Chitnis , 2006; Henrique et al . , 1997 ) . We performed chick somite-electroporation at the forelimb level ( Figure 5B ) using a stable vector that can be integrated into the genome in the presence of a transposase ( Bourgeois et al . , 2015 ) . This vector co-expresses the Tomato reporter gene and DELTA1/DN under the control of the mouse Myosin Light Chain ( MLC ) promoter that drives expression in chick-differentiated muscle cells and not in muscle progenitors ( Wang et al . , 2011 ) . In electroporated Tomato-expressing muscles , 44 . 94% ( ±12 . 88 ) of MF20+ cells displayed Tomato expression ( Figure 5D–E ) . We observed that the lack of ligand activity in around 45% of differentiated muscle cells significantly decreased the number of PAX7+ cells by 32 . 42% ( ±7 . 23 ) compared to contralateral limbs ( Figure 5F–K ) . This shows that a diminution of NOTCH ligand activity in muscle fibers suffices to induce a decrease in the number of PAX7+ cells . We conclude that NOTCH ligand activity in muscle fibers is required to maintain the pool of fetal muscle progenitors . 10 . 7554/eLife . 15593 . 010Figure 5 . NOTCH-ligand activity in differentiated muscle cells is required to maintain the pool of fetal muscle progenitors . ( A ) Schematic representation of the recombinant vector co-expressing the Tomato reporter gene and a dominant-negative form of DELTA1 ( DELTA1/DN ) under the control of the Myosin Light Chain ( MLC ) promoter between two Tol2 transposons , and of the transient vector containing the transposase . ( B ) E2 . 5 chick embryos were electroporated at the level of forelimb somites in order to target limb muscle cells . Electroporated and contralateral limbs from the same animals ( N = 4 ) were analyzed 7 days after electroporation , at E9 . 5 . ( C , D , F–J ) Transverse sections of electroporated and contralateral limbs were immunostained with PAX7 and MF20 antibodies and labeled with Hoechst to visualize nuclei in blue . ( D , E ) In electroporated muscles displaying Tomato expression ( D ) , an average of 44 . 94% ( ±12 . 88 ) of MF20+ cells displayed Tomato expression ( E ) . The size and shape of the electroporated muscles was not affected ( D ) compared to control muscles ( C ) . Electroporated muscles displayed a decrease in the number of PAX7+ cells ( G , H , J ) compared to contralateral limbs ( F , I ) . ( K ) The number of PAX7+ cells per unit area and the muscle area were analyzed in electroporated muscles and equivalent muscles of the contralateral limbs ( originating from the same experimental animals ) . Results were presented as percentage of the contralateral limbs ( control ) . The graph shows means ± standard errors of the mean of 14 sections originating from four electroporated embryos . The p-value was calculated using the Wilcoxon test . Limb sections are dorsal to the top and posterior to the left . u , ulna . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 010 We next aimed to identify the signal that could sense mechanical forces and regulate JAG2 expression in muscle fibers . We focused on the transcriptional co-activator YAP that has been shown to sense mechanical signals independently of the Hippo pathway ( Aragona et al . , 2013; Dupont et al . , 2011 ) . YAP1 transcripts and YAP protein were expressed ubiquitously in chick limbs ( Figure 6—figure supplement 1A–E ) . Since the subcellular localization of YAP protein reflects its transcriptional activity ( Dupont et al . , 2011 ) , we examined nuclear YAP staining in MF20+ and PAX7+ cells . Unexpectedly , given the proliferative role of YAP in organ formation and tumorigenesis ( Zanconato et al . 2015 , Piccolo et al . , 2014 ) , we found that 89 . 6% ( ±3 . 65 ) of myonuclei of post-mitotic MF20+ cells displayed nuclear YAP protein ( Figure 6A-C , G , H; Figure 6—figure supplement 1F , G ) . Moreover , YAP nuclear staining was stronger than YAP cytoplasmic staining in MF20+ cells; the cytoplasmic domains being delineated with MF20 staining ( Figure 6G , Figure 6—figure supplement 1F ) . In contrast , only a subset of PAX7+ cells showed nuclear YAP staining ( Figure 6—figure supplement 1H ) . Myoblast proliferation has been shown to be associated with increased nuclear YAP in cell cultures ( Judson et al . , 2012; Watt et al . , 2010 ) . However , we did not detect an obvious correlation between nuclear YAP protein and the proliferative state of PAX7+ cells in chick limb fetal muscles , in vivo ( Figure 6—figure supplement 1I ) . ANKRD1 and CTGF are two recognized direct target genes of YAP in many cell types ( Lai et al . , 2011; Zhao et al . , 2008 ) . ANKRD1 is a muscle ankyrin-repeat protein that binds sarcomeric proteins ( Kojic et al . , 2011 ) , and CTGF is a secreted matricellular protein involved in multiple cellular processes ( Malik et al . , 2015 ) . ANKRD1 and CTGF expression was observed at the tips of MF20+ muscle fibers ( Figure 6I-M , Q ) . In fetal limbs , ANKRD1 was exclusively expressed in muscle fibers ( Figure 6I , K , L ) , whereas CTGF transcripts were observed also in cartilage ( Figure 6M , Q ) . YAP staining in myonuclei and the expression of YAP target genes in muscle fibers show that YAP is active in differentiated muscle cells of chick fetal limbs . The regionalized location of the YAP target gene transcripts at muscle tips along fibers suggests that additional proteins are regionalized at muscle tips to regulate ANKRD1 and CTGF transcription . In immobilized fetuses , ANKRD1 and CTGF expression was lost in muscle fibers ( Figure 6I–R ) , but CTGF expression was maintained in cartilage ( Figure 6M , P–R ) . Consistently , the mRNA levels of ANKDR1 showed a drastic diminution , while those of CTGF displayed a significant but less strong reduction in immobilization conditions ( Figure 6S ) . CYR61 is another matricellular protein of the same family as CTGF and is also a YAP target gene ( Lai et al . , 2011 ) . CYR61 mRNA levels were also decreased in limbs 48 hr after DMB treatment ( Figure 6S ) . Moreover , the number of YAP+ myonuclei was significantly decreased in muscle fibers of paralyzed muscles compared to control muscles ( dropping from 89 . 6% ± 3 . 66 to 15 . 9% ± 4 . 36 ) ( Figure 6H ) . We did not observe any obvious modification of nuclear YAP staining in PAX7+ cells of DMB-treated versus control animals ( Figure 6—figure supplement 1J , K ) , indicating that the nuclear YAP staining in the subpopulation of PAX7+ cells is unrelated to mechanical signals . We conclude that YAP activity is observed in muscle fibers of contracting muscles and that this activity is lost in paralyzed muscles . 10 . 7554/eLife . 15593 . 011Figure 6 . YAP activity is observed in contracting muscle fibers and lost in paralyzed muscles . ( A–F ) Transverse limb muscle sections of E9 . 5 control ( A–C ) and DMB-treated fetuses ( D–F ) were immunostained with YAP and MF20 antibodies and labeled with Hoechst to visualize nuclei in blue ( N = 3 ) . ( A–C ) In myofibers , YAP was preferentially localized in myonuclei ( arrows ) . ( D–F ) In paralyzed limb muscles , the nuclear localization of YAP protein in muscle fibers ( MF20+ cells in red ) was lost ( D–F , arrows ) compared to control muscles ( A–C , arrows ) . ( G ) Focus on YAP+ myonuclei in MF20+ cells in control muscles and YAP- myonuclei in paralyzed muscles ( DMB ) . ( H ) Quantification of the percentage of YAP+ myonuclei in MF20+ cells in control and paralyzed muscles . In situ hybridization to YAP target genes , ANKRD1 ( I , K , L , N , O ) and CTGF ( J , M , Q , R ) followed by immunohistochemistry with MF20 antibody in control limbs ( I-M , Q ) and paralyzed muscles ( N , O , P , R ) ( N = 4 ) . ( I–M , Q ) The YAP target genes were expressed at the tips of muscle fibers ( blue staining in MF20+ cells in brown ) visualized on longitudinal ( I–K , Q ) and transverse ( L , M ) muscle sections . ANKRD1 was exclusively expressed in limb muscles ( L , K ) , while CTGF ( M , Q ) displayed additional expression in cartilage . ( N , O , P , R ) In the absence of muscle contraction , the expression of ANKRD1 and CTGF was lost in muscles but not in cartilage for CTGF . u , ulna , r , radius . For transverse limb sections , dorsal is to the top and posterior to the left . For longitudinal sections , dorsal is to the top and proximal to the left . ( S ) RT-q-PCR analyses of the expression levels of YAP target genes in limbs of 12 hr and 48 hr DMB-treated embryos . For each gene , the mRNA levels of control limbs were normalized to 1 . The graph shows means ± standard errors of the mean of nine biological replicates . The p-values were calculated using the Mann-Whitney test . Asterisks indicate the p-value , **p<0 . 01 , ***p<0 . 001 , ****p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 01110 . 7554/eLife . 15593 . 012Figure 6—figure supplement 1 . YAP expression and activity in limb muscles of control and immobilized fetuses . Limbs from E9 . 5 chick fetuses were longitudinally ( A , B , D ) or transversally ( C , E–G ) sectioned and analyzed for YAP1 expression by in situ hybridization ( blue ) followed by immunochemistry with the MF20 antibody to visualize myosins ( brown ) ( A–C ) or for YAP protein expression ( green ) by double immunohistochemistry with MF20 antibody ( red ) ( D–G ) . Hoechst was used to visualize nuclei . YAP1 transcripts and YAP protein were ubiquitously expressed in limbs . ( F , G ) High magnifications of muscle transverse sections showed YAP protein in myonuclei of postmitotic MF20+ muscle fibers ( G , arrowheads ) . ( F ) YAP staining was higher in myonuclei compared to cytoplasms in MF20+ cells . ( H ) Muscle transverse sections co-immunostained with PAX7 ( red ) and YAP ( green ) showed that PAX7+ cells could be either nuclear YAP+ ( arrowheads ) or nuclear YAP− ( arrows ) . ( I ) Chick limbs from E9 . 5 embryos treated with EdU were analyzed by immunohistochemistry for PAX7 , YAP and EdU . Transverse limb sections co-immunostained with PAX7 ( red ) , YAP ( green ) and EdU ( grey ) antibodies showed that the PAX7+/EdU+ cells were either nuclear YAP+ ( I , arrows ) or nuclear YAP- ( I , arrowheads ) . ( J , K ) Transverse limb sections of control ( J ) or DMB-treated ( K ) E9 . 5 embryos were immunostained with YAP and PAX7 antibodies . PAX7+ cells displaying nuclear YAP staining ( J , arrows ) or no nuclear YAP staining ( J , arrowheads ) were observed in control and paralyzed muscles ( J , K , arrows and arrowheads ) . u , ulna , r , radius . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 012 The concomitant loss of YAP activity and JAG2 expression in muscle fibers of paralyzed muscles prompted us to test whether YAP controlled JAG2 expression . We overexpressed chick fetal myoblasts in a constitutively active form of mouse Yap that cannot be phosphorylated at Ser112 and is therefore preferentially translocated to the nucleus ( Xin et al . , 2013 ) . Transfection of YapS112A/RCAS into fetal myoblasts increased the expression levels of YAP target genes ( CTGF and CYR61 ) ( Figure 7—figure supplement 1A ) . In addition , we also observed changes in muscle gene expression , previously described in C2C12 cells and satellite cell-derived myoblasts ( Judson et al . , 2012; Watt et al . , 2010 ) , i . e . concomitant increase of PAX7 and MYF5 and reduction of MYOD and MYOG expression . YapS112A also activated JAG2 expression and NOTCH target genes ( Figure 7—figure supplement 1A ) . JAG2 expression was also upregulated in differentiated muscle cells after transfection with MLC-Tomato-YapS112A ( Tomato+ FACS-sorted cells ) ( Figure 7—figure supplement 1B ) . Thus , YAP positively regulated JAG2 expression in cultured fetal muscle cells . In order to assess whether YAP would rescue JAG2 expression and the muscle phenotype observed in immobilization conditions , in vivo , we electroporated the MLC-Tomato-YapS112A construct in chick limb somites of embryos that were then treated with DMB . We observed that YAP forced-activity in differentiated muscle cells activated ANKRD1 expression in paralyzed muscles compared to the loss of ANKRD1 in paralyzed muscles ( Figure 7A , B , D , E ) . This shows that YapS112A activates YAP target gene expression in paralyzed muscles . YAP forced-activity also rescued JAG2 expression in paralyzed muscles compared to the loss of JAG2 in paralyzed muscles ( Figure 7A , C , D , F ) . Moreover , the number of PAX7+ cells was increased of 55 . 19% ( ± 10 . 2 ) in muscles displaying YapS112A expression in differentiated muscle cells ( visualized with Tomato expression ) compared to muscles of contralateral limbs ( Figure 7G–N ) . We conclude that YAP forced-activity in muscle fibers prevents the loss of JAG2 expression in muscle fibers and the decrease in the number of adjacent muscle progenitors in immobilization conditions . 10 . 7554/eLife . 15593 . 013Figure 7 . Forced YAP activity in differentiated muscle fibers prevents the decrease of ANKRD1 and JAG2 expression and in the number of fetal muscle progenitors in immobilized fetuses . E2 . 5 chick embryos were electroporated at the level of forelimb somites with MLC-Tomato-YapS112A to force YAP activity in differentiated muscle cells , and then treated with DMB . Contralateral and electroporated limbs from the same immobilized animals ( N = 4 ) were analyzed 7 days after electroporation , at E9 . 5 . ( A–F ) Adjacent sections of contralateral ( A–C ) and electroporated ( D–F ) limbs were immunostained with MF20 and labeled with Hoechst to visualize nuclei in blue ( A , D ) or hybridized with ANKRD1 ( B , E ) and JAG2 ( C , F ) probes . ( D ) Tomato indicates the electroporated muscle fibers . ( A–F ) ANKRD1 ( E ) and JAG2 ( F ) expression was activated in paralyzed muscles in the presence of YapS112A ( D ) in differentiated muscle cells compared to paralyzed muscles ( A–C ) . ( I ) Tomato expression in wholemount MLC-Tomato-YapS112A-electroporated limbs . ( G , H , J–L ) Transverse sections of contralateral left ( G , J ) and MLC-Tomato-YapS112A electroporated right ( H , K , L ) limbs of DMB-treated fetuses were co-immunostained with PAX7 and MF20 antibodies . ( L ) Tomato expression in the same limb section as ( H , K ) to visualize electroporated muscle fibers . ( M ) High magnifications of dorsal muscle areas of DMB and DMB+YapS112A fetuses showing PAX7/TOMATO/HOECHST and MF20/HOECHST . ( N ) PAX7+ cell number was counted per unit area in muscles displaying Tomato expression in electroporated limbs and in equivalent muscles of contralateral limbs of four embryos . Results are shown as percentage of control . Error bars indicate standard deviations . The p-value was obtained using the Mann-Withney test . Limb sections are dorsal to the top and posterior to the left . u , ulna , r , radius . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 01310 . 7554/eLife . 15593 . 014Figure 7—figure supplement 1 . Forced YAP activity increases JAG2 expression in chick fetal myoblasts . ( A ) RT-q-PCR analysis of the mRNA levels of muscle genes , YAP target genes and components of the NOTCH pathway in primary cultures of fetal myoblasts transfected with YapS112A/RCAS ( N = 6 ) . Graph shows means ± standard errors of the means . For each gene , the mRNA levels of cultured fetal myoblasts transfected with a control vector ( Empty/RCAS ) was normalized to 1 . The relative expression levels of YAP target genes CTGF and CYR61 were significantly increased . The relative expression levels of PAX7 and MYF5 were increased , while those of MYOD and MYOG were downregulated . The expression levels of the NOTCH ligand JAG2 were increased upon forced YAP activation as those of the HES4 and HEYL compared to control cultures . ( B ) RT-q-PCR analysis of the mRNA levels of ANKRD1 and JAG2 of differentiated muscle cells obtained after Tomato FACS-sorting of primary cultures of fetal myoblasts after transfection of MLC-Tomato-YapS112A . For each gene , the mRNA levels of cultured fetal myoblasts transfected with a MLC-Tomato was normalized to 1 . The p-values were calculated using the Mann-Whitney test . Asterisks indicate the p-value , *p<0 . 05; **p<0 . 01; ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 014 To define whether YAP directly regulates JAG2 transcription , we performed in vivo ChIP experiments using chick limbs as chromatin source . ChIP sequencing data on chick limb cells analyzing promoter-associated histone marks allowed us to characterize the promoter of the chick JAG2 gene ( Figure 8—figure supplement 1 ) . We identified a putative regulatory region ( −629 bp; −1023 bp ) in the JAG2 promoter , which contained a MCAT element ( CATTCC ) , the known binding motif for TEAD complexes ( Davidson et al . , 1988 ) . We found that YAP was recruited to this regulatory region based on PCR ( Figure 8B ) and RT-q-PCR ( Figure 8D ) analyses . Further sequences , upstream of the JAG2 transcription initiation site , containing ( −8481 −8960 bp ) or not containing ( −4500 bp −4894 bp ) MCAT elements were not occupied by YAP ( Figure 8A , B ) . This result showed that YAP was recruited to the JAG2 promoter region in limb fetal skeletal muscles . The YAP occupancy to this region was decreased in immobilized fetuses based on PCR ( Figure 8C ) and RT-q-PCR ( Figure 8D ) analyses , consistent with the decrease of JAG2 expression in muscle fibers in immobilization conditions ( Figure 2 ) . The YAP recruitment to JAG2 promoter ( containing MCAT elements ) provides a possible mechanism for the contraction-dependent control of JAG2 expression in fetal muscle fibers . 10 . 7554/eLife . 15593 . 015Figure 8 . YAP is recruited to a MCAT element-containing promoter region of the chick JAG2 gene in fetal muscles upon muscle contraction . ChIP assay was performed from eight limbs of E9 . 5 chick control or immobilized fetuses with antibodies against YAP , AcH4 for positive control , or without antibody as a negative control in two independent biological experiments . ChIP products were analyzed by PCR ( B , C ) ( N = 4 ) or by RT-q-PCR ( N = 2 ) ( D ) . Primers targeting a 394 pb fragment named region 1 ( −629 bp −1023 bp ) in the JAG2 promoter region ( A ) identified a DNA sequence immunoprecipitated by YAP by PCR ( B ) or by RT-q-PCR ( D ) , while primers targeting regions 2 and 3 did not show any immunoprecipitation by PCR ( B ) . ( D ) Experiment showing the signal of relative YAP recruitment to JAG2 regulatory region 1 in control and immobilized limbs . Results were represented as percentage of the input . Error bars show standard deviations . The YAP recruitment to JAG2 regulatory region 1 was lost in the absence of muscle contractions assessed by PCR ( C ) and RT-q-PCR ( D ) analyses . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 01510 . 7554/eLife . 15593 . 016Figure 8—figure supplement 1 . ChiP sequencing data with promoter histone marks on the chick JAG2 gene . ChiP sequencing was performed from chick limb cell cultures with H3K4me2 and H3Kme3 , which are promoter and active promoter marks , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 016
The influence of muscle contraction has been extensively studied for skeleton ( reviewed in Schwartz et al . ( 2013 ) and Shea et al . ( 2015 ) ) but not for skeletal muscle development . We show here that mechanical signals are required to maintain the PAX7+ muscle progenitor pool during fetal myogenesis . However , muscle contraction does not change cell proliferation but rather affects the maintenance of PAX7+ cells by preventing their differentiation . Although chondrocyte proliferation is reduced in growth plates of long bone in the absence of muscle contraction ( Roddy et al . , 2011 ) , skeletogenesis is also affected by proliferation-independent mechanisms after immobilization . Muscle contraction is necessary to maintain progenitors of the joint in an undifferentiated state and prevents their differentiation into chondrocytes ( Kahn et al . , 2009 ) . Muscle contraction also controls chondrocyte convergence extension during cartilage development in zebrafish and mice ( Shwartz et al . , 2012 ) . This indicates that progenitor cells of the musculoskeletal system are sensitive to mechanical signals generated by muscle contraction during development and respond to mechanical activity by several mechanisms . Interestingly in plants , physical forces contribute to stem cell maintenance acting on a master regulator ( STM homeobox ) of Arabidopsis shoot meristems , and this is achieved by a mechanism that is independent of cell proliferation ( Landrein et al . , 2015 ) . In the adult , changes in mechanical loading are known to cause variation of muscle size , but the contribution of satellite cells ( adult muscle stem cells ) to muscle atrophy or hypertrophy is a debated issue [reviewed in Brooks and Myburgh ( 2014 ) ] . However , a reduction in the number of satellite cells has been described in adult muscle disuse/unloading animal models ( Mitchell and Pavlath , 2004; Verdijk et al . , 2012 ) . Conversely , training exercises result in an increase of satellite cell number in humans ( Crameri et al . , 2004 , 2007; Suetta et al . , 2013 ) . Moreover , physiological exercise has been shown to increases satellite cell numbers and to be beneficial for sarcopenic muscles [reviewed in Arthur and Cooley ( 2012 ) ] . Thus , our data and the available information in literature indicate that muscle stem cells are sensitive to mechanical forces during muscle development , homeostasis and ageing . It remains to be determined whether similar molecular or cellular mechanisms are active in these diverse settings . Muscle paralysis causes a reduction of NOTCH activity and consequently muscle changes that are similar to those observed in NOTCH loss-of-function experiments . This shows that NOTCH activity is sensitive to mechanical signals during developmental myogenesis . Furthermore , ligand-induced NOTCH activation prevents the concomitant reduction of the fetal progenitor pool and increase muscle differentiation that is observed in the absence of muscle contraction . NOTCH components are also downregulated in bones of muscleless limbs of Spd mutant mice ( Rolfe et al . , 2014 ) , although there is no described NOTCH loss-of-function phenotype in bones of muscleless limbs or immobilized embryos . NOTCH signaling is known to control the development of the cardiovascular system , which experiences mechanical forces . The NOTCH pathway has been proposed as an intermediary between mechanical forces and heart development [reviewed in Granados-Riveron and Brook ( 2012 ) ] , although the precise mechanotransduction pathway has not been identified . In adult muscle , NOTCH signaling has been extensively studied during muscle homeostasis , ageing and regeneration [reviewed in Mourikis and Tajbakhsh ( 2014 ) ] , and correlations between mechanical forces and NOTCH have been established ( Carlson et al . , 2009; Conboy et al . , 2003 ) . In summary , there is evidence for a change of NOTCH activity in different tissues during development and adult life upon muscle loading , but a functional link between mechanical signals and NOTCH has not been reported . We provide evidence that NOTCH signaling emerges as a molecular pathway that acts downstream of mechanical forces to regulate the pool of muscle progenitors during development . We believe that muscle activity acts on NOTCH ligand expression in fetal muscle fibers , since JAG2 expression is lost in fibers upon unloading and the blockade of NOTCH ligand function in muscle fibers suffices to decrease the number of adjacent muscle progenitors . In zebrafish , swimming-induced exercises lead to extensive transcriptional changes in fast muscles , including modification of the expression of NOTCH ligands , Dll1 , Jag1 and Jag2 in addition to other NOTCH components ( Palstra et al . , 2014 ) . Interestingly , on a molecular level , mechanical forces exerted by the signal-sending cell are required for ligand-induced NOTCH activation in a receiving cell that does not appear to sense forces directly ( Gordon et al . , 2015; Wang and Ha , 2013 ) . Lastly , the NOTCH ligand Jagged1 rescues the Duchenne muscular dystrophy phenotype in zebrafish embryos and is upregulated in mildly affected dystrophin-deficient dogs ( Vieira et al . , 2015 ) . All these data converge to the idea that mechanical forces provided by muscle contraction are sensed by NOTCH ligands in muscle fibers . We show here that YAP activity is sensitive to mechanical stimuli in fetal muscles in vivo . The preferential loss of YAP protein in myonuclei and YAP target gene expression in differentiated muscle cells in immobilized fetuses indicate that muscle fibers directly sense mechanical signals . Our results do not provide evidence that muscle progenitors directly sense mechanical signals , since nuclear YAP protein , proliferation rate and apoptosis were not modified in PAX7+ cells in immobilization conditions . Moreover , within fetal muscles , we did not detect any expression of the YAP target genes , ANKRD1 and CTGF elsewhere than in MF20+ muscle fibers . YAP is , however , known to be expressed in activated satellite cells and has been shown to enhance satellite cell proliferation and to inhibit their differentiation in culture ( Judson et al . , 2012; Watt et al . , 2010 ) . Consistently , YAP is upregulated in alveolar rhabdomyosarcoma and is a potent modulator of rhabdomyosarcoma formation ( Crose et al . , 2014; Tremblay et al . , 2014 ) . However , in addition to YAP function in muscle cell proliferation , YAP was recently shown to be upregulated in muscle fibers by mechanical overload and to be critical for the size of skeletal muscle fibers in adult mice ( Goodman et al . , 2015; Watt et al . , 2015 ) , highlighting an additional YAP function in post-mitotic muscle fibers upon loading . YAP and NOTCH pathways have been shown to converge in many biological systems in mice ( Li et al . , 2012 , Yimlamai et al . , 2014 , Manderfield et al . , 2015 ) . Oncogenic YAP variants activate the NOTCH ligand JAG1 and consequently NOTCH signaling in hepatocellular carcinoma cells ( Tschaharganeh et al . , 2013 ) . Moreover , TEAD binds to regulatory sequences of the JAG1 gene in human breast cancer cells ( Zhao et al . , 2008 ) . The current view is that YAP and NOTCH pathways converge to promote proliferation , cell fate and differentiation in a cell-autonomous manner in vertebrates . Using muscle unloading during fetal myogenesis , we provide evidence that YAP acts on muscle progenitors in a NOTCH–dependent and non-cell autonomous manner . This mechanism is reminiscent of that of the Drosophila YAP equivalent , Yorkie , during crystal cell differentiation in hematopoiesis , where Yorkie activates Serrate to induce responses in neighboring cells ( Ferguson and Martinez-Agosto , 2014 ) . We demonstrate a mechanistic link between YAP activity in muscle fibers and the regulation of the muscle progenitor pool and show that this is mediated by a transcriptional regulation of the NOTCH ligand JAG2 ( Figure 9 ) . It remains to be determined if a similar link exists between YAP and NOTCH ligands in adult muscle fibers that could explain changes in satellite cell number during unloading or uploading of muscles . We believe that the molecular mechanism that we identify during fetal development might be central to control muscle maintenance upon mechanical loading . 10 . 7554/eLife . 15593 . 017Figure 9 . Schematic representation of YAP and NOTCH signaling pathways in normal contracting muscles and paralyzed muscles . ( A ) In contracting muscles , nuclear YAP ( green myonuclei ) and YAP target gene transcripts ( green ) are present in post-mitotic muscle fibers . YAP positively regulates JAG2 transcription upon muscle contraction . Ligand-dependent NOTCH activation regulates the muscle progenitor pool , by preventing muscle progenitors to differentiate . ( B ) In paralyzed muscles , nuclear YAP , YAP target genes and JAG2 transcripts are lost in post-mitotic muscle fibers . The absence of the NOTCH ligand JAG2 in fibers , due to the loss of mechanical signals , induces a NOTCH loss-of-function phenotype i . e . a diminution in the number of muscle progenitors and a shift toward differentiation , in a non-cell autonomous manner . DOI: http://dx . doi . org/10 . 7554/eLife . 15593 . 017
Fertilized chick eggs from commercial sources ( White Leghorn , HAAS , Strasbourg , France and JA57 strain , Morizeau , Dangers , France ) were incubated at 37 . 5°C . Chick embryos were staged according to days in ovo . Decamethonium bromide ( DMB ) ( Sigma , France ) and pancuronium bromide ( PB ) ( Sigma ) solutions were freshly prepared before each experiment at 12 mM or 11 mM , respectively , in Hank’s solution ( Sigma ) with Penicillin-Streptomycin at 1% ( Gibco , France ) . The control solution was prepared using Hank’s solution with 1% of Penicillin-Streptomycin . 100 µl of DMB , PB or control solutions were administrated in chick embryos at E7 . 5 and E8 . 5 . Embryos were fixed at E8 for the 12 hr time point , at E8 . 5 for the 24 hr time point and at E9 . 5 for the 48 hr time point . Chick embryonic fibroblasts ( CEFs ) obtained from E10 chick embryos were transfected with DELTA1/RCAS using the Calcium Phosphate Transfection Kit ( Invitrogen , France ) . Cell pellets of approximately 50–100 µm in diameter were grafted into limb buds of E3 . 5 embryos as previously described ( Bonnet et al . , 2010; Delfini et al . , 2000 ) . DELTA1/RCAS-grafted embryos were either harvested at E9 . 5 or treated with DMB or control solution and harvested at E9 . 5 . The pT2AL-MLC-Tomato-2A-DELTA1/DN was designed as follows: the dominant negative form of DELTA1 was defined as a truncated form of DELTA1 that lacks the intracellular domain ( Henrique et al . , 1997 ) and was amplified by PCR from the DELTA1/RCAS ( Delfini et al . , 2000 ) using the following primer sequences: Fw GACTTCGAAATGGGAGGCCGCT and Rv CACGTGTTACTATCACCTGCAGGCCTCG . The DELTA1/DN PCR product was inserted in the pT2AL-MLC-Tomato-2A-GFP ( Bourgeois et al . , 2015 ) after GFP removal to obtain pT2AL-MLC-Tomato-2A-DELTA1/DN ( named as pT2AL-MLC-Tomato-DELTA1/DN ) . The pT2AL-MLC-Tomato-2A-mYapS112A was designed as follows: mYapS112A ( Xin et al . 2013 ) was amplified by PCR from the mYapS112A/RCAS ( McKey et al . , 2016 ) using the following primer sequences: Fw CAATTCGAAATGGAGCCCGCG and Rv GGCCACGTGCTATAACCACGTGAGAAA . The mYapS112A PCR product was inserted in the pT2AL-MLC-Tomato-2A-GFP ( Bourgeois et al . , 2015 ) after GFP removal to obtain pT2AL-MLC-Tomato-2A-mYapS112A ( named as pT2AL-MLC-Tomato-YapS112A ) . Forelimb somite electroporation was performed as previously described ( Wang et al . , 2011 ) . The DNA solution was composed of the pT2AL-MLC-Tomato-DELTA1/DN or pT2AL-MLC-Tomato-YapS112A vectors and a transient transposase-containing vector pCAGGS-T2TP , at a molar ratio of 3:1 . This vector set allows the stable integration of the MLC-Tomato-DELTA1/DN or the MLC-Tomato-YapS112A cassettes into the chick genome . Embryos electroporated with pT2AL-MLC-Tomato-YapS112A were immobilized with DMB treatment . Primary myoblasts were obtained from hindlimbs of E10 chick embryos , as previously described ( Havis et al . , 2012 ) . DMB and PB were applied to proliferating myoblast cultures in high-serum conditions ( 10% ) , at a final concentration of 50 µM and 5 µM , respectively , for 48 hr . Myoblasts were then analyzed by immunohistochemistry for PAX7+ muscle progenitors and MF20+ differentiated cells and for muscle gene expression by RT-q-PCR . To force YAP activity in muscle cells , myoblasts were transfected with YapS112A/RCAS ( McKey et al . , 2016 ) . Muscle cells were amplified in high-serum conditions until confluence , collected and analyzed for gene expression by RT-q-PCR . To obtain differentiated muscle cells that overexpress YapS112A , chick fetal myoblasts were transfected with pT2AL-MLC-Tomato-YapS112A or pT2AL-MLC-Tomato as control and cultured for 4 days . Cells were collected into PBS 1X 10% fetal bovine serum ( FBS ) , stained with DAPI to exclude dead cells and purified via FACS Aria gating on the Tomato+ cell fraction . Total RNAs were extracted from control limbs , experimental limbs or primary fetal myoblast cultures . 500 ng to 1 µg of RNA was reverse-transcribed using the High-Capacity Retrotranscription kit ( Applied Biosystems , France ) . RT-q-PCR was performed using SYBR Green PCR Master Mix ( Applied Biosystems ) . Primer sequences used for RT-q-PCR are listed in Supplementary file 1 . The relative mRNA levels were calculated using the 2^-ΔΔCt method ( Livak and Schmittgen , 2001 ) . The ΔCts were obtained from Ct normalized with GAPDH and RPS17 levels in each sample . Twelve DMB , PB or control limbs , originating from four independent experiments , were used as independent RNA samples . Six samples of cultured myoblasts originating from three independent experiments were used as independent RNA samples for DMB , PB , YapS112A/RCAS or pT2AL-MLC-Tomato-YapS112A experiments . Each RNA sample was analyzed in duplicate . Errors bars in RT-q-PCR results represent standard error of the mean . Fifty microliters of EdU ( Invitrogen ) solution ( 5 mg/ml ) was injected into the circulation in E9 . 5 chick embryos for 1 . 5 hr . Embryos were then fixed and immunohistochemistry was performed on limb transverse sections . Forelimbs of control or manipulated ( DMB , PB , DELTA1/RCAS , DELTA1/RCAS+DMB , pT2AL-MLC-Tomato-DELTA1/DN , pT2AL-MLC-Tomato-YapS112A+DMB ) chick embryos were fixed in 4% paraformaldehyde overnight at 4°C and then processed in gelatin/sucrose for 12 µm cryostat sections . The monoclonal antibodies , MF20 that recognizes sarcomeric myosin heavy chains and PAX7 that recognizes muscle progenitors , developed by D . A . Fischman and A . Kawakami , respectively , were obtained from the Developmental Studies Hybridoma Bank developed under the auspices of the NICHD and maintained by The University of Iowa , Department of Biology Iowa City , IA 52242 . The Desmin antibody was obtained from Sigma ( dilution 1/100 ) . The YAP mouse monoclonal antibody was obtained from Santa Cruz Biotechnology ( dilution 1/50 ) . The MEP21 antibody originates from ( McNagny et al . , 1997 ) . Proliferation analysis ( EdU ) was performed using the Click-iT kit ( Thermo Fisher Scientific , France ) . Apoptosis was detected using the ApoTag kit ( Millipore , France ) . Secondary antibodies were conjugated with Alexa-488 or Alexa-555 ( Invitrogen ) . Nuclei were detected with Hoechst staining ( Molecular Probes ) . Chick forelimbs of control or manipulated ( DMB , PB , DELTA1/RCAS , DELTA1/RCAS+DMB , pT2AL-MLC-Tomato-DELTA1/DN , pT2AL-MLC-Tomato-YapS112A+DMB ) embryos were fixed in Farnoy ( 60% ethanol , 30% formaldehyde ( stock at 37% ) and 10% acetic acid ) overnight at 4°C , and processed for in situ hybridization on wax tissue sections , as previously described ( Wang et al . , 2010 ) . The digoxigenin-labeled mRNA probes were obtained as follows: DLL1 , JAG2 and MYOD ( Delfini et al . , 2000 ) ; HES5 ( Henrique et al . , 1997 ) . CTGF ( EST clones ) and YAP1 ( McKey et al . , 2016 ) probes were linearized with NotI and SacII and synthetized with T3 and SP6 , respectively . ANKRD1 probe was obtained by PCR from E9 . 5 limb tissues using the following primers: Fw: TGGCTCACGGGAAGGAGAAG; Rv: GGTGCTCGGCACAGTCG , cloned into the pCRII-TOPO vector ( Invitrogen ) , linearized with KpnI and synthetized with T7 . ChIP assay was performed as previously described ( Havis et al . , 2012 ) . Eight limbs from E9 . 5 chick embryos were homogenized using a mechanical disruption device ( Lysing Matrix A , Fast Prep MP1 , 40 s at 6 m/s ) . 10 µg of the YAP rabbit polyclonal antibody ( Santa Cruz Biotechnology ) or 10 µg of the anti-acetylated histone H4 ( AcH4 ) antibody ( Upstate Biotechnology ) was used to immunoprecipitate 20 µg of sonicated chromatin . ChIP products were analyzed by PCR to amplify three regions upstream the JAG2 coding sequence or by RT-q-PCR to amplify the region 1 ( Figure 8A ) . The primer list is displayed in Supplementary file 1 . ChIP assay was performed with or without DMB treatment in two independent biological replicates . After immunohistochemistry or in situ hybridization experiments , images were obtained using a Nikon epifluorescence microscope , a Leica DMI600B fluorescence microscope or a Leica SP5 confocal system . All cell number and muscle area measurements were performed using the free software ImageJ ( Rasband , W . S . , ImageJ , U . S . National Institutes of Health , Bethesda , Maryland , USA , http://imagej . nih . gov/ij/ , 1997–2012 ) . To quantify the number of PAX7+ cells in limb muscles of DMB- and PB-treated fetuses , the number of PAX7+ cells was counted per unit area in dorsal and ventral muscles on three sections of each three different limbs originating from either DMB-treated , PB- treated or control embryos . The quantification of PAX7+ cells and muscle area ( delineated with PAX7 and MF20 expression domains ) in DELTA1/RCAS-grafted embryos was performed in three sections of each grafted ( right ) and contralateral ( left ) limbs of three immobilized and three control embryos . Quantification was performed in ectopic DLL1-expressing muscles of grafted right limbs and compared with equivalent muscles of the contralateral left limbs originating from the same embryos . The quantification of the Tomato+/MF20+ cells , MF20+ cells , PAX7+ cells and muscle area in pT2AL-MLC-Tomato-DELTA1/DN electroporated embryos was performed in five sections of each electroporated ( right ) and contralateral ( left ) of four embryos . Quantification was performed in Tomato-expressing muscles of electroporated right limbs and compared with equivalent muscles of the contralateral left limbs originating from the same embryos . The quantification of PAX7+ cells in pT2AL-MLC-Tomato-YapS112A electroporated and immobilized fetuses was performed in four sections of each electroporated ( right ) and contralateral ( left ) of four immobilized fetuses . Quantification was performed in Tomato-expressing muscles of electroporated right limbs and compared with equivalent muscles of the contralateral left limbs originating from the same embryos . Hoechst+ nuclei overlapping with MYOD expression were counted in four different muscles over four different sections and compared with the total number of Hoechst+ nuclei in paralyzed ( DMB 48 hrs ) and control muscles . Data were analyzed using non-parametric two-tailed tests , Mann-Withney test for unpaired samples or Wilcoxon test for paired samples using Graphpad Prism V6 . Results were shown as means ± standard deviations or standard errors of the mean , depending on the size of the samples . The p-values are indicated either with the number on the graphs or with asterisks . Asterisks indicate the different p-values *p<0 . 05 , **p<0 . 01 and ***p<0 . 001 . | Skeletal muscle is attached to the skeleton and allows the body to move . Making a new muscle , or repairing an existing one , relies on stem cells that are present inside muscles . A major goal of skeletal muscle research is to understand the signals that regulate the abilities of muscle stem cells to divide and give rise to more stem cells or to become muscle cells . Molecular signals are known to regulate the numbers of stem cells in the muscle . Skeletal muscles become larger if they are exercised , but it is not clear if mechanical forces generated by muscle contractions directly affect the number of muscle stem cells . The NOTCH signaling pathway contributes to maintaining the population of stem cells in muscles by forcing the stem cells to divide and preventing them from becoming muscle cells . Here , Esteves de Lima et al . investigated whether muscle contraction regulates NOTCH signaling during muscle formation in chick fetuses . The experiments show that muscle contraction stimulates the activity of a protein called YAP in muscle cells , which in turn , activates a gene in the NOTCH signaling pathway known as JAG2 . This increases NOTCH signaling activity in the neighboring stem cells and maintains the number of stem cells in the muscle . The next step following this work will be to establish if this mechanism also operates during muscle formation and regeneration in other animals such as mice and zebrafish . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2016 | Muscle contraction is required to maintain the pool of muscle progenitors via YAP and NOTCH during fetal myogenesis |
Deep transcriptome sequencing has revealed the existence of many transcripts that lack long or conserved open reading frames ( ORFs ) and which have been termed long non-coding RNAs ( lncRNAs ) . The vast majority of lncRNAs are lineage-specific and do not yet have a known function . In this study , we test the hypothesis that they may act as a repository for the synthesis of new peptides . We find that a large fraction of the lncRNAs expressed in cells from six different species is associated with ribosomes . The patterns of ribosome protection are consistent with the translation of short peptides . lncRNAs show similar coding potential and sequence constraints than evolutionary young protein coding sequences , indicating that they play an important role in de novo protein evolution .
Studies performed over the past decade have unveiled a richer and more complex transcriptome than was previously appreciated ( Okazaki et al . , 2002; Carninci et al . , 2005; Kapranov et al . , 2007; Ponjavic et al . , 2007 ) . Thousands of long RNA molecules ( >200 nucleotides ) that do not display the typical properties of well-characterized protein-coding RNAs , and which have been named intergenic or long non-coding RNAs ( lncRNAs ) , have been discovered in several eukaryotic genomes ( Okazaki et al . , 2002; Ponting et al . , 2009; Cabili et al . , 2011; Liu et al . , 2012; Pauli et al . , 2012; Ulitsky and Bartel , 2013 ) . There are several lncRNAs that have regulatory functions ( Guttman and Rinn , 2012; Ulitsky and Bartel , 2013 ) . For example the X-inactive-specific transcript Xist regulates X chromosome inactivation in eutherian mammals ( Brockdorff et al . , 1992 ) . However , the vast majority of lncRNAs do not have a known function . Intriguingly , several recent studies have noted that a large fraction of lncRNAs associate with ribosomes ( Ingolia et al . , 2011; Bazzini et al . , 2014; Juntawong et al . , 2014; van Heesch et al . , 2014 ) . Deep sequencing of ribosome-protected fragments , or ribosome profiling , provides detailed information on the regions that are translated in a transcript ( Ingolia , 2014 ) . According to some studies , the patterns of ribosome protection indicate that lncRNAs are capable of translating short peptides ( Ingolia et al . , 2011; Bazzini et al . , 2014; Juntawong et al . , 2014 ) although others have reached different conclusions ( Guttman et al . , 2013 ) . Many lncRNAs have the same structure as classical mRNAs: they are transcribed by polymerase II , capped and polyadenylated , and accumulate in the cytoplasm ( van Heesch et al . , 2014 ) . However , in contrast to typical protein-coding genes , they tend to contain few introns , are expressed at low levels , exhibit weak sequence constraints , and show limited phylogenetic conservation ( Cabili et al . , 2011; Derrien et al . , 2012; Kutter et al . , 2012; Necsulea et al . , 2014 ) . The association of lncRNAs with ribosomes , and the fact that many of them appear to have arisen relatively recently in evolution , indicate that they could be an important source of new peptides . Levine et al . , who described the first examples of de novo originated genes in Drosophila melanogaster , already noted that non-coding RNAs expressed at low levels could contribute to the birth of novel protein coding genes ( Levine et al . , 2006 ) . Cai et al . found a new protein coding gene in Saccharomyces cerevisiae likely to have been formed from a previously transcribed non-coding sequence ( Cai et al . , 2008 ) . Wilson and Masel observed that ribosome profiling reads from a yeast experiment often mapped to intergenic transcripts ( Wilson and Masel , 2011 ) , and they proposed that this could help provide the raw material for the birth of new protein-coding genes . Another study in yeast found evidence of translation of short species-specific ORFs located in non-genic regions ( Carvunis et al . , 2012 ) . More generally , it is important to consider that de novo protein-coding gene evolution , which was once thought to be a very rare event , is now believed to be relatively common ( Khalturin et al . , 2009; Toll-Riera et al . , 2009; Tautz and Domazet-Lošo , 2011; Long et al . , 2013; Reinhardt et al . , 2013 ) . Recently emerged proteins tend to be very short and evolve under weak evolutionary constraints ( Albà and Castresana , 2005; Levine et al . , 2006; Cai et al . , 2009; Liu et al . , 2010; Xie et al . , 2012; Palmieri et al . , 2014 ) , properties that we also expect to find in the putative ORFs of lncRNAs . The idea that lncRNAs serve as a repository for the evolution of new peptides is appealing but the evidence is still fragmented . In this study , we have analyzed ribosome profiling experiments performed in six different species and measured the sequence coding potential and selective constraints of the putatively translated ORFs in lncRNAs and codRNAs . We have discovered that lncRNAs show very similar characteristics to evolutionary young protein coding genes ( lineage-specific proteins ) . The results strongly support a role for lncRNAs in the production of new peptides .
We obtained polyA+ RNA and ribosome profiling sequencing data from six different published experiments performed in diverse eukaryotic species , mouse ( Mus musculus ) , human ( Homo sapiens , HeLa cells ) , zebrafish ( Danio rerio ) , fruit fly ( D . melanogaster ) , Arabidopsis ( A . thaliana ) , and yeast ( S . cerevisiae ) ( Table 1 ) . After read mapping and transcript assembly , we classified the expressed transcripts longer than 200 nucleotides into coding and long non-coding classes ( codRNAs and lncRNAs , respectively , Table 2 ) . 10 . 7554/eLife . 03523 . 003Table 1 . Data sets used in the studyDOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 003SpeciesGEO AccessionMapped reads ( millions ) Max read length ( bp ) DescriptionReferenceMouse M . musculusRNA-seqGSE30839226 . 043ES cells , E14Ingolia et al . , 2011Ribosome profilingGSE3083939 . 247Human H . sapiensRNA-seqGSE2200429 . 836HeLa cellsGuo et al . , 2010Ribosome profilingGSE2200478 . 336Zebrafish D . rerioRNA-seqGSE329001382 . 22 × 75Series of developmental stagesChew et al . , 2013Ribosome profilingGSE465121040 . 044Fruit fly D . melanogasterRNA-seqGSE491971317 . 9500–2hr embryos , wild typeDunn et al . , 2013Ribosome profilingGSE49197105 . 750Arabidopsis A . thalianaRNA-seqGSE5059779 . 851No stress conditions , TRAP purificationJuntawong et al . , 2014Ribosome profilingGSE50597140 . 351Yeast S . cerevisiaeRNA-seqGSE5211920 . 5450GSY83 , diploidMcManus et al . , 2014Ribosome profilingGSE521196 . 835010 . 7554/eLife . 03523 . 004Table 2 . Fraction of transcripts associated with ribosomesDOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 004codRNAlncRNAExpressedAssociated with ribosomes ( RP ) ExpressedAssociated with ribosomes ( RP ) TotalStringentTotalStringentMouse14 , 24514 , 196 ( 99 . 7% ) 13 , 918 ( 97 . 7% ) 476390 ( 81 . 9% ) 367 ( 77 . 1% ) Human17 , 01116 , 630 ( 97 . 8% ) 16 , 617 ( 97 . 7% ) 934403 ( 43 . 1% ) 343 ( 36 . 7% ) Zebrafish12 , 59511 , 643 ( 92 . 4% ) 11 , 637 ( 92 . 4% ) 2392726 ( 30 . 4% ) 684 ( 28 . 6% ) Fruit fly80418031 ( 99 . 9% ) 7623 ( 94 . 8% ) 2822 ( 78 . 6% ) 10 ( 35 . 7% ) Arabidopsis19 , 16218 , 879 ( 98 . 5% ) 10 , 329 ( 53 . 9% ) 13993 ( 66 . 9% ) 68 ( 48 . 9% ) Yeast47404547 ( 95 . 9% ) 4335 ( 91 . 5% ) 216 ( 28 . 6% ) 6 ( 28 . 6% ) Stringent: number of transcripts significant at p < 0 . 05 using 3′UTRs as a null model ( see ‘Materials and methods’ for more details ) . We detected hundreds of annotated lncRNAs in the vertebrate species ( mouse , human and zebrafish ) , the number being lower ( <150 ) in the other species ( fruit fly , Arabidopsis and yeast ) . In addition , we identified a large number of novel lncRNAs not annotated in the databases , 2488 taking all species together ( Supplementary file 1A ) . The inclusion of such lncRNAs resulted in a sixfold increase in the number of lncRNAs amenable for study in zebrafish and a twofold increase in mouse . In yeast , we only found two annotated lncRNAs , but there were 19 novel ones . In the majority of the analyses , we merged the annotated and the novel lncRNAs . As expected , lncRNAs tended to be much shorter than codRNAs in all the species studied ( Figure 1A ) . We found that most lncRNAs contained at least one short ORF ( ≥24 amino acids ) and often several ORFs . The average ORF size in lncRNAs was between 43 and 68 amino acids depending on the species ( Supplementary file 1B ) . Consistent with previous studies , lncRNAs were expressed at significantly lower levels than codRNAs ( Figure 1B , Wilcoxon test , p < 10−5 ) . 10 . 7554/eLife . 03523 . 005Figure 1 . General characteristics of codRNA and lncRNA transcripts . ( A ) Density plots of transcript length . ( B ) Box-plots of transcript expression level in log2 ( FPKM ) units . lncRNA_ribo: lncRNAs associated with ribosomes; lncRNA_noribo: lncRNAs for which association with ribosomes was not detected . codRNA: coding transcripts encoding experimentally validated proteins except for zebrafish in which all transcripts annotated as coding were considered . The area within the box-plot comprises 50% of the data and the line represents the median value . In all studied species , codRNAs were expressed at higher levels than lncRNAs ( Wilcoxon test , p < 10−5 ) , and lncRNA_ribo at higher levels than lncRNA_noribo ( Wilcoxon test , p < 0 . 005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 005 The analysis of ribosome profiling sequencing data showed that the percentage of expressed coding transcripts associated with ribosomes was >90% in all species , with the highest values ( >99% ) in mouse and fruit fly ( Table 2 ) . Pseudogenes had a lower rate of association with ribosomes than coding RNAs , but surprisingly , in species with many annotated pseudogenes , such as human , mouse , and Arabidopsis , the majority of them showed association with ribosomes ( Supplementary file 1A ) . This appeared to be a true signal; while pseudogenes will typically show sequence similarity to other functional copies in the genome , we only considered uniquely mapped reads with no mismatches . Ribosome profiling is based on deep sequencing , and thus provides an unmatched level of resolution of the translated peptides when compared with current proteomics techniques . This is especially important for short proteins , which are difficult to detect by standard mass spectrometry methods ( Slavoff et al . , 2013 ) . We used the ribosome-associated protein-coding RNA data to investigate the relationship between peptide detection by proteomics and protein length . We found that human and mouse translated proteins between 24 and 80 amino acids long were more difficult to identify in proteomics databases than longer proteins ( Table 3 ) . 10 . 7554/eLife . 03523 . 006Table 3 . Fraction of translated proteins of different size detected in proteomics databasesDOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 006Protein size ( amino acids ) Species24–8081–130131–180>180Mouse27/58 ( 46 . 6% ) 222/286 ( 77 . 6% ) 256/330 ( 77 . 6% ) 3716/4786 ( 77 . 7% ) Human116/272 ( 42 . 6% ) 536/748 ( 71 . 7% ) 669/875 ( 76 . 5% ) 6757/8964 ( 75 . 4% ) Yeast27/30 ( 90 . 0% ) 168/207 ( 81 . 1% ) 234/265 ( 88 . 3% ) 2934/3224 ( 91 . 0% ) Only transcripts encoding experimentally validated proteins ( codRNAe ) were considered . The percentage of lncRNAs scanned by ribosomes ( lncRNA_ribo ) was surprisingly high in all the species studied ( Table 2 ) . The values ranged from 28 . 6% in yeast to 81 . 9% in mouse . This affected the main lncRNA classes described in Ensembl v . 70 , including long intervening non-coding RNAs ( lincRNAs ) or antisense transcripts ( Supplementary file 1C ) . Short transcript size may hinder ribosome association detection ( Aspden et al . , 2014 ) . We also found that the ribosome profiling signal was more difficult to detect in poorly expressed transcripts than in highly expressed ones , both for lncRNAs and codRNAs ( Figure 2 ) . As lncRNAs tend to be expressed at low levels and are short when compared to codRNAs ( Figure 1 ) , we might be underestimating their association with ribosomes . 10 . 7554/eLife . 03523 . 007Figure 2 . Effect of transcript expression level on the detection of ribosome association . The percentage of transcripts associated with ribosomes is shown for several transcript expression intervals . codRNA: annotated coding transcripts encoding experimentally verified proteins ( except in zebrafish for which all coding transcripts were considered ) . lncRNA: annotated and novel long non-coding RNAs . Only species with at least 20 transcripts in each expression bin were plotted . In the rest of species , the data were consistent with the trends shown . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 007 In order to determine if the ribosome profiling signal in lncRNAs was different from noise , we compared ribosome density in the transcripts it to that in 3′untranslated regions ( 3′UTRs ) . More specifically , the null model consisted in a size-matched set of sequences containing randomly taken 3′UTR from annotated coding transcripts . Ribosome density was calculated as the number of ribosome profiling reads divided by RNA-seq reads , a ratio defined as translational efficiency ( TE ) ( Ingolia et al . , 2011 ) . Both codRNAs and lncRNAS displayed much higher TE values than 3′UTRs in all species studied ( Wilcoxon test p < 10−5 , Figure 3 ) . We could reject the null model for 90 . 12% of the lncRNAs and 87 . 19% of the codRNAs associated with ribosomes ( p < 0 . 05 ) ( see details by species in Table 2 , Stringent set ) . Therefore , we concluded that the density of ribosomes in lncRNAs is much higher than expected by spurious ribosome binding . 10 . 7554/eLife . 03523 . 008Figure 3 . TE distribution in human transcripts and 3′UTRs ( null-model ) . Cumulative distribution of TE values in human codRNAs , lncRNAs , and 3′UTR sequences . We randomly selected 3′UTRs with a minimum length of 30 nucleotides to build a set of 3′UTR sequences with the same size distribution as the complete transcripts . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 008 Next , we compared ribosome density in lncRNAs and codRNAs in each of the species focusing on regions covered by ribosome profiling reads to accommodate for any differences in the length of the putatively translated regions . In human , fruit fly , and yeast , TE was higher in codRNAs than in lncRNAs ( Wilcoxon test , p < 0 . 005 ) , but in mouse and zebrafish the opposite trend was observed ( Wilcoxon test , p < 0 . 05 ) ( Figure 4 ) . Despite the differences between the species , which may be due to technical issues , it is clear that lncRNAs can show TE values that are similar or even higher than codRNAs . The results were similar when we restricted the analysis to genes encoding a single transcript to avoid any possible biases due to multiple read mapping or when we employed the maximum TE in 90 nucleotide windows ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 03523 . 009Figure 4 . Ribosome association profiles for codRNAs and lncRNAs . Box-plots of transcript translational efficiency ( TE ) in log2 ( TE ) units . The area within the box-plot comprises 50% of the data , and the line represents the median value . lncRNA: lncRNAs for which association with ribosomes was detected . codRNA: coding RNAs transcripts encoding experimentally validated proteins except for zebrafish in which all transcripts annotated as coding were considered . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 00910 . 7554/eLife . 03523 . 010Figure 4—figure supplement 1 . Additional translational efficiency ( TE ) measures . Single isoforms correspond to data for genes with a single transcript . The number of such genes was 2961 codRNA and 246 lncRNA_ribo for mouse , 2853 codRNA and 150 lncRNA_ribo for human , 9352 codRNA and 412 lncRNA_ribo for zebrafish , 836 codRNA and 18 lncRNA_ribo for fruit fly , and 3024 codRNA and 92 lncRNA_ribo for Arabidopsis . In the case of yeast , all genes were taken . TE max is the maximum TE value taking 90 nucleotide windows . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 010 For comparison , we collected a set of 29 human genes with non-coding functions described in several recent reviews ( Supplementary file 2A; Ponting et al . , 2009; Ulitsky and Bartel , 2013; Fatica and Bozzoni , 2014 ) . Many of these genes play roles in the regulation of gene expression in the nucleus and are thus unlikely to be translated . We only detected expression for five of these genes: Malat1 , Pvt1 , Neat1 , Meg8 , and Cyrano . Transcripts encoded by the first three genes showed ribosome association . In the case of Malat1 , this was also consistently observed in mouse and zebrafish ( in the latter species Malat1 was identified as a novel transcript ) and in the case of Pvt1 in mouse . Given the small number of expressed transcripts , we could not draw any general conclusions for this set . The exact positions of ribosome profiling reads on the RNA can be used to delineate the regions that are being actively translated or to discover new functional ORFs ( Chew et al . , 2013; Guttman et al . , 2013; Ingolia , 2014 ) . Because the ribosome is released after encountering a stop codon , this technique can also be employed to identify novel C-terminal protein extensions ( Dunn et al . , 2013 ) or to evaluate if a predicted ORF is likely to correspond to a translated peptide ( Guttman et al . , 2013 ) . We next aimed at comparing the TE values in different transcript regions , including open reading frames ( ORFs ) , putative 5′ and 3′ untranslated regions ( UTRs ) , and the regions between ORFs . In order to obtain an unbiased picture , it was important to define the different regions in the same way in lncRNAs and codRNAs . In typical codRNAs there is a main translated ORF that covers a large fraction of the transcript , sometimes accompanied by short upstream ORFs in the 5′UTR ( Chew et al . , 2013 ) . However , lncRNAs may potentially encode several short peptides ( Ingolia et al . , 2011 ) . The minimum size of ORFs was set at 24 amino acids ( 75 nucleotides counting the STOP codon ) , as peptides of this size have been identified in genetic screen studies in humans ( Hashimoto et al . , 2001 ) . To simplify the comparisons , we employed the same ORF size cut-off in all species . We also considered both a primary ORF , defined as the ORF with the largest number of ribosome profiling reads , as well as any additional non-overlapping ORFs that mapped to ribosome profiling reads ( rest of ORFs ) . In codRNAs , the primary ORF showed a nearly perfect degree of agreement with the annotated protein , indicating that it was an appropriate metric for the main translated product . Primary ORFs in lncRNAs typically occupied a shorter fraction of the transcript than in codRNAs ( Figure 5A ) . The relative length of the ORF with respect to transcript length did not seem to be a strong predictor of ribosome association , as it did not help distinguish lncRNAs associated with ribosomes ( lncRNA_ribo ) to those not associated with ribosomes ( lncRNA_noribo ) . In lncRNAs , most of the primary ORFs corresponded to proteins less than 100 amino acids long ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 03523 . 011Figure 5 . Ribosome association in different transcript regions . ( A ) Density plot of the relative length of the primary ORF in lncRNA_ribo and codRNA with respect to transcript length . For comparison data for the longest ORF in lncRNA_noribo is also shown ( except for fruit fly due to insufficient data ) . ( B ) Box-plots of TE distribution in primary ORF , 5′UTR , and 3′UTR regions . The area within the box-plot comprises 50% of the data , and the line represents the median value . The analysis considered all transcripts with 5′UTR and 3′UTR longer than 30 nucleotides and >0 . 2 FPKM in all three regions . The number of transcripts was 1956 codRNA and 159 lncRNA_ribo in mouse , 3558 codRNA and 139 lncRNA_ribo in human , 5216 codRNA and 252 lncRNA_ribo in zebrafish , and 2019 codRNA and 33 lncRNA_ribo in Arabidopsis . ( C ) Box-plots of TE distribution in primary ORFs , rest of ORFs with ribosome profiling reads and non-ORF regions ( interORF ) . The analysis considered all transcripts with at least two ORFs and more than 30 nucleotides interORF . The number of transcripts was 3264 codRNA and 204 lncRNA_ribo in mouse , 3104 codRNA and 168 lncRNA_ribo in human , 1646 codRNA and 212 lncRNA_ribo in zebrafish , and 1098 codRNA and 25 lncRNA_ribo in Arabidopsis . Fruit fly and yeast were not included in the last two analyses due to insufficient data ( <8 lncRNA_ribo meeting the conditions ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 01110 . 7554/eLife . 03523 . 012Figure 5—figure supplement 1 . Absolute nucleotide length of ORFs in different kinds of transcripts . In codRNAs and lncRNA_ribo , we selected the primary ORF ( the ORF with the largest number of ribosome profiling reads ) , whereas in lncRNA_noribo we selected the longest ORF . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 01210 . 7554/eLife . 03523 . 013Figure 5—figure supplement 2 . Translational efficiency in single-isoform genes . ( A ) Box-plots of TE distribution in primary ORF , 5′UTR , and 3′UTR regions . The analysis considered only genes with one isoform , with UTR and ORF regions expressed at >0 . 2 FPKM and with 5′UTR and 3′UTR longer than 30 nucleotides . The number of transcripts was 980 codRNA and 97 lncRNA_ribo in mouse , 758 codRNA and 36 lncRNA_ribo in human , 3763 codRNA and 117 lncRNA_ribo in zebrafish , and 1495 codRNA and 32 lncRNA_ribo in Arabidopsis . ( B ) Box-plots of TE distribution in primary ORFs , other ORFs with ribosome profiling reads and non-ORF regions ( interORFs ) . The analysis only considered genes with one isoform in which these regions were longer than 30 nucleotides and with expression >0 . 2 FPKM . The number of transcripts was 1691 codRNA and 113 lncRNA_ribo in mouse , 763 codRNA and 54 lncRNA_ribo in human , 1170 codRNA and 108 lncRNA_ribo in zebrafish , and 817 codRNA and 25 lncRNA_ribo in Arabidopsis . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 01310 . 7554/eLife . 03523 . 014Figure 5—figure supplement 3 . Translational efficiency in annotated transcripts . ( A ) Box-plots of TE distribution in primary ORF , 5′UTR , and 3′UTR regions . The analysis considered only annotated transcripts , with UTR and ORF regions expressed at >0 . 2 FPKM and with 5′UTR and 3′UTR longer than 30 nucleotides . The number of transcripts was 1956 codRNA and 92 lncRNA_ribo in mouse , 3558 codRNA and 138 lncRNA_ribo in human , 5216 codRNA and 54 lncRNA_ribo in zebrafish , and 2019 codRNA and 22 lncRNA_ribo in Arabidopsis . ( B ) Box-plots of TE distribution in primary ORFs , other ORFs with ribosome profiling reads ( rest ORFs ) and non-ORF regions ( interORF ) . The analysis only considered annotated transcripts in which these regions were longer than 30 nucleotides and with expression >0 . 2 FPKM . The number of transcripts was 3264 codRNA and 128 lncRNA_ribo in mouse , 3104 codRNA and 167 lncRNA_ribo in human , 1646 codRNA and 58 lncRNA_ribo in zebrafish , and 1098 codRNA and 18 lncRNA_ribo in Arabidopsis . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 01410 . 7554/eLife . 03523 . 015Figure 5—figure supplement 4 . Translational efficiency in transcripts expressed at different levels . We restricted this analysis to transcripts with ORF and UTR regions expressed at >0 . 2 FPKM and with 5′UTR and 3′UTR longer than 30 nucleotides . ( A ) Expressed at low levels: transcripts expressed at 0 . 5–2 FPKM , ( B ) expressed at high levels: transcripts expressed at 2–10 FPKM . codRNAs were sampled in such a way as to have the same gene expression distribution as the corresponding lncRNA set . Results for species in which all sets contained at least 20 transcripts are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 015 Next , we focused our attention on the differences between the primary ORF and the 5′UTR and 3′UTR regions in codRNAs and lncRNAs . We defined the 3′ untranslated region ( 3′UTR ) as the sequence located immediately after the STOP codon of the primary ORF or the most downstream ORF associated with ribosomes . We used the same criteria to define the 5′UTR upstream from the initiation codon . In this analysis , we included all transcripts containing at least one ORF associated with ribosomes ( the primary ORF ) and sufficiently long UTR regions as to detect ribosome profiling reads ( >30 nucleotides ) ; insufficient data for fruit fly and yeast precluded the analysis for these species . In both codRNAs and lncRNAs , the 5′UTR showed a ribosome density ( translational efficiency , TE ) comparable to that of the primary ORF ( Figure 5B ) . In contrast , the 3′UTR showed very little ribosome association and often we could not find a single read mapping to this region ( 31–91% of cases in codRNAs and 46–68% in lncRNAs ) . Using genes with a single isoform or considering only annotated transcripts produced similar results ( Figure 5—figure supplements 2 and 3 ) . We also controlled for expression level by dividing the data set in transcripts with low ( 0 . 5–2 FPKM ) and high expression ( >2 FPKM ) , and by sampling the codRNAs in such a way as to have a similar expression distribution as lncRNAs . The results were very similar to those obtained with the complete data set ( Figure 5—figure supplement 4 ) , indicating that the analysis is robust to transcript expression differences . As transcripts may contain several ORFs , we performed a separate analysis in which we compared the translational efficiency of the primary ORF , any additional ORFs with mapped ribosome profiling reads , and the regions between ribosome-protected ORFs ( interORF ) ( Figure 5C ) . InterORF regions showed little signal when compared to the primary ORF , both in codRNAs and lncRNAs ( Wilcoxon test , p < 10−9 in human , mouse , and zebrafish , p < 0 . 05 in Arabidopsis , insufficient data for fruit fly and yeast precluded the analysis for these species ) . The data also indicated that ribosome binding is not always restricted to the primary ORF , especially in lncRNAs , as ribosome protection could sometimes be observed for additional ORFs . Taken together , these results indicate that lncRNAs have ribosome profiling signatures consistent with translation , with a strong decrease of ribosome density in the 3′UTR but not the 5′UTR region , and preferential binding of ribosomes to the primary ORF . There exists the possibility that the translated peptides are degraded soon after being produced . However , we estimate that the percentage of cases that may undergo nonsense-mediated decay ( NMD , see ‘Materials and methods’ for more details ) is low , between 4 . 47 and 14 . 11% depending on the species . For comparison , the percentage for protein-coding transcripts showing the same patterns ( including transcripts annotated as NMD in Ensembl ) is between 0 . 34 and 13 . 33% . Are the putatively translated ORF in lncRNAs conserved ? We performed sequence similarity searches using BLASTP ( E-value < 10−4 ) against all annotated coding transcripts in Ensembl , as well as against the primary ORFs in lncRNAs , for the six species studied here ( Supplementary files 1D and 2B ) . The number of lncRNA_ribo with homologues in other species was remarkably low ( 0–15 . 6% ) except for zebrafish ( 49 . 4% ) . In contrast , the majority of codRNAs had homologues in other species ( >95% for vertebrates and fruit fly and 70–73% for Arabidopsis and yeast ) . After we discarded lncRNAs that showed cross-species conservation , association with ribosomes was still very prevalent ( 80 . 4% of mouse , 40 . 3% of human , and 22 . 1% of zebrafish lncRNAs were associated with ribosomes ) . We also investigated whether the ribosome-associated ORFs in lncRNAs showed homology to annotated proteins in the same species . The values were very low for all the species ( 0–12 . 4% ) except for zebrafish ( 47 . 5% ) . Therefore , in general lncRNAs are not truncated duplicated copies ( pseudogenes ) . The case of zebrafish is an exception probably because of missing protein-coding annotations in this species . Subsequently , we compared the sequence coding properties of the primary ORF in lncRNAs with those in bona fide coding and non-coding sequences using a hexamer-based coding score ( see ‘Materials and methods’ ) . In all species the coding scores of the primary ORF in lncRNAs , while lower than that of codRNAs , were significantly higher than the coding score of ORFs in introns ( Figure 6 , Wilcoxon test lncRNA_ribo vs intron , human , mouse , zebrafish , and Arabidopsis p < 10−16; fruit fly and yeast p < 10−5 ) . This clearly shows that ORFs in lncRNAs are more coding-like than random ORFs . We repeated the same comparison using 100 different randomly sampled intronic sequence sets , and in >95% of the cases , we obtained the same result . lncRNAs associated with ribosomes ( lncRNA_ribo ) showed higher coding scores than those not associated with ribosomes ( lncRNA_noribo ) , even when we did not use the ribosome profiling information and compared the longest ORF in both types of transcripts ( Figure 6—figure supplement 1 ) . We reached similar conclusions when we restricted the analysis to annotated lncRNA transcripts ( Figure 6—figure supplement 2 ) , when we used ORFs from gene deserts as an alternative non-coding sequence set ( differences with lncRNAs significant by Wilcoxon test , p < 10−16 , see ‘Materials and methods’ for more details ) , and when we restricted the analysis to lncRNAs for which we did not find protein coding homologues in the other species studied ( Figure 6—figure supplement 3 ) . Because a high proportion of lncRNAs contained small ORFs , we repeated the comparison only considering transcripts with ORFs shorter than 100 amino acids to avoid any length biases , again obtaining similar results ( Figure 6—figure supplement 4 ) . The use of other coding scores , for example based on codon frequencies instead of hexamer frequencies or related metrics such as GC content produced consistent results ( Figure 6—figure supplement 5; Supplementary file 1E ) . 10 . 7554/eLife . 03523 . 016Figure 6 . Coding scores in ORFs from different types of transcripts . Intron: randomly selected intronic regions; lncRNA_noribo: lncRNAs not associated with ribosomes; lncRNA_ribo: lncRNAs associated with ribosomes; pseudogene: pseudogenes associated with ribosomes; codRNAne: coding transcripts encoding non-validated proteins associated with ribosomes; codRNAe: coding transcripts encoding experimentally validated proteins . The coding score was calculated as the log ratio of hexamer frequencies in coding vs intronic sequences . In lncRNA_noribo and introns , we considered the longest ORF and in the rest of transcripts the primary ORF . The Class ‘pseudogene’ was only included in species with more than 20 expressed pseudogenes with mapped ribosome profiling reads . The coding score of the primary ORF in lncRNAs ( lncRNA_ribo ) was significantly higher than the coding score in ORFs defined in introns ( Wilcoxon test , human , mouse , zebrafish , and Arabidopsis p < 10−16; fruit fly and yeast p < 10−4 , Wilcoxon test ) and in lncRNA_ribo it was significantly higher than in lncRNA_noribo in four species ( Wilcoxon test , human , mouse and zebrafish p < 10−5 , and Arabidopsis p < 0 . 05 ) . Transcripts from genes of different evolutionary age were taken from the literature ( see manuscript text ) . The number of transcripts was 68 for rodent , 127/123 for mammalian ( mouse/human as reference species ) , 11 , 203/13 , 423/9812 for metazoan ( mouse/human/zebrafish ) , 162 for fish , 208 for Crucifera , 28 for S . cerevisiae and 84 for Saccharomyces . The youngest class of codRNAs displayed similar scores than lncRNA_ribo in mouse , zebrafish , and yeast ( classes rodent , fish and S . cerevisiae , respectively ) , being only significantly higher in human and Arabidopsis ( Wilcoxon test , p < 0 . 005; classes primate and Cruciferae ) . We did not analyze young genes in fruit fly due to lack of a suitable young set of codRNAs in this species . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 01610 . 7554/eLife . 03523 . 017Figure 6—figure supplement 1 . Coding scores for the longest ORF . Comparison between lncRNAs associated and not associated with ribosomes using the longest ORF in both cases ( lncRNA_ribo and lncRNA_noribo , respectively ) . Differences between lncRNA_ribo and lncRNA_noribo are significant by a Wilcoxon test ( p < 10−10 in human , mouse , and zebrafish; p < 0 . 005 in Arabidopsis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 01710 . 7554/eLife . 03523 . 018Figure 6—figure supplement 2 . Coding scores in different classes of annotated sequences . Comparison between different transcript classes using only annotated lncRNAs . Yeast transcriptome is composed of very few annotated lncRNAs , and this analysis could not be performed . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 01810 . 7554/eLife . 03523 . 019Figure 6—figure supplement 3 . Coding scores in lncRNAs without homologues in other species . Comparison between different transcript classes using only lncRNA with no homologues ( noH ) in other species . Only species in which several lncRNA_ribo and lncRNA_noribo had homology matches were considered . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 01910 . 7554/eLife . 03523 . 020Figure 6—figure supplement 4 . Coding scores in small ORFs from different types of transcripts . Here we only employed lncRNAs in which the primary ORF was shorter than 100 amino acids . codRNA refers to joined codRNAe and codRNAne sets , since experimentally verified proteins are usually longer than 100 amino acid . The number of transcripts is shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 02010 . 7554/eLife . 03523 . 021Figure 6—figure supplement 5 . Use of different coding statistics in human transcripts . Equal dicodon was based on the observed hexamer frequencies in coding sequences vs hexamer equiprobability , intron dicodon was based on the differences between hexamer frequencies in coding vs non-coding sequences and intron_monocodon was based on the observed codon frequences in coding sequences vs codon equiprobability . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 02110 . 7554/eLife . 03523 . 022Figure 6—figure supplement 6 . Ribosome protection patterns in transcripts containing short ORFs . ( A ) Mouse CUFF . 34338 . 1 ( chr5:113183493–113188347 ) is a novel lncRNA , it contains an ORF encoding a 169 amino acid protein associated with ribosomes and with protein-coding homologues in human , zebrafish , and yeast . ( B ) ENSMUST00000107081 is an annotated codRNA in mouse which evolved recently since no homologues were found in any other species . It has a small ORF that translates a 55 amino acid protein . ( C ) AT1G34418 . 1 is an annotated lncRNA in Arabidopsis showing abundant association with ribosomes in the 5′UTR region , the primary ORF ( 34 amino acid ) and the final region of the transcript , which contains two redundant ORFs ( in red ) coding the sequence: MGLGFVN ( V/F ) LLGM . RNAseq: profile of RNAseq reads . RPFs: profile of ribosome profiling reads . Exon-intron transcript structures are represented; the thickest boxes on the exons are the primary ORFs . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 022 At the individual transcript level , a sizeable fraction of lncRNAs associated with ribosomes displayed significantly higher coding scores than expected for non-coding sequences ( p < 0 . 05 in all 100 intronic random sets; data in Supplementary file 2C; examples in Figure 6—figure supplement 6 ) . These transcripts are comprised of 143 human lncRNAs ( 35 . 5% of the lncRNAs , score > 0 . 0189 ) , 137 mouse lncRNAs ( 35 . 1% , score > 0 . 0377 ) , 379 zebrafish lncRNAs ( 52 . 1% score > 0 . 0095 ) , 7 fruit fly lncRNAs ( 31 . 8% , score > −0 . 0483 ) , 43 Arabidopsis lncRNAs ( 46 . 2% , score > −0 . 0202 ) , and 5 yeast lncRNAs ( 83 . 3% , score > 0 . 03387 ) . Annotated and novel lncRNAs were present in similar proportions in these sets , supporting the validity of our strategy of merging the two types of transcripts from the beginning . We also noted that the fraction of lncRNAs with coding homologues in other species increased in these sets . For example , whereas the proportion of total human lncRNA_ribo with homologues in other species was 15 . 6% , in the set with significant coding scores it was 29 . 3% . This number increased to 57 . 3% when we performed searches against the NCBI non-redundant peptide database ‘nr’ , as some of the ORFs in lncRNAs are annotated as predicted peptides in this database . If ORFs in lncRNAs are being translated this is likely to be a relatively recent evolutionary event , as many lncRNAs are lineage-specific ( Pauli et al . , 2012; Necsulea et al . , 2014; our data ) . It is well established that proteins of different evolutionary age display distinct sequence properties , including different codon usage ( Toll-Riera et al . , 2009; Carvunis et al . , 2012; Palmieri et al . , 2014 ) . We retrieved sets of annotated protein-coding transcripts of different evolutionary age from human , mouse , zebrafish , Arabidopsis , and yeast available from various studies ( Ekman and Elofsson , 2010; Donoghue et al . , 2011; Neme and Tautz , 2013 ) and expressed in the systems studied here . We found that the coding score was always lower in the youngest group than in older groups ( Figure 6 , Wilcoxon test , p < 0 . 05 ) . Remarkably , the youngest codRNAs showed a very similar coding score distribution to lncRNAs ( Figure 6 ) . We obtained similar results when we discarded lncRNAs that had homologues in any of the other species ( Figure 6—figure supplement 3 ) . We also collected information from young protein coding genes encoding experimentally verified proteins according to Swiss-Prot ( Supplementary file 2D ) . We observed that these proteins were short and the ORF occupied a relatively small fraction of the transcript , features typically observed in lncRNAs . For example , the average size of proteins encoded by primate-specific transcripts was 148 amino acids and the average transcript coverage 47% . The coding score was remarkably low and again similar to that of lncRNAs ( median 0 . 008 for primate-specific human transcripts , 0 . 046 for rodent-specific mouse transcripts , and 0 . 089 for yeast-specific coding transcripts ) . An important measure of the strength of purifying selection acting on a coding sequence is the ratio between the number of non-synonymous and synonymous single nucleotide polymorphisms ( PN/PS ) . Given the nature of the genetic code , there are more possible non-synonymous mutations than synonymous mutations . Under neutrality ( no purifying selection ) , the PN/PS ratio is expected to be approximately 2 . 89 ( Nei and Gojobori , 1986 ) . Here , we applied the large amount of available polymorphism data for human , mouse , and zebrafish to compare the level of purifying selection in primary ORFs from codRNAs and lncRNAs ( Figure 7; Supplementary file 1F ) . In general , human sequences showed higher PN/PS ratios than sequences from the other analyzed species , probably due to the presence of many slightly deleterious mutations segregating in the population ( Eyre-Walker , 2002 ) . However , despite the intrinsic differences between organisms , we observed the same general trends . First , the PN/PS was significantly lower in codRNAs than in lncRNAs ( proportion test , p < 10−5 ) , denoting stronger purifying selection in the former . Second , there was a very clear inverse relationship between the strength of purifying selection and the age of the gene ( p < 10−15 between the youngest and rest of codRNAs in mouse and zebrafish ) , in agreement with previous studies ( Liu et al . , 2008; Cai et al . , 2009 ) . High PN/PS values were also observed in the subset of young genes encoding experimentally validated proteins in human ( primate-specific transcripts median PN/PS of 3 . 10 ) and mouse ( rodent-specific transcripts median PN/PS 1 . 42 ) , confirming this tendency . Third , the distribution of PN/PS values in lncRNAs was very similar to that of young protein-coding genes . In human and mouse , there were no significant differences , and in the case of zebrafish the lncRNAs had even slightly lower PN/PS values than the fish-specific protein coding genes ( p < 0 . 01 ) . 10 . 7554/eLife . 03523 . 023Figure 7 . Selective pressure in ORFs from different types of transcripts . PN/PS: ratio between the number of non-synonymous and synonymous single nucleotide polymorphisms ( SNPs ) in the complete set of primary ORFs for a given class of transcripts ( in lncRNA_noribo the longest ORF was considered ) . In blue , data for different coding and non-coding transcript classes . In brown , data for different age codRNA classes . The bars represent the 95% confidence interval for the PN/PS value . For the species not shown there was not sufficient data to perform this analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 03523 . 023
Here , we analyzed the patterns of ribosome protection in polyA+ transcripts from cells belonging to six different eukaryotic species . Among the expressed transcripts , we identified many lncRNAs in the different species . The vast majority of transcripts annotated as coding showed association with ribosomes ( >92% in all species ) . Remarkably , a very large number of transcripts annotated as long non-coding RNA ( lncRNAs ) also showed such association ( 30–82% depending on the data set ) . Considering that lncRNAs are typically much shorter and expressed at lower levels than codRNAs , which may hinder the identification of ribosome association , this is a very significant fraction . In addition , the patterns of ribosome protection along the transcript are similar to those of protein-coding genes . Therefore , many lncRNAs appear to be scanned by ribosomes and are likely to translate short peptides . Long non-coding RNAs are classified as such in databases because , according to a number of criteria , they are unlikely to encode functional proteins . These criteria include the lack of a long ORF , the absence of amino acid sequence conservation , and the lack of known protein domains ( Harrow et al . , 2012 ) . Moreover , we expect lncRNAs not to have matches to proteomics databases , as this should classify them as coding . Annotated lncRNAs are typically longer than 200 nucleotides because this is the cutoff size normally implemented to differentiate them from other RNA classes such as microRNAs and small nuclear RNAs . In practice , it is difficult to classify a transcript as coding or non-coding on the basis of the ORF size ( Dinger et al . , 2008 ) . Some true coding sequences may be quite small , and by chance alone non-coding transcripts may have relatively long ORFs . The majority of lncRNAs contain ORFs longer than 24 amino acids , which can potentially correspond to real proteins . Short proteins are more difficult to detect than longer ones and consequently they are probably underestimated in databases . In recent years , the use of comparative genomics ( Frith et al . , 2006; Ladoukakis et al . , 2011; Hanada et al . , 2013 ) , proteomics ( Slavoff et al . , 2013; Vanderperre et al . , 2013; Ma et al . , 2014 ) , and a combination of evolutionary conservation and ribosome profiling data ( Crappé et al . , 2013; Bazzini et al . , 2014 ) have shown that the number of short proteins is probably much higher than previously suspected ( Andrews and Rothnagel , 2014 ) . In yeast , gene deletion experiments have provided evidence of functionality for short open reading frames ( sORFs < 100 amino acids ) ( Kastenmayer et al . , 2006 ) ; in zebrafish , several newly discovered sORFs appear to be involved in embryonic development ( Pauli et al . , 2014 ) and other examples exist in insects ( Magny et al . , 2013 ) and humans ( Lee et al . , 2013; Slavoff et al . , 2014 ) . In many cases , the transcripts containing sORFs will be classified as non-coding , especially if the ORF is not well conserved across different species . One approach to identify potential coding transcripts is ribosome profiling ( Ingolia et al . , 2009 ) , which has been used to study translation of proteins in a wide range of organisms ( Guo et al . , 2010; Ingolia et al . , 2011; Brar et al . , 2012; Michel et al . , 2012; Chew et al . , 2013; Dunn et al . , 2013; Huang et al . , 2013; Artieri and Fraser , 2014; Bazzini et al . , 2014; Juntawong et al . , 2014; McManus et al . , 2014; Vasquez et al . , 2014 ) . In several of these studies it has been noted that lncRNAs can be protected by ribosomes ( Ingolia et al . , 2011; Chew et al . , 2013; Bazzini et al . , 2014; Juntawong et al . , 2014 ) . However , there is no consensus on whether the observed patterns are consistent with translation . For example in the original analysis of mouse stem cells , which we reanalyzed here , it was reported that many lncRNAs were polycistronic transcripts encoding short proteins ( Ingolia et al . , 2011 ) , but in another paper where the same data were processed in a different way , they concluded that lncRNAs were unlikely to be protein-coding ( Guttman et al . , 2013 ) . A zebrafish ribosome profiling study reported resemblance between lncRNAs and 5′leaders of coding RNAs; the authors suggested that translation may play a role in lncRNA regulation ( Chew et al . , 2013 ) . Nevertheless , in the same study dozens of lncRNAs were proposed to be bona fide protein-coding transcripts . In Arabidopsis , the translational efficiency values of highly expressed lncRNAs ( >5 FPKM ) were similar to those of coding RNAs and some lncRNAs had profiles consistent with initiation and termination of translation ( Juntawong et al . , 2014 ) . Finally , using yeast data , Wilson and Masel . ( 2011 ) found many cases of non-coding transcripts bound to ribosomes and suggested that this facilitates the evolution of novel protein-coding genes from non-coding sequences . The disparity of results obtained in different systems motivated us to retrieve the original data and perform exactly the same analyses for six different species . As lncRNA catalogues are still very incomplete for most species , we also defined sets of novel lncRNAs using the RNA-seq sequencing reads for de novo transcript assembly . We discovered many novel , non-annotated , lncRNAs , especially in zebrafish , mouse , and fruit fly ( Table 2 ) . After the analysis of the ribosome profiling data , the same general picture emerged for the different biological systems , indicating that we are detecting very fundamental properties . In transcripts classified as lncRNAs , the ribosome profiling reads tend to cover a smaller fraction of the transcript than in typical codRNAs , in agreement with a shorter relative size of the ORF accumulating the largest number of ribosome profiling reads ( primary ORF ) . We also find that the translational efficiency of regions corresponding to the primary ORF is much higher than that of 3′UTRs , both in codRNAs and lncRNAs , consistent with translation of the transcripts . Furthermore , the primary ORF of lncRNAs showed significantly higher coding score than the longest ORF extracted from randomly selected non-coding regions . lncRNAs often contain several potentially translated ORFs ( Ingolia et al . , 2011 ) . Transcripts encoding multiple short proteins have been reported in insects ( Savard et al . , 2006 ) and could be common in other species as well ( Tautz , 2009 ) . One such candidate is AT1G34418 . 1 in Arabidopsis , an annotated lncRNA which contains a primary ORF followed by two instances of a 12 amino acid ORF also covered by ribosome profiling reads ( Figure 6—figure supplement 6 ) . This case is reminiscent of the gene pri in fruit fly , which regulates tarsal development ( Galindo et al . , 2007 ) and translates several small redundant ORFs ( Kondo et al . , 2007 ) . lncRNAs are poorly conserved across species and so , if translated , they will produce species- or lineage-specific proteins . Recently evolved proteins are markedly different from widely distributed ancient proteins; they are shorter , subject to weaker selective constraints and expressed at lower levels ( Albà and Castresana , 2005; Cai et al . , 2009; Liu et al . , 2010; Donoghue et al . , 2011; Carvunis et al . , 2012; Xie et al . , 2012; Wissler et al . , 2013; Neme and Tautz , 2014 ) . Here for the first time , we have compared the properties of the ORFs in lncRNAs associated with ribosomes with the properties of annotated , and in some cases experimentally validated , young protein-coding genes . lncRNAs and young protein-coding transcripts are virtually indistinguishable regarding their coding score and ORF selective constraints ( Figures 6 and 7 ) , which is consistent with the idea that many lncRNAs encode new peptides . Although it is unclear how many of these peptides are functional , the data indicate that at least a fraction of them may be functional . Sequences that translate functional proteins are expected to display signs of selection related to preferential usage of certain amino acids and codons . This can be used to differentiate between coding and non-coding entities , especially in the absence of cross-species conservation , as is the case of many lncRNAs . About 35–40% of primary ORFs in human and mouse lncRNAs displayed coding scores that were significantly higher than those expected for non-coding sequences , making them excellent candidates for translating functional proteins . In fact , five human lncRNAs associated with ribosomes that exhibited high coding scores in our study were re-annotated as protein-coding transcripts in a subsequent Ensembl gene annotation release ( version 75 , Supplementary file 2C ) . Gene knock-out experiments in fly have discovered that young proteins , even if rapidly evolving , are often essential for the organism and can cause important defects when deleted ( Chen et al . , 2010; Reinhardt et al . , 2013 ) . Similarly , some peptides translated from lncRNAs may have important cellular functions yet to be discovered . lncRNAs tend to be expressed at much lower levels than typical codRNAs , so , everything else being equal , the amount of translated peptide is also expected to be smaller . It may be that some of these peptides are not functional , but their translation does not produce a large enough deleterious effect for them to be eliminated via selection . Pseudogenes also showed extensive association with ribosomes in our study , indicating that the translation machinery is probably not very selective or that some pseudogenes produce functional proteins . This question may be worth revisiting , as a recent proteomics study has also found that dozens of human pseudogenes produce peptides ( Kim et al . , 2014 ) . The data also indicate that a fraction of lncRNAs have not acquired the capacity to be translated . Depending on the experiment analyzed , a number of lncRNAs did not show any significant association with ribosomes . As previously discussed , this is probably affected by a lack of sensitivity; it is also true that the lncRNAs not associated with ribosomes tended to show lower coding scores than lncRNAs associated with ribosomes , even when we did not use the ribosome profiling data and simply compared the longest ORF in both kinds of transcripts . Recently , it has been reported that human-specific protein-coding genes are often related to non-coding transcripts in macaque , pointing to a non-coding origin for many newly evolved proteins ( Xie et al . , 2012 ) . More generally , one may view de novo protein-coding gene evolution as a continuum from non-functional genomic sequences to fully-fledged protein-coding genes ( Albà and Castresana , 2005; Toll-Riera et al . , 2009; Carvunis et al . , 2012 ) . Therefore , many lncRNAs could be in intermediate states in this process , their pervasive translation serving as the building material for the evolution of new proteins . It may be difficult to obtain functional proteins from completely random ORFs ( Jacob , 1977 ) , but the effect of natural selection preventing the production of toxic peptides ( Wilson and Masel , 2011 ) , and the high number of transcripts expressed in the genome , may facilitate this process .
We downloaded the original data from Gene Expression Omnibus ( GEO ) for six different ribosome profiling experiments that had both ribosome footprinting and polyA+ RNA-seq sequencing reads: mouse ( M . musculus ) ( Ingolia et al . , 2011 ) , human ( H . sapiens , HeLa cells ) ( Guo et al . , 2010 ) , zebrafish ( D . rerio ) ( Chew et al . , 2013 ) , fruit fly ( D . melanogaster ) ( Dunn et al . , 2013 ) , Arabidopsis ( A . thaliana ) ( Juntawong et al . , 2014 ) , and yeast ( S . cerevisiae ) ( McManus et al . , 2014 ) . We retrieved genome sequences and gene annotations from Ensembl v . 70 and Ensembl Plants v . 21 ( Flicek et al . , 2012 ) . Raw ribosome and RNA-seq sequencing reads underwent quality filtering using Condentri ( v . 2 . 2 ) ( Smeds and Künstner , 2011 ) with the following settings ( -hq=30 –lq=10 ) . Adaptors described in the original publications were trimmed from filtered reads if at least five nucleotides of the adaptor sequence matched the end of each read . In zebrafish , reads from different developmental stages were pooled to improve read coverage . In all experiments , reads below 25 nucleotides were not considered . Clean ribosome short reads were filtered by mapping them to the corresponding species reference RNA ( rRNA ) using the Bowtie2 short-read alignment program ( v . 2 . 1 . 0 ) ( Langmead et al . , 2009 ) . Unaligned reads were aligned to a genomic reference genome with Bowtie2 allowing one mismatch in the first 'seed' region ( the length of this region was selected according to the descriptions provided in each individual experiment ) . RNA-seq short reads were mapped with Tophat ( v . 2 . 0 . 8 ) ( Kim et al . , 2013 ) to the corresponding reference genome . We allowed two mismatches in the alignment with the exception of zebrafish , for which we allowed three mismatches since the reads were significantly longer . Multiple mapping was allowed unless specifically stated . Expressed transcripts were assembled using Cufflinks ( v 2 . 2 . 0 ) ( Trapnell et al . , 2010 ) . We initially considered a transcript as expressed if it was covered by at least four reads and its abundance was higher than 1% of the most abundant isoform of the gene . We also discarded assembled transcripts in which >20% of reads were mapped to several locations in the genome . Gene annotation files from Ensembl ( gtf format , v . 70 ) were provided to Cufflinks to guide the reconstruction of already annotated transcripts . Annotated transcripts were divided into coding RNAs and long non-coding RNAs ( lncRNAs ) , we only considered lncRNAs that were not part of genes with coding transcripts . Novel isoforms corresponding to annotated loci were not analyzed . Transcripts that did not match or overlapped annotated genes were labeled 'novel’ lncRNAs . We used a length threshold of 200 nucleotides to select novel long non-coding RNAs , as in ENCODE annotations ( Djebali et al . , 2012 ) . Strand directionality of multiexonic transcripts was inferred using the splice site consensus sequence . We only considered monoexonic transcripts in the case of Arabidopsis and yeast , provided the transcripts were intergenic . The inclusion of novel lncRNAs made it possible to perform analyses of species for which there are very few annotated lncRNAs . Annotations of UTR regions in yeast genes were missing from Ensembl because of the variability observed in transcription start sites ( TSS ) . However , we downloaded a set of available 5′ and 3′UTRs obtained by deep transcriptomics ( Nagalakshmi et al . , 2008 ) and added them to the existing yeast Ensembl annotations before assembling the transcriptome . Coding transcripts were classified into different subclasses depending on the existing annotations: ( a ) Annotated protein-coding transcripts ( codRNA ) , ( b ) Annotated transcripts with surveillance mechanisms ( nonsense mediated decay , nonstop mediated decay , and no-go decay ) , ( c ) Annotated pseudogenes . We removed protein-coding transcripts in which annotated coding sequences ( CDS ) are still incomplete . Subsequently , we defined an additional subset of annotated protein-coding transcripts with well-established coding properties based on the existence of an experimentally verified protein in Swiss-Prot for the gene ( ‘evidence at protein level’ , downloaded 29 October 2013 , UniProt Consortium , 2014 ) . These transcripts were labeled codRNAe . The rest of annotated protein-coding transcripts were abbreviated codRNAne . In zebrafish , most proteins are not yet experimentally validated; and therefore , we generated a single group . We built a data set of human lncRNAs with described non-coding functions using data obtained from several recent reviews ( Ponting et al . , 2009; Ulitsky and Bartel , 2013; Fatica and Bozzoni , 2014 ) . This data set included 29 different genes ( Supplementary file 2A ) . We used cufflinks to estimate the expression level of a transcript in FPKM units ( Fragments Per Kilobase per total Million mapped reads ) . We used a threshold of >0 . 5 FPKM except in yeast , in which the average read coverage per transcript was much higher than in the other species and the threshold was set up at >5 FPKM . These thresholds guaranteed detection of ribosome association for the majority of expressed coding transcripts ( >92% ) , while yielding proportions of transcripts comparable to those reported in the original papers . We predicted all possible open reading frames ( ORFs ) in the expressed transcripts . We defined an ORF as any sequence starting with an AUG codon and finishing with a stop codon ( TAA , TAG , or TGA ) , and at least 75 nucleotides long . This would correspond to a 24 amino acid protein , which is the size of the smallest complete human polypeptide found in genetic screen studies ( Hashimoto et al . , 2001 ) . This ORF definition will not detect non-canonical ORFs with different start or stop codons , although these ORFs often correspond to regulatory ORFs ( uORFs ) in the 5′UTR region . In monoexonic transcripts ( Arabidopsis and yeast ) , we considered all six possible different frames . We also defined each transcript 5′UTR as the region between the transcription start site and the AUG codon from the left-most predicted ORF , and the 3′UTR the region from the stop codon in the right-most predicted ORF to the transcript end . UTRs with lengths below 30 nucleotides were not analyzed since ribosome reads could not be properly aligned to these regions due to their small size . Regions between two consecutive putatively translated ORFs ( with ribosome profiling reads ) were termed interORF . We only analyzed this region when the length of the interORF sequence in a transcript was 30 nucleotides or longer . We defined a set of bona fide non-coding sequences sampled from intronic fragments . We used the introns of the genes expressed in each experiment , provided they did not overlap to any exons from other overlapping genes . We randomly selected fragments in such a way as to simulate the same size distribution as in the complete set of expressed transcripts . We performed 100 simulations of intron sampling to ensure the results were robust to the randomization process . We selected the longest ORF in each intronic fragment for the calculation of coding scores and GC content . We computed the number of reads overlapping each feature of interest ( transcript , UTR , ORF , and interORF ) using the BEDTools package ( v . 2 . 16 . 2 ) ( Quinlan and Hall , 2010 ) . We only considered ribosome reads in which more than half of their length spanned the considered region . This was considered appropriate because the ribosome P-site is usually detected at the central region of the read , with only slight variations depending on the experimental setting . We set up a minimum ribosome profiling coverage of 75 nucleotides per transcript to define the transcript or transcript region ( e . g . , ORF ) as associated with ribosomes . This is significantly longer than the length of the ribosome profiling sequencing reads ( 36–51 nucleotides ) and is consistent with the minimum ORF length threshold . The translational efficiency ( TE ) of a sequence has been previously defined as the density of ribosome profiling ( RPF ) reads normalized by transcript abundance ( Ingolia et al . , 2009 ) . We calculated it by dividing the FPKM of the ribosome profiling experiment by the FPKM of the RNA-seq experiment . In transcripts , we also obtained the maximum TE by dividing the sequence in 90 nucleotide windows and selecting the window with the highest TE value . In order to have a null model of ribosome binding against which to compare the ribosome profiling signal in codRNA and lncRNA transcripts , we extracted annotated 3′ untranslated regions ( 3′UTRs ) from codRNAs in genes in which UTRs did not overlap with coding sequences from other transcripts , and by randomly selecting 3′UTRs with a minimum length of 30 nucleotides , we built a set of 3′UTR sequences with the same size distribution as the complete transcripts . For each species , we calculated the TE values for codRNAs , lncRNA , and 3′UTR sequences . We used the empirical distribution of TE values in the 3′UTRs to calculate the number of codRNAs and lncRNAs that showed significantly higher TE value than expected under the null model at a p < 0 . 05 . These corresponded to TE values higher than 0 . 1043 in mouse , 0 . 2556 in human , 0 . 0004 in zebrafish , 0 . 7164 in fruit fly , 0 . 1800 in Arabidopsis , and 0 . 0527 in yeast . We defined the primary ORF in a transcript as the ORF with the largest number of RPF reads with respect to the total RPF reads covering the transcript . The rest of ORFs ≥24 amino acids associated with ribosomes were considered as well; when two or more ORFs overlapped , we selected the longest one . In ORFs , interORFs , and UTRs , we computed the TE along the whole region . For comparing the TE in different regions , we only considered transcripts in which all regions had >0 . 2 FPKM . We downloaded all peptide sequences from the PeptideAtlas database: 338 , 013 human peptides ( August 2013 ) , 101 , 695 mouse peptides ( June 2013 ) , and 86 , 836 yeast peptides ( March 2013 ) . We investigated if the number of ribosome-associated protein-coding transcripts that matched the peptides in these databases varied with protein length . We omitted this analysis in zebrafish and Arabidopsis due to the lack of sufficiently large peptide databases . The matches were identified using BLASTP searches ( v . 2 . 2 . 28+ ) ( Altschul et al . , 1997 ) . We selected perfect matches only . We investigated how many primary ORFs may be candidates for being regulated via non-sense mediated decay ( NMD ) surveillance pathways , whose main function is to eliminate transcripts containing premature stop codons . We defined NMD candidates as all cases in which the stop-codon from a predicted ORF was located ≥55 nucleotides upstream of a splice junction site , provided the stop-codon was not in the terminal exon ( Scofield et al . , 2007 ) . This mechanism is well characterized in protein-coding genes and it has been proposed as a way to degrade non-functional peptides translated in lncRNAs ( Tani et al . , 2013 ) . Other surveillance mechanisms , such as non-stop-mediated decay or no-go decay , were not considered since all predicted ORFs finished at a stop codon , and we did not analyze RNA secondary structures . We utilized existing gene age classifications in human , mouse , and zebrafish ( Neme and Tautz , 2013 ) to identify young gene classes: human primate-specific ( ∼55 . 8 My ) , mouse rodent-specific ( ∼61 . 7 My ) , human and mouse mammalian-specific ( ∼225 My ) , zebrafish actinopterygii-specific ( ∼420 My ) ( abbreviated fish ) and metazoan ( ∼800 My ) . In yeast , we used predefined genes specific to S . cerevisiae ( 1–3 My ) ( abbreviated S . cerevisiae ) and the Saccharomyces group ( ∼100 My ) ( Ekman et al . , 2007 ) . In Arabidopsis , we retrieved Cruciferae ( Brassicaceae ) -specific genes ( 20–40 My ) ( Donoghue et al . , 2011 ) . These genes are believed to have arisen primarily by de novo mechanisms , as no homologies in other species have been detected despite the fact that many closely related genomes have now been sequenced . In humans , we obtained a set of gene desert sequences as defined in Ovcharenko et al . ( 2005 ) . We selected two stable and two flexible gene deserts ( the definition depends on the degree of conservation in other species ) . They belonged to chromosome 4 ( flexible located in coordinates 136 , 000 , 001–138 , 000 , 000; stable located in coordinates 180 , 000 , 001–182 , 000 , 010 ) that has a high number of gene deserts; and chromosome 17 ( flexible located in coordinates 51 , 100 , 001–51 , 900 , 000; stable located in coordinates 69 , 300 , 001–70 , 000 , 000 ) that has a high gene density . We ensured that no protein-coding genes were annotated in subsequent Ensembl versions in these regions . We predicted all possible ORFs in these regions and evaluated their coding score and GC content . The examination of nucleotide hexamer frequencies has been shown to be a powerful way to distinguish between coding and non-coding sequences ( Sun et al . , 2013; Wang et al . , 2013 ) . We computed one coding score ( CS ) per hexamer:CShexamer ( i ) =log ( freqcoding ( hexamer ( i ) ) freqnon−coding ( hexamer ( i ) ) ) . The coding hexamer frequencies were obtained from the open reading frame of all transcripts in a species encoding experimentally validated proteins ( except for zebrafish in which all protein-coding transcripts were considered ) . The non-coding hexamer frequencies were calculated using the longest ORF in intronic regions , which were selected randomly from expressed protein-coding genes . Next , we used the following statistic to measure the coding score of an ORF:CSORF=∑i=1i=nCShexamer ( i ) n , where i is each sequence hexamer in the ORF , and n the number of hexamers considered . The hexamers were calculated in steps of three nucleotides in frame ( dicodons ) . We did not consider the initial hexamers containing a Methionine or the last hexamers containing a STOP codon , since they are not informative . Given that all ORFs were at least 75 nucleotides long the minimum value for n was 22 . We calculated other related statistics in a similar way . This included using an equiprobable hexamer distribution instead of the distribution obtained from non-coding sequences , or using codon frequencies instead of hexamer frequencies . These statistics showed somewhat lower power to distinguish between coding and non-coding sequences . As a complementary measure , we quantified the GC content in different coding and non-coding transcripts and ORFs . We employed BLASTP with an E-value cutoff of 10−4 to compare the amino acid sequences encoded by ORFs in different kinds of transcripts . We enabled SEG to mask low complexity regions in protein sequences before doing the homology searches . We also searched for homologues in the NCBI non-redundant ( nr ) protein database ( Pruitt et al . , 2014 ) . BLAST sequence similarity search programs are based on gapped local alignments ( Altschul et al . , 1997 ) . We downloaded all available single-nucleotide polymorphisms ( SNPs ) from dbSNP ( Sherry et al . , 2001 ) for human ( ∼50 million ) , mouse ( ∼64 . 2 million ) , and zebrafish ( ∼1 . 3 million ) . We did not consider other species due to insufficient data for the analysis . We classified SNPs in ORFs as non-synonymous ( PN , amino acid altering ) and synonymous ( PS , not amino acid altering ) . We computed the PN/PS ratio in each sequence data set by using the sum of PN and PS in all sequences . The estimation of PN/PS ratios of individual sequences was in general not reliable due to lack of sufficient SNP data . We obtained confidence intervals using the proportion test in R ( see below ) . The analysis of the data , including generation of plots and statistical tests , was done with R ( R Development Core Team , 2010 ) . Supplementary file 1 contains additional Tables and Supplementary file 2 data subsets . The genomic coordinates of all transcripts used in this study ( GTF files ) and the amino acid sequences corresponding to primary ORFs in lncRNA with coding scores significant at p < 0 . 05 ( FASTA files ) are available at figshare ( http://dx . doi . org/10 . 6084/m9 . figshare . 1114969 ) . | Despite the terms being largely interchangeable in modern language , ‘DNA’ and ‘gene’ do not mean the same thing . A gene is made of DNA and contains the instructions to make a protein , and it is the protein that performs the function of the gene . However , cells in the body also contain DNA that does not form genes . Far from being ‘junk’ DNA with no biological purpose; this DNA has a variety of roles , including affecting how other genes are used . To produce a protein , the DNA sequence of a gene is transcribed into an intermediate molecule called RNA , which is then translated to produce a protein . So-called long non-coding RNA ( lncRNA ) molecules are also transcribed from DNA , but whether these are translated to make proteins has been a subject of much debate . Indeed , the function of the vast majority of lncRNA molecules is unknown . Ruiz-Orera et al . analyzed RNA sequences collected from earlier experiments on six different species—humans , mice , fish , flies , yeast , and a plant—and found nearly 2500 as yet unstudied lncRNAs in addition to those previously identified . Many of the lncRNAs that Ruiz-Orera et al . investigated could be found lodged inside the cellular machinery used to translate RNA into proteins . Furthermore , these lncRNA molecules are oriented in the machinery as if they are primed and ready for translation , suggesting that many lncRNAs do produce proteins . However , it is unclear how many of these proteins have a useful function . Very few lncRNAs were found in more than one species , suggesting that they have evolved recently . The properties of lncRNA molecules also show many similarities with the properties of ‘young’—recently evolved—genes that are known to produce proteins . The combined findings of Ruiz-Orera et al . therefore suggest that lncRNAs are important for developing new proteins . The emergence of proteins with new functions has been an important driving force in evolution , and this work provides important clues into the first steps of this process . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology"
] | 2014 | Long non-coding RNAs as a source of new peptides |
Plant cells divide using the phragmoplast , a microtubule-based structure that directs vesicles secretion to the nascent cell plate . The phragmoplast forms at the cell center and expands to reach a specified site at the cell periphery , tens or hundreds of microns distant . The mechanism responsible for guiding the phragmoplast remains largely unknown . Here , using both moss and tobacco , we show that myosin VIII associates with the ends of phragmoplast microtubules and together with actin plays a role in guiding phragmoplast expansion to the cortical division site . Our data lead to a model whereby myosin VIII links phragmoplast microtubules to the cortical division site via actin filaments . Myosin VIII's motor activity along actin provides a molecular mechanism for steering phragmoplast expansion .
To divide , plant cells must build a new cell wall . This is accomplished by a dynamic and complex structure known as the phragmoplast , comprising the cytoskeletal polymers , microtubules and actin filaments . The phragmoplast assembles from the anaphase spindle and directs the traffic of vesicles carrying cell wall components to and from the nascent wall , called the cell plate ( Seguí-Simarro et al . , 2004; Austin et al . , 2005 ) . The phragmoplast forms at the center of the cell and expands outward to reach the parental cell wall , tens or even hundreds of microns distant . When the phragmoplast reaches the parental cell wall , the cell plate and parental membranes fuse , completing cytokinesis . Active guidance is required during phragmoplast expansion . However , the molecular basis for this steering has been elusive . It is clear that microtubules are essential for cell division , since in their absence the cell plate does not form . And while it has been known for decades that the phragmoplast also contains actin filaments ( Clayton and Lloyd , 1985; Kakimoto and Shibaoka , 1987 ) , it remains unclear how actin contributes to phragmoplast function . For one , plant cells still divide in the absence of actin ( Baluska et al . , 2001; Nishimura et al . , 2003 ) . Additionally division still occurs in the presence of mutations in genes encoding actin or various actin-associated proteins ( Jürgens , 2005 ) . However , based on drug treatments and localization studies , actin has been proposed to stabilize the phragmoplast and link the phragmoplast to the cell cortex ( Lloyd and Traas , 1988; Molchan et al . , 2002 ) . But beyond implicating actin in a steering mechanism somehow , these studies have provided little if any mechanistic details . Specifically dissecting the role of phragmoplast actin is further complicated because actin is also present at the preprophase band ( Pickett-Heaps and Northcote , 1966; Kakimoto and Shibaoka , 1987; Palevitz , 1987 ) , a microtubule-based structure that is present on the cell cortex just prior to nuclear envelope breakdown . As mitosis proceeds , the microtubules in the band disassemble while actin filaments become depleted from the band itself but enriched on either side of it ( Cleary , 1995 ) . Although the cytoskeletal polymers are lost from the band , a number of proteins remain at the site of the band throughout mitosis and cytokinesis , thereby marking the site where the new cell plate will fuse to the parental cell wall ( Walker et al . , 2007; Xu et al . , 2008 ) . Interfering with preprophase band development invariably interferes with cell plate positioning ( Rasmussen et al . , 2011b ) . Because actin is present in both the band and the phragmoplast , discovering actin's function specifically in the latter has been challenging . However , not all dividing plant cells have a preprophase band . Moss spores germinate into a branched network of filaments , known as protonemata . All dividing cells , both apical and branching , divide without benefit of a preprophase band ( Doonan et al . , 1985 ) . While depolymerization of the actin cytoskeleton halts cell expansion in protonemata , it has little if any effect on cell division . The fact that moss protonemata do not make a preprophase band , but have actin in the phragmoplast provides a unique opportunity to study the role of actin in phragmoplast guidance . Here , we use a combination of genetics and live-cell imaging to probe the role for guiding the phragmoplast of actin and a family of actin-based molecular motors , the class VIII myosins .
Physcomitrella patens has five identified class VIII myosin genes , named myo8A through E . Taking advantage of facile homologous recombination in this species , Wu et al . ( 2011 ) constructed a line in which all five genes were disrupted ( Δmyo8ABCDE ) . Protonemata from this line have multiple , unevenly distributed branches . Upon further inspection , we found that cell plate placement at branch sites is often affected ( Figure 1A ) . Cell plates are aberrantly positioned with respect to the filament axis ( Figure 1A , arrows ) . Since branch patterning and cell division plane specification are linked , we reasoned that non-branching cells in the myosin VIII null plants might also have cell division defects . In young wild-type plants , apical cells position their new cell plates perpendicular to the long axis of the cell: more than 84% of apical cell plates are within 15° of the perpendicular plane . In contrast in myosin VIII null plants , less than 35% of the apical cell plates are within 15° of the perpendicular axis and nearly 40% have cell plates with angles greater than 25° , some as high as 45° ( Figure 1B ) . 10 . 7554/eLife . 03498 . 003Figure 1 . Cell plate defects in Δmyo8ABCDE can be restored by expression of Myo8A-GFP . ( A ) 10-day-old wild type and myosin VIII null plants stained with calcofluor . Scale bar , 100 µm . Arrows indicate mis-positioned cell plates . ( B ) Histograms of cell plate angles of apical cells from 5-day-old plants regenerated from protoplasts . Images of apical cells were acquired as in Figure 1A and cell plate angles were measured manually using ImageJ . Number of cells analyzed: wild type ( n = 151 ) , Δmyo8ABCDE ( n = 180 ) , Myo8A-GFP in Δmyo8ABCDE ( n = 167 ) . All distributions are significantly different from each other ( Wilcoxon-Mann-Whitney Rank Sum Test , p < 0 . 001 ) . ( C ) 8-day old plants regenerated from protoplasts were imaged with a stereo microscope . Scale bar , 100 µm . ( D ) Measurements of cell length were made on images of the apical cells from calcofluor stained 5 and 6-day old plants regenerated from protoplasts . Average apical cell lengths with standard deviation are indicated below each image . n indicates the number of cells measured . Scale bar , 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 003 To investigate how myosin VIII regulates cell plate positioning , we generated a construct encoding Myo8A fused to three tandem copies of monomeric enhanced GFP ( hereafter referred to as Myo8A-GFP ) and transformed Myo8A-GFP into the myosin VIII null plant . Since myosin VIII's are partially redundant ( Wu et al . , 2011 ) , we reasoned that expression of Myo8A should be sufficient to partially rescue the myosin VIII null phenotype . To test this , we measured cell plate positioning in young plants and found that expression of Myo8A-GFP results in plants with 63% of the apical cell plates within 15° of the perpendicular axis . Importantly , cell plates with angles greater than 35° are never observed in the Myo8A-GFP expressing plants ( Figure 1B ) , indicating that Myo8A-GFP partially restores cell plate positioning in the myosin VIII null plants . Additionally , Myo8A-GFP expression partially rescues a number of other defects in myosin VIII null plants , including protonemal branching defects ( Figure 1C ) , apical cell length ( Figure 1D ) , and timing of gametophore formation ( data not shown ) . Taken together , our data indicate that Myo8A-GFP is functional . Myo8A-GFP localizes diffusely and as small particles throughout the cytoplasm ( Figure 2A ) as well as at the cell cortex ( Figure 2B ) . In apical cells Myo8A-GFP particles are enriched at the cell tip ( Figure 2A ) . Using variable angle epifluorescence microscopy ( VAEM ) , we simultaneously imaged Myo8A-GFP and actin labeled with lifeact-mCherry . We observed that Myo8A-GFP cortical particles appear to move along actin filaments , consistent with actin-based motility ( Figure 2B; Video 1 ) . Movement was observed in both directions along an actin cable ( Figure 2C ) . We measured the velocity of the particles and found that Myo8A-GFP moves at 0 . 65 ± 0 . 57 µm/s ( n = 249 particles from 7 cells , Figure 2E , Figure 2—figure supplement 1 ) . In the presence of 25 µM latrunculin B ( LatB ) , which depolymerizes the actin cytoskeleton , Myo8A-GFP still localizes to the cell cortex but no longer exhibits directed motility ( Figure 2D; Video 2 ) , as expected for an active , actin-based molecular motor . 10 . 7554/eLife . 03498 . 004Figure 2 . Myo8A moves on cortical actin filaments . ( A ) Myo8A-GFP localizes to punctate structures throughout the cytosol as well as on the cell cortex . Images are maximum projections of z-stacks acquired with a spinning disc confocal . The punctate structures accumulate near the apex of the growing cell . Large globular structures are chloroplasts , which autofluorescence under these imaging conditions . Scale bar , 10 µm . ( B ) Images of Myo8A-GFP and Lifeact-mCherry in moss protonemata were simultaneously acquired with VAEM . In the merge Myo8A-GFP is green and Lifeact-mCherry is red . Scale bars , 2 µm . See also Video 1 . Yellow box indicates the enlarged area shown in ( C ) . Red line marks the trace for making the kymograph in ( C ) . ( C ) An example of Myo8A-GFP particle moving along actin filaments . Six consecutive frames with 76 ms time interval are shown . Arrowhead indicates the starting position of a Myo8A-GFP particle , and arrows indicate the last position of that Myo8A-GFP . In the last frame , a new Myo8A-GFP particle binds to the same position indicated by the arrowhead . Linear movement of Myo8A-GFP is evident in kymograph . Scale bar , 1 µm . Scale bar in t , 1 s . ( D ) Moss protonemal cells expressing Myo8A-GFP , were treated with or without 25 µm Latrunculin B ( LatB ) and imaged with VAEM . In control samples , Myo8A-GFP linear trajectories are apparent in a frame average of 25 frames from approximately 2 s of real time , but absent in cells treated with LatB . Scale bars , 2 µm . See also Video 2 . ( E ) Distribution of Myo8A-GFP velocities on actin filaments . Inset is a dot plot of the measured Myo8A-GFP velocities . Average velocity is 0 . 65 ± 0 . 57 µm/s; n = 249 events from seven cells . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 00410 . 7554/eLife . 03498 . 005Figure 2—figure supplement 1 . Myo8A moves on cortical actin filaments . Images of Myo8A-GFP and Lifeact-mCherry in moss protonemata were simultaneously acquired with VAEM . In the merge Myo8A-GFP is green and Lifeact-mCherry is red . Scale bars , 2 µm . Scale bar in t , 1 s . Red line marks the trace for making the kymographs shown below . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 00510 . 7554/eLife . 03498 . 006Video 1 . Myo8A moves on cortical actin filaments ( see Figure 2B ) . Myo8A-GFP ( green ) and Lifeact-mCherry ( red ) were simultaneously imaged with VAEM ( acquired at 13 fps ) . Video is playing at 15 fps . Scale bar , 2 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 00610 . 7554/eLife . 03498 . 007Video 2 . Myo8A motility at the cell cortex depends on actin ( see Figure 2D ) . Moss protonemal cells expressing Myo8A-GFP were imaged with VAEM continuously at 12 . 233 fps . Video is playing at 25 fps . Scale bar , 2 µm . Bottom panels show the same time series with a 25 frame rolling average . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 007 To investigate the role of myosin VIII in cell division , we first examined the localization of Myo8A-GFP in divisions that form branches since the myosin VIII null plants have a strong phenotype in branching cells . To stage division , we introduced into the Myo8A-GFP line mCherry-PpTUA1 ( hereafter referred to as mCherry-tubulin ) . In branch forming cells , Myo8A-GFP accumulates prominently at the cell cortex at the neck region of the emerging bulge , and this accumulation happens prior to prophase ( Figure 3 arrows; Figure 3—figure supplement 1; Video 3 ) . The tubulin images confirm that these cells lack a preprophase band . During branch cell mitosis , Myo8A-GFP appears on the spindle and phragmoplast , and as the phragmoplast matures , Myo8A-GFP accumulates at the phragmoplast periphery . In the later stages of cytokinesis , there are two populations of Myo8A-GFP: an inner ring on the phragmoplast ( Figure 3 , arrow heads; Figure 3—figure supplement 1 ) and an outer ring at the cell cortex ( Figure 3 , arrows; Figure 3—figure supplement 1 ) . As the phragmoplast expands , the inner Myo8A-GFP ring eventually reaches the outer one . Together with an increased frequency of cell plate positioning defects in myosin VIII null plants , these data suggest that myosin VIII plays a role in ensuring that the phragmoplast expands to the pre-determined cortical division site . 10 . 7554/eLife . 03498 . 008Figure 3 . Myosin VIII localizes to the phragmoplast and cortical division site in moss . A protonemal branching cell expressing Myo8A-GFP ( green ) and mCherry-tubulin ( red ) . Images are maximum intensity projections of z-stacks from a spinning disc time-series acquisition . Before mitosis , Myo8A-GFP accumulates at the neck of the emerging cell ( top , arrows ) . Myo8A-GFP accumulates at the spindle midzone ( middle , arrow heads ) and forms a ring at the edge of the phragmoplast that expands out to the cell cortex ( bottom ) . Scale bar , 10 µm . See also Video 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 00810 . 7554/eLife . 03498 . 009Figure 3—figure supplement 1 . Myo8A-GFP localizes to the cortical division site and the phragmoplast in moss . Images from a 3D rotation of the spinning disc confocal images presented in Figure 3 . Rotations along the plane of the phragmoplast equator show that in the early phragmoplast Myo8A-GFP is found throughout the phragmoplast midzone and the cell cortex . As the phragmoplast matures , Myo8A-GFP tightens into a ring along the leading edge of the phragmoplast . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 00910 . 7554/eLife . 03498 . 010Video 3 . Myo8A-GFP marks the future site of cell division in moss branching cells ( see Figure 3 ) . Moss branch cell expressing Myo8A-GFP ( green ) and mCherry-tubulin ( red ) was imaged with a spinning disc confocal microscope . Images are maximum projections of a z-stack acquired every minute . Video is playing at 4 fps . Scale bar , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 010 To investigate whether myosin VIII has a similar role in seed plants , we generated a tobacco BY-2 cell line stably transformed with the moss Myo8A-GFP . BY-2 cells are commonly used to study the localization of plant proteins during mitosis and cytokinesis ( Van Damme et al . , 2004; Rasmussen et al . , 2011a; Lipka et al . , 2014 ) . Alignment of myosin VIII proteins from tobacco , Arabidopsis and moss , revealed that moss myosin VIII proteins are ∼50% identical to tobacco and Arabidopsis myosin VIIIs ( Figure 4—source data 1 , 2 ) . Except for the closely related tobacco and Arabidopsis Myo8B and D , the percent identity between the remaining four tobacco and Arabidopsis myosin VIIIs is 51–58% ( Figure 4—source data 1 and 2 ) . With such high sequence identity , we reasoned that Myo8A-GFP might serve as a proxy for the localization of seed plant myosin VIIIs . We found that similar to the interphase localization described above for moss , Myo8A-GFP localizes in tobacco to dynamic cortical particles ( Figure 4A; Video 4 ) . In cells about to enter mitosis , cortical Myo8A-GFP accumulates at the presumptive position of the preprophase band ( Figure 4B ) . As cells enter mitosis , cortical Myo8A-GFP tightens into a thin band and remains at the cortical division site . In early cytokinesis , Myo8A-GFP appears also at the phragmoplast midzone ( Figure 4C ) . As the phragmoplast expands , the Myo8A-GFP at the phragmoplast midzone expands out eventually reaching the Myo8A-GFP at the cortical division site ( Figure 4D; Video 5 ) . Evidently , myo8A-GFP is capable of localizing to the phragmoplast and future site of division in a seed plant as well as in a bryophyte . 10 . 7554/eLife . 03498 . 011Figure 4 . Myosin VIII localizes to the preprophase band , phragmoplast and cortical division site in tobacco BY-2 cells . ( A ) Myo8A-GFP localizes to dynamic punctate cortical structures on the cell cortex of BY-2 cells . VAEM image of a single frame from a time-lapse acquisition is shown on the left . On the right is a maximum projection of frames from 5 s of real time . Linear trajectories are readily apparent in the maximum projection . See also Video 4 . Scale bar , 5 µm . ( B and C ) Left , z-projection . Right , midplane . Tobacco BY-2 cell in preprophase ( B ) and cytokinesis ( C ) expressing moss Myo8A-GFP . Myo8A-GFP accumulates on the preprophase band ( B , arrow heads ) . n denotes the nucleus . Myo8A-GFP remains at the cortical division site ( C , arrow heads ) and is at the phragmoplast midzone ( C , arrow ) . ( D ) Images from a single focal plane of a dividing BY-2 cell expressing Myo8A-GFP ( green ) and stained with FM4-64 ( red ) acquired over time . FM4-64 labels membrane added to the expanding cell plate ( asterisk ) . Myo8A-GFP localizes to the phragmoplast midzone ( arrow ) and the cortical division site ( arrow heads ) . See also Video 5 . Images in ( B–D ) were acquired with a spinning disc confocal . ( B–D ) Scale bars , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 01110 . 7554/eLife . 03498 . 012Figure 4—source data 1 . Multiple sequence alignment of class VIII myosins from Arabidopsis thaliana ( At ) , Nicotiana benthamiana ( Nb ) , and Physcomitrella patens ( Pp ) generated with Clustal O . Nicotiana benthamiana sequences can be found on CyMobase . The following sequences can be found on Phytozome Physcomitrella patens v1 . 6: PpMyo8A ( Pp1s228_18V6 . 1 ) , PpMyo8C ( Pp1s199_21V6 . 1 ) . The following sequences can be found on NCBI: PpMyo8B ( AEM05967 ) , PpMyo8D ( AEM05968 ) , PpMyo8E ( AEM05969 ) , AtMyo8A ( NP_194467 ) , AtMyo8B ( NP_175453 ) , AtMyo8C ( NP_001078755 ) , AtMyo8D ( NP_188630 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 01210 . 7554/eLife . 03498 . 013Figure 4—source data 2 . Table shows the amino acid sequence comparison between Arabidopsis thaliana ( At ) , Nicotiana benthamiana ( Nb ) and Physcomitrella patens ( Pp ) class VIII myosins . Percent identity from Clusal O multiple sequence alignment is reported . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 01310 . 7554/eLife . 03498 . 014Video 4 . Myo8A-GFP localizes to dynamic punctate cortical structures on the cell cortex of BY-2 cells ( see Figure 4A ) . BY-2 cells expressing Myo8A-GFP were imaged with VAEM ( acquired at 11 . 3 fps ) . Video is playing at 11 . 3 fps . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 01410 . 7554/eLife . 03498 . 015Video 5 . Myo8A-GFP localizes to the phragmoplast and cortical division site in tobacco BY-2 cells ( see Figure 4D ) . BY-2 cell expressing Myo8A-GFP ( green ) stained with FM4-64 ( red ) was imaged with spinning disc confocal microscope . Images are a single focal plane acquired every minute . Video is playing at 3 fps . Scale bar , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 015 Because division in apical cells is more frequent than in branching cells , we imaged moss apical cells to investigate the mechanism of myosin VIII function during phragmoplast expansion . Before nuclear envelope break down , we found a population of Myo8A-GFP that localizes to cytoplasmic microtubules surrounding the nucleus ( Figure 5A ) . Myo8A-GFP remains localized along microtubules as the spindle is assembled ( Figure 5B; Video 6 ) . During mitosis , Myo8A-GFP continues to associate with the mitotic spindle ( Figure 5C; Video 7 ) , enriched at the midzone and to some extent at the poles . Prior to anaphase , a small population of Myo8A-GFP accumulates at the cortex near the spindle midzone ( Figure 5C , arrows; Video 7 ) . During anaphase , Myo8A-GFP concentrates at the midzone . Initially Myo8A-GFP is found throughout the phragmoplast midzone ( Figure 5C—figure supplement 1 ) . As the phragmoplast matures , Myo8A-GFP accumulates on the leading edge of the phragmoplast , ultimately forming a ring ( Figure 5C , Figure 5C—figure supplement 1; Video 7 ) . Interestingly in apical cells , Myo8A-GFP does accumulate at the cell cortex , but in contrast to branching cells , this accumulation occurs later and is significantly more dynamic . 10 . 7554/eLife . 03498 . 016Figure 5 . Myo8A-GFP localizes to the mitotic spindle and phragmoplast . ( A and B ) Moss protonemal apical cells expressing Myo8A-GFP ( green ) and mCherry-tubulin ( red ) imaged on a scanning confocal microscope . Images are single focal planes acquired over time . Scale bar , 5 µm . ( A ) Two examples of Myo8A-GFP associating with cytoplasmic microtubules surrounding the nucleus before mitosis . ( B ) Myo8A-GFP stays associated with microtubules throughout mitosis . See also Video 6 . Scale bar , 5 µm . ( C ) Myo8A-GFP accumulates in the midzone . Arrows indicate cortical accumulation . Images were acquired with a spinning disc confocal microscope and are maximum projections of z-stacks acquired over time . See also Video 7 . Scale bar , 5 µm . In all cases , large globular structures are chloroplasts that auto-fluoresce in the GFP channel . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 01610 . 7554/eLife . 03498 . 017Figure 5—figure supplement 1 . Myo8A-GFP localization in the phragmoplast . Images from a 3D rotation of the spinning disc confocal images presented in Figure 5C . Rotations along the plane of the phragmoplast equator show that in the early phragmoplast Myo8A-GFP is found throughout the phragmoplast midzone and the cell cortex . As the phragmoplast matures , Myo8A-GFP tightens into a ring along the leading edge of the phragmoplast . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 01710 . 7554/eLife . 03498 . 018Video 6 . Myo8A-GFP localizes to the cytoplasmic microtubules around nucleus and remains on the spindle ( see Figure 5B ) . Moss apical cell expressing Myo8A-GFP ( green ) and mCherry-tubulin ( red ) was imaged with a scanning confocal microscope . Images are a single focal plane acquired every minute . Video is playing at 4 fps . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 01810 . 7554/eLife . 03498 . 019Video 7 . Myo8A-GFP associates with mitotic spindle and phragmoplast ( see Figure 5C ) . Moss apical cell expressing Myo8A-GFP ( green ) and mCherry-tubulin ( red ) was imaged with a spinning disc confocal microscope . Images are maximum projections of a z-stack acquired every minute . Video is playing at 4 fps . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 019 To test whether actin is involved in recruiting Myo8A-GFP to the mitotic spindle , we first imaged Myo8A-GFP and lifeact-mCherry in dividing cells ( Figure 6A; Video 8 , top ) . When Myo8A-GFP appears on the spindle , there is little to no accumulation of lifeact-mCherry . During the transition from spindle to phragmoplast , Myo8A-GFP concentrates at the midzone and lifeact-mCherry fluorescence rises above background levels in the vicinity of the phragmoplast . However , when Myo8A-GFP fluorescence tightens into a thin band on the phragmoplast leading edge , the lifeact-mCherry fluorescence accumulates significantly around the phragmoplast ( Figure 6A; Video 8 , top ) . The timing of the appearance of actin suggests that early localization of Myo8A-GFP to the mitotic spindle is independent of actin . To test this , we imaged cells entering mitosis in the presence of 25 µM Latrunculin B , which depolymerizes the actin cytoskeleton ( Figure 6—figure supplement 1 ) . Strikingly , Myo8A-GFP still accumulates on the mitotic spindle . Similar to control cells , Myo8A-GFP in drug-treated cells accumulates in the spindle midzone at anaphase and tightens into a thin band at the leading edge of the phragmoplast during cytokinesis ( Figure 6B; Video 8 , bottom ) . Thus , Myo8A-GFP recruitment to the mitotic spindle and behavior during mitosis is apparently independent of actin . 10 . 7554/eLife . 03498 . 020Figure 6 . Myo8A-GFP localizes to the mitotic spindle and phragmoplast independent of actin . Moss protonemal apical cells imaged on a spinning disc confocal microscope . All images are maximum projections of z-stacks acquired over time . ( A ) Cell expressing Myo8A-GFP ( green ) and Lifeact-mCherry ( red ) . ( B ) Cell expressing Myo8A-GFP ( green ) and mCherry-tubulin ( red ) treated with 25 µm LatB . See also Video 8 . Scale bars , 5 µm . In all cases , large globular structures are chloroplasts that auto-fluoresce in the GFP channel . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 02010 . 7554/eLife . 03498 . 021Figure 6—figure supplement 1 . Dose response of latrunculin B in apical protonemal moss cells . VAEM images of protonemal apical cells expressing lifeact-mRuby2 were acquired with increasing concentrations of LatB . At 25 µM , the concentration used for all drug treatments , it is no longer possible to observe actin filaments ( n = 25 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 02110 . 7554/eLife . 03498 . 022Video 8 . Myo8A-GFP associates with the mitotic spindle and phragmoplast independent of actin ( see Figure 6A , B ) . Moss apical cells expressing Myo8A-GFP ( green ) and lifeact-mCherry/mCherry tubulin ( red ) were imaged with a spinning disc confocal microscope . Images are maximum projections of a z-stack acquired every minute . Video is playing at 4 fps . Scale bar , 5 µm . Top , Myo8A-GFP and lifeact-mCherry in control cell . Bottom , Myo8A-GFP and mCherry-tubulin in LatB treated cell . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 022 The fact that Myo8A-GFP localization is scarcely changed in the absence of actin raises the possibility that myosin VIII functions independently of actin . To test this , we imaged phragmoplasts labeled with GFP-tubulin and FM4-64 . Since actin is essential for polarized expansion , it was not possible to perform long-term Latrunculin B treatments . Instead , plants were treated for 2 hr before imaging , ensuring that essentially all observed phragmoplasts had been formed in the absence of actin . The majority of untreated , wild-type phragmoplasts deposit membrane uniformly , appearing as smooth FM4-64 staining in the midzone ( 80% of wild type cells , n = 55; Figure 7A; Video 9 , ) . In contrast , most myosin VIII null dividing cells have non-uniform FM4-64 staining ( 70% of Δmyo8ABCDE cells , n = 79; Figure 7B; Video 10 ) . Interestingly , treatment of wild type dividing cells with Latrunculin B results in cells with similar defects in cell plate assembly as those observed in the myosin VIII null cells ( 70% of wild type cells treated with LatB , n = 50; Figure 7C; Video 11 ) , indicating that the observed changes in the assembling cell plate result from compromising actin and myosin . 10 . 7554/eLife . 03498 . 023Figure 7 . Actin is required for Myo8 function in cytokinesis . Phragmoplasts from wild type . ( A ) Δmyo8ABCDE ( B ) and wild type treated with 25 µM LatB ( C ) expressing GFP-tubulin ( green ) and stained with FM4-64 ( red ) . Cells were imaged on a scanning confocal microscope . For LatB treatment , wild type plants were treated with 25 µM LatB for 2 hr , then stained with FM4-64 and imaged in the presence of 25 µM LatB . Images are single focal planes taken from a time series . Scale bars , 5 µm . See also Video 9 ( for A ) , Video 10 ( for B ) , and Video 11 ( for C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 02310 . 7554/eLife . 03498 . 024Video 9 . New membrane is deposited uniformly in a wild type dividing cell ( see Figure 7A ) . A wild type cell expressing GFP-tubulin was stained with FM4-64 and imaged in a single focal plane on a scanning confocal microscope . Video is playing at 10 fps . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 02410 . 7554/eLife . 03498 . 025Video 10 . New membrane is deposited non-uniformly in a Δmyo8ABCDE dividing cell ( see Figure 7B ) . A Δmyo8ABCDE cell expressing GFP-tubulin was stained with FM4-64 and imaged in a single focal plane on a scanning confocal microscope . Video is playing at 10 fps . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 02510 . 7554/eLife . 03498 . 026Video 11 . New membrane is deposited non-uniformly in a wild type dividing cell treated with LatB ( see Figure 7C ) . A wild type cell expressing GFP-tubulin was treated for two hours with 25 µM LatB and then stained with FM4-64 and imaged in 25 µM LatB . Images are from a single focal plane taken on a scanning confocal microscope . Video is playing at 10 fps . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 026 During phragmoplast expansion , the vast majority of microtubules are tightly focused at the phragmoplast midzone , where there is a strong accumulation of Myo8A-GFP . However , we also observed that Myo8A-GFP localizes to the ends of peripheral phragmoplast microtubules ( Figure 8A , B , arrows; Video 12 ) that initiate from the edge of the phragmoplast and are initially unattached to the phragmoplast midzone . We investigated the behavior of these peripheral microtubules in control and latrunculin B treated cells ( Figure 8; Video 12 ) . Before the phragmoplast reaches the cell cortex , we found that there were three times fewer peripheral microtubules in control cells as compared to cells treated with latrunculin B . We suspected that this difference likely results from the fact that peripheral microtubules more rapidly focus at the phragmoplast midzone in control cells . To test this , we identified peripheral microtubules in control and latrunculin B treated cells and followed them for 20 s . We found in control cells that only 20% of the peripheral microtubules were still present after 20 s ( n = 4 cells ) . In contrast , 89% of peripheral microtubules were still present after 20 s in latrunculin B treated cells ( n = 4 cells ) . In some cases ( Figure 8D , yellow arrow ) peripheral microtubules were observed to remain unattached for more than a minute . Taken together these data indicate that in the presence of actin peripheral microtubules are swiftly integrated into the expanding phragmoplast , suggesting that actin filaments exist between the edge of the expanding phragmoplast , peripheral microtubules , and the cell cortex . 10 . 7554/eLife . 03498 . 027Figure 8 . Myo8A-GFP associates with the ends of phragmoplast microtubules . Images of a protonemal apical cell expressing Myo8A-GFP ( green ) and mCherry-tubulin ( red ) were acquired with a spinning disc confocal microscope . Images are a single focal plane from a time series . Arrows indicate enrichment of Myo8A-GFP at the ends of peripheral microtubules . Arrow heads indicate where peripheral microtubules are incorporated into the phragmoplast midzone . ( A ) In a control cell , peripheral microtubules focus at the phragmoplast midzone . Scale bar , 5 µm . ( B ) Zoom-in of the phragmoplast periphery from the control cell . Peripheral microtubules with Myo8A-GFP are evident in this area . In the presence of actin , peripheral microtubules are incorporated into the phragmoplast midzone rapidly . Scale bar , 2 µm . ( C ) In a cell treated with 25 µM LatB , peripheral microtubules are no longer focused at the midzone . Scale bar , 5 µm . ( D ) Zoom-in of the phragmoplast periphery in the LatB treated cell . Peripheral microtubules stay associated with the cell cortex for more than a minute . Scale bar , 2 µm . See also Video 12 . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 02710 . 7554/eLife . 03498 . 028Video 12 . Peripheral microtubules require actin to be efficiently incorporated into the phragmoplast ( see Figure 8 ) . Moss apical cell expressing Myo8A-GFP ( green ) and mCherry-tubulin ( red ) were imaged with a spinning disc confocal microscope . Images are a single focal plane acquired every 2 s . Video is playing at 10 fps . Scale bar , 5 µm . Top , control . Bottom , LatB treated cell . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 028 In support of this , we found that an actin nucleator , For2A-GFP ( van Gisbergen et al . , 2012 ) , is enriched on the phragmoplast as soon as it forms from the late spindle ( Figure 9A; Video 13 ) . For2A-GFP remains on the edge of the phragmoplast throughout cytokinesis ( Figure 9B; Video 14 ) , suggesting that actin is actively polymerized on the phragmoplast edge . To test this , we simultaneously imaged lifeact-mEGFP and mCherry-tubulin using a laser scanning confocal microscope . We confirmed that actin accumulates at the midzone once the phragmoplast forms . As the phragmoplast expands out from the center of the cell towards the cell cortex , we discovered that actin filaments are present at the midzone and between the leading edge of the phragmoplast and the cell cortex ( Figure 9C; Video 15 ) . Moreover , peripheral phragmoplast microtubules intersect the actin filaments that span the distance between the phragmoplast leading edge and the cell cortex ( Figure 9C , arrowhead ) . Our data suggest that microtubules may interact with actin filaments bridging the cell cortex and the phragmoplast during the time that myosin VIII-mediated motility guides the phragmoplast . 10 . 7554/eLife . 03498 . 029Figure 9 . Actin is polymerized on the phragmoplast edge . Single focal plane images acquired with a laser scanning confocal microscope . ( A and B ) Protonemal apical cell expressing For2A-GFP ( green ) and mCherry-tubulin ( red ) . ( A ) In metaphase through anaphase ( A , t = 0–75 s ) , For2A-GFP is not associated with the spindle . See also Video 13 . ( B ) For2A-GFP is enriched at the phragmoplast midzone and remains on the edge of the phragmoplast throughout cytokinesis . See also Video 14 . ( C ) Protonemal cell expressing Lifeact-mEGFP ( green ) and mCherry-tubulin ( red ) . Microtubules intersect actin filaments between the leading edge of the phragmoplast and the cell cortex ( arrow heads ) . See also Video 15 . Scale bars , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 02910 . 7554/eLife . 03498 . 030Video 13 . For2A-GFP does not associat with the mitotic spindle but is present in the phragmoplast ( see Figure 9A ) . Moss apical cell expressing For2A-GFP ( green ) and mCherry-tubulin ( red ) was imaged with a scanning confocal microscope . Images are a single focal plane acquired every 5 s . Video is playing at 5 fps . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 03010 . 7554/eLife . 03498 . 031Video 14 . For2A-GFP remains at the edge of phragmoplast ( see Figure 9B ) . Moss apical cell expressing For2A-GFP ( green ) and mCherry-tubulin ( red ) was imaged with a scanning confocal microscope . Images are a single focal plane acquired every 5 s . Video is playing at 5 fps . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 03110 . 7554/eLife . 03498 . 032Video 15 . Actin polymerizes between the edge of the phragmoplast and the cortical division site during cytokinesis ( see Figure 9C ) . Moss apical cell expressing Myo8A-GFP ( green ) and lifeact-mCherry ( red ) was imaged with a scanning confocal microscope . Images are a single focal plane acquired continuously at 3 . 75 fps . Video is playing at 75 fps . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 032
Based on our data , we propose the following model for myosin VIII function in phragmoplast guidance during division of a protonemal apical cell ( Figure 10 ) . A population of myosin VIII localizes to microtubules and this population incorporates into the mitotic spindle upon spindle formation . In early mitosis , as microtubules are dynamically searching to make chromosomal attachments , Myo8A-GFP decorates the entire spindle ( Figure 5B ) . At metaphase , Myo8A-GFP concentrates in the spindle midzone and at anaphase an additional population of Myo8A-GFP dynamically concentrates at the cell cortex where the cell plate ultimately fuses with the parental cell wall . Once the phragmoplast assembles , Myo8A-GFP forms a tight ring at the edge of the expanding phragmoplast ( Figure 5 ) . At this point , the class II formin , For2A , localizes to the phragmoplast ( Figure 9B; van Gisbergen et al . , 2012 ) and polymerizes actin filaments between the leading edge of the phragmoplast and the cell cortex ( Figure 9C ) . Since For2A remains associated with the phragmoplast , it suggests that the barbed ends of the actin filaments are anchored at the phragmoplast midzone and the pointed ends are in the cytoplasm . Myosin VIII at the cortex can hold onto these actin filaments and walk towards the barbed end thereby aligning the actin filaments to the cortical division site . We propose that myosin VIII at the ends of peripheral phragmoplast microtubules moves along these actin filaments from the cortical division site towards the expanding phragmoplast , thereby translocating microtubules and ensuring that phragmoplast expansion occurs along a plane defined by the cortical division site ( Figure 10 ) . 10 . 7554/eLife . 03498 . 033Figure 10 . A model for myosin VIII function in phragmoplast guidance . In prometaphase , myosin VIII localizes to plus ends throughout the mitotic spindle . In metaphase , myosin VIII accumulates at the spindle midzone and on the poles . During anaphase , myosin VIII is observed on peripheral microtubules and at the cell cortex . As the phragmoplast forms , the midzone myosin VIII accumulation tightens into a thin band on the phragmoplast edge . For2A is at the phragmoplast midzone . Actin filaments are generated between the phragmoplast and cortical myosin VIII . Peripheral microtubules with plus-end associated myosin VIII translocate on actin filaments and are incorporated into the expanding phragmoplast . DOI: http://dx . doi . org/10 . 7554/eLife . 03498 . 033 There is precedent for myosin-based motility translocating microtubules on actin filaments . In the budding yeast , Saccharomyces cerevisiae , the class V myosin , Myo2p , binds Kar9p , which localizes to cytoplasmic microtubule plus ends by binding the yeast EB1 homolog , Bim1p ( Beach et al . , 2000 ) . These cytoplasmic microtubules emanate from the spindle pole body embedded in the nuclear envelope and Myo2p mediates their motility along actin cables directed into the bud , moving the nucleus toward the bud neck ( Beach et al . , 2000; Yin et al . , 2000 ) . We imagine a similar mechanism could be at work in plant cells , whereby myosin VIII associates with microtubule ends , and subsequently translocates microtubules on actin filaments to guide phragmoplast expansion . While the localization of myosin VIII is striking , myosin VIII in protonemata is mostly dispensable for phragmoplast guidance in apical cells . These cells are narrow , with diameters not much greater than the sizes of the nucleus and spindle , leaving little room for those two structures to be misplaced . Due to these geometric constraints , the mitotic spindle always forms along the long axis of the cell . Thus , any cell plate positioning defects that occur are mild because the spindle midzone is always roughly perpendicular to the long axis of the cell . In contrast , myosin VIII is needed for branch formation , insofar as cell plates are often aberrantly positioned in the branching cells of the myosin VIII nulls . Arguably , in comparison to apical cell division , side-branch formation needs to specify the cell division plane more accurately . Branching involves an asymmetric cell division in an L-shaped cell . The nucleus migrates toward the junction of the parental cell and the branch and the spindle is oriented along the longitudinal axis of the new emerging cell . The phragmoplast builds the new cell plate at the junction of the two cells . We found that in branching cells , a relatively static population of Myo8A-GFP accumulates at the cell cortex at the future site of cell division ( Figure 3 ) . Myo8A-GFP accumulates at this site early , prior to mitosis , and remains throughout cytokinesis . As in apical cells , Myo8A-GFP also accumulates on the mitotic spindle , ultimately forming a tight band on the leading edge of the phragmoplast . The new cell plate fuses with the parental cell membranes at the cortical site defined by the presence of Myo8A-GFP . Since there are many ways to orient the spindle in a branching cell , in the absence of myosin VIII defects in cell plate positioning are frequent , suggesting that myosin VIII functions to guide phragmoplast expansion . Notably , our data suggests that a myosin VIII-mediated phragmoplast-guidance mechanism also exists in seed plants , since we found that moss Myo8A-GFP localizes to the preprophase band , cortical division site , and the phragmoplast midzone in BY-2 cells ( Figure 4 ) . Our result is consistent with a previous report in which a GFP fusion of one of the Arabidopsis class VIII myosins , ATM1-GFP , was shown to localize to the phragmoplast in BY-2 cells ( Van Damme et al . , 2004 ) . The discovery that Myo8A-GFP localizes to the cortical division site gives us a portal to connect our observations in moss to the current model of cell division in seed plants ( Van Damme , 2009; Rasmussen et al . , 2011b ) . In that model , cells with preprophase bands use a microtubule-dependent mechanism to position the nucleus such that it is bisected by the future plane of division and to form the spindle perpendicularly ( Mineyuki and Furuya , 1986; Venverloo and Libbenga , 1987; Katsuta et al . , 1990 ) . Once the spindle is aligned , defects in phragmoplast guidance would little alter cell plate positioning and subsequent tissue morphogenesis , obscuring the role of myosin VIII . Nevertheless , when BY-2 cells are treated with actin inhibitors , phragmoplasts are disorganized , generating wrinkled cell plates that are often skewed with respect to the cortical division site ( Hoshino et al . , 2003; Yoneda et al . , 2004; Sano et al . , 2005; Higaki et al . , 2008; Kojo et al . , 2013 ) . Thus , we predict that for fine tuning myosin VIII and actin guide phragmoplast expansion in cells with preprophase bands . Our data provide evidence that myosin VIII and actin steer phragmoplast expansion during cytokinesis in both moss and tobacco , suggesting that myosin VIII function is conserved throughout plant evolution . In fact , myosin VIII provides a physical link between phragmoplast microtubules and the cortical division site via actin filaments . We propose that myosin VIII's motor activity along actin provides a molecular mechanism for steering phragmoplast expansion .
All expression constructs were constructed using Multisite Gateway recombination technology from Invitrogen ( Carlsbad , CA ) . Generation of entry clones 3XmEGFP-L5L2 ( 29 ) , Lifeact-L1R5 ( 30 ) and mCherry-L5L2 ( 15 ) were described previously . To construct the entry clone Myo8A-L1R5 , total RNA was extracted from 7-day-old moss protonemal tissues using RNeasy plant mini kit ( Qiagen ) , followed by DNase I treatment according to the manufacturer's protocol . cDNA was synthesized from total RNA using SuperScript II reverse transcriptase ( Invitrogen ) and oligo ( dT ) according to manufacturer's protocol . Full length Myo8A coding sequence was amplified from moss cDNA using Myo8A specific primers ( P1 & P2 ) , and cloned into pGEM-T easy from Promega ( Madison , WI ) . The full-length Myo8A coding sequence was then amplified from the pGEM-Myo8A clone using primers ( P3 & P4 ) containing attB1 and attB5r sites , and cloned into pDONR-P1P5r with a BP reaction ( Invitrogen ) . The mCherry and mEGFP coding sequence was amplified using primers ( P5 & P6 ) with attB1 and attB5r sites and cloned into pDONR-P1P5r to generate entry clones mCherry-L1R5 and mEGFP-L1R5 . Moss α−tubulin coding sequence was amplified from pAct-GFP-TUA1 ( 31 ) with primers ( P7 & P8 ) containing attB5 and attB2 sites and cloned into pDONR-P5P2 to generate entry clone α-tubulin-L5L2 . Combinations of entry clones were assembled with destination vectors generating constructs for stable expression in moss or tobacco BY-2 cells using LR clonase II plus reactions ( Invitrogen ) as follows: Myo8A-L1R5 and 3XmEGFP-L5L2 with pTKUbi-gate generating pTKUbi-Myo8A-3mEGFP; Lifeact-L1R5 and mCherry-L5L2 with pTZUbi-gate generating pTZUbi-Lifeact-mCherry; TUA1-L5L2 and mCherry-L1R5 with pTZUbi-gate generating pTZUbi-mCherry-tubulin; TUA1-L5L2 and mEGFP-L1R5 with pTKUbi-gate generating pTKUbi-mEGFP-tubulin; Myo8A-L1R5 and 3XmEGFP-L5L2 with pMDC32 ( 32 ) generating pMDC32-Myo8A-3XmEGFP . The pTKUbi-gate vector has an expression cassette derived from pTHUbi-Gate ( 33 ) , which contains the maize ubiquitin promoter , Gateway cassette and NOS terminator . Following this expression cassette is a 35S::NptII::ter cassette flanked by lox sites ( 14 ) . The expression and antibiotic resistance cassettes are flanked by moss genomic sequence from the Pp1s249_67V6 . 1 locus . Nucleotides −2 to −1153 and nucleotides 660 to 1757 are on the 5′ and 3′ ends , respectively , with Pme I sites incorporated such that digestion with Pme I releases the moss genomic DNA targeting arms as well as the expression and resistance cassettes . The pTZUbi-gate vector is similar except that it contains a 35S::Zeo::ter cassette flanked by lox sites ( 14 ) as the antibiotic resistance cassette , uses moss genomic sequence from the Pp1s141_25V6 . 1 locus ( nucleotides +908 to +2021 and −35 to −1532 ) , and has Swa I for release of the moss genomic DNA targeting arms and expression and resistance cassettes . The coding sequence of moss α-tubulin ( Pp1s215_51V6 locus ) was amplified from moss cDNA using primers P9 and P10 and cloned into pENTR/D-TOPO ( Invitrogen ) . After verification by sequencing , the coding sequence was cloned into the L5L4-mCherry plasmid ( modified with restriction sites C-terminal to the mCherry L5L4-mCherry-AscI-SpeI ) using the restriction sites AscI and SpeI , generating L5L4-mCherry-αtub215-51-1 . Sequences upstream and downstream of the locus were amplified as targeting sequences for homologous recombination . The 5' targeting sequence was amplified using primers ( P11 & P12 ) and cloned into pDONR-P1P5r using BP Clonase II ( Invitrogen ) , generating L1R5-αtub215-51-1-5′tarm . Similarly , the 3′ targeting sequence was amplified with primers ( P13 & P14 ) and cloned into pDONR-P3P2 using BP Clonase II ( Invitrogen ) , generating L3L2-αtub215-51-1-3′tarm . The L1R5-αtub215-51-1-5′tarm , L5L4-mCherry-αtub215-51-1 and L3L2-αtub215-51-1-3′tarm were recombined with R4R3 NOSter-Lox-Hygro-Lox ( 34 ) and pGEM-Gate ( 30 ) using LR Clonase II plus ( Invitrogen ) to generate the final construct for homologous recombination in moss , mCherry-αtub215-51-1AR . ( P1 ) ATGTATTCTACGAATGGCATTGAGG; ( P2 ) CTAACCTTGGAGCGCTCTTGAGG; ( P3 ) GGGGACAAGTTTGTACAAAAAAGCAGGCTTCATGTATTCTACGAATGGC; ( P4 ) GGGGACAACTTTTGTATACAAAGTTGTACCTTGGAGCGCTCTTGAGG; ( P5 ) GGGGACAAGTTTGTACAAAAAAGCAGGCTTCATGGTGAGCAAGGGCGAGGAG; ( P6 ) GGGGACAACTTTTGTATACAAAGTTGTCTTGTACAGCTCGTCCATGCC; ( P7 ) GGGGACAACTTTGTATACAAAAGTTGTTATGAGAGAGATTATCAGCATCCAC; ( P8 ) GGGGACCACTTTGTACAAGAAAGCTGGGTATCAGTAGTCGTCGTCCTCC; ( P9 ) ACTAGTTTAGTACTCGTCGTCGTCCTGTCCTCCGTCGGTGGATTCAGC; ( P10 ) GGCGCGCCATGAGAGAGATCATCAGTATCCATATAGGTCAGG; ( P11 ) GGGGACAAGTTTGTACAAAAAAGCAGGCTCTGGCGCGCCACTTCATAATCTACCTGTGC; ( P12 ) GGGGACAACTTTTGTATACAAAGTTGTGGAAGAGTACGAGCAGCAGC; ( P13 ) GGGGACAACTTTGTATAATAAAGTTGTGGGCTTTTATTTTGAGGCGGAAACGG; ( P14 ) GGGGACCACTTTGTACAAGAAAGCTGGGTAGGCGCGCCGTTAACTGTGGAGTTCTG; ( P15 ) GCAATACAACACACTGTGCTTGGG; ( P16 ) GGTGTTGAAAGCATCATCACCACC . All moss tissue culture , protoplasting and transformation were performed as described previously ( 15 ) . Moss protonemal tissues were propagated weekly on PpNH4 medium ( 1 . 03 mM MgSO4 , 1 . 86 mM KH2PO4 , 3 . 3 mM Ca ( NO3 ) 2 , 2 . 7 mM ( NH4 ) 2-tartrate , 45 μM FeSO4 , 9 . 93 μM H3BO3 , 220 nM CuSO4 , 1 . 966 μM MnCl2 , 231 nM CoCl2 , 191 nM ZnSO4 , 169 nM KI , and 103 nM Na2MoO4 ) containing 0 . 7% agar . For imaging , 1-week-old protonemal tissues were protoplasted , plated in plating medium at a density of ∼20 , 000 cells/9 cm2 plate , regenerated on protoplast regeneration medium ( PRM ) for 4 days and transferred to PpNH4 plates . For moss transformation , protoplasts were isolated from 7-day-old moss protonemal tissue , and transformed with linearized DNA via PEG-mediated transformation ( 15 ) . At least 30 µg of plasmid DNA was linearized , ethanol precipitated , and dissolved in sterile TE buffer . pTKUbi constructs were linearized with Pme I and pTZUbi constructs were linearized with Swa I . Protoplasts were transformed with linearized DNA via PEG-mediated transformation ( 15 ) . Protoplasts were plated in top agar , regenerated on PRM ( PpNH4 medium supplemented with 8 . 5% mannitol and 10 mM CaCl2 . ) for 4 days and transferred to PpNH4 medium containing the appropriate antibiotics ( G418 , 20 µg/ml; zeocin , 50 µg/ml; hygromycin , 15 µg/ml ) . To select for stable transformants , transformations were cycled on and off antibiotic selection for three 1-week intervals . Stable transgenic lines were visually screened on a confocal microscope for expression of the transgene . pTKUbi-mEGFP-Tub was transformed into WT and Δmyo8ABCDE ( 14 ) generating the mEGFP-Tub lines . pTKUbi-Myo8A-3mEGFP was transformed into Δmyo8ABCDE generating the Myo8A-GFP line . Myo8A-GFP was subsequently transformed with pTZUbi-mCherry-tubulin and pTZUbi-Lifeact-mCherry to generate Myo8A-GFP/mCherry-tub and Myo8A-GFP/Lifeact-mCherry . Lifeact-mEGFP line ( 30 ) was transformed with the Asc I linearized mCherry-αtub215-51-1AR construct , to generate Lifeact-mEGFP/mCherry–tubulin line . Stably transformed lines were identified by visually screening for mCherry expression and then verified by genotyping using primers P15 and P16 . Tobacco BY-2 tissue culture and Agrobacterium-mediated transformation were performed as previously described ( 35 ) . Tobacco BY-2 cells were transformed with pMDC32-Myo8A-3XmEGFP . The GV3101 Agrobacterium tumefaciens strain was used for infection . For measurement of cell plate angle and apical cell length , plants regenerated from protoplasts were stained with 0 . 1 mg ml−1 calcofluor solution and visualized by epifluorescence microscopy ( Leica MZ16FA , Leica Microsystems , Buffalo Grove , IL ) using a UV filter ( excitation 360/40 nm , emission 420 long pass ) or a Violet filter ( excitation 425/40 nm , emission 400 long pass ) . Cell plate angles and apical cell lengths were measured manually using ImageJ software . For imaging cell division , 5- to 8-day-old plants regenerated from protoplasts were placed unto an agar pad in Hoagland's ( 4 mM KNO3 , 2 mM KH2PO4 , 1 mM Ca ( NO3 ) 2 , 89 μM Fe citrate , 300 μM MgSO4 , 9 . 93 μM H3BO3 , 220 nM CuSO4 , 1 . 966 μM MnCl2 , 231 nM CoCl2 , 191 nM ZnSO4 , 169 nM KI , 103 nM Na2MoO4 , and 1% sucrose ) , covered by a glass cover slip and sealed with VALAP ( 1:1:1 parts of Vaseline , lanoline , and paraffin ) . For cell plate staining , 15 µM FM4-64 was added in the Hoagland's solution in the agar pad . For latrunculin B treatment , plants were transferred to PpNH4 medium containing 25 µM latrunculin B for 2 hr , then transferred to slides containing 25 µM latrunculin B in both the agar pad and the Hoagland's solution . Samples were mounted on an inverted microscope ( model Ti-E; Nikon Instruments , Melville , NY ) equipped with a mirror-based T-FL-TIRF illuminator ( Nikon ) and imaged with a 1 . 49 NA 100× oil immersion TIRF objective ( Nikon ) . The 1 . 5× optivar was used for all images to increase magnification . 488 and 561 nm laser illumination was used for GFP and mCherry excitation , respectively . The laser illumination angle was adjusted individually for each sample to achieve the maximum signal-to-noise ratio . Signals for each channel were captured simultaneously with a 1024 × 1024 electron-multiplying CCD camera ( iXON3; Andor Technology USA , South Windsor , CT ) equipped with a dual-view adaptor ( Photometrics , Tucson , AZ ) . Emission filters were 525/50 nm for GFP and 595/50 for mCherry . Image acquisition process was controlled by NIS-Elements AR 3 . 2 software ( Nikon ) . The slides were mounted on an inverted microscope ( model Ti-E; Nikon ) equipped with a Yokogawa spinning disk head ( model CSU-X1 ) and a 512 × 512 electron multiplying CCD camera ( iXON; Andor Technology ) . Images for each channel were collected sequentially with a 1 . 4 NA 100× oil immersion objective ( Nikon ) at room temperature . 488 and 561 nm laser illumination was used for GFP/FM4-64 and mCherry excitation , respectively . Emission filters were 515/30 nm for GFP and 600/32 nm for mCherry/FM4-64 . Image acquisition was controlled by MetaMorph software ( Molecular Devices , Sunnyvale , CA ) . Images for each channel were acquired simultaneously on a Nikon A1R confocal microscope system with a 1 . 4 NA 100× oil immersion objective ( Nikon ) at room temperature . 488 and 561 nm laser illumination was used for GFP/FM4-64 and mCherry excitation , respectively . Emission filters were 525/50 nm for GFP and 595/50 for mCherry/FM4-64 . Image acquisition was controlled by NIS-Elements software ( Nikon ) . All images were processed using ImageJ software with enhanced contrast and background subtraction ( rolling ball diameter of 50 was used for spinning disc images; for scanning confocal images a rolling ball diameter of 50 was used for actin filaments and 75 for all other labels ) . We also performed smoothing and applied the unsharp mask filter to all images . All settings were standard . | Plant cells are surrounded by a membrane , which controls what enters and leaves the cell , and a cell wall , which provides rigidity . When a plant cell is ready to divide , it needs to produce two new cell membranes , with a new cell wall sandwiched between them , to split the cell contents into two daughter cells . During the division process the cell builds a scaffold called the phragmoplast that guides the delivery of the materials that are needed to make the new cell wall and membranes . The phragmoplast—which is made of rod-like proteins called microtubules and actin filaments—starts at the centre of the cell and expands towards a pre-determined site on the existing cell wall . The question is: how does the phragmoplast target this site , which can be tens or hundreds of microns away ? Wu and Bezanilla have now found that a protein called myosin VIII has a central role in guiding the growing phragmoplast to the cell wall . Myosin VIII is a motor protein that moves along actin filaments . Wu and Bezanilla propose that myosin VIII can guide the expansion of the phragmoplast by pulling microtubules along the actin filaments . The experiments were carried out on two distantly-related plant species , tobacco and a moss called Physcomitrella patens . Similar results were found in both species so it is possible that myosin VIII may play the same role in cell division in all plants . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology",
"cell",
"biology"
] | 2014 | Myosin VIII associates with microtubule ends and together with actin plays a role in guiding plant cell division |
We report that the Gm7068 ( CatSpere ) and Tex40 ( CatSperz ) genes encode novel subunits of a 9-subunit CatSper ion channel complex . Targeted disruption of CatSperz reduces CatSper current and sperm rheotactic efficiency in mice , resulting in severe male subfertility . Normally distributed in linear quadrilateral nanodomains along the flagellum , the complex lacking CatSperζ is disrupted at ~0 . 8 μm intervals along the flagellum . This disruption renders the proximal flagellum inflexible and alters the 3D flagellar envelope , thus preventing sperm from reorienting against fluid flow in vitro and efficiently migrating in vivo . Ejaculated CatSperz-null sperm cells retrieved from the mated female uterus partially rescue in vitro fertilization ( IVF ) that failed with epididymal spermatozoa alone . Human CatSperε is quadrilaterally arranged along the flagella , similar to the CatSper complex in mouse sperm . We speculate that the newly identified CatSperζ subunit is a late evolutionary adaptation to maximize fertilization inside the mammalian female reproductive tract .
Sperm hyperactivation , characterized by a large asymmetric lateral displacement of the flagellum ( Ishijima et al . , 2002 ) , is required for normal mammalian sperm navigation ( Demott and Suarez , 1992 ) , rheotaxis ( Miki and Clapham , 2013 ) , and zona pellucida ( ZP ) penetration ( Stauss et al . , 1995 ) . Calcium influx through the flagellar Ca2+ ion channel , CatSper , triggers hyperactivation ( Carlson et al . , 2003; Kirichok et al . , 2006; Ren et al . , 2001 ) and leads to changes in the flagellar envelope during capacitation ( Chung et al . , 2011; Quill et al . , 2003 ) . In hyperactivated spermatozoa , the transverse flagellar force is larger than the propulsive flagellar force due to the increase in mid-piece curvature ( α angle ) , which enables a larger range of motion and typical figure-of-eight swimming trajectories compared to the nearly straight paths of non-hyperactivated spermatozoa ( Ishijima , 2011 ) . Transverse force facilitates sperm penetration through the cumulus and ZP ( Ishijima , 2011; Yanagimachi , 1966 ) . Spermatozoa from all CatSper-null ( 1–4 or d ) males have smaller α angles than wild-type ( wt spermatozoa upon capacitation ( Chung et al . , 2011; Qi et al . , 2007 ) . Consistently , CatSper-null mutant spermatozoa migrate inefficiently in vivo ( Chung et al . , 2014; Ho et al . , 2009 ) and fail to penetrate the ZP ( Ren et al . , 2001 ) . Sperm rheotax against Fallopian tubular and isthmus fluid flow ( Miki and Clapham , 2013 ) . Rheotactic turning to reorient to directional flow depends on flagellar rolling , not the sperm head or its geometry , as demonstrated by the rheotaxis of headless mouse sperm ( Miki and Clapham , 2013 ) . CatSper channels form unique Ca2+ signaling domains in linearly quadrilateral arrays along the principal piece of sperm flagella . The integrity of these domains is necessary to time and/or maintain hyperactivated motility ( Chung et al . , 2014 ) . Thus , CatSper1-null sperm cannot rheotax due to defects in rolling ( Miki and Clapham , 2013 ) , and presumably exert less lateral force in escaping from epithelial walls ( Ho et al . , 2009 ) or in pushing cumulus cells aside . In general , however , there is a lack of understanding of the steps between CatSper-mediated calcium entry , Ca2+-modified phosphorylation cascades , and the resulting structural changes underlying orchestrated flagellar movement . Here , we reveal that the murine Gm7068 ( C1orf101-like ) and Tex40 genes encode two new subunits of the CatSper ion channel complex , CatSper epsilon ( ε ) and zeta ( ζ ) , respectively . In this study , we focus primarily on CatSperζ’s function . Genetic disruption of mammalian-specific CatSperζ reduces the CatSper current in the sperm flagellum and hyperactivated motility , resulting in severe subfertility . We use high speed video microscopy and digital image analysis to determine swimming trajectory and the flagellar waveform in detail . Surprisingly , abrogation of CatSperζ renders the proximal flagellum inflexible but preserves overall motility , thus resulting in restriction of the 3D flagellar envelope , inefficient sperm rheotaxis in vitro , and delayed sperm migration in vivo . Using super-resolution microscopy , we demonstrated that the structurally distinct CatSper Ca2+ signaling domains along the flagellum ( Chung et al . , 2014 ) becomes fragmented in the absence of CatSperz . We demonstrate that IVF failure of CatSperz-null spermatozoa is partially rescued by using ejaculated sperm recovered from the uterus of mated females , explaining the discrepancy between in vitro and in vivo fertilizing ability . Finally , we show that mouse and human spermatozoa have a similar macroscopic organization of the CatSper complex .
We previously identified seven protein components of the CatSper channel complex ( CatSper1-4 , β , γ , and δ ) from mouse testis using tandem affinity purification ( Chung et al . , 2011 ) . As the most biochemically complex ion channel known to date , it has not been possible to express functional CatSper channels in heterologous systems . This includes many attempts in many cell types , including simultaneous injection of all 7 CatSper mRNAs into Xenopus oocytes ( data not shown ) . Therefore , we continued to seek potential additional components to more thoroughly understand CatSper channel assembly and trafficking . We identified a mouse homolog of human C1orf101 ( C1orf101-like , currently Gm7068 ) ( Figure 1A ) based on its sequence homology to the C-terminal extracellular domain of CatSperδ ( Figure 1—figure supplement 1A ) . This testis-specific gene ( Figure 1—figure supplement 2A ) is predicted to encode a single transmembrane ( TM ) protein ( Figure 1C and Figure 1—figure supplement 1B , C ) . In addition , a small soluble protein encoded by another testis-specific gene , Tex40 ( Figure 1—figure supplement 2A ) , was found to be associated with the CatSper channel complex ( Figure 1B and C , and Figure 1—figure supplement 1D ) . In this study , we refer to the C1orf101-like and Tex40 genes as CatSpere and CatSperz , respectively ( see Molecular Cloning , Materials and methods ) . Like the other CatSper accessory subunits ( Chung et al . , 2011 ) , both CatSpere and CatSperz mRNAs express specifically in germ cells and are detected before CatSper1 expression during postnatal development ( Figure 1—figure supplement 2B , C ) . Moreover , mouse CatSper ε and ζ proteins partition into the testis microsome fraction ( P ) ( Figure 1—figure supplement 2D ) , complex with CatSper1 , and exhibit interdependence with the expression of the other CatSper subunits ( Figure 1D–1F ) . In both human and mouse sperm cells , CatSper ε and ζ proteins are localized to the principal piece of the tails ( Figure 1G and H and Figure 1—figure supplement 2E–G ) . 10 . 7554/eLife . 23082 . 003Figure 1 . CatSper ε and ζ , two new accessory proteins of CatSper channel complex . ( A and B ) Mouse protein sequences of CatSper ε ( A ) and ζ ( B ) . ( C ) Cartoon of the predicted topology of 9 CatSper subunits . ( D ) Association of CatSperε with CatSper1 in testis . ( E and F ) Dependence of CatSper ε ( E ) and ζ ( F ) proteins on CatSper1 in mouse sperm cells . ( G and H ) Confocal fluorescence and the corresponding phase-contrast images of immunostained human CatSperε ( G ) and mouse CatSperζ ( H ) . ( I ) 3D STORM images of mouse CatSperζ in capacitated wt sperm . x-y projection ( left ) and a y-z cross-section ( right ) at 0 . 5 um from the annulus . The color encodes the relative distance from the focal plane along the z axis ( color scale bar in x-y projection ) . ( J ) 3D STORM images of human CatSperε in x-y projection ( left ) , in y-z cross-sections ( right ) . Colors indicate the z positions ( see color scale bar ) . See also Figure 1—figure supplements 1–2 . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 00310 . 7554/eLife . 23082 . 004Figure 1—figure supplement 1 . Identification of CatSper ε and ζ , two novel accessory proteins of the CatSper channel complex , related to Figure 1 . ( A ) Multiple sequence alignment of CatSperδ identifies C1orf101 as δ homologs from various species . Alignments originate ~165 amino acids before the predicted single transmembrane ( TM ) domain , showing the highly conserved region in the proteins’ C-terminal half . Identical ( black ) and similar ( gray ) residues highlighted . ( B ) CatSperε is a protein containing a putative single transmembrane ( TM ) domain localized to the sperm tail . Pairwise alignment of the predicted human ( upper , C1orf101 isoform 1 ) and mouse ( lower , C1orf101-like isoform X2 ) CatSper ε protein sequences . The predicted signal peptide ( SP ) ( Frank and Sippl , 2008 ) and TM domain are boxed . ( C ) von Heijne hydrophilicity plot ( window size = 11 ) of human and mouse CatSperε proteins . ( D ) Sequence alignment between the human ( upper ) and mouse ( lower ) CatSperζ proteins encoded by Tex40 genes . The four peptides from mouse CatSperζ ( identified by mass spectrophotometry from CatSper1 affinity purification but not annotated in the previous study ) ( Chung et al . , 2011 ) , are underlined . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 00410 . 7554/eLife . 23082 . 005Figure 1—figure supplement 2 . Expression of CatSper e and z mRNAs and proteins , related to Figure 1 . ( A ) Tissue expression profile of CatSper e and z . Reverse transcription PCR of CatSper e ( upper ) , z ( middle ) , and G3pdh ( control; lower ) from 12 mouse cDNAs . CatSper e and z are enriched in testis . ( B ) Spatial localization of CatSpere and z mRNA in the testis . Representative fields of in situ hybridization by gene-specific oligonucleotides against CatSper e ( left ) and z ( right ) in mouse testis ( RNAscope ) . ( C ) Temporal expression of CatSper1 , CatSpere , and CatSperz mRNAs during postnatal testis development . The mRNA levels of CatSper1 ( orange ) , CatSpere ( green ) , and CatSperz ( blue ) are measured by real-time RT PCR , normalized to HPRT and expressed as ratios relative to 80-day old adult mouse testis . The data are presented as mean ± SEM . N = 3 . ( D ) Partitioning of CatSperε and CatSperζ in fractionated extracts of testis from wt mice . Both CatSperε and CatSperζ are enriched in the microsomal pellet ( P ) , not in supernatant ( S ) . ( E–G ) Specific recognition of CatSperε and CatSperζ in human spermatozoa . Immunoblotting of ( E ) total human sperm extracts and ( F ) recombinant human CatSperζ by rabbit polyclonal CatSperε ( hε31 ) and CatSperζ ( hζ11 ) antibodies , respectively . d1 and d2 indicate sperm from donors 1 and 2 . ( G ) Confocal image and the corresponding phase-contrast image of CatSperζ in human sperm , immunostained with hζ11 . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 00510 . 7554/eLife . 23082 . 006Figure 1—figure supplement 2—source data 1 . Temporal expression of CatSper1 , CatSpere , and CatSperz mRNAs during postnatal testis development . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 006 Mouse CatSper proteins form a unique pattern of four linear ( ‘racing stripes’ ) Ca2+ signaling domains running down the four quadrants of the principal piece of the flagellum ( Chung et al . , 2014 ) . We examined whether ε and ζ share this distinctive compartmentalization . The antibodies , anti-hε31 , recognizing the N-terminal extracellular region of human CatSperε , and anti-mζ174 , against the very C-terminus of mouse CatSperζ , were suitable for 3D stochastic optical reconstruction microscopy ( STORM ) ( Figure 1F–1H and Figure 1—figure supplement 2E ) . CatSperζ and CatSperε show the apparent four-fold arrangement of CatSper1 , β and δ subunits in mouse ( Figure 1I ) and human ( Figure 1J ) spermatozoa . The lack of functional expression of CatSper channels in heterologous systems requires that genetic manipulation be used to study the function of each component . CatSpere has the same ancient origin at the root of early eukaryotic evolution as those of CatSpers1-4 , b , and g and the same pattern of extensive lineage-specific gene loss as CatSperb and g through metazoan evolution ( Figure 2—figure supplement 1A ) ( Cai et al . , 2014 ) . While CatSper δ and ε share high C-terminal sequence homology ( Figure 1—figure supplement 1A ) , CatSperδ appears later in evolution ( Figure 2—figure supplement 1A ) . In contrast , CatSperζ has no conserved domains and , like hyperactivated motility , is only present in mammals ( Figure 2—figure supplement 1A ) , leading us to speculate that CatSperζ is a required evolutionary adaptation to mammalian fertilization . Based on sequence homology and conservation , we anticipated that deletion of CatSperε would likely be the same as the existing knockout of other CatSper subunits , but deletion of CatSperζ might provide new insights into spermatozoan adaptations to changes concomitant with the evolution of mammalian fertilization . To test this idea , we began by generating a CatSperz-null mouse line from Tex40 gene targeted ES cell clones . Tex40 is a small gene composed of 5 exons that spans only ~3 kb on chromosome 11 ( Figure 2—figure supplement 1B ) . Deletion of exons 2–4 was confirmed in the homozygous null mouse ( Figure 2—figure supplement 1C ) . No CatSperζ protein was detected in Tex40-null spermatozoa by immunoblotting and immunocytochemistry ( Figure 2—figure supplement 1D , E ) . CatSperz-null mutant mice are indistinguishable from their wt or heterozygous ( het ) littermates in appearance , gross behavior , or survival . In addition , no morphological differences were observed by histological examination of testis and epididymis ( data not shown ) . Sperm morphology and epididymal sperm number from CatSperz-null mice were not significantly different from those of 2–3 month old paired heterozygous littermates ( Figure 2—figure supplement 1E and Figure 2—figure supplement 2A ) . CatSperz-null female mice exhibited normal mating behavior and gave birth to litters comparable to those of het females when mated with wt or het males ( Figure 2—figure supplement 2B ) . However , when CatSperz-null male mice were mated with wt or het females , they were severely subfertile: 20% ( 5/25 ) CatSperz-null males were completely infertile over six months ( Figure 2A ) , and progeny of the fertile paternal CatSperz-null mice were significantly fewer in number ( Figure 2B and Figure 2—figure supplement 2B ) . The latency from pair formation to the birth of these offspring from CatSperz-null males was ≥10 days compared to those from wt or het males ( data not shown ) . 10 . 7554/eLife . 23082 . 007Figure 2 . Deletion of the mouse CatSperζ subunit severely impairs male fertility . ( A ) Percent pregnancy rate over three months . ( B ) Average litter size resulting from CatSperz+/- ( 7 . 4 ± 0 . 5 ) and CatSperz-/- ( 4 . 4 ± 0 . 3 ) males . ( C ) Sperm number per egg at the fertilization site 8 hr after 1 hr window-timed coitus with CatSperz+/- ( 0 . 58 ± 0 . 15 ) and CatSperz-/- ( 0 , none ) males , quantified from eggs collected from ampullae . ( B ) and ( C ) Data are mean ± SEM . ****p<0 . 0001 . ( D ) In vivo fertilization rate: Scatter plot with mean % of 2 cell fertilized eggs from CatSperz+/- ( 70% and 94 . 4% ) and CatSperz-/- ( 21 . 3% and 24 . 6% ) mated females at 20 and 27–30 hr after coitus , respectively . ( E ) Head trace of free swimming CatSperz+/- ( top ) and CatSperz-/- ( bottom ) sperm cells at 10 min ( left ) and 90 min ( right ) after capacitation . Traces are from 1 s movies taken at 37°C . ( F ) Flagellar waveform traces . Movies recorded at 200 fps: CatSperz+/- ( top ) and CatSperz-/- ( bottom ) sperm cells attached on glass coverslips before capacitation ( left ) , and 10 min ( middle ) , and 90 min ( right ) after capacitation . Overlays of flagellar traces from two beat cycles are generated by hyperstacking binary images; time coded in color . See also Figure 2—figure supplements 1–2 and Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 00710 . 7554/eLife . 23082 . 008Figure 2—source data 1 . Impaired male fertility in CatSperz-/- mice: pregnancy rate , litter size , sperm number per egg , and in vivo fertilization rate . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 00810 . 7554/eLife . 23082 . 009Figure 2—figure supplement 1 . Generation of CatSperz-/- mice , related to Figure 2 . ( A ) Distribution of CatSper subunits in eukaryotes . ( B ) ES cells ( Project ID: CSD33943 ) from the KOMP Repository were used to produce KO mice . Exons 2–4 deleted by gene trap . ( C and D ) Genotyping ( primers F/R1/R2 ) ( C ) and immunoblotting ( D ) analysis of CatSperz-/- . ( E ) Normal sperm morphology despite the absence of ζ protein in CatSperz-/- spermatozoa . Overlay of confocal images and the corresponding phase-contrast images of mouse sperm cells from CatSperz+/- and CatSperz-/- mice immunostained with mζ174 ( also used in ( D ) and Figure 1F and H . The principal piece labeling is not observed in CatSperz-null sperm , validating the specific subcellular distribution of the signal to the sperm tail . Hoechst dye stains the sperm head DNA ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 00910 . 7554/eLife . 23082 . 010Figure 2—figure supplement 2 . Sperm count and fertility of CatSperz-/- mice; sperm motility analysis and development of P-Tyr , related to Figures 2 and 3 . ( A ) Epididymal sperm count ( mean ± SEM ) from littermates at ages 2–3 months . CatSperz het ( +/- , blue; 2 . 4 ± 0 . 1 ) versus null ( -/- , green; 2 . 4 ± 0 . 2 ) cells ( 107 ) . ( B ) Average litter size from all males in the mating test , grouped by male and female genotype . ( C ) IVF rate calculated by counting fertilized eggs ( 2 cell stage ) 20 and 27–30 hr after coitus . Data are expressed as a scatter plot of mean percentage from CatSperz+/- ( 20 hr , 70 ± 15; 27–30 hr , 94 . 4 ± 2 . 5 ) and CatSperz-/- ( 20 hr , 21 . 3 ± 7 . 6; 27–30 hr , 24 . 6 ± 7 . 4 ) . **p=0 . 0097 ( One-way ANOVA and F test ) . See also Figure 2D . ( D ) Sperm motility parameters measured by computer assisted sperm analysis ( CASA ) from CatSperz het ( +/- ) versus null ( -/- ) male mice . 5 min ( basal , gray ) and 90 min ( capacitated , black ) after incubation in HTF . ALH of CatSperz+/- ( basal , 15 . 5 ± 0 . 4; 120 min , 17 . 3 ± 0 . 4 , p=0 . 0002 ) . Data are mean ± SEM . N = 4 . ( E ) Capacitation-associated protein tyrosine phosphorylation of CatSperz -/- spermatozoa . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 01010 . 7554/eLife . 23082 . 011Figure 2—figure supplement 2—source data 1 . Impaired male fertility in CatSperz-/- mice: sperm count , litter size per genotype , in vivo fertilization rate , and CASA parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 01110 . 7554/eLife . 23082 . 012Figure 3 . ICatSper , but not ATP-activated P2X2 current , is reduced in CatSperz-null spermatozoa . ( A ) CatSperz-/- and ( B ) CatSperz+/- ICatSper . Left panels show the current-voltage relations of monovalent ICatSper in response to voltage ramps at the time points indicated . Right traces are representative time courses of ICatSper measured in the standard bath solution ( 1 , HS ) , ATP-activated P2X2 current ( 2 , ATP ) , and nominally divalent-free solution ( 3 , DVF ) at −100 mV ( gray circles ) and +100 mV ( black circles ) . ICatSper in CatSperz-null sperm cells is ~60% of that recorded from wt . Inward IATP current induced by 100 µM ATP is similar in both phenotypes and indistinguishable from previously published wt IATP ( Navarro et al . , 2011 ) . ( C ) Average ICatSper measured from CatSperz+/- ( −683 ± 77 pA ) and CatSperz-/- ( −426 ± 50 pA ) sperm cells at −100 mV . Data are mean ± SEM . p=0 . 0297 . Cartoon shows the standard pipette solution ( mM ) ; internal Cs used to block K+ currents . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 01210 . 7554/eLife . 23082 . 013Figure 3—source data 1 . Inward CatSper current at −100 mV . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 013 We examined the number of sperm within cumulus oocyte complexes ( COCs ) after copulation and checked in vivo fertilization rates by isolating the COCs and/or embryos from the female ampullae . At 8 hr after coitus , no sperm was found in the COCs when mated with CatSperz-null male mice ( Figure 2C ) . In contrast , the majority of the COCs from CatSperz-het mated females had one or more sperm cells within the complex . When mated with CatSperz-het males , more 2 cell eggs were observed over time after coitus , while the fertilization rate by CatSperz-null males did not change significantly ( Figure 2D and Figure 2—figure supplement 2C ) . These data suggest that CatSperz-null sperm migration is delayed in the female reproductive tract . To understand why CatSperz-null spermatozoa did not efficiently migrate in the female reproductive tract , we first investigated sperm motility using computer assisted sperm analysis ( CASA ) ( Figure 2—figure supplement 2D ) . The percentage of motile spermatozoa was not significantly different and most motility parameters of z-null spermatozoa were comparable to those of z-het sperm cells . However , the characteristic increase of lateral head displacement upon capacitation was not observed in z-null spermatozoa ( Figure 2—figure supplement 2D ) , suggesting that hyperactivated motility was reduced . Ninety minutes after capacitation , there was a less marked difference in swimming trajectories of z-null spermatozoa compared to z-het spermatozoa , supporting this notion ( Figure 2E and Video 1 ) . Further analysis of flagellar amplitude and waveforms of tethered spermatozoa revealed a striking rigidity of z-null spermatozoa from their midpiece to midway down the principal piece ( Figure 2F and Video 2 ) . This phenotype was also observed from hyperactivation-deficient CatSper2-null patients ( Smith et al . , 2013 ) . After incubation under capacitating conditions for 90 min , we observed that z-null spermatozoa beat only at the very distal end of a flagellum ( Video 3 ) . Moreover , CatSperz-null spermatozoa remain bent in the anti-hook direction ( Ishijima et al . , 2002 ) ( Figure 2F , Videos 2 and 3 ) . The anti-hook bend predominates as the pro-hook bend ( initiated by the CatSper-mediated Ca2+ signaling pathway ( Chang and Suarez , 2011 ) is dysregulated in CatSperz-null spermatozoa . 10 . 7554/eLife . 23082 . 014Video 1 . Movement of free swimming CatSperz +/- and -/- spermatozoa; related to Figure 2 . Uncapacitated ( left ) and 90 min capacitated ( right ) spermatozoa were allowed to disperse for 10 min pre-incubation in a 37°C chamber containing HEPES-HTF; free swimming sperm cells recorded within the next 5 min; video rate 20 fps ( 1/5 speed ) , 1 s movies; head trace to track swimming trajectory . ( A ) CatSperz+/- and ( B ) CatSperz-/- spermatozoa . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 01410 . 7554/eLife . 23082 . 015Video 2 . Motility of tethered CatSperz +/- and -/- spermatozoa; uncapacitated , related to Figure 2 . Uncapacitated epididymal spermatozoa in non-capacitating M2 media were tethered to the fibronectin-coated glass bottom dish; sperm motility was recorded at 37°C; video rate 200 fps , 2 s movies . ( A ) CatSperz+/- and ( B ) CatSperz-/- spermatozoa . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 01510 . 7554/eLife . 23082 . 016Video 3 . Motility of tethered CatSperz +/- and -/- spermatozoa; 90 min capacitated , related to Figure 2 . After 90 min incubation in HTF , capacitated epididymal spermatozoa were tethered to a fibronectin-coated glass bottom dish; sperm motility was recorded at 37°C; video rate 100 fps ( 1/2 speed ) , 1 s movies . ( A ) CatSperz+/- and ( B ) CatSperz-/- spermatozoa . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 016 To examine how Ca2+ signaling in CatSperz-null spermatozoa is impaired , we first examined ICatSper , the sperm-specific Ca2+-selective ion current . Since Ca2+ has high affinity to calcium-selective pores ( Almers et al . , 1984 ) , CatSper permeation of monovalents increases when external calcium is removed ( Kirichok et al . , 2006; Navarro et al . , 2007 ) . In divalent-free ( DVF ) solutions , wt spermatozoan ICatSper conducts a large Na+ current , which is completely absent in mice lacking CatSpers1 , 2 , 3 , 4 , or d ( Chung et al . , 2011; Kirichok et al . , 2006; Qi et al . , 2007 ) . However , in CatSperz-null spermatozoa , monovalent CatSper current is present but reduced to ~60% of normal ( −426 ± 50 pA at −100 mV; Figure 3A ) compared to control CatSperz-het spermatozoa ( −683 ± 77 pA at −100 mV; Figure 3B ) . Thus , in the absence of CatSperζ , the CatSper channel complex is still targeted to the flagellar membrane and forms functional channels . We hypothesize that the reduction in CatSper current reflected decreased protein expression levels . P2X receptors are nonselective ion channels gated by purines such as ATP . The ATP-activated cation-nonselective current in the midpiece of murine sperm is mediated by the P2X2 purinergic receptor ( Navarro et al . , 2011 ) . In CatSperz-null spermatozoa , IATP did not differ substantially from heterozygous spermatozoa ( Figure 3A and B ) , supporting the assumption that there is selective down regulation of CatSper channels . Smaller ICatSper explains , in part , the attenuated hyperactivated motility , delayed sperm migration , and male subfertility ( Figure 2A–2E ) . Protein tyrosine phosphorylation ( P-Tyr ) , a hallmark of sperm capacitation , is potentiated and delocalized in CatSper knockout mice ( Chung et al . , 2014 ) or when Ca2+ influx is pharmacologically blocked ( Navarrete et al . , 2015 ) . Upon capacitation , P-Tyr was more prominent in CatSperz-null spermatozoa than wt , but to a lesser extent than CatSper1-null spermatozoa ( Figure 2—figure supplement 2E ) , consistent with the reduced calcium current . It is , however , also possible that an altered arrangement of the CatSper complex and/or its interaction with target proteins in the linear domains could have contributed to these functional deficits . To better understand why ICatSper is reduced in CatSperz-null spermatozoa , we examined levels of protein expression in CatSperz-null spermatozoa ( Figure 4A ) . Expression of other CatSper subunits was detected in CatSperz -null spermatozoa , albeit at 30–60% lower levels than that of wt ( Figure 4B ) , consistent with reduced ICatSper ( Figure 3 ) . This contrasts with the complete absence of other CatSper subunits in CatSper1- and CatSperd-null spermatozoa . mRNA and protein levels of other CatSper subunits were not reduced in the testis of CatSperz-null mice ( Figure 4C and D ) , suggesting that the defect occurs during or after assembly of the protein complex . 10 . 7554/eLife . 23082 . 017Figure 4 . CatSper proteins are reduced in sperm from CatSperz-null mice despite protein expression during spermatogenesis . ( A and B ) Reduced expression of CatSper subunits in sperm cells of CatSperz homozygous null mice compared with their complete absence in CatSper1 and d-null mice . Immunoblotting of ( A ) total mouse sperm extracts and ( B ) protein expression ratio ( z-KO/wt ) of CatSper 1 ( 0 . 5 ± 0 . 1 ) , 3 ( 0 . 6 ± 0 . 08 ) , 4 ( 0 . 3 ± 0 . 07 ) , β ( 0 . 4 ± 0 . 1 ) , δ ( 0 . 4 ± 0 . 07 ) , and ε ( 0 . 5 ± 0 . 03 ) . Data are mean ± SEM . ( C ) Increased expression of CatSper1 and ε in mouse testis in CatSperz-null mutants . ( D ) Quantitative gene expression analysis ( qRT-PCR ) from adult CatSperz-het and null testes: expression ratio ( 2-ddCT ) and mean ddCt ( null-het ) ; TATA binding protein ( TBP ) is the internal control . The expression ratio of CatSper1 ( 1 . 1 ± 0 . 1 ) and all accessory g ( 1 . 0 ± 0 . 2 ) , b ( 1 . 3 ± 0 . 07 ) , d ( 1 . 1 ± 0 . 08 ) , and e ( 0 . 95 ± 0 . 07 ) subunits are mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 01710 . 7554/eLife . 23082 . 018Figure 4—source data 1 . Protein and mRNA expression of CatSper subunits . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 018 Loss of CatSperz resulted in fragmentation of CatSper1 staining on the flagellar membrane and these defects are large enough to be resolved by confocal imaging ( Figure 5A ) . These gaps were not observed in wt/het ( Figure 1H and Figure 2—figure supplement 1E ) or previous wt and CatSper knockout studies ( Chung et al . , 2011; 2014; Liu et al . , 2007; Ren et al . , 2001 ) . 3D STED and 3D STORM super-resolution microscopies clearly demonstrate that structural continuity is interrupted in CatSperz-null spermatozoa - each ‘stripe’ of the CatSper domains is fragmented , while the overall quadrilateral structure is maintained ( Figure 5B and C ) . Cross-sections of the 3D STORM image of wt flagellum show the normal four tight clusters ( Figure 5C , lower ) , represented as four lines in the 2D angular profiles of surface localizations ( Figure 5—figure supplement 1A , E; inset ) as previously observed ( Chung et al . , 2014 ) . In CatSperz-null spermatozoa , however , the four lines in the 2D angular profiles were interrupted ( Figure 5—figure supplement 1B ) . To examine whether the interruptions were periodic , we performed autocorrelation analysis and Fourier transform of STORM images of CatSperz-null sperm flagella ( Figure 5—figure supplement 1C–F ) . Autocorrelation analysis of the CatSperz-null sperm flagella exhibited enhanced periodicity compared to the wt flagellum , with the first peak at ∼850 nm ( Figure 5—figure supplement 1C , D ) . The Fourier transform shows a fundamental frequency of ( 800 nm ) −1 ( Figure 5—figure supplement 1F ) . We assume this thinning of one or more linear domains reflects an underlying structural periodicity that regulates CatSper complex trafficking or membrane insertion . 10 . 7554/eLife . 23082 . 019Figure 5 . CatSperz deletion disrupts the continuity of the CatSper linear domains . Application of different modes of fluorescence microscopy to observe CatSper localization . ( A ) Deconvolved confocal image of α-CatSper1 immunostained CatSperz -null spermatozoa . Scale bar , 1 μm . ( B and C ) 3D super-resolution images of CatSper1 . 3D STED ( B ) and 3D STORM ( C ) images of CatSperz -null ( top ) and wt ( bottom ) sperm flagella , respectively . x-y projection colors encode the relative distance from the focal plane along the z axis . Scale bar , 500 nm . Arrowheads indicate the junction between the mid-piece and the principal piece ( annulus ) of the tail . 3D STORM , y-z cross-section images are shown on the right . Scale bar , 200 nm . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 01910 . 7554/eLife . 23082 . 020Figure 5—figure supplement 1 . Subcellular distribution of immunolocalized CatSper proteins; related to Figure 5 . ( A and B ) Angular distributions ( left ) and profiles ( right ) of the surface-localized molecules of CatSper1 in wt ( A ) and CatSperz -/- spermatozoa ( B ) of Figure 5 . ( C and D ) Averaged autocorrelation functions along the longitudinal axis ( x-axis shown in E , inset ) calculated from multiple CatSper domains in wt ( C ) and CatSperz -/- ( D ) spermatozoa . ( n = 8 ) . The longitudinal axis ( x ) is placed at the flagellar center and the origin at the annulus . ( E and F ) Fourier transformation of the 1D localization distribution shown in ( A ) and ( B ) , showing a fundamental frequency of ( 800 nm ) −1 in CatSperz-/- spermatozoa . ( E , inset ) Cartoon of cylindrical coordinate system for defining the radius and angles of molecular coordinates in STORM images . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 020 Thus far , our results show that CatSperz-null sperm have reduced ICatSper , dysregulated structural continuity of the CatSper Ca2+ signaling domains , beat in an atypical pattern , and are delayed in migrating in the female reproductive tract , resulting in reduced male fertility . Rheotactic guidance for sperm over long distances requires rotational motion during CatSper-mediated hyperactivated motility ( Miki and Clapham , 2013 ) . We thus measured rheotactic parameters and the rotation rate of CatSperz-null spermatozoa with a particular focus on whether their proximal tail rigidity and subsequent low amplitude lateral movement ( Figure 2 ) affects sperm movement . In flow-directed capillary tubes ( Miki and Clapham , 2013 ) , we observed that the rheotactic ability of CatSperz-null spermatozoa was significantly reduced ( Figure 6A and B and Figure 6—figure supplement 1A ) . At all flow rates tested , most motile CatSperz-null spermatozoa were unable reorient to swim against the flow and were swept out of the tube ( Video 4 ) . In contrast , 85% of motile heterozygous spermatozoa displayed rheotactic behaviors by maintaining their position or swimming upstream for more than 2 s of the 9 s period of recording ( Figure 6B and Video 4 ) . 10 . 7554/eLife . 23082 . 021Video 4 . In-capillary rheotaxis of CatSperz +/- and -/- spermatozoa; capacitated , related to Figure 6 . Capacitated epididymal spermatozoa in HTF for 90 min were loaded into the capillary and transferred to a 37°C chamber; sperm cells swimming against the flow and down were recorded; video rate 33 fps , 9 s movies . ( A ) CatSperz+/- and ( B ) CatSperz-/- spermatozoa . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 02110 . 7554/eLife . 23082 . 022Figure 6 . CatSperz-null sperm rheotax poorly due to low torque . ( A ) In-capillary sperm rheotaxis . Rheotactic ability is reduced in sperm lacking CatSperz at all flow rates tested ( 12–65 µm/s ) . ( B ) Rheotactic sperm cells are expressed as the % of total motile spermatozoa counted from 9 s-movies ( CatSperz+/- , n = 32; CatSperz-/- , n = 24 ) . Data are expressed in scatter plots; mean ± SEM ( colored bars ) of CatSperz+/- ( 88 ± 2 ) and CatSperz-/- ( 24 ± 4 ) as well as median with interquartile ranges ( black boxes ) of CatSperz+/- ( 96 , IQR 78–100 ) and CatSperz-/- ( 22 . 5 , IQR 0–40 ) . ****p<0 . 0001 . ( C ) Trajectory of free-swimming sperm in 0 . 3% methyl cellulose . Movies were taken at 50 fps to compare CatSperz+/- ( left ) and CatSperz-/- ( right ) sperm cells; bottom of glass dish , 37°C , 5 min after incubation in capacitation medium ( HTF ) . Overlays of flagellar traces ( 20 frames , 2 s movie ) are generated by hyperstacking binary images with gray intensity scale; end frame in black . Arrows indicate sperm heads in each trace . ( D ) Sperm rotation rate from CatSperz+/- ( 5 min , 2 . 4 ± 0 . 2; 90 min , 3 . 1 ± 0 . 2 , p=0 . 0064 ) and CatSperz-/- ( 5 min , 3 . 8 ± 0 . 2; 90 min , 3 . 4 ± 0 . 3 ) males after incubation in HTF . The sperm rotation rate is calculated as previously reported ( Miki and Clapham , 2013 ) . Data are mean ± SEM . ****p<0 . 0001 . ( E and F ) IVF with epididymal and/or ejaculated CatSperz+/- and CatSperz-/- spermatozoa . 2 cell stage eggs were counted 24 hr after insemination . ( E ) IVF rate with cumulus-intact oocytes from CatSperz+/- ( epididymal , 71 ± 6 ) and CatSperz-/- ( epididymal , 0 . 7 ± 0 . 7; ejaculate , 20 ± 6 , p=0 . 0051 ) . ( F ) IVF rate of cumulus-free/ZP-intact eggs with ( CatSperz+/- , 80 ± 8; CatSperz-/- , 7 ± 7 , p=0 . 0005; wt , 78 ± 6 ) or without ( CatSperz+/- , 88 ± 6; CatSperz-/- , 0 . 8 ± 0 . 8 , p=0 . 0002 ) glutathione-containing ( GSH; 2 mM ) media . Data are mean ± SEM . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 02210 . 7554/eLife . 23082 . 023Figure 6—source data 1 . In-capillary sperm rheotaxis and in vitro fertilization . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 02310 . 7554/eLife . 23082 . 024Figure 6—figure supplement 1 . CatSper-mediated Ca2+signaling and development of the flagellar envelope; related to Figure 6 . ( A ) In-capillary sperm rheotaxis . Number of rheotactic sperm cells from each 9 s movie at all flow rates in the range of 12–65 µm/s; from Figure 6A . Data are expressed as a scatter plot with mean ± SEM ( colored bars ) from CatSperz+/- ( 7 . 1 ± 0 . 6 ) and CatSperz-/- ( 1 . 1 ± 0 . 2 ) . ( B ) Working model illustrating the integrity of CatSper Ca2+ signaling domains and their relation to flagellar envelopes during sperm rotation . In wt spermatozoa , the CatSper channel forms four linear continuous Ca2+ signaling domains confined to the principal piece of the flagella . Ca2+ entry through the CatSper channels potentiates sperm rotation during capacitation ( Miki and Clapham , 2013 ) . With the resulting increased asymmetry and change in wave amplitude , the flagellar envelope is mapped out as a 3-dimensional cone in space , orienting sperm into the flow . Deficits in ICatSper and loss of the continuity of the linear domains in CatSperz-/- null spermatozoa compromise Ca2+ signaling and result in rigidity in the proximal region . The inflexibility of CatSperz-null spermatozoa from midpiece to halfway through the principal piece constrains the flagellar envelope to a narrower , rod like spatial map . The still active distal tail rotation then drives the more static rod-like structure . This causes the sperm to rotate faster but with less torque , thereby inefficiently orienting them into the flow and yielding less force in orthogonal directions needed to push aside the cumulus cells . ( C and D ) Flagellar waveform traces of spermatozoa . Movies recorded at 200 fps: CatSperz+/- ( top ) and CatSperz-/- ( bottom ) sperm cells tethered on glass coverslips . Overlays of flagellar traces from two beat cycles are generated by hyperstacking binary images; time coded in color . ( C ) Spermatozoa capacitated under low ( 0 . 5 , 0 . 1 mM ) or high ( 4 mM ) extracellular calcium . ( D ) Sperm motility after Ca2+ transients by A23187 treatment followed by incubation under capacitation conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 024 We next examined the rotational motion of CatSperz-null spermatozoa . At high viscosities ( 0 . 3% methyl cellulose ( MC ) , cP = 6 . 7 ) , uncapacitated z-het spermatozoa swim in circles ( Figure 6C , left and Video 5 ) , while capacitated z-het spermatozoa swim in a more linear path as they rotate around a longitudinal axis , like wt spermatozoa ( Video 6 ) ( Miki and Clapham , 2013 ) . Interestingly , z-null spermatozoa exhibit linear migration as they can rotate along the tail axis regardless of capacitating conditions , even at higher viscosities ( Figure 6C , right and Videos 5 and 6 ) . Indeed , uncapacitated CatSperz-null spermatozoa rotate ~50% faster than z-het spermatozoa ( Figure 6D ) . This indicates that CatSperz-null spermatozoa have less lateral motion and are subject to less torque by the moving stream . Spatially , the spermatozoa trace out a less conical 3D envelope ( Figure 2F and Figure 6—figure supplement 1B ) . In short , the rigidity of the CatSperz-null sperm proximal tail constrains its motion to that of a propeller-driven rod . 10 . 7554/eLife . 23082 . 025Video 5 . Movement of CatSperz +/- and -/- sperm in viscous medium; uncapacitated , related to Figure 6 . Uncapacitated spermatozoa were allowed to disperse for 10 min pre-incubation in a 37°C chamber containing HEPES-HTF supplemented with 0 . 3% methylcellulose; swimming sperm cells were recorded within the next 5 min; video rate 50 fps , 2 s movies . ( A ) CatSperz+/- and ( B ) CatSperz-/- spermatozoa . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 02510 . 7554/eLife . 23082 . 026Video 6 . Movement of CatSperz +/- and -/- sperm in viscous medium; 90 min capacitated , related to Figure 6 . Spermatozoa capacitated in HTF were allowed to disperse for 10 min pre-incubation in a 37°C chamber containing HEPES-HTF supplemented with 0 . 3% ( left ) , 0 . 4% ( middle ) , or 0 . 5% ( right ) methylcellulose; swimming sperm cells were recorded within the next 5 min; video rate 50 fps , 2 s movies . ( A ) CatSperz +/- and ( B ) CatSperz-/- spermatozoa . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 026 We next examined the relation between external calcium entry and sperm function . First , we tested whether increasing extracellular [Ca2+] could rescue z-null sperm motility . After incubation for 90 min with a two-fold greater [Ca2+] , most CatSperz-null sperm remain bent in the anti-hook orientation with a rigid proximal tail ( Figure 6—figure supplement 1C , z ( -/- ) middle , and Video 7 ) . A few z-null spermatozoa partially recover , bending occasionally in the pro-hook direction with hyperactivated motility ( Figure 6—figure supplement 1C , z ( -/- ) right , and Video 7 ) . Conversely , a 20-fold reduction of extracellular [Ca2+] alone did not significantly alter the flagellar waveforms of z-het spermatozoa within 90 min ( Figure 6—figure supplement 1C and Video 8 ) . 10 . 7554/eLife . 23082 . 027Video 7 . Motility of tethered CatSperz-/- sperm in high external calcium; 90 min capacitated , related to Figure 6—figure supplement 1 . After 90 min incubation in Ca2+-HTF under capacitating conditions , CatSperz-/- spermatozoa were tethered to a fibronectin-coated glass bottom dish; sperm motility was recorded within the next 5 min at 37°C; video rate 100 fps ( 1/2 speed ) , 1 s movies . ( A ) 2 Ca2+-HTF and ( B ) 4 Ca2+-HTF ( in mM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 02710 . 7554/eLife . 23082 . 028Video 8 . Motility of tethered CatSperz+/- sperm in low external calcium; 90 min capacitated , related to Figure 6—figure supplement 1 . After 90 min incubation in Ca2+-HTF under capacitating conditions , CatSperz+/- spermatozoa were tethered to a fibronectin-coated glass bottom dish; sperm motility was recorded within the next 5 min at 37°C; video rate 100 fps ( 1/2 speed ) , 1 s movies . ( A ) 2 Ca2+-HTF , ( B ) 0 . 5 Ca2+-HTF and ( C ) 0 . 1 Ca2+-HTF ( in mM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 028 A transient Ca2+ pulse induced by Ca2+ ionophore , A23187 , significantly reduces the time required for wt sperm to develop hyperactivated motility ( Tateno et al . , 2013 ) . Moreover , a short ( 10 min ) exposure to A23187 can rescue defects in hyperactivated motility and the fertilizing capability of CatSper1-null sperm in vitro ( Navarrete et al . , 2016 ) . We tested the relation of calcium transients to hyperactivated motility in CatSperz-het and null sperm . In z-het spermatozoa , an A23187-induced Ca2+ pulse followed by washout , enables full hyperactivation , characterized by wide lateral displacement with large midpiece α angle within 30 min ( Figure 6—figure supplement 1D and Videos 9 and 10 ) . However , in z-null sperm , the same treatment improved the flexibility of the proximal flagella , particularly in the principal piece , but the midpiece remained largely inflexible ( Figure 6—figure supplement 1D and Videos 9 and 10 ) . Building on our previous work , the present study suggests that calcium entry through CatSper channels has time-dependent , complex effects on the coordination of motility and that loss of CatSperz results in reduced ICatSper , changes in calcium signaling , and structural alterations of the flagellum . 10 . 7554/eLife . 23082 . 029Video 9 . Motility of tethered CatSperz +/- and -/- sperm after A23187 treatment; 5 min after wash , related to Figure 6—figure supplement 1 . Spermatozoa treated with 20 µM A23187 in H-HTF for 10 min were washed and incubated in HTF under capacitating conditions for 5 min; sperm were tethered to a fibronectin-coated glass bottom dish and the motility was recorded in H-HTF within the next 5 min at 37°C; video rate 100 fps ( 1/2 speed ) , 1 s movies . ( A ) CatSperz+/- and ( B ) CatSperz-/- spermatozoa . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 02910 . 7554/eLife . 23082 . 030Video 10 . Motility of tethered CatSperz +/- and -/- sperm after A23187 treatment; 30 min after wash , related to Figure 6—figure supplement 1 . Spermatozoa treated with 20 µM A23187 in H-HTF for 10 min were washed and incubated in HTF under capacitating conditions for 30 min; sperm were tethered to a fibronectin-coated glass bottom dish and the motility was recorded in H-HTF within the next 5 min at 37°C; video rate 100 fps ( 1/2 speed ) , 1 s movies . ( A ) CatSperz+/- and ( B ) CatSperz-/- spermatozoa . DOI: http://dx . doi . org/10 . 7554/eLife . 23082 . 030 We performed in vitro fertilization ( IVF ) to determine how the low rotational torque generated by CatSperz-null spermatozoa affects sperm-egg interactions . We found that these spermatozoa cannot fertilize cumulus-intact oocytes ( Figure 6E ) , but could dissociate the cumulus cell layers and bind to the ZP ( data not shown ) . Cumulus removal did not change the fertilization rate of ZP intact oocytes by CatSperz-null spermatozoa . Furthermore , this rate was only marginally enhanced by destabilization of the ZP by 2 mM glutathione ( Figure 6F ) ( Miyata et al . , 2015 ) . This indicates that reduced hyperactivated motility alone does not explain the failure of CatSperz-null spermatozoa in IVF . One possibility is that the kinetics of capacitation in vitro is different from that in vivo , resulting from fluctuations in timing or amplitude of known factors ( e . g . , HCO3 , pH ) or from unknown factors present in seminal and/or female fluids . We then compared IVF rates with ejaculated and epididymal spermatozoa of CatSperz-null mice . When ejaculated spermatozoa flushed from the uterus of the mated females were used in IVF trials , 20% of oocytes incubated with z-null spermatozoa developed into two-cell embryos ( Figure 6E ) . This compares to 50% of oocytes incubated with z-het ejaculated ( data not shown ) , or epididymal sperm . Thus , additional factors may be functionally relevant in in vivo fertilization .
Sperm hyperactivation and normal fertility in mammals requires the unique CatSper channel complex . With four distinct pore-forming gene products ( CatSper 1–4 ) and , now , five accessory subunits ( β , γ , δ , ε , and ζ ) , the CatSper channel is the most complex of known ion channels . This may reflect the relatively high evolutionary pressure on spermatozoan evolution ( Swanson and Vacquier , 2002; Torgerson et al . , 2002 ) , and various adaptations to different modes of fertilization . Like many gamete-specific proteins and the other CatSper proteins reported so far ( Cai and Clapham , 2008; Chung et al . , 2011 ) , mouse and human CatSper ε and ζ show signs of rapid evolutionary change with only 50% and 45% amino acid sequence identity , respectively . In particular , the sequence regions outside TM segments and the pore loop of CatSper proteins are poorly conserved across species , indicating these regions possibly convey species-specific modulation of flagellar motility ( Miller et al . , 2015 ) . This is illustrated by striking differences in progesterone-elicited ICatSper responses in mouse and human ( Lishko et al . , 2011 ) . Here we have shown that CatSper ε and ζ are components of the highly organized CatSper complex , that CatSperζ is required for proper continuity of this complex along the flagellum , and that loss of ζ alters hyperactivation waveforms and reduces fertilizing capacity . The conservation pattern of the lineage-specific gain and loss of the CatSpere gene is identical to those of b and g , suggesting that they likely belonged to an ancient CatSper channel Ca2+ signaling network before the divergence of unikonts and bikonts . Since their protein expression is strictly interdependent , we speculate that CatSpere-null mice will have the phenotype of CatSper1-4 , or d-null mice . In contrast , CatSperz is conserved only in mammals , suggesting that this protein imparts some adaptation , perhaps as a method enabling rheotaxis in the mammalian female reproductive tract . Interestingly , although CatSperζ has no putative transmembrane domains , it is localized in the same quadrilateral pattern as other CatSpers , but is not present elsewhere in sperm . An intriguing aspect of our observations is that , unlike CatSper1-4 and d-null mice , which produce complete infertility , CatSperz-null males exhibit an incomplete loss of fertility . The CatSper current is reduced in CatSperz-null spermatozoa , and may have similar permeation properties ( likely dominated by the CatSper1-4 pore subunits ) , but the effects of CatSperz on channel gating remain to be determined in future studies . This is reminiscent of the non-spermatozoan , voltage-gated Cav channel auxiliary subunits , which are not required for expression but modulate expression levels and gating ( Catterall et al . , 2005 ) . Most tantalizing is the thinning and disruption of the linear CatSper signaling domains at repeat intervals in the absence of ζ . Further detailed examination via mutagenesis experiments has been stymied by our inability to heterologously express functional CatSper channels . New rapid genome editing techniques should enable more mice to be generated that will further the study of CatSper trafficking , subunit interactions , and localized signaling pathways . Functional Ca2+ signaling domains are common adaptations in many biological systems , such as synapses and muscle . They enable specific and fast triggering of downstream events ( Clapham , 2007 ) . CatSper channels are compartmentalized into a unique multilinear arrangement and form Ca2+ signaling nanodomains with other Ca2+ signaling molecules along the sperm flagellum ( Chung et al . , 2014 ) . The mechanisms involved in the delivery of the CatSper channels to these specific domains are currently unknown , and we suspect will be as interesting and complex as those in primary and motile cilia ( Sung and Leroux , 2013 ) . We found that abrogation of CatSperz not only retards targeting of the CatSper complex to flagella , but also disrupts continuity of the linear domains , resulting in repeated fragmented domains with ~800 nm periodicity . In order for CatSper domains to form and function properly , interactions are needed between the CatSper channel complex in the flagellar membrane and the underlying cytoskeletal proteins . One speculation is that CatSperζ might adapt to cytoskeletal structures that traffic , distribute , and enable membrane insertion of CatSper . The fibrous sheath ( FS ) , a cytoskeletal structure unique to the mammalian sperm flagellum , defines the extent of the tail’s principal piece , in which all the CatSper proteins are specifically localized . The FS closely lies under the plasma membrane and its two longitudinal columns are connected by circumferential ribs . Immunogold electron microscopy demonstrated that the CatSper channels are distributed on the end of ribs , where they merge with the column ( Chung et al . , 2014 ) . It seems likely that the timing of occurrence and localization of CatSper Ca2+ signaling domains is coordinated with the assembly of FS proteins along the axoneme . The column appears early in spermiogenesis , forming from the distal tip of the tail along the axoneme , followed by subsequent rib formation in the opposite direction ( Oko , 1998; Oko and Clermont , 1989 ) . Based on scanning electron micrographs ( Danshina et al . , 2010; Miki et al . , 2004 ) , we find that the distance between ribs is about 800 nm in mouse spermatozoa . Thus , it seems likely that the repeated disruption in the absence of CatSperz is related to rib spacing of the FS . Genetic abrogation of CatSper disrupts hyperactivated motility as manifested by changes in movement symmetry , amplitude , and rolling ( Carlson et al . , 2003; Chung et al . , 2011; Miki and Clapham , 2013; Qi et al . , 2007 ) . Here we report that the flagellar envelope is significantly altered in the absence of CatSperz , in part due to the inflexibility of the proximal tail . We previously reported that the catalytic subunit of calcineurin , PP2B-Aγ , expresses throughout the tail but localized to the CatSper quadrilateral structures and axoneme ( Chung et al . , 2014 ) . In CatSper1-null spermatozoa , PP2B-Aγ remains localized primarily to the axoneme but disappears from the quadrilateral structures . Recently , a similar but not identical phenotype ( inflexible midpiece , reduced hyperactivated motility , and impaired ZP penetration ) was reported in testis-specific calcineurin Ppp3cc-null and Ppp3r2-null spermatozoa ( Miyata et al . , 2015 ) . Note that the principal piece of both CatSper1-null and Ppp3cc-null spermatozoa are not rigid . The integrity and distribution of CatSper channels in Ppp3cc-null spermatozoa remain to be examined and may clarify midpiece/principal piece disparities . In any case , inflexibility in the proximal regions of flagellum results in a flagellar envelope approximated as a rod with a distal propeller . The sperm can rotate faster but the smaller lateral deviation reduces torque . This limits the sperm’s ability to orient into the flow , as well as penetrate the cumulus and ZP . Gene-manipulated mice highlight the importance of in vivo observations and have reshaped the landscape of fertilization science ( Okabe , 2015 ) . In vitro capacitation and fertilization systems underpin much of the study of sperm motility and fertilization potential . While ejaculated sperm are preferred for fertilization studies in larger animals and humans , epididymal sperm are commonly used in genetically tractable mouse studies . Notably , these sperm are not exposed to accessory sex gland secretions and female fluids . This may explain why CatSperz-null spermatozoa are completely infertile in an IVF setting ( COCs ) , but in vivo are merely subfertile . Perhaps natural modulators , absent in epididymal sperm IVF studies , partially rescue the fertilizing potential of CatSperz-null spermatozoa by activating Ca2+ signaling activity . Interestingly , a transient pulse of Ca2+ can greatly reduce the capacitation time required for wt sperm to develop hyperactivated motility ( Tateno et al . , 2013 ) . Moreover , Navarrete et al recently demonstrated that a short exposure to A23187 rescued the defects in motility and fertilizing capability of CatSper1-null sperm in vitro ( Navarrete et al . , 2016 ) . These independent studies were interpreted to mean that the initial priming by Ca2+ influx , perhaps above a certain threshold , is essential for sperm function . However , the linear quadrilateral CatSper complexes are not present in CatSper1-null spermatozoa and in CatSperz-null spermatozoa are disrupted by gaps . We hypothesize that the linear quadrilateral structure in vivo likely maintains , regulates , and distributes CatSper Ca2+ signaling during hyperactivated motility . But it is important to point out that alterations in the structure should also result in changes in mechanical properties , movement of the flagellum , distribution of entering calcium , and downstream kinase activity and the motor elements they regulate . This complexity is illustrated in vivo sperm swimming trajectories , which are modulated by switching between pro- and anti-hook beating patterns . In the absence of CatSperζ , anti-hook beating predominates . Pro-hook motions are associated with intact CatSper-mediated Ca2+ signaling pathways ( Chang and Suarez , 2011 ) . Finally , ejaculated sperm display more pro-hook hyperactivation than epididymal sperm ( Li et al . , 2015 ) . Future areas for investigation are the functional positioning of the remaining accessory subunits of the CatSper channel in assembly and domain organization , the testing of potential modifiers present in accessory sex gland secretions that may activate CatSper channels , and the determination of Ca2+ dependent molecules in the axoneme which eventually determine flagellar bending and its envelope . CatSperz-null mice , which are hypomorphic to the null-mutation of other CatSper genes with abrogated hyperactivation , and newly expanding animal models from recent advances in genome editing will serve as a foundation to this end . Advanced imaging techniques with higher time and spatial resolution will be necessary to carry this out . The present results also suggest that alterations of Ca2+ current and/or dysregulated downstream Ca2+ signaling affecting dynamic structures may be sufficient to compromise sperm function . CatSper’s unique composition and central role in hyperactivated motility make it an ideal target for contraception .
CatSper1 and d-null mice were previously described ( Chung et al . , 2011; Ren et al . , 2001 ) . Lines were backcrossed and maintained on a C57BL/6 background . WT C57BL/6 male , B6D2F1 female ( Jackson laboratory , Bar Harbor , ME ) , and CD1 ( Charles River Laboratories , Wilmington , MA ) female mice were purchased . Mouse caudal epididymal sperm were collected by swim-out in HEPES buffered saline ( HS ) containing ( in mM ) : 135 NaCl , 5 KCl , 2 CaCl2 , 1 MgSO4 , 20 HEPES , 5 glucose , 10 lactic acid , 1 Na pyruvate , pH 7 . 4 ( with NaOH ) ( Chung et al . , 2011 ) . To induce capacitation in vitro , sperm cells were incubated ( 2 × 106 cells ml−1 ) in human tubular fluid ( HTF ) media ( in mM ) : 102 NaCl , 4 . 7 KCl , 2 CaCl2 , 0 . 2 MgCl2 , 0 . 37 KH2PO4 , 2 . 78 glucose , 18 . 3 lactic acid , 0 . 33 Na pyruvate , 25 HCO3- and 4 mg ml−1 BSA ) ( Millipore ) for 90 min at 37°C ( 5% CO2 ) . All experiments using human samples were approved by the Committee of Clinical Investigation , Boston Children’s Hospital CCI/IRB ( IRB-P00000538 ) . Human semen samples were obtained from fertile donors . Human spermatozoa were collected by the swim-up method with the use of modified human tubal fluid medium ( HTF ) . HEK293T cells were purchased from ATCC . In this study , they were used to overexpress recombinant human CatSperζ in order to test antibodies . The cells were tested negative for mycoplasma and validated as of human origin . The identity was authenticated by confirming their negative expression of testis-specific genes including CatSper . The cell line was cultured in DMEM/F12 containing 10% FBS . Rabbit polyclonal CatSper1 , CatSper4 , β , and δ antibodies were previously described ( Chung et al . , 2011; Ren et al . , 2001 ) . To produce antibodies to new CatSper subunits , peptides were synthesized and conjugated to KLH carrier protein ( Open Biosystems , Lafayette , CO ) as follows: mouse CatSperε , 968–985 ( αm-ε968: RQFIIEPLHKRPAKQKKN ) ; mouse CatSperζ , 174–195 ( αm-ε174: GYIEGIRKRRNKRLYFLDQ ) ; human CatSperε , 31–50 ( αh-ε31: RIFSTRSTIKLEYEGTLFTE ) ; and human CatSperζ , 11–29 ( αh-ζ11: KSSDRQGSDEESVHSDTRD ) . Antisera were affinity purified on the immobilized resin of the corresponding peptide ( Amino Link Plus or Sulfo Link Plus ) ( Pierce , Waltham , MA ) . Anti-phosphotyrosine ( clone 4G10 ) , anti-Flag ( clone M2 ) , anti-calmodulin ( 05-173 ) and anti-acetylated tubulin ( T7451 ) antibodies were from EMD Millipore ( Germany ) . All chemical compounds were from Sigma-Aldrich ( St . Louis , MO ) unless indicated . Annotated orthologs in the NCBI gene database ( http://www . ncbi . nlm . nih . gov/gene/ ) and/or homologous amino acid sequences of reported protein databases were screened in 17 eukaryotes for the presence of genes for CatSper auxiliary subunits . Non-annotated orthologs in the NCBI gene database were identified by comparing sequences of the annotated orthologs to those in the protein database of species by Phmmer implemented on HMMER 3 . 1 ( default option , http://hmmer . org/ ) . The longest amino acid sequences among all the isoforms of the orthologs annotated in each species and protein sequence databases from 15 eukaryotes , except human and mouse , were downloaded from the NCBI Genome database ( http://www . ncbi . nlm . nih . gov/genome; Tinamus guttatus , GCA000705375 . 2; Anolis carolinensis , GCA000090745 . 2; Salmo salar , GCA000233375 . 4; Callorhinchus milii , GCA000165045 . 2; Branchiostoma floridae , GCA000003815 . 1; Caenorhabditis elegans , GCA000002985 . 3; Crassostrea gigas , GCA000297895 . 1; Exaiptasia pallida , GCA001417965 . 1; Trichoplax adhaerens; GCA000150275 . 1 , Salpingoeca rosetta , GCA_000188695 . 1 ) , Ensembl genome browser ( http://ensembl . org; Strongylocentrotus purpuratus , GCA000002235 . 2; Drosophila melanogaster , GCA000001215 . 4; Thecamonas trahens , GCA000142905 . 1 ) , and JGI genome portal ( http://genome . jgi . doe . gov; Allomyces macrogynus; Aurantiochytrium limacinum ) . Aligned phmmer hits of expected values <10−10 were considered as candidate orthologs of the corresponding CatSper subunits in each species . PCR was performed according to standard protocols using a commercial multiple panel cDNA template ( MTC ) , Clontech ) . PCR primers amplified Gm7068 ( forward: 5′-CTATGGCTCAAGTGTAATGACC-3′ , reverse: 5′-GCTCTTATTGAATCCTCGAACC-3′ ) , Tex40 ( forward: 5′-GAAACAGGATTCGCAAGTACAG-3′ , reverse: 5′-TCGTGGACCTATATGTGATGAG-3′ ) using mouse GAPDH ( forward: 5′-TGAAGGTCGGTGTGAACGGATTTGGC-3′ , 5′-ATGTAGGCCATGAGGTCCACCAC-3′ ) as a control . The initial mouse Tex40 cDNA sequence ( NM_001039494 ) was identified from database searches using novel peptide sequences from MS . The full-length human Tex40 cDNAs was obtained by PCR with primers ( forward: 5′-GGGCAGAACCATGGAGGAAA-3′ , reverse: 5′-AGGACTCAAATTCCACTCGGATG-3′ ) using the human testis cDNA library ( Clontech ) . Sequencing the TOPO-cloned PCR products into pCR4-TA ( Invitrogen ) confirmed the full-length human Tex40 ORF , which was subcloned into pCMV-Tag2A ( Stratagene ) to express recombinant N-terminal Flag-tagged human CatSperζ in mammalian cells . Mouse Gm7068 was identified by homologous amino acid sequence to C-terminal CatSperd ( Tmem146 ) . There are six transcript variants ( Almers et al . , 1984 ) ; XM_006497083 , 2; XM_006497084 , 3; XM_017314031 , 5: XM_006497085 , 6; XM_017314033 , and 8; XM_006497087 ) . Variants 1 , 3 , 5 , 6 , and 8 are predicted to encode polypeptides with the same C-terminal sequence that can be detected by anti-mε−968 . Among them , the predicted polypeptides from longer splicing variant 1 ( isoform X1; XP_006497147 , 985 aa ) and variant 3 ( isoform X3; XP_017169520 , 914 aa ) are consistent with the apparent molecular weight of the band observed in testes microsomes ( Figure 4C and Figure 1—figure supplement 2D ) . The predicted polypeptides from shorter variant 5 ( isoform 4; XP_006497148 , 805 aa ) and variant 6 ( isoform 5; XP_017169522 , 770 aa ) are consistent with that of the band detected in CatSper1-IP from testis and total sperm lysate ( Figures 1D , E and and 4A ) . It is likely that mouse Gm7068 expresses at least four potential splice variants that can encode protein isoforms and/or undergo cleavage during spermatogenesis . In situ hybridization experiments were carried out with an RNAscope ( Advanced Cell Diagnostics , Newark , CA ) . Testes from three month old wild-type mice were fixed in 10% ( vol/vol ) neutral-buffered formalin at room temperature for 24 hr , dehydrated , and embedded in paraffin . Paraffin sections ( 10 μm thick ) were processed according to the manufacturer’s instructions for in situ detection in the Rodent Histopathology Core Facility at Harvard Medical School . Sequences of the probes used in this study are: Gm7068 ( XM_982472 . 3 , 645–1072 ) and Tex40 ( NM_001039494 . 2 , 41–456 ) . After the DAB ( 3 , 3 , -diaminobenzidine ) reaction , slides were counterstained using hematoxylin . Real-time PCR was carried out with first strand cDNAs ( iScript cDNA Synthesis ) ( Bio-Rad , Hercules , CA ) synthesized from 2 µg total mouse testis RNA using the SYBR Green ( iTaq Universal SYBR Green Supermix ) ( Bio-Rad; CFX96 ) . Quantitative analysis by the ddCt method employed c-Jun as an amplification control . Three independent sets of experiments were performed to calculate fold changes ( 2-ddCt ) of CatSpers mRNA . The primers used for qRT-PCR were: CatSper1 ( forward: 5′-CTGCCTCTTCCTCTTCTCTG-3′ , reverse: 5′-TGTCTATGTAGATGAGGGACCA-3′ ) , CatSperb ( forward: 5′-CCTTA TTGACCAAGAAACAGAC-3′ , reverse: 5′-TGAAACCCATATTTGACTGCC-3′ ) , CatSperg ( forward: 5′-TGAGCAATAGAGGTGTAGAC-3′ , reverse: 5′-CAGGA TGTAGAAGACAACCAG-3′ ) , CatSperd ( forward: 5′-GCTGACATTTCTGTGTATCTAGG-3′ , reverse: 5′-CTGATATACCTTCCAATTTACGCC-3′ ) , CatSpere ( forward: 5′-GTCTCATGCTTCTTCAGTTCC-3′ , reverse: 5′- CAGAAGTTCCTTGTCCATCAC-3′ ) , CatSperz ( forward: 5′-GAGACCTCCTTAGCATCGTC-3′ , reverse: 5′-TCGTGGACCTATATGTGATGAG-3′ and c-Jun ( forward: 5′-CTCCAGACGGCAGTGCTT-3′ , reverse: 5′-GAGTGCTAGCGGAGTCTTAACC-3′ ) . Testes ( 200 mg , normally two testicles ) from 8- to 12-wk-old male mice were homogenized on ice using a Dounce homogenizer in 2 mL 0 . 32 M sucrose solution with protease inhibitor cocktails ( Roche ) . The tissue suspension was centrifuged at 300 g for 10 min at 4°C and the supernatant was then transferred to an ultra-speed centrifuge tube . The microsome faction was isolated by centrifuging the tube at 105 , 000 g for 60 min . Mouse sperm total protein was prepared as described before ( Chung et al . , 2011; 2014 ) . For total protein from human spermatozoa , purified swim-up sperm were then lysed ( 0 . 1% SDS , 0 . 5% sodium deoxycholate , 1 mM DTT , 1 mM EDTA in PBS with protease inhibitors ) followed by sonication for 5 min and centrifuged at 15 , 000 g for 10 min . The supernatants were further denatured by adding DTT to 10 mM and heated at 75°C for 10 min before SDS-PAGE . For immunoprecipitation , the testis microsome pellet was resuspended in 10 mL 1% Triton X-100 in PBS with protease inhibitors ( Roche ) . The suspension was rocked at 4°C for 1 hr and then centrifuged at 15 , 000 g for 30 min . 1 . 5 mL of the solubilized testis microsome were mixed with 1–2 μg antibody and 25 μL Protein A/G-bead slurry ( Santa Cruz Biotechnology ) at 4°C overnight . The IP products were finally eluted in 50 μL LDS loading buffer containing 50 mM DTT . Antibodies used for Western blotting were rabbit anti-mouse CatSperε ( αm-ε968; 1 . 6 μg/mL ) , mouse CatSperζ ( αm-ε174; 2 . 7 μg/mL ) , human CatSperε ( αh-ε31; 2 . 7 μg/mL ) , human CatSperζ ( αh-ζ11: 1 μg/mL ) . Monoclonal anti-phosphotyrosine ( clone 4G10; 1 μg/mL ) , anti-Flag ( clone M2; 1 μg/mL ) , anti-calmodulin ( 05–173 , 1 μg/mL ) , and anti-acetylated tubulin ( T7451 , 1: 20 , 000 ) . Secondary antibodies were anti-rabbit IgG-HRP ( 1:10 , 000 ) and anti-mouse IgG-HRP ( 1: 10 , 000 ) from Jackson ImmunoResearch ( West Grove , PA ) . Caudal epididymal mouse sperm cells attached to glass coverslips were fixed in 4% paraformaldehyde ( PFA ) in PBS , permeabilized with 0 . 1% TrixonX-100 for 10 min . Human sperm cells from swim-up purification were fixed 4% PFA in PBS for 10 min followed by 100% MeOH . Fixed human sperm cells were permeabilized in 0 . 1% saponin for 10 min . Permeabilized sperm cells were washed in PBS and blocked with 10% goat serum for 1 hr . Mouse samples were stained overnight with primary antibody against CatSper1 ( 10 µg ml−1 ) and CatSperζ ( mζ174 , 20 µg ml−1 ) as were human samples with primary antibodies against CatSperε ( hε31 , 20 µg ml−1 ) and CatSperζ ( hζ11 , 10 µg ml−1 ) , in 10% goat serum in PBS , 4°C . After PBS wash , goat-anti-rabbit Alexa488 conjugate ( Invitrogen ) served as the secondary antibody . Images were acquired on laser scanning confocal microscopes ( Olympus Fluoview 1000; Figure 1G and H , Figure 1—figure supplement 2G , and Figure 2—figure supplement 1E and Leica TCS SP8; deconvolved image in Figure 5A ) . For timed coitus , females were introduced to single-caged CatSperz-het or -null males for 1 hr and checked for the presence of a vaginal plug . To examine sperm migration to the fertilization site in vivo , ampullae were removed from the mated females at 8 hr after coitus and COCs were released . A series of z-stacked images ( 2 µm step size ) of the COCs was taken and number of sperm within each COC was recorded according to the presence of a sperm head . To calculate in vivo fertilization rate , eggs were gently flushed from oviducts and ampullae from the mated females at 20 hr and 27–30 hr after coitus . The total number of eggs and the number of 2 cell eggs were counted . Two females were caged with each male for three months to track pregnancy and litter production . For IVF assays , oocytes were recovered from superovulated 5–6-week-old B6D2F1 female mice 13 hr after injection of 5 U human chorionic gonadotropin . For standard IVF , sperm were collected from the cauda epididymis . For ejaculate IVF , sperm were retrieved from the uterus of a 1 hr window-timed coitus . Both epididymal and ejaculated sperm were capacitated in vitro at 37°C for 1 hr , and coincubated with eggs at ∼105 sperm/mL . After about 4 . 5 hr , unbound sperm were washed away . After 24 hr incubation the embryos were observed under light microscopy ( Olympus IX-70 ) to check for development of the two-cell stage . Spermatozoa from the dissected cauda epididymis ( swim up method ) were collected in HEPES buffered saline ( HS ) media . Spermatozoa were plated on 35 mm fibronectin-coated coverslips for 15 min ( 22°C ) ; unattached sperm were removed by the gentle pipette wash ( time 0 ) and basal motility recorded . Activated motility was recorded within the first 10 min after adding pre-warmed human tubal fluid ( HTF ) -capacitating medium ( Millipore ) . To induce hyperactivation , attached sperm cells were incubated in HTF media for 90 min at 37°C ( 5% CO2 ) . All subsequent images were recorded at 37°C . The flagellar waveform was analyzed by stop-motion digital imaging collected at 200 fps ( HC Image software , Hamamatsu Photonics or Zen Blue , Zeiss; 2 s movies ) . Overlay of flagellar traces from two complete flagellar beats were generated by hyperstacking binary images using open-source FIJI software ( Schindelin et al . , 2012 ) and time coded in color . Cauda epididymal spermatozoa were suspended and incubated in non-capacitating M2 medium ( Specialty Media , Millipore ) or in HTF medium for capacitation . Sperm motility was then measured using the IVOS sperm analysis system ( Hamilton Thorne Biosciences , Beverly , MA ) in an 80 µm ( depth ) chamber to obtain various parameters ( Figure 2—figure supplement 2D ) . Sperm motility was also analyzed with an Olympus IX-70 microscope equipped with a high-speed sCMOS camera ( Orca-Flash4 . 0 ) and a 10x objective . 1–2 × 105 mouse sperm before and after capacitation were added to the 37°C chamber ( Delta T culture dish controller; Bioptechs ) containing 1 ml HEPES-HTF medium ( H-HTF: 92 mM NaCl , 2 mM CaCl2 , 4 . 7 mM KCl , 0 . 2 mM MgCl2 , 0 . 37 mM KH2PO4 , 25 mM NaHCO3 , 18 . 3 mM Na lactate , 2 . 78 mM glucose , 0 . 33 mM Na pyruvate , 0 . 4% [w/v] bovine serum albumin [BSA] , and 10 mM HEPES [pH 7 . 4] ) . In some experiments , the medium was supplemented with methylcellulose ( MC ) ( M0512 , 4000 cP in 2% solution; Sigma ) at 0 . 3% , 0 . 4% , or 0 . 5% ( w/v ) . Sperm swimming 3–5 mm from the rim were recorded after a 10 min preincubation period that allowed spontaneous dissociation of sperm clumps . To inhibit convective flow , 1 ml of medium was overlaid by 1 ml of mineral oil and covered by a heated glass lid ( Bioptechs ) . Sperm motility at 37°C was videotaped at 100 fps . Images ( HC Image software , Hamamatsu Photonics ) were analyzed for swimming trajectory from a 1 s playback movie at 1/5 speed , by head tracing via Computer Assisted Sperm Analysis ( CASA; http://rsbweb . nih . gov/ij/plugins/casa . html ) . To track swimming trajectory in viscous medium , the sperm motility was videotaped at 50 fps . The images were analyzed using Fiji software ( Schindelin et al . , 2012 ) by assembling overlays of the flagellar traces generated by hyperstacking binary images of 20 frames of 2 s movies coded in a gray intensity scale . Mouse sperm incubated in HTF medium for 90 min at 2 × 106/ml yielded capacitated sperm . Capacitated sperm were transferred and concentrated for capillary loading by centrifugation at 900 g for 3 min . The loose sperm pellet at the bottom of the microcentrifuge tube was resuspended in HEPES-HTF at 4 × 106/ml , and loaded into the capillary by suction via an air-pressure microinjector ( IM-5B; Narishige; 22 C , ~200 µm/s ) . While applying gentle positive pressure , the sperm in the tip of the capillary were moved out of the sperm drop . The tip of the capillary is transferred to a 37°C chamber ( Delta T culture dish controller; Bioptechs ) and placed into a 50 µl drop of HEPES-HTF medium covered with mineral oil . Negative pressure was applied slowly and sperm cells swimming against the flow and down to the H-HTF drop was video-recorded at 33 fps . Whole-cell recording of corpus epididymal spermatozoa from 3–5 month-old CatSperz+/- or CatSperz-/- mice was performed blind as to genotype ( Kirichok et al . , 2006; Navarro et al . , 2011 ) . HS was the bath medium . The standard pipette solution was ( mM ) : 120 Cs- Methanesulfonate ( Cs-MeSO4 ) , 5 CsCl , 5 Cs-BAPTA , 10 HEPES and 10 MES , pH 7 . 2 with H-MeSO3 . To record IATP , we used a low Cl- bath solution ( to reduce background Cl- conductance ) in the following ( mM ) : 150 Na- methanesulfonate ( Na-MeSO3 ) , 2 CaCl2 , 10 Na-HEPES , and 10 MES ( pH 7 . 4 or 6 . 0 ) . To measure ICatSper , we used divalent-free ( DVF ) solution , in mM: 150 Na-MeSO3 , 2 Na3HEDTA [ ( hydroxyethyl ) ethylenediaminetriacetic acid] , 2 EGTA , and 20 HEPES ( pH 7 . 4 ) with NaOH . Solutions were applied to sperm cells ( lifted from the coverslips ) initially by bath perfusion . After break-in , the access resistance was 25–80 MΩ . All experiments were performed at 22–24°C . The whole-cell currents were recorded using an Axopatch 200B amplifier ( Molecular Devices , Sunnyvale , CA ) , acquired with Clampex 9 ( pClamp9 Software; Molecular Devices ) , and analyzed with Origin software ( OriginLab ) . Signals were low-pass filtered at 2 kHz and sampled at 10 kHz . Data are given as mean ± SD . All the experiments are repeated at least three times . Sample size and number of replicates are described in each figure and the figure legends . Statistical analyses were performed using Student’s t-test unless indicated; e . g . F-test in one-way ANOVA . Differences were considered significant at *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , and ****p<0 . 0001 . When ****p<0 . 0001 , actual P value is not indicated . | Male mammals ejaculate millions of sperm cells each time they mate with a female . Only a few of these cells manage to travel up the female’s reproductive tract to reach the egg , and usually only one sperm fertilizes it . Freshly ejaculated sperm are incapable of fertilizing eggs and have to undergo several changes within the female to become able to do so . One crucial change occurs in the sperm tail , which starts to beat vigorously in a whip-like motion . This type of movement – known as hyperactivated motility – enables the sperm to swim towards the egg , push through a sticky coating that surrounds it , and then burrow into it . Hyperactivated motility is triggered when calcium ions enter the sperm cell via a specific channel protein known as CatSper , which is found in the membrane that surrounds the cell . CatSper channels form groups ( known as complexes ) with several other proteins that are arranged in a unique pattern of four straight ‘stripes’ running down the tail of the sperm . This arrangement is necessary for hyperactivated motility and mutations in the genes that encode these proteins can lead to infertility in males . The CatSper channel complex is known to contain seven proteins: four that form a pore through which calcium ions can enter , and three accessory proteins whose roles in hyperactivated motility are less clear . Chung et al . identified two genes in mice that encode new accessory proteins in the CatSper channel complex named CatSper epsilon and CatSper zeta . Further experiments analyzed the role of CatSper zeta in more detail . Mutant males that lack CatSper zeta have fragmented patterns of CatSper stripes in the tails of their sperm . Moreover , fewer calcium ions were able to pass through the channels to enter the cell . Together , this made the sperm tail more rigid , which prevented it from moving efficiently within the female , resulting in reduced fertility . Chung et al . also found that the mutant sperm were less able to penetrate the egg than normal sperm . During evolution , the gene that encodes CatSper zeta appeared first in mammals and may represent an adaptation that improved the chances of a sperm fertilizing the egg inside the reproductive tract of female mammals . Future challenges will be to explore how the CatSper channel assembles on the membrane of sperm and find out exactly how calcium ions trigger hyperactivated motility . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2017 | CatSperζ regulates the structural continuity of sperm Ca2+ signaling domains and is required for normal fertility |
In an effort to identify human endothelial cell ( EC ) -enriched lncRNAs , ~500 lncRNAs were shown to be highly restricted in primary human ECs . Among them , lncEGFL7OS , located in the opposite strand of the EGFL7/miR-126 gene , is regulated by ETS factors through a bidirectional promoter in ECs . It is enriched in highly vascularized human tissues , and upregulated in the hearts of dilated cardiomyopathy patients . LncEGFL7OS silencing impairs angiogenesis as shown by EC/fibroblast co-culture , in vitro/in vivo and ex vivo human choroid sprouting angiogenesis assays , while lncEGFL7OS overexpression has the opposite function . Mechanistically , lncEGFL7OS is required for MAPK and AKT pathway activation by regulating EGFL7/miR-126 expression . MAX protein was identified as a lncEGFL7OS-interacting protein that functions to regulate histone acetylation in the EGFL7/miR-126 promoter/enhancer . CRISPR-mediated targeting of EGLF7/miR-126/lncEGFL7OS locus inhibits angiogenesis , inciting therapeutic potential of targeting this locus . Our study establishes lncEGFL7OS as a human/primate-specific EC-restricted lncRNA critical for human angiogenesis .
Angiogenesis plays a critical role in tissue development and homeostasis . Aberrant angiogenesis has been associated with numerous diseases , including heart disease , tumor growth , metastasis and age-related macular degeneration ( AMD ) ( Carmeliet , 2003 ) . Defective vascularization , usually associated with compensatory angiogenesis and vasculogenesis , has been observed in human dilated cardiomyopathy ( DCM ) patients ( Roura et al . , 2007; Gavin et al . , 1998; De Boer et al . , 2003 ) . Methods to augment angiogenesis have been tested clinically for DCM ( Ylä-Herttuala et al . , 2017 ) . Anti-angiogenic therapy , such as antibodies to vascular endothelial growth factors ( VEGF ) , has shown efficacy clinically in treating wet AMD , the leading blinding disease in the elderly ( Brown et al . , 2006; Rosenfeld et al . , 2006; Zampros et al . , 2012; Hurwitz et al . , 2004 ) . However , some patients failed to respond to anti-VEGF treatment . Similarly , anti-angiogenic therapies have shown efficacy in certain cancers when used alone or combined with chemotherapy ( Miller et al . , 2007; Sandler et al . , 2006 ) . However , anti-angiogenic therapy has met several hurdles on its way to be an main option for cancer therapy , mainly due to drug resistance ( Shojaei and Ferrara , 2007 ) . Identifying novel human angiogenesis mechanism would provide important insights and potential therapeutic options for angiogenesis-related diseases . It is now established that up to 90% of the human genome is transcribed , and the majority of these transcripts are non-coding RNAs ( ncRNAs ) that do not encode proteins ( Kapranov et al . , 2007; Gerstein , 2012; Ecker , 2012 ) . NcRNAs can be classified as short noncoding RNAs such as microRNAs ( miRNAs ) , long noncoding RNAs ( lncRNAs ) and other classic ncRNAs . miRNAs include a group of small noncoding RNAs sized ~22 nucleotides that play important regulatory functions in numerous physiological and pathological processes , including angiogenesis ( Wang and Olson , 2009 ) . LncRNAs represent a large group of long ( typically >200 nt ) noncoding RNAs , whose function is still largely enigmatic ( Ulitsky and Bartel , 2013 ) . The study of lncRNAs in vascular biology is still in its infancy ( Yu and Wang , 2018; MM and Goyal , 2016 ) . Several lncRNAs , including MALAT1 ( Liu et al . , 2014; Michalik et al . , 2014 ) , MANTIS ( Leisegang et al . , 2017 ) , PUNISHER ( Kurian et al . , 2015 ) , MEG3 ( He et al . , 2017; Qiu et al . , 2016 ) , MIAT ( Yan et al . , 2015 ) , SENCR ( Boulberdaa et al . , 2016 ) , GATA6-AS ( Neumann et al . , 2018 ) and STEEL ( Man et al . , 2018 ) , have been shown to regulate angiogenesis . Dependent on their subcellular localizations , these lncRNAs function by regulating promoter and enhancer activities of angiogenesis-related genes in cis , or modulating gene expression by in trans mechanism through interaction with DNA/RNA-binding proteins or chromatin modifying proteins , or functioning as antisense RNAs to mRNAs or sponge for miRNAs in the cytoplasm . By profiling more than 30 , 000 lncRNAs in several primary human EC lines , we have identified ~500 human EC-restricted lncRNAs . Among them , we focused on lncEGFL7OS , which is located in the opposite strand of the EGFL7/miR-126 gene . Through a series of in vitro and in vivo experiments , we established lncEGFL7OS as a disease-relevant , human/primate-specific , EC-enriched lncRNA that is critical for angiogenesis through regulating MAX transcription factor activity at the EGFL7/miR-126 locus .
To identify lncRNAs specific in ECs , a microarray was performed to profile ~30 , 000 lncRNAs and ~26 , 000 coding transcripts using an Arraystar human LncRNA microarray v3 . 0 system ( Arraystar , Rockville , MD ) . Three primary human EC lines and two non-EC lines at low passages , namely , human umbilical vein EC ( HUVEC ) , human retinal EC ( HREC ) , human choroidal EC ( HCEC ) , human dermal fibroblast cell ( HDF ) and human retinal pigment epithelial ( RPE ) cell lines , were used in the array . Purity of EC lines was confirmed by acetyl-LDL uptake and EC marker staining ( Figure 1—figure supplement 1 ) . Hierarchical cluster analysis of the array results validated the clustering of EC lines , which clearly separates from the HDF and RPE cell lines based on lncRNA and mRNA expression ( Figure 1A ) . Moreover , lncRNAs appeared to be a stronger classifier to distinguish between EC and non-ECs than mRNAs . 498 lncRNAs are enriched in all three EC lines for more than two folds compared to the non-ECs ( see Figure 1B for top 50 hits , Supplementary file 1 ) . Among them , 308 are intergenic lncRNAs , 62 are sense overlapping lncRNAs , 83 are antisense lncRNAs , 23 are bidirectional lncRNAs , and 22 lncRNAs were previously identified as pseudogenes ( Figure 1C ) . When these lncRNAs were cross-referenced with the enhancer-like lncRNAs , 19 of them are known enhancer-like lncRNAs with nearby coding genes within 300 kb ( Supplementary file 2 ) ( Ørom et al . , 2010 ) . We also took advantage of our microarray system in profiling both lncRNAs and mRNAs , and examined the lncRNA/mRNA regulation relationship for the EC-restricted lncRNAs . Since many lncRNAs have been shown to exert locus-specific effect on nearby genes , we first did a bioinformatics search for protein-coding genes that are within 10 kb of the 498 EC-restricted lncRNAs . 91 lncRNAs have protein-coding genes within 10 kb of the lncRNA gene ( Supplementary file 3 ) . Moreover , 27 of the 91 lncRNAs exhibited parallel expression pattern to the neighboring mRNAs in all 5 cell lines tested , while three of them showed inverse expression pattern relationship with the neighboring mRNAs . For some lncRNAs , including those near to SRGN , FOXC2 , STEAP1B , ECE1 , GOT2 , EGFL7 and PRKAR1B , the specificity for lncRNA in ECs is more robust than the neighboring mRNAs; for some other lncRNAs , including those near to HHIP , ESAM , and UBE2L3 , their EC-specificity is less robust than their neighboring mRNAs . These results suggest that some lncRNAs can serve as robust EC-restricted gene expression markers . We also carried out a functional enrichment analysis based on the EC-restricted lncRNAs and the associated genes . The following biological processes and genes are highly represented in the associated lncRNAs with a false discovery rate ( FDR ) of less than 10% ( Figure 1—figure supplement 2A ) : ( 1 ) heart development ( NRP1 , ECE1 , FOXC2 , PKD1 , ZFPM2 , FKBP1A , FOXP4 ) ; ( 2 ) chordate embryonic development ( GATA2 , SATB2 , ECE1 , LMX1B , FOXC2 , PKD1 , ZFPM2 ) ; ( 3 ) embryonic development ending in birth ( GATA2 , SATB2 , ECE1 , LMX1B , FOXC2 , PKD1 , ZFPM2 ) ; ( 4 ) blood vessel development ( NRP1 , EGFL7 , ROBO4 , FOXC2 , PKD1 , ZFPM2 ) ; ( 5 ) vasculature development ( NRP1 , EGFL7 , ROBO4 , FOXC2 , PKD1 , ZFPM2 ) ; and ( 6 ) metallopeptidase activity ( ECE1 , ADAMTS16 , LTA4H , MMP25 , ADAM15 ) . From above , genes involved in embryonic development , especially vascular development , are associated with the EC-restricted lncRNAs . Taken together , we have established the lncRNA expression profile in ECs by comparative lncRNA microarray , and identified hundreds of EC-restricted lncRNAs , with a list of them having associated genes involved in vascular development . Quantitative ( q ) RT-PCR was used to confirm a selected list of EC-enriched lncRNAs from the microarray . Friend leukemia integration 1 ( FLI1 ) antisense lncRNA ( FLI1AS , also named as SENCR ( Bell et al . , 2014 ) , ASHGA5P026051 ) , GATA binding protein 2 ( GATA2 ) antisense lncRNA ( lncGATA2 , ASHGA5P019223 , RP11-475N22 . 4 ) , endothelial converting enzyme 1 ( ECE1 ) intron sense-overlapping lncRNA ( lncECE1 , ASHGA5P032664 , AX747766 ) , endothelial cell-selective adhesion molecule ( ESAM ) bidirectional lncRNA ( lncESAM , ASHGA5P021448 , RP11-677M14 . 3 ) , roundabout homolog 4 ( ROBO4 ) nature antisense RNA ( lncROBO4 , ASHGA5P026882 , RP11-664I21 . 5 ) , and epidermal growth factor-like domain 7 ( EGFL7 ) opposite strand lncRNA ( lncEGFL7OS , ASHGA5P045551 , RP11-251M1 . 1 ) were chosen because of their EC restriction and potential relevance to EC function . As shown in Figure 1D , the expression of lncECE1 , lncGATA2 , lncESAM , lncROBO4 , lncFLI1 and lncEGFL7OS was found to be highly enriched in EC cell lines compared to the non-EC lines . Among different EC lines , lncECE1 and lncESAM were more enriched in HUVECs , while FLI1AS and lncEGFL7OS were more enriched in HCECs , supporting heterogeneity of ECs and suggesting differential expression of the lncRNAs in different ECs . We also used a bioinformatics approach to determine the tissue distribution of the EC-restricted lncRNAs . The tissue expression information of the top 50 EC-restricted lncRNAs was obtained from the Stanford Source database ( Diehn et al . , 2003 ) . Figure 1—figure supplement 2B shows the tissue distribution heatmap of the candidate lncRNAs with information available . The majority of the lncRNAs are enriched in the lung and placenta , which are highly vascularized tissues . Taken together , these data support the EC- and vasculature- restriction of the candidate lncRNAs from our microarray . Given the involvement of EGFL7/miR-126 locus in regulating angiogenesis , we focused on lncEGFL7OS , which partially overlaps with EGFL7/miR-126 gene but is transcribed in opposite direction ( Figure 2A ) ( Fish et al . , 2008; Wang et al . , 2008a; Kuhnert et al . , 2008; Durrans and Stuhlmann , 2010; Parker et al . , 2004; Schmidt et al . , 2007 ) . The existence of lncEGFL7OS was confirmed by RT-PCR cloning using human placental RACE-ready cDNAs and subsequent sequencing , and the size of lncEGFL7OS is consistent with deposited gene AF161442 ( Figure 2—figure supplement 1A ) . Interestingly , conserved homologous sequence of lncEGFL7OS only exists in humans and primates Rhesus monkey , but not in other lower vertebrate species including mice , suggesting lncEGFL7OS is an evolutionarily new gene in mammals . We performed qRT-PCR to examine the tissue expression pattern of lncEGFL7OS . LncEGFL7OS was found to be highly enriched in the human lung , placenta and heart , which are highly vascularized tissues ( Figure 2B ) . Since lncEGFL7OS overlaps with EGFL7/miR-126 , the expression of EGLF7 and miR-126 was also examined in parallel to lncEGFL7OS . Human EGFL7 has four isoforms , named as EGFL7A-D , but only EGFL7B and EGFL7C are detectable by RT-PCR in human tissues . By qRT-PCR , both EGFL7B and EGFL7C are enriched in heart , kidney , bone marrow , uterus , thymus , thyroid , small intestine and placenta . Besides that , EGFL7B is more enriched in prostate , while EGFL7C is more enriched in lung and brain , suggesting a differential expression pattern of EGFL7 isoforms in humans ( Figure 2—figure supplement 1B ) . miR-126 is highly enriched in the bone marrow , lung and heart ( Figure 2—figure supplement 1C ) . Taken together , these results suggest there are both common and differential expression pattern of lncEGFL7OS and EGFL7/miR-126 in different human tissues . We also examined the subcellular localization of lncEGFL7OS using both semi-quantitative RT-PCR and high-resolution RNA fluorescence in situ hybridization ( FISH ) . By RT-PCR , lncEGFL7OS was shown to be expressed in both the cytoplasm and nucleus , but more in the nucleus of HUVECs ( Figure 2C ) . SENCR was used a marker for cytoplasmic-enriched lncRNA , while NEAT-1 was used as a marker for nuclear enriched- lncRNA ( Bell et al . , 2014; Zhang et al . , 2013 ) . These results were confirmed by high-resolution RNA FISH experiment . RNA FISH with single-molecule sensitivity was performed using oligonucleotide ( oligo ) probe pools specific for lncEGFL7OS ( Cabili et al . , 2015 ) . We observed variable numbers of lncEGFL7OS molecules in both the nucleus and cytoplasm of HUVECs ( Figure 2D ) . RNaseA-treated samples were used as negative control and adeno-lncEGFL7OS-overexpressed HUVECs were used as positive control for specificity of the probe . By quantification , the average copy number of lncEGFL7OS RNA in HUVECs is ~19 , which is in agreement with the copy number ( 23–28 copies ) by qRT-PCR using in vitro transcribed lncEGFL7OS as control for copy number calculation ( Supplementary file 4 ) . Taken together , these data indicate that lncEGFL7OS is expressed at relatively low copy numbers in both the nucleus and cytoplasm of HUVEC cells . To study the involvement of lncEGFL7OS in cardiovascular disease , we asked whether lncEGFL7OS expression correlates with human dilated cardiomyopathy ( DCM ) , a disease with defective vascularization ( Roura et al . , 2007; Gavin et al . , 1998; De Boer et al . , 2003 ) . Increased expression of proangiogenic factors , including hypoxia-inducible factor 1α ( HIF-1α ) and VEGF-A , have been found in DCM , likely due to the compensatory angiogenesis and/or increased mobilization of endothelial progenitor cells ( EPCs ) to the diseased heart ( Roura et al . , 2007 ) . The expression of lncEGFL7OS was examined by qRT-PCR in the hearts of 7 DCM patients , with five healthy hearts used as controls . In the DCM hearts , the expression of atrial natriuretic peptide ( ANP ) , a prominent marker for heart failure , was drastically upregulated ( Figure 2—figure supplement 1D ) . In line with the increased angiogenic factors , the expression of EC/EPC marker PECAM-1 was also marginally increased . We found lncEGFL7OS expression was significantly elevated in the hearts of DCM patients ( Figure 2E ) . Interestingly , the expression of EGFL7B and EGFL7C , as well as pri-miR-126 , was also significantly upregulated in the hearts of DCM patients . To dissect the lncEGFL7OS regulation mechanism in relation to its host gene EGFL7/miR-126 , we aimed to identify the potential regulatory elements for lncEGFL7OS . We have analyzed the cell type-specific active element of the locus from online database UCSC genome browser ( Figure 2F ) . A critical regulatory element is located on EGFL7B promoter between lncEGFL7OS and EGFL7/miR-126 . Bioinformatics data from ENCODE indicate that LncEGFL7OS DNA contains a region positive for epigenetic marks including histone H3 trimethylated lysine four methylation ( H3K4Me1 ) and H3K27Ac ( mark poised and active enhancers ) , H3K4Me3 ( marks promoter of protein coding genes ) , and binding sites for transcription factors MAX , MYC and RNA Polymerase ( PolR ) II . Several binding sites for ETS transcription factors were found in region . We have shown that its homologous region drives the EC-enriched LacZ reporter gene expression in mice ( Wang et al . , 2008a ) . Consistently , chromatin immunoprecipitation ( ChIP ) PCR assay using antibodies against MAX/MYC , RNA Pol II and histone H3 trimethylated lysine 4 ( H3K4me3 ) demonstrated the binding of these factors specifically to the region but not a non-relevant nearby region , indicating that this region is transcriptionally active ( Figure 2—figure supplement 2A ) . Additional potential promoters were not found in the region between lncEGFL7OS and EGFL7 transcripts by bioinformatics approach . Instead , CpG islands were found in the region . CpG islands in mammalian promoter regions tend to show bidirectional promoter activity ( Antequera , 2003 ) . Bidirectional promoters have been proposed to drive head-to-head gene transcription involving ncRNAs ( Uesaka et al . , 2014 ) . Based on these , we tested a novel hypothesis that a bidirectional promoter ( lncEGFL7OS/EGFL7/miR-126 promoter ) regulated by ETS factors drives the expression of both lncEGFL7OS and EGFL7/miR-126 in human ECs . The putative lncEGFL7OS promoter was cloned into a promoter-less luciferase reporter construct in either sense or anti-sense direction . By luciferase assay , the promoter in either direction exhibited similar activity under baseline in 293 T cells ( Figure 2G ) . Moreover , ETS1 transcription factor significantly activated the promoter activity in either direction , while the ETS1mut that lacks the DNA-binding domain showed significantly reduced activation of the promoter ( Wang et al . , 2008a ) . ETS factors have been shown to regulate miR-126 expression in ECs ( Harris et al . , 2010 ) . To further test whether ETS factors are required to regulate lncEGFL7OS expression , ETS1 and ETS2 genes were silenced in HUVEC cells , and lncEGFL7OS and pri-miR-126 expression were examined by qRT-PCR . Both genes were significantly reduced by ETS1/2 silencing , suggesting ETS factors control the expression of both lncEGFL7OS and EGFL7/miR-126 ( Figure 2H ) . To define the potential role for lncEGFL7OS in angiogenesis , we performed EC-fibroblast co-culture assays after silencing lncEGFL7OS using two independent siRNAs in HUVEC cells ( Hetheridge et al . , 2011 ) ( Figure 3—figure supplement 1A–B ) . When ECs are cultured on the top of a confluent fibroblast cell layer , ECs will proliferate to form ‘islands’ of ECs , and then sprout to form three-dimensional vascular tubules resembling capillaries which can be visualized by immunostaining with an antibody to EC-enriched human PECAM-1 ( Figure 3A ) . Of note , control siRNA has a mild but not significant effect in angiogenesis in this model . Compared to the control siRNA , si-lincEGFL7#1 or si-lncEGFL7OS#2 significantly repressed the formation of vascular tubules at 7 days after co-culture as shown by PECAM-1 staining and the subsequent quantification of the vascular tube length ( Figure 3A–B ) . Taken together , we conclude that lncEGFL7OS is required for proper angiogenesis in vitro . To examine the requirement of lncEGFL7OS in vasculogenesis/angiogenesis in vivo , si-lncEGFL7OS or control transfected HUVEC cells were mixed with Matrigel and injected subcutaneously on the back midline of nude mice , and the primary vascular network was stained with antibody against human PECAM-1 at 14 days after Matrigel implantation . Compared to the well-connected vessel structure in the controls , fewer networking was observed in the lncEGFL7OS-silenced EC group ( Figure 3C–D ) . Red blood cells and smooth muscle cells recruiting was detected in the formed vessels as proved by co-staining of human PECAM-1 and mouse Ter-119 ( red blood cell marker ) or mouse α-SMA ( smooth muscle marker ) staining , which suggests functionality of the vessels ( Figure 3C and Figure 3—figure supplement 2A–B ) . These results indicate that lncEGFL7OS is required for proper angiogenesis in vivo . To directly test the function of lncEGFL7OS in angiogenesis in human tissues , we developed a unique human choroid sprouting assay based on a previous publication ( Shao et al . , 2013 ) . Briefly , human choroids were dissected from the donor eyes from the eye bank , and were cut into approximately 4 mm2 pieces and transfected with control or lncEGFL7OS siRNAs overnight . The choroids were then seeded in the Matrigel and cultured in EGM-2 medium for up to 10 days . Silencing of lncEGFL7OS by siRNAs ( a mix of siRNA #1 and 2 at half concentration used for other assays ) in the system was confirmed by qRT-PCR ( Figure 3E ) . In the control choroid , significant sprouting was observed at day 10 with an average distance of ~1200 µm ( Figure 3F ) . Compared to the control , lncEGFL7OS siRNAs drastically repressed human choroid sprouting , establishing a critical role for lncEGFL7OS in angiogenesis in human tissues ( Figure 3F–G ) . The EC identity of the sprouting cells was confirmed by ICAM-2 and isolectin B4 co-staining ( Figure 3H ) . To dissect the cellular mechanism whereby lncEGFL7OS regulates angiogenesis , a BrDU incorporation assay was carried out to analyze EC proliferation upon lncEGFL7OS silencing . Under starvation condition , si-lincEGFL7#2 significantly decreased EC proliferation as shown by BrdU incorporation compared to the random control , while the effect from si-lncEGFL7OS#1 was not statistically significant ( Figure 4A ) . However , the EC proliferation induced by VEGF treatment was significantly repressed by either si-lncEGFL7OS#1 or si-lncEGFL7OS#2 . To further characterize the reduced EC proliferation after lncEGFL7OS knockdown , the cell cycle profile was quantified after flow cytometry under normal culture conditions . A significant increase in the percentage of cells in the G0/G1 phase was observed upon lncEGFL7OS knockdown ( Figure 4B–C ) . Accordingly , cells in the S and G2/M phase are significantly decreased . This indicates a G1 arrest in the si-lncEGFL7OS treated cells . We also determined whether EC migration is affected by lncEGFL7OS knockdown . Using a scratch wound assay , we found that compared to the control , lncEGFL7OS silencing significantly repressed EC migration in response to VEGF treatment after wound scratch ( Figure 4D–E ) . To assess whether lncEGFL7OS silencing results in EC death , TUNEL assay was performed . In the control condition , ~0 . 4% of EC cells undergo cell death , silencing of lncEGFL7OS by siRNA#1 and #2 significantly increased EC death to ~0 . 55% and~0 . 64% , respectively ( Figure 4—figure supplement 1 ) . Therefore , the increase of EC death by si-lncEGFL7OS is statistically significant , but probably not biologically important with regard to the angiogenic phenotypes observed . These results indicate that lncEGFL7OS is required for proper EC proliferation and migration in vitro . We further examined whether overexpression of lncEGFL7OS in ECs enhances angiogenesis . To do so , lncEGFL7OS or control LacZ adenoviruses were generated , and used to infect HUVEC cells at multiplicity of infection at 50 . Infected ECs were cultured on a fibroblast mono layer , and their angiogenic response was examined by staining with an antibody to PECAM-1 at 7 days after co-culture . The efficiency of the virus was verified by qRT-PCR . Over 2000-fold lncEGFL7OS was achieved in ECs after virus infection ( Figure 4—figure supplement 2A ) . No significant differences were observed in Ad-lacZ infected samples compared to noninfection controls . LncEGFL7OS overexpression enhanced angiogenesis as shown by the significantly increased total tube length compared to the controls ( Figure 4—figure supplement 2B–C ) . These data indicate that overexpression of lncEGFL7OS is sufficient to enhance EC angiogenesis . lncRNAs could exert regulatory function in cis on the neighboring genes in the nucleus ( Ørom et al . , 2010 ) . Since lncEGFL7OS is located in the opposite strand neighboring EGFL7/miR-126 , we surmised that lncEGFL7OS regulates angiogenesis by controlling EGFL7/miR-126 expression . The expression of EGFL7B-C and miR-126 was examined by qRT-PCR upon lncEGFL7OS knockdown . As shown in Figure 5A , EGFL7B and C expression was dramatically decreased upon lncEGFL7OS knockdown . The downregulation of EGFL7 at protein level by lncEGFL7OS knockdown was confirmed by Western blot analysis ( Figure 5—figure supplement 1 ) . Similarly , the expression of both miR-126 and miR-126* , a microRNA located in the intron of EGFL7 gene , is also downregulated by lncEGFL7OS knockdown ( Figure 5B ) . miR-126 has been shown to modulate MAP kinase signaling and PI3K-AKT signaling by targeting Spred-1 and PI3KR2 , respectively ( Fish et al . , 2008; Wang et al . , 2008a; Kuhnert et al . , 2008 ) . Consistent with the downregulation of miR-126 , phosphorylation of ERK1/2 and AKT induced by VEGF was significantly reduced in ECs transfected with si-lncEGFL7OS#1 or si-lncEGFL7OS#2 compared to the controls ( Figure 5C ) . We also examined whether lncEGFL7OS overexpression increases the expression of EGFL7 and miR-126 . As expected , a ~ 2 fold upregulation of miR-126 and a ~ 3 fold increase of EGFL7B were observed when lncEGFL7OS is overexpressed in ECs ( Figure 5—figure supplement 2A–B ) . To determine whether EGFL7 and miR-126 can mediate the angiogenic response of lncEGFL7OS , we tested the capability of miR-126 expressing adenovirus and EGFL7 protein in rescuing the anti-angiogenic phenotype of si-lncEGFL7OS . The combination of miR-126 and EGFL7 enhanced angiogenesis in the wild-type HUVECs , and rescued the anti-angiogenic effect of lncEGFL7OS silencing to a great extent in an EC-Fibroblast cell co-culture model ( Figure 5D–E ) . These results indicate that lncEGFL7OS is critical for maintaining maximal expression of EGLF7/miR-126 , which is required for VEGF signaling and angiogenesis through MAPK and PI3K/AKT pathways . To study the mechanism whereby lncEGFL7OS regulates EGFL7/miR-126 expression , we hypothesized that lncEGFL7OS regulates EGFL7/miR-126 promoter/enhancer activity by interacting with MAX transcription factor . MAX was predicted as one of the top lncEGFL7OS-interacting proteins by lncRNA interaction prediction program catRAPID ( Bellucci et al . , 2011 ) . Online database UCSC genome browser predicts the existence of MAX binding sites between lncEGFL7OS and EGFL7/miR-126 genes ( Figure 6A ) . We first tested whether lncEGFL7OS interacts with MAX protein in ECs . RNA immunoprecipitation ( RIP ) assays showed that lncEGFL7OS RNA was pulled down in the nuclear lysate by a Chip-grade antibody to MAX , and this interaction was increased by lncEGFL7OS overexpression ( Figure 6B ) . To dissect the domains in lncEGFL7OS that interact with MAX , lncEGFL7OS was separated into three fragments according to the predicted secondary structure ( Figure 6C ) . Three different fragments ( F1 to F3 ) were cloned into expression vectors , and transfected into RPE cells that have undetectable endogenous lncEGFL7OS expression . Similar RIP RT-PCR assays demonstrated that F1 fragment in the 5’ end of lncEGFL7OS is the major domain that interacts with MAX protein ( Figure 6D ) . We further examined whether MAX protein binds to the bidirectional lncEGFL7OS/EGFL7/miR-126 promoter/enhancer . ChIP-PCR assays confirmed the specific binding of MAX to this region in ECs ( Figure 6E ) . Moreover , overexpression of lncEGFL7OS significantly increased MAX binding to this region . As control , MAX protein was not enriched in a non-relevant control DNA region ( Figure 6—figure supplement 1A ) . MAX has been shown to dimerize with MYC and stimulate histone acetylation and gene transcription ( Vervoorts et al . , 2003 ) . Our co-immunoprecipitation assay confirmed the interaction of MAX with p300 , a component in the p300/CBP co-activator complex that has intrinsic histone acetyltransferase activities , in ECs ( Figure 6—figure supplement 1B ) . We therefore determined whether acetylated H3K27 ( H3K27ac ) , a marker for active enhancer , is enriched in this region , and found H3K27ac was indeed enriched in the region , and this enrichment was further increased by lncEGFL7OS overexpression ( Figure 6F ) . To confirm whether the interaction of lncEGFL7OS with MAX is required for angiogenesis , lncEGFL7OS-F ( 2 + 3 ) that does not contain the F1 region was cloned and used to make adenovirus . Overexpression of lncEGFL7OS-F ( 2 + 3 ) by adenovirus neither affected EGFL7B and miR-126 expression , nor impacted angiogenesis in an EC-fibroblast co-culture assay ( Figure 6—figure supplement 1C–F ) , suggesting the requirement of lncEGFL7OS/MAX interaction in angiogenesis . Together , these results suggest that lncEGFL7OS promotes the binding of MAX protein to the bidirectional promoter/enhancer region of lncEGFL7OS/EGFL7/miR-126 , and enhances their transcription , and therefore angiogenesis . To examine whether MAX is required for regulating lncEGFL7OS/EGFL7/miR-126 expression , two specific siRNAs were used to silence MAX expression ( Figure 6G ) . MAX silencing resulted in significantly decreased expression of EGFL7 , lncEGFL7OS and miR-126 ( Figure 6H–J ) . Consistently , MAX silencing led to repressed angiogenesis as shown by EC-Fibroblast co-culture assays ( Figure 6K ) . We further determine whether MAX silencing overrides the increased expression of miR-126 induced by adenovirus expressing lncEGFL7OS . As shown in Figure 6L , the induction of miR-126 expression by lncEGFL7OS overexpression was blunted by MAX knockdown . To determine whether lncEGFL7OS is required for MAX recruiting to the EGFL7/miR-126 promoter/enhancer , similar ChIP-PCR was performed after lncEGFL7OS knockdown . As shown in Figure 7A–B , silencing of lncEGFL7OS significantly reduced MAX binding to the EGFL7/miR-126 promoter/enhancer as well as H3K27 acetylation at the locus . Together , our data indicate that lncEGFL7OS regulates EGFL7/miR-126 expression by interaction with MAX transcription factor , which enhances H3K27 acetylation in the lncEGFL7OS/EGFL7/miR-126 enhancer/promoter region . Since lncEGFL7OS interacts with MAX , we asked whether other known MAX target genes , including Cyclin D2 and DHFR , are regulated by lncEGFL7OS ( Mai and Jalava , 1994; Bouchard et al . , 2001 ) . These two genes were confirmed to be MAX targets in ECs by siRNA experiments and ChIP assays ( Figure 7C–D and G–H ) . Overexpression of lncEGFL7OS enhanced the expression of Cyclin D2 and DHFR ( Figure 7E–F ) , which could be explained by the increased binding of MAX and increased H3K27 acetylation at their respective promoters ( Figure 7G–J ) . However , neither Cyclin D2 nor DHFR expression was repressed by lncEGFL7OS knockdown ( Figure 7K–L ) . These data suggest that , although lncEGFL7OS is capable of regulating other MAX target genes when overexpressed , lncEGF7OS does not act in trans to regulate angiogenesis through MAX under normal conditions . To further study the regulatory mechanism and the therapeutic targeting potential of the EGFL7/miR-126/lncEGFL7OS locus , a dCas9-KRAB system , in which a catalytically inactive Cas9 is fused to KRAB transcriptional repressor , was utilized to test the effect of silencing this locus on angiogenesis ( Qi et al . , 2013 ) . Two guide RNAs ( sgRNAs ) , with one targeting the genomic region between the EGFL7B and lncEGFL7OS transcription start sites and the other targeting the lncEGFL7OS intron region , were designed to guide sequence-specific transcription repression mediated by dCas9-KRAB ( Figure 8A ) . By EC-fibroblast co-culture assay , lentivirus expressing sgRNA-1 or sgRNA-2 significantly repressed EC angiogenesis only when dCas9-KRAB was co-expressed ( Figure 8B–C ) . Of note , Lenti-dCas9-KRAB alone did not significantly impact angiogenesis , ruling out the potential side-effects of dCas9-KRAB overexpression . When gene expression near this locus was examined , the expression of EGFL7B , miR-126 and lncEGFL7OS was drastically repressed by sgRNA-1 , and to a less extent by sgRNA-2 ( Figure 8D ) . These data support the co-regulation of EGFL7/miR-126 and lncEGFL7OS in the locus , and suggest the potential of therapeutic targeting angiogenesis by simultaneously targeting these three genes using a CRISPR-mediated approach .
Our data areconsistent with a recent publication that identified EC-restricted lncRNAs ( Man et al . , 2018 ) . Several lncRNAs , including lncEGFL7OS , HHIP-AS1 and SENCR , were in the short list from both microarrays . The difference from our results may reflect the different cell types used in the microarrays . We found 498 lncRNAs are enriched in three different primary EC lines compared to non-EC lines using a cutoff of 2 . By hierarchical cluster analysis , lncRNA-based clustering appeared to be a stronger classifier for EC lines than mRNA clustering . This is consistent with the general perception that lncRNAs exhibit better tissue specificity than mRNAs ( Derrien et al . , 2012 ) . We also found significant variability in lncRNA expression among EC lines , consistent the observed heterogeneity among ECs . Given the central importance of ECs in vascular biology , this dataset may provide a foundation to study the regulation and function for lncRNAs in various vascular development and disease models . Of note , we also found many lncRNAs are highly expressed in ECs , but those lncRNAs are not necessarily EC-specific ( data not shown ) . Those lncRNAs may also important function in cell types including ECs . Looking deep into the gene list , 91 lncRNAs of the 498 EC-restricted genes have protein coding genes within 10 kb , and about a third of them showed parallel or inverse expression pattern to the associated genes . Functional enrichment analysis indicates that EC-restricted lncRNAs are associated with genes involved in vascular development . Those lncRNAs may be good candidates for further functional studies . The evolution of EGFL7/miR-126 locus exemplifies the evolution of the vascular system . EGFL7 encodes an EGF-like domain containing protein that is specifically secreted by vascular ECs ( Parker et al . , 2004 ) . It is conserved among vertebrates but an orthologue is also found in Drosophila melanogaster ( CG7447 ) ( Nikolic et al . , 2010 ) . miR-126 and miR-126* are encoded by the intron of EGFL7 , and are conserved from Fugu in vertebrates to homo sapiens ( Wang et al . , 2008a ) . They are the only miRNAs that are known to be specifically in EC lineage and hematopoietic stem cells . Loss-of-function studies in mice and zebrafish revealed an important function of miR-126 in governing vascular integrity and angiogenesis ( Fish et al . , 2008; Wang et al . , 2008a ) . Egfl7-/- mice display similar vascular abnormalities to MiR126-/- mice , including edema , defective cranial vessel and retinal vascularization ( Schmidt et al . , 2007 ) . However , an independent study suggests that the vascular phenotype of Egfl7-/- mice could be attributed to the MiR126 deletion ( or downregulation ) in the mice ( Kuhnert et al . , 2008 ) . The important regulatory function of miR-126 in vascular integrity and angiogenesis is correlated with its appearance during the evolution of vascular system in vertebrates . Besides , miR-126 also has documented functions in vascular inflammation , as well as innate and adaptive immunity ( Harris et al . , 2008; Mattes et al . , 2009; Agudo et al . , 2014 ) . That also correlates with the evolutionary innovation of adaptive immune system in vertebrates . These support an important function of EGFL7/miR-126 locus from the evolutionary point of view . To further dissect the function and regulation of the locus during evolution from vertebrates to humans , we identified lncEGFL7OS , which is located in the opposite strand neighboring the EGFL7/miR-126 gene . It only exists in humans and several other primates , including rhesus monkeys , but not in other lower vertebrate species including mice . Although we showed significant function of lncEGFL7OS in human angiogenesis , the full spectrum of lncEGFL7OS function remains to be established . The expression of lncEGFL7OS is restricted to ECs and highly vascularized tissues , which is consistent with the expression of its host genes EGFL7 and miR-126 . As to its regulatory mechanisms , we found that both lncEGFL7OS and miR-126 are regulated by ETS1/2 factors in ECs through a bidirectional promoter . We found that lncEGFL7OS is required for proper angiogenesis in vitro by using EC-fibroblast co-culture vasculogenesis/angiogenesis assays . Conversely , overexpression of lncEGFL7OS enhances angiogenesis . Using a human choroid sprouting angiogenesis model we developed , we further demonstrated that lncEGFL7OS is required for human sprouting angiogenesis . This study indicates that three different transcripts from the EGFL7/miR-126 locus , including lncEGFL7OS , EGFL7 and miR-126 , have important functions in angiogenesis . EGFL7 and miR-126 have been previously shown to regulate angiogenesis ( Nikolic et al . , 2010 ) . EGFL7 is essential for vascular tube formation during vasculogenesis in zebrafish ( Parker et al . , 2004 ) . The importance of miR-126 in angiogenesis was demonstrated by loss-of-function studies in both mouse and zebrafish . Targeted deletion of miR-126 in mice or miR-126 knockdown in zebrafish resulted in loss of vascular integrity and defective angiogenesis , while overexpression of miR-126 regulates angiogenesis in a cell-type and strand-specific manner ( Fish et al . , 2008; Wang et al . , 2008a; Kuhnert et al . , 2008; Zhou et al . , 2016 ) . It is intriguing that , in contrast to EGFL7 and miR-126 , lncEGFL7OS represents a human/primate-specific mechanism in regulating angiogenesis , since lncEGFL7OS only exists in human and several other primates . New angiogenesis mechanism through lncEGF7OS has evolved during evolution , which underscores the importance and delicacy of EFGL7/miR-126 locus in angiogenesis . This study also highlights the importance of using human ( and/or primate ) system to study the mechanism of angiogenesis . We showed that the action of lncEGFL7OS reflects at least partially the regulation of expression of EGFL7 and miR-126 . miR-126 has been shown to promote MAP kinase and PI3K signaling in response to VEGF and FGF by targeting negative regulators of these signaling pathways , including Spred-1 and PIK3R2 . Consistent with the downregulation of miR-126 by lncEGFL7OS silencing , we found that the phosphorylation of ERK1/2 and AKT in response to VEGF is repressed by lncEGFL7OS silencing . Mechanistically , MAX transcription factor was identified as a lncEGFL7OS interaction protein required for lncEGFL7OS-regulated gene expression and angiogenesis in ECs . Under normal conditions , the lncEGFL7OS/MAX interaction is likely locus dependent since several other MAX target genes were not affected by lncEGFL7OS silencing . This is possibly due to the low expression of lncEGFL7OS . LncEGFL7OS enhances the transcription of EGFL7/miR-126 by binding to MAX protein that is recruited to the bidirectional promoter/enhancer region in EGFL7/miR-126 . MAX knockdown blunts the induction of miR-126 by lncEGFL7OS in ECs . MAX transcription factor has been shown to interact with MYC to control cell proliferation and cell death ( Amati and Land , 1994 ) . MYC has been shown to stimulate histone acetylation and gene transcription by recruitment of cAMP-response-element-binding protein ( CBP ) and p300 ( Vervoorts et al . , 2003 ) . Based on our results showing interaction of MAX and p300 , the enrichment of H3K27 acetylation by lncEGFL7OS likely result from the recruitment of CBP and P300 by MAX/MYC . Taken together , lncEGFL7OS acts in cis by interacting with MAX transcription factor to enhance H3K7 acetylation and promote EGFL7/miR-126 expression . Identifying angiogenic mechanisms that are conserved to human is critical for developing therapeutics for human vascular disorders . Our studies have demonstrated that lncEGFL7OS is a human/primate-specific lncRNA critical for human angiogenesis . This may be directly translatable for human diseases involving abnormal angiogenesis . Our studies showed increased expression of both lncEGFL7OS and EGFL7/miR-126 in the heart of DCM patients . Although the causative role of lncEGFL7OS in DCM is still unclear , lncEGFL7OS upregulation may reflect the compensatory vascularization/angiogenesis in DCM . It would be intriguing to test whether manipulating the lncEGFL7OS/EGFL7/miR-126 axis has therapeutic benefits for DCM patients . AMD is the leading cause of blindness in the elderly , and choroidal neovascularization is a hallmark for wet AMD ( Jager et al . , 2008 ) . Although anti-VEGF agents can markedly improve the clinical outcome of wet AMD , they have been unable to induce complete angiogenesis regression , and only 30–40% of individuals experienced vision improvement after treatment ( Folk and Stone , 2010; Krüger Falk et al . , 2013 ) . We developed a human choroid sprouting angiogenesis model and showed that silencing of lncEGFL7OS represses human choroid sprouting angiogenesis . It would be appealing to develop and test lncEGFL7OS-based therapy to treat choroidal neovascularization in wet AMD and other vascular disorders in the future . In this regard , our data that CRISPR-mediated targeting of EGLF7/miR-126/lncEGFL7OS locus inhibits angiogenesis could have therapeutic implications in angiogenesis-related diseases . Targeting this locus could be a potent approach for inhibiting angiogenesis than targeting the three genes individually .
Animal studies were conducted in accordance with the ARVO statement for the Use of Animals in Ophthalmic and Vision Research and were approved by the Institutional Animal Care and Use Committees at the Tulane University . BALB/cAnN-nu ( Nude ) female mice ( 6 to 8 weeks of age ) from Jackson lab were used for in vivo angiogenesis assay . In vivo Matrigel analysis was performed as described ( Skovseth et al . , 2007 ) . HUVEC cells transfected with control si-RNA , or mix of si-LncEGFL7OS#1 and si-LncEGFL7OS#2 ( 50nM each ) for 2 days . Cells were then trypsinized and about 5 × 105 cells were mixed with 50 μl EBM-2 medium and 350 μl ice-cold Matrigel ( BD Biosciences ) . The mixture was then applied under the back skin of 8 week-old BALB/cAnN-nu ( Nude ) female mice ( Jackson lab ) . After 14 days , The Matrigel plugs were extracted and snap-frozen in OCT and processed for immunostaining with human EC marker PECAM-1 ( DAKO ) , mouse red blood cell marker Ter-119 ( Thermo Fisher ) , mouse smooth muscle marker αSMA ( Abcam ) , and tube length quantification using image J ( National Institute of Health ) . HUVEC ( ATCC ) cells were grown in EC growth medium EGM-2 ( Lonza ) . HCEC and HREC cells were kindly provided by Dr . Ashwath Jayagapol from Vanderbilt University and grown in EGM2 media ( Lonza ) . EC identity of cells has been confirmed by immunostaining and acetyl-LDL uptake assay ( Figure 1—figure supplement 1 ) . ARPE-19 ( ATCC ) cells were growth in DMEM/F12 ( HyClone ) media with 10% FBS . HDF ( ATCC ) cells were grown in DMEM ( HyClone ) with 10% FBS . All cells have been tested negative for mycoplasma contamination . For VEGF treatment , HUVECs were starved with EC basal medium-2 with 0 . 1% FBS for 24 hr and then treated with VEGF ( 20 ng/mL ) for the indicated periods of time . SiRNA transfection in cell culture was performed as described ( Zhou et al . , 2014 ) . SiRNAs for LncEGFL7OS were purchased from sigma . Sequences for siRNAs are as follows: si-lncEGFL7OS#1: 5’-GCGUUUCCCUAGCAAUGUUdTdT-3’ and 5’-AACAUUGCUAGGGAAACGCdTdT-3’; si-lncEGFL7OS#2: 5’-CAGCUUUGCCCUAUCCCAUdTdT-3’ and 5’-AUGGGAUAGGGCAAAGCUGdTdT-3’ . Two pair of siRNAs for MAX gene include: 5’-CCAGUAUAUGCGAAGGAAAdTdT-3’ and 5’-UUUCCUUCGCAUAUACUGGdTdT-3’ , 5’-CACACACCAGCAAGAUAUUdTdT-3’ and 5’-AAUAUCUUGCUGGUGUGUGdTdT-3’ . SiRNAs for ETS1 include: 5’-CCGACGAGUGAUGGCACUGAAdTdT-3’ and 5’-UUCAGUGCCAUCACUCGUCGG-3’ . SiRNAs for ETS2 include: 5’-CAGUCAUUCAUCAGCUGGA[dT][dT]−3’ and 5’-UCCAGCUGAUGAAUGACUG[dT][dT]−3’ . RNAs from five cell lines were purified by mirVanaTm total RNA Isolation Kit ( Ambion , Invitrogen ) . These RNAs were subjected to microarray-based global transcriptome analysis ( Arraystar Human LncRNA array ( version 2 . 0 ) , Arraystar Inc , Rockville , MD ) . The lncRNA microarray is designed to detect about 30 , 586 LncRNAs and 26 , 109 coding transcripts . The lncRNAs were constructed using the most highly-respected public transcriptome databases ( Refseq , UCSC known genes , Gencode , etc ) , as well as landmark publications . The lncRNA probes include 19590 intergenic lncRNAs ( lincRNAs ) , 4409 intronic lncRNAs , 1299 bidirectional lncRNAs , 1597 sense overlapping lncRNAs and 3691 antisense lncRNAs . Data analyses , including hierarchy clustering analysis and functional enrichment analysis , were performed using Genescript software . The data have been deposited into NCBI GEO database ( GSE105107 ) . Tissue distribution data of the top-50 candidates was downloaded from the Stanford Source database ( Diehn et al . , 2003 ) . LncEGFL7OS- , lncEGFL7OS-F ( 2 + 3 ) , miR-126- , GFP- , or LacZ-expressing adenoviruses were generated as described ( Wang et al . , 2008b ) . Briefly , lncEGFFL7 cDNA was amplified by PCR using Phusion High-Fidelity DNA Polymerase from HUVEC cDNAs ( ThermoFisher Scientific ) and cloned into TOPO vector using the following primers: lncEGFL7up: 5’-GCCCTTTGGGCTCAGGCCCAGA-3’ and lncEGFL7dn: 5’-GCCCTTTGGGTTTGAGTAATAATTAC-3’ . After confirmation by sequencing , the fragment was cloned into pshuttle-CMV vector after HindIII/XhoI digestion . For lncEGFL7OS-F ( 2 + 3 ) cloning into pshuttle-CMV vector , the following primers were used: 5’-aaaagatctATGGCGTGTGAGTGCATGGCGAGC-3’ and 5’-tataagcttTGGGTTTGAGTAATAATTACATCAT-3’ . For making miR-126 adenovirus , miR-126-containing genomic DNA was PCR amplified from mouse using the following primers: 5’-ATGCGAATTC GAGTGAAAGAGCCCCACACTG-3’ and 5’-ATGCAAGCTT AGTGCCAGCCGTGGTCCTTAC-3’ , and cloned into pshuttle-CMV vector after ECORI/HindIII digestion . The positive clones were cut with PmeI and transformed into E . coli with adenovirus vector for recombination . Positive clones were then cut with PacI and transfected into Ad-293 cells using ViralPack Transfection Kit from Stratagene . Viral titers were determined by End-Point Dilution Assay . For adenovirus infection , the cells were switched to serum free EBM-2 medium and adenovirus was added at an MOI of 10 . The infection medium was removed after 3 hr . Cells were washed with PBS and overlaid with fresh growth medium and cultured for 48 hr before further experiments . EC cell proliferation , TUNEL assay and scratch-wound assays were performed using HUVEC cells as described ( Zhou et al . , 2014 ) . For cell proliferation assay , about 2 × 103 transfected HUVECs were seeded in 96-well plates . After starvation with 0 . 1% serum for overnight , the cells were stimulated with 20 ng/mL VEGF-A for 20 hr and then subjected to BrDU labeling for 4 hr . DNA synthesis as determined by BrDU incorporation was quantified using a commercial ELISA kit from Roche according to the manufacturer’s instructions . Cell cycle analysis was performed using Guava Cell Cycle Reagents ( Guava Technologies ) on a Guava instrument and analyzed using Cytosoft software according to the manufacturer's manual . For scratch wound assay , scratch-wound was made using a 200 μL pipette tip in lncRNA or control siRNA–transfected HUVEC monolayer before VEGF ( 20 ng/mL ) stimulation . 1 μM of 5-fluouracil ( Sigma ) was then added to the cells right after scratch wound to block cell proliferation . Post-scratch EC migration was scored at 14 hr after wound scratch . For in vitro angiogenesis assay , at 3 days after lncRNA or control siRNA transfection with Liptofectamine RNAiMAX reagent ( Invitrogen ) , cells were harvested for RNA or in vitro Matrigel assay and branch point analysis as described before . In vitro EC-Fibroblast co-culture was performed as described ( Hetheridge et al . , 2011 ) . Briefly , human dermal fibroblast cells ( HDF ) were seeded into each well of a 24 well plate and maintained in DMEM at 6 × 103 cells/well until they developed confluent monolayers . HUVECs were maintained as described above and transfected with siRNA one day prior to seeding on HDF monolayers . Approximately 6 × 103 HUVECs were seeded onto each monolayer and the HDF/HUVEC co-culture was maintained for 7 days in EGM-2 medium with medium changes every 2–3 days to allow endothelial cell polarization , migration , networking , and the formation of an in vitro primitive vascular plexus . For rescue experiments , some wells were transfected with Ad-miR-126 ( MOI of 10 ) and EGFL7 ( Abcam ) protein was added to the medium at 10 nM every other day . After 7 days the wells were fixed with 100% Methanol at −20° C for 20 min and then stained with anti-PECAM-1 ( DAKO ) . After hybridizing a secondary antibody , the endothelial tissue was visualized and imaged under a Nikon microscope . Multiple images were automatically stitched with Nikon software to provide a large image ( several mm [Roura et al . , 2007] ) and the resulting image was analyzed on ImageJ software to determine the degree of vascularization . Three wells were used for each condition and results are representative of the mean of each three well group . The experiments were repeated for at least three repeats with similar results . Ex vivo human choroid sprouting assay was adapted from a mouse protocol ( Shao et al . , 2013 . Donated human eye balls were obtained from Southern eye bank ( New Orleans , LA ) . The use of deceased human eye balls for the study was EXEMPT under DHHS regulations ( 46 . 101 ( b ) ) after consultation with the Tulane IRB committee . Informed consent has been obtained from all subjects by Southern eye bank . Eyes were collected within 24 hr of decease of the donors , and cleaned and kept in sterile ice-cold PBS with Penicillin/Streptomycin before dissection . Using fine forceps , the cornea and the lens from the anterior of the eye were removed . The peripheral choroid-scleral complex was separated from the retina and the RPE layer was peeled away using fine forceps . The choroid-scleral complex was then cut into approximately 4 mm2 pieces using sterile scalpel blade under laminar airflow . The choroid was then washed with sterile ice-cold PBS and transferred into endothelial base medium ( EBM2 ) with 0 . 1% FBS ( 300 µl/well in 24-well plates ) . The choroid was transfected with control si-RNA , or mix of si-LncEGFL7OS#1 and si-LncEGFL7OS#2 ( 50nM each ) for overnight . Choroid fragments were then washed by EGM2 media then placed in growth factor-reduced MatrigelTM ( BD Biosciences ) in 24-well plate . Briefly , 30 µl of matrigel was used to coat the bottom of 24 well plates without touching the edge of the well . After seeding the choroid , the plate was incubated in a 37 °C cell culture incubator to make the Matrigel solidify . 500 µl EC growth medium ( EGM-2 ) were added slowly to the plate without disturbing the Matrigel , and the plate was incubated at 37 °C cell culture incubator with 5% CO2 . Cell culture medium was changed every 48 hr . The EC sprouts normally start to appear on the day five and grow rapidly between day 7 and 10 . Phase contrast photos of individual explants were taken using a Nikon microscope . The sprouting distance was quantified with computer software ImageJ ( National Institute of Health ) . Sprouting ECs were stained with ICAM-2 ( BD Pharmingen ) or isolectin B4 ( Vector Lab ) . Human total RNA master panel II was purchased from clontech ( Takara ) . Total RNA was isolated from human choroid tissues or cell lines using TRIzol reagent ( Invitrogen ) . Cytoplasmic and nuclear RNA was purified using a Cytoplasmic and Nuclear RNA Purification Kit ( Norgen Biotek Corp . , Thorold , ON , Canada ) according to manufacturer’s supplied protocol . In brief , cells growing in monolayer were rinsed with 1xPBS and lysed directly on the plate with ice-cold Lysis Buffer . Next cell lysate was transferred to the RNase-free microcentrifuge tube and spun for 3 min at 14 , 000 x g . Supernatant containing cytoplasmic RNA was mixed with manufacturer’s supplied buffer ( Buffer SK ) and 100% ethanol , and applied onto a spin column . The pellet containing the nuclear RNA was mixed with Buffer SK and 100% ethanol , and applied onto a spin column . Both columns were washed with supplied Wash Solution , and RNA was eluted with supplied elution buffer ( Elution Buffer E ) . For maximum recovery two rounds of elution were performed . Quantitative ( q ) RT-PCR or regular RT-PCR was performed using iScript cDNA Synthesis system ( BioRad ) , miRNA qRT-PCR was performed using qScript cDNA Synthesis and microRNA Quantification System ( Quanta Biosciences ) . lncEGFL7OS RACE PCR was performed using Marathon –ready cDNA from human placenta ( Clontech , Mountain View , CA ) . 5’RACE and 3’RACE PCR was carried out using lncEGFL7OS primers and primers from the kit . Then a second round of PCR was performed using the combination of the RACE products and the RACE primers from the kit . The derived PCR product was then cloned using TOPO vector and sequenced . Primers for real-time PCRs include human β-actin , 5’-GAGCAAGAGATGGCCACGG-3’ and 5’-ACTCCATGCCCAGGAAGGAA-3’; lnc-FLI1-AS1 ( also named SENCR ) , up: 5’- CCTGAGGCCATCTTACCACC-3’ , down: 5’- AATCCGCTTCGATGAGTGGG-3’; SENCR ( for regular PCR ) , up: 5’-GCGCATTGTTAGGAGAAGGG-3’ , down: 5’- CCTGCTGACTGTCCTAGAGG-3’; lnc-GATA2-AS , up: 5’-CGGGCAGCTTACGATTCTTC-3’ , down: 5’- CGGTGTCTTTCAGAGGGTCT-3’; lnc-ECE1 , up: 5’- CCATGTCGCCTCAGCCTAAA−3’ , down: 5’- GGGCAGTCTCAGGGTAACAC-3’; lnc-ESAM , up: 5’-CTCGGAAAACGGAGGGTTGA-3’ , down: 5’- CGCTGCCCTTAATTCCTTGC-3’; lnc-ROBO4-AS , up: 5’- ACCAGCAGACCCTGAAACTC-3’ , down: 5’-GGCAGGGATCAGGCATTCAT-3’; lnc-EGFL7OS , up: 5’- AGTGCCAGCTTTGCCCTATC-3’ , down: 5’- GAGAACACAGGACGTCCACA-3’; EGFL7-A , up: 5- CTTCAGAGGCCAAAAGCACC-3’ , down: 5’- GAATCAGTCATCCCCCGGAC-3’; EGFL7-B , up: 5’- AAGGGAGGCTCCTGTGGA-3’ , down: 5’- CCTGGGGGCTGCTGATG-3’; EGFL7-C , up: 5’- CGGATCCGGCGGCCA-3’ , down: 5’- CGAACGACTCGGAGACAGG-3’; Neat1 , up: 5’-AGATACAGTGTGGGTGGTGG-3’ , down: 5’-AGTCTTCCCCACCTTGTAGC-3’ . Human primiR-126 , up: 5’-TGGCGTCTTCCAGAATGC-3’ , down: 5’-TCAGCCAAGGCAGAAGT-3’ . Human Cyclin D2 , up: 5’-GCTGTGCATTTACACCGACA-3’; down: 5’-TGCGCAAGATGTGCTCAATG-3’ . Human DFHR , up: 5’-ATTTCGCGCCAAACTTGACC-3’; down: 5’-TCTGAATTCATTCCTGAGCGGT-3’ . For western blot analysis , protein lysates were resolved by SDS-PAGE and blotted using standard procedures . Antibodies used were as follows: ERK1/2 ( Cell signaling ) , Phospho-ERK1/2 ( Cell signaling ) , AKT ( Cell signaling ) , Phospho-AKT ( Cell signaling ) , EGFL-7 ( Abcam ) and β-Tublin ( Abcam ) . For immunofluorescence experiments , samples were fixed with 4% paraformaldehyde or methanol for 30 min . After treatment with 1% Triton X-100 in PBS , samples were incubated in PBS containing 4% goat serum for 30 min . The samples were then incubated with primary antibodies overnight at 4°C , followed by incubation with appropriate secondary antibodies . Antibody used for immunofluorescence include: ICAM-2 ( BD Pharmingen ) , PECAM-1 ( DAKO ) . Single-cell lncEGFL7OS RNA copy number was determined as modified from a previous publication ( Wagatsuma et al . , 2005 ) . Briefly , 106 HUVEC cells were harvested for total RNA isolation using Trizol . 16% ( 8µl out of 50 µl ) of the total RNA was used for reverse transcription reaction as described above , and 1/100 of the cDNA was used as template in each well for the subsequent qRT-PCR . Therefore , for each well , the total lncEGFL7OS came from about ~1600 cells . For establishing the standard curve , pCRII-TOPO-lncEGFL7OS plasmid was linearized for generating lncEGFL7OS RNA by in vitro transcription . After concentration determination and copy number calculation , a given amount of RNA was employed to carry out the reverse transcription under the same conditions for HUVEC total RNA . The derived cDNA was diluted for PCR to generate a standard curve for lncEGFL7OS PCR . The copy number of RNA per cell was calculated based on the CT number ( Supplementary file 4 ) . 25 Stellaris RNA Fluorescence In Situ Hybridization ( FISH ) probes for lncEGFL7OS were designed according to Stellaris FISH probe designer ( https://www . biosearchtech . com/ Account/Login ? return=/stellaris-designer ) ( BiosearchTech , Supplementary file 5 ) . RNA-FISH was performed following the manufacturer’s protocol . Briefly , HUVECs cultured on 18 mm coverglasses were fixed and permeabilized by methanol-acetic acid solution for 10 min . After removing the fixation solution , cells were washed by Wash Buffer A ( Biosearch Tech ) at room temperature for three minutes , and then transferred to a humidified chamber to incubate with Hybridization Buffer ( Biosearch Tech ) containing the probes . The coverglasses were put upside-down on Parafilm for overnight . After washing with Wash Buffer A ( Biosearch Tech ) at 37°C for 30 min , the cells were incubated with Wash Buffer A containing 5 ng/ml DAPI in the dark at 37°C for 30 min . Finally , Wash Buffer B was added and the cells were incubated at room temperature for 5 min before mounting coverglass onto the slides with mounting medium . Pictures were taken under a Nikon A1 confocal microscope . For RNA copy quantification , hybridization signals and DAPI positive nucleus were counted manually . Co-immunoprecipitation assay was carried out following the Abcam protocol . Briefly , 107 HUVEC cells were scraped and resuspended in ice-cold lysis buffer ( 20 mM Tris . Hcl pH8 , 137 mM NaCl , 1% NP-40 , 2 mM EDTA , 10mM beta-mercaptoethanol , 15 U/ml DNAse I , protease Inhibitors ) . After 30 min on ice , cell lysate was centrifuged at 12000 g for 15 min at 4°C . The supernatant was transferred to another pre-chilled tubes and pre-cleared by 2 µg off-target rabbit antibody ( Santa Cruz ) followed by 40 µl of protein G magnetic bead slurry ( Bio-rad ) at 4°C . 25 µl pre-cleared cell lysate was reserved as input control . The rest was divided into two parts and added 2 µg of off-target rabbit IgG ( Santa Cruz ) and anti-P300 antibody ( Abcam ) respectively . The samples were incubated with antibodies at 4°C for overnight under gentle rotation . Then , 60 µl of protein G magnetic bead slurry ( Bio-rad ) was added into each sample . Incubate the lysate beads mixture at 4°C under rotation for 4 hr , then centrifuge the tubes and discard supernatant . The beads were washed with lysis buffer gently for three times . The proteins were eluted by SDS loading buffer ( supplemented with 10 mM beta-mercaptoethanol and protease Inhibitors ) . Western blot was used to analyze the content of samples . ChIP experiments were performed as described with some modifications ( Nelson et al . , 2006 ) . Briefly , HUVEC cells were cultured in the 10 cm dishes to 80–90% of confluence . After adding 400 µl of 37% formaldehyde to 10 ml medium and incubation for 15 min to fix the cells , cells were rinsed by pre-chilled PBS buffer and collected in 1 ml IP buffer ( 150 mM NaCl , 50 mM Tris-HCl ( pH 7 . 5 ) , 5 mM EDTA , NP-40 ( 0 . 5% vol/vol ) , Triton X-100 ( 1 . 0% vol/vol ) , 1% proteinase inhibitor cocktails ) . After half an hour of sonication , 2 µg of antibodies were added into cell lysate and incubated in ultrasonic bath for 30 min . Protein G Magnetic Beads were used to pull down antibodies in 4°C rotating platform for 2 hr . Once beads were washed for 5 times by cold IP buffer , 100 µl 10% ( wt/vol ) Chelex-100 was mixed with washed beads , and the mixture was boiled for 10 min . Each sample was added 1 µl of 20 µg/µl proteinase K and incubated at 55°C for 30 min . Samples were boiled for 10 min again and centrifuged . Supernatant were collected for real-time PCR . ChIP grade antibodys used in ChIP assay: Max ( Santa Cruz , sc-197 ) , Myc ( Sigma-Aldrih , c3956 ) , Anti-RNA Polymerase II ( Abcam , ab5408 ) , Tri-Methel-Histon H3 ( Lys4 ) ( Cell Signaling , #9751 ) , ETS1 ( Santa Cruz , sc-111 ) , H3K27Ac antibody ( Abcam , Ab4729 ) , Normal Rabbit IgG ( Cell Signaling , #2729 ) . ChIP samples were analyzed by using normal PCR with following parameters: ( 1 ) initial denaturation at 94°C for 10 min , ( 2 ) denaturation at 94°C for 20 s , ( 3 ) anneal at 58°C for 30 s , ( 3 ) extension at 72°C for 1 min . Steps from 2 to 4 were repeated 35 times . Primers to amplify conserved transcription factors binding region in the lncEGFL7OS enhancer/promoter region were as follows: Primers 1: 5′- CTGGCTGTTTTGGGGCTAGA-3′ and 5′- CCTGTGTGTGTTCTCCGCT-3′ . Primers 2 ( control region ) : 5′- AGATCCCAGGGCTGTTTAGC-3′ and 5′- AACACTCCTCCCAGCGAATC-3 . Primers for Cyclin D2 and DFHR promoter regions are as follows: Cyclin D2 promoter-F: 5’-GCAGGGAACCTAGTGTACGG-3’; Cyclin D2 promoter-R: 5’-CGCGCCCTTTGGTGTATTTC-3’; DHFR promoter-F: 5’-CGGGGCTACAAATTGGGTGA-3’; DHFR promoter-R: 5’-TAAAAGACGCACCCCTTGCC-3’ . RNA immunoprecipitation ( RIP ) was performed following a protocol from Abcam . Briefly , 107 Ad-GFP or Ad-lncEGFL7OS-infected HUVEC cells were harvested by trypsinization , and resuspended in PBS buffer respectively when the confluence was about 90% . Freshly prepared nuclear isolation buffer ( 1 . 28 M sucrose , 40 mM Tris-HCL pH7 . 5 , 20 mM MgCl2 , 4% Triton X-100 ) was diluted by 3× ddH2O and used to resuspend the above cell pellets . After incubation on ice for 20 min with frequent mixing , cell nuclei were collected by centrifugation at 2500 g for 15 min at 4°C , and resuspended in 1 ml freshly prepared RIP buffer ( 150 mM KCl , 25 mM Tris pH7 . 4 , 5 mM EDTA , 0 . 5 mM DTT , 0 . 5% NP40 , 100 U/ml RNAase inhibitor , protease inhibitors ) . After chromatin shearing , RNA supernatants were collected by centrifugation at 13000 rpm for 10 min to remove nuclear membrane and debris . 2 µg mock and anti-Max IgG were added into 500 µl supernatant respectively and incubated overnight at 4°C . 40 µl protein G magnetic beads ( Bio-rad ) was added and incubated for 2 hr at 4°C with gentle rotation . Coprecipitated RNAs were resuspended in 1 ml TRIzol reagent ( Invitrogen ) according to manual . Extracted RNAs were employed for subsequent reverse transcription and cDNA analysis . Some RNA samples were used as controls . LncEGFL7OS was separated into three domains according to its predicted secondary structure . Briefly , F1 domain contains 1-239nt of lncEGFL7OS , F2 domain contains 208-393nt and F3 domain contains 377-557nt . The separated domains were PCR amplified and sub-cloned into pShuttle-CMV vectors ( Agilent Technologies ) respectively , and transfected into APRE-19 cells together at 3 µg per vector per dish . After 48 hr , cells were harvested , the expression of the lncRNA fragments was confirmed by RT-PCR , and RNA immunoprecipitation ( RIP ) was performed by using MAX antibody as described above . Wild type ARPE 19 cells were harvested as background control since its lncEGFL7OS level is under the detection threshold . Dnase I was used to remove potential DNA contamination from the RNA samples before first-strand cDNAs were synthesized . Primers for construction and detection as below: F1-5’: 5’-AATAGATCT TGGGCTCAGGCCCAGAGTGCCA-3’; F1-3’:5’-AAAAAGCTT CT GGAGGCGCTCGCCATGCAC-3’; F2-5’: 5’ AATAGATCT ATGGCGTGTGAGTG CATGGC-3’; F2-3’: 5’-AAAAAGCTT TCAGGTAGCTGCGAGTTCAAG-3’; F3-5’: 5’-AATAGATCTACTCGCAGCTACCTGAGTCAGA-3’; F3-3’: 5’-AAAAAGCTT TG GGTTTGAGTAATAATTACATC-3’ . CRISPRi ( dCas9-KRAB ) assay was perform as described ( Larson et al . , 2013 ) . pHR-SFFV-dCas9-BFP-KRAB ( Addgene:46911 ) and control ( pLJM1-EGFP ) vectors were packaged into lentivirus vectors respectively . sgRNA-1 ( TGCTTACAGGCAAGGGGCGA ) and sgRNA-2 ( AAGAATTGCTTCAGCTCGGA ) , which target lncEGFL7OS promoter and intron respectively , were subcloned into lentiGuide-Puro vector ( Addgene: 52963 ) , which could express sgRNAs to assemble with dCas9-Krab . Empty lentiGuide-Puro vector serves as control . For the assay , HUVEC cells were transduced by control or dCas9-Krab vector , combing with lentiGuide-gRNA1 , lentiGuide-gRNA2 and empty lentiGuide-Puro , respectively . All lentivirus vectors were employed at 10 MOI . EC-fibroblast co-culture was performed as described above . Luciferase assays were performed as described ( Wang et al . , 2008a ) . The putative bidirectional promoter for lncEGFL7OS/EGFL7 was PCR amplified from human DNA and cloned into promoterless PGL3 Basic luciferase vector ( Promega ) . Primers include: plncEGFL7OSup ( XhoI ) : 5’-atcgCTCAGATAGACTCTGATGGCCCAGG-3’ and plncEGFL7OSdn ( XhoI ) : 5’ –atcgCTCAGACCAGCTTGGTGCAGGGAG-3’ . 293 T cells in 24-well plates were transfected with 50 ng of reporter plasmids in the presence or absence of increasing amount of Ets1 or Ets1 DNA-binding mutant expression plasmid . The human study was performed according to the principles of the Declaration of Helsinki . Patient information was described previously ( Huang et al . , 2015 ) . The procedure was approved by the Institutional Ethics Committee of the National Institute of Cardiovascular Diseases , Bratislava , Slovakia . Briefly , left ventricular tissues from seven patients with terminal-stage heart failure and five control healthy donors were dissected and snap frozen , and used for RNA isolation and gene expression study . In the bar graphs without P-value analysis , the central values are the means , and the error bars are standard deviation . In the bar graphs with P-value analysis , the central values are the means , and the error bars are standard error of means . Significant differences between groups were analyzed via Student’s unpaired t-test ( default ) . For multiple group analysis , significances between multiple groups were analyzed by ordinary ANOVA followed by Tukey honest significant difference testing . P-values of less than 0 . 05 were considered to be statistically significant . | A well-networked supply of blood vessels is essential for delivering nutrients and oxygen to the body . To do so , new blood vessels need to form throughout life , from embryonic development to adult life . This process , known as angiogenesis , also plays a critical role in exercise , menstruation , injury and disease . If it becomes faulty , it can lead to conditions such as the ‘wet’ version of age-related macular degeneration , where leaky blood vessels grow under the retina . This can lead to rapid and severe loss of vision . One way to treat this condition is to stop the growth of new blood vessels into this area using anti-angiogenic therapy , but not all patients respond to it . Identifying new mechanisms at play in human angiogenesis could provide insight into potential new therapies for this disease and other angiogenesis-related conditions . A large amount of our genetic material is made up of a group of molecules called long non-coding RNAs or lncRNAs for short , which normally do not code for proteins . However , they are thought to play a role in many processes and diseases , but it has been unclear if they also influence angiogenesis . Now , Zhou , Yu et al . set out to study these RNA molecules in different types of human vessel-lining cells and to identify their role in angiogenesis . Out of the 30 , 000 lncRNAs measured , about 500 of them were more abundant in these cells than other types of cells . One of the lncRNAs , called lncEGFL7OS , can be found on two human genes known to be relevant in angiogenesis ( EGFL7 and miR-126 ) . The results showed that patients with a condition called dilated cardiomyopathy , in which the heart muscle overstretches and becomes weak , had elevated levels of lncEGFL7OS . Other experiments analyzing human eye tissue revealed that lncEGFL7OS is required for angiogenesis by increasing the concentration of the EGFL7 and miR-126 gene products in cells . To achieve this , it binds together with a specific protein to the regulatory regions of the two genes to control their activity . The discovery of a new control mechanism for angiogenesis in humans could lead to new therapies for conditions such as macular degeneration and other diseases in which angiogenesis is affected . A next step will be to see if the same RNA molecules and genes are also elevated in other diseases . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2019 | LncEGFL7OS regulates human angiogenesis by interacting with MAX at the EGFL7/miR-126 locus |
The small molecule EMD 57033 has been shown to stimulate the actomyosin ATPase activity and contractility of myofilaments . Here , we show that EMD 57033 binds to an allosteric pocket in the myosin motor domain . EMD 57033-binding protects myosin against heat stress and thermal denaturation . In the presence of EMD 57033 , ATP hydrolysis , coupling between actin and nucleotide binding sites , and actin affinity in the presence of ATP are increased more than 10-fold . Addition of EMD 57033 to heat-inactivated β-cardiac myosin is followed by refolding and reactivation of ATPase and motile activities . In heat-stressed cardiomyocytes expression of the stress-marker atrial natriuretic peptide is suppressed by EMD 57033 . Thus , EMD 57033 displays a much wider spectrum of activities than those previously associated with small , drug-like compounds . Allosteric effectors that mediate refolding and enhance enzymatic function have the potential to improve the treatment of heart failure , myopathies , and protein misfolding diseases .
We used microscale thermophoresis to show that EMD 57033 binds purified β-cardiac myosin S1 with a stoichiometry of one molecule of EMD 57033 per myosin head with an affinity of 7 . 3 ± 1 . 9 µM ( Figure 1A ) . Direct binding of EMD 57033 to the motor domain was observed for Dictyostelium discoideum ( Dd ) myosin-2 and myosin-5 motor domain constructs but not for Dd myosin-1B , Dd myosin-1C , Dd myosin-1D , and Dd myosin-1E motor domain constructs . Myosin constructs that bind EMD 57033 display increased basal and actin-activated ATPase activity in the presence of the compound . The most potent interaction was observed with β-cardiac myosin with half-maximal activation ( AC50 ) observed at 7 . 0 ± 1 . 5 µM EMD 57033 ( Table 1 ) . The activation of skeletal muscle myosin-2 and Dd myosin-5 is twofold and fivefold weaker . Neither binding nor activation was observed for the ( − ) -enantiomer EMD 57439 ( Figure 1B ) . The consequences of EMD 57033 binding on the turnover of ATP by actin-activated β-cardiac myosin were followed in single-turnover experiments . Using the extrinsic fluorescence probe 2’-/3’-O- ( N’-methylanthraniloyl ) -ATP ( mantATP ) instead of ATP as substrate , we observed signal changes that can be attributed to three phases . The initial fast rise of fluorescence reflects the binding of the ATP analogue to the myosin active site , the following plateau phase monitors the duration of the hydrolysis reaction , and the third phase , corresponding to a decrease in fluorescence signal intensity , monitors the rate of product release . In the presence of EMD 57033 , we observed an approximately twofold net increase in the rate of ATP turnover ( Figure 1C ) . 10 . 7554/eLife . 01603 . 003Figure 1 . EMD 57033 binds to the myosin motor domain and activates ATPase and motor activities . ( A ) Direct interaction study between fluorescently labeled β-cardiac myosin S1 , Dd myosin-2 motor domain , and Dd myosin-1C motor domain and EMD 57033 . The myosin concentration was kept constant at 100 nM and EMD 57033 was titrated from 10 nM to 200 µM . The normalized thermophoresis signals were plotted against the EMD 57033 concentration . KD values were obtained by fitting the data to the Hill equation . Error bars indicate SD ( n = 3 ) . ( B ) Dose-dependent activation of the actin-activated ATPase of β-cardiac myosin S1 by EMD 57033 . Control measurements with the ( − ) -enantiomer EMD 57439 show no activation . Errors indicate s . d . ( n = 4 ) . ( C ) Single-turnover analysis of mantATP binding , hydrolysis , and product release by β-cardiac myosin in the absence ( blue curve ) and presence of 25 µM EMD 57033 ( orange curve ) . ( D ) Fraction of active myosin heads in the absence and presence of saturating concentrations of EMD 57033 determined by active site titrations . Four β-cardiac myosin and four Dd myosin-2 preparations were compared . In the absence of EMD 57033 ( blue bars ) β-cardiac myosin preparations display between 34% and 74% and Dd myosin-2 preparations 76–92% activity . The fraction of active protein increased to 87–100% following the addition of EMD 57033 ( 25 µM for β-cardiac myosin and 100 µM for Dd myosin-2 ) . Addition of EMD 57033 to fully denatured myosin aggregates followed by incubation for 6 hr on ice or at 20°C produced no detectable recovery of enzymatic activity . ( E ) Activation of basal ATPase activity by EMD 57033 . The observed turnover of ATP is 1 . 4-fold ( Dd myosin-2 ) and 1 . 5-fold ( β-cardiac myosin ) higher than expected for preparations of 100% active protein ( dashed line ) . ( F ) EMD 57033-mediated activation of motor activity . The histograms and Gaussian fits show the distribution and average sliding velocity of actin filaments on lawns of β-cardiac myosin in the absence ( blue histograms ) and presence of EMD 57033 ( orange histograms ) . Errors indicate SD ( G ) EMD 57033-mediated increase in force production . A constant frictional load of 8 µg/ml α-actinin was applied to stall β-cardiac myosin-based motility in the absence of EMD 57033 . Filament movement is restarted after the addition of EMD 57033 . Errors indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 00310 . 7554/eLife . 01603 . 004Figure 1—figure supplement 1 . EMD 57033-mediated activation of motor activity for a recombinant Dd myosin-2 motor domain construct . The histograms and Gaussian fits show the distribution and average sliding velocity of actin filaments on lawns of Dd myosin-2 motor ( blue histograms ) . In the presence of EMD 57033 ( 100 µM , orange histograms ) the sliding velocity is increased from 0 . 86 ± 0 . 15 µm s−1 to 1 . 35 ± 0 . 2 µm s−1 . Errors indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 00410 . 7554/eLife . 01603 . 005Figure 1—figure supplement 2 . Frictional loading experiments . Faster filament movment is observed in the presence of EMD 57033 and higher concentrations of α-actinin are required to stall filament movement on surfaces decorated with β-cardiac myosin . Errors indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 00510 . 7554/eLife . 01603 . 006Table 1 . Interaction of EMD 57033 with myosin isoformsDOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 006Myosin constructAC50 ATPaseNormalized maximal ATPase activation ( basal ) Normalized maximal ATPase activation ( with 30 µM actin ) Binding affinity ( MST ) β-Cardiac myosin-2 ( S1 , full-length ) 7 . 0 µM1 . 52 . 57 . 3 µMSkeletal muscle myosin-2 ( HMM ) 15 . 1 µM1 . 62 . 2–Dd myosin-2 motor domain25 . 8 µM1 . 42 . 823 . 0 µMDd myosin-5b motor domain35 . 4 µM1 . 41 . 6–Dd myosin-1B , -1C , -1D , -1E motor domainsn . a . no effectno binding ( myosin-1C/-1D/-1E ) Dd myosin-2 ΔSH3 22 ( lacks residues 33–79 ) n . a . no effectn . a . The number of experimentally accessible active sites is always smaller in preparations of purified enzymes than the number of active sites calculated based on protein concentration . The observed deviation results from incomplete folding , the presence of impurities , and the stress-induced partial loss of function during purification and storage . In the case of myosin , the amplitude of the fast rise in fluorescence intensity that follows binding of mantATP can be used to estimate the fraction of functional protein by active site titration ( Tsiavaliaris et al . , 2002 ) . Figure 1D shows the results of active site titrations for four typical preparations of β-cardiac myosin and a Dd myosin-2 motor domain construct . Dependent on the type of myosin , purification conditions , and length of storage , the fraction of active myosin heads varies between 32% and 92% . The fraction of active sites increased to 87–98% for the same enzyme preparations following the addition of EMD 57033 . The increase in the number of binding-competent active sites is followed by a correlated increase in myosin ATPase activity ( Figure 1D , E ) . In addition to the apparent conversion of inactive to catalytically competent myosin , ATPase activities measured in the presence of EMD 57033 correlate with the total number of myosin heads rather than the initial number of active myosin heads . Moreover , the measured ATPase activities exceed the values expected for a 100% active population of myosin heads by 30–50% ( Figure 1D ) . To probe the effect of EMD 57033 on myosin motor activity , we performed in vitro motility assays . The sliding velocity of actin filaments increased 1 . 8-fold for β-cardiac myosin ( Figure 1F ) and 1 . 6-fold for a Dd myosin-2 motor domain construct ( Figure 1—figure supplement 1 ) . In addition , the fraction of moving filaments increased from <75% to >95% . To examine whether EMD 57033 affects force production by β-cardiac myosin as well , we performed frictional loading experiments ( Greenberg and Moore , 2010 ) . In initial experiments , we determined the minimal concentration of α-actinin molecules that generate sufficient load to stall the movement of actin filaments for a given surface density of myosin heads . In the presence of EMD 57033 , the α-actinin concentration required to stall filament movement increased from 8 µg/ml to ≥12 µg/ml ( Figure 1—figure supplement 2 ) . Next , we generated in vitro motility flow cells with actin filaments held at stall force . The addition of 25 μM EMD 57033 to these flow cells restarts filament movement . The time-dependence of the recovery of motile activity is best fit by a hyperbola . A plateau value of 0 . 4 μm s−1 is reached after more than 2 hr ( Figure 1G ) . Steady-state kinetic assays performed in the absence and the presence of EMD 57033 indicate an allosteric mode of action ( Figure 2—figure supplement 1 ) . The binding site of EMD 57033 in the human β-cardiac myosin motor domain was further defined by molecular docking experiments in combination with direct binding studies using myosin motor domain constructs with alterations in the predicted binding region . Initial blind docking to homology models of the β-cardiac myosin motor domain in the post-rigor state indicated preferred binding of the drug to a region near the small N-terminal , SH3-like βbarrel subdomain . Important contacts are predicted to involve the SH3-like subdomain ( residues 34–72 ) and two α-helices ( residues 20–28 and 98–111 ) . The mode of binding and the orientation of EMD 57033 in this pocket were further analyzed using a flexible , targeted docking procedure . Rearrangements in the side chains of residues Arg29 and Lys34 are predicted to bring these two amino acids into the close vicinity of polar groups in EMD 57033 . The identified binding pocket exhibits a Y shape , supporting two slightly different clusters of binding poses of EMD 57033 with predicted binding free energies in the range of −11 to −12 kcal mol−1 . The sets of binding poses differ mainly in the orientation of the dimethylated catechol group , occupying either of the two possible branches of the Y , while the thiadiazinone moiety appears little affected and resides in the root of the Y shaped cavity ( Figure 2A , B ) . Potential residues involved in the binding of the thiadiazinone and tetrahydroquinoline moieties of the compound include Arg23 , Asp85 , Lys86 , Asp107 , Arg108 , and Ser111 . The dimethylated catechol moiety is mainly bound by residues Lys34 , Lys48 , and Gln79 , and thus , interacts with the SH3-like β-barrel . The contribution of residues from the SH3-like β-barrel to EMD 57033 binding is in good agreement with our observation that all class-1 myosins tested , which lack this subdomain , fail to bind EMD 57033 ( Table 1 ) . To verify that the SH3-like β-barrel is necessary to mediate EMD 57033 binding , we performed direct binding , ATPase , and in vitro motility assays with Dd myosin-2 ΔSH3 , a recombinant motor domain construct that lacks this subdomain ( Fujita-Becker et al . , 2006 ) . We observed neither binding of EMD 57033 to Dd myosin-2 ΔSH3 nor EMD 57033-induced activation of the constructs chemomechanical activity ( Figure 2C , D and Table 1 ) . Further studies using point mutations that eliminate key residues predicted to interact with EMD 57033 are needed to confirm the binding site . 10 . 7554/eLife . 01603 . 007Figure 2 . Localization of the EMD 57033 binding pocket . ( A ) Predicted binding of EMD 57033 near the base of the lever arm as obtained by molecular docking . The overview of the Hs β-cardiac myosin head domain includes residues 1–800 . Two slightly different clusters of binding poses were identified for EMD 57033 with comparable binding affinities . ( B ) Close-up of the allosteric binding pocket shown in surface representation . The Y shape of the pocket is outlined and the two residues—Arg29 and Lys34—that were allowed full conformational flexibility during docking are colored red . For clarity only the best poses for the two identified clusters of binding modes are shown . Hydrogen bonds of EMD 57033 to the motor protein are shown in magenta . ( C ) The SH3-like subdomain is required for EMD 57033 binding and myosin activation . A truncated Dd myosin-2 motor domain construct without SH3-like β-barrel shows no significant activation in the presence of 100 μM EMD 57033 . Errors indicate SD ( n = 4 ) . ( D ) In contrast to the wild-type Dd myosin-2 motor domain construct , no actin filament sliding velocity enhancement was observed upon addition of 100 µM EMD 57033 to the construct with truncated N-terminal SH3-like β-barrel . Errors indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 00710 . 7554/eLife . 01603 . 008Figure 2—figure supplement 1 . Kinetic analysis of the effect of EMD 57033 binding using the Lineweaver–Burk plot . Rates measured in the absence ( blue circles ) and in the presence of 25 µM EMD 57033 ( orange rectangles ) . KM , ATP remains unaffected , whereas kcat is 1 . 5-fold increased . Errors indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 008 To gain a better understanding of its mechanism of action , we examined the effect of EMD 57033-binding on the kinetic activity of the myosin motor . The effect of EMD 57033 on individual steps of the ATPase cycle of β-cardiac myosin was analyzed with the help of transient kinetic measurements and evaluated according to the model shown in Figure 3A , Scheme 1 . Upon addition of 25 µM EMD 57033 to β-cardiac myosin , we observed a 1 . 6-fold increase in kcat , the maximum rate of ATP turnover , and a 10-fold decrease in KM ( actin ) , the F-actin concentration necessary for half-maximal activation ( Figure 3B ) . The extent to which coupling between the actin and nucleotide binding sites is improved in the presence of EMD 57033 is indicated by the resulting 16-fold increase in the apparent second-order rate constant ( kcat/KM ( actin ) ) for the actin-activated ATPase activity ( Table 2 ) . Fast mixing of excess ATP with β-cardiac myosin results in an increase of the protein’s intrinsic tryptophan fluorescence . The kinetic model predicts the double exponential ‘behavior’ of the observed transients to be associated with two discrete processes ( Figure 3C ) . The fast process corresponds to the binding of ATP to the active site . The rate of the slow process saturates at ATP concentrations above 100 µM and can be attributed to the ATP hydrolysis step ( Deacon et al . , 2012 ) . Accordingly , EMD 57033 triggers a 3 . 4-fold increase in the second order rate constant for ATP binding ( K1k+2 ) , a 3 . 8-fold increase in the rate of the subsequent conformational change ( k+2 ) , and a 12 . 9-fold increase in the rate of ATP hydrolysis ( k+3 + k−3 ) ( Figure 3D , E and Table 2 ) . Phosphate release is the rate-limiting step of the β-cardiac myosin actin-activated ATPase cycle . It can be monitored by the fluorescence increase that is associated with the much faster binding of phosphate to the A197C mutant of Escherichia coli phosphate binding protein , labeled at Cys197 with the thiol reactive coumarin dye MDCC ( Brune et al . , 1994 ) ( Figure 3F ) . The observed transients in the absence and presence of EMD 57033 were best fitted by single exponential functions . The resulting values for k+4 indicate a 1 . 7-fold increase in the presence of the small compound . The subsequent ADP-release step ( k−D , k−AD ) appears unaffected by EMD 57033-binding ( Table 2 ) . 10 . 7554/eLife . 01603 . 009Figure 3 . EMD 57033-mediated changes in the interaction with nucleotides and F-actin . ( A ) Kinetic reaction scheme of the actomyosin ATPase cycle . ‘A’ refers to actin , ‘M’ to myosin , ‘T’ to ATP , and ‘D’ refers to ADP . Rate constants are referred to as k+n and k−n and are assigned to the corresponding forward and reverse reactions . An additional notation is used that distinguishes between the constants in the absence and presence of actin by italic type ( k+1 , K1 ) and bold ( k+1 , K1 ) , respectively; subscript A refers to actin ( KA ) and subscript D ( KD ) refers to ADP ( B ) EMD 57033-mediated increase in the actin-dependence of ATP-turnover . In the presence of 25 µM EMD 57033 , the plateau value representing kcat is 1 . 6-fold increased and the apparent actin affinity KM ( Actin ) , indicated by the half-maximal activation of ATPase activity , is 10-fold increased ( values are given in Table 2 ) . Errors indicate SD ( n = 5 ) . ( C ) EMD 57033-mediated increase in the rate of ATP binding and hydrolysis . The fast phase of the biphasic transients correspond to ATP binding , the slow phase can be attributed to ATP hydrolysis . ( D ) EMD 57033-mediated changes in the rate of ATP binding . The observed rates for the fast phase show a linear dependence on ATP concentration . Differences in the slopes indicate a threefold increase of the second order rate constants for ATP binding in the presence of EMD 57033 . Errors indicate SD . ( E ) EMD 57033 mediates a 10-fold increase in the rate of ATP hydrolysis . The plateau values from a hyperbolic fit of the observed slow components of the fluorescence transients plotted against ATP concentrations define the rate of ATP hydrolysis . Errors indicate SD . ( F ) The rate of phosphate release increased 1 . 7-fold in the presence of EMD 57033 . Phosphate release from β-cardiac myosin was observed following rapid mixing with excess ATP and F-actin in the presence of the fluorescent phosphate-binding protein MDCC-PBP . Errors indicate SD . ( n = 6 from three different protein preparations ) . ( G ) Arrhenius analysis of the temperature dependence of β-cardiac myosin ATPase activity . The activation energy for the reaction is 1 . 5-fold reduced in the presence of EMD 57033 . Errors indicate SD . ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 00910 . 7554/eLife . 01603 . 010Table 2 . EMD 57033-mediated changes in the kinetic behavior of β-cardiac myosinDOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 010Control25 µM EMD 57033∼Change ( −fold ) KM ( Actin ) 36 . 8 ± 2 . 67 µM3 . 6 ± 0 . 8 µM10kcat0 . 12 ± 0 . 02 s−10 . 19 ± 0 . 02 s−11 . 6kcat/KM ( Actin ) *0 . 00326 µM−1 s−10 . 0528 µMv1 s−116kcat/KM ( Actin ) †0 . 00193 µM−1 s−10 . 0304 µM−1 s−115 . 8K1k+20 . 19 ± 0 . 02 µM−1s−10 . 65 ± 0 . 04 µM−1s−13 . 4k+246 ± 3 s−1174 ± 7 s−13 . 8k+3 + k−31 . 9 ± 0 . 2 s−124 . 5 ± 1 . 9 s−112 . 9k+4‡0 . 09 ± 0 . 01 s−10 . 15 ± 0 . 01 s−11 . 7k−D0 . 15 ± 0 . 03 s−10 . 16 ± 0 . 04 s−1n . a . k−AD25 . 5 ± 3 s−126 . 9 ± 4 s−1n . a . Ea47 ± 4 kJ mol−131 ± 3 kJ mol−11 . 5*The apparent second order rate constant for actin binding ( kcat/KM ( Actin ) ) was obtained from the calculated ratio of both values . †kcat/KM ( Actin ) was obtained from the initial slope of the steady-state ATPase activity vs the F-actin plot . ‡Measured at F-Actin concentration of 25 µM and 25 µM ATP . To assess the effect of EMD 57033 on the activation energy of ATP turnover by β-cardiac myosin , we recorded the temperature dependence of the reaction over the range 25–45°C . The resulting Arrhenius plot shows that EMD 57033 affects the slope of the recorded linear dependencies , indicating that the Arrhenius energy ( Ea ) is decreased from 47 kJ mol−1 to 31 kJ mol−1 in the presence of the small-compound ( Figure 3G and Table 2 ) . Equilibrium binding of ligands leads to a shift of the midpoint of the thermal transition Tm , which corresponds to the temperature at which half of all protein molecules are in the native state and the remaining half are in the denatured state . Typically , the extent of the observed increase in protein thermal stability is proportional to the concentration and affinity of the added ligand . The Tm measured for the Dd myosin-2 motor domain corresponds to 45 . 6 ± 0 . 2°C in the absence of nucleotide and 52 . 7 ± 0 . 2°C in the presence of 400 µM Mg2+-ADP•BeF3 ( Ponomarev et al . , 2000 ) . In the case of the β-cardiac myosin motor domain , we measured Tm values of 45 . 8 ± 2 . 2°C ( nucleotide-free ) and 52 . 0 ± 0 . 8°C ( +100 µM Mg2+-ATP ) in low salt buffer and 46 . 5 ± 1 . 1°C ( nucleotide-free ) and 54 . 1 ± 0 . 9°C ( +100 µM Mg2+-ATP ) in high salt buffer , respectively . The addition of a saturating concentration of EMD 57033 to nucleotide-free β-cardiac myosin motor domain shifts the Tm to 53 . 6 ± 1 . 9°C ( low salt ) ( Figure 4—figure supplement 1 ) and for the complex with F-actin from 55 . 5 ± 3°C to 66 . 4 ± 3°C ( Figure 4—figure supplement 2 ) . However , the EMD 57033-mediated activation and conversion of inactive to catalytically competent myosin ( Figure 1C , D ) and the greatly improved preservation of enzymatic activity during storage ( Figure 4A ) indicate that EMD 57033 is more potent in preventing the precipitation and irreversible denaturation of the protein than suggested by the shift in Tm alone . 10 . 7554/eLife . 01603 . 011Figure 4 . EMD 57033 acts as a pharmacological chaperone . ( A ) The presence of 25 µM EMD 57033 extends the shelf life of Hs β-cardiac myosin at 4°C . Errors indicate SD ( n = 3 ) . ( B ) EMD 57033 renders β-cardiac myosin more heat-stable . The SDS-PAGE gel shows supernatants ( S ) and pellets ( P ) after 10-min incubation of β-cardiac myosin at the indicated temperatures . In the presence of EMD 57033 β-cardiac myosin remained in the supernatant over the entire temperature range tested . Incubation for 10 min at 49°C leads to the almost complete loss of catalytic activity . The ability to bind mantATP is gradually recovered following the addition of 25 µM EMD 57033 . The dotted line represents the fluorescence amplitude for mantATP binding to native β-cardiac myosin . ( C ) Rescue and activation of heat-treated β-cardiac myosin basal ATPase activity by EMD 57033 . The orange columns indicate changes observed in the presence of EMD 57033 . Errors indicate SD ( n = 3 ) . ( D ) Time-dependent recovery of the capacity to bind nucleotide monitored by ATP-induced changes in intrinsic tryptophan fluorescence . Refolding was initiated by the addition of 50 µM EMD 57033 . Transients obtained after rapid mixing with ATP were measured at the indicated times following the addition of EMD 57033 . ( E ) The observed rate constants for ATP binding to heat-treated β-cardiac display a hyperbolic dependence upon the incubation with EMD 57033 . Errors indicate s . d . ( F ) Effect of EMD 57033 concentration on the time-dependent recovery of heat-treated β-cardiac myosin ATPase activity . Errors indicate SD ( n = 3 ) . ( G ) Time and EMD 57033 concentration dependence of the recovery of β-cardiac myosin motor activity . ( H ) Determination of the second order rate constant for the EMD 57033-mediated refolding reaction . The observed rate constants for the recovery of ATPase activity and motility were extracted from the data shown in Figure 4F , G . The slope of a linear fit to the data defines krescue as 0 . 02 × 10−3 μM−1s−1 . The y-intercept gives a first order dissociation rate constant for EMD 57033 of 0 . 23 × 10−3 s-1 ( I ) Proposed model for the mode of action of EMD 57033 . T Conformationally trapped soluble aggregates , nucleotide binding incompetent; R Rescued , hydrolysis and motility competent; 0 Compromised , hydrolysis incompetent , nucleotide binding competent; A Activated; † Insoluble protein aggregates ( irreversible ) ; ↓ Precipitation . DOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 01110 . 7554/eLife . 01603 . 012Figure 4—figure supplement 1 . Melting curves of the β-cardiac myosin motor domain as determined by circular dichroism spectroscopy at 222 nm ( c = 0 . 3 mg/ml ) . Circular Dichroism was followed along a temperature gradient from 25°C to 90°C in the absence and presence of 100 µM EMD 57033 . Data curves are averaged from 3 individual measurements , normalized and fitted according to the model of Boltzmann . [0] displays the folded state whereas [1] is attributed to the unfolded state . The melting temperatures are 45 . 1 ± 2 . 2°C in the absence of EMD 57033 and 53 . 6 ± 1 . 9°C in its presence , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 01210 . 7554/eLife . 01603 . 013Figure 4—figure supplement 2 . Melting curves of the complex formed by the β-cardiac myosin motor domain with F-actin followed by the change in light-scattering signal . Light scattering was followed along a temperature gradient from 25°C to 90°C in the absence and presence of 100 µM EMD 57033 . The mid-points of the observed transitions correspond to approximately 55 . 5°C in the absence and 66 . 4°C in its presence of EMD 57033 . DOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 013 To elucidate the mode of action of EMD 57033 , we submitted myosin constructs in the absence and presence of the compound to heat stress and followed the resulting changes in solubility and functional competence . The fraction of myosin that is able to bind mantATP drops to below 32% when the temperature is raised to 49°C for more than 5 min . Incubation for 10 min at 51°C removes the myosin completely and irreversibly from the soluble fraction . Preincubation of the myosin with 25 μM EMD 57033 before exposure to heat-stress prevents precipitation ( Figure 4B , insert ) . Moreover , based on the amplitude of the rise in mantATP fluorescence , the competence of β-cardiac myosin to bind nucleotide is preserved to more than 91% after 5-min incubation at 49°C and to approximately 82% after 10-min incubation at 49°C ( Figure 4B ) . The apparent large reduction in the rate of mantATP binding and the extended plateau phase suggest that the observed transients report a rate-limiting step in the refolding of the β-cardiac myosin motor rather than the rate of nucleotide binding ( see Figure 1B for comparison ) . This view is supported by experiments that monitor the recovery of basal activity over a 2 hr period ( Figure 4C ) . To obtain more detailed information about the kinetics of the EMD 57033-mediated refolding reaction , we performed three sets of experiments recording the time and concentration dependence of EMD 57033-mediated increases in the fraction of myosin molecules that are able to undergo nucleotide induced conformational changes , hydrolyze ATP , and produce movement in the in vitro motility assay . Changes in intrinsic protein fluorescence are tightly linked to ATP binding and ATP-induced conformational changes in the myosin motor domain ( Batra and Manstein , 1999 ) . Therefore , the progress of the refolding reaction can be estimated by incubating heat-inactivated β-cardiac myosin with EMD 57033 and recording the change in intrinsic protein fluorescence that follows the addition of substoichiometric amounts of ATP ( Figure 4D ) and excess ATP ( Figure 4E ) at predefined time points over the next two hours . The dependence of the kobs values on the incubation time was best fit by a hyperbola ( Figure 4E ) . Refolding and reactivation of ATPase activity and motility were measured in the presence of 10–100 μM EMD 57033 . In each experimental setting the data obtained for a specific EMD 57033 concentration were best fit by hyperbolae ( Figure 4F , G ) . The recovery of motor activity upon addition of EMD 57033 to flow cells is shown in Video 1 and Video 2 . The kobs values derived from the hyperbolae determined for EMD 57033-mediated changes in ATPase activity and motility ( Figure 4F , G ) and from changes in intrinsic protein fluorescence ( data not shown ) indicate a linear dependence between the progress of the refolding reaction and the concentration of EMD 57033 . The slope defines a value of 0 . 02 × 10-3 μM−1 s−1 for the second order rate constant for the EMD 57033-mediated refolding reaction ( Figure 4H ) . 10 . 7554/eLife . 01603 . 014Video 1 . Video showing fluorescently labeled actin filaments attached to a lawn of catalytically inactive β-cardiac myosin . ( MT state , Figure 4I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 01410 . 7554/eLife . 01603 . 015Video 2 . Reactivation of β-cardiac myosin by the addition of EMD 57033 . Video showing motile actin filaments on the same lawn of β-cardiac myosin three hours after the addition of 10 µM EMD 57033 to the flow cell ( MR state , Figure 4I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 015 Our studies provide evidence for a pro-hypertrophic effect of EMD 57033 ( Figure 5A , B ) . After 1 day exposure to 10 μM EMD 57033 in serum-free growth medium , neonatal rat cardiomyocytes ( NRCM ) displayed a 1 . 8- and 1 . 4-fold increase in cell size without significant changes in α-cardiac MHC transcripts and a reduction in β-cardiac MHC transcripts without significant changes in total cardiac MHC protein content compared to controls ( Figure 5C–F ) . Atrial natriuretic peptide ( ANP ) has been described as a marker that rapidly responds to increased cardiac strain and stress induced by hyperthermia ( Aggeli et al . , 2002; Chen et al . , 2012 ) . ANP expression was not significantly changed in NRCM treated with 10 μM EMD 57033 ( Figure 5G ) . 10 . 7554/eLife . 01603 . 016Figure 5 . Effects of EMD 57033 on neonatal rat cardiomyocytes ( NRCM ) in vitro . ( A ) Cell size and organization of the sarcomere in representative NRCM ( DMSO control ) at 37°C . ( B ) Representative NRCM after 24 hr incubation with EMD 57033 . Cardiomyocytes were immunostained for α-actinin ( green ) and DAPI was used for nuclear staining . ( C ) Bar graph depicting cumulative measurements of cell size of NRCM after 24 hr incubation with EMD 57033 or DMSO at 37°C or 42°C . NRCM display a comparable increase in cell size in response to EMD 57033 stimulation independent of the incubation temperature . ( D ) Bar graph showing expression of α-cardiac MHC mRNA normalized to b2m in NRCM after 24 hr treatment with EMD 57033 or DMSO . EMD 57033 does not alter α-cardiac MHC mRNA expression compared to DMSO controls . Incubation at 42°C significantly reduces expression of α-cardiac MHC mRNA in NRCM compared to cells that were incubated at 37°C . The decrease in α-cardiac MHC mRNA upon induction of hyperthermia was not affected by EMD 57033 treatment . Errors indicate SD ( n = 3 ) . ( E ) Bar graph summarizing mRNA expression of β-cardiac MHC normalized to b2m in EMD 57033 or DMSO-treated NRCM after 24-hr incubation at 37°C or 42°C . Treatment with EMD 57033 significantly reduces β-cardiac MHC expression at 37°C compared to DMSO controls . Incubation at 42°C for 24 hr results in downregulation of β-cardiac MHC mRNA . Downregulation is significantly more pronounced in EMD 57033-treated cells . Errors indicate SD ( n = 3 ) . ( F ) Representative immunoblot depicting expression of total cardiac MHC protein content in NRCM treated with EMD 57033 or DMSO . EMD 57033 treatment does not significantly alter MHC protein levels at 37°C or 42°C and MHC protein content is comparable in NRCM incubated at 37°C or 42°C . ( G ) Bar graph depicting mRNA expression of ANP normalized to b2m in NRCM treated with EMD 57033 or DMSO for 24 hr at 37°C or 42°C . EMD 57033 treatment does not affect expression of ANP mRNA at 37°C . Incubation of NRCM at 42°C induces mRNA expression of ANP in DMSO-treated control cells , which is completely suppressed by treatment with EMD 57033 . Errors indicate SD ( n = 3 ) . NRCM were treated with EMD 57033 ( 10 µM ) or the solvent DMSO ( 1 µl/ ml ) alone for 24 hr in serum-free conditions . Significances are coded as follows: *p<0 . 05 , **p<0 . 01 und ***p<0 . 001 for the DMSO control vs EMD 57033 at a specific temperature; #p<0 . 05 , ##p<0 . 01 und ###p<0 . 001 for a specific group at different temperatures . DOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 016 Hyperthermic stress for 24 hr did not significantly alter cell size or MHC total protein content in NRCM but reduced transcript levels of both α- and β-cardiac MHC ( Figure 5C–F ) . The reduction of β-cardiac MHC remained more pronounced in the EMD-treated NRCM without affecting MHC protein content ( Figure 5 ) . Hyperthermic stress-induced ANP expression in controls , while the presence of 10 μM EMD 57033 fully suppressed the expression of ANP during hyperthermic stress ( Figure 5G ) .
Our results show that EMD 57033 is a member of a new class of pharmacological chaperones that stabilize , enhance the activity , and correct stress-induced misfolding of their target protein . EMD 57033 binds to an allosteric pocket in the myosin motor domain . The site appears to be close to the site where cardiac myosin activator omecamtiv mecarbil is predicted to bind ( Malik et al . , 2011 ) . The greater isoform specificity of omecamtiv mecarbil and preliminary results with other compounds suggest that it is feasible to identify compounds with improved properties for the treatment of aberrant myosin motor activity that are derived from thiadiazinone derivatives and unrelated scaffolds . The efficacy of the resulting drugs may be further enhanced by combining them with drugs that increase the abundance and activity of cellular chaperones ( Willis and Patterson , 2013 ) . The actin-affinity of cardiac β-myosin and the coupling efficiency between motor domain and actin filament are increased in the presence of saturating concentrations of EMD 57033 . The selective shortening of the lifetime of weakly bound states and the associated increase in the fraction of strongly bound motor domains explain the leftwards shift of the force–pCa relation and the reduced oxygen cost of contractility observed in previous studies ( Solaro et al . , 1993; Kraft and Brenner , 1997; Senzaki et al . , 2000 ) . The expected concomitant increase in maximal isometric force production is confirmed by the results obtained using the frictional load assay . In addition to promoting the formation of a super-active MEMD 57033A state , binding of EMD 57033 increases the thermal stability of the cardiac β-myosin motor domain . The observed shifts in melting temperature are significant but do not exceed levels observed with high-affinity binders such as enzyme substrates , substrate analogues , and competitive and allosteric inhibitors . However , in the presence of EMD 57033 the formation of insoluble M† aggregates is reduced to a much greater extent than observed for competitive inhibitors and other stabilizing agents . This more potent effect of EMD 57033 on the retention of enzymatic activity and maintenance of near structural integrity appears to be mediated by its interaction with stress-induced misfolded but soluble MsolubleT intermediates as described in scheme 2 ( Figure 4I ) . EMD 57033 binding to MsolubleT induces the transition of the enzymatically inactive myosin to a nucleotide binding-competent state MEMD 570330 , followed by the transition to a super-active MEMD 57033R state that displays similar enzymatic activity as the EMD 57033-activated native enzyme . The need for separate MEMD 57033A and MEMD 57033R states arises from our observation that only myosin that has not been exposed to any kind of stress can be returned to the Mnative state , when the compound is washed out . Many genetically transmitted diseases lead to mutant proteins with reduced stability . The resulting proteotoxicity is frequently of equal or greater importance for the clinical expression of the associated disease states as the direct , mutation-induced loss in enzymatic activity . Misfolded proteins including mutation-induced misfolded myosins are known to play a central role in the pathophysiology of neurodegenerative diseases . The role of misfolded proteins in the development of numerous other disorders such as type-2 diabetes and heart failure has become evident in recent years ( Willis and Patterson , 2013 ) . Moreover , cardiac dysfunction is a frequent complication associated with elevated body temperature due to fever or hyperthermia ( Marijon et al . , 2012 ) . It is well known that pathophysiologic alterations of the heart are associated with a decrease in α-MHC and an increase in β-MHC expression ( Mercadier et al . , 1983 ) . This is normally associated with an increase in ANP ( Aggeli et al . , 2002 ) . In fact , ANP levels in the myocardium are greatly augmented in patients with congestive heart failure and animal models of ventricular hypertrophy or cardiomyopathy ( Ogawa et al . , 1995 ) . Such changes are indicative of re-expression of a fetal gene expression program , as it is induced by the α-adrenergic agonist phenylephrine ( Hilfiker-Kleiner et al . , 2006 ) . Physiological stress such as exercise induces a shift towards α-cardiac MHC ( Allen et al . , 2001; Baldwin and Haddad , 2001 ) . EMD 57033 induces hypertrophic growth under normothermic conditions that involves a reduction in β-MHC without affecting α-cardiac MHC and thereby changes in the transcriptional regulation that are clearly distinct from phenylephrine-mediated cardiomyocyte hypertrophy ( Hilfiker-Kleiner et al . , 2006 ) . Therefore , EMD 57033 appears to promote a more physiological type of hypertrophy . This notion is further supported by the observation that ANP expression is not induced in the presence of EMD 57033 . In addition , the observed reduction of ANP expression in heat-treated cardiomyocytes suggests that EMD 57033-mediates a reduction in cellular stress responses by stabilizing and refolding denatured MHC proteins . Our results provide evidence for the refolding effect exerted by a small drug-like molecule on a complex protein such as myosin . It remains to be shown whether the refolding activity is shared by other compounds binding to the same region of myosin and to what extent it can be separated from the activation of chemomechanical activity . Small chemical compounds with similar activity profile but increased specificity for their target protein have the potential to improve the treatment of protein misfolding diseases , myopathies , and heart failure . The functionalization of biohybrid devices that exploit actomyosin-based cargo transport for molecular diagnostics and other nanotechnological applications ( Amrute-Nayak et al . , 2010; Lard et al . , 2013; Persson et al . , 2013 ) is another area that will certainly benefit from the use of EMD 57033 and the development of compounds that stabilize their target proteins against stress-induced denaturation , precipitation , and proteolytic degradation . The effective and simple long-term storage of chip-based nanobiosensors , where actomyosin-driven transport substitutes microfluidics and forms the basis for novel detection schemes , is a precondition for the commercial viability of such devices . The addition of EMD 57033 extends the shelf life of protein-functionalized surfaces from a couple of hours to several months , making the widespread and routine use of biohybrid devices feasible .
Stopped-flow experiments were performed using PiStar ( Applied Photophysics , Leatherhead , UK ) and Hi-Tech SF61 ( TgK Scientific Limited , Bradford on Avon , UK ) spectrophotometers at 20°C as described previously ( Taft et al . , 2008 ) . Mant-nucleotides ( Jena Bioscience , Germany ) were excited at 365 nm and detected after passing through a KV 389 nm cut-off filter . Tryptophane fluorescence was excited at 297 nm and detected using a WG 320 nm cut-off filter . Long-time experiments were carried out using a shutter to avoid photo-bleaching . To account for possible absorption or fluorescence effects , control measurements were carried out in the absence of EMD 57033 , as well as in the presence of the non-binding ( − ) -enantiomer EMD 57439 . Phosphate release was measured using MDCC-PBP ( N-[2- ( 1-maleimidyl ) ethyl]-7- ( diethylamino ) coumarin-3-carboxamide fused to phosphate binding protein; obtained from Life Technologies , Carlsbad , CA ) as described previously ( Brune et al . , 1994 ) . Coumarin fluorescence was excited at 430 nm and detected using 455 nm cut-off-filter . We used recombinant myosin constructs fused to yellow fluorescent protein ( Dd myosin-2 , Dd myosin-1B , Dd myosin-1C , Dd myosin-1D , Dd myosin-1E ) or myosins chemically labeled with an amine reactive RED-NHS dye ( Dd myosin-2 , β-cardiac myosin ) . Experiments were performed using hydrophobic capillaries in a NanoTemper Monolith NT . 115 instrument according to Duhr et al . ( Duhr and Braun , 2006; Wienken et al . , 2010 ) . Rabbit fast skeletal muscle heavy meromyosin ( HMM ) was prepared as described by Margossian and Lowey ( 1982 ) . Porcine β-cardiac myosin was prepared essentially as described by Pant et al . ( 2009 ) and Jacques et al . ( 2008 ) and further purified by gel filtration . S1 and HMM fragments were produced by proteolytic digestion of the full-length proteins with 0 . 05 mg/ml papain ( 10 min , 25°C ) for S1 and with 0 . 1 mg/ml TLCK-treated α-chymotrypsin ( 15 min , 25°C ) for HMM . Reactions were stopped with specific protease inhibitors for papain ( 10 µM E−64 ) and chymotrypsin ( 10 µM chymostatin , 1 mM benzamidine ) . Proteolytic fragments were further purified by actin-coprecipitation . Pellets were washed twice and myosin was extracted from the pellets using ATP containing buffer . Gel filtration was used to further increase purity of myosin S1 and HMM fragments . His-tagged motor domain constructs of Dd myosin-2 , Dd myosin-1B , Dd myosin-1C , Dd myosin-1D , Dd myosin-1E , Dd myosin-5b fused to two Dd α-actinin repeats substituting the light-chain binding domains ( Anson et al . , 1996 ) were expressed in Dd AX3-ORF + cells and purified by Ni2+-chelate affinity chromatography ( Manstein and Hunt , 1995 ) . F-actin was prepared by the method of Lehrer and Kerwar ( 1972 ) . We used the molar absorption coefficients at 280 nm to determine the concentration of the myosin constructs in solution . Fast mixing active-site titrations were performed to follow the fluorescence increase of mantATP binding to myosin . We determined the minimal mantATP concentration necessary to fully saturate the binding amplitude . This mantATP concentration defines the actual concentration of active myosin heads in the respective protein preparation . Basal and actin-activated Mg2+-ATPase activities were measured using the NADH-coupled assay described previously ( Furch et al . , 1998 ) adapted for higher throughput screens for a temperature controlled plate reader ( Multiscan FC , Thermo Scientific , Waltham , MA ) at 25°C . EMD 57033 was added to the reaction mixture in the absence of nucleotide and incubated before the reaction was started by the addition of ATP . Control measurements were carried out in the absence of drug , as well as in the presence of according concentrations of the non-binding ( − ) -enantiomer EMD 57439 to account for possible absorption effects . Each reaction mixture including the controls contained 2 . 5% DMSO that was used as solvent for EMD 57033 . Data points are mean values from 3 to 5 independent experiments . Error bars indicate standard deviations . Actin-sliding motility was performed at 20°C using an Olympus IX81 ( Olympus , Hamburg , Germany ) inverted fluorescence microscope as described previously ( Fujita-Becker et al . , 2005 ) . Average sliding velocities were determined from the Gaussian distribution of automatically tracked actin filaments using DiaTrack 3 . 01 ( Semasopht , Chavannes , Switzerland ) and Origin 7 . 0 ( OriginLab Corporation , Northampton , MA ) . Errors indicate standard deviations . Circular dichroism and differential static light scattering experiments were performed to determine Tm . Homology models of the Hs β-cardiac myosin-2 motor domain ( residues 1–800 ) were built using Modeller ( Sali and Blundell , 1993 ) and the X-ray crystal structure of Gg myosin-2 Subfragment-1 ( PDB code: 2MYS ) as template . Prior to molecular docking , the protein models were subjected to geometry optimization using MacroModel ( MacroModel , version 9 . 9; Schrödinger , LLC , New York , NY , 2011 ) and the OPLS2005 force field . Blind and local docking of EMD 57033 was carried out using Autodock4 ( Morris et al . , 1998 ) , employing the Lamarckian Genetic Algorithm . The ligand was prepared and energy-minimized using MacroModel ( MacroModel , version 9 . 9; Schrödinger , LLC , New York , NY , 2011 ) and the OPLS2005 force field , as well as AutodockTools ( Sanner , 1999; Morris et al . , 2009 ) . During flexible docking , the side chains of Arg29 and Lys34 were allowed full conformational flexibility . In addition , local docking was performed using models that are based on the X-ray structure of Hs β-cardiac myosin-2 experimentally solved by the laboratory of I Rayment ( PDB code: 4DB1 ) . Cell culture media , fetal bovine serum ( FBS ) , and horse serum ( HS ) were purchased from Biochrome ( Berlin , Germany ) ; all other chemicals were from Sigma-Aldrich ( St Louis , MO ) . Isolation of neonatal rat cardiomyocytes ( NRCM ) was performed by enzymatic digestion of neonatal rat hearts as described previously ( Hilfiker-Kleiner et al . , 2004 ) . NRCM were cultured in plating medium containing 5% FBS and 10% HS for 24 hr followed by serum-starvation for 48 hr . NRCM were cultured in serum-free medium containing either EMD ( 10 µM ) or the corresponding volume of the solvent DMSO ( 1 µl/ ml ) for additional 24 hr at 37°C or 42°C with 5% CO2 . For indirect immunofluorescence NRCM were fixed with 95% ethanol . Monoclonal antibodies against sarcomeric α-actinin ( 1:50; Sigma-Aldrich ) or sarcomeric Myosin heavy chain ( 1:50; Santa Cruz Biotechnology , Dallas , TX ) were used . Fluor-488 goat anti-mouse IgG ( 1:250; Jackson ImmunoResearch , West Grove , PA ) or Alexa Fluor 639 goat anti-rabbit ( 1:250; Jackson ImmunoResearch ) was employed as secondary antibody . Immunofluorescence was detected by the use of the AxioVison Rel 4 . 1 package ( Carl Zeiss GmbH ) . Realtime measurement of PCR amplification was performed using the Stratagene MX4000 multiplex QPCR System with the SYBR green dye method ( Brilliant SYBR Green Mastermix-Kit , Stratagene , La Jolla , CA ) as described ( Hilfiker-Kleiner et al . , 2010; Hoch et al . , 2011 ) . Total RNA was extracted from NRCM using Trizol reagent according to the manufactures protocol ( Invitrogen , Carlsbad , CA ) . Reverse transcriptase ( RT ) -PCR was performed of ANP and b2m using 2 μg of total RNA . The primer sequences and PCR conditions are described in Table 3 . PCR products were size-fractionated by 2% agarose gel electrophoresis . 10 . 7554/eLife . 01603 . 017Table 3 . Primer sequences and PCR conditions used for realtime PCRDOI: http://dx . doi . org/10 . 7554/eLife . 01603 . 017PrimerPrimer-SequenceAnnealing temperature ( °C ) r-ANP forward5′-GCCGGTAGAAGATGAGGTCA-3′60r-ANP reverse5′-GGGCTCCAATCCTGTCAATC-3′60r-aMHC forward5′-GGAAGAGCGAGCGGCGCATCAAGG-3′55r-aMHC reverse5′-GTCTGCTGGAGAGGTTATTCTCG-3′55r-bMHC forward5′-CAAGTTCCGCAAGGTGC-3′55r-bMHC reverse5′-AAATTGCTTTATTGTGTTTCT-3′55r-b2m forward5′-CATGGCTCGCTCGGTGACC-3′60r-b2m reverse5′-AATGTGAGGCGGGTGGAACTG-3′60 Protein expression in NRCM was determined by standard immunoblotting techniques under denaturing conditions , as described previously ( Hilfiker-Kleiner et al . , 2004 ) . A monoclonal primary antibody against sarcomeric MHC ( 1:1 . 000; Abcam , Cambridge , UK ) was used and signal detection was achieved by the usage of a horseradish peroxydase-conjugated secondary antibody ( GE ) and ECL-detection . PonceauS staining was carried out and used as a loading control . | Our muscles contain large numbers of ‘motor proteins’ called myosins . To contract a muscle , many myosin molecules expend energy to ‘walk’ along a filament made from another molecule , called actin , and generate a pulling force . Like other proteins , myosins must fold into the correct shape to work , but high temperatures or other types of stress can disrupt their ability to adopt or maintain the correct shape . Misfolding of myosins , for example , can result in muscular diseases , including those that affect the heart; so there is an ongoing effort to find compounds that can stabilize protein folding and treat these diseases . The small molecule EMD 57033 was discovered over 20 years ago , and its ability to increase the strength of muscle contractions suggested that it could be used to treat chronic heart failure , but the risk of side effects limited its clinical use . The effectiveness of other compounds that improve cardiac muscle function is still routinely compared to EMD 57033 , however the exact mechanism responsible for its effect on muscle tissue remained unknown . Now Radke , Taft et al . have identified the part of the myosin protein that EMD 57033 binds to , and shown how this activates muscle contraction . The experiments also , unexpectedly , revealed that EMD 57033 is able to convert misfolded myosin back into the fully functional form . By revealing this refolding effect , the findings of Radtke , Taft et al . suggest that similar small molecules could be used as drugs for the treatment of protein misfolding diseases , muscular diseases , and heart failure . | [
"Abstract",
"Introduction",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | Small molecule-mediated refolding and activation of myosin motor function |
Multidrug ATP binding cassette ( ABC ) exporters are ubiquitous ABC transporters that extrude cytotoxic molecules across cell membranes . Despite recent progress in structure determination of these transporters , the conformational motion that transduces the energy of ATP hydrolysis to the work of substrate translocation remains undefined . Here , we have investigated the conformational cycle of BmrCD , a representative of the heterodimer family of ABC exporters that have an intrinsically impaired nucleotide binding site . We measured distances between pairs of spin labels monitoring the movement of the nucleotide binding ( NBD ) and transmembrane domains ( TMD ) . The results expose previously unobserved structural intermediates of the NBDs arising from asymmetric configuration of catalytically inequivalent nucleotide binding sites . The two-state transition of the TMD , from an inward- to an outward-facing conformation , is driven exclusively by ATP hydrolysis . These findings provide direct evidence of divergence in the mechanism of ABC exporters .
ATP binding cassette ( ABC ) transporters harness the energy of ATP to traffic a wide spectrum of molecules across cell membranes . In prokaryotes , ABC importers drive accumulation of nutrients in the cytoplasm against their concentration gradients while ABC exporters remove toxic substrates out of the cytoplasm and may function as flippases of lipids ( Higgins and Linton , 2004; Rees et al . , 2009; Sharom , 2011; George and Jones , 2012 ) . Mammalian ABC transporters , such as P-glycoprotein ( Pgp ) and cystic fibrosis transmembrane conductance regulator ( CFTR ) , are exclusively of the exporter class , play critical physiological roles and are associated with disease ( Higgins and Linton , 2004 ) . Importers and exporters share a modular molecular architecture featuring two nucleotide binding domains ( NBDs or ATP binding cassettes ) that turnover ATP and two transmembrane domains ( TMDs ) that are presumed to form a translocation pathway across the bilayer . The four modules of ABC transporters can be encoded by separate genes and assembled as homo- or hetero-dimers , or expressed as a single polypeptide chain ( Higgins and Linton , 2004; Rees et al . , 2009 ) . Mapping the conformational motion that transduces the energy of ATP binding and hydrolysis in the NBDs to the mechanical work of substrate translocation in the TMDs is central to understanding the mechanism of ABC transporters . Crystallographic snapshots of ABC importers have revealed inward- and outward-facing states ( Locher et al . , 2002; Hollenstein et al . , 2007; Oldham et al . , 2008; Korkhov et al . , 2012 ) in the nomenclature of Jardetzky's alternating access model ( Jardetzky , 1966 ) . Determined in the presence of substrates , substrate binding proteins and/or nucleotides , these structures were cast as representing catalytic intermediates in the ATP binding and hydrolysis cycle . In contrast , the proposed structural mechanism of ABC exporters is less elaborate invoking two states captured by crystallography: Inward-facing devoid of substrates and/or nucleotides ( referred to as apo ) ( Ward et al . , 2007; Aller et al . , 2009; Jin et al . , 2012 ) and outward-facing with bound nucleotides ( Dawson and Locher , 2006 , 2007; Ward et al . , 2007 ) . While these structures highlight the possible range of conformational motion , there is no consensus regarding the suite of conformational steps that couple ATP hydrolysis to substrate translocation ( George and Jones , 2012 ) . The quest for a unified mechanism of transport by ABC exporters has been hampered by seemingly conflicting structural and biochemical models . Inward-facing structures of the bacterial homodimer MsbA ( Ward et al . , 2007 ) and eukaryotic Pgp ( Aller et al . , 2009; Jin et al . , 2012 ) have inverted V-shapes wherein the two leaflets of the transporters are separated by a large chamber open to the cytoplasm and the bilayer . In these nucleotide and substrate-free structures , the two NBDs are disengaged and separated by 10–50 Å . In contrast , nucleotide-bound , outward-facing structures of MsbA ( Ward et al . , 2007 ) and its homolog Sav1866 ( Dawson and Locher , 2006 , 2007 ) have the two NBDs in close contact , bringing together the ABC signature motif from one subunit and the Walker A and Walker B sequences from the other subunit to form the nucleotide binding sites ( NBSs ) . Inward- and outward-facing exporter structures were interpreted to imply that transport entails cycles of association and dissociation of the NBDs powered by ATP binding and hydrolysis . In this model , substrate partitions into the large chamber cradled by the two TMDs and is pushed along to the outer membrane leaflet by ATP-induced alternating access of the chamber . However , central elements of this model are considered inconsistent with mechanistic studies , primarily of the mammalian ABC exporter Pgp , implying catalytic asymmetry between the two NBSs and constant contact between the NBDs during transport ( Tombline and Senior , 2005; Siarheyeva et al . , 2010; George and Jones , 2012 ) . Furthermore , it would appear that an apo state , devoid of nucleotides , is unlikely to be populated or is transient under cellular ATP concentrations which are typically an order of magnitude above the Km of ATP . The ‘alternating site hydrolysis’ model posits that the two NBSs turnover ATP in an alternating manner constantly holding an intact ATP molecule in one of the NBSs ( Tombline and Senior , 2005 ) . The conjecture that the large inward-facing cavity observed in the apo structures is required to accommodate the large substrates of exporters has also been challenged ( George and Jones , 2012 ) by the observation of smaller openings in the outward-facing conformations of MsbA and Sav1866 ( Dawson and Locher , 2006 , 2007; Ward et al . , 2007 ) . Further confounding the structural perspective , a recent structure of an ABC heterodimer TM287/288 from the hyperthermophile Thermotoga maritima offered the first view of an ABC exporter in an inward-facing conformation where the two NBDs are partially engaged ( Hohl et al . , 2012 ) . Unlike MsbA , Sav1866 and Pgp , TM287/288 is a heterodimeric ABC transporter with non-canonical sequences in the Walker B and switch motifs of one subunit resulting in a catalytically impaired NBS . In the TM287/288 structure , the impaired NBS , also referred to as the degenerate NBS , is bridged by an AMP-PNP molecule while the intact NBS , referred to as the consensus NBS , is more open creating an asymmetric NBD interface . Despite the partially engaged NBDs , the TMDs still form a chamber open to partitioning of substrates from the cytoplasm but does not appear to be accessible from the bilayer ( Hohl et al . , 2012 ) . This structure is consistent with biochemical studies of bacterial ABC heterodimers ( Lubelski et al . , 2006; Zutz et al . , 2011 ) that suggest catalytic asymmetry between the NBSs . However , while the TM287/288 structure provides a structural basis for asymmetric nucleotide binding , it does not elucidate the mechanistic role of such an intermediate in the transport cycle in the presence of substrate and under conditions of ATP turnover . Importantly , the broader question of how the mechanism of homodimers and heterodimers differ as a consequence of impairment of an NBS remains unanswered . In addition to their importance to understanding the fundamental mechanism of transport , addressing these questions provides insight into the role of catalytic asymmetry in mammalian ABC transporters such as CFTR and TAP1/2 , which play fundamental physiological roles and have been directly associated with diseases ( Abele and Tampé , 2004; Aleksandrov et al . , 2007 ) . Here , we have investigated the conformational changes associated with specific steps in the substrate-coupled ATPase cycle of an ABC heterodimeric transporter , BmrCD from Bacillus subtilis ( Torres et al . , 2009; Galián et al . , 2011 ) . Similar to other ABC heterodimers , sequence modifications in the consensus motifs of the NBDs , including the replacement of the Walker B catalytic glutamate by an aspartate in BmrD , suggest that ATP turnover is impaired . BmrC and BmrD genes are upregulated in the response of B . subtilis to antibiotics exposure and BmrCD confers multidrug transport activity on inside-out Escherichia coli membrane vesicles ( Torres et al . , 2009 ) and in reconstituted giant unilamellar vesicles ( Dezi et al . , 2013 ) . Spin label pairs were introduced at strategic locations to monitor the movement of the NBDs and TMDs by Double electron–electron resonance spectroscopy ( DEER ) ( Jeschke and Polyhach , 2007; Mchaourab et al . , 2011 ) . The experimental design reconstructs the conformational dynamics of the transporter by comparing distance distributions obtained under turnover conditions with those measured in trapped catalytic intermediates . Our results reveal structural asymmetry at the NBSs presumably reflecting asymmetric binding and hydrolysis of ATP . The ATPase cycle proceeds through multiple conformations of the NBD dimer while the TMDs undergo an inward- to outward-facing transition powered exclusively by ATP hydrolysis . These are distinct mechanistic steps compared to ABC homodimers where a two-state association/dissociation cycle of the NBD is tightly coupled to the transition of the TMD from inward- to outward-facing conformations with ATP binding providing the power stroke ( Dong et al . , 2005; Zou and Mchaourab , 2009 ) . To our knowledge , this is the first study to define the structural dynamic consequences of the selective impairment of an NBS and to uncover divergence in the conformational cycles within the ABC exporter superfamily .
In the absence of nucleotide or substrates , hereafter referred to as the apo state , BmrCD NBDs are disengaged as evident from distances between spin labels that are larger than those predicted on the basis of the nucleotide-bound TM287/288 crystal structure and the BmrCD homology model ( Figure 1—figure supplement 3 ) . Featureless DEER decays at spin label pairs 533/625 and 555/647 , which were designed to monitor the NBD dimer interface , firmly reflect distances longer than 50 and 70 Å , respectively ( black trace , Figure 1B , C , Figure 1—figure supplement 4 ) . The average distance and broad distribution at 533/625 are similar to those observed at the equivalent MsbA site 539 ( Borbat et al . , 2007 ) . Accordingly , they report a relatively large separation between the NBDs in the apo state similar to that of inward-open apo MsbA ( Borbat et al . , 2007; Zou et al . , 2009 ) and are suggestive of considerable conformational flexibility . ATP turnover followed by Vanadate ( Vi ) trapping stabilizes the transporter in an otherwise high-energy post-hydrolysis state ( HES ) , also referred to as the transition state of ATP hydrolysis ( Sharma and Davidson , 2000 ) . The energy input from ATP hydrolysis leads to a large relative movement that brings the two NBDs closer together as manifested by the reduction in the average distances at 533/625 and 555/647 to about 25 and 45 Å , respectively ( red trace , Figure 1B , C ) . The narrower distance distributions indicate a restriction of NBD conformational flexibility . Compared to the equivalent sites in MsbA ( Borbat et al . , 2007; Zou et al . , 2009 ) ( 539 , 561 ) , these distance distributions are consistent with the formation of a closed NBD sandwich similar to that observed in the AMP-PNP bound crystal structures of MsbA ( Ward et al . , 2007 ) and Sav1866 ( Dawson and Locher , 2007 ) . Binding of the non-hydrolyzable ATP analog AMP-PNP stabilizes a conformation distinct from that observed under trapped or apo conditions ( blue trace , Figure 1C ) and well outside the range predicted by the crystal structure and the homology model ( Figure 1—figure supplement 3 ) . This is in stark contrast to the ABC homodimer , MsbA , where ATP binding and HES trapping stabilize the same conformation of the NBD dimer ( Figure 1—figure supplement 5 ) . The 555/647 pair suggests an arrangement of the NBDs where the spin labels are separated by a longer average distance than in the HES but not as disengaged as apo ( Figure 1C ) . These characteristics suggest that ATP binding , mimicked by AMP-PNP , nucleates the formation of an NBD dimer distinct from the canonical closed NBD sandwich observed in the crystal structures of nucleotide-bound MsbA ( Ward et al . , 2007 ) and Sav1866 ( Dawson and Locher , 2007 ) . Remarkably , a similar distance is observed in the presence of ADP ( Figure 1C ) suggesting that both nucleotides stabilize this intermediate state of the NBD dimer . ATP hydrolysis requires the assembly of the NBSs for optimal positioning of the catalytic residues of the conserved motifs ( George and Jones , 2012 ) . Therefore , we monitored the conformation of each NBS in the various catalytic intermediates of BmrCD outlined above . In order to measure distances across the NBSs , spin label pairs were introduced at non-symmetric sites in the BmrC and BmrD protomers . Three pairs of residues monitoring the degenerate and consensus NBSs ( Figure 2 ) were selected to be structurally equivalent based on sequence alignment between BmrC and BmrD and verified by inspection of the structure of TM287/288 ( Hohl et al . , 2012 ) and the BmrCD homology model . 10 . 7554/eLife . 02740 . 009Figure 2 . Structural asymmetry of the NBSs . ( A–C ) Close up side view of the NBD dimer and distance distributions for spin label pairs monitoring the consensus NBS . The overlapping distributions in the AMP-PNP bound and ADP-Vi trapped intermediates demonstrate that ATP binding can induce formation of the HES state . In contrast , distance distributions of structurally equivalent pairs monitoring the degenerate NBS ( D–F ) are predominantly non-overlapping . The views in ( D–F ) are obtained by rotating the views of A–C by 180° . See also Figure 1—figure supplement 2B and Figure1—figure supplement 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 00910 . 7554/eLife . 02740 . 010Figure 2—figure supplement 1 . DEER data analysis for spin-labeled pairs in the NBSs . Close up view of BmrCD homology model highlighting the location of the pairs ( A ) ( 440/441 ) , ( B ) ( 348/532 ) along with the baseline-corrected DEER decays and the corresponding distance distributions . Addition of Hoechst does not affect the distance distribution in detergent micelles . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 010 We found that nucleotide-induced changes in distance distributions reported at the NBSs ( Figure 2 ) are consonant with those reported by the symmetry-related 555/647 which monitors NBD dimerization ( Figure 1 ) . Three distinct conformations for apo , AMP-PNP-bound and the HES intermediates are deduced from the shifts in distance distributions at the NBSs for all the pairs . However , the shape of the distributions uncovers fundamental structural differences between consensus and degenerate NBS . Consistently , spin labels reporting on the consensus NBS have overlapping distributions in the AMP-PNP-bound and trapped HES intermediates whereas at the degenerate site the distributions are by and large distinct ( Figure 2 , compare panels A , B , and C with D , E , and F ) . The breadth of the AMP-PNP distribution at the consensus NBS demonstrates that ATP binding enables a range of conformations to be populated that includes the HES . In contrast , ATP hydrolysis is required to observe an HES population at the degenerate NBS as inferred from the predominantly distinct distributions in the presence of AMP-PNP and ADP-Vi . In the presence of excess ADP , the degenerate NBS has a predominantly ATP-bound like conformation while the consensus NBS has a significant population of apo-like conformation ( Figure 2 , Figure 1—figure supplement 4 ) . Both NBSs disengage in the absence of nucleotides consistent with the relative distance and conformational flexibility observed at the NBD interface ( Figure 2 , black traces ) . Addition of the transport substrate Hoechst ( Torres et al . , 2009 ) did not change the distance distributions of the AMP-PNP and ADP-Vi states ( Figure 2—figure supplement 1 ) . Conformational changes in the TMDs , as a consequence of ATP binding and hydrolysis , were monitored by spin label pairs on the intracellular and extracellular sides of the transporter ( Figure 3 ) . These pairs were introduced at equivalent residues to MsbA spin label pairs which reported structural rearrangements associated with the inward- to outward-facing transition ( Borbat et al . , 2007 ) . Specifically , the 55/146 pair monitors the extracellular loops between helices 1 and 2 at a structurally similar position to MsbA residue 61 while the pair 96/188 monitors the movement of the intracellular region of transmembrane helix 2 ( TM2 ) similar to residue 103 in MsbA ( Figure 3B , C ) . 10 . 7554/eLife . 02740 . 011Figure 3 . The inward- to outward-facing transition of the TMD requires ATP hydrolysis . ( A ) Ribbon representation of the BmrCD homology model showing the spin-labeled pairs designed to monitor the conformation of the TMD . Residue 147 is shown since 146 was not included in the homology model . ATP turnover and trapping with Vi , but not AMP-PNP binding , induces opening of ( B ) the extracellular side and ( C ) the closing of the intracellular side of the transporter . See also Figure 1—figure supplement 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 01110 . 7554/eLife . 02740 . 012Figure 3—figure supplement 1 . CW-EPR and DEER data analysis for spin-labeled pairs in the TMDs . ( A ) Close up view of BmrCD homology model highlighting the location of the spin-labeled pairs at the extracellular ( upper panel ) and intracellular ( lower panel ) side . * residue 146 was not modeled therefore residue 147 is displayed . ( B ) Superposition of the CW-EPR spectra demonstrates minimal changes in the EPR lineshape , and by extension the rotamer preferences of the spin label pairs , following addition of nucleotides . ( C ) Baseline-corrected DEER signals along with the fits corresponding to the distance distributions in panel ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 012 The pattern of average distances observed in the apo state of BmrCD is consistent with an open intracellular side and a relatively more closed extracellular side ( Figure 3B , C , black trace , Figure 3—figure supplement 1 ) . Furthermore , the intracellular side of apo BmrCD appears to be more closed than that of apo MsbA consistent with TM287/288 structure and the homology model . For instance , the distance between the symmetry-related residues 103 in the MsbA homodimer ( previous work Borbat et al . , 2007 and see below ) is outside the range of measurable distances by DEER , at least 10 Å longer than the distance measured at the equivalent BmrCD pair , 96/188 . While ATP binding , mimicked by addition of AMP-PNP , induces the formation of a distinct NBD intermediate in BmrCD , we found that this conformational change is not propagated to the TMD . Spin label pairs at the intracellular and extracellular sides have identical distance distributions in apo and AMP-PNP-bound ( Figure 3 , black and blue traces respectively ) . This finding suggests relative flexibility between the NBD and TMD that allows large independent movement of the former upon ATP binding . In contrast , when ATP is hydrolyzed and ADP subsequently trapped by the addition of Vi , we observed concomitant shifts in the average distance at the intra- and extracellular sides of the transporter . Coupled to the formation of the NBD closed dimer , ATP hydrolysis induces closing of the intracellular side and opening of the extracellular side consistent with an alternating access mechanism involving inward-facing and outward-facing conformations . Compared to MsbA ( residue 103 ) ( Borbat et al . , 2007 ) the distance change at the intracellular side is of smaller magnitude ( approximately 15 Å for BmrCD 96/188 vs 25 Å for MsbA 103 ) primarily due to a closer distance between the TMDs in the apo conformation of BmrCD ( Figure 3 ) . More importantly , while conformational changes in MsbA are induced by the binding of AMP-PNP , the inward- to outward-facing transition in BmrCD requires ATP hydrolysis revealing a distinct power stroke between homodimers and heterodimers . To test whether the substrate-coupled ATP turnover cycle in lipid bilayers involves the conformations identified in detergent micelles , we determined distance distributions in the NBD and TMD pairs following reconstitution of BmrCD double mutants in nanodiscs of PC/PA ( Figure 4 , Figure 4—figure supplement 1 ) . This particular lipid mixture was selected on the basis of previous results showing optimal ATPase activity ( Galián et al . , 2011 ) . Reconstitution in nanodiscs stimulated the ATP turnover rates of BmrCD-WT and BmrCD-WT* by 11–14-fold relative to detergent micelles ( Figure 4 , Figure 4—figure supplement 2 ) . The rates were further stimulated two-fold in the presence of Hoechst . 10 . 7554/eLife . 02740 . 013Figure 4 . Conformational states of BmrCD in lipid bilayers . ( A ) Close up view from cytoplasmic side for spin label pairs monitoring the NBD dimer , the consensus NBS and the degenerate NBS along with the corresponding distance distributions . ( B ) Close up view from the membrane for spin label pairs monitoring the closing on intracellular side and ( C ) the opening on extracellular side of the transporter . The arrow points to the distance corresponding to HES . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 01310 . 7554/eLife . 02740 . 014Figure 4—figure supplement 1 . Reconstitution of BmrCD-WT* ( Cysteine-less BmrCD ) and its spin-labeled mutants in PC/PA nanodiscs . ( A ) Size-exclusion chromatography of BmrCD in nanodiscs and a cartoon depicting BmrCD-WT* assembly in nanodiscs . Similar size-exclusion chromatographs were obtained for spin-labeled mutants . ( B ) SDS-PAGE of BmrCD-WT* and its spin-labeled mutants assembled in nanodiscs showing the presence of MSP and BmrCD . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 01410 . 7554/eLife . 02740 . 015Figure 4—figure supplement 2 . The ATPase activity of BmrCD-WT ( wild type BmrCD ) , BmrCD-WT* ( Cysteine-less BmrCD ) and its spin-labeled mutants in PC/PA nanodiscs is stimulated by Hoechst ( 10 µM ) and inhibited by vanadate ( 5 mM ) . ( A ) Reconstitution of BmrCD-WT , BmrCD-WT* and its spin-labeled mutants in PC/PA nanodiscs stimulates the basal ATPase activity . The solid line is a non-linear least-squares fit that yields Vmax ( 527 . 6 ± 14 . 9 nmol/min/mg ) , Km ( 1 . 53 ± 0 . 2 mM ) for BmrCD-WT ( basal ) ; Vmax ( 1065 ± 32 . 9 nmol/min/mg ) , Km ( 1 . 65 ± 0 . 18 mM ) for BmrCD-WT ( in presence of Hoechst ) and Vmax ( 494 . 2 ± 45 . 41 nmol/min/mg ) , Km ( 2 . 19 ± 0 . 39 mM ) for BmrCD-WT* ( basal ) ; Vmax ( 1067 . 2 ± 98 . 3 nmol/min/mg ) , Km ( 2 . 13 ± 0 . 37 mM ) for BmrCD-WT* ( in presence of Hoechst ) . Vmax and Km values for BmrCD-WT , BmrCD-WT* and spin-labeled BmrCD are the average of three independent measurements . ( B ) All spin-labeled BmrCD pairs used for this study turnover ATP well above the background observed from inhibition by vanadate . Vmax increased for all pairs following the addition of Hoechst although the level of stimulation was somewhat variable . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 01510 . 7554/eLife . 02740 . 016Figure 4—figure supplement 3 . CW-EPR and DEER data analysis for BmrCD spin-labeled pairs reconstituted in nanodiscs . ( A ) Close up view of BmrCD homology model highlighting the location of the spin-labeled pairs in the NBDs and NBSs . ( B ) Superposition of the CW-EPR spectra demonstrates minimal changes in the lineshape , and by extension the spin label rotamer preferences . ( C ) Baseline-corrected DEER signals along with the fits corresponding to the distance distributions in panel ( D ) . The arrow points to the distance corresponding to HES . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 01610 . 7554/eLife . 02740 . 017Figure 4—figure supplement 4 . CW-EPR and DEER data analysis for BmrCD spin-labeled pairs reconstituted in nanodiscs . ( A ) Close up view of BmrCD homology model highlighting the location of the spin-labeled pairs at the extracellular ( upper panel ) and intracellular ( lower panel ) side . * residue 146 was not modeled therefore residue 147 is displayed . ( B ) Superposition of the CW-EPR spectra demonstrates minimal changes in the EPR lineshape , and by extension the rotamer preferences of the spin label pairs , following addition of nucleotides . ( C ) Baseline-corrected DEER signals along with the fits corresponding to the distance distributions in panel ( D ) . The arrow points to the distance corresponding to the HES . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 017 Distance distributions in nanodiscs reveal components almost identical to those in detergent micelles ( Compare Figure 4 with Figures 1–3 , ) . Spin label pairs in the NBD show distinct distances in the apo state , AMP-PNP bound state , and the HES consistent with three conformations of the NBDs ( Figure 4 , top panel , Figure 4—figure supplement 3 ) . In contrast , distance distributions of spin label pairs in the TMD identify only two conformational states corresponding to inward and outward-facing conformations ( Figure 4 , lower panel , Figure 4—figure supplement 4 ) . ATP hydrolysis is required to transition between these states whereas BmrCD bound to AMP-PNP- is in the inward-facing conformation , in agreement with the observations in detergent micelles described above ( Figure 3 ) . Thus , not only is there a correspondence between conformational states in detergent micelles and lipid bilayers but also the power stroke is essentially identical . Remarkably , the transition between inward- and outward-facing conformations in nanodiscs shows a more stringent dependence on the presence of substrate ( Figure 4 , Figure 4—figure supplements 3 and 4 ) . Thus , although ATP is turned over in the absence of substrate , transmission of this hydrolysis step is enhanced by the presence of the substrate Hoechst . To frame the stable intermediates of the NBSs identified above in the context of the transport cycle , distance distributions were obtained under conditions that reflect cellular ATP concentrations and allow the transporter to sample intermediates associated with ATP turnover ( Figure 5 , Figure 5—figure supplement 1 ) . Following incubation with ATP in the presence and absence of the substrate Hoechst , distance distributions were compared at two time points , where the ATP concentration is calculated to remain above the Km . At the degenerate NBS , we observed predominantly two-component distance distributions that reflect two distinct transporter populations ( Figure 5C ) . By comparing distance distributions for the same spin label pair in trapped intermediates and under turnover conditions , the two components are identified as arising from nucleotide-bound ( ATP or ADP ) and HES-like states . The population ratio of the two components shifts from the HES-like towards the nucleotide-bound state with little if any apo population observed at the two time points . In contrast , multicomponent distributions at the consensus NBS ( Figure 5B ) show , in addition to the HES-like and nucleotide-bound states , clear evidence of an apo-like population where the two NBDs are locally disengaged . 10 . 7554/eLife . 02740 . 018Figure 5 . Structural asymmetry of the NBSs under turnover conditions . ( A ) Ribbon diagram of BmrCD NBDs showing the location of spin label pairs monitoring the consensus and the degenerate NBS ( the NBD dimer is viewed from the cytoplasm , along the membrane normal ) . Following 1 or 5 min of incubation with 10 mM ATP at 30°C in ( B ) and ( C ) in presence ( dashed traces ) or absence ( solid traces ) of the substrate Hoechst ( H ) , samples of spin-labeled BmrCD were immediately cooled to 4°C and then analyzed by DEER spectroscopy ( Jeschke and Polyhach , 2007; Mchaourab et al . , 2011 ) . The two panels labeled ‘Catalytic intermediates’ are identical to those in Figure 2 and serve as a reference to assign the components in the distance distributions under turnover conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 01810 . 7554/eLife . 02740 . 019Figure 5—figure supplement 1 . DEER data analysis for spin-labeled pairs in the NBSs under turnover conditions . Panels from left to right: location of spin-labeled pairs on BmrCD homology model , baseline-corrected DEER decays along with the fits and the resulting distance distributions . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 019 Thus , the two NBSs cycle between conformational states as ATP is bound , hydrolyzed and Pi and ADP are released . However , the differences in the population amplitude of these states reveal that the two NBSs do not have to concurrently be in the same catalytic intermediate . For example , while a significant population of the consensus NBS is in apo-like conformation , the degenerate NBS is predominantly in either the nucleotide-bound or HES-like conformations . Such an asymmetric disengagement of the signature and Walker A motifs at the consensus NBS , which requires their separation by more than 20–40 Å , is not structurally compatible with an HES-like conformation at the degenerate NBS . Therefore , the distributions under turnover conditions imply the existence of a transporter intermediate wherein the degenerate NBS is ATP-bound while the consensus NBS assumes an apo-like conformation . Similarly , we can infer the presence of another state of the transporter wherein the degenerate NBS is in a HES-like state and the consensus NBS is in the ATP-bound conformation . Longer incubation times shift the distribution towards the ATP/ADP-bound population at the degenerate NBS ( compare 1 and 5 min time points in Figure 5C ) while at similar incubation times , the consensus site shows a considerably larger population in the ATP-bound and the apo states ( Figure 5B ) . The addition of the substrate Hoechst , which accelerates the rate of ATP hydrolysis ( Figure 1—figure supplement 2 ) , simply shifts the populations at both NBSs towards those obtained at the longer incubation time . Thus , the substrate-coupled ATP turnover cycle of BmrCD involves the same conformational states sampled in the absence of substrate . Coupled to the ATP hydrolysis cycle of the NBSs , the TMD undergoes an inward- to outward-facing transition as evidenced by the time dependence of the distance distributions ( Figure 6 , Figure 6—figure supplement 1 ) . Similar to the NBSs , the TMDs show multi-component distance distributions . However , the equilibrium favors the inward-facing conformation as demonstrated by the ratio of the two populations at high ATP concentrations . This presumably reflects the failure of ATP binding to switch TMD accessibility from inward-facing to outward-facing ( Figure 3 ) . 10 . 7554/eLife . 02740 . 020Figure 6 . In the presence of excess ATP , the inward-facing state of BmrCD is favored . Distance distributions of spin label pairs monitoring the ( A ) extracellular and ( B ) intracellular side demonstrate that in the presence of 10 mM ATP the predominant population is inward-facing . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 02010 . 7554/eLife . 02740 . 021Figure 6—figure supplement 1 . DEER data analysis for spin-labeled pairs in the TMDs under turnover conditions . Panels from left to right: location of spin-labeled pairs on BmrCD homology model , baseline-corrected DEER decays along with the fits and the resulting distance distributions . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 021 While the previous body of spin labeling data on MsbA establishes that ATP binding is the power stroke of transport ( Dong et al . , 2005; Zou and Mchaourab , 2009 ) , we sought to conclusively investigate if the NBDs can adopt conformations besides the disengaged apo and the closed dimer stabilized by ATP binding or ADP-Vi trapping . Pairs of spin labels were introduced in each MsbA protomer with the purpose of monitoring the two NBSs following the design principles described above for BmrCD . Because MsbA is a homodimer , this strategy introduces four spin labels in the functional unit related by six distances ( Figure 7A , B ) . If the NBD interface is twofold symmetric , three of these distances are unique ( d1 , d2 , d3 , in Figure 7B ) . Two distances relate labels at distinct sites: one is short range ( d1 ) arising from pairs monitoring the NBS while the other represents symmetry-related sites that interact across the dimer interface and is considerably longer ( d2 ) . Therefore , in addition to considerations of ATPase activity ( Figure 7—figure supplement 1 ) and surface localization , the two sites , 350 and 471 , were selected to maximize the differences between spin labels reporting on the NBS conformation vs those reporting the distance between symmetry-related spin labels in the dimer . These selection criteria simplify the interpretation of the distance distributions and allow for the identification of the components that arise from dipolar coupling between spin labels across the NBS . However , the projection of the spin labels relative to the Cα breaks the symmetry leading to distinct distances in the 350/350 and 471/471 pairs . This complication does not affect the interpretation of the short component which monitors the NBSs . 10 . 7554/eLife . 02740 . 022Figure 7 . MsbA conformational dynamics under turnover conditions reveal a two-state equilibrium . ( A ) Ribbon representation of the AMP-PNP structure of MsbA ( PDB:3B60 ) showing the spin-labeled pairs . ( B ) Binding of ATP induces the formation of the closed NBD dimer and concomitantly drives the transition to an outward-facing conformation of the TMD ( C and D ) In the presence of excess ATP , the equilibrium favors the outward-facing conformation ( green trace ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 02210 . 7554/eLife . 02740 . 023Figure 7—figure supplement 1 . ATPase activity of MsbA . ( A ) Hoechst33342 ( hereafter referred to as Hoechst ) does not stimulate , but vanadate inhibits the ATPase activity of cysteine-less MsbA ( MsbA-WT* ) and ( B ) spin label mutants . ATPase activity of MsbA-WT* in the absence and presence of vanadate ( 1 mM ) was measured as described in the methods section . The solid line is a non-linear least-squares fit that yields Vmax ( 1 . 32 ± 0 . 059 µmol/min/mg ) , Km ( 0 . 38 ± 0 . 06 mM ) for MsbA-WT* ( basal ) ; Vmax ( 1 . 4 ± 0 . 09 µmol/min/mg ) , Km ( 0 . 33 ± 0 . 06 mM ) for MsbA-WT* ( in presence of Hoechst ) . Vmax and Km values for MsbA-WT* and spin label pairs are derived from three independent measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 02310 . 7554/eLife . 02740 . 024Figure 7—figure supplement 2 . DEER data analysis for spin-labeled pairs . ( A ) Ribbon representation of the TMDs and NBDs of MsbA showing the spin label pairs ( B ) CW-EPR spectra ( C ) baseline-corrected DEER signals along with the fits and ( D ) the resulting distance distributions . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 024 Distance distributions were obtained in the apo , AMP-PNP-bound , ADP-Vi trapped and under turnover conditions ( Figure 7B , C , D ) . As expected , MsbA NBDs completely dissociate in the absence of nucleotides as evidenced by the absence of a short distance component that would be expected from the assembly of the signature motifs into a NBS . In contrast , this short distance component is present to the same extent in the AMP-PNP-bound and ADP-Vi intermediates supporting the previous conclusion that ATP binding is the power stroke of transport for MsbA ( Dong et al . , 2005; Zou and Mchaourab , 2009; Figure 7B ) . We verified that this component arises from the formation of the NBS by measuring distance distributions for the corresponding single mutants ( Figure 7—figure supplement 2 ) . No other short distance component is detected as would be expected if the NBS were structurally asymmetric; although the detection of such asymmetry is limited by the intrinsic width of the distance distributions . Thus , we conclude that the two NBSs are structurally equivalent in the HES . The longer distance components in the distance distribution can be accounted for by the other distances ( d2 , d3 ) relating the two pairs of spin labels in the homodimer ( Figure 7—figure supplement 2 ) . The two conformations of the NBDs , that is the disengaged and closed , are coupled to inward- and outward-facing conformations of the TMD respectively ( Figure 7C , D ) . As previously reported , spin label pairs at the intra- and extracellular sides of the transporter report concurrent but opposite changes in average distance upon ATP binding suggesting that there is tight coupling between the movements of the NBDs and TMDs ( Borbat et al . , 2007; Zou et al . , 2009 ) in contrast to the relative flexibility between the two domains in BmrCD . Under turnover conditions , the distance distributions were distinctly bimodal in the presence of ATP consistent with a partitioning of the transporter between inward- and outward-facing bound conformations . In contrast to BmrCD , the equilibrium of MsbA favors an outward-facing conformation . The predominance of the outward-facing conformation of the TMD is consistent with the persistence of the NBS short distance component under turnover conditions ( Figure 7B ) .
The major novel finding of this paper is that sequence asymmetry in the NBDs shapes the mechanism of energy transduction in ABC exporters . Modifications of the Walker B , signature and/or switch motifs sequences , which lead to selective catalytic impairment of one of the NBSs in a subclass of ABC exporters ( Gao et al . , 2000; Aleksandrov et al . , 2002; Zhang et al . , 2006; Procko et al . , 2009; Boncoeur et al . , 2012 ) , are associated with fundamental differences in the conformational dynamics of the NBDs and in the power stroke of transport . In contrast to the ATP-induced association/dissociation cycle of NBDs in bacterial homodimers reported previously ( Borbat et al . , 2007; Zou et al . , 2009 ) and further confirmed here , the NBDs of heterodimers form structurally asymmetric dimers as a consequence of ATP binding and its subsequent hydrolysis . We uncovered conformations wherein the impaired NBS holds on to an ATP molecule while the consensus NBS disengages enabling product release and rebinding of ATP . Asymmetric NBD conformations were hypothesized from the elegant mechanistic analysis of Pgp by Senior et al . but were not directly detected ( Tombline and Senior , 2005; Tombline et al . , 2005 ) . We propose that the distinct mechanism of heterodimers requires a less efficient coupling between the NBD and TMD , a ‘conformational leak’ , to enable NBD movement without transmission to the TMDs . This would require a tuning of the sequences and interactions of the IH1 and IH2 coupling loops and possibly the cytoplasmic extension of TMs , regions hypothesized to form the transmission interface between the NBDs and TMDs of ABC transporters ( Hollenstein et al . , 2007; Oancea et al . , 2009; Loo et al . , 2013 ) . There is a substantial impact of lipids on the coupling between the NBD and TMD . Transmission of the conformational changes to the TMD is more stringently dependent on the presence of substrate in lipid bilyaers . This finding is consistent with previous work revealing that lipids can shift the conformational preferences of ion-coupled transporters ( Hanelt et al . , 2013 ) . This novel insight was derived from experiments that monitor transporter conformational states under turnover conditions . Unlike previous studies , which investigated exclusively stable intermediates ( Borbat et al . , 2007; Zou et al . , 2009 ) , here the mechanistic implications of these intermediates were challenged under substrate and nucleotide conditions that mimic those of the cell . Another unique experimental design is the selective monitoring of individual NBSs rather than the NBD interface facilitated by the heterodimeric nature of BmrCD . NBS distance distributions capture the interconversion of the transporter between conformations corresponding to those observed under trapped conditions , thereby establishing the direct relevance of the trapped conformations to the substrate-coupled ATPase cycle . Furthermore , the observation of a steady state population of apo-like conformation of the consensus NBS , even in the presence of excess ATP , prompts us to conclude that disengagement of this NBS is required for ADP-Pi dissociation and subsequent binding of ATP , thereby establishing the mechanistic significance of a local apo-like conformation . Our results can be framed in a model of how asymmetric ATP binding and hydrolysis in the NBDs drive transition of the TMDs between inward- and outward-facing conformations ( Figure 8A ) . Distinguishing features of this model include the formation of structurally asymmetric conformations of the NBDs and the association of the transporter power stroke with ATP hydrolysis . In the presence of cellular ATP concentrations , BmrCD will have at least one ATP molecule bound , most likely at the degenerate NBS as evidenced by the fact that an apo-like conformation was nearly undetectable at this NBS in the presence of excess ATP . Such an ATP-bound resting state does not hinder substrate access to the putative binding chamber at the TMD interface , since ATP binding does not induce closure of the cytoplasmic side of BmrCD . ATP binding to the consensus NBS yields an intermediate with two ATP molecules bound ( ‘1’ in Figure 8A ) . We refer to this conformation of the NBD dimer ( Figure 2 ) as a pre-hydrolysis intermediate reminiscent of the pre-hydrolysis complex observed in a crystal structure of the maltose transporter , MalK ( Oldham and Chen , 2011 ) . In this conformation , the NBDs are engaged but the NBSs have not formed the canonical configuration and are therefore not catalytically competent . ATP turnover is initiated by fluctuations at the consensus NBS which enables transition to a hydrolysis-competent configuration ( HES-like ) as deduced from the overlapping distance distributions in Figure 2 . This is a relatively minor population of the transporter which likely explains the relatively slow ATP turnover by BmrCD ( Figure 1—figure supplement 2 ) . ATP hydrolysis switches TMD accessibility thereby catalyzing the extrusion of the substrate . Turnover rates are stimulated if the substrate binds prior to hydrolysis but there is a considerable basal consumption of ATP ( Figure 1—figure supplement 2 ) . While there is evidence of ATP turnover by impaired NBSs , consideration of the reaction mechanism as well as previous biochemical studies of ABC heterodimers support an intrinsically slower rate at the degenerate NBS ( Procko et al . , 2006; Zutz et al . , 2011; Boncoeur et al . , 2012 ) . Therefore , ADP and Pi are released from the consensus site prior to hydrolysis at the degenerate site enabling multiple rounds of asymmetric ATP hydrolysis without the complete disengagement of the NBDs to an apo conformation . 10 . 7554/eLife . 02740 . 025Figure 8 . Distinct mechanisms of ABC heterodimers and homodimers . ( A ) Asymmetric hydrolysis of ATP in an ABC heterodimer . The conformation of the transporter and a top view of the NBD dimer are shown for each catalytic state . Substrate binding can precede or follow ATP binding to the NBSs . The cycle is triggered when ATP is bound at both NBSs ( 1 ) . The top view of the NBDs shows the partially engaged ATP-bound intermediate state . Hydrolysis proceeds preferentially at the consensus NBS ( 2 ) driving substrate extrusion and leading to another asymmetric state of the NBD wherein the degenerate NBS is ATP-bound while the consensus NBS is ADP-bound ( 3 ) . Dissociation of ADP occurs via a local apo-like conformation of the consensus NBS ( 4 ) thereby enabling a new transport cycle . ( B ) Two-state cycle of an ABC homodimer . Here substrate must precede ATP binding , which induces the parallel formation of the closed NBD and the outward-facing state of the TMD . Sequence symmetric NBSs hydrolyze two ATP molecules prior to disengagement of the NBDs . DOI: http://dx . doi . org/10 . 7554/eLife . 02740 . 025 Slow ATP hydrolysis at the degenerate NBS would require release of ADP and Pi and rebinding of ATP most likely through the population of a local apo-like conformation . Whether ATP hydrolysis at the degenerate NBS could occur prior or concomitant with release of ADP and Pi from the consensus NBS cannot be conclusively addressed from our data . However , the negligible population of apo-like degenerate NBS and the fact that ADP-bound NBDs form a partially engaged dimer argue against significant population of an apo state of the transporter where the two NBDs are separated by more than 50 Å . Thus , this ‘global’ apo state ( as in Figure 1 ) appears not to be mechanistically relevant in the ATP hydrolysis cycle of ABC heterodimers . We propose that ADP-Pi release , as a consequence of infrequent hydrolysis at the degenerate site , occurs while ATP is bound at the consensus NBS . If the degenerate site is not turning over ATP at levels comparable to the consensus NBS , how is the significant HES-like population observed under turnover conditions ( Figure 5B ) accounted for ? We speculate that hydrolysis at the consensus NBS requires transition of the degenerate NBS to an ATP-occluded state along the lines of the model proposed for Pgp on the basis of experiments with catalytically impaired NBSs ( Tombline and Senior , 2005 ) . This state would relax to the nucleotide-bound conformation following hydrolysis at the consensus NBS ( state 3 in Figure 8A ) . The mechanistic elements of ATP turnover for ABC heterodimers and homodimers are contrasted in Figure 8 . For the latter , the NBSs are identical in sequence and there is no experimental evidence of catalytic inequivalence . Analysis of MsbA conformational dynamics reported previously ( Dong et al . , 2005; Borbat et al . , 2007; Zou and Mchaourab , 2009 ) and extended here is entirely consistent with a model of turnover that invokes engaged and disengaged NBD states , symmetric NBS conformations , and tight structural coupling between the NBDs and the TMDs . We conclude from the comparative analysis of BmrCD and MsbA conformational cycles that the controversy regarding NBD disengagement vs constant contact ( George and Jones , 2012 ) is a consequence of extrapolation of models between mechanistically distinct classes of ABC exporters . The structural mechanism of ABC heterodimers described here is a remarkable example of variations on a common conformational cycle in the context of conserved transporter architecture . While all ABC transporters derive energy from an ATP binding and hydrolysis cycle , fundamental differences exist in the regulation of ATP turnover by catalytic elements and/or structural intermediates as evidenced here . Although it was anticipated that the inward- to outward-facing transition of ABC exporters would follow the structural mechanics defined by the structures of MsbA ( Ward et al . , 2007 ) , we show in this work that coupling of the TMD transition to the ATPase cycle is not identical between subclasses of exporters . Evidence of more variation in the coupling mechanism has emerged recently from a structure of a human ABC homodimer where the NBSs are disengaged even in the presence of AMP-PNP ( Shintre et al . , 2013 ) . Finally , if indeed TMD conformational changes require hydrolysis of a single ATP molecule in heterodimers , it raises the conundrum of why two molecules are hydrolyzed in ABC homodimers to achieve what would appear to be similar type of mechanical work . Addressing these questions is the next frontier in understanding the mechanism of ABC exporters .
A homology model of the inward-facing BmrCD was constructed with MODELLER ( Marti-Renom et al . , 2000 ) , using the crystal structure of the asymmetric ABC transporter TM287/288 ( PDB: 3QF4 ) ( Hohl et al . , 2012 ) as the template . In order to correctly map the consensus and degenerate nucleotide-binding sites , TMD287 was chosen as the template for BmrC and TMD288 as a template for BmrD . The alignment between BmrC and TM287 was directly generated by pairwise Blastp ( Figure 1—figure supplement 1A ) . However , due to low sequence identity , the homology of the first ∼150 amino acids of BmrD could not be determined through direct pairwise alignment . The BmrD/TM288 alignment ( Figure 1—figure supplement 1B ) was instead obtained by manually optimizing a multiple alignment involving sequences of BmrC , BmrD , TM287 , TM288 , and SAV1866 ( Dawson and Locher , 2006; Figure 1—figure supplement 1C ) , which was generated using Clustal-Ω ( Sievers et al . , 2011 ) . In the multiple alignment , the N-terminal region of BmrD ( Met1-Leu48 ) is aligned to the elbow helix and the TM1 of other ABC transporters , followed by a heterologous insertion of ∼100 residues . Assuming the aligned sequence within a transmembrane helix is ungapped , the BmrD alignment near the heterologous insertion was then manually optimized by ungapped extension from the aligned sequence immediately C-terminal to the inserted region , which covers most of the TM2 of TM288 . Similarly , the alignment of BmrD/TM288 near the TM5 ( Asn333-Gly365 of BmrD ) was also manually optimized due to its lower local sequence identity and the presence of inserted residues in the transmembrane region of the BmrD sequence . The inserted region of BmrD ( Cys49–Ile146 ) was not modeled due to the lack of homologous sequence in any protein structure . The final sequence alignment used for homology modeling is shown in ( Figure 1—figure supplement 1B ) . To maintain the structural stability and the global conformation during the refinement of the initial homology model , Cα atoms in ungapped aligned regions ( Tyr15-Asp46 , Leu57-Gly84 , Leu93-Asp124 , Leu137-Leu316 , Gly335-Gln573 of BmrC; and Met1-Ile47 , Phe162-Arg328 , Ser368-Glu411 , Arg429-Ala673 of BmrD ) were excluded from the refinement , that is , their positions remain fixed in the final model . Cysteine-less BmrCD ( BmrCD-WT* ) was constructed from wild type BmrCD ( BmrCD-WT ) in pET21b ( + ) ( a kind gift of Dr JM Jault ) by substitution of three native cysteines of BmrD by alanines using QuikChange site directed mutagenesis ( Stratagene , La Jolla , USA ) . BmrCD-WT* template was then used to make double cysteine mutants . All substitutions were confirmed by DNA sequencing . BmrCD-WT , BmrCD-WT* and cysteine mutant plasmids were transformed into E . coli BL21 ( DE3 ) cells . A single transformant colony was inoculated into 20 ml LB for the primary culture which subsequently was used to start the main culture in 1 l minimal media supplemented with glycerol ( 0 . 5% ) , thiamin ( 2 . 5 μg/ml ) , ampicillin ( 100 μg/ml ) , MgSO4 ( 1 mM ) , and 50 × MEM amino acids ( 1 ml ) . Cultures were grown at 37°C with shaking to an OD600 of 1 . 2 , and then expression of BmrCD was induced by addition of 0 . 7 mM isopropyl β-D-1-thiogalactopyranoside . BmrCD cultures were incubated at 25°C with shaking for another 5 . 5 hr . The cells were harvested by centrifugation and stored at −80°C . E . coli cell pellets were resuspended in 20 ml of lysis buffer ( 50 mM Tris–HCl , 5 mM MgCl2 , pH 8 . 0 ) , including 10 mM DTT , 10 μg/ml DNAase , 0 . 1 mM PMSF , 1/3 of a Complete EDTA-free protease inhibitor cocktail tablet ( Roche , Indianapolis , USA ) and were lysed by five passes through an Avestin C3 homogenizer at 15-20 , 000 PSI . The lysate was centrifuged at 9000×g for 10 min to remove cell debris and the membranes were isolated by ultracentrifugation at 200 , 000×g for 1 hr . Membranes were solubilized in resuspension buffer ( 50 mM Tris–HCl , 100 mM NaCl , 15% [vol/vol] glycerol , pH 8 . 0 ) including 1 mM DTT , 1% wt/vol n-dodecyl-β-D-maltopyranoside ( β-DDM ) with constant stirring on ice for 1 hr . Solubilized membranes were then centrifuged at 200 , 000×g for 45 min to 1 hr to remove insoluble particulates . The solubilized fraction was then incubated for 1 hr with 300 µl of pre-washed Ni-NTA resin ( QIAGEN , Venlo , Limburg ) pre-equilibrated with Ni buffer ( 50 mM Tris–HCl , 100 mM NaCl , 15% [vol/vol] glycerol , 0 . 05% β-DDM , pH 8 . 0 ) . BmrCD bound Ni-NTA resin was loaded onto a column , washed with five column volumes of Ni buffer containing 20 mM imidazole and was eluted with 250 mM imidazole . Eluted BmrCD was incubated with a 20-fold excess of ( 1-Oxyl-2 , 2 , 5 , 5-tetramethylpyrroline-3-methyl ) methanethiosulfonate ( MTSSL , Enzo Life Sciences , Farmingdale , USA ) for 4 hr at 23°C and placed on ice for ∼12 hr . The labeled protein was then separated from free label by size-exclusion chromatography on a Superdex 200 column in buffer containing 50 mM Tris–HCl , 150 mM NaCl , 10% ( vol/vol ) glycerol , 0 . 02% β-DDM , pH 7 . 4 . BmrCD concentration was determined by absorbance at 280 nm ( Mean extinction coefficient = 68 , 077 . 5 M−1 cm−1 ) . Cysteine-less MsbA ( MsbA-WT* ) template ( Dong et al . , 2005 ) was used to generate single and double-cysteine substitutions . MsbA-WT* and cysteine substituted mutants were expressed , purified and labeled as previously described ( Smriti et al . , 2009; Zou and Mchaourab , 2009 ) . Membrane scaffold protein ( MSP1D1E3 ) is expressed and purified as described earlier ( Boldog et al . , 2007 ) with the following modifications . Briefly , MSP1D1E3 gene in pET-28a ( + ) was obtained from Genscript ( Piscataway , USA ) and transformed in E . coli BL21 ( DE3 ) cells . A dense starter culture was used to inoculate secondary culture of 500 ml Terrific broth supplemented with 30 μg/ml of kanamycin . Cultures were grown at 37°C with shaking to an OD600 of ∼2 . 2–2 . 5 , and then expression of MSP1D1E3 was induced by addition of 1 mM IPTG . Cultures were further grown for 4 hr at 37°C , and cells were harvested by centrifugation . Cell pellets were resuspended in 15 ml of lysis buffer ( 20 mM sodium phosphate , 1% Triton X-100 , pH 7 . 4 ) , including 1 mM PMSF , 1/3 of a Complete EDTA-free protease inhibitor cocktail tablet ( Roche ) and were lysed by sonication . The lysate was centrifuged at 30 , 000×g for 30 min , and the supernatant was loaded onto a Ni-NTA column equilibrated with lysis buffer , followed by washing with four bed volumes of wash buffer-1 ( 40 mM Tris/HCl , 0 . 3 M NaCl , 1% Triton X-100 , pH 8 . 0 ) , four bed volumes of wash buffer-2 ( wash buffer-1 with 50 mM sodium cholate ) , four bed volumes of buffer A ( 40 mM Tris/HCl , 0 . 3 M NaCl , pH 8 . 0 ) , four bed volumes of buffer A containing 20 mM imidazole , and eluted with buffer A containing 300 mM imidazole . The eluted MSP1D1E3 was passed over a desalting column into MSP buffer ( 50 mM Tris , 0 . 1M NaCl , 0 . 5 mM EDTA , pH 7 . 5 ) and the concentration was determined by absorbance at 280 nm ( extinction coefficient = 29 , 910 M−1 cm−1 ) . PC ( L-α phosphatidylcholine ) and PA ( L-α phosphatidic acid ) ( Avanti Polar Lipids , Alabaster , USA ) were combined in a 9:1 molar ratio , dissolved in chloroform , evaporated to dryness on a rotary evaporator and desiccated overnight under vacuum . The lipids were hydrated in MSP buffer containing 0 . 5% ( wt/vol ) ß-DDM , filtered through 0 . 2 µm polycarbonate membrane ( Whatman , Florham Park , USA ) and stored in small aliquots at −80°C . For reconstitution into nanodiscs , purified BmrCD* or spin-labeled proteins in ß-DDM micelles were mixed with PC/PA lipid mixture , MSP1D1E3 and ß-DDM in the following molar ratios: lipid:MSP1D1E3 , 120:1; MSP1D1E3:BmrCD , 3:1; ß-DDM:lipid , 5:1 . Mixtures were rocked at room temperature for 30 min . Biobeads ( 0 . 8–1 g/ml ) were then added to the solution and incubated overnight at 4°C with rocking . The nanodiscs assembly solution was filtered using 0 . 45 µm filter to remove biobeads . Full nanodiscs were separated from empty nanodiscs by size-exclusion chromatography . Nanodiscs were concentrated using Amicon Ultra-50K centrifugal filter units ( Millipore , Billerica , USA ) . Nanodiscs having BmrCD* or spin-labeled mutants were then characterized using SDS-PAGE to verify reconstitution and estimate reconstitution efficiency . In another measure , concentration of spin-labeled mutants in nanodiscs was determined as described previously ( Zou and McHaourab , 2010 ) by comparing the intensity of the integrated CW-EPR spectrum to that of the same mutant in detergent micelles . The specific ATPase activity of BmrCD and MsbA was determined as previously described ( Smriti et al . , 2009 ) with the following modifications . Briefly , BmrCD in detergent micelles ( 20 µg ) , in nanodiscs ( 1 µg ) and MsbA in detergent micelles ( 1 µg ) samples were incubated with increasing concentrations of ATP at 30°C ( detergent micelles ) , 37°C ( nanodiscs ) for 30 min and at 37°C for 20 min respectively in presence or absence of Hoechst and vanadate . The reaction was stopped by adding 1% SDS and the color was developed using a 1:1 solution of ammonium molybdate ( 2% in 1M HCl ) and ascorbic acid ( 12% in 1M HCl ) . The absorbance of samples was measured at a wavelength of 850 nm on a BioTek Synergy H4 microplate reader . The amount of phosphate released was determined by comparison to inorganic phosphate standards . For CW-EPR , spin-labeled BmrCD and MsbA samples were loaded in capillaries and spectra were collected on a Bruker EMX spectrometer using 10 mW microwave power level and a modulation amplitude of 1 . 6 G . DEER spectroscopy was performed on a Bruker 580 pulsed EPR spectrometer operating at Q-band frequency ( 33 . 9 GHz ) with a standard four-pulse protocol at 83 K ( Jeschke , 2002 ) . A 30% ( wt/wt ) glycerol was added to samples as a cryoprotectant . Raw DEER decays were analyzed using home-written software operating in the Matlab environment . The software implements a number of well-established approaches to DEER data analysis ( Sen et al . , 2007; Brandon et al . , 2012 ) . The motivation for developing this software was to carry out global analysis ( Beechem , 1992 ) of the DEER decays obtained under different conditions for the same spin label pair . The distance distribution is assumed to consist of a sum of Gaussians , the number of which is determined based on a statistical criterion . The optimal number of Gaussian components and the center , width and amplitude of each component were allowed to vary between conditions . The slope of the background was assumed to be uniform across the data set but was fit rather than set by the user . Care was taken to ensure that the concentration of samples for the same spin label pair were identical within experimental error . The statistical significance between fits with increasing number of Gaussian distributions was determined using an F test . The root-mean squared difference between the data and fit was minimized using the trust-region-reflective algorithm implemented in the MATLAB routine ‘lsqnonlin’ . We compared results from this analysis to those obtained by using the package DeerAnalysis 2011 ( Jeschke et al . , 2006 ) for data obtained in the trapped conditions ( Figures 1–3 ) and found that the resulting distributions are very similar . The main advantage of this approach is the accurate determination of changes in the amplitude of distance components under turnover conditions where a subjective determination of the background in DeerAnalysis can distort these changes . This software is available from Matlab Central ( http://www . mathworks . com/matlabcentral/fileexchange/46729-deera2012-zip ) . | Cells are surrounded by a membrane that acts like a barrier to many molecules . This membrane either stops molecules from entering or exiting the cell , or at least slows their movement . However , it is important that cells can remove some molecules , such as toxins , and that nutrients and certain other molecules can get into cells . As such , cells rely on ‘transporter’ proteins embedded within the membrane to move these molecules through the membrane . Transporters called ‘Multidrug ABC exporters’ are found in almost all living things , and use the energy released by breaking down molecules of adenosine triphosphate ( ATP for short ) to pump toxins out of cells . Although the three-dimensional shapes of many transporters are known , it is not clear how the energy released from ATP molecules allows the transporter to move a toxin from one side of the membrane to the other . Here , Mishra et al . have looked at how the shape of an ABC exporter from a bacterium called Bacillus subtilis changes as it interacts with ATP . Most bacterial ABC exporters are made from two copies of the same protein , but the B . subtilis exporter is made from two slightly different proteins , one of which is less able to bind to and break down ATP . Mishra et al . found that those parts of the two proteins that bind to ATP can adopt a range of different shapes that had not been seen before . Moreover , the parts of the proteins that extend across the cell membrane face into the cell when the ATP binds , and switch to face out of the cell when the ATP is broken down . This movement of the proteins would allow toxic molecules inside the cell to enter the exporter , and then be pushed to the outside of the cell . The findings of Mishra et al . show that not all ABC exporters work by the same mechanism . Future work could extend this new understanding to multidrug ABC transporters from humans , which remove waste and harmful molecules from our cells and have been implicated in resistance to chemotherapy in cancer cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | Conformational dynamics of the nucleotide binding domains and the power stroke of a heterodimeric ABC transporter |
The majority of mammalian promoters are CpG islands; regions of high CG density that require protection from DNA methylation to be functional . Importantly , how sequence architecture mediates this unmethylated state remains unclear . To address this question in a comprehensive manner , we developed a method to interrogate methylation states of hundreds of sequence variants inserted at the same genomic site in mouse embryonic stem cells . Using this assay , we were able to quantify the contribution of various sequence motifs towards the resulting DNA methylation state . Modeling of this comprehensive dataset revealed that CG density alone is a minor determinant of their unmethylated state . Instead , these data argue for a principal role for transcription factor binding sites , a prediction confirmed by testing synthetic mutant libraries . Taken together , these findings establish the hierarchy between the two cis-encoded mechanisms that define the DNA methylation state and thus the transcriptional competence of CpG islands .
Multiple levels of regulation control correct expression level of a gene . In addition to transcription factors ( TF ) , epigenetic signals enable temporal integration of regulatory events through dynamic processes including cell division and organism development . Considerable progress has been made in mapping the occurrence of various epigenetic marks during the course of mammalian development . This has refined our picture of the spatio-temporal occurrence of epigenetic marks , yet our mechanistic understanding on how their deposition is regulated remains limited . In mammals , methylation of DNA occurs mainly at cytosines lying in a CG context and its presence correlates with transcriptionally repressed states ( Deaton and Bird , 2011; Jones , 2012 ) . Unmethylated CGs are concentrated in regions that are unusually rich in CG dinucleotides as compared to the rest of the genome ( Bird , 1980 ) . The extrapolation of this observation led to the concept of CG islands ( CGI ) , as an operational definition of regions that are likely unmethylated based on their sequence composition . Recently generated methylation maps at basepair resolution from several tissues and organisms have experimentally identified unmethylated regions at unprecedented detail ( Hodges et al . , 2011; Molaro et al . , 2011; Stadler et al . , 2011; Xie et al . , 2013; Ziller et al . , 2013 ) . These datasets revealed that many unmethylated regions ( UMRs ) extend far beyond the CGI definition and that CGI can show variable levels of methylation ( Meissner et al . , 2008; Mohn et al . , 2008; Doi et al . , 2009; Hodges et al . , 2011; Molaro et al . , 2011; Stadler et al . , 2011 ) . Moreover , these studies identified CG poor regions outside of CGIs having low methylation levels ( Low Methylated Regions–LMRs ) ( Hodges et al . , 2011; Stadler et al . , 2011; Hon et al . , 2013; Xie et al . , 2013; Ziller et al . , 2013 ) . These findings challenged the notion that a simple sequence definition provides the most accurate prediction for the methylation state of a DNA sequence ( Hodges et al . , 2011; Molaro et al . , 2011; Long et al . , 2013b ) and motivated to derive improved models to predict CGIs that integrate multiple genomic features ( Bock et al . , 2007; Wrzodek et al . , 2012; Zheng et al . , 2013 ) . Several molecular mechanisms have been proposed that contribute to an unmethylated state on the basis of their correlative occurrence with unmethylated regions of the genome ( Deaton and Bird , 2011 ) . From sequence perspective , these include binding of unmethylated CGs by CXXC domain containing proteins , which have been proposed to inhibit or counteract methyltransferase activity directly or through the establishment of a specific chromatin state ( Ooi et al . , 2007; Cedar and Bergman , 2009 ) . This model would predict that local CG content is the sole determinant of the unmethylated state . Recent experiments suggested that classical transcription factors that bind motifs more complex than CG generally promote a hypomethylated states within CG poor region ( Hodges et al . , 2011; Stadler et al . , 2011 ) . Notably transcription factors have been previously implicated to impact activity but also methylation state of CGI ( Brandeis et al . , 1994; Macleod et al . , 1994; Dickson et al . , 2010; Lienert et al . , 2011 ) . Yet , it is inherently difficult to dissociate the effects of CGs and TF binding sites since both coincide at UMRs . Transgenic studies have argued that both local CG concentration and binding of transcription factors ( TF ) have a role in promoting low methylation levels ( Brandeis et al . , 1994; Macleod et al . , 1994; Dickson et al . , 2010; Lienert et al . , 2011 ) . The cumbersome nature of targeted genetics in mammals limited the scale of these experiments preventing generalization of the observed effects as well their translation into predictive models . Nevertheless , these results demonstrate that genetic information is necessary and sufficient to instruct methylation states , opening the possibility to study the regulation of methylation states through genetic perturbation . However this requires a large number of sequence variations in order to be comprehensive , which is a general prerequisite and experimental bottleneck in the analysis of DNA sequence contribution to regulation of biological processes . In order to be informative , such experiments have to be performed in a controlled environment to minimize context related interference . Recently such approaches have been successfully developed to dissect the organization of cis-regulatory elements by generating large pools of sequence variants and measuring their effect on transcription using transient reporter gene assays ( Melnikov et al . , 2012; Patwardhan et al . , 2012; Sharon et al . , 2012; Arnold et al . , 2013 ) . Such transient assays however are not suitable to study chromatin regulation , which requires stable genomic integration of the sequences of interest at the same chromosomal locus to account for influences of copy number and local chromatin environment . To move beyond these limits , we developed an assay that allows parallel insertion of thousands of DNA fragments in a defined locus in murine embryonic stem cells ( ESC ) . We isolated and synthesized various collections of DNA sequences in order to separately test the quantitative effect of sequence features proposed to influence methylation states . Parallel profiling of the methylation status of this catalogue of sequences allowed us to derive a comprehensive dataset that permits the systematic association between sequence motifs and methylation states . Using this information , we derive quantitative models describing the sequence determinants that govern the establishment of DNA methylation states . This surprisingly reveals that transcription factor binding sites are essential to explain the unmethylated state of the majority of CpG islands . We further demonstrate the utility of such datasets by explaining methylation changes observed in the course of normal differentiation . Moreover , we observe that deviation from the derived models is a characteristic hallmark of methylation changes associated with cancer states suggesting that this phenomenon is mechanistically distinct from differentiation related methylation changes .
In order to combine genomic targeting and high-throughput measurements , we developed a method that allows parallel insertion of hundreds of DNA fragments in a defined locus in murine ESCs ( Figure 1A ) . This approach entails the generation of a plasmid library of sequences of interest , which can be generated by selection or synthesis . The inserted sequences are flanked by lox sites in inverted orientation to subsequently enable cre mediated targeting ( Feng et al . , 1999 ) . The library is transfected as a pool into ES cells together with an expression plasmid for the cre recombinase . Clones that underwent targeted exchange are selected solely based on loss of a negative selection marker at the target site . The resulting targeted cells can be analyzed through the use of universal primers flanking the fragments and subsequent sequencing . This approach yielded reproducibly up to several thousand integrants using a standardized transfection protocol ( Figure 1—figure supplement 1 ) . To our knowledge this generated unprecedented sequence diversity at a single genomic site in a higher eukaryote . 10 . 7554/eLife . 04094 . 003Figure 1 . High throughput genome engineering methodology . ( A ) A pool of diverse DNA fragments is ligated into a plasmid that contains a set of inverted lox-P sites ( triangles ) and universal priming sequences ( pink boxes ) flanking the cloning site . After transformation in E . coli the library composition is determined by paired-end sequencing of the fragment boundaries . The plasmid library is inserted at the β-globin locus by Recombination Mediated Cassette Exchange ( RMCE ) . Methylation of the fragments is determined by high throughput sequencing of the bisulphite PCR product produced using the universal primer sites . ( B ) Comparative distribution of methylation and densities of CG dinucleotides in the mouse genome . CGs of the mouse genome were classified based on local CG density and the average of their methylation status in mESC was plotted ( blue line ) . The proportion of single CGs within UCSC CGIs having a certain CG density are plotted as filled red line , revealing the spread of densities observed in umethylated islands . The average CG density of islands is also plotted ( dashed red line ) , revealing the heterogeneity between islands . ( C ) Single locus example of unmethylated CG rich regions with heterogeneous CG density . ( D ) Application of the genome engineering methodology to test the sequence contribution to methylation states . Sequences of CG rich regions are fragmented into smaller entities and the ability of theses sub-fragments to acquire methylation is assayed . DOI: http://dx . doi . org/10 . 7554/eLife . 04094 . 00310 . 7554/eLife . 04094 . 004Figure 1—figure supplement 1 . Evaluation of the efficiency of the developed method . ( A ) Evaluation of the library insertion efficiency for four of the tested libraries shows up to 1 , 800 insertions per experiment . Compared are the numbers of unique fragments in the initially cloned pool as determined by sequencing of the native PCR performed on the library containing plasmid and the number of unique fragments inserted in the mESC genome as detected by sequencing of the native PCR performed using the universal primers flanking the fragments . The number of uniquely detected fragments depends on the initial library complexity . ( B ) Evaluation of the proportion of inserted fragments recovered during the methylation profiling as measured by sequencing of the PCR performed on the bisulfite converted gDNA . This reveals that around 30% of the initially inserted fragments are efficiently covered suggesting that bisulphite PCR is the limiting step of the experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 04094 . 004 CG rich unmethylated regions differ in their average CG content ( Figure 1B ) . Moreover CG content varies within individual regions that contain short stretches of differential CG density ( Figure 1C ) . In order to ask how this local heterogeneity translates into an unmethylated state and if different mechanisms function in different parts of islands , we inserted fragments of CGIs to define their individual potency to regulate methylation and to potentially identify DNA sequence determinants of this regulation ( Figure 1D ) . To create libraries allowing the exploration of CG rich regions , we took advantage of our previously described murine ESC methylome ( Stadler et al . , 2011 ) . We performed an in silico digestion of the mouse genome using all available methylation sensitive restriction enzymes and chose a set of enzymes and defined size range of fragments to maximize the enrichment for fragments from CG rich unmethylated regions . The library was cloned and amplified in Escherichia coli and its composition determined by paired-end sequencing . Subsequently the libraries were targeted into a defined genomic site in murine stem cells as described above . DNA extracted from the pool of cells after selection served as a template for PCR after bisulfite conversion using universal primer binding sites that flank all inserts . The resulting PCR products were analyzed by next generation sequencing resulting in high coverage methylation measurements for ∼30% of the initially inserted fragments ( Figure 1—figure supplement 1 ) . This translates into 100–600 fragments per transfection depending on initial library complexity . We initially inserted three independent libraries enriched for sequences from unmethylated CGIs . This resulted in high-resolution methylation measurements for ∼400 fragments with variable length ( 100–400 bp ) . The acquisition of methylation of fragments between different insertions was reproducible for most fragments between independent library insertions ( R > 0 . 6 ) ( Figure 2—figure supplement 1 ) . Next we compared the methylation state of fragments after insertion with that of the matching endogenous sequence ( Stadler et al . , 2011 ) ( Figure 2—figure supplement 1 ) . This analysis reveals that the majority ( 63 . 5% ) of tested sequences are methylated similarly to their endogenous counterparts despite the fact that they only represent short sub-fragments of CGIs . Interestingly , the length of individual fragments did not appear to be critical for this autonomy . For example , elements as small as 100 bp were found to be sufficient to establish an unmethylated state . However 40% of the inserted fragments gained methylation relative to their endogenous site . Surprisingly , this gain of methylation does not result in a completely hypermethylated state but covers a broad range in frequency from 20–100% despite the fact that these fragments originate from regions devoid of methylation . Thus , dividing unmethylated regions into shorter entities transforms a methylation frequency that is mostly binary in the genome to a continuous variable . Having derived this observation from a large pool of sequences opens the possibility to infer quantitative relationships between sequence composition and resulting methylation . We initially assayed how much dinucleotide frequency could explain the observed differential methylation patterns ( Figure 2—figure supplement 2 ) . In fact the frequency of CG explains only 14% of the observed variation ( R = −0 . 37 ) , suggesting that differences in CG density between fragments contribute to a certain extend to their differential methylation . A comparison of the average ectopic methylation for all fragments relative to their CG density reveals that at the upper end of CG densities ( ≥12CGs/100 bp ) the inserted sequences tend to behave as the cognate endogenous sequence since they show little to no methylation ( Figure 2D ) . Importantly however , at CG densities that are more representative for islands ( Figure 2D , upper panel ) , the methylation of endogenous and inserted fragments starts to significantly deviate . While some stay unmethylated , others gain methylation as indicated by the spread of methylation levels . This trend of increased methylation over a wide range becomes stronger with reduced CG density ( Figure 2D ) . Thus at very high frequencies , CGs appear sufficient to explain an unmethylated state of small fragments after insertion , while at lower densities numerous fragments of similar CG density show highly variable methylation . 10 . 7554/eLife . 04094 . 005Figure 2 . Systematic determination of the autonomy of DNA sequences to acquire DNA methylation patterns . ( A ) Comparison of methylation levels of inserted fragments with their methylation at endogenous locus ( n = 394; grey transparent dots ) . Histograms depict the proportion of fragments in each area of the plot illustrating the prevalence of unmethylated regions within the library . A majority of these fragments maintain their state when inserted . Orange arrows indicate fragments displayed as single locus examples . ( B–C ) Examples of regions that loose or maintain their unmethyated status when inserted at the ectopic site . Single CG methylation levels for the same sequence are compared between endogenous ( blue dots ) and ectopic ( red dots ) context . Vertical lines show the boundaries of each fragment . Black box indicates UCSC CpG island definition . Black vertical bars depict the positions of CGs . ( D ) Comparison of methylation levels of DNA fragments between endogenous and ectopic context plotted against CG content . Center panel: data were binned according to the CG density of fragments and the distribution of endogenous ( blue ) and ectopic ( grey ) methylation within each bin is depicted in boxplots . Upper panel: Comparative distribution of the CG density in a 300 bp surrounding all CGs within ( red ) and outside ( black ) CpG islands throughout the genome illustrating that the vast majority of tested fragments have a CG density within the range observed at CGIs . Lower panel: statistical significance of the differences between endogenous and ectopic methylation for each CG density bin . p-values derived from a t test are displayed using the indicated color code for each bin . DOI: http://dx . doi . org/10 . 7554/eLife . 04094 . 00510 . 7554/eLife . 04094 . 006Figure 2—figure supplement 1 . Evaluation of the reproducibility of the developed method . ( A ) Estimation of the measurement error in methylation levels determination for the Mouse and the E . coli samples . Technical replicates represent the same library insertion profiled twice using the same starting gDNA as a template for PCR . ( B ) Estimation of the biological variation in methylation acquisition for the mouse and E . coli based libraries . Biological replicates represent separate library insertion events . Pearson correlation coefficients R are depicted . ( C ) Validation of single fragments methylation levels by standard bisulphite sequencing . Black dots represent the single CG endogenous methylation levels as measured by shotgun bisulphite sequencing ( Stadler et al , 2012 ) . Blue boxes represent the endogenous location of the inserted fragment . Red dots represent the ectopic methylation levels of the inserted fragments as measured by next generation sequencing ( upper track ) or Sanger sequencing ( lower track ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04094 . 00610 . 7554/eLife . 04094 . 007Figure 2—figure supplement 2 . Systematic analysis of the relationship between all dinucleotide frequency and methylation of the mouse fragments . Shown are the pearson correlations ( R – upper panel ) and the 1/p-value ( upper panel ) of individual linear regression between fragment methylation levels and each dinucleotide . DOI: http://dx . doi . org/10 . 7554/eLife . 04094 . 007 CGIs represent functional regulatory regions that are under selection for the presence of transcription factor binding sites ( Deaton and Bird , 2011; Jones , 2012 ) . In order to ask how the presence or absence of such motifs affects the methylation after insertion , we next introduced libraries of sequences with different CG densities derived from the E . coli genome . This prokaryotic DNA is not under selection for binding sites for mammalian transcription factors , allowing us to measure the effect of CG density in an isolated context . The combination of two library designs , allowed us to measure a total of 183 prokaryotic sequences covering a broad range of CG densities . We compared their acquired methylation once inserted in the mouse genome in relation to their CG density ( Figure 3A ) . Similar to the inserted mouse fragments the unmethylated state can be observed at the highest CG densities ( ≥12CG/100 bp ) consistent with previous observations that CG density can also protect prokaryotic sequences from de novo methylation ( Lienert et al . , 2011 ) . At a lower CG density however prokaryotic sequences get readily methylated , as revealed by higher average methylation and a significantly reduced spread compared to mouse fragments with similar CG content ( Figure 3A ) . Moreover , we observe a much stronger negative correlation between CG density and methylation for these fragments than the mouse fragments ( Figure 3—figure supplement 1 , R = 0 . 65 ) . The reduced spread and the high number of fragments that cover a broad range of CG densities and methylation states enable us to model the relationship between CG frequency and methylation quantitatively in a sequence context where the influence of TF sequence motifs is limited ( Figure 3—figure supplement 2 ) . We fitted a standard sigmoidal model , which accounts for the finite asymptotes inherent to methylation data ( Figure 3B , see methods for complete description ) . The resulting graph properly describes the data ( R2 = 0 . 51 ) . 10 . 7554/eLife . 04094 . 008Figure 3 . Quantification and modeling of the influence of CG content on methylation levels . ( A ) Comparison of methylation level of inserted DNA fragments derived from mouse ( grey ) or prokaryotic ( E . coli-purple ) genomes plotted against their CG content . Data were binned according to the CG density of the fragments and the distribution of methylation within each bin is depicted in boxplots . Lower panel: statistical significance of the differences between E . coli and mouse methylation for each CG density bin . p-values derived from a t test are displayed using the indicated color code for each bin . ( B ) Average DNA methylation levels acquired by E . coli DNA fragments relative to their CG density ( n = 183 ) . The methylation state of E . coli derived fragments is anti-correlated to its CG density . The dashed red line represents the sigmoidal model fitted to the data ( Coefficient of determination of the sigmoidal fit is displayed R2 = −0 . 51 ) . ( C ) Comparison of the CG density based prediction and the observed methylation levels for the mouse fragments ( Pearson correlation R = 0 . 38 ) . The red color scale depicts the DHS signal within the fragments in their endogenous context . The two boxes show subsets used for selecting fragments to be mutated . ( D ) Boxplot comparing the endogenous DHS signal ( log2 of the counts ) for fragments predicted to be methylated by the CG based model and observed either unmethylated ( left ) or methylated ( right ) when inserted ectopically . The difference observed between the two groups is significant as indicated by the p-value derived from a t test . ( E ) Comparison of the methylation of mouse fragments to their mutated versions in which all non-CG positions were substituted by E . coli sequence . Boxplots for each predicted category were plotted separately for WT ( grey ) and mutated sequences ( purple ) . Dots representing the methylation values for individual fragments were overlaid . Pearson correlation between the predicted value and the E . coli transformed is R = 0 . 64 . p-value derived from a t test as a measure of statistical significance of the observed differences is displayed . ( F ) Evaluation of the effect of protein binding to methylation by insertion of a REST perfect motif ( red dots ) , or the lowest score randomization motif ( black dots ) in the middle of one of the CG rich E . coli fragment previously found to be fully methylated . DOI: http://dx . doi . org/10 . 7554/eLife . 04094 . 00810 . 7554/eLife . 04094 . 009Figure 3—figure supplement 1 . Systematic analysis of the relationship between all dinucleotide frequency and methylation of the E . coli fragments . Shown are the pearson correlations ( R – upper panel ) and the 1/p-value ( upper panel ) of individual linear regression between fragment methylation levels and each dinucleotide . DOI: http://dx . doi . org/10 . 7554/eLife . 04094 . 00910 . 7554/eLife . 04094 . 010Figure 3—figure supplement 2 . Estimation of the influence of transcription factor motifs within the tested sequences . ( A–B ) Estimation of the abundance of transcription factor binding motifs in mouse ( grey line ) and E . coli ( purple lines ) sequences used in the comparisons . Depicted are the mean abundance per CG density bin of the total counts of motifs for the fragment of each origin . The motifs counts were calculated separately for all motifs ( A ) and for GC rich motifs only ( B ) . ( C ) Comparison of the distance between predicted and observed values and the presence of DNase Hypersensitivity within fragments . The difference between the observed methylation levels and the CG density based prediction were calculated and used to bin fragments . The median of DHS signal for each bin was computed and plotted as a dotted line showing that fragments that negatively deviate from the prediction have DHS signal that could explain their unmethylated state . DOI: http://dx . doi . org/10 . 7554/eLife . 04094 . 010 We refer to this as the CG only model derived from prokaryotic sequences and use it to predict the methylation state of mouse fragments ( Figure 3C ) and subsequently contrast it with the actual measurement . By doing so we hope to subtract the effect of CG density on mouse fragments and furthermore isolate those that contain additional sequence cues that influence methylation . This comparison reveals that the ‘CG only’ model indeed has predictive power ( Figure 3C , R = 0 . 38 ) , however numerous mouse fragments are significantly less methylated than predicted based on their CG content , arguing that additional sequence cues contribute to their unmethylated state . Next we wanted to ask if transcription factor binding could explain differential behavior of fragments with similar CG density . As an indirect indicator of transcription factor binding , we annotated the presence of DNaseI hypersensitive ( DHS ) sites in mouse ESCs at the endogenous loci from which the inserted fragments were derived ( Supplementary file 1 ) . Despite being predicted to be methylated based on their CG content , those fragments that are unmethylated indeed show significantly higher DHS enrichment ( Figure 3C , D , Figure 3—figure supplement 2 ) . In turn , this is compatible with the notion that TF binding motifs other than CG can contribute to the observed deviation from the CG only model . To test this hypothesis we synthesized mutated sequences of mouse fragments that were identical in CG position and density but where all non-CG nucleotides are replaced by E . coli sequence in order to alter putative TF motifs . This assay was performed for fragments that we expect to have strong ( Figures 3C–1 ) as well as weak ( Figures 3C–2 ) protection based on their CG concentration . This reveals that those fragments for which we predict a minor role for CG density indeed display a strong shift in their methylation states while the ones where we predict CG density to have a major role only shift slightly . Thus the amplitude of methylation gain upon removal of TF binding sites is related to the CG density of the fragment , resembling the prediction from the CG only model ( Figure 3E , R = 0 . 64 ) . We conclude that in the absence of complex sequence motifs , methylation tends to approach the prediction of the ‘CG only’ model . Next we asked whether insertion of TF binding motifs leads to reduced methylation in a CG rich context . We inserted both the perfect and the lowest score motif ( with identical CG composition , see ‘Materials and methods’ for details ) for the well-studied TF REST in an E . coli fragment that we previously observed to be fully methylated ( from Figure 3B ) . Insertion of the high score motif leads to a loss of methylation while the low score motif has no effect ( Figure 3F ) . Thus as previously shown for CG poor regions ( Stadler et al . , 2011 ) , protein binding at regions with high CG densities contributes to a spatially constrained reduction of the methylation acquired by the inserted fragment . Taken together , these results suggest that mouse fragments derived from CG rich regions are maintained unmethylated by the combined action of two different mechanisms . This argues that while CG rich regions appear homogeneously unmethylated in the genome , they are maintained at this state by cumulative effects of CG density and sequence specific protein binding . After assaying separately the contribution of CG dinucleotide frequency and the effect of TF binding , we tested the relative ability of each parameter to explain methylation patterns genome-wide . First we applied the ‘CG only’ model to predict methylation for all CGs in the genome ( Figure 4A ) . Consistent with the results obtained with the inserted mouse fragments ( Figure 3C ) , we found that considering only CG densities in the prediction overestimates the methylation levels for a large fraction of the CGs in the genome ( Figure 4A , R2 = 0 . 47 , Figure 4—figure supplement 1 ) . This overestimation occurs not only at CG poor low methylated regions , for which a function of transcription factors in reducing local methylation has been shown but also at CG rich UMRs . Indeed >50% of CGs within UMRs show lower methylation than predicted by their CG density ( Figure 4D ) . This confirms our observations made with individual fragments and further argues for the contribution of sequences motifs other than the CG dinucleotide . 10 . 7554/eLife . 04094 . 011Figure 4 . Genome wide modeling of the methylation levels combining CG density prediction and DNase hypersensitivity data . ( A ) Comparison of CG density based prediction and observed methylation levels in mouse ESC throughout the genome . Methylation is predicted using the model derived from the E . coli fragments . CG density is calculated in a 300 bp window around each CG in the mouse genome . The predicted value is compared to measured methylation at the single CG level in mESC . The reference line is shown in black . ( B ) Similar comparison of measured methylation at the single CG level genome wide in mESC but using a prediction model combining CG density and DHS . ( C ) Barplot comparing the explained variance by the CG only based model ( CG ) and the model using DHS data either alone ( DHS ) or in combination with CG density prediction ( CG + DHS ) . ( D ) Proportion of CGs predicted accurately for each model relative to their genomic context . The prediction of each model was compared to methylation as measured by bisulfite sequencing and prediction accuracy was quantified ( with a precision of 20% methylation ) . The barplot illustrates the improvement gained by each variable used in the modeling . It shows that the combination of CG density and DHS is particularly important to accurately predict methylation at CG rich regions . DOI: http://dx . doi . org/10 . 7554/eLife . 04094 . 01110 . 7554/eLife . 04094 . 012Figure 4—figure supplement 1 . Performance evaluation of the derived models . ( A ) Pearson correlation ( R ) between the complete model prediction and the measured methylation in human stem cells and four in vitro derived cell types ( Mch-Mesenchymal; Ms-Mesoderm; NP-Neuronal Progenitor; T-Trophoblastic ) . ( B ) Quantification of the proportion of CGs predicted accurately for each model depending on their genomic context . The prediction of each model was compared to methylation as measured by bisulfite sequencing and prediction accuracy was quantified ( with a precision of 20% methylation ) . The barplot illustrates the improvement gained by each variable used in the modeling . It shows that the combination of CG density and DHS is particularly important to accurately predict methylation at CG rich regions . ( C–E ) Influence of the size of collecting window for quantifying DHS signal on modeling performance . The coefficient of determination R2 for the DHS only ( DHS ) or the combined model ( CG+DHS ) was calculated as a function of the collection window in the DHS dataset . The analysis was performed for all CGs in the genome or for CGs within particular genomic regions . This shows that 300bp is the optimal collection window for DHS regardless of the type of region considered . ( F ) Comparison of the prediction by the DHS only ( DHS ) and the combined model ( CG + DHS ) within CG rich unmethylated regions ( UMR ) . For a significant fraction of the CGs the predicted methylation is lower in for the complete model . ( G ) Comparison of the prediction accuracy for the two models . The delta between predicted and measured value is calculated for both model and plotted against each other . This reveals that the DHS only model overestimates DNA methylation for a significant part of the CGs within UMRs that are more accurately predicted by the combined model . DOI: http://dx . doi . org/10 . 7554/eLife . 04094 . 012 Based on our previous finding of a general link between TF binding and hypomethylation of distal regulatory regions ( LMRs ) ( Stadler et al . , 2011 ) and our experiments with mutated mouse fragments described above , we hypothesized that the remaining variance could be caused by the binding of TFs . In order to account for this effect , we attempted to define the quantitative relationship between protein binding and methylation by creating models that integrate the presence of DHS to predict hypomethylated regions . Such model indeed outperforms and complements a CG based model ( Figure 4B , C and Figure 4—figure supplement 1 ) . It improves predictions for CGs residing in LMRs as well as UMRs ( Figure 4D ) , which reinforces the idea that DNA binding factors significantly contribute to the existence of unmethylated states at CG rich regions . Combining CG and DHS into one model readily explains over two thirds ( 68% genome wide and 60% at CG rich regions ) of the methylation variations observed in mESCs suggesting that CG density and TF binding are the major determinants of genomic methylation . Similar modeling in different human methylomes ( Xie et al . , 2013 ) revealed that these regulatory principles apply similarly in other somatic cell types and organism ( Figure 4—figure supplement 1 ) . Thus information derived from inserting a large number of short sequences enables to model the methylation variation observed throughout mammalian genomes . It suggests that CG density and binding of proteins to more complex transcription factor motifs explain most regions with reduced methylation in the genome . Having established a predictive model consisting of two determinants to explain methylation states in mESCs , we wondered how effects driven by CG density or TF recruitments relate to the dynamics of DNA methylation observed during cellular differentiation . To do so , we compared the methylation prediction from the CG only model with the measured differences in methylation between murine stem cells and neuronal progenitors ( NP ) . To better illustrate the influence of CG density on methylation changes , we contrasted CG rich unmethyated regions ( UMRs ) ( Figure 5A ) , with CG poor low methylated regions ( LMRs ) ( Figure 5B ) . As a whole CG rich regions show little variation between both cell states in line with continuous hypomethylation of CGIs during cellular differentiation . This is in contrast to CG poor regions that experience extensive methylations changes across cell lines ( Figure 5B ) ( Stadler et al . , 2011 ) . However , detailed analysis within CG rich regions ( Figure 5A ) identifies a subgroup of cytosines that show dynamics ( primarily hypermethylation ) in their methylation status during differentiation . Interestingly , cytosines that change their methylation status reside within fragments where a CG only model predicts a methylated state , and little changes are observed within regions where an unmethylated state is predicted ( Figure 5A ) . Notably the observed increased methylation during differentiation approaches the methylation state as predicted by the CG only model ( Figure 5A ) . One likely explanation is that the unmethylated state of these sites depends on binding of TFs that are present in stem cells but not in the neuronal progenitors . Indeed , motifs for stem cell or neuron specific TFs are enriched around differentially methylated CGs ( Figure 5A , Figure 5—figure supplement 2 ) . For example we observe a methylation increase in neuronal progenitors within CG rich regions at binding sites of the stem cell specific pluripotency factor Oct4 ( Pou5f1 ) , while regions bound by factors expressed in both cell types such as REST do not change their status ( Figure 5C , D ) . This effect is reminiscent of the effect of sequence mutations that abolish TF recruitment within individual fragments shown above , resulting in increased methylation that follows the CG-only model ( Figure 3D ) . We conclude that variation in DNA methylation within subparts of islands is a function of TF binding and local CG density . 10 . 7554/eLife . 04094 . 013Figure 5 . CG concentration restricts the amplitude of methylation changes during cellular differentiation . ( A–B ) Methylation gain during differentiation reaches a maximum that can be predicted by the local CG density of the region . CG density based prediction is plotted against the methylation difference between stem cells ( ES ) and neuronal progenitors ( NP ) for CGs located within stem cells ( A ) CG rich unmethylated regions ( UMRs ) or ( B ) low methylated regions ( LMRs ) . Black dots represent the 99th percentile of the changes observed , as a proxy of the maximal amplitude of changes observed at a given CG density . TF motif enrichments were calculated around the most changing CGs and heatmaps depicting the expression changes for the top enriched motifs were plotted . This reveals enrichment for pluripotency factor motifs next to the hypermethylated CGs and of neuronal specific factors in the surroundings of hypomethyayed CGs . ( C ) A gain of methylation is observed at Oct4 binding sites within CG rich regions during differentiation . Composite plot depicting the average DNA methylation around Oct4 ( upper panel ) and REST ( lower panel ) binding sites at CG rich-UMRs . The red doted line represent methylation averages in embryonic stem cells ( ES ) . The blue doted line represent methylation averages in neuronal progenitors ( NP ) , where Oct4 is not expressed , while REST is expressed in both cell types . ( D ) Single locus example of a CG rich region that changes methylation status during differentiation , and which contains a binding site for Oct4 . Lower density track represent Oct4 binding as measured by ChIP-seq ( E ) Methylation changes coincide with changes in DHS but the amplitude of change is limited by CG density . CGs present at UMRs were classified based on the CG density of their surroundings ( x-axis ) and the amplitude of DHS changes from ES to NP ( box plot color correspond to log2 ( delta DHS ) ) . The distribution of methylation changes ( y-axes ) is depicted by a box plot for each category ( methylation difference % ) . The figure illustrates that DHS changes correlate with methylation changes , this correlation is lost at very CG rich stretches . ( F–G ) Cancer related hypermethylation is not restricted by CG density . CG density based prediction is plotted against methylation differences between ( F ) hES cells and normal colon , ( G ) normal colon and cancer colon for CGs located within hES CG rich unmethylated regions . Black dots represent the 99th percentile of the changes observed . DOI: http://dx . doi . org/10 . 7554/eLife . 04094 . 01310 . 7554/eLife . 04094 . 014Figure 5—figure supplement 1 . Characterization of the sub-regions within CpG islands that are independent from transcription factors for being unmethylated . ( A ) Single locus examples of constitutively unmethylated regions ( highlighted by blue boxes ) . Regions were defined as stretches of successive CGs with low predicted score based on their CG density ( <20% ) . Shown are single CG methylation levels in mESC ( red dots ) and the predicted methylation based on CG density only ( dotted line ) . The examples illustrate the diversity of distribution of those regions in the context of CGI definition , illustrating the heterogeneity of methylation regulation within islands . ( B ) Distribution of the constitutively unmethylated regions relative to CpG islands ( CGI from UCSC ) . While the vast majority of the constitutively unmethylated regions are contained within islands , less than 50% of CGI contain such a constitutive CGs . DOI: http://dx . doi . org/10 . 7554/eLife . 04094 . 01410 . 7554/eLife . 04094 . 015Figure 5—figure supplement 2 . Motif enrichment analysis at differentially methylated regions within CpG islands . ( A–B ) Motif MA plots representing the counts for known TF motifs in the surroundings ( 300bp ) of hypomethylated ( A ) or hypermethylated ( B ) CGs versus counts in CGs not changing methylation . For each set , control sequences with similar CG density distribution were used . The name of enriched TF matrices were depicted and matrices for which expression data could be mapped were depicted in red and represented in Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 04094 . 01510 . 7554/eLife . 04094 . 016Figure 5—figure supplement 3 . Hypermethylation at CGs with low predicted methylation is a widespread and specific mark of cancer . Heatmap depicting the 99th percentile of methylation gain in each bin of CG density based predicted methylation for 486 human primary tissues ( 161 normal and 326 cancer samples ) ( Fernandez et al . 2012 ) . The heatmap was subjected to hierarchical clustering , which accurately separates between normal ( white ) and cancer samples ( black and grey ) ( side colorbar ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04094 . 016 To further explore the relationship between differential protein binding and methylation differences within islands , we compared dynamics in methylation and DHS formation using an existing dataset of human ES cells and differentiated neuronal progenitors ( Xie et al . , 2013 ) . We observe that the methylation changes are tightly anti-correlated to DHS changes at CG poor regions ( Figure 5E ) . However , when DHS changes of similar amplitude are detected at higher CG densities , methylation levels do not change . This suggests that methylation levels at these regions are independent from the binding of TFs , and that a CG dependent mechanism is sufficient to explain their methylation . These data are consistent with the model that TF binding contributes to the unmethylated state at CG poorer regions within CGIs , while TF binding is non-essential within highly CG rich regions of islands . We observe that , unlike the TF driven effect , the CG dependent effect is remarkably stable across cell types . One potential explanation is that the responsible trans-acting factors are widely expressed , as it is the case for some of the CXXC domain containing proteins , which are bona fide CG binders ( Long et al . , 2013a ) . A direct consequence of this observation is that within islands , a subset of CGs embedded in short CG dense patches are unmethylated regardless of the binding by TFs in their neighborhood ( Figure 5—figure supplement 1 ) . Importantly however , these patches are not essential to form unmethylated regions , as they are absent in 50% of all CGIs ( Figure 5—figure supplement 1 ) and only contain 30% of all CGs within islands . They further do not show any particular positioning pattern within islands ( Figure 5—figure supplement 1 ) . This suggests that for the majority of CGs ( 67 . 5% ) within islands TF binding could contribute to their unmethylated state . Our data suggest that methylation dynamics observed within islands during differentiation are mostly the result of changes in the binding profiles of transcription factors . Additionally , we observed that the local concentration of CGs restricts the amplitude of these TF driven changes . Aberrant methylation , including hyper-methylation of CGIs is an established hallmark of cancer ( Baylin and Jones , 2011 ) . Since the origin and the regulation of these changes during transformation are largely unknown , we wondered how these changes relate to our model . To this end , we contrasted the methylation gain at UMRs between human stem cells ( hES ) and normal colon cells and between normal colon and related cancer cells ( Berman et al . , 2012 ) . This reveals that normal colon cells show methylation changes that follow our model , while colon cancer cells display a distinct hyper-methylation phenotype ( Figure 5F , G ) . We observe a significant gain in methylation even at CGs that are constitutively unmethylated in normal tissues due to their high CG density ( Figure 5F , G ) . Interestingly , the same is observed when comparing a large set of healthy tissues and cancer types arguing that this is not unique to colon cancers ( Figure 5—figure supplement 3 ) . This argues that cancer hyper-methylation at islands is mechanistically distinct as it cannot be modeled with the same local determinant that describe methylation changes during normal development . Since both CG driven and TF mediated effects are affected this can unlikely be explained by alterations in transcription factor binding patterns .
In this study , we report a high-throughput genome engineering protocol and demonstrate its potential to determine the contribution of DNA sequence to the establishment of epigenetic states . We establish that DNA insertions at a given locus in mammalian genomes through RMCE can be performed with high complexity DNA libraries , opening the possibility to dissect regulatory mechanisms such as the ones that govern the establishment of DNA methylation patterns . In principle this approach can be adapted for any genomic readout ( e . g . transcription or replication ) in order to understand the interplay between multiple regulatory layers that function in cis . It further circumvents the limitations of previously used transient transfections ( Melnikov et al . , 2012; Patwardhan et al . , 2012; Sharon et al . , 2012; Arnold et al . , 2013 ) , which lack a proper chromosomal context and are not controlled for amount of sequence variants per cell and their copy number . We measured the ability of several thousand DNA sequence variants to acquire DNA methylation in mouse embryonic stem cells . The specific design and scale of the tested libraries enabled us to gain insights into the sequence determinants that drive the establishment of unmethylated states at CG rich regions and to build predictive models based on these methylation measurements and the underlying DNA sequence characteristics . One of our key findings is the quantification of the effect of CG density on methylation states . Surprisingly , while we confirm that CG density is sufficient to create an unmethylated state , only every second island harbors sequence stretches above that threshold ( Figure 5—figure supplement 1 ) . Combined , these CG rich stretches only cover 30% of the total number of CGs that reside in islands . Mechanistically , it remains to be determined how CG richness drives unmethylated states , yet several scenarios are compatible with our observations . Proteins that specifically recognize unmethylated CG via a CXXC domain could antagonize de novo DNA methylation as previously proposed ( Long et al . , 2013a ) . Alternatively , presence of dinucleotide concentrations such as CG stretches could alter nucleosomal organization ( Brogaard et al . , 2012; Struhl and Segal , 2013 ) and thereby impact DNA methylation . Since , transcription factors and CG richness can both impact nucleosome positioning ( Struhl and Segal , 2013 ) , it is tempting to speculate that nucleosome depletion could be a unifying principle underlying the formation of unmethylated regions . Our data suggest that a key driver of hypomethylated states is the local binding of transcription factors . Others and we have recently shown that TF binding creates reduced methylation at CG poor regulatory regions such as tissue specific enhancers ( Hodges et al . , 2011; Stadler et al . , 2011 ) , a finding that extends to CG rich regions . Their methylation however is less dynamic during cellular differentiation since most CGIs are constitutively active as promoters of housekeeping genes . Moreover the effect driven by CG density is stable across cell types restricting the amplitude of the changes . Our data argue that the higher the CG content , the lower the TF dependent changes in methylation and vice versa . This scenario explains mechanistically why unmethylated regions frequently extend beyond CGI definitions ( Hodges et al . , 2011; Molaro et al . , 2011; Long et al . , 2013b ) and predict that the methylation changes at borders of CG rich regions ( also referred to as CGI shores ( Doi et al . , 2009 ) ) are a function of TF binding . Importantly however the observed variability in methylation is not restricted to any particular location within CGIs but is only defined by local CG frequency . We believe it is important to account for this property when studying differential methylation patterns . We identified a striking contrast in the type of CGs affected by methylation changes within CG rich regions during normal differentiation vs those that occur during transformation to a cancerous state . Notably it is not individual cytosines that are predictive for cancer , as these vary widely between types , but it is rather the class of CGs affected as defined by our CG density model . While the nature of this predictive difference remains unknown , we note that unlike methylation changes occurring during normal cellular changes , the loss of TF binding is unlikely to explain the observed differences . Hypermethylation in cancer particularly targets CGIs that control genes that are inactive ( Gebhard et al . , 2010; Berman et al . , 2012; Sproul et al . , 2012 ) . Moreover , the study of cancer methylomes revealed that hyper-methylated CGIs were embedded within larger domains of intermediate methylation ( Partially Methylated Domains–PMDs ) ( Hansen et al . , 2011; Berman et al . , 2012 ) . If CGI methylation changes within PMD's are a general phenomenon it might indicate that the observed changes at CGIs in cancer reflect the loss of the local sequence autonomy in determining correct DNA methylation . The observation that TF binding contributes substantially to hypomethylation at CGIs has potential implications for their evolutionary origin as it is compatible with the idea that the emergence of CG rich regions could have occurred indirectly through TF-driven demethylation and resulting reduced C to T transition . This is in line with reports of limited positive selection for the CG content of islands ( Cohen et al . , 2011; Molaro et al . , 2011 ) and that nucleotide composition including CG frequencies vary substantially between unmethylated regions across vertebrates ( Long et al . , 2013b ) . Thus our findings support the notion that CGIs arose as an evolutionary footprint of ancient regulatory regions . Such scenario is still compatible with a subsequent specialization of proteins such as CFP1 or KDM2A to recognize unmethylated CGs by CXXC domains in order to target chromatin processes to regulatory regions ( Blackledge et al . , 2010; Thomson et al . , 2010 ) .
For targeted insertion , DNA libraries were cloned into a plasmid containing a multiple cloning site flanked by priming regions for a pair of universal primers and two inverted L1 Lox sites ( pL1-LPP1-1L ) . For constructing the mouse primary libraries , a set of enzymes was selected on the basis of a screen of all available CG methyl-sensitive enzymes ( NEB - Ipswich , MA ) . An in silico digest of the mouse genome was performed masking all methylated regions in ESCs ( Stadler et al . , 2011 ) , and multiple size selection were tested to optimize the enrichment in CG rich unmethylated regions and the number of unique fragments isolated . Three enzymes ( NarI , BstUI , BssHII ) were selected on the prediction that they would produce a complex library ( >1000 fragments ) of which over 80% of the fragments would overlap with CGIs . 100 μg of mES cells gDNA ( background: ES 129S6/SvEvTac ) was digested and resulting fragmented DNA was size selected ( 100–600 bp ) on a 1% agarose gel . The isolated DNA inserts were directly cloned in the receiving vector ( L1-LPP1-L1 ) . The plasmid pool was transformed and amplified in XL1-competent cells . Library complexity was estimated based on size diversity of colony PCR products prior composition determination by sequencing . For prokaryotic libraries , a similar approach was employed , digesting E . coli DNA ( NC_010473 . 1 ) with MspI and size selecting fragments ( 100–600 bp ) . Additionally , to be able to cover the lower and higher extremes of CG densities in a focused library , a PCR based library was cloned . 96 pairs of primers were in silico designed to target these regions . After PCR amplification , the products were pooled and cloned in the receiving vector . For synthetic libraries , the sequence was designed in silico , custom synthesized , sequence verified ( IDTechnologies , Coralville , IA or GeneArt , Life Technologies , Carlsbad , CA ) and cloned in the receiving vector . The Recombinase-mediated Cassette Exchange ( RMCE ) insertion protocol ( Feng et al . , 1999; Lienert et al . , 2011 ) was refined in order to scale the needs of inserting large number of fragments in parallel . Briefly , TC-1 ES cells were selected under hygromycin ( 250 μg/ml , Roche , Switzerland ) for 10 days . Next , 12 × 106 cells were electroporated ( Amaxa nucleofection , Lonza , Switzerland ) with 75 μg of L1-library-1L plasmid and 45 μg of pIC-Cre . Negative selection with 3 μM Ganciclovir ( Roche , Switzerland ) was started 2 days after transfection and continued for 10 days . Pools of selected cells were tested for successful insertion of DNA libraries by PCR using primers recognizing the universal priming region flanking the insertion site . Direct sequencing of the fragments ends by paired end sequencing was used to determine the sequence composition of the DNA libraries derived by restriction digest of the mouse and prokaryotic genomes . To this end , the library containing plasmids were used as a template for 15 cycles of PCR using the universal set of primers flanking the fragment insertion site . The purified product was then used for standard Illumina library preparation and sequenced on a MiSeq instrument ( Illumina , San Diego , CA ) . Reads were aligned against the corresponding reference genome using Bowtie ( Langmead et al . , 2009 ) and fragments identity was called using read pairing information . A reference set of regions was established where only fragments without overlaps within the library were retained ( to avoid ambiguous read assignments during methylation call of sonicated material ) . A similar procedure than above was used to call genomic insertion rates . PCR was performed with primers annealing to the non bisulfite-converted DNA ( 5′-CCAACCTGACTGTGGTGGACAA-3; 5′-ACATGCACCTTCCCAGGGC-3′ ) . The product was sonicated , gel purified and a sequencing library was prepared . Generation of sequence mutants of the mouse fragments: Fragments were selected based on their high ( >40% ) or low ( <20% ) predicted methylation according to their CG density . For each fragment all non CG positions of the sequence were replaced using a non CG containing stretch of DNA from E . coli as a template . Generation of TF motif insertions: The receiving cassette was derived from an E . coli fragment observed to be methylated in the E . coli library with a CG density typical for CGI . A BamHI-XbaI entry site was in silico inserted in the middle of the fragment and fragment was synthetized . REST position weight matrix was extracted from the JASPAR database ( Portales-Casamar et al . , 2010 ) and was used to derive the best score motif ( most frequent base at each position of the PWM–GACTTTCAGCACCATGGACAGCGCCACTG ) ; and the lowest score motif ( lowest score randomized motif with identical base composition and CG content–CCTCAGGTTGGCACACCTCTAAGAGCCGA ) . These sequences were used to synthetize pairs of oligonucleotides with flanking 5′and 3′sequences to form sticky ends for BamHI , XbaI respectively ( CTAG-5′; 3′-GATC ) . Oligonucleotides were annealed and cloned into the receiving cassette . Genomic DNA ( 2 μg ) of ES cells carrying the libraries was bisulfite converted with the EpiTec Bisulfite Kit ( QIAGEN , Germantown , MD ) . Libraries were amplified by PCR ( AmpliTaq Gold Life Technologies , Carlsbad , CA ) using bisulfite compatible primers ( 5′-AACCTAACTATAATAAACAACC-3′; 5′-GGTATATGTATTTTTTTAGGGT-3′ ) annealing to the universal priming region flanking the fragments cloning site . The PCR product was gel purified and fragmented by sonication ( Covaris S220 , Woburn , MA ) . The sonicated material was used to construct sequencing libraries following Illumina's recommendations . Samples were sequenced as barcoded pools on Illumina GAII or MiSeq instruments . Bismark/Bowtie 0 . 12 . 7 ( Langmead et al . , 2009; Krueger and Andrews , 2011 ) were used to align bisulfite reads against an in silico converted reference genome ( C > T and G > A ) and call methylation state for each CG . Only CGs covered by at least 10 reads were used for analysis . Strain specific SNPs were masked . Methylation was called per CG and fragment averages were derived using the previously established reference set of regions for the library . Only fragments where >50% of the CG and a minimum of four CGs were covered were considered in the analysis . Bowtie 0 . 12 . 7 was used for aligning the non-bisulfite reads from native PCR experiments used to call insertion rates . Fragments were called inserted when >50% of the fragment sequenced was covered by the reads . Data modeling was conducted stepwise by first integrating information from prokaryotic insertions , and then combining it with mouse genomic data . For CG content analysis , only fragments ≥250 bp were considered in order to avoid scoring instability while size normalizing low CG counts . A sigmoidal model was fitted to the prokaryotic data describing the relationship of methylation to CG density at the level of fragments averages . y=1001+e−b ( x−c ) Both higher and lower asymptotes were fixed prior model fitting ( 100% and 0% methylation ) since these are known in the case of DNA methylation data . The best-fit model was then retained ( b = −0 . 337 , c = 6 . 917 ) . Note that a linear model was also tested and performed equally well on the linear part of the data , however the sigmoidal model out performs it to describe both the lower and higher ends of CG densities . Coefficients derived from this model fitted on the prokaryotic data were then used to predict methylation of ( 1 ) the mouse fragments ( 2 ) all CGs genome wide ( considering a 300 bp window around each CG for CG density calculation ) . In the second step of modeling , a linear model was used to combine the prokaryotic based model and transcription factor binding information as measured by DNAse-seq . The model inputs were the single CG mESC methylation levels , prediction of methylation for each CG of the genome based on the prokaryotic model and DNAse-cuts collected in a 300 bp window surrounding the CG . Prior to regression , DNAse-seq data were pre-processed to remove outliers and categorize the data . Fitting of the models were conducted on two chromosomes and performance was assessed on the rest of the genome using R-squared values . For the analysis of methylation dynamics , segments ( UMR , LMR , FMR , PMDs ) were called using MethylSeekR ( Burger et al . , 2013 ) in the different cell types . Then single CG average methylation was compared between the two cell types in each segment type . | Regions of DNA called genes produce the proteins and other molecules that are essential for life . The act of making these molecules is known as gene expression , and being able to switch this process on and off allows cells to adapt to changing conditions . For example , some genes may be turned on in response to injury or may only turn on during waking hours . There are several ways gene expression can be switched on and off . Proteins called transcription factors can bind to DNA and act like a switch that affects nearby genes . Alternatively , special tags called methyl groups can attach to the ‘letters’ that make up the DNA code and turn off gene expression . However , it is not understood how these tags work with transcription factors and other forms of gene regulation . Regions of DNA that boost the expression of a neighboring gene are called promoters . Many promoters in mammals contain repeating patterns of the DNA letters ‘C’ ( which is a chemical called cytosine ) and ‘G’ ( guanine ) , and these regions are tagged less often than other regions of DNA . This led scientists to wonder whether the DNA sequence itself controls where the tags are placed , but existing experimental techniques made it difficult to establish if DNA sequence alone can prevent tagging . Krebs et al . created a technique that allows thousands of different DNA sequences to be inserted into the same part of the genome of mouse stem cells . Comparing the tagging across these different sequences revealed that the CG pattern is not as closely associated with tagging as was thought . If the CG pattern is repeated many times it does seem to prevent tagging , but sequences with fewer repeats also sometimes escape tagging . Krebs et al . found that a sequence was much less likely to be tagged if the nearby DNA also contains a site that transcription factors can bind to . However , regions with a very high number of CG repeats are able to avoid tagging without help from transcription factors . Krebs et al . found that this behavior is not seen in cancer cells . DNA in cancer cells is heavily tagged , even in CG-rich regions , and transcription factors do not appear to play a major role in directing tagging . The new approach developed by Krebs et al . should benefit researchers working to understand the multiple mechanisms that control gene activity . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"genetics",
"and",
"genomics"
] | 2014 | High-throughput engineering of a mammalian genome reveals building principles of methylation states at CG rich regions |
Ribosomal RNA transcription mediated by RNA polymerase I represents the rate-limiting step in ribosome biogenesis . In eukaryotic cells , nutrients and growth factors regulate ribosomal RNA transcription through various key factors coupled to cell growth . We show here in mature adipocytes , ribosomal transcription can be acutely regulated in response to metabolic challenges . This acute response is mediated by PTRF ( polymerase I transcription and release factor , also known as cavin-1 ) , which has previously been shown to play a critical role in caveolae formation . The caveolae–independent rDNA transcriptional role of PTRF not only explains the lipodystrophy phenotype observed in PTRF deficient mice and humans , but also highlights its crucial physiological role in maintaining adipocyte allostasis . Multiple post-translational modifications of PTRF provide mechanistic bases for its regulation . The role of PTRF in ribosomal transcriptional efficiency is likely relevant to many additional physiological situations of cell growth and organismal metabolism .
The ribosome is the engine for protein synthesis and it is vital for cell growth and survival . Ribosome biogenesis , the process of making ribosomes , is a complex and energy intensive process that involves transcriptional regulators , protein modification , and the assembly of multi-component macromolecular complexes ( reviewed in ( Grummt , 2003; Kusnadi et al . , 2015; Russell and Zomerdijk , 2005 ) . Synthesis of ribosomal RNA ( rRNA ) by RNA polymerase I ( Pol I ) is generally considered as a major rate-limiting step in ribosome biogenesis , and it produces the common 47S precursor of mature 5 . 8S , 18S , and 28S rRNAs ( review in [Moss et al . , 2007] ) . The 47S pre-rRNA is transcribed from hundreds of copies of ribosomal DNA ( rDNA ) genes distributed in repeated arrays among 5 acrocentric chromosomes in humans ( Henderson et al . , 1972; Stults et al . , 2008 ) . Given the large numbers of rDNA genes , rRNA transcription can be regulated by both varying the number of transcriptionally active genes and/or by varying the rate of activated transcription . The former can be regulated through chromatin remodeling and related factors , and the latter can be acutely regulated through the factors assembled in Pol-I transcription complex machinery ( reviewed in ( Grummt , 2010 ) . Ribosomal transcriptional activity has been tightly linked to cell growth and proliferation . Any perturbation that changes cell growth or protein synthesis , such as nutrient and growth factor availability , senescence , toxin exposure or viral infection , leads to changes in rDNA transcription activity ( reviewed in ( Kusnadi et al . , 2015 ) . This regulatory process comprises a series of coordinated steps including transcription initiation , promoter escape , elongation and termination ( reviewed in ( Russell and Zomerdijk , 2005; Schneider , 2012 ) . Regulation is achieved via coordinated multiple signaling pathways that modulate the expression and activities of many key factors , such as Pol I-specific transcription factors ( RRN3 , also known as transcription initiation factor 1A , TIF-1A ) , selectivity factor 1 ( SL-1; also known as TIF-1B ) , upstream binding factor ( UBF ) and others ( reviewed in ( Bywater et al . , 2013; Grummt , 2010 ) . Most of these studies were performed in cultured cancer or cancer-like cell lines , where ribosomal transcriptional regulation was coupled to cell proliferation or closely related cell growth . For cell mass growth in mature non-proliferating cells , it’s less clear if and how ribosomal transcription is regulated , and the physiological relevance of ribosomal RNA transcription in these cells has been little studied . Adipocytes are a highly metabolically dynamic cell type that can respond rapidly to alterations in nutrient excess and deprivation , thereby fulfilling its major role in whole body energy homeostasis ( reviewed in ( Rosen and Spiegelman , 2014; Scherer , 2006; Sun et al . , 2011 ) . As a mature non-proliferating cell type , it undergoes dramatic changes upon metabolic challenges . In obesity due to excess calorie loading , adipocytes need to develop not only corresponding cellular structures and functions for the increasing needs in lipid storage and metabolic capacity , but also the machinery for the secretion of adipokines and other proteins . These cells also have to maintain an insulin sensitive functional response in order to avoid the development of type 2 diabetes . Given the importance of homeostatic protein synthesis as a basic cellular function to maintain structure and activity , and to ensure normal cellular physiological functions , it becomes obvious that 'healthy' adipocyte expansion has to be supported by fundamental processes such as protein synthesis , which in turn , can be determined by ribosome biogenesis . Changes in ribosomal RNA synthesis by long term starvation and re-feeding were in fact reported soon after ribosomes were first described ( Benjamin and Gellhorn , 1966 ) although many mechanistic details of ribosome composition and function were unknown at that time . The effect of insulin on protein synthesis and ribosome biogenesis in adipocytes was also reported ( Hansson and Ingelman-Sundberg , 1985; Vydelingum et al . , 1983 ) . A precisely controlled ribosomal DNA transcriptional response to changes in nutrient and insulin levels would therefore seem essential for healthy adipocytes . We and others have shown that PTRF ( polymerase I transcription and release factor , also known as Cavin-1 , herein after , PTRF ) , plays a critical role in caveolae formation ( Hill et al . , 2008; Liu et al . , 2008; Liu and Pilch , 2008 ) , structures that are particularly abundant in adipocytes . Moreover , a lipodystrophic phenotype was observed in PTRF null mice ( Ding et al . , 2014; Liu et al . , 2008 ) that is similar or identical to that of human patients with inactivating PTRF mutations who also display a type of muscular dystrophy ( Ardissone et al . , 2013; Dwianingsih et al . , 2010; Hayashi et al . , 2009; Jelani et al . , 2015; Shastry et al . , 2010 ) . The molecular mechanisms underlying these phenotypes that have been proposed , principally alterations in lipid metabolism/transport and perturbations of the cell surface membrane ( Parton and del Pozo , 2013; Pilch and Liu , 2011 ) cannot fully explain both the adipose and muscular dystrophy phenotypes . In fact PTRF/Cavin-1 as PTRF was first characterized by its Pol-I related regulatory function ( Jansa et al . , 1998 , 2001; Jansa and Grummt , 1999 ) . These in vitro studies established a role for PTRF in the efficiency of rRNA transcription ( Jansa et al . , 1998 , 2001; Jansa and Grummt , 1999 ) , but since then no further experiments concerning this function have been performed that we are aware of . Moreover , the physiological relevance of this activity was never established in cells or in vivo . Consequently , we used primary mouse and cultured adipocyte experimental systems to show that PTRF localized to the nucleus and associated with the pol I transcription complex , playing a direct role on metabolically regulated ribosomal DNA transcription . A number of PTRF post-translational modifications and motifs can explain its nuclear translocation and the role in mediating insulin and nutrient-regulated ribosomal transcription . PTRF also plays a critical role in maintaining the active rDNA transcriptional 'loop' formation . Our studies not only show a specific role of PTRF on metabolism-regulated ribosomal DNA transcription in the adipocyte , they add another layer of regulation to rDNA transcriptional complexity . They also highlight the role of ribosome biogenesis in adipocyte allostasis .
Using immunofluorescence staining , we showed that PTRF has a clear insulin stimulated nuclear translocation in cultured 3T3-L1 adipocytes ( Figure 1A ) . After insulin stimulation , PTRF but not caveolin-1 showed significant nuclear localization in isolated primary ( Top panels ) and cultured 3T3-L1 adipocytes ( Bottom panels ) ( Figure 1B ) . To further confirm this , we fractionated 3T3-L1 adipocytes into nuclear and non-nuclear fractions by sucrose gradient centrifugation , and as shown in Figure 1C , a significant portion of PTRF translocated to nucleus upon insulin stimulation . In contrast other caveolae component proteins such as caveolin-1 , Cavin-2 ( SDPR ) , and Cavin-3 ( SRBC ) did not show any nuclear localization or stimulated translocation ( Figure 1C ) , suggesting a specific role of PTRF on rRNA transcriptional regulation . We performed an immunoprecipitation from the nuclear fraction with PTRF antibody , and consistent with the prior in-vitro assay ( Jansa et al . , 1998 ) , our data indicated PTRF binds to TTF1 ( polymerase I transcription termination factor 1 ) in cells ( Figure 1D ) . We also performed chromatin immunoprecipitation using a PTRF antibody followed by qPCR using primers probing the ribosomal transcription termination region . As shown in Figure 1E , the occupancy of PTRF in the termination complex is significantly regulated by the nutrient availability and insulin . These results show PTRF nuclear localization and binding to the rDNA transcription complex , suggesting that PTRF might play a functional role in ribosomal transcription activity in cells . 10 . 7554/eLife . 17508 . 003Figure 1 . A functional role of PTRF in nucleus . ( A ) Immunofluorescence staining of PTRF in basal and insulin stimulated 3T3-L1 adipocytes . Scale bars , 50 μm . ( B ) Higher magnificent images of insulin stimulated single isolated primary ( top ) and 3T3-L1 adipocyte ( bottom ) are shown , green: PTRF , and red: caveolin-1 . Scale bars , 10 μm . ( C ) Nuclear and non-nuclear fractions from basal or stimulated primary adipocytes were subject to western blots by indicated antibodies . ( D ) Co-immunoprecipitation by PTRF antibody from basal and stimulated adipocytes nuclear lysate followed by western blots using TTF1 ( Transcription Termination Factor , RNA Polymerase I ) and PTRF antibodies , PTRF null adipocytes ( KO ) serving as a negative control for the co-IP . ( E ) Chip-qPCR assay using PTRF antibody from wild type control ( WT ) and PTRF null adipocytes ( KO ) . *p<0 . 05 , **p<0 . 01 , and ***p<0 . 001; Student’s test . Error bars indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 17508 . 003 Previous studies have shown that rDNA transcription can be acutely regulated by nutrients and growth factors ( reviewed in [Moss et al . , 2007] ) . After pre ribosomal RNA ( pre-rRNA ) is synthesized , the first step is to remove the 5’ portion of the 5’-externally transcribed spacer ( 5’-ETS ) to generate 45S pre-rRNA ( Craig et al . , 1987; Gurney , 1985; Miller and Sollner-Webb , 1981 ) . The length of the removed region , known as the leader fragment of the 5’-ETS , is 650 bases in mouse and 414 bases in human . Because this leader sequence is rapidly excised and degraded after transcription ( Puvion-Dutilleul et al . , 1997 ) , its quantity in the nucleolus is a sensitive and highly selective marker of the amount of the nascent 47S pre-rRNA and hence , a widely used indicator of the rate of ribosomal transcription . Here we let 3T3-L1 fibroblasts and differentiated adipocytes 'rest' in PBS with 1% BSA for 3–4 hr , then switched them back to full culture medium ( 10% FBS in high glucose DMEM ) with insulin . As shown in Figure 2A , pre-rRNA 47S levels were significantly up-regulated by this culture condition change . Compared to fibroblastic cells , the response in adipocytes is 2-fold higher , suggesting an adipocyte-specific rDNA transcription regulation mechanism as the levels of PTRF are much higher in the latter as compared to the former ( Bastiani et al . , 2009 ) . To see if PTRF played any role on this regulation , we isolated primary adipocytes from 6–8 weeks old wild type and PTRF null mice and challenged them with similar medium changes by switching from 1% BSA/KRP buffer to 10% FBS in high glucose DMEM with 100nM insulin . A dramatic defect of rDNA transcription up-regulation was observed in PTRF null cells as compared to wild type cells ( Figure 2B ) . A time-course experiment further showed the peak of rDNA transcription in response to a metabolic challenge was significantly blunted in PTRF null cells ( Figure 2C ) . 10 . 7554/eLife . 17508 . 004Figure 2 . PTRF mediated ribosomal transcription regulations . ( A ) Relative pre-rRNA ( 47S ) levels were detected by RT-qPCR from 3T3-L1 fibroblast and fully differentiated adipocytes stimulated by switching culture medium from 1% BSA in PBS to full growth medium with insulin ( Fed ) . ( B ) Basal ( KRP buffer with 1% BSA ) or 45 min stimulated ( Fed: high glucose DMEM , 10% FBS , and insulin ) pre-rRNA levels were detected by RT-qPCR in primary adipocytes isolated from wild type ( WT ) and PTRF null ( KO ) mice . ( C ) A stimulation time-course of B . ( D ) Pre-rRNA levels were detected by RT-qPCR from basal or stimulated wild type ( WT ) , PTRF null ( KO ) , PTRF re-transfected PTRF null ( KO+PTRF ) , and Cav-1 re-transfected PTRF null ( KO+Cav1 ) MEFs cells . ( E ) Relative levels of total proteins , RNA , 18S , and 47S from wild type or PTRF null mice ( n = 6 ) were measured and normalized to cell numbers . *p<0 . 05 , **p<0 . 01 , and ***p<0 . 001; Student’s test . Error bars indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 17508 . 00410 . 7554/eLife . 17508 . 005Figure 2—figure supplement 1 . Characterization of ribosome biogenesis in 6–8 weeks old PTRF null and control mice . Total proteins , RNAs , 18S , and 47S levels were measured from 6–8 weeks old wild type control and PTRF null mice ( n = 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17508 . 00510 . 7554/eLife . 17508 . 006Figure 2—figure supplement 2 . Cell growth curve of cultured WT and PTRF null MEFs cells . Primary PTRF null ( KO ) and wild type control ( WT ) MEFs cell numbers were determined by cell counting on indicated days . **p<0 . 01; Student’s test . Error bars indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 17508 . 006 To further confirm this is a PTRF-dependent effect , similar experiments were performed in cultured primary ( within 2–3 passages ) wild type , PTRF null , and PTRF or caveolae-1 rescued null MEF cells . To exclude cell proliferation associated ribosomal transcription , cells were pre-treated with nocodazole to induce cell arrest by depolymerizing the microtubule network , and culture media were switched from 1% BSA in PBS to full culture medium with insulin . Consistent with the results from adipocytes , PTRF null MEFs showed a dramatically smaller response to the culture condition change . PTRF transfection rescued this defect , whereas caveolin-1 ( Cav-1 ) cannot , indicating a role of PTRF independent of its function in caveolae ( Figure 2D ) . To determine if loss of this acute PTRF-dependent rDNA transcriptional regulation caused any long term ribosome biogenesis defect , we measured total protein , RNA and pre-rRNA contents in primary adipocytes from PTRF null and control wild type mice . Although 6–8 week old mice did not show any differences ( Figure 2—figure supplement 1 ) , in 6-month old mice , total protein and RNA were significantly reduced in PTRF null primary adipocytes ( Figure 2E ) . A significantly lower basal level of 47S was also observed . In addition , PTRF null MEFs showed a slower growth rate comparing to wild type control ( Figure 2—figure supplement 2 ) . These data support a ribosome degeneration phenomenon , which may explain the in vivo dystrophic phenotype we observed in PTRF null mice ( Ding et al . , 2014; Liu et al . , 2008 ) , also seen in human patients with PTRF mutations ( Hayashi et al . , 2009 ) . To determine if PTRF played a direct causal effect on rDNA transcription , we created a PTRF null stable 3T3-L1 cell line using CRISPR/cas9 genome editing technology . Three different Ptrf gene exon loci were targeted . Similar results were obtained from all three cell-lines ( Figure 3—figure supplement 1A ) and results from one of these ( KO3 ) are shown here ( Figure 3A ) . From the cellular lipid content and adiponectin ( Adpn ) secretion ( Figure 3B ) and the expression of key adipocyte differentiation markers of 3T3-L1 by western blot ( Figure 3C ) and qPCR ( Figure 3—figure supplement 1B ) , we concluded that the PTRF null 3T3-L1 cells do not have any significant deficit in the degree of 3T3-L1 adipocyte differentiation as compared to control cells . We also did not see any significant changes for total protein , RNA and 47S rRNA levels ( Figure 3—figure supplement 1C ) . However when we subject the cells to metabolic challenges by switching culture medium from PBS with 1% BSA ( 'fasting' ) to high glucose DMEM with 10% FBS and insulin ( 'feeding' ) , 47S levels in PTRF null adipocytes were not up-regulated as efficiently as in control cells ( Figure 3D ) . A further time-course study showed a delayed response upon loss of PTRF . Conversely , 47S levels in PTRF null adipocyte were not down-regulated as efficiently as in wild type cells under nutritional challenge by switching from full culture medium to PBS with 1% BSA ( fasting ) . These data further support a crucial functional role of PTRF on metabolically regulated rDNA transcription . 10 . 7554/eLife . 17508 . 007Figure 3 . PTRF plays critical role mediating ribosomal transcription response to metabolic challenges in CRISPR/Cas9 genomic editing cell model , without any affect on differentiation . ( A ) Indicated protein levels were detected by Western blots from CRISPR/Cas9 genomic edited PTRF null and control 3T3-L1 adipocytes . ( B ) Relative total TG content and adiponectin secretion levels were measured from CRISPR/Cas9 genomic edited PTRF null and control 3T3-L1 adipocytes . ( C ) Expression levels changes of indicated proteins during 3T3-L1 differentiation were measured by western blot from CRISPR/Cas9 genomic edited PTRF null and control 3T3-L1 adipocytes . ( D ) Pre-rRNA levels from nutrients/insulin stimulated ( feeding: switching from PBS with 1% BSA to high-glucose DMEM with 10% FBS and insulin ) or starved ( Fasting: switching the growth medium to PBS with 1% BSA ) were measure by RT-qPCR from CRISPR/Cas9 genomic edited PTRF null and control 3T3-L1 adipocytes . ( E ) After CRISPR/Cas9 genomic edited PTRF null and control 3T3-L1 adipocytes were subject to 'feeding' 12 hr followed by 'fasting' 12 hr cycle for seven days , total TG content , proteins , RNAs and 47S levels were measure as described before . *p<0 . 05 , **p<0 . 01 , and ***p<0 . 001; Student’s test . Error bars indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 17508 . 00710 . 7554/eLife . 17508 . 008Figure 3—figure supplement 1 . Characterizations of CRISPR/Cas9 genomic edited PTRF null 3T3-L1 adipocytes . ( A ) Whole cell lysates from control and 3 cell lines of CRISPR/Cas9 genomic edited PTRF null 3T3-L1 adipocytes were separated on SDS-PAGE followed by immuno-blots using indicated antibodies . ( B ) Gene expression levels of some adipocyte specific markers were measured from control and one cell line ( KO3 ) of CRISPR/Cas9 genomic edited PTRF null 3T3-L1 adipocytes by RT-qPCR . ( C ) Total proteins , RNAs and 47S were measured from control and one cell line ( KO3 ) of CRISPR/Cas9 genomic edited PTRF null 3T3-L1 adipocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 17508 . 008 Unlike the cell culture model , humans and mice are constantly challenged by daily feeding or fasting cycles . To mimic these physiological conditions in vitro , we repeatedly challenged WT and PTRF null 3T3-L1 adipocytes by repeated 'feeding' , and 'fasting' at 12-hr intervals . After seven days , the PTRF null adipocytes showed a dramatically lower lipid content , less total protein and less total and 47S RNA levels ( Figure 3E ) , which essentially recapitulated the lipodystrophy characteristics observed in the PTRF knockout mouse model and human patients with PTRF deficiencies . These data suggest a PTRF-dependent ribosomal transcriptional response may be the early and direct causal mechanism for the pathogenesis of in vivo lipodystrophy . Interestingly we saw a subunit of RNA polymerase I , PolR1A dramatically decreased upon 3T3-L1 differentiation , whereas PTRF and TTF1 levels increased ( Figure 3C ) , suggesting a possible adipocyte-specific termination factor-dependent mechanism for the rDNA transcription regulations ( See Discussion ) . It has been reported that deregulated ribosomal biogenesis can induce p53 pathway activation ( Woods et al . , 2015; Zhang and Lu , 2009 ) , which will lead to cell growth arrest , apoptosis and cell death ( Berkers et al . , 2013; Vousden and Prives , 2009 ) . Here we first examined p53 activation in PTRF null 3T3-L1 adipocytes . We did not observe a significant p53 and MDM2 protein level change at basal condition ( Figure 4B , left ) . However when we challenged the cells with 'feeding' medium , PTRF null cells showed a poor rRNA response ( Figure 4A ) , higher ribosomal protein rpl5 levels , lower mdm2 and higher p53 nuclear accumulation ( Figure 4B , right ) , indicating p53 pathway activation . This is probably due to insufficient rRNA synthesis causing a relative increased accumulation of ribosomal proteins in the nucleus , which is also called nuclear stress . When we challenged the cells by 'fasting' medium , the inefficient rRNA transcription regulation was confirmed in PTRF null cells comparing to WT ( Figure 4C ) . However we saw opposite responses of p53 activation , which was transiently up-regulated in control wild type cells and lost in PTRF null adipocytes ( Figure 4D ) . This response in wild type cells seems to be only at cell arrest levels ( activated p21 , gadd45 , and cdkn1a ) , not cell apoptosis ( unchanged levels of atm , Bax and caps2 ) ( Figure 4E ) , which probably represents some level of cell response under nutrient starved conditions . When we examined the adipose tissue from PTRF null mice , a full range of p53 activation targets including both cell arrest and apoptosis all were up regulated ( Figure 4F–G ) , indicating a long term adipocyte apoptosis . 10 . 7554/eLife . 17508 . 009Figure 4 . The p53 pathway is dys-regulated upon loss of PTRF . Pre-rRNA levels from fed ( A ) and fasted ( C ) control and CRISPR/Cas9 edited PTRF null adipocytes were determined by RT-qPCR . Nucleus fraction lysates from fed ( B ) or fasted ( D ) adipocytes were separated in SDS-page followed by western blot by using the indicated antibodies , RPL5: ribosomal protein L5; MDM2: mouse double minute 2 homolog; H3: histone-H3 . ( E ) p53 pathway related marker gene expression levels were measured from basal and fasted control or CRISPR/Cas9 edited PTRF null adipocytes by RT-qPCR . Gadd45: Growth arrest and DNA-damage-inducible protein GADD45; Cdkn2a: cyclin-dependent kinase inhibitor 2; Atm: Ataxia telangiectasia mutated; Bax: bcl-2-like protein 4; Caps2: Calcyphosine 2 . ( F ) p53 pathway related marker gene expression levels were measured from six-month old wild type control and PTRF null mice adipose tissues . ( G ) Wild type and PTRF null mice adipose tissue lysate were subject to western blots by using indicated antibodies . **p<0 . 01; Student’s test . Error bars indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 17508 . 009 Previously we and others have shown PTRF can be tyrosine phosphorylated by EGF or insulin signaling pathway activations ( Guha et al . , 2008; Humphrey et al . , 2013; Pilch and Liu , 2011; Pilch et al . , 2007 ) . However the functional relevance of this modification in a physiological context remains unknown . Consistent with previous results , we show here that insulin stimulates PTRF tyrosine phosphorylation ( Figure 5A ) . Interestingly another caveolae protein in the cavin family , SDPR ( Serum deprivation response protein , Cavin-2 ) co-immunoprecipitated with PTRF ( Bastiani et al . , 2009 ) also showed increased insulin-dependent tyrosine phosphorylation but it does not undergo nuclear translocation ( Figure 1C ) . Next we investigated whether this modification plays any role in rDNA transcription regulation . Since PTRF has only 4 tyrosine sites , we substituted each of these with phenylalanine , Y158F showed a significant decrease in phosphotyrosine levels ( Figure 5B ) . Accordingly , this mutant form ( Y158F ) shows a significant inhibitory effect on rDNA transcription activity when transfected into PTRF null MEF cells ( Figure 5C ) , suggesting a critical role for this phosphorylation site in mediating the up-regulation of rDNA transcription activity in response to feeding medium . 10 . 7554/eLife . 17508 . 010Figure 5 . PTRF regulates ribosomal transcription through tyrosine and Ser/Thre phosphorylations . ( A ) Isolated mouse primary adipocytes were stimulated by insulin from 0–60 min . The whole cell lysates were immunoprecipitated by PTRF antibody followed by SDS-PAGE and blotted with indicated antibodies . ( B ) Whole cell lysates from wild type PTRF and various single tyrosine mutants ( 1: Y13F; 2: Y158F; 3: Y310F; 4: Y318F ) transfected HEK293 cell were immunoprecipitated by PTRF antibody followed by SDS-PAGE and blotted with indicated antibodies . ( C ) Basal and nutrients/insulin stimulated 47S levels were measured from control , wild type PTRF and various single tyrosine mutants transfected PTRF null MEFs cells . ( D ) Whole cell lysates from 3T3-L1 adipocyte stimulated by isoproterenol ( ISO ) for 0–90 min were subject to SDS-PAGA and blotted with indicated antibodies . ( E ) 47S levels were measured from ISO stimulated 3T3-L1 adipocytes by RT-qPCR . ( F ) Isolated mouse primary adipocyte were nutrients starved ( fasted ) or stimulated by insulin with nutrients . The whole cell lysates were immunoprecipitated by PTRF antibody followed by SDS-PAGE and blotted with indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 17508 . 010 PTRF has also been shown to be serine/threonine phosphorylated ( Aboulaich et al . , 2004 , 2011 ) . Upon PKA activation by isoproterenol ( ISO ) in adipocytes , PTRF shows a significant size shift on SDS/PAGE . Consistent with this we showed a slower PTRF band migration in SDS page following ISO stimulation ( Figure 5D ) . Furthermore we showed this PKA pathway-dependent phosphorylation associated with a down-regulation of rDNA transcriptional activity measured by 47S rRNA levels ( Figure 5E ) , suggesting a potential mechanism for negatively regulated ribosome transcription through PTRF Ser/Thre-phosphorylation under catabolic conditions . We also show this Ser/Thre-phosphorylation can be reversed by insulin ( Figure 5F ) . These data support the hypothesis that PTRF is a key metabolism-dependent transcriptional efficiency regulator that can be both positively and negatively regulated through different signaling pathways , thus fulfilling its physiological functions in adipocytes under various nutrient and hormonal conditions . Next we probed the potential mechanism ( s ) for PTRF translocation between different cellular localizations . PTRF has two putative nuclear localization signals ( Aboulaich et al . , 2004 , 2006; Kalderon et al . , 1984; Wei et al . , 2015 ) . Besides this , PTRF contains a protein sequence ( 54VLVLSLLDKII64 ) , which is highly similar to nuclear export signals ( NESs ) ( Figure 6A ) . The NES is a short amino acid sequence of hydrophobic residues in a protein that targets it for export from the cell nucleus to the cytoplasm through the nuclear pore complex by interacting with exportin , and the common NES sequence motif is 'LxxLxxLxL' , where 'L' is a hydrophobic residue ( often leucine ) and 'x' is any other amino acid ( Bogerd et al . , 1996 ) . To investigate if this NES signal plays any role in PTRF nuclear localization , we generated a mutant by deletion of this region and expressed it in HEK293 cells . As shown in Figure 6B this mutant was exclusive localized in the nucleus , probably due to the loss of nuclear export signal . When we transfected it into PTRF null primary MEFs cells , unlike wild type , this mutant cannot rescue 'feeding' medium 47S up-regulation under cell growth arrest condition ( Figure 6C ) , indicating a nuclear export 'recycling' process is needed for maintaining PTRF ribosomal transcription regulatory activity . 10 . 7554/eLife . 17508 . 011Figure 6 . PTRF nuclear localization is regulated by its nuclear export signals and other post-translational modifications . ( A ) Schematic illustration of nuclear export signals ( NES ) and nuclear localization signal ( NLS ) motifs in mouse PTRF protein . ( B ) NES deleted PTRF constructs and wild type control were transfected to HEK293 cells . The nuclear and cytoplasmic fractions were separated on SDS-PAGE following by immunoblot using indicated antibodies . ( C ) Basal and nutrients/insulin-stimulated 47S levels were measured from empty vector controls ( C ) , wild type ( WT ) and NES motif deleted PTRF cDNA constructs transfected PTRF null MEFs cells . ( D and E ) Total rat adipocyte lysate were immunoprecipitated by IgG , or PTRF or ubiquitin ( Ui ) followed by SDS-PAGE and blotted with indicated antibodies . ( F ) A scheme of 8 designed PTRF fragment mutants ( C1-4 , N1-4 ) . ( G ) Whole cell lysates from Wild type HEK293 cells transfected by ( WT ) and 8 PTRF fragment mutants were immunoprecipitated by myc-tag magnetic beads followed by immuno-blots with indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 17508 . 011 Mouse PTRF has 392 amino acids and the predicted size is 43 kDa . However the band in SDS-PAGE always migrates at 55–60 kDa . By immunoprecipitation and immunoblotting , we showed that PTRF is mono-ubiquitinated ( Figure 6D and E ) . Based on the size of 8 kDa for each ubiquitin protein , PTRF seems to be mono-ubiquitinated at two sites . Thus we expressed 8 PTRF truncation constructs ( Figure 6F ) in HEK293 cells and immunoprecipitated them with anti-myc antibody followed by western blot using an anti-ubiquitin antibody . As shown in Figure 6G , it seems that two regions ( 166-192aa and 193-249aa ) are responsible for one mono-ubiquitination each . Protein monoubiquitination has recently emerged as an important posttranslational modification regulating transcription , endocytic vesicle trafficking , histone modification , and DNA repair ( Friedberg , 2006; Pemberton and Paschal , 2005 ) . For example , the addition of a mono-ubiquitin tag to p53 is sufficient for its nuclear export; p53 monoubiquitination may cooperatively interact with its nuclear export signal ( NES ) for cytoplasmic targeting ( Carter et al . , 2007 ) . NLS can also be regulated by monoubiquitination signals . Studies have suggested that monoubiquitination can mask its NLS resulting in cytoplasmic retention ( Chen and Mallampalli , 2009 ) . Our data in Figures 5 and 6 , taken together , suggest the localization of PTRF in caveolae or in the nucleus might be coordinated by monoubiquitination , NLS and NES motifs . To test if intact caveolae are required for PTRF ribosomal functions we treated mouse primary adipocyte with beta-methyl-cyclodextrin ( beta-CD ) , which has been widely used to disrupt caveolae structure due to cholesterol depletion ( Wiesmann et al . , 1975; Liu and Pilch , 2008 ) . As shown in Figure 7A , beta-CD treatment caused PTRF and caveolin-1 redistribution out of lipid raft fractions , consistent with previous reports . However insulin-stimulated PTRF nuclear translocation was completely blocked ( Figure 7B ) , indicating the localization of PTRF in caveolae is critical for this translocation . Furthermore immunoblot analyses of tyrosine-phosphorylated proteins in cell lysates showed that PTRF tyrosine phosphorylation was largely inhibited although the insulin stimulated IRS-1/2 ( 180 kDa ) and the beta-subunit of insulin receptor ( 95 kDa ) ( arrows ) were not affected by beta-CD treatment ( Figure 7C ) . PTRF mediated ribosomal transcription regulation was diminished as shown in Figure 7D . These results suggest intact caveolae structures provides a platform for insulin stimulated PTRF tyrosine phosphorylation , which is essential for its nuclear translocation and ribosomal function . 10 . 7554/eLife . 17508 . 012Figure 7 . Intact caveolae are required for PTRF tyr-phosphorylation , nuclear translocation and ribosomal function . ( A ) Isolated mouse primary adipocytes treated with beta-cyclodextrin ( Beta-CD ) or not were fractionated into crude lipid rafts and non lipid rafts fractions . Lysate from both fractions were run on SDS-PAGE and blotted with indicated antibodies . ( B ) Nuclear fraction lysates from beta-CD ( 45-min ) and insulin ( 30-min ) treated or not adipocyte were run on SDS-PAGE and blotted with indicated antibodies , histone-H3:H3 . ( C ) Whole cell lysates from B were separated by SDS-PAGE and blotted with phosphotyrosine antibody ( 4G10 ) . Arrows indicate the size of 180 and 95 kDa , which possibly corresponds to IRS-1/2 and beta-subunit of insulin receptor respectively . The whole cell lysates were also immunoprecipitated by PTRF antibody followed by SDS-PAGE and blotted with indicated antibodies . ( D ) 47S levels of the samples from B were measured by RT-qPCR . DOI: http://dx . doi . org/10 . 7554/eLife . 17508 . 012 Previous in vitro assays suggested PTRF up-regulates rDNA transcription rate through 'releasing' Pol-I from the transcription termination complex , which results in more efficient transcriptional re-initiation ( Jansa and Grummt , 1999 ) . The active rDNA transcription complex forms nucleotide loops by jointing initiation and termination regions together ( Grummt and Langst , 2013 ) . Here we tested if PTRF plays any role on this active loop formation . By using a chromosome conformation capture assay ( 3C assay ) ( Nemeth et al . , 2008 ) , we pulled down the transcriptional complex and ligated the DNA fragments followed by PCR using primers targeting the initiation and termination loci ( Figure 8A ) . As shown in Figure 8B the PCR product was almost completely absent in PTRF null cells comparing to control WT cells , indicating the existence of a PTRF-dependent active rDNA transcription loop formation upon nutrient and insulin stimulation . 10 . 7554/eLife . 17508 . 013Figure 8 . PTRF dependent active ribosomal transcription complex formation . ( A ) Schematic illustration of chromatin confirmation capture ( 3C ) assay , details are described in Method . ( B ) Following 3C assay , the relative levels of active transcription were quantified by RT-qPCR using nutrition/insulin stimulated primary mouse adipocytes from wild type ( WT ) and PTRF null ( KO ) mice ( n = 4 ) . ***p<0 . 001; Student’s test . Error bars indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 17508 . 013
Publications on PTRF/Cavin-1 in the past 8 years have focused almost exclusively on its essential role as a structural component of caveolae , and these structures are especially abundant in fat cells ( Parton and del Pozo , 2013 ) . When PTRF/Cavin-1 is knocked out in mice or lost through mutation in humans , caveolae are absent and the metabolic consequences are severe , including muscular and lipodystrophies . The possible explanation ( s ) proposed for the adipocyte lipodystrophic phenotype include alterations in lipid transport and/or altered plasma membrane adaptations ( Pilch and Liu , 2011 ) , but these do not fully explain the progressive nature of the lipodystrophy observed in vivo . The transport/membrane defects would be present at birth presumably causing an immediate fat cell deficiency . On the other hand and given that the lipodystrophy of PTRF/Cavin-1 null mice and humans has a progressive severity , PTRF’s eponymous function in rRNA synthesis does indeed provide a possible explanation whereby decreased rRNA transcription as a function of time would lead to adipocyte loss due to the inability to maintain adequate protein levels , and we present data here in support of this hypothesis . It was recognized some time ago that a fraction of PTRF could be found in the nucleus of human fat cells consistent with it having two nuclear localization sequences , but the possible functional significance of this finding was not pursued ( Aboulaich , 2004 ) . Here we confirm that PTRF is present in the nucleus of fat cells , and indeed , it functions in these cells to promote rRNA transcription ( Figures 1–3 ) . The data of Figures 1–3 derive from a comparison of WT and PTRF null primary mouse fat cells and gene edited cultured fat cells , and taken together , support the hypothesis that inefficient rRNA transcription underlies the lipodystrophy observed in mice and humans . Moreover , the absence of PTRF would be expected to lead to ribosomal stress , which in turn would lead to fat cell loss over time , and we find evidence for this process in fat cells from PTRF null mice ( Figure 4 ) . The nuclear localization of PTRF is dynamic and regulated by insulin-dependent tyrosine phosphorylation under anabolic conditions and Ser/Thre phosphorylation under catabolic conditions ( Figure 5 ) . Mutation of tyr-158 to phe abrogates the regulation of rRNA transcription and prolonged exposure to the beta-adrenergic agent , isoproterenol , leads to ser/thre phosphorylation and diminution of 47S transcription , the catabolic condition ( Figure 5 ) . These results make physiological sense by adjusting potential ribosomal activity to the nutritional status of the cell . A further level of regulation may be achieved by ubiquitination and we have verified the necessity of the nuclear export signal in PTRF dynamics ( Figure 6 ) . Our data ( Figure 7 ) support the idea that intact caveolae are required for PTRF tyrosine phosphorylation and nuclear function . Further conceptual support for this derives from the fact that the relevant kinases are very likely to be located at the plasma membrane but further experiments regarding this and the other PTRF modifications will be necessary to complete the picture . Lastly for ribosome transcription , it’s also proposed that 'loop' formation between initiation and termination complexes is essential for activation ( Denissov et al . , 2011; Nemeth et al . , 2008; Nemeth and Langst , 2011; Sander and Grummt , 1997; Shiue et al . , 2009 ) . Our studies support that PTRF play an essential role in this loop formation ( Figure 8 ) . We show that in 3T3-L1 pre-adipocytes , there is a high level of Pol-I expression in order to ensure the needs of rRNA syntheses by increasing the number of active transcription sites . This is closely correlated with pre-adipocyte proliferation . Upon adipocyte differentiation and consistent with previous results ( Li et al . , 2006 ) Pol-I levels were dramatically decreased ( Figure 3C ) , whereas PTRF and TTF-1 levels were increased correspondingly , indicating ribosomal transcription in mature adipocyte became less Pol-I dependent , and more reliant on the regulation of transcriptional efficiency through the termination machinery . Our results support the notion that cells have developed at least these two different mechanisms to ensure proper control of ribosomal transcription in a cell type and physiological condition-specific manner . One mechanism is Pol-I dependent active transcriptional number regulation , which plays major role in proliferating cells . The other is termination factor , PTRF-dependent transcriptional rate regulation , which plays a critical role in mature differentiated cells in response to nutrient challenges . In fact , all cells must maintain a certain level of ribosomal activity to maintain protein expression upon their normal turnover , and nutritionally stressed cells such as adipocytes have a particular need for this activity . These cells are subject to repeated cycles of feeding and fasting , and indeed , we show in vitro ( Figure 3E ) , the consequences of PTRF deficiency in this regard . We see no apparent differences in protein and fat content between WT and PTRF null fat cells until we expose them to the metabolic stress of repeated fasting and refeeding cycles . The PTRF null cells show a dramatic reduction in total triglyceride , RNA and protein content which essentially recapitulates the lipodystrophy in vitro ( Figure 3E ) that is occurring in vivo in PTRF deficient mice and humans . The PTRF null status obviously represents the extreme , but variations in PTRF expression would be predicted to have phenotypic consequences , in particular , in response to overnutrition , a ubiquitously prevalent condition in current times . What has been observed is that some individuals respond to over-nutrition by fat cell expansion requiring ribosomal activity to amass the additional proteins required to increase the size of fat cells and accommodate the excess nutrients as triglyceride , the so-called healthy obese individuals who lack obvious metabolic pathology . On the other hand , most people apparently fail to mount an adequate adipocyte adaptive response , and this results in fat spilling over to liver and muscle resulting in hepatosteatosis and type 2 diabetes . We hypothesize that the capacity of adaptational ribosomal DNA transcription may determine the maximal limit of adipocyte functional homeostasis and may coincide with these metabolic perturbations . In support of this idea , mice were exposed to conditions of normal , over and under nutrition from birth to weaning and then subjected to the same diet and followed for up to 112 days . The mice subject to undernourishment show an apparent inability for adipose tissue expansion that correlates with PTRF expression ( Kozak et al . , 2010 ) . Together these data and ours suggest the PTRF-dependent ribosomal transcriptional response to metabolic challenges may be the early and direct causal molecular mechanism for the pathological developments of adipose tissue function in vivo , lipodystrophy in the worst case scenario .
Ptrf−/− mice were created as described in ( Liu et al . , 2008 ) . They were backcrossed for at least 8 generations with the C57 black lineage . The mice used in the present study were homozygous male Ptrf−/− and their wild-type ( WT ) littermates generated from breeding of Ptrf+/− mice . Animals were maintained in a pathogen-free animal facility at 21°C under a 12-hr light/12-hr dark cycle with access to a chow diet ( CD , 2918; Harlan Teklad Global Diet , Madison , WI ) . For the preparation of isolated adipocytes , freshly harvested adipose tissue was digested by collagenase in Krebs-Ringer bicarbonate ( KRB ) buffer . Isolated rat adipocytes were prepared by the collagenase method as described in ( Liu et al . , 2006 ) from epididymal adipose tissues of Sprague-Dawley rats ( from Charles-River , 170–220 g ) . All animal studies were performed in accordance with the guidelines and under approval of the Institutional Review Committee for the Animal Care and Use of Boston University . 3T3-L1 fibroblasts culture and differentiation were previously described in ( Liu and Pilch , 2008 ) . PTRF null and control wild type primary MEFs cells used in all experiments were maintained within 3–5 passages . When they reached 90% confluence , they were transfected with the cDNA of interest by means of the X-tremeGENE HP reagent ( Roche , Indianapolis , IN ) . HEK293 cells were cultured and transfected as describe in ( Liu and Pilch , 2008 ) . The PTRF null 3T3-L1 stable cell lines were generated using lentiviral plasmid based CRISPR/Cas9 genome editing , in which U6 promoter driven guide RNA , CMV promoter driven cas9 and puromycin selection marker were supplied by one single-vector ( all-in-one ) . Four vectors , including three mus musculus Ptrf targets ( K1: 5’-TCACGCTCCATATCGTTGAG-3’; K2: 5’-GTCAACGTGAAGACCGTGCG-3’; K3: 5’-GGTCAGCTGGATCTGGTCAA-3’ ) and one non-targeting control were custom designed and made by Transomic Technologies ( Huntsville , AL ) . Lentivirus were packing used third generation packing system . The stable cell lines were obtained after lentivirus transduction and puromycin selection . These cells were cultured and differentiated as described above . Dexamethasone , 3-isobutylmethylxanthine , insulin , sodium fluoride , sodium orthovanadate , fetal bovine serum ( Australian origin ) , benzamidine , and mouse immunoglobulin G ( IgG ) were purchased from Sigma ( St . Louis , MO ) . LB base , ampicillin , kanamycin , aprotinin , leupeptin , and pepstatin A were obtained from American Bioanalytical ( Natick , MA ) . Calf serum was purchased from Life Science ( Cambridge , MA ) , and Dulbecco’s modified Eagle’s medium ( DMEM ) was from Mediatech ( Herndon , VA ) . Transfection reagent and the pcDNA 3 . 1 expression vector were purchased from Life Science . A BCA protein assay kit was from Pierce . Protein A or G magnetic beads was from Santa Cruz Biotechnology ( Santa Cruz , CA ) . Penicillin , streptomycin , and trypsin were purchased from Life Science . Monoclonal antibodies recognizing PTRF ( 2F11 ) , caveolin-1 ( 7C8 ) , have been previously described ( Souto et al . , 2003; Vinten et al . , 2001 ) . The following antibodies were commercially acquired: anti-caveolin-1 was from BD Transduction Laboratories ( San Jose , CA ) , anti-actin was from Sigma; anti-transferrin receptor was from Zymed Laboratories ( South San Francisco , CA ) . Additional anti-PTRF antibodies were purchased from BD Transduction . Polyclonal rabbit anti-PTRF antibody was also produced against a peptide sequence at the C terminus of the protein ( 21st Century Biochemicals , Hopkinton , MA ) . Primary antibodies were detected in Western blots using secondary antibodies conjugated to horseradish peroxidase ( Sigma ) diluted 1:3000 and chemiluminescent substrate ( PerkinElmer Life Sciences , Boston , MA ) , followed by detection by Fujifilm LAS-4000 Image Analyzer . Nucleus fraction was prepared according to sucrose centrifugation method ( Nuclei Isolation Kit , 'Nuclei PURE Prep' , Sigma , NUC201 ) . This protocol incorporates centrifugation through a dense sucrose cushion to protect nuclei and strip away cytoplasmic contaminants . The whole cell/tissue lysates or soncicated nucleus fraction was solubilized with 1% Triton X-100 . Insoluble material was removed by pelleting for 10 min in a microcentrifuge . Indicated antibodies and nonspecific mouse or rabbit IgGs were incubated with the supernatant 1 hr at 4°C , then 20–40 µl of protein A/G magnetic beads was added for 2 hr to overnight . The supernatant with unbound proteins was collected , and the beads were washed four times and eluted with SDS-PAGE loading buffer containing 2% SDS . This protocol for isolated rat adipocyte and 3T3-L1 cultured adipocyte was performed as described by ( Liu et al . , 2006 ) and ( Liu and Pilch , 2008 ) . Total RNA was isolated from indicated tissues or cells with TRIzol reagent ( Life Science ) , and the cDNA was synthesized using Reverse Transcription System ( Promega ) . Real-time PCR was performed with the ViiA7 detection system ( Applied Biosystems ) using Fast SYBR Green Master Mix ( Applied Biosystems ) . Gene expression levels were presented relative to the wild type . The primer sequences are listed in supplementary file . Briefly , control and stimulated cells were fixed in 1% paraformaldehyde for 10 min at room temperature followed by adding glycine ( 0 . 6 M , final concentration ) to quench the cross-linking reaction . Then cells were suspended in lysis buffer ( 10 mM Tris-HCl , pH 7 . 2 , 10 mM NaCl , 0 . 2% NP-40 and protease inhibitor cocktail; Sigma ) for 60 min on ice . The nuclei were pelleted , resuspended in nuclear buffer with 0 . 3% SDS and incubated for 1 hr at 37°C . Triton-X100 ( 1 . 8% ) was added to sequester the SDS followed by the addition of high concentration BglII and BamHI ( NEB ) to digest the chromatin overnight at 37°C with gentle shaking . The restriction enzymes were inactivated by the addition of SDS to 1 . 6% and incubation at 70°C for 15 min . The reaction mixture was diluted with 1x ligase buffer ( NEB ) and incubated for 1 hr at 37°C . Ligation of DNA was done using T4 ligase for overnight at 16°C . Following reversal of cross-linking and DNA extraction , the samples were subjected to qPCR quantification . The various primer combinations were tested for their amplification efficiency using a control template prepared by mixing mouse rDNA plasmids in equimolar amounts , followed by digestion with BglII and BamHI and subsequent ligation . ChIP assays were performed according to the manufacturer’s protocol ( MAGnify Chromatin Immunoprecipitation System , Thermo Fisher Scientific ) . Briefly antibodies were incubated with crosslinked chromatin overnight at 4 degree and collected with protein A magnetic beads . After reversal of the crosslink and digestion with proteinase K , DNA was extracted and amplified by PCR . PCR products were first visualized on ethidium bromide-stained agarose gels then subjected to real-time PCR for quantification . All results are presented as mean ± SD . p values were calculated by unpaired Student’s t-test . *p<0 . 05 , **p<0 . 01 , and ***p<0 . 001 . P<0 . 05 was considered significant throughout . For cultured and primary isolated cells , all experiments were performed independently at least three times . Animal studies were from 4–6 animal per group . | Obesity can cause several other health conditions to develop . Type 2 diabetes is one such condition , which arises in part because fat cells become unable to store excess fats . This makes certain tissues in the body less sensitive to the hormone insulin , and so the individual is less able to adapt to changing nutrient levels . Without treatment or a change in lifestyle , this insulin resistance may develop into diabetes . However , “healthy obese” individuals also exist , who can accommodate an overabundance of fat without developing insulin resistance and diabetes . Some forms of rare genetic disorders called lipodystrophies , which result in an almost complete lack of body fat , can also lead to type 2 diabetes . This raises the question of whether lipodystrophy and obesity share some common mechanisms that cause fat cells to trigger insulin resistance . One possible player in such mechanisms is a protein called PTRF . In rare cases , individuals with lipodystrophy lack this protein , and mice that have been engineered to lack PTRF also largely lack body fat and develop insulin resistance . Fat cells can respond rapidly to changes in nutrients during feeding or fasting , and to do so , they must produce new proteins . Structures called ribosomes , which are made up of proteins and ribosomal RNA , build proteins; thus when the cell needs to make new proteins , it also has to produce more ribosomes . PTRF is thought to play a role in ribosome production , but it is not clear how it does so . Liu and Pilch analyzed normal mice as well as those that lacked the PTRF protein . This revealed that in response to cycles of fasting and feeding , PTRF increases the production of ribosomal RNA in fat cells , enabling the cells to produce more proteins . By contrast , the fat cells of mice that lack PTRF have much lower levels of ribosomal RNA and proteins . Liu and Pilch then examined mouse fat cells that were grown in the laboratory . Exposing these cells to insulin caused phosphate groups to be attached to the PTRF proteins inside the cells . This modification caused PTRF to move into the cell’s nucleus , where it increased the production of ribosomal RNA . Overall , the results show that fat cells that lack PTRF are unable to produce the proteins that they need to deal with changing nutrient levels , leading to an increased likelihood of diabetes . The next steps are to investigate the mechanism by which PTRF is modified , and to see whether the mechanisms uncovered in this study also apply to humans . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2016 | PTRF/Cavin-1 promotes efficient ribosomal RNA transcription in response to metabolic challenges |
Microglia are brain-resident macrophages that function as the first line of defense in brain . Embryonic microglial precursors originate in peripheral mesoderm and migrate into the brain during development . However , the mechanism by which they colonize the brain is incompletely understood . The retina is one of the first brain regions to accommodate microglia . In zebrafish , embryonic microglial precursors use intraocular hyaloid blood vessels as a pathway to migrate into the optic cup via the choroid fissure . Once retinal progenitor cells exit the cell cycle , microglial precursors associated with hyaloid blood vessels start to infiltrate the retina preferentially through neurogenic regions , suggesting that colonization of retinal tissue depends upon the neurogenic state . Along with blood vessels and retinal neurogenesis , IL34 also participates in microglial precursor colonization of the retina . Altogether , CSF receptor signaling , blood vessels , and neuronal differentiation function as cues to create an essential path for microglial migration into developing retina .
Microglia are the resident macrophages of brain . These dedicated CNS phagocytes form the innate immune system of embryonic and adult brain . Microglia eliminate cellular debris to prevent neuro-inflammation and to promote neuronal protection in vertebrates ( Ashwell , 1991; Calderó et al . , 2009; Lawson et al . , 1990; Neumann et al . , 2009; Sierra et al . , 2010 ) . They also prune unnecessary synapses to establish functional , mature neural circuits during brain development , performing a variety of cellular functions ( Paolicelli et al . , 2011; Tremblay et al . , 2010 ) . In contrast to other CNS cells , like neurons and astrocytes , microglia do not originate from neural plate , but are derived from mesoderm ( Ashwell , 1991; Boya et al . , 1979 ) through hematopoiesis ( Ginhoux et al . , 2013 ) . In developing zebrafish , embryonic hematopoiesis occurs in successive waves that are separated anatomically and temporally . The primitive or first wave of microglial precursors is generated from myeloid cells originating in the rostral blood island ( RBI ) at about 11 hr post-fertilization ( hpf ) ( Stachura and Traver , 2011; Xu et al . , 2012 ) . The definitive wave is contributed by the ventral wall of the dorsal aorta ( VDA ) , giving rise to hematopoietic stem cells ( HSCs ) ( Xu et al . , 2015 ) . In addition , a short intermediate wave also originates from the posterior blood island ( PBI ) ( Bertrand et al . , 2007 ) . After 2 weeks post-fertilization , VDA-derived microglia progressively replace RBI-derived microglia throughout the CNS ( Ferrero et al . , 2018; Xu et al . , 2015 ) . Thus , primitive and definitive hematopoiesis contribute embryonic and adult microglia , respectively , during zebrafish development . Generation of embryonic microglial precursors and their colonization of brain areas has been extensively described in zebrafish ( Herbomel et al . , 2001 ) . In zebrafish , embryonic microglial precursors are initially specified in lateral plate mesoderm and then spread on yolk . They start to migrate into the cephalic mesenchymal region after 22 hpf . At 26–30 hpf , a few microglia are observed in the vitreous space or choroid fissure of the optic cup , and around 30 microglia colonize the neural retina by 48 hpf . Microglial colonization of the optic tectum and other regions of zebrafish brain occurs after 48 hpf , indicating that the retina is one of the first brain regions to be colonized by microglia during development . Previous studies have suggested various signals that promote microglial colonization in brain . In mice , Cxcl12/CxcR4 signaling orchestrates microglial migration into developing cerebral cortex ( Arnò et al . , 2014; Hattori and Miyata , 2018 ) . In zebrafish , microglia migrate from the yolk-sac and colonize the brain in an apoptosis-dependent manner ( Casano et al . , 2016; Xu et al . , 2016 ) . Microglial precursors also migrate into the cephalic mesenchymal area in a Colony Stimulating Factor-Receptor ( CSF-R ) -dependent manner ( Herbomel et al . , 2001; Wu et al . , 2018 ) . Zebrafish fms mutants carry a genetic mutation in CSF-R and show severe delays in microglial colonization of both brain and retina , as well as an increase in neuronal apoptosis ( Herbomel et al . , 2001 ) . Recently , it was reported that brain colonization by microglial precursors depends primarily on one zebrafish CSF-R , CSF1ra , and one CSF-R ligand , IL34 , and that this combination of CSF ligand and receptor dominates this process ( Wu et al . , 2018 ) . Importantly , the number of microglia in the brain and retina is reduced in zebrafish il34 mutants that overexpress anti-apoptotic protein , Bcl2 . Thus , apoptosis and the IL34-CSF1ra signaling pathway cooperate to promote microglial colonization of the brain and retina during zebrafish development . In developing zebrafish retina , neurogenesis is initiated in the ventro-nasal retina , adjacent to the optic stalk at 25 hpf and progresses to the whole region of the neural retina , suggesting a spatio-temporal pattern of retinal neurogenesis in zebrafish ( Hu and Easter , 1999; Masai et al . , 2000 ) . Retinal progenitor cells are multipotent and give rise to six major classes of neurons and one type of glial cells . Two types of photoreceptors , rods , and cones , form the outer nuclear layer ( ONL ) . Three interneurons , amacrine cells , bipolar cells , and horizontal cells form the inner nuclear layer ( INL ) . Retinal ganglion cells ( RGCs ) form the RGC layer . Synaptic connections between photoreceptors and bipolar/horizontal cells form the outer plexiform layer ( OPL ) , and synaptic connections between RGCs and bipolar/amacrine cells form the inner plexiform layer ( IPL ) . Cell fate determination is less dependent on the cell lineage of retinal progenitor cells , suggesting that both extrinsic and intrinsic mechanisms influence the status of retinal progenitor multipotency , leading to generation of diverse retinal cell types ( He et al . , 2012 ) . These developmental profiles of retinal neurogenesis and cell differentiation may be coupled with microglial colonization . Although apoptosis and CSF-R signaling are suggested in microglial colonization of the retina in zebrafish ( Wu et al . , 2018 ) , the mechanism underlying microglial colonization of the retina remains to be determined . In this study , using zebrafish , we examined the developmental profile of retinal colonization by microglia precursors . The number of ocular microglial precursors progressively increases from 32 to 54 hpf . Most microglial precursors do not proliferate , suggesting that microglial colonization of the retina depends on cell migration from outside the optic cup . We found three guidance mechanisms driving microglial precursor colonization of the retina . First , IL34 initiates microglial precursor movement from yolk toward the brain and the retina . Second , microglia precursors enter the optic cup via ocular hyaloid blood vessels in the choroid fissure , suggesting that these blood vessels guide microglia to the retina . Third , microglial precursors infiltrate the neural retina preferentially through the neurogenic region , suggesting that the neurogenic state of retinal tissue acts as an entry signal for microglial precursors to infiltrate the retina . Thus , a series of guidance mechanisms promote microglial colonization from yolk to the neural retina in zebrafish .
In zebrafish , early macrophages are generated from myeloid cells originating in the RBI around 11 hpf and these macrophages colonize the brain and retina by 55 hpf ( Xu et al . , 2015 ) . Around 60 hpf , brain and retina-resident macrophages undergo a phenotypic transition , which indicates expression of mature microglial markers , such as apolipoprotein E ( apo E ) and phagocytic behavior toward dead cells ( Herbomel et al . , 2001 ) . Importantly , early macrophages outside the brain never express apo E ( Herbomel et al . , 2001 ) , suggesting that only brain and retina-resident macrophages give rise to microglia . Thus , early macrophages localized in the brain and retina by 60 hpf are generally accepted as microglial precursors in zebrafish . In this study , we focused on two macrophage markers , macrophage expressing gene 1 . 1 ( mpeg1 . 1 ) ( Ellett et al . , 2011 ) and microfibrillar-associated protein 4 ( mfap4 ) ( Walton et al . , 2015 ) , and define mpeg1 . 1; mfap4-positive cells inside the optic cup as microglial precursors colonizing the zebrafish retina . To ascertain how microglia precursors migrate from peripheral tissues into the neural retina during development , we generated a zebrafish transgenic line , Tg[mpeg1 . 1:EGFP] , using the original DNA construct ( Ellett et al . , 2011 ) . As previously reported ( Ellett et al . , 2011 ) , our established transgenic line visualized ocular microglial precursors and enabled us to monitor their number and location in the optic cup from 24 to 54 hpf . Accordingly , we obtained 3D images using confocal laser scanning microscopy ( LSM ) ( Figure 1A and B ) . The first microglial precursor cells appeared near the choroid fissure and lens around 30–32 hpf . After that , the number of ocular microglial precursors increased to 19 . 1 ± 1 . 26 at 42 hpf and 31 . 0 ± 4 . 44 at 54 hpf ( Figure 1C ) , indicating a progressive increase in the number of ocular microglial precursors . We also confirmed a similar progressive increase in the number of ocular L-plastin-positive cells , although L-plastin is expressed in microglial precursors and neutrophils in zebrafish brain ( Figure 1—figure supplement 1 ) . Next , to determine more precisely the spatial distribution of microglial precursors in the optic cup , we generated another transgenic line , Tg[mfap4:tdTomato-CAAX] , using the original DNA construct ( Walton et al . , 2015 ) . As previously reported ( Walton et al . , 2015 ) , our established transgenic line efficiently labeled ocular microglial precursor membranes . We labeled this transgenic embryo using Bodipy ceramide conjugated with fluorescent Alexa-488 , which visualizes retinal layer structures ( Figure 1—figure supplement 2 ) . From 32 to 36 hpf , mfap4+ cells were mostly located in the vitreous space between the neural retina and lens , and possibly associated with ocular blood vessels , which develop around the lens . In 42–44-hpf retina , a few microglial precursor cells start to enter the neural retina and spread toward the emerging IPL , where they are associated with newly born amacrine cells ( Figure 1—figure supplement 3 ) . By 54 hpf , IPL formation is complete and microglial precursors were observed throughout all retinal tissue , except the OPL . Thus , microglial precursors enter the optic cup along the choroid fissure at 30 hpf , remain temporarily in the vitreous space between the lens and the retina , and then begin spreading into differentiating retinal tissue after 42 hpf . Next , to evaluate the contribution of cell proliferation to the increasing number of ocular microglial precursors , we labeled ocular microglial precursors with markers of DNA replication . Here , we used a zebrafish transgenic line , Tg[EF1α: mCherry-zGem] that specifically marks proliferative cells in S and G2 phases ( Mochizuki et al . , 2017; Mochizuki et al . , 2014 ) . We combined this Tg[EF1α: mCherry-zGem] system with Tg[mpeg1 . 1:EGFP] to calculate the fraction of proliferative microglial precursors undergoing S phase ( Figure 1D and Video 1 ) . First , we observed mCherry-zGem; mpeg1 . 1:EGFP double-positive cells in the peripheral tissue ( Figure 1—figure supplement 4A-C ) and found that more than 60 % of mpeg1 . 1:EGFP-positive cells expressed mCherry-zGem ( Figure 1—figure supplement 4D ) , confirming that this Tg[EF1α: mCherry-zGem] system works in early macrophages in zebrafish . However , in the retina , the fraction of mCherry-zGem; mpeg1 . 1:EGFP double-positive cells was less than 2 % of all microglial precursors from 32 to 54 hpf ( Figure 1E ) . Furthermore , more than 80 % of mpeg1 . 1:EGFP-positive cells did not incorporate BrdU at 48 hpf ( Figure 1F–H ) , suggesting that a majority of ocular microglial precursors do not undergo S phase . Thus , microglial colonization of the retina mostly depends on cell migration from outside the optic cup . The zebrafish retina receives its blood supply from two blood vessel systems , intraocular hyaloid blood vessels encapsulating the lens ( Hartsock et al . , 2014 ) and superficial choroidal blood vessels ( Kaufman et al . , 2015 ) . Developing hyaloid blood vessels start to enter the space between the lens and retina through the ventral fissure at 18–20 hpf . Loop formation occurs around the lens at 24–28 hpf , and a branched hyaloid network forms after 35 hpf ( Hartsock et al . , 2014 ) . Our live imaging showed that microglial precursors enter the optic cup through the choroid fissure and remain temporarily in the vitreous space between the lens and the retina before they infiltrate the neural retina ( Figure 1—figure supplement 2 ) . Furthermore , microglial precursors start to enter the optic cup after loop formation of hyaloid blood vessels is completed , suggesting a guiding role of blood vessels in microglial precursor colonization of the optic cup . To confirm whether microglial precursors entering the ocular space are associated with developing hyaloid blood vessels , we conducted time-lapse imaging of Tg[kdrl:EGFP; mfap4:tdTomato-CAAX] transgenic embryos , which visualizes endothelial cells of blood vessels ( Jin et al . , 2005 ) and ocular microglial precursors , respectively . The first microglial precursor was always associated with ocular hyaloid blood vessels around 30 hpf ( Figure 2A ) and moved along blood vessel surfaces ( Video 2 ) , so it is very likely that microglial precursors use blood vessels as a scaffold to enter the vitreous space between the lens and the neural retina . Microglial precursors move along hyaloid blood vessels in the ventral fissure , gradually leave vessel surfaces , and invade the neural retina through the basement membrane ( Figure 2B , Figure 2—figure supplement 1 , and Video 3 ) . Troponin T2A ( tnnt2a; silent heart ) is specifically expressed in heart and is essential for heart contraction ( Sehnert et al . , 2002 ) . In zebrafish brain and mouse retina , hemodynamics drive blood vessel pruning , and loss of blood circulation causes blood vessel regression ( Chen et al . , 2012; Lobov et al . , 2011; Yashiro et al . , 2007 ) . To examine whether the entry of microglial precursors into retina is altered upon blood vessel regression , we blocked blood circulation by injecting morpholino antisense oligos against tnnt2a ( tnnt2a MO ) . When blood circulation is inhibited , ocular hyaloid blood vessels do not develop fully and microglial precursors are less likely to be associated with these thin blood vessels ( Figure 2C ) . The number of ocular microglial precursors was significantly reduced at 36 hpf ( Figure 2D ) , showing that microglial colonization of the optic cup depends upon normal development of the blood vessel network . This is in contrast to the case of microglial colonization of zebrafish midbrain and optic tectum , which is independent of the blood vessel network ( Xu et al . , 2016 ) . Indeed , we confirmed that the number of microglial precursors in the optic tectum was not significantly different between tnnt2a morphants and standard MO-injected embryos at 72 hpf , although microglial precursor colonization of the optic tectum was enhanced in tnnt2a morphants at 48 hpf ( Figure 2—figure supplement 2 ) . Recent studies indicate that microglia facilitate ocular blood vessel development ( Checchin et al . , 2006; Fantin et al . , 2010; Rymo et al . , 2011 ) , and that macrophages initiate a cell-death program in endothelial cells for blood vessel regression in developing mouse retina ( Lang and Bishop , 1993; Lobov et al . , 2005 ) . However , we eliminated microglial precursors with morpholino antisense oligos against pu . 1 ( pu . 1-MO ) or interferon regulatory factor 8 ( irf8 ) mutation ( irf8 gene knockdown causes apoptosis of pu . 1-positive myeloid cells ) ( Shiau et al . , 2015 ) , and confirmed that microglial precursor elimination did not affect hyaloid blood vessel formation in zebrafish at least by 48 hpf ( Figure 2—figure supplement 3 ) . In zebrafish , retinal neurogenesis occurs at the ventronasal retina adjacent to the optic stalk at 25 hpf and propagates into the entire region of the neural retina at 33 hpf ( Masai et al . , 2000 ) . Microglial precursors start to migrate from the vitreous space into the neural retina after 42 hpf , when the earliest differentiating retinal neurons , RGCs , start to form the IPL ( Mumm et al . , 2006 ) . To examine the role of retinal neurogenesis and RGC differentiation in microglia precursor infiltration of the neural retina , we used double transgenic lines , Tg[EF1α: mCherry-zGem; mpeg1 . 1:EGFP] , which enable us to examine the relationship between microglial precursor migration and retinal progenitor cells ( Mochizuki et al . , 2014 ) . Live imaging of Tg[EF1α: mCherry-zGem; mpeg1 . 1:EGFP] retinas at 42 and 48 hpf clearly showed that microglial precursors avoid mCherry-zGem-positive proliferating regions and are preferentially positioned in the region of mCherry-zGem-negative post-mitotic cells ( Figure 3A–B , Figure 3—figure supplement 1 ) . The fraction of microglial precursors that infiltrated mCherry-zGem-positive proliferating regions was 7 . 37 % at 42 hpf and 6 . 13 % at 48 hpf ( Figure 3C ) , suggesting that >90% of microglial precursors infiltrate the retina through the mCherry-zGem-negative post-mitotic cell region . We also used another transgenic line Tg[ath5:EGFP; mfap4:tdTomato-CAAX] . In the Tg[ath5:EGFP] line , EGFP starts to be expressed in G2 phase of the final neurogenic cell division of retinal progenitor cells and is inherited by their daughter cells , which are negative for BrdU incorporation ( Poggi et al . , 2005; Yamaguchi et al . , 2010 ) , suggesting that ath5:EGFP specifically marks early differentiating retinal neurons . We conducted live imaging of Tg[ath5:EGFP; mfap4:tdTomato-CAAX] retinas at 36 , 42 , 48 hpf , and found that infiltration of mfap4-positive microglia preferentially occurs in the ath5:EGFP-positive region ( Figure 3D ) . These data suggest that microglial precursors infiltrate the neural retina preferentially through the neurogenic area , raising the possibility that the neurogenic retinal region acts as a gateway through which microglial precursors move from the vitreous space into the neural retina . Colonization of the optic tectum by microglial precursors depends on neuronal apoptosis in zebrafish ( Casano et al . , 2016; Xu et al . , 2016 ) . Therefore , it is still possible that microglial precursors preferentially infiltrate the neural retina through the neurogenic region , because of neuronal apoptosis . We inhibited retinal apoptosis by injecting morpholino antisense oligos against p53 ( p53 MO ) and confirmed that p53 MO effectively suppresses retinal apoptosis at 24 and 36 hpf ( Figure 3—figure supplement 2 ) . However , the number of microglial precursors did not differ between p53 morphant retinas and standard-MO-injected retinas at 48 hpf ( Figure 3—figure supplement 3A-B ) , whereas the number of microglial precursors was significantly decreased in p53 morphant optic tectum compared with standard-MO-injected optic tectum at 96 hpf ( Figure 3—figure supplement 3C-D ) . Thus , in contrast to microglial colonization of the optic tectum , neuronal apoptosis is not the major cue for microglial precursor colonization of the retina , at least prior to 54 hpf . To confirm the possibility that the neurogenic retinal region functions as a gateway for microglial precursors to infiltrate the retina , we examined whether microglial precursor migration into the retina is compromised when retinal neurogenesis is affected . Previously , we found that histone genesis slowed in zebrafish stem loop binding protein 1 ( slbp1 ) mutants , leading to severe delays in retinal neurogenesis ( Imai et al . , 2014 ) . Our bulk RNAseq analysis confirmed that retinal neurogenesis and subsequent neuronal differentiation were markedly delayed in zebrafish slbp1 mutants ( Figure 4—figure supplement 1 ) , such that ath5 expression spread into the entire slbp1 mutant retina only at 48 hpf , an event that occurs in wild-type retina at 33 hpf ( Imai et al . , 2014 ) . We combined slbp1 mutants with transgenic lines Tg[ath5:EGFP; mfap4: tdTomato-CAAX] and examined the number of ocular mfap4:tdTomato-CAAX-positive microglial precursors ( Figure 4A , Figure 4—figure supplements 2A and 3 ) . In 48-hpf slbp1 mutant retinas , the number of ocular microglial precursors was 4 . 67 ± 2 . 42 ( Figure 4A and B ) , which is similar to the number in wild-type retinas at 32 hpf ( Figure 1B ) , whereas the number of ocular microglial precursors in wild-type siblings was 17 . 60 ± 5 . 13 at 48 hpf ( Figure 4A and B ) . To confirm whether the slbp1 mutation interferes with genesis of early macrophages , we examined peripheral mfap4+ cells in the tail/trunk region of slbp1 mutants and wild-type sibling embryos . There was no significant difference in mfap4+ cells between slbp1 mutants and wild-type siblings in the trunk/tail region ( Figure 4C and D ) , indicating that the slbp1 mutation does not influence early macrophage specification in zebrafish embryos . Although inhibition of retinal apoptosis by p53 MO does not influence microglial precursor colonization of the retina ( Figure 3—figure supplement 3A-B ) , we examined the level of retinal apoptosis in slbp1 mutants . TUNEL revealed that apoptosis was increased in slbp1 mutant retinas compared with wild-type sibling retinas ( Figure 4—figure supplement 4 ) . These data exclude the possibility that decreased retinal apoptosis affects microglial precursor colonization of the retina in slbp1 mutants , and again confirm that neuronal apoptosis is not the major cue for microglial precursor colonization of the retina . Mouse brain cortex colonization by microglia depends on the Cxcl12a-CxcR4 signaling axis ( Arnò et al . , 2014 ) . We previously reported that cxcl12a expression is absent in the optic stalk of zebrafish slbp1 mutants ( Imai et al . , 2014 ) . To exclude the possibility that the absence of cxcl12a expression in the optic stalk affects microglial colonization of the retina in zebrafish slbp1 mutants , we examined zebrafish cxcl12a morphants . Injection of cxcl12a-MO at 500 μM , which effectively induces RGC axon trajectory defects reported in zebrafish odysseys mutants carrying mutations in cxcl12a receptor , cxcr4b ( Li et al . , 2005 ) , did not affect the number of ocular microglial precursors ( Figure 4—figure supplement 5 ) . Thus , Cxcl12a-CxcR4 signaling is not involved in microglial colonization defects in slbp1 mutants . We also confirmed that elimination of microglial precursors with pu . 1 MO did not affect the rate of retinal neurogenesis or cell differentiation by 72 hpf ( Figure 4—figure supplement 6 ) . We previously showed that overexpression of Notch1 intracellular domain ( NICD ) suppresses retinal neurogenesis in zebrafish ( Yamaguchi et al . , 2005 ) . We confirmed that overexpression of NICD suppresses retinal neurogenesis in zebrafish by injecting a DNA expression construct encoding UAS:myc-NICD ( Scheer and Campos-Ortega , 1999 ) into Tg[hsp:gal4; ath5:EGFP] double transgenic embryos ( Figure 4—figure supplement 7 ) . Next , we examined whether microglial precursor infiltration of the retina is compromised in retinas overexpressing NICD . We established a zebrafish transgenic line , Tg[rx1:gal4-VP16] , which expresses Gal4-VP16 under control of a retinal progenitor-specific promoter rx1 ( Chuang et al . , 1999 ) , and then injected two DNA expression constructs encoding UAS:EGFP ( Köster and Fraser , 2001 ) and UAS:myc-NICD into Tg[rx1:gal4-VP16; mfap4:tdTomato-CAAX] double-transgenic embryos . Embryos injected with only the DNA construct of UAS:EGFP served as a positive control . We selected embryos in which EGFP was expressed in most retinal cells at 24 hpf and used them for further analysis . The number of ocular microglial precursors was significantly reduced in embryos overexpressing NICD and EGFP , compared with control embryos overexpressing EGFP , at 44 hpf ( Figure 4E and F , Figure 4—figure supplements 2B and 3 ) . These data support the possibility that the retinal neurogenic region functions as a gateway for microglia to infiltrate the retina . The blockade of retinal neurogenesis delays differentiation of the first-born retinal cell-type , RGCs . To examine whether blockade of RGC differentiation affects microglial precursor colonization of the neural retina , we applied an antisense morpholino against ath5 ( known as atoh7 ) ( ath5 MO ) . As with the zebrafish ath5 mutant , lakritz ( Kay et al . , 2001 ) , RGC differentiation was specifically inhibited in ath5 morphant retinas ( Figure 4—figure supplement 8A , B ) . In ath5 morphants , the timing of the first appearance of microglial precursors in the ocular vitreous space was not altered , but the number of ocular microglial precursors was significantly decreased at 49 hpf ( Figure 4G and H , and Figure 4—figure supplements 2C and 3 and 8 C ) , suggesting that RGC differentiation or RGC-mediated IPL formation is required for microglial precursor infiltration into the neural retina . To determine whether microglia precursors have greater affinity for differentiating neurons than for retinal progenitor cells , we carried out two sets of experiments . First , we conducted cell transplant experiments using a wild-type donor line and an slbp1 mutant recipient line carrying Tg[mfap4:tdTomato-CAAX] . Wild-type donor cells were transplanted into slbp1 mutant recipient embryos at the blastula stage . We selected slbp1 mutant and wild-type sibling embryos in which wild-type , donor retinal cell columns were introduced in a mosaic manner at 48 hpf ( Figure 5A ) . Host microglial precursors and donor retinal cells were visualized with mfap4:tdTomato-CAAX and Alexa-488 Dextran , respectively . In slbp1 mutant host retinas , microglial precursors were likely to be associated with donor wild-type retinal columns more frequently than in wild-type host retinas ( Figure 5B ) . To analyze these data statistically , we compared eyes in which wild-type donors were transplanted into wild-type hosts with those in which wild-type donors were transplanted into slbp1 mutant hosts ( Figure 5—figure supplement 1A-B ) . The fraction of microglial precursors associated with donor wild-type retinal columns in total ocular microglial precursors was significantly higher in slbp1 mutant host retinas than in wild-type sibling host retinas at 48 hpf ( Figure 5C ) , suggesting that microglial precursors are more attracted by wild-type donor neurogenic retinal columns than surrounding slbp1 mutant proliferative retinal cells . Since the fraction of microglial precursors associated with donor retinal columns in total microglial precursors may depend on the number of donor retinal columns incorporated into host retinas , we next estimated trapping efficiency of microglial precursors per donor column by dividing the fraction of microglial precursors associated with donor columns with the transplanted donor column number in each eye ( Figure 5—figure supplement 1B ) . Trapping efficiency of microglial precursors per donor column was significantly higher in slbp1 mutant host retinas than in wild-type sibling host retinas ( Figure 5D ) , suggesting that microglial precursors are preferentially associated with donor-derived wild-type retinal cells than with host-derived slbp1 mutant retinal cells . Second , we injected two DNA constructs encoding UAS:EGFP and UAS:mycNICD into Tg[hsp:gal4; mfap4:tdTomato-CAAX] double-transgenic wild-type embryos . Two rounds of heat-shock treatment at 18 and 30 hpf induced expression of NICD and EGFP in a mosaic manner in the retina ( Figure 5E ) . We examined the fraction of mfap4:tdTomato-CAAX-positive microglial precursors associated with EGFP-expressing retinal columns in the total number of mfap4:tdTomato-CAAX-positive microglial precursors ( Figure 5F ) . This fraction was significantly lower in retinas overexpressing NICD and EGFP than in control retinas overexpressing only EGFP ( Figure 5G , Figure 5—figure supplement 1C ) . We also confirmed that trapping efficiency of mfap4:tdTomato-CAAX-positive microglial precursors per EGFP-positive retinal column was significantly lower in retinas overexpressing NICD and EGFP than in retinas overexpressing only EGFP ( Figure 5H , Figure 5—figure supplement 1C ) . Thus , microglial precursors are less attracted by retinal columns in which neurogenesis is arrested . Taken together , these data suggest that microglial precursors preferentially associate with neurogenic retinal columns as opposed to proliferative retinal columns . Recently , it was reported that microglial colonization of zebrafish brain , including retina , depends on CSF-R , and that one of the CSF-R ligands , IL34 , dominates this process ( Wu et al . , 2018 ) . In adult mouse retina , RGCs express IL34 , which attracts one subset of microglia and retains them around the IPL niche ( O’Koren et al . , 2019 ) . First , we confirmed that retinal cell differentiation proceeds normally until 72 hpf in zebrafish il34 mutants , although pyknotic nuclei were stochastically observed in RGC and amacrine cell layers ( Figure 6—figure supplement 1 ) . Next , we examined microglial precursor colonization of the retina . The number of ocular microglial precursors was significantly lower in il34 homozygous mutants than in wild-type siblings at 34 hpf ( Figure 6—figure supplement 2 ) and 48 hpf ( Figure 6A and B ) . Thus , consistent with the previous report ( Wu et al . , 2018 ) , IL34 is required for microglial precursor colonization of the retina in zebrafish . However , il34 mRNA expression is comparable in slbp1 mutant heads and wild-type sibling heads at 48 hpf ( Figure 6—figure supplement 3 ) , suggesting that il34 mRNA expression is not linked to retinal neurogenesis . Since the number of ocular microglial precursors in il34 homozygous mutants was zero at 34 hpf ( Figure 6—figure supplement 2 ) and no more than two , if any , at 48 hpf ( Figure 6B ) , it is likely that Csf1r-il34 signaling promotes microglial precursor movement from yolk to the optic cup upstream of the blood-vessel-mediated guidance mechanism ( Figure 6C ) .
In zebrafish , primitive microglia originate from the RBI , which is a hematopoietic tissue equivalent to mouse yolk sac , whereas definitive microglia are generated from hematopoietic stem cells that are specified in the VDA ( Ferrero et al . , 2018; Xu et al . , 2015 ) . Primitive and definitive waves of hematopoiesis generate embryonic and adult microglia , respectively . Using zebrafish as an animal model , several groups investigated microglial colonization from the periphery into developing brain , especially the optic tectum , which is part of the midbrain ( Casano et al . , 2016; Herbomel et al . , 2001; Svahn et al . , 2013; Wu et al . , 2018; Xu et al . , 2016 ) . Colonization of the optic tectum by microglial precursors depends on neuronal apoptosis , probably through attraction by an apoptotic cell-secreted phospholipid , lysophosphatidylcholine ( LPC ) ( Casano et al . , 2016; Xu et al . , 2016 ) . In addition , microglial colonization of brain is CSF receptor-dependent ( Herbomel et al . , 2001; Wu et al . , 2018 ) . In mice , microglial colonization of brain requires functional blood circulation ( Ginhoux et al . , 2010 ) . However , in zebrafish , microglial colonization of the optic tectum is independent of blood circulation ( Xu et al . , 2016 ) . A series of elegant studies revealed the molecular network that promotes microglial colonization of the midbrain . However , it remains to be seen whether this mechanism fully explains colonization of other brain regions by microglial precursors . In this study , we focused on zebrafish retina and investigated the mechanism that regulates migration of embryonic microglial precursors into developing retina . We first conducted live imaging of zebrafish microglial precursors from 24 to 54 hpf . Microglial precursors progressively increase in number during embryonic development . Interestingly , almost all microglial precursors enter the optic cup through the choroid fissure . However , peripheral macrophages located in the mesenchymal region between the eye and the brain did not enter the optic cup across the ciliary marginal zone . This may be consistent with the observation that these peripheral macrophages never enter the retina following rod cell death ( White et al . , 2017 ) , suggesting a functional difference between peripheral macrophages and ocular microglia . Next , we found that the majority of ocular microglial precursors do not undergo S phase and are probably in G1 phase . Thus , the increase of ocular microglial precursors is due to migration from outside the eye . In developing mouse retina , microglial precursors appear from the vitreous area near the optic disk at E11 . 5 , progressively increase in number , and then infiltrate the neural retina . These retinal microglia were also negative for a proliferative marker , Ki67 ( Santos et al . , 2008 ) , suggesting that mouse embryonic retinal microglia are also non-proliferative . Another interesting finding is that entry of microglial precursors into the optic cup through the choroid fissure depends on ocular blood vessels . We observed that migrating microglial precursors are closely associated with hyaloid blood vessels after loop formation . These microglial precursors pass along these vessels , which traverse the choroid fissure and surround the posterior region of the lens . Furthermore , the number of ocular microglial precursors was reduced when blood circulation was blocked . Since inhibition of blood circulation compromises the structural integrity of blood vessels in zebrafish , we conclude that ocular blood vessel formation is required for microglial precursor entry into the optic cup through the choroid fissure . One possibility is that blood vessels function as a path upon which microglial precursors enter the optic cup . Membrane proteins or extracellular matrix proteins on blood endothelial cells may facilitate the association of microglial precursors with blood vessel surfaces . Alternatively , substances that attract microglial precursors may be released from hyaloid blood endothelial cells . Previous studies on human and murine microglia demonstrated that microglial colonization of the retina takes place prior to retinal vascularization , and that microglia facilitate ocular blood vessel development ( Checchin et al . , 2006; Fantin et al . , 2010; Rymo et al . , 2011 ) . Macrophages initiate endothelial cell death for blood vessel regression in developing mouse retina ( Lang and Bishop , 1993; Lobov et al . , 2005 ) . However , in contrast to mammals , elimination of microglia by pu . 1 MO or irf8 mutation did not affect ocular blood vessel formation in zebrafish , suggesting that microglia do not regulate ocular blood vessel formation in zebrafish . Interestingly , classic histological studies on mouse retinas showed that early emerging ocular microglia are associated with the hyaloid artery ( Hume et al . , 1983; Santos et al . , 2008 ) , which is located in the vitreous area and regresses in later stages before retinal vasculature formation ( Ito and Yoshioka , 1999 ) . Thus , further investigation will be necessary to determine whether the hyaloid artery guides microglial precursors into the optic cup in vertebrate species such as mice . In zebrafish , colonization of the optic tectum by microglia is independent of blood circulation ( Xu et al . , 2016 ) . We confirmed that the number of microglia in the optic tectum did not differ between tnnt2a morphants and control embryos at 72 hpf; however , microglial colonization of the optic tectum was enhanced and microglial shape was round rather ramified in tnnt2a morphants at 48 hpf . Further study will be necessary to clarify the role of blood circulation in microglial colonization of the optic tectum . After 42 hpf , microglial precursors detach from hyaloid blood vessels and start to infiltrate the neural retina . Interestingly , we found that more than 90 % of microglial precursors enter the neural retina through the neurogenic area . Indeed , the number of microglial precursors is reduced in slbp1 mutant retinas and NICD-overexpressing retinas , in both of which retinal neurogenesis is severely delayed . Furthermore , we conducted two sets of experiments: the first was cell transplantation from wild-type donor cells into slbp1 mutant host retinas , which introduced neurogenic wild-type retinal cell columns in proliferative slbp1 mutant retinas , and the second was overexpression of NICD in wild-type retina , which introduced proliferative retinal cell columns in neurogenic retinas . Consistently , in both cases , microglial precursors were preferentially associated with neurogenic retinal cell columns . Thus , neurogenesis is required for infiltration of microglial precursors into the neural retina after 42 hpf . We observed that the number of microglial precursors is diminished in ath5 morphant retinas , suggesting that RGCs are required for infiltration of microglial precursors into the neural retina . There are at least three possible mechanisms for this infiltration . First , the basal region of retinal progenitor cells may function as a physical barrier that inhibits microglial precursor infiltration of the neural retina . Second , microglial precursors may be attracted to surfaces of differentiating retinal neurons or RGCs . Third , differentiating retinal neurons or RGCs may release a specific attractant for microglia . There are several candidate molecules that suggest the third possibility . In adult mice , RGCs express IL34 , which attracts microglia and retains them around the IPL niche ( O’Koren et al . , 2019 ) . Indeed , microglial colonization of zebrafish brain depends on CSF-R , and one of the CSF-R ligands , IL34 , dominates this process ( Wu et al . , 2018 ) . We confirmed that microglial precursor colonization of retina is severely affected in il34 mutants . However , il34 mRNA expression is comparable in slbp1 mutants and their wild-type siblings , suggesting that IL34 is not linked to neurogenesis-mediated microglial precursor infiltration . Rather , the number of ocular microglial precursors in il34 mutants was almost zero at 34 and 48 hpf , so it is very likely that Csf1r-il34 signaling initiates microglial precursor movement from yolk toward brain and retina , followed by blood vessel- and neurogenesis-mediated guidance . It was reported that apoptosis attracts microglia in zebrafish developing brain ( Casano et al . , 2016; Xu et al . , 2016 ) . However , microglial precursor colonization of the retina is normal in zebrafish p53 morphants , suggesting that apoptosis does not promote microglial precursor colonization of the retina . Why are microglial precursors insensitive to retinal apoptosis ? We found that apoptosis is enhanced in zebrafish slbp1 mutant retinas , in which microglial precursor colonization is severely affected due to a delay of retinal neurogenesis . It is likely that spontaneous apoptotic cells fail to be eliminated because of the reduced number of microglial precursors in slbp1 mutant retinas; however , interestingly , these increased dead cells did not promote microglial precursor infiltration into slbp1 mutant retinas , suggesting that neurogenesis primarily opens the gate through which microglial precursors enter the neural retina . Since retinal neurogenesis normally occurs from 24 to 48 hpf in zebrafish , microglial precursors could not be attracted by apoptosis without the infiltration path opened by neurogenesis before 48 hpf . Further studies will be necessary to unveil the molecular mechanism underlying microglial infiltration into neural retina . In summary , there are three mechanisms for microglial colonization of developing zebrafish retina ( Figure 6C ) . IL34-CSF-R signaling initiates microglial precursor movement from yolk toward brain and retina . Microglial precursors further use ocular hyaloid blood vessels as a pathway to enter the optic cup and then infiltrate the neural retina preferentially through the neurogenic region . In the future , it remains to identify molecules involved in blood-vessel- and neurogenesis-mediated guidance mechanisms , and to assess whether these mechanisms are used for microglial colonization of other brain regions in other vertebrate species .
Zebrafish ( Danio rerio ) were maintained using standard procedures ( Westerfield , 1993 ) . RIKEN wako ( RW ) was used as a wild-type strain for mutagenesis ( Masai et al . , 2003 ) . slbp1rw440 ( Imai et al . , 2014 ) , irf8st96 ( Shiau et al . , 2015 ) and il34hkz11 ( Wu et al . , 2018 ) were used . Transgenic lines Tg[ath5:EGFP]rw021 were used to monitor ath5 gene expression ( Masai et al . , 2005 ) . Tg[EF1α:mCherry-zGem]oki011 ( Mochizuki et al . , 2014 ) was used for visualization of cell-cycle phases . Tg[mfap4:tdTomato-CAAX]oki058 and Tg[mpeg1 . 1:EGFP]oki053 were used to visualize microglial precursors . Tg[kdrl:EGFP]s843Tg was employed to visualize blood vessels ( Jin et al . , 2005 ) . Tg[hsp:gal4]kca4 ( Scheer et al . , 2002 ) and Tg[rx1:gal4-VP16]oki065 were used for UAS-mediated expression of target genes . For confocal scanning , embryos were incubated with 0 . 003 % phenyltiourea ( PTU ) ( Nacalai tesque , 27429–22 ) to prevent melanophore pigmentation . The zebrafish pigmentation mutant , roy orbison ( roy ) ( D’Agati et al . , 2017 ) was used to remove iridophores . The DNA construct encoding mpeg1 . 1:EGFP was kindly provided by Dr . Graham Lieschke and we are indebted to Dr . David Tobin for the construct encoding mfap4:tdTomato-CAAX . These DNA constructs were injected into fertilized eggs with Tol2 transposase mRNA , to establish transgenic lines , Tg[mpeg1 . 1:EGFP] and Tg[mfap4:tdTomato-CAAX] in our lab . Plastic sectioning and immunolabeling of cryosections were carried out as previously described ( Masai et al . , 2003 ) . Anti-GFP ( Themo Fisher Scientific , A11122 ) , anti-myc-tag ( Invitrogen , R950-25 ) , zn5 ( Oregon Monoclonal Bank ) and zpr1 ( Oregon Monoclonal Bank ) antibodies were used at 1:200; 1:250 , 1:100 , and 1:100 dilutions , respectively . For detection of BrdU incorporation , BrdU ( Nacalai , tesque , 05650–95 ) was applied to 52-hpf embryos , chased for 2 hr at 28 . 5 °C and fixed with 4 % paraformaldehyde ( PFA ) . Labeling of retinal sections with anti-BrdU antibody ( BioRad , MCA2060 ) was carried out as previously described ( Yamaguchi et al . , 2005 ) . TUNEL was performed using an In Situ Cell Death Detection Kit ( Roche , 11684795910 ) . Bodipy-ceramide ( Thermo Fisher Scientific , B22650 ) was applied to visualize retinal layers as previously described ( Masai et al . , 2003 ) . Nuclear staining was performed using 1 nM TOPRO3 ( Thermo Fisher Scientific , T3605 ) . Morpholino antisense oligos were designed as shown below . Morpholino antisense oligos were injected into fertilized eggs at 500 µM for tnnt2a MO and cxcl12a MO; 250 µM for ath5 MO and pu . 1 MO and 100 µM for p53 MO . The same concentration was used for Standard MO in each MO experiment . Cell transplantation was performed as previously described ( Masai et al . , 2003 ) . Wild-type zygotes were injected with Alexa-488 dextran ( Thermo Fisher Scientific , D22910 ) and used for donor embryos . slbp1 mutant embryos carrying Tg[mfap4:tdTomato-CAAX] were used as host embryos . Host embryos carrying donor retinal cells were selected by observing Alexa 488 fluorescence at 24 hpf . slbp1 mutant and wild-type sibling embryos were sorted based on the slbp1 mutant morphological phenotype at 48 hpf and used for live imaging . After confocal images were obtained , the number of ocular mfap4-positive microglial precursors associated with Alexa-488 dextran-labeled donor transplanted retinal columns was counted . The fraction of ocular mfap4-positive microglial precursors associated with donor transplanted retinal columns in total ocular microglial precursors was calculated . The trapping efficiency of ocular mfap4-positive microglial precursors per transplanted donor retinal column was calculated using the total number of donor transplanted retinal columns in the retina . Detailed information on each transplanted eye is shown in Figure 5—figure supplement 1A-B . Transgene lines Tg[mpeg1 . 1:EGFP] or Tg[mfap4:tdTomato-CAAX] , and Tg[kdrl:EGFP] , were used for time-lapse imaging of microglial precursors and blood vessels . 3D confocal images were obtained using a confocal LSM , LSM710 ( Zeiss ) or an FV3000RS ( Olympus ) , and analyzed using ImageJ ( 2 . 0 . 0-rc-69/1 . 52 p ) and Imaris software ( ver . 9 . 1 . 2 Bitplane ) . The DNA construct encoding Ptf1a:EGFP was used for visualizing amacrine cells or their progenitors ( Jusuf and Harris , 2009 ) . Heads of 48 hpf wild-type sibling and slbp1 mutant embryos were dissected and transferred to 100 μL Sepasol ( Nacalai tesque , 09379 ) on ice . Heads were then homogenized using a hand homogenizer ( ~20 pulses ) . Twenty μL CHCl3 were then added to samples and mixed gently . After centrifugation at 15 , 000 g for 15 min , the aqueous phase was collected and mixed with 100 μL isopropanol . One μL of RNase-free glycogen ( Nacalai tesque 11170–11 , 20 mg/mL ) was added to all samples to increase the yield . After incubating at room temperature for 10 min , samples were centrifuged at 15 , 000 g at 4 °C for 15 min . Supernatant was removed and the pellet was washed three times with 500 μL of 75 % ethanol and centrifuged at 8000 g at 4 °C . The pellet was then resuspended in a desired amount of nuclease-free water and stored at –80 °C . RNA concentration and purity of samples were determined using a Nanodrop . RNA samples with RIN >7 were subjected to paired-end sequencing using an Illumina HiSeq4000 . First , a quality check was performed using FastQC and read trimming was done with Trimomatic ( Bolger et al . , 2014 ) . PRINSEQ lite ( Schmieder and Edwards , 2011 ) was used for PolyA trimming and quality filtering . Trimmed sequences were then mapped to the zebrafish reference genome ( GRCz11 ) using hisat2 . 1 . 0 ( Kim et al . , 2019 ) and mapped reads are counted using featureCounts ( Liao et al . , 2014 ) . With the R package , EdgeR ( Robinson et al . , 2010 ) , differentially expressed genes with Log2FC > |2| and FDR values < 0 . 01 were extracted . EnhancedVolcano package ( Blighe et al . , 2018 ) was used to draw volcano plots . A heat map was generated with the pheatmap package ( Kolde , 2019 ) . Extracted RNA from 48-hpf wild-type sibling and slbp1 mutant heads was used to prepare cDNA , using ReverTra Ace qPCR RT master mix with gDNA remover ( Toyobo , FSQ-301 ) . The expression level of il34 mRNA was evaluated with quantitative PCR using the primers below . mRNA of cytoplasmic actin β2 , namely actb2 ( ZFIN ) , was used for normalization . The DNA fragment that covers a 2892 bp genomic region upstream from the start codon of rx1 cDNA ( Chuang et al . , 1999 ) , was amplified by PCR and inserted between XhoI and BamHI sites of the Tol2 base expression vector , pT2AL200R150G ( Urasaki et al . , 2006 ) . Next , DNA fragments encoding gal4-VP16 ( Köster and Fraser , 2001 ) were further inserted between BamHI and ClaI sites of pT2AL200R150G to fuse the rx1 promoter . The plasmid was injected with Tol2 transposase mRNA into fertilized eggs of the UAS:EGFP transgenic line to establish a transgenic line , Tg[rx1:gal4-VP16]oki065 . A mixture of plasmids of UAS:EGFP ( Köster and Fraser , 2001 ) and UAS:myc-NICD ( Scheer and Campos-Ortega , 1999 ) ( each 10 ng/μL ) were injected into fertilized eggs of the Tg[mfap4:tdTomato-CAAX; rx1:gal4-VP16] or Tg[mfap4:tdTomato-CAAX; hsp:gal4] transgenic line . In the case of the Tg[mfap4:tdTomato-CAAX; hsp:gal4] transgenic line , two rounds of heat shock at 37 °C for 1 hr were applied at 18 and 30 hpf . Embryos expressing EGFP in the optic cup were selected at 24 hpf , fixed with PFA at 48 hpf and used to prepare serial retinal sections for imaging analysis . After confocal images were obtained , the number of ocular mfap4-positive microglial precursors associated with EGFP-expressing columns was counted . The fraction of ocular mfap4-positive microglial precursors associated with EGFP-expressing columns in total microglial precursors was calculated and the trapping efficiency of ocular mfap4-positive microglial precursors per EGFP-expressing column was calculated using the total number of EGFP-expressing columns in the retina . Detailed information on each injected eye is shown in Figure 5—figure supplement 1C . To confirm that NICD inhibits retinal neurogenesis , UAS:myc-NICD or UAS:mCherry ( each 10 ng/μL ) was injected into zebrafish transgenic embryos Tg[ath5:EGFP; hsp:gal4] . Three rounds of heat shock at 37 °C for 1 hr were applied at 18 , 24 , and 30 hpf . Embryos were fixed at 36 hpf and labeled with anti-myc tag antibody to visualize myc-NICD expressing retinal cells with Alexa-543-conjugated secondary antibody . Whole retinas were used for confocal scanning with an FV3000RS ( Olympus ) . Controls were UAS:mCherry-injected samples and used directly for live confocal scanning . Confocal 3D retinal images were used to count the number of ath5:EGFP-positive and negative retinal columns in myc-NICD or mCherry expressing retinal columns from five independent embryos . The il34hkz11 allele ( Wu et al . , 2018 ) was combined with the Tg[mfap4:tdTomato] transgenic line and used for analysis . Embryos were generated by pair-wise crosses between heterozygous mutant male and female fish , and maintained with N-phenyl thiourea ( PTU ) -containing water to prevent melanophore pigmentation . Whole retinas of 19 embryo at 34 hpf and 29 embryos at 48 hpf were scanned with confocal microscopy , using an LSM710 ( Zeiss ) or an FV3000RS ( Olympus ) . Embryos were fixed with 4 % PFA and used for genotyping . A DNA fragment containing the 4-base deletion mutation of the il34hkz11 allele was amplified by PCR and sequenced to determine genotypes . Primers used for PCR amplification and sequencing are below . Using the surface rendering tool of Imaris software ( Bitplane , ver . 9 . 1 . 2 ) , we eliminated signals of iridophore-derived noise or peripheral macrophages around the optic cup and extracted only ocular microglial precursors . The number of ocular microglial precursors was counted in each retina and compared between genotypes . Statistical analyses were performed using GraphPad Prism version 8 . 2 . 1 . Statistical significance was determined using two-tailed unpaired Student’s t-tests for Figure 1G; Figure 2D; Figure 4B , D , H; Figure 5C , D , G , H; Figure 1—figure supplement 2; Figure 2—figure supplement 2B; Figure 3—figure supplement 3B-D; Figure 4—figure supplement 4B; Figure 4—figure supplement 5D; Figure 6—figure supplement 3 , Tukey’s multiple comparison test for Figure 4F; Figure 6B; Figure 6—figure supplement 2B , and Bonferroni’s multiple comparison test for Figure 3—figure supplement 2B . Chi square tests were used for Figure 4—figure supplement 7C . Detailed information on each dataset is provided in Excel files in Raw data . | The immune system is comprised of many different cells which protect our bodies from infection and other illnesses . The brain contains its own population of immune cells called microglia . Unlike neurons , these cells form outside the brain during development . They then travel to the brain and colonize specific regions like the retina , the light-sensing part of the eye in vertebrates . It is poorly understood how newly formed microglia migrate to the retina and whether their entry depends on the developmental state of nerve cells ( also known as neurons ) in this region . To help answer these questions , Ranawat and Masai attached fluorescent labels that can be seen under a microscope to microglia in the embryos of zebrafish . Developing zebrafish are transparent , making it easy to trace the fluorescent microglia as they travel to the retina and insert themselves among its neurons . Ranawat and Masai found that blood vessels around the retina act as a pathway that microglia move along . Once they reach the retina , the microglia remain attached and only enter the retina at sites where brain cells are starting to mature in to adult neurons . Further experiments showed that microglia fail to infiltrate and colonize the retina when blood vessels are damaged or neuron maturation is blocked . These findings reveal some of the key elements that guide microglia to the retina during development . However , further work is needed to establish the molecular and biochemical processes that allow microglia to attach to blood vessels and detect when cells in the retina are starting to mature . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2021 | Mechanisms underlying microglial colonization of developing neural retina in zebrafish |
How producers of public goods persist in microbial communities is a major question in evolutionary biology . Cooperation is evolutionarily unstable , since cheating strains can reproduce quicker and take over . Spatial structure has been shown to be a robust mechanism for the evolution of cooperation . Here we study how spatial assortment might emerge from native dynamics and show that fluid flow shear promotes cooperative behavior . Social structures arise naturally from our advection-diffusion-reaction model as self-reproducing Turing patterns . We computationally study the effects of fluid advection on these patterns as a mechanism to enable or enhance social behavior . Our central finding is that flow shear enables and promotes social behavior in microbes by increasing the group fragmentation rate and thereby limiting the spread of cheating strains . Regions of the flow domain with higher shear admit high cooperativity and large population density , whereas low shear regions are devoid of life due to opportunistic mutations .
Cooperation is the cement of biological complexity . A combined investment brings larger returns . However , while cooperating populations are fitter , individuals have evolutionary incentive to cheat by taking advantage of available public goods without contributing their own . Avoiding the cost of these goods allow larger reproduction rates , causing cheaters to proliferate until the lack of public goods compromise the fitness of the entire population . In other words , while cooperating populations are fitter than non–cooperating ones , cooperation is not evolutionarily stable . How then can social behavior emerge and persist in microbial colonies ? The evolution of cooperation is an active field of research , with multiple theories resolving this dilemma ( Axelrod and Hamilton , 1981; Sachs et al . , 2004; Sachs and Simms , 2006; Nowak , 2006 ) . According to ( Fletcher and Doebeli , 2009 ) the fundamental mechanism is assortment . That is , in order for cooperation to evolve , cooperators and cheaters must experience different interaction environments . How this assortment is achieved is a central question . Possibilities include positive and negative reciprocity ( Trivers , 1971; Clutton-Brock and Parker , 1995; El Mouden et al . , 2010 ) , where cooperators are rewarded later by others , or where cheaters are inflicted a cost , via policing or reputation . For example , quorum signals reveal whether available public goods add up to the population density . In this case , altruists cut back public good production to eliminate cheaters ( albeit with collateral damage ) ( Allen et al . , 2016; Sandoz et al . , 2007; Diggle et al . , 2007 ) . Another idea is group selection ( Wynne-Edwards , 1962; Haldane , 1932; Traulsen and Nowak , 2006; Wilson , 1975 ) and its modern incarnation , multi-level selection , ( Wilson and Sober , 1994 ) which propose that cooperating groups ( or groups of groups ) will reproduce faster than non-cooperating ones and prevail . Kin-selection theory ( Hamilton , 1964a , 1964b; Williams , 1966; Smith , 1964; Hamilton , 1975; Lion et al . , 2011 ) provides a mechanism that arises from individual level dynamics . Kin-selection proposes that individuals cooperate with those to which they are genetically related , and thus , a cooperative genotype is really cooperating with itself . Hamilton conjectured that kin selection should promote cooperation if the population is viscous , that is when the mobility of the population is limited ( Hamilton , 1964a; Hamilton , 1964b ) . This helps ensure that genetically related individuals cooperate with each other . However , competition within kin can inhibit altruism ( Taylor , 1992; Wilson et al . , 1992 ) . One solution to this is if individuals disperse as groups , also known as budding dispersal . This was shown to promote cooperation theoretically by Gardner and West ( 2006 ) and demonstrated experimentally by ( Kümmerli et al . , 2009 ) . Budding dispersal has also been studied by ( Pollock , 1983; Goodnight , 1992; Kelly , 1994 ) and by ( Wilson et al . , 1992 ) from a group selection perspective . There may be multiple and overlapping mechanisms underlying assortment . There is much debate in the literature over which theories best explain the evolution of cooperation and under which conditions each theory may be applicable . There is still not agreement , for example , on whether kin-selection and group selection can be viewed as equivalent theories ( Lion et al . , 2011; Kramer and Meunier , 2016 ) . According to ( Simon et al . , 2012; Simon et al . , 2013 ) , since relatedness need not impact certain group level selection events , such as various games between groups , group selection is distinct from kin selection . Also , individual and group level selection events are generally asynchronous in nature and therefore cannot be equivalent . However , the debate still goes on ( Kramer and Meunier , 2016; West et al . , 2007; Gardner , 2015; Goodnight , 2015 ) . Typically , evolution of cooperation is quantitatively analyzed with the aid of game theoretic models applied to well-mixed populations , networks and other phenomenological spatial structures ( Szabó and Fáth , 2007; Allen et al . , 2013; Nowak and Sigmund , 2004; Vural et al . , 2015 ) . While there are few models that take into account spatial proximity effects , ( Medvinsky et al . , 2002; Nadell et al . , 2010; Nadell et al . , 2013; Dobay et al . , 2014; Driscoll and Pepper , 2010 ) and the influence of decay and diffusion of public goods ( Dobay et al . , 2014; Wakano et al . , 2009; Hauert et al . , 2008 ) , how advective fluid flow influences social evolution remains mostly unexplored . The present study aims to fill this gap . A flowing habitat can have a drastic effect on population dynamics ( Tél et al . , 2005; Nickerson et al . , 2004; Koshel' and Prants , 2006; Sandulescu et al . , 2008 ) . For example , a flowing open system can allow the coexistence of species despite their differential fitness ( KarolyiKárolyi et al . , 2000 ) . Interactions between fluid shear and bacterial motility has been shown to lead to shear trapping ( Rusconi et al . , 2014; Berke et al . , 2008 ) which causes preferential attachment to surfaces ( Berke et al . , 2008; Li et al . , 2011 ) . Turbulent flows can also lead to a trade-off in nutrient uptake and the cost of locomotion due to chemotaxis ( Taylor and Stocker , 2012 ) , and can drastically effect the population density ( Pigolotti et al . , 2012; Perlekar et al . , 2010 ) . Most importantly , the reproductive successes of species ( and individuals within a single species ) are coupled over distance , through the secretion of toxins , goods , and signals ( Mimura et al . , 2000; Allison , 2005; Hibbing et al . , 2010 ) . The spatial distributions of all such fitness altering intermediaries depend on flow . Indeed , the experimental study by Drescher et al . ( 2014 ) has shown that flow can help promote cooperation in bacterial biofilms . Thus , we are motivated to find out how flow plays a role in the evolution of cooperation . Here we theoretically study how fluid dynamics molds the social behavior of a planktonic microbial population . Qualitatively stated , our evolutionary model has three assumptions: ( 1 ) Individuals secrete one waste compound and one public good . The former has no cost , whereas the latter does . ( 2 ) Mutations can vary the public good secretion rates of microbes , thereby producing a continuum of social behavior . ( 3 ) Microbes and their secretions diffuse and flow according to the laws of fluid dynamics . Under these assumptions , we find , through computer simulations and analytical theory , that bacteria self organize and form patterns of spots , which then exhibit an interesting form of budding dispersal when sheared by ambient fluid flow . The dispersal process preserves the group structure , thereby enabling evolutionarily stable social behavior . Our model is applicable to a wide variety of social ecosystems ranging from phytoplankton flowing in oceanic currents to opportunistic bacteria colonizing blood or industrial pipelines . Our findings imply that greater social complexity amongst planktonic species would be observed in regions of large shear , such as by rocks and river banks . We might even speculate that multicellularity may have originated near fluid domains with large shear flow , rather than the bulk of oceans or lakes . This paper is organized as follows: We first establish that under certain conditions our physical model gives rise to spatially organized cooperative structures . The structures are a natural byproduct of the dynamics of the system . Furthermore , these social structures reproduce in whole to form new identical structures . Variants of such structures have already been studied in ecological settings ( Tian et al . , 2011; Camara , 2011; Baurmann et al . , 2007; Wilson et al . , 2003 ) and growth patterns of microbial populations havebeen explored ( Ben-Jacob et al . , 1994; Chang-Li et al . , 1988 ) . We then study the effects of mutation . We first start with a simplified model with only two phenotypes: cheater and altruist . We then generalize to a continuum of public good secretion rates . In both cases , we observe that above a certain mutation rate , cheating strains take over groups which leads to total extinction . The latter finding is consistent with other empirical ( Rainey and Rainey , 2003; Diggle et al . , 2007 ) and theoretical studies ( Nowak and May , 1992 ) . Through the fragmentation of social groups , and death of cheating groups , we recover the results of Simpson’s paradox ( Chuang et al . , 2009 ) where individual groups may decrease in sociality , but the population as a whole becomes more social . After setting up the stage for naturally forming social groups , we present our central and novel finding , that flow shear can lead to evolutionarily stable cooperative behavior within the population . Specifically , we demonstrate and study the evolution ofsociality of a microbial population ( 1 ) subjected to constant shear , ( 2 ) embedded in a cylindrical laminar flow and ( 3 ) in a Rankine vortex . We find that in all three cases population density and cooperative behavior scales with flow shear . The mechanism of action is that shear distortion enhances the fragmentation of cooperative clusters , thereby increasing the group fragmentation rate and limiting the spread of cheaters . If the shear is large enough that groups are torn apart at a larger rate than the mutation rate , then cooperation will prevail . Otherwise , groups will become dominated by cheaters , and eventually die out ( Figure 2 , Videos 1–3 ) .
Diffusion can cause an instability that leads to the formation of intriguing patterns ( Turing , 1990 ) , which among other fields , have been investigated in ecological context ( Tian et al . , 2011; Camara , 2011; Baurmann et al . , 2007; Wilson et al . , 2003 ) . These so called Turing patterns typically form when an inhibiting agent has a diffusion length greater than that of an activating agent . For our model system , the waste compound and public good play the role of inhibiting and activating agents , and patterns manifest as cooperating microbial clusters Figure 2 . The size and reproduction rate of these clusters , in terms of system parameters , can be estimated from a Turing analysis ( cf . appendix 1 ) . Figure 3 shows the values of diffusion constants that gives rise to Turing patterns , as well as the size of the groups . In the homogeneous phase , the system is evolutionarily very unstable , since as soon as one cheating mutant emerges , it quickly takes over the entire population , ultimately causing the population to go extinct ( Video 1 ) . The group phase tolerates cheaters better , since once a cheater emerges it will take over and compromise the fitness of only one group , while the others will live on . However , in the absence of group fragmentation , novel cheating mutations will ultimately emerge in all groups , and annihilate the population one group at a time ( Video 2 ) . The main contribution of this paper is to demonstrate that stable social cooperation can be induced or enhanced by fluid flow gradients . Specifically , shear forces induced by advective flow distorts and fragments microbial clusters , leading to a kind of budding dispersal , which in turn enables evolutionary stable cooperation ( Figure 2 , Video 3 ) . It is well known that evolutionary outcomes can depend on individuals being discrete ( Durrett and Levin , 1994 ) . In our model , having a continuous population density can allow for the existence of ‘micro-mutant populations’ which can spread easier between adjacent groups . The discreteness further separates the clusters of microbes from each other , since there cannot exist fractional individuals . In reality microbes are quantized , and we thus expect a discrete simulation to better model the biology . In Figure 3 we present the phase diagram of the system , as obtained by analytical theory , discrete agent based simulations ( where microbes are discrete , self-replicating brownian particles ) , and continuous simulations ( where Equations 1 , 2 , 3 are solved numerically ) . In order not to obfuscate the biology , we report our detailed mathematical treatment in the appendix . We quantitatively determine the effect of different flow velocity profiles on the social evolution of the system . A constant fluid flow merely amounts to a change in reference frame , which of course , does not change the evolutionary fate of the population . However , we find that velocity gradients cause significant changes to the social structure , both spatially and temporally . Specifically , we find that large shear rate causes microbial groups to distort and fragment , which in turn facilitates group reproduction . To investigate this effect in detail , we ran simulations for three fluid velocity distributions: Couette flow , Hagen-Poiseuille flow , and Rankine vortex . We first look at a simplified system with just two phenotypes , cheater and altruist , to gain a basic understanding of the mechanism involved . Mutations can cause a switch in social behavior . To see the effect of shear on social evolution , we introduced Couette flow to the microbial habitat . In this case , the flow velocity takes the formv ( x ) =vmaxrRz^ , where R is the radius of the pipe , and z^ is the longitudinal direction . The shear rate is the derivative of the flow velocity and is related to the maximum flow rate vmax , σ=dvdr=vmaxR . We ran simulations for various shear rates and diffusion constants and observed that shear does not significantly influence the region of parameter space that gives rise to cooperating groups . However if the system parameters are conducive to the formation of groups , shear tears groups apart and increases the rate at which spatially distinct cooperative clusters form . We find that the group fragmentation rate ω ( σ ) , depends linearly on the shear rate σ ( Figure 4 ) ω ( σ ) =mσ+ω0 , where ω0 is the fragmentation rate solely due to microbial diffusion and can approximately be given by the Turing eigenvalue ω0≈Λmax ( see appendix ) . The constant of proportionality m is given empirically from our simulations and depends on diffusion lengths and group density . This holds in the low density regime . Once the population density becomes large , group-group interactions slow the group reproduction rate and the population saturates . Larger shear corresponds to a faster rate of group fragmentation , thus enabling or enhancing social behavior in the microbial population . Given the group size and fragmentation rate as a function of shear , we can calculate roughly where the critical shearrate is for sociality . We also find that the group population N increases linearly with shear , N ( σ ) =nσ+N0 . where n and N0 are the slope and intercept of the line in Figure 4B . This also influences where the critical shear rate will be . Cooperation is stable if a group is able to fragment before a cheating strain emerges and proliferates in the group . Therefore , for stability , we need the take-over time to exceed the time it takes for a group to reproduce . A mutant emerges at a rate of μN ( σ ) ( we emphasize that μ is not the generic mutation rate , but the rate at which a particular social gene mutates ) . Once a mutant emerges , it takes some time τd to spread to where the daughter group forms . τd will depend on where the mutant first emerges . Assuming a uniform distribution , and taking the diffusion time in two dimensions as a function of radius r , τd ( r ) =r2/4db , we obtain<τd>=∫0R r24db2rR2dr=R28db , where R is the group radius . The take-over rate is then given by taking the inverse of the total take-over time , and the critical shear rate σc necessary for social cooperation is given by equating the take-over rate with the reproduction rate , ( 4 ) ω ( σc ) =[ 1μN ( σc ) +<τd> ]−1 . The critical shear rate σc above which the system can maintain stable cooperation is then given by the positive root , ( 5 ) σc=−bσ+bσ2−4aσcσ2aσwhere aσ=<τd>μmn , bσ=<τd>μnω0+<τd>μmN0+m−μn , and cσ=<τd>μN0ω0+ω0−μN0 . The values obtained from Equation 5 is indicated by the vertical dashed lines in Figure 4 and agrees with the computationally observed critical shear reasonably well . It may be possible to improve this formula further by taking into account additional factors , such as the non-uniform spatial distribution of population within a group and the elongation of groups with shear . Furthermore , as the mutants increase in numbers it becomes more likely that one of them crosses over the daughter group , thereby reducing further the expected <τd> . We see better agreement with analytical theory and simulations at lower mutation rates , since these corrections are mainly to the diffusion time <τd> , and become more significant at higher mutation rates , where <τd>≫1/μN , ( cf . Equation 4 ) . If shear is below the critical value ( Equation 5 ) , the system will be in a non-social state . Ultimately , cheaters will take over , and wipe out all groups . When shear is increased above the critical value however , the system will transition to a stable social state , thereby maintaining its fitness and dense population indefinitely . Figure 4 shows the long-time population of the system versus the shear rate . The population goes extinct under larger mutation rates unless the shear rate is above the critical value . When is shear necessary , and when is it just a sufficient condition for cooperation ? By setting σc=0 in Equation 5 and solving for μ we can also obtain the critical mutation rate above which shear is necessary in order to have social cooperation . We get μc=6 . 9×10−7 analytically and our simulations show a critical mutation rate around μc=5 . 5×10−7 . As we will see , we obtain similar results when the available phenotypes include a continuum of social behaviors . In this case , a mutation changes the secretion rate of a microbe by a uniformly chosen random number between 0 and 1 s−1 . We observe from our simulations , that mutations that increase the secretion rate of a microbe do not fixate , since the microbe now pays a higher cost and is less fit than its neighbors . However , once a mutation that lowers the secretion rate of a microbe occurs within a group , it quickly takes over the entire group , leaving individual groups homogeneous in secretion rate . We show the diversity of social behaviors across groups and within individual groups in greater detail in Appendix 1—figure 1 . Since less cooperative phenotypes always dominate more cooperative phenotypes , we find no diversity of social behavior within a group . However we do see a large variation across groups , which increases with shear . Groups with different secretion rates reproduce at different rates . Groups with too low of a secretion rate are not stable and die off . In general , the system will evolve to a distribution of groups with secretion rates centered around the value of secretion rate that maximizes the group reproduction . For larger mutation rates , the system will tend towards lower average secretion rate and/or go extinct . The average secretion rate of the population can be maintained at a higher value by introducing shear flow , Figure 5 . We therefore have the same qualitative result as in the two phenotype case , if shear is below some critical value , the system will be in a non-social state . Ultimately , cheaters will take over , and wipe out all groups . As before , the social state of the population can transition from a non-cooperative state to a cooperative one with increased flow shear . We now further generalize our results by looking at laminar flow with fixed boundaries , and with a continuum of public good secretion rates . For Hagen-Poiseuille flow , the shear rate varies linearly with the radius , taking its maximum value adjacent to the boundaries , when r=R . The flow and shear profiles are given as , v=vmax ( 1−r2R2 ) , dvdr=−2vmaxrR2 . We therefore expect to see groups fragment quicker at the boundary , leading to larger cooperation , higher average secretion rate , and larger population , which is indeed what we do see ( Figure 6 , Video 4 ) . Earlier studies have proposed and shown that shear trapping due to the interaction between bacterial motility and fluid shear can result in preferential attachment to surfaces , ( Rusconi et al . , 2014; Berke et al . , 2008; Li et al . , 2011 ) . In a similar spirit , we suggest that inhabiting surfaces may have the additional advantage of enhanced sociality , due to shear driven group fragmentation and dispersal . In a vortex , the region above the critical shear value constitutes an annulus . Thus , we expect social behavior to be localized . Any point in the fluid outside this annulus will be taken over and destroyed by cheaters . In our simulations , at steady state we indeed see clusters whirling around exclusively within annulus , neither too near , nor too far from the vortex core ( Video 5 ) . Life cannot exist outside this annulus , as cheaters kill these groups . The Rankine vortex in two dimensions is characterized by a vortex radius R and a rotation rate Γ . The shear rate acting on a group acts tangential to the flow . The velocity profile and shear magnitude are given as , v={Γr2πR2θ^ , r≤RΓ2πrθ^ , r>Rσ={0 , r≤RΓ2πr2 , r>Rwhere r2=x2+y2 . The shear rate is then a maximum at the minimum value of r which occurs at the vortex radius R . We therefore expect to see the largest concentration of groups at the vortex radius , which is what we observe in our simulations ( Figure 6 ) . While we paid close attention to physical realism , we also made important simplifying assumptions which under certain circumstances , may lead to incorrect conclusions . We caution the reader by enumerating the limitations of our model . First , since many microorganisms live in a low Reynolds number environment , we have chosen to neglect the inertia of microorganisms . However in reality , the microorganisms influence the flow around them . This effect will be particularly significant for a dense microbial population , especially when the microbes stick onto one other , or integrate via extracellular polymers . A more sophisticated model would include the coupling of the microbes to the flow . Secondly , the finite size and shapes of the microorganisms have been neglected . Instead , we have treated microbes as point particles , which will also invalidate our model in the dense population limit . Lastly , real microbes display a large number of complex behaviors such as biofilm formation and chemotactic migration . Here we have ignored the active response of microorganisms to the chemical gradients that surround them and to the surfaces they might attach and migrate . Instead , we took them as simple Brownian particles .
It is well known that spatial structure is crucial in the evolution of cooperation , ( Wilson et al . , 1992; Taylor , 1992; Lion and Baalen , 2008 ) . Many of these studies introduce these mechanisms ‘manually’ , for example density regulation and migration are enforced by applying carrying capacities and migration rates to groups . In this study we have distanced ourselves from the typical game theoretic abstractions used to investigate evolution of cooperation . Instead we adopted a mechanical point of view . We investigated in detail , the fluid dynamical forces between microbes and their secretions , to understand how cooperation evolves among a population of planktonic microbes inhabiting in a flowing medium . In our first principles model , the spatial structuring and dispersion occur naturally from the physical dynamics . We found that under certain conditions , microbes naturally form social communities , which then procreate new social communities of the same structure . More importantly , we discovered that regions of a fluid with large shear can enhance the formation of such social structures . The mechanism behind this effect is that fluid shear distorts and tears apart microbial clusters , thereby limiting the spread of cheating mutants . Our proposed mechanism can be seen as shear flow enhanced budding dispersal . This can also be viewed under the phenomenon of Simpson’s paradox ( Chuang et al . , 2009 ) where individual groups may decrease in sociality , but the population as a whole becomes more social . In our investigation , we found only certain regions of the fluid domain admits life , social or otherwise , as governed by the domain geometry and flow rate . From this perspective , it appears that evolution of sociality is a mechanical phenomenon . In our physics-based model , groups emerge from individual-level dynamics and selection . Groups with cheaters are negatively selected , and give way to those without cheaters . On the other hand the ensemble of groups do not exhibit any variation in their propensity to progenerate cheaters , neither is such propensity heritable . Rather , the progeneration of cheating is a ( non-genetic ) symptom that inevitably manifests in every group that has been around long enough . In this sense , it might be appropriate to view the emergence and spread of cheaters in a microbial population as a phenomenon of ‘aging’ , in the non-evolutionary and mechanical sense , that any system consisting of a large number of interdependent components will inevitably and with increasing likelihood , fall apart ( Vural et al . , 2014 ) .
The analytical conclusions we derive from our system ( Equations 1 , 2 , 3 ) has been guided and supplemented by an agent based stochastic simulation . Videos of simulations are provided in Videos 1–5 . Our simulation algorithm is as follows: at each time interval , Δt , the microbes ( 1 ) diffuse by Brownian motion , with step size δ derived from the diffusion constant and a bias dependent on the flow velocity , δ=4dbΔt+vΔt , ( 2 ) secrete chemicals locally that then diffuse and advect using a finite difference scheme , and ( 3 ) reproduce or die with a probability dependent on their local fitness given by f=Δt[ α1c1c1+k1−α2c2c2+k2−β1s1 ] . If f is negative , the microbes die with probability 1 , if f is between 0 and 1 they reproduce with probability f , and if f is larger than 1 , they produce number of offspring given by the integer part of f and another with probability given by the decimal part of f . Upon reproduction , random mutations may alter the secretion rate of the public good –and thus the reproduction rate– of the microbes . Mutations occur with probability μ and can change the secretion rate by a random number between 0 and s1 . The secretion rate is assumed to be heritable , and constant in time . Numerical simulations for figures were performed by implementing the model described above using the Matlab programming language and simulated using Matlab ( Mathworks , Inc . ) . The source code for discrete simulations is provided in Source code 1 and the source code for continuous simulations used in Figure 3B is provided in Source code 2 . A summary of the system parameters is given in Table 1 , along with typical ranges for their values used in the simulations . The relevant ratios of parameters are consistent with those observed experimentally ( Kim , 1996; Ma et al . , 2005 ) . | According to the principle of the ‘survival of the fittest’ , selfish individuals should be better off compared to peers that cooperate with each other . Indeed , even though a population of organisms benefits from working together , selfish members can exploit the cooperative behavior of others without doing their part . These ‘cheaters’ then use their advantage to reproduce faster and take over the population . Yet , social cooperation is widespread in the natural world , and occurs in creatures as diverse as bacteria and whales . How can it arise and persist then ? One idea is that when individuals form distinct groups , the ones with cheaters will perish . Even though a selfish individual will fare better than the rest of its team , overall , cooperating groups will survive more and reproduce faster; ultimately , they will be favored by evolution . This is called group selection . Here , Uppal and Vural examine how the physical properties of the environment can influence the evolution of social interactions between bacteria . To this end , mathematical models are used to simulate how bacteria grow , evolve and drift in a flowing fluid . These are based on equations worked out from the behavior of real-life populations . The results show that flow patterns in a fluid habitat govern the social behavior of bacteria . When different regions of the fluid are moving at different speeds , ‘shear forces’ are created that cause bacterial colonies to distort and occasionally break apart to form two groups . As such , cooperative groups will rapidly form new cooperating colonies , whereas groups with cheaters will reproduce slower or perish . Furthermore , results show that when different areas of the fluid have different shear forces , social cooperation will only prevail in certain places . This makes it possible to use flow patterns to fine tune social evolution so that cooperating bacteria will be confined in a certain region . Outside of this area , these bacteria would be taken over by cheaters and go extinct . Bacteria are both useful and dangerous to humans: for example , certain species can break down pollutants in the water , when others cause deadly infections . These results show it could be possible to control the activity of these microorganisms to our advantage by changing the flow of the fluids in which they live . More broadly , the simulations developed by Uppal and Vural can be applied to a variety of ecosystems where microscopic organisms inhabit fluids , such as plankton flowing in oceanic currents . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"computational",
"and",
"systems",
"biology"
] | 2018 | Shearing in flow environment promotes evolution of social behavior in microbial populations |
Detachment of newborn neurons from the neuroepithelium is required for correct neuronal architecture and functional circuitry . This process , also known as delamination , involves adherens-junction disassembly and acto-myosin-mediated abscission , during which the centrosome is retained while apical/ciliary membranes are shed . Cell-biological mechanisms mediating delamination are , however , poorly understood . Using live-tissue and super-resolution imaging , we uncover a centrosome-nucleated wheel-like microtubule configuration , aligned with the apical actin cable and adherens-junctions within chick and mouse neuroepithelial cells . These microtubules maintain adherens-junctions while actin maintains microtubules , adherens-junctions and apical end-foot dimensions . During neuronal delamination , acto-myosin constriction generates a tunnel-like actin-microtubule configuration through which the centrosome translocates . This movement requires inter-dependent actin and microtubule activity , and we identify drebrin as a potential coordinator of these cytoskeletal dynamics . Furthermore , centrosome compromise revealed that this organelle is required for delamination . These findings identify new cytoskeletal configurations and regulatory relationships that orchestrate neuronal delamination and may inform mechanisms underlying pathological epithelial cell detachment .
Delamination involves extraction of a cell from within a proliferative tissue . It is a fundamental process underlying epithelial tissue morphogenesis that is linked to cell state change during normal differentiation and also to cancer cell dispersal . Cells undergoing neuronal differentiation delaminate from the proliferative domain of the neuroepithelium and this involves loss of adhesion between neighbouring cells at the ventricular surface . This process is required for correct neuron placement ( Kriegstein and Noctor , 2004; Singh and Solecki , 2015 ) , and this in turn is necessary for subsequent formation of functional neuronal circuitry . Neuronal delamination defects are collectively known as periventricular heterotopias and lead to a spectrum of deficits including epilepsy , dyslexia and intellectual disability ( Lian and Sheen , 2015; Passarelli and Moreira , 2014 ) . A genetic basis for human periventricular heterotopia has been mapped to the actin cross-linking protein , FilaminA and the ADP-ribosylation factor guanine exchange factor 2 ARFGEF2/BIG2 ( Lian and Sheen , 2015 ) . The interaction between these proteins has implicated them in vesicle trafficking and stability/turnover of cell adhesion proteins ( Zhang et al . , 2013; Zhang et al . , 2012 ) . These data are consistent with work linking mutation of cadherins FAT4 and DCHS1 with a periventricular heterotopia phenotype ( Badouel et al . , 2015; Cappello et al . , 2013 ) . Experiments in animal models implicate further regulators of cell adhesion in neuronal delamination , including Slit/Robo , which also acts by attenuating N-cadherin activity ( Wilsch-Bräuninger et al . , 2016; Wong et al . , 2012 ) ( Borrell et al . , 2012 ) . Overall , many such proteins associated with apically localised adherens junctions ( AJs ) have been linked to the delamination process ( Cappello et al . , 2006; Imai et al . , 2006; Kadowaki et al . , 2007; Singh and Solecki , 2015; Stocker and Chenn , 2009; Stocker and Chenn , 2015 ) . AJs are required for the integrity of the entire neuroepithelium and so delamination defects and precocious neuronal differentiation are most readily seen following cell-autonomous deletion of associated proteins ( Stocker and Chenn , 2009; Stocker and Chenn , 2015; Woodhead et al . , 2006; Zhang et al . , 2010 ) . However , despite such manipulations we know little about the cell biological mechanisms that mediate delamination as AJs disassemble . Recent high-resolution live tissue-imaging of chick spinal cord has revealed that detachment of the newborn neuron from the ventricle is mediated by a novel cell sub-division mechanism , apical abscission , which leads to shedding of the apical tip of the cell ( Das and Storey , 2014; Das and Storey , 2014b ) . The apical poles of neuroepithelial cells are characterised by the presence of a contractile sub-apical acto-myosin cable which is mechanically and biochemically linked to cadherin-containing AJs ( Abe and Takeichi , 2008; Marthiens and ffrench-Constant , 2009; Maul et al . , 2003; Miyamoto et al . , 2015 ) . Apical abscission is triggered by acto-myosin cable constriction following attenuation of N-cadherin; this process is blocked by N-cadherin mis-expression ( Das and Storey , 2014 ) while repression of N-cadherin transcription downstream of the neurogenic transcription factor cascade , which promotes neuronal differentiation , leads to loss of cell–cell contact at the ventricular surface ( Rousso et al . , 2012 ) . Similar transcription factor activity that promotes neuronal delamination in the brain involves regulation of cadherin/apical polarity proteins by Snail superfamily members ( and others ) ( Acloque et al . , 2009; Itoh et al . , 2013; Singh et al . , 2016; Singh and Solecki , 2015 ) . Importantly , such proteins also induce cell-cell detachment during epithelial to mesenchymal transition in other tissues and in oncogenic contexts suggesting operation of shared downstream cell biological mechanisms . In some respects , apical abscission resembles cytokinesis , where a contractile acto-myosin ring generates the forces that separate the two daughter cells . A key structure regulating this cytokinetic ring is the central spindle , which consists of an array of antiparallel microtubules as well as de novo synthesized microtubules ( Fededa and Gerlich , 2012 ) . This raises the possibility that microtubules regulate the apical acto-myosin cable in neuroepithelial cells during delamination . Like actin , microtubules are also associated with AJs ( Bellett et al . , 2009; Ligon et al . , 2001; Meng et al . , 2008; Stehbens et al . , 2006 ) and cadherin-mediated adhesion can recruit and stabilize microtubules ( Stehbens et al . , 2006; Waterman-Storer et al . , 2000 ) . Conversely , AJs are destabilized by microtubule de-polymerisation in a variety of cell types in vitro ( Mary et al . , 2002; Yap et al . , 1995 ) . This microtubule support for AJs involves kinesin-based transport of cadherin containing vesicles ( Mary et al . , 2002 ) and specifically in neuroepithelial cells by the KIF3 motor complex ( Teng et al . , 2005 ) , although this transport role is context dependent ( Stehbens et al . , 2006 ) . Furthermore , microtubule de-polymerisation or stabilisation can block AJ disassembly ( Ivanov et al . , 2006 ) suggesting a more complex relationship between cadherin supply and AJ integrity . Little is known about the organisation of microtubules and their relationship with actin and AJs in the neuroepithelial cells or how they might regulate neuronal delamination . A relationship between regulation of AJs and cell cycle exit is suggested by findings that link AJs to mitogenic signalling via Notch and Wnt pathways ( Hatakeyama et al . , 2014; Zhang et al . , 2010 ) . In the chick spinal cord , apical abscission is preceded by dis-assembly of the primary cilium ( Das and Storey , 2014 ) and loss and or retraction of ciliary membrane is also associated with delaminating zebrafish retinal neuroblasts ( Lepanto et al . , 2016 ) . Mediators of the mitogenic Sonic hedgehog pathway are processed into activated forms in the primary cilium ( Guemez-Gamboa et al . , 2014; Kim et al . , 2009 ) and so this may be a further way in which cell biological mechanisms associated with delamination link this process to cell state change . Following cilium disassembly , the centrosome is retained in the withdrawing neuronal cell process while ciliary and apical membrane are shed ( Das and Storey , 2014 ) . Centrosome retention is then critical for subsequent neuronal differentiation: for neuronal migration to form the cortical plate ( Higginbotham and Gleeson , 2007; Tsai and Gleeson , 2005; Xie et al . , 2003 ) , as a microtubule organising centre during axonogenesis ( de Anda et al . , 2005; Zmuda and Rivas , 1998 ) , and in defining where dendrites will elongate ( Puram and Bonni , 2013; Puram et al . , 2011 ) , although this is context dependent ( Kuijpers and Hoogenraad , 2011 ) . The role of the centrosome in delamination and the mechanism that ensures its retention in newborn neurons are , however , not known . Here , we use live-tissue imaging and super-resolution microscopy to elucidate the cytoskeletal architecture of the apical end-foot of neuroepithelial cells and to dissect the regulatory relationships which underpin cytoskeletal dynamics underlying neuronal delamination .
To localise microtubules within neuroepithelial cells , we carried out immunocytochemistry in sections of chick spinal cord ( at Hamburger and Hamilton stage HH17-8 ) ( Hamburger and Hamilton , 1951 ) to detect the stable microtubule marker , acetylated α-tubulin ( Perdiz et al . , 2011 ) , the AJ-associated protein N-cadherin and the actin cytoskeleton using phalloidin . The microtubule cytoskeleton was enriched apically and overlapped with the actin cable and the AJs ( Figure 1A–A’’’’ ) . Closer examination of microtubule architecture in neuroepithelial cell apical end-feet using en face imaging , revealed a sub-apical ring-like structure ( 2 . 57 ± 0 . 5 μm in diameter , 21 cells , in 2 explants from 2 embryos ) and associated microtubules radiating from the centrosome of the primary cilium , identified by γ-tubulin and IFT88 , respectively ( Figure 1B–B’’ , C–C’ and D–D’ ) . A similar microtubule configuration was observed by en face imaging of the ventricular surface in E12 . 5 mouse spinal cord and cortex ( Figure 1E , in 4 explants from 2 embryos ) indicating conservation of this apical microtubule architecture across species and different regions of the central nervous system . To place these apical microtubules in the context of known apical sub-cellular organisation , we next captured the three-dimensional relationship between the alpha-tubulin-labelled microtubules , the actin cable and associated N-cadherin-containing AJs , imaging from the apical surface of the chick spinal cord in en face orientation ( Figure 1F , Figure 1—video 1 , n = 167 cells , in 4 explants from 4 embryos ) . The alignment of actin and tubulin was then measured at the Z-level defined by N-Cadherin localisation; this revealed actin-tubulin co-alignment in the majority of cells ( 71% ) ( Figure 1—figure supplement 1 , 31 cells in 3 explants from 3 embryos ) . A subset of microtubules was also observed to extend basal to the actin/N-cadherin junctional region deep into the cell-process ( Figure 1F ) . To capture the overall microtubule configuration in individual cells , we next mis-expressed a GFP-tagged microtubule binding protein MAP7/Ensconsin ( EMTB-GFP ) ( Bulinski et al . , 1999 ) along with a plasmid expressing the proneural factor Neurog2 ( pCAGGS-Neurog2_IRES-nucGFP , pCIG-Neurog2 ) to promote neuronal differentiation ( Ma et al . , 1996 ) in a scattering of cells in the developing chick spinal cord ( HH 10–12 ) . Mis-expression of EMTB at high levels can stabilise microtubules ( Bulinski et al . , 1999 ) and this facilitated use of structured illumination microscopy ( SIM ) to generate super-resolution images of extensive microtubule structures within neuroepithelial cells . Analysis of such individual cells in transverse embryo slices revealed a more elaborate microtubule meshwork and also continuity between sub-apical microtubules and apico-basal orientated microtubules that extend towards and around the cell nucleus ( Figure 1G–G’’’ , and Figure 1—video 2 , 4 cells from 2 embryos ) . Together these two and three-dimensional analyses suggest the presence of a sub-apical wheel-like microtubule organisation , composed of radial microtubules emanating from the centrosome and rim microtubules aligned with actin/N-cadherin , which is further continuous with apico-basal microtubules that extend the length of the cell , summarised in Figure 1H . To substantiate the centrosomal origin of the radial and rim microtubules , we next used live tissue imaging to monitor microtubule nucleation patterns in the apical end-foot . This involved mis-expression of PACT-TagRFP to label the centrosome and EB3-GFP to identify microtubule plus-ends ( Gillingham and Munro , 2000; Stepanova et al . , 2003 ) in chick spinal cord and monitoring cell behaviour in an adapted en face version of ex-vivo embryo slice cultures using high-resolution wide-field microscopy ( Das et al . , 2012 ) . Tracking the trajectory of EB3-GFP comets revealed that radial microtubules emanate in an evenly spaced fashion from the centrosome of the primary cilium in the end-foot ( Figure 2A; Figure 2—video 1 , 51 cells in 3 explants from 3 embryos ) . By combining EB3-GFP and F-tractin-mKate2 to monitor the relationship between these microtubules and the actin cable , we further observed some EB3-GFP comets running along the actin cable ( Figure 2B; Figure 2—video 2 , 95 cells in 4 explants from 4 embryos ) . To quantify this relationship , we followed the 2D trajectories of EB3-GFP comets and measured the EB3-GFP/F-tractin-mKate2 inter-peak distance over time . This analysis indicated a close alignment of polymerising microtubules with the actin belt ( Figure 2D–F; Figure 2—video 3; trail tracking: 10 cells in 3 explants from 3 embryos ) . Tracking comet movements also delineated microtubule shapes and revealed that radial microtubules bend as they reach the periphery and turn to run along the actin cable ( Figure 2C; Figure 2—video 4 , 12 cells in 4 explants from 4 embryos ) . These dynamic data further support the case for a wheel-like organisation of apical microtubules , demonstrate that the centrosome is the source of both radial and rim microtubules and confirm the close alignment of rim microtubules with the actin cable and AJs . To test the regulatory relationships between apical microtubules , actin and AJs , we next assessed the consequences of microtubule depolymerisation following exposure to Nocodazole for 1 hr . This treatment depleted apical microtubules as expected ( Figure 3A , B ) and reduced N-Cadherin at AJs ( Figure 3A’ , B’ ) quantified by fluorescence intensity measurements ( Figure 3C , C’ ) . Depletion of microtubules also increased distribution of actin within the cell ( Figure 3A” , B” , D ) , however , this did not significantly alter actin levels at the adhesion belt ( Figure 3D , D’ ) nor reduce apical end-foot area ( Figure 3A’’’ , B’’’ , E ) . These findings indicate that apical microtubules maintain AJs as defined by N-Cadherin levels and that they influence actin localisation , although this did not impact the actin cable nor apical end-foot size . We next tested the effects of actin depletion on AJs and apical microtubules . Brief exposure ( 15 mins ) to Latrunculin-A which binds actin monomers and so prevents their polymerisation ( Coué et al . , 1987 ) dramatically reduced apical actin as expected ( Figure 3F , G , H , H’ ) . This treatment depleted apical microtubules ( Figure 3F” , G” , J ) and consistent with this also reduced N-Cadherin at AJs ( Figure 3F’ , G’ ) and quantified in Figure 3I , I’ . Actin depletion additionally led to a decrease in apical end-foot size ( Figure 3F’’’ , G’’’ , K ) . These findings indicate that an intact actin cable is required for maintenance of apical microtubule structures as well as AJs in neuroepithelial cells and that the actin cytoskeleton serves to define apical end-foot dimensions . Together , the above findings uncover a wheel-like organisation of sub-apical microtubules that is nucleated by the centrosome of the primary cilium and which aligns with the actin cable , maintains AJs and stabilises the apical cytoskeleton in neuroepithelial cells of the developing embryo . The tissue analysed at these early stages comprises largely neural progenitors in interphase and so we next addressed how this cytoskeletal configuration alters during neuronal delamination . To assess the apical cytoskeletal configuration in delaminating cells , we next monitored EB3-GFP and F-tractin-mKate2 in cells with a small apical end-foot diameter ( typically 1–2 . 5 μm ) , characteristic of delaminating cells ( Figure 4A , Figure 4—video 1 , 6 cells in 4 explants from 4 embryos ) . This revealed that microtubule growing tips still emanated towards and along the now constricted actin cable in such cells . This suggests that despite declining N-Cadherin in delaminating cells microtubules remain closely associated with peripherally located actin . To elucidate further the spatial organisation of the cytoskeleton in delaminating cells , we used Stimulated Emission Depletion ( STED ) microscopy to generate super-resolution images of cells expressing F-tractin-mKate2 and EMTB-GFP ( along with the neuronal differentiation gene Neurog2 , as above ) . This revealed a close sub-apical alignment of actin and microtubules in such cells and , observed in 3-dimensions , these two cytoskeletal components appeared to form a composite tunnel-like configuration ( Figure 4B , Figure 4—video 2 , 4 cells from 2 embryos ) . To monitor overall microtubule dynamics during delamination in live tissue , spinal cord cells were next co-transfected with EMTB-GFP , pCIG-Neurog2 and mKate2-GPI to label cell membranes and to monitor the changing morphology of individual cells . We then observed neurogenesis in ex-vivo embryo transverse slice cultures as described in Das et al . , 2012 . We monitored cells with moderate levels of EMTB-GFP transfection and observed that the prominent sub-apical EMTB-GFP labelling was highly dynamic and its intensity progressively increased as delamination proceeded . Following completion of abscission , the condensed band of EMTB-GFP was then rapidly lost from the tip of the withdrawing cell-process ( Figure 4C , Figure 4—video 3 , 22 cells in 15 slices; in each experiment slices are taken from 2 or 3 embryos , this applies here and in all similar experiments below ) . This dynamic pattern of enrichment and subsequent loss following abscission is very similar to that we observed previously for actin during this process ( Das and Storey , 2014 ) . These findings further support the coordinated condensation of apical actin and microtubules during delamination and raised the possibility that apical microtubule re-organisation plays a role in this process . To test whether microtubules are required for neuronal delamination neural tubes were first co-transfected with GFP-GPI and pCIG-Neurog2; following 18 hr of incubation , many transfected cells were found to have adopted a configuration with a basally located nucleus and long cell-process contacting the ventricular surface , indicative of imminent neuronal differentiation . In control DMSO treated slices 19/61 cells ( 31% in 29 slices ) then abscised within 4 hr ( Figure 5A; Figure 5—video 1 ) . However , fewer labelled cells exposed to nocodazole delaminated during this period ( 8/51 cells , 16% in 35 slices ) ( Figure 5B , Figure 5—video 2 ) ( an effective nocodazole concentration ( 8 . 5 μM ) for this embryo slice culture assay was determined by monitoring mitotic arrest see Figure 5—figure supplement 1 , Figure 5—videos 3 and 4 ) . These data suggest that microtubules are required for delamination . We next used the microtubule stabilising agent taxol ( Jordan and Wilson , 1998 ) , which reduces microtubule plus end growth ( Kleele et al . , 2014; Marx et al . , 2013 ) , to determine whether this process relies on dynamic microtubules . The effectiveness of taxol concentration in this embryo slice assay ( 10 μM ) was also first determined using live imaging to assess induction of mitotic arrest ( Figure 5—figure supplement 2 , Figure 5—video 5 ) . Cells were transfected as above and cell behaviour monitored following exposure to control DMSO or taxol . While many cells abscised in DMSO treated slices within 6 hr ( 24/55 cells , 44% in 23 slices ) , fewer cells cultured in the presence of taxol exhibited this behaviour ( 13/51 cells , 25% in 26 slices ) ( Figure 5C , Figure 5—video 6 ) . To test this requirement for microtubules during neuronal delamination further we additionally used a genetic approach . This involved mis-expression of Stathmin , which binds to soluble/free tubulin doublets ( Jourdain et al . , 1997 ) and so can be used to deplete soluble tubulin available for microtubule polymerisation ( Gavet et al . , 1998 ) . Cells transfected with Stathmin-GFP and pCIG-Neurog2-NLS were monitored for 12–15 hr and delamination was quantified in cells poised to abscise . We found that few cells delaminated in the presence of Stathmin-GFP ( 3/11 cells , 27% in 6 slices , 5 experiments ) ( Figure 5—video 7 ) , while many more cells underwent this step when only the vector control EGFP was expressed ( 10/17 cells , 59% , in 9 slices , 6 experiments ) ( Figure 5—figure supplement 3A and B ) ( Figure 5—video 8 ) . These data indicate that microtubules and their turnover and active growth are required for neuronal delamination . To confirm that the association between microtubules and actin continues throughout abscission , we performed further live imaging of cells co-transfected with F-tractin-td-Tomato , EMTB-GFP and pCIG-Neurog2 . We again observed that sub-apical actin and microtubules accumulated in and were closely associated at the abscission site and that this remained until final abscission , following which both actin and microtubules were rapidly depleted from the cell-process tip ( Figure 6A , Figure 6—video 1 , 12 cells in 6 slices from 5 embryos ) . To investigate the potential regulatory interactions between actin and microtubules , we next used taxol to stabilise microtubules in cells expressing mKate2-GPI , pCIG-Neurog2 and EMTB-GFP that were poised to delaminate . This confirmed cessation of EMTB-GFP accumulation and subsequent failure to detach from the apical surface . These EMTB-GFP dynamics were then quantified by measuring GFP fluorescence intensity at the sub-apical poles of these cells following exposure to this drug ( Figure 6B , quantified in B’’ grey dashed line , Figure 6—video 2 , 12 cells in 10 slices ) . These clearly contrasted with control cells imaged in medium containing only DMSO which displayed normal accumulation and subsequent loss of EMTB-GFP during abscission ( Figure 6B’ , quantified in B’’ ( black dashed line ) , Figure 6—video 3 , 12 cells in 11 slices ) . We then monitored overall actin dynamics in cells expressing GFP-GPI , pCIG-Neurog2 and F-tractin-mKate2 that were poised to delaminate . We observed that while some cells in taxol-treated slices exhibited sub-apical constriction as judged by local cell shape change ( Figure 6C white arrowheads , 16/34 cells in 21 slices ) , this dynamic of sub-apical actin accumulation and subsequent loss ceased as indicated by fluorescence intensity measurements and such cells remained attached at the ventricular surface ( Figure 6C , quantified in C’’ ( grey dashed line ) , Figure 6—video 4 , 20 cells in 16 slices ) . In contrast , actin intensity in cells imaged in medium containing only DMSO increased at the abscission site and was then rapidly lost from the withdrawing cell-process ( Figure 6C’ , quantified in C’’ ( black dashed line ) , Figure 6—video 5 , 12 cells in 11 slices ) , consistent with our previous report of actin dynamics during this process ( Das and Storey , 2014 ) . We then carried out the converse experiment , in which cells expressing mKate2-GPI , pCIG-Neurog2 and EMTB-GFP that were poised to delaminate were cultured in medium containing 20 μM ML-7 to inhibit acto-myosin constriction ( Saitoh et al . , 1987 ) . We observed that the majority of the EMTB-GFP electroporated cells were now unable to initiate sub-apical constrictions ( 19/25 cells in 18 slices ) and progress through to abscission , and that sub-apical EMTB-GFP labelling no longer exhibited its characteristic pattern of accumulation followed by loss . This was confirmed by measuring sub-apical GFP fluorescence intensities in these cells ( Figure 6D and quantified in D’’ ( grey dashed line ) , Figure 6—video 6 , 16 cells in 13 slices ) . This profile contrasted with control cells , in which EMTB-GFP accumulated and was subsequently lost following abscission ( Figure 6D’ and quantified in D’’ ( black dashed line ) , Figure 6—video 7 , 10 cells in 10 slices ) . These findings indicate that microtubule and actin conformational dynamics are inter-dependent; loss of actively growing microtubules blocked stable accumulation of actin at the presumptive abscission site and loss of acto-myosin activity abolished enrichment of microtubules at this location . Consistent with this inter-dependence , inhibition of either acto-myosin ( Das and Storey , 2014 ) or microtubule activity ( Figure 5C ) reduced the incidence of neuronal delamination . A number of proteins have been proposed to link actin and microtubules ( Coles and Bradke , 2015 ) . These include Drebrin , which was initially identified as an actin-binding protein ( Ishikawa et al . , 1994 ) and later shown to interact with the +TIP protein EB3 ( Geraldo et al . , 2008 ) . To address whether Drebrin is a candidate mediator for actin-microtubule interaction during neuronal delamination we first assessed localisation of endogenous protein using IHC in transverse sections of the neural tube ( Figure 7A ) . We found widespread cytoplasmic localisation of endogenous Drebrin , including in the apical end-foot ( Figure 7A–A’’’ , 3 sections from each of 4 embryos ) . To look more closely at Drebrin localisation in end-feet we mis-expressed Drebrin-mCherry and EMTB-GFP and stained for actin in individual cells ( Figure 7B–B’’ , 18 cells , 6 embryos ) . This analysis confirmed cytoplasmic localisation but also revealed co-localisation with the actin belt and the apical EMTB-GFP-labelled microtubules , quantified by measuring fluorescence intensities across the actin cable in a subset of cells ( Figure 7C , 7 cells , 3 embryos , see Materials and methods ) . Similar co-localisation of Drebrin-YFP and actin was also apparent in en face images ( Figures 7D , 5 explants from 5 embryos ) . These localisation studies support the possibility that Drebrin is involved in the coordination of actin and microtubule dynamics during neuronal delamination To test the requirement for Drebrin in this process , we next mis-expressed a Drebrin short-hairpin ( Sh ) construct ( Dun et al . , 2012 ) in the developing neural tube along with Neurog2 . This led to a marked reduction in the number of delaminating cells ( 4/27 cells in 12–15 hr , 15% in 9 slices ) compared to the scrambled GFP control ( 7/10 cells in 12–15 hr , 70% in 7 slices ) ( Figure 7—figure supplement 1 , Figure 7—videos 1 and 2 ) . This requirement for Drebrin during neuronal delamination is consistent with a role for this protein in regulating cytoskeletal dynamics during this process and supports the possibility that Drebrin acts here as a link between actin-microtubules . Apical abscission is characterised by dis-assembly of the centrosome-primary cilium complex which is followed by a basal translocation of the centrosome and so its retention in the withdrawing cell-process ( Das and Storey , 2014 ) . To investigate the relationship between this translocation and the sub-apical constriction , neural tube cells were transfected with GFP-GPI , pCIG-Neurog2 and PACT-TagRFP , which labels centrosomes and cells were then subjected to live imaging . We observed that differentiating neurons first constricted their sub-apical membranes and that this was then strikingly followed by basal translocation of the centrosome . This event therefore takes place late in the delamination process; indeed in some cells this movement was visible within a thinned membrane bridge between the withdrawing cell-process and the abscising particle ( Figure 8A , Figure 8—video 1 , 10 cells , 9 slices in 9 embryos ) . Monitoring centrosome translocation in cells expressing PACT-TagRFP and EMTB-GFP further revealed that the translocating centrosome moves basally before the resolution of the sub-apical microtubules ( Figure 8B , Figure 8—video 2 , 8 cells in 8 slices and see Figure 4B ) . This suggests that it passes through the sub-apical actin/microtubule tunnel-like configuration that we observed in cells poised to delaminate ( Figure 4B , Figure 4—video 2 ) . To investigate this possibility further , we measured the diameter of the ring formed by rim microtubules visualised with acetylated alpha tubulin ( 17 cells in 2 explants from 2 embryos , Figure 8—figure supplement 1 ) . This gave an average diameter of 0 . 89 ± 0 . 18 μm with an average centrosome diameter measured with IFT88 , at the base of the ciliary membrane , of 0 . 32 ± 0 . 06 μm ( 36 cells in 2 explants from 2 embryos ) . However , the latter only identifies the ciliary axoneme and centrosome ( Robert et al . , 2007 ) and so may under-estimate the full extent of the centrosomal material . Measurement of centrosomal γ-tubulin ( which includes peri-centriolar material ) revealed an average size of 0 . 98 ± 0 . 12 ( 21 cells , data not shown ) , consistent with centrosome size of 0 . 82 ± 0 . 17 μm in other contexts ( Fu and Glover , 2012 ) . These data therefore support the possibility that the centrosome moves through a tunnel-like cytoskeletal configuration formed by apical microtubules and the constricting actin cable . These observations raised the further possibility that sub-apical constriction , which depends on acto-myosin activity , is required for subsequent centrosome translocation . To test this , we observed cells in slices transfected with GFP-GPI , pCIG-Neurog2 and PACT-TagRFP that were cultured in medium containing ML-7 , to block acto-myosin constriction . In such conditions , few cells delaminated and exhibited sub-apical constrictions or centrosome translocation within 6 hr ( Figure 8C , Figure 8—video 3 , 5/31 cells in 9 slices ) . To determine whether centrosome translocation also required active microtubules , slices transfected with the same constructs were exposed to 10 µM taxol , and again fewer cells exhibited centrosome translocation and abscised within 6 hr ( Figure 8D , Figure 8—video 4 , 4/24 cells in 12 slices ) compared with DMSO control conditions ( Figure 8E , Figure 8—video 5 , 9/26 cells in 14 slices ) . These experiments indicate that centrosome translocation and hence its retention in the newborn neuron depends on both microtubule turnover and acto-myosin constriction . The centrosome is important for subsequent morphogenesis of the newborn neuron , but it is unclear whether it is also involved in the delamination process . Indeed , while the centrosome has been implicated in the final stages of cytokinetic abscission ( Piel et al . , 2001 ) it is also possible that ablating this organelle might hasten loss of microtubule-actin/cadherin interactions and so trigger delamination . To investigate the involvement of the centrosome in this process , this structure was disrupted using chromophore assisted light inactivation ( CALI ) mediated by the phototoxic fluorescent protein KillerRed ( Bulina et al . , 2006 ) linked to the pericentrin derived PACT domain . To verify centrosome disruption using this approach , cells were first transfected with PACT-KillerRed and PACT-YFP . Following irradiation with green light , we observed photo-bleaching of the PACT-KillerRed labelling and a corresponding reduction in PACT-YFP labelling ( Figure 9A , 5/5 cells in 5 slices ) , indicating that photoactivation of KillerRed compromised neighbouring centrosomal protein complexes . Conversely , cells transfected with PACT-TagRFP and PACT-YFP and exposed to the same regime did not display reduced YFP labelling ( Figure 9B , Figure 9—video 1 , 25/25 cells in 4 slices from 4 embryos ) , supporting the conclusion that CALI mediated by PACT-KillerRed targeted centrosomal proteins . To assess the functional significance of this manipulation we then carried out this CALI experiment and monitored PACT-KillerRed and production of EB3-GFP comets . This revealed a dramatic reduction in the number of comets ( assessed at a 3 hr time point post-CALI ) ( 22 cells in 7 explants from 7 embryos , Figure 9E ) indicating that this regime significantly compromised centrosome-mediated microtubule nucleation . We then performed CALI on cells transfected with PACT-KillerRed , GFP-GPI and pCIG-Neurog2 that were poised to delaminate . This resulted in fewer cells detaching from the ventricular surface during the subsequent 8 hr imaging period ( 3/12 cells , 25% , in 5 slices Figure 9C , Figure 9—video 2 ) , compared with control PACT-TagRFP transfected cells ( 15/25 cells , 60% , in 10 slices , Figure 9D , Figure 9—video 3 ) . This suggests that centrosome-mediated microtubule nucleation is required for delamination .
One of the major challenges in neural development as well as cell biology is to elucidate the mechanisms regulating cytoskeletal interactions that direct neuroepithelial integrity and neuronal morphology . We provide evidence here for a microtubule wheel-like organization nucleated by the centrosome of the primary cilium in neuroepithelial apical end-feet and for conservation of this configuration across species and regions of the central nervous system . A similar wheel-like arrangement of microtubules has been observed in kidney epithelial ( MDCK ) cells in vitro and cochlear epithelial cells ( Bellett et al . , 2009 ) . Here centrosomal microtubules were orientated with plus-ends towards the AJ ( Bellett et al . , 2009 ) and by tracking the trajectories of EB3-GFP comets we observed a similar configuration in neuroepithelial cells . One explanation for this structure is the recruitment of microtubule plus-ends by AJs/cell cell contact , which has been demonstrated in several epithelial cell lines in vitro ( Stehbens et al . , 2006; Waterman-Storer et al . , 2000 ) . However , in myoblasts , microtubules are directed towards cell contacts by their plus-ends , and here they are then locally repelled at N-cadherin adhesion sites ( Plestant et al . , 2014 ) . This indicates that AJ capture of microtubules is context dependent; indeed this can involve association with minus- rather than plus-ends ( Meng et al . , 2008 ) and that other mechanisms might also account for plus-end growth towards the cell periphery . We show here that in neuroepithelial cells microtubule wheel-like ‘rim’ microtubules interface with the actin cable and that this configuration is generated by dynamic centrosome generated microtubules that bend and grow along the actin cable . This may reflect bio-physical properties of microtubules when they encounter the epithelial cell periphery ( Gomez et al . , 2016 ) and/or regulation by proteins transported by microtubules ( Mata and Nurse , 1997 ) , but it also suggests that interaction between these two cytoskeletal components influences overall microtubule conformation . Our data support such a regulatory relationship , demonstrating microtubule depletion in the apical end-foot following inhibition of actin polymerisation and increased accumulation of actin within the cell following depletion of microtubules; indicating that microtubules regulate actin localisation , although levels of actin at the adhesion belt were unaffected in the timeframe of our assay . Such interactions may be mediated directly by proteins that bind actin and microtubules , these may include formins , IQGAP , dynein/dynactin complex and unconventional myosins as well as Drebrin ( Bazellières et al . , 2012; Brown , 1999; Geraldo et al . , 2008; Goode et al . , 2000; Merriam et al . , 2013; Rodriguez et al . , 2003; Trivedi et al . , 2017 ) . We demonstrate here that Drebrin is localised in apical end-feet of neuroepithelial cells in a distribution similar to that in apical intestinal epithelia ( Bazellières et al . , 2012 ) that includes the sub-apical actin cable , which we show is also aligned with apical microtubules . Drebrin is therefore in a position to link and so coordinate changes in the acto-myosin cytoskeleton and microtubules during neuronal delamination . Furthermore , Drebrin knock-down clearly indicated that this protein is required for neuronal delamination . Experiments should now be focused on elucidating Drebrin dynamics during this process in relation to those of actin and microtubules . In particular , it will be important to establish whether Drebrin serves to direct EB3 comets emerging from the centrosome to actin cable and so create the interface between apical microtubules and sub-apical actin , much as observed during neuronal cell nucleokinesis and migration movements ( Trivedi et al . , 2017 ) . Drebrin binding of the AJ protein Afadin ( Rehm et al . , 2013 ) also raises the interesting possibility that changes in Drebrin localisation as these junctions disassemble , underpins coordinated condensation of the actin and microtubule cytoskeleton during delamination . In other cellular contexts , emphasis has been placed on microtubule regulation of AJs . There is evidence that microtubules promote accumulation of E-cadherin at epithelial cell-cell contacts ( Stehbens et al . , 2006; Waterman-Storer et al . , 2000 ) , but this did not reflect a role in conveying E-cadherin to the cell surface ( Stehbens et al . , 2006 ) . However , these researchers demonstrated a requirement for microtubules for myosin phosphorylation at sites of E-cadherin accumulation in MCF7 cells and so linked microtubules to actin-mediated organisation of AJs . In contrast , N-cadherin transport to the cell membrane requires the microtubule kinesin based motor in a range of cell types ( Mary et al . , 2002; Teng et al . , 2005 ) and neuroepithelial cells in mice mutant for the KIF3 motor complex protein KAP3 , lack membrane localised N-Cadherin ( Teng et al . , 2005 ) . Our data demonstrate that within an hour of microtubule depletion N-Cadherin levels drop dramatically at AJs , consistent with microtubule transport of N-cadherin in the neuroepithelial end-foot . Importantly , actin is required to maintain these apical microtubules and both actin and microtubules maintain the AJs , so actin may act directly and/or indirectly to promote these junctions . Unlike nocodazole treatment , acute inhibition of actin filament assembly reduced actin levels at the adhesion belt and resulted in a smaller end-foot size and so indicated that it is the actin cable that determines apical end-foot dimensions . This delicately balanced apical cytoskeletal architecture changes dramatically as newborn neurons delaminate from the neuroepithelium . This involves the process of apical abscission , which takes place following N-cadherin downregulation ( Das and Storey , 2014; Rousso et al . , 2012 ) . We show here that this includes enrichment of microtubules as well as actin in a composite tunnel-like configuration at the presumptive abscission site . It is interesting that blocking microtubule growth with taxol , while not abolishing acto-myosin contractility , interferes with stable accumulation of actin and that this correlates with reduced cell delamination . This regulatory relationship appears similar to that of the central spindle during cytokinesis , which specifies assembly of the acto-myosin ring by delivering the small GTPase RhoA to the equatorial cortex , that in turn triggers local actin polymerisation and acto-myosin contractility ( Eggert et al . , 2006; Piekny et al . , 2005 ) . This relationship is also consistent with failure to disassemble AJs and impaired acto-myosin constriction in calcium-free conditions ( which disrupt trans-cadherin dimers ) in renal and intestinal cells treated with taxol in vitro , ( Ivanov et al . , 2006 ) ; which additionally suggests a further role for active microtubules in AJ disassembly . Importantly , downregulation of N-Cadherin during neuronal delamination involves not simply transcriptional repression downstream of the neurogenesis transcription factor cascade ( Rousso et al . , 2012 ) , but also mechanism ( s ) that remove N-Cadherin protein , as plasmid driven N-Cadherin is attenuated by such proneural gene activity ( Das and Storey , 2014 ) . One possibility is that rearrangement of apical microtubules during apical abscission may reduce microtubule-AJ association and so further facilitate loss of N-cadherin protein . This may additionally involve regulation of endocytosis/cadherin turnover and there is evidence that actin can also influence this process ( Cavey and Lecuit , 2009; Georgiou et al . , 2008; Ivanov et al . , 2004; Izumi et al . , 2004; West and Harris , 2016 ) . For example , in a cell free assay trans-acting E-Cadherin activates the actin Rac1/Cdc42/IQGAP1 pathway that inhibits E-Cadherin endocytosis and so maintains AJs ( Izumi et al . , 2004 ) ; when such cell-cell interactions are lost then cadherin endocytosis increases . This mechanism is consistent with the phenotype of Cdc42 deletion in the developing mouse cortex , which leads to loss of AJs and mis-localisation of neuroepithelial cells away from the ventricular/apical surface ( Cappello et al . , 2006 ) and with the involvement of heterotopia-associated genes FilaminA and ARFGEF2/BIG2 in endocytosis ( Sheen , 2014 ) . In previous work , we established that acto-myosin constriction was required for apical abscission and here we show that inhibition of acto-myosin activity with ML-7 also blocks accumulation of microtubules at the presumptive abscission site . Together with the requirement for microtubules for stable actin accumulation , these findings suggest that active actin is upstream of microtubule conformational change during this process and that these microtubules then act back to promote effective acto-myosin constriction . Importantly , these data demonstrate that microtubules and actin continue to influence each other even when N-Cadherin/AJs are disassembled in a delaminating cell , further supporting involvement of cross-linking proteins which directly coordinate these cytoskeletal components . An intriguing possibility is that microtubules act here during delamination to augment myosin II phosphorylation , as reported at cell-cell contacts in MCF7 cells in vitro ( Stehbens et al . , 2006 ) . The continued generation of radial comets from the centrosome in cells with small apical end-feet suggests that microtubule nucleation persists as the actin cable constricts and that this may result in formation of the microtubule/actin tunnel-like configuration through which the centrosome eventually passes . This is supported by our finding that both acto-myosin contractility and microtubule turnover are required for centrosome translocation . Furthermore , by specifically compromising centrosome-mediated microtubule nucleation using targeted CALI , we demonstrate that delamination requires centrosome generated microtubules . To our surprise , we further found that centrosome translocation takes place late in the abscission process , in highly constricted cells . Together these findings suggest a mechanism which places the centrosome at the centre of the abscission process and its own retention during neuronal delamination ( Figure 10 ) . This sequence of events has some similarity to that taking place during cytokinesis observed in Hela cells ( Piel et al . , 2000; Piel et al . , 2001 ) ; here , following cleavage furrow and midbody formation , movement of the mother centriole into the midbody bridge triggers release of central spindle microtubules , while disassembly of the actin ring and plasma membrane scission take place after it moves away ( Piel et al . , 2000 ) . Furthermore , experiments which compromise the centrosome inhibited final cytokinetic abscission ( Piel et al . , 2001 ) or , in our experiments , neuronal delamination and this suggests that the centrosome provides molecular cues that prompt common final abscission steps . A critical function for the centrosome in neuronal delamination predicts that mouse mutants affecting the centrosome should exhibit heterotopias in which neurons remain ectopically attached in the region of the ventricle . Phenotypes in such mice vary depending on which centrosomal gene is targeted as well as the timing and extent of gene loss ( Buchman et al . , 2010; Insolera et al . , 2014; Lizarraga et al . , 2010 ) . However , Sas4/Cenp2 mutant mouse cortex exhibits mis-localisation of mitotically stalled neural progenitors away from the ventricle and also some neuronal heterotopias ( Insolera et al . , 2014 ) , consistent with the findings reported here following compromise of the centrosome specifically in presumptive neurons . Cell delamination from within epithelial sheets is a fundamental cell behaviour linked to both differentiation and disease ( e . g . Kesavan et al . , 2014; Slattum and Rosenblatt , 2014; Nikitas and Cossart , 2012; Vasioukhin , 2012 ) . Our data uncover novel cytoskeletal architecture and cell biological mechanisms that mediate this process in the neuroepithelium . It is important now to determine whether these cytoskeletal configurations and regulatory relationships are conserved in other cell types and if they are perturbed in pathological contexts . Indeed , our findings are consistent with recent work demonstrating microtubule network remodelling prior to centrosome reorientation in cells undergoing EMT-like polarity inversions ( Burute et al . , 2017 ) . It therefore seems likely that the apical microtubule-actin alignment uncovered here is a common feature of epithelial cells and that the interdependency of effective acto-myosin constriction and dynamic microtubules during apical constriction is a shared mechanism which may ensure retention of the apically localised centrosome characteristic of many epithelial cell types . These findings further show that delamination is an active process downstream of AJ loss and may open up new opportunities to manipulate delamination by targeting context specific proteins that orchestrate actin and microtubule interactions .
Fertilized chicken ( Gallus gallus domesticus ) eggs were obtained from Winter Egg Farm ( Hertfordshire - Royston SG8 7RF , UK ) and incubated at 38°C to Hamburger and Hamilton stages 10–12 . This was followed by neural tube electroporation of plasmids as described previously ( Das et al . , 2012 ) . Minimal plasmid concentrations were used to enable visualisation of the marker being analysed ( typically within the range of 25–100 ng/μl ) . Only cells that expressed low levels of the markers were chosen for subsequent analysis . EMTB-GFP was a kind gift from Professor WA Harris , University of Cambridge , UK ( Norden et al . , 2009 ) , F-tractin-mKate2 from Alwyn Dady , University of Dundee , UK , Stathmin-GFP from Lynne Cassimeris ( Addgene plasmid # 86782 ) , Drebrin-shRNA , scrambled control , Drebrin-YFP and Drebrin-mCherry constructs from Dr John Chilton ( University of Exeter , UK ) , PACT-KillerRed was generated by replacing TagRFP in PACT-TagRFP with KillerRed on an AgeI/NotI fragment . The KillerRed construct was obtained from Evrogen ( FP962 ) . Hamburger and Hamilton Stage 17–18 embryos were fixed in 4% paraformaldehyde and equilibrated overnight in 30% sucrose at 4°C . These were then embedded in 1 . 5% LB agar ( Sigma , L7025 ) and 5% sucrose , dissolved in MilliQ water . Mounted tissue was dehydrated again for 24 hr in 30% sucrose and snap frozen on dry ice . 20 μm thick sections were then collected using a Leica cryostat ( maintained at −25°C ) . To visualise endogenous microtubules , the spinal cord region of E3 chick embryos were fixed with pre-warmed ( 37°C ) PHEMO fix solution ( 68mMPIPES , 25mMHEPES , 15 mM EGTA , 3 mM MgCl2 , 3 . 7% PFA , 0 . 05% Glutaraldehyde , 0 . 5% tritonX ) for 40 min , washed twice with PHEMO buffer ( ( 68 mM PIPES , 25 mM HEPES , 15 mM EGTA , 3 mMMgCl2 , 10% [v/v] DMSO , pH 6 , with 10M KOH ) and quenched with 100 mM Glycine for 60 min ( Wagstaff et al . , 2008 ) before being equilibrated in 30% sucrose overnight . For immunofluorescence of EMTB-GFP , neural tubes were fixed with pre-warmed ( 37°C ) 4% PFA for 30 min . For en-face imaging of endogenous microtubules , the neural tube of E3 chick or mouse embryos ( E12 . 5 ) was halved sagittaly ( dorsoventrally ) along the ventricle and fixed in pre-chilled ( −20°C ) 100% methanol for 10 min at −20°C . To investigate the effect of microtubule depolymerisation and actin polymerisation inhibition on fixed tissue , neural tube explants for en face imaging were incubated in pre-warmed neurobasal medium containing nocodazole ( 8 . 5 μM , Calbiochem , CAS 31340-18-9 ) for 1 hr ( in this explant assay , microtubule depolymerisation is not observed at 30 min Nocodazole treatment , data not shown ) or latrunculin-A ( 1 μM , Abcam , ab144290 ) for 15 min ( severe tissue collapse at 20 min incubation , data not shown ) . They were then fixed in PHEMO fix solution for 30 min and processed for immunofluorescence imaging in whole mount . For all fixation methods , E3 embryos were handled in pre-warmed ( 37°C ) Leibovitz’s L-15 media ( ThermoFisher , 11415049 ) to maintain microtubule integrity . E12 . 5 Mouse tissue for en-face imaging was blocked overnight with donkey anti-mouse IgG ( 1:200 , Jackson Immunoresearch , 715-005-151 ) . Primary antibody dilutions in blocking buffer ( 0 . 1% Triton-X-100% and 1% heat inactivated donkey serum , in PBS ) : Acetylated alpha tubulin ( Sigma , T7451; RRID:AB_609894 ) 1:150 , alpha tubulin ( YL1/2 ) 1:200 , alpha tubulin ( Abcam , ab7291 ) 1:150 , N-Cadherin ( ThermoFisher , 13–2100; RRID:AB_2533007 ) 1:300 , GFP ( Abcam , ab6673; RRID:AB_305643 ) 1:500 , IFT88 ( Proteintech , 13967–1-AP; RRID:AB_2121979 ) 1:200 , γ-tubulin ( Sigma , T5326;RRID:AB_532292 ) 1:300 , Drebrin ( Abcam , ab11068; RRID:AB_2230303 ) 1:200 . All secondary antibodies used were Alexa Fluor conjugates at 1:500 ( Donkey anti-goat 488 [ThermoFisher , A-11055; RRID:AB_2534102] , Donkey anti-rat 568 [Abcam , ab175475; RRID:_AB2636887] , Donkey anti-rabbit 568 [ThermoFisher , A-10042; RRID:AB_2534017] , Donkey anti- mouse 488 [ThermoFisher , A-21202; RRID:AB_141607] ) . Actin was stained with conjugated CF640R Phalloidin ( Biotum , 00050 ) . Sections were mounted on Prolong Gold antifade mountant ( ThermoFisher , P36930 ) . Neural tube expants were mounted in 0 . 6% low gelling temperature agarose ( Sigma , A9045 ) . Images were acquired using a 40 × 1 . 3 NA or 60 × 1 . 42 NA objective on a Deltavision Core microscope system ( Applied Precision LLC , Issaquah , WA ) . Cover-slips of 0 . 17 mm thickness ( no . 1 . 5 ) were coated with poly-l-lysine ( Sigma , P8920 ) for 30 min at 37°C , washed twice with MilliQ water and left to dry overnight at room temperature . Cryosections of 20 μm thickness were mounted directly on the cover-slips . The EMTB-GFP and F-tractin-mKate2 fluorescent signals were amplified with anti-GFP ( Abcam , ab6673 ) and anti-tRFP ( Evrogen , AB233; RRID:AB_2571743 ) primary antibodies ( both at 1:300 ) , respectively . Secondary antibodies were conjugated with Alexa 488 and Alexa 568 ( ThermoFisher , A-11055 , ThermoFisher , A-10042 ) . Tissue sections were mounted on Slowfade Gold antifade ( ThermoFisher , S36936 ) or Prolong Diamond ( ThermoFisher , P36965 ) mounting media . Structured illumination microscopy was carried out on the OMX Blaze system ( GE Healthcare ) equipped with a UPlanSApochromat 63 × 1 . 42 NA , oil immersion objective lens ( Olympus , Center Valley , PA ) , scientific CMOS camera ( PCO AG , Germany ) and a 488 nm solid-state laser . Samples were illuminated by a coherent scrambled laser light source that had passed through a diffraction 10 . 755410 . 755410 . 7554grating to generate the structured illumination by interference of light orders in the image plane to create a 3D sinusoidal pattern , with lateral stripes approximately 0 . 2 µm apart . The pattern was shifted laterally through five phases and through three angular rotations of 60° for each Z-section , separated by 0 . 125 µm . Exposure times were typically between 10 and 50 ms , and laser power was adjusted to achieve optimal intensities of between 500 and 1000 counts in a raw image of 15-bit dynamic range , at the lowest possible laser power to minimize photo bleaching . Raw images were processed and reconstructed to reveal structures with greater resolution ( Gustafsson et al . , 2008 ) implemented using SoftWorx , ver . 6 . 0 ( Applied Precision , Inc . ) . The channels were then aligned in x , y , and rotationally using predetermined shifts as measured using 100 nm TetraSpeck ( Invitrogen ) beads with the SoftWorx alignment tool ( Applied Precision , Inc . ) . STED imaging was carried out using a Leica Microsystems TCS SP8 STED system equipped with a 100 × 1 . 4 NA oil immersion STED objective . Images in the green channel were acquired using a 488 nm excitation laser and 592 nm depletion laser . Images in the red channel were acquired using a 568 nm excitation laser and 660 nm depletion laser . Z-sections were separated by 0 . 2 µm and images were scanned at 10 Hz using 2x line averaging . The resulting images were deconvolved using Huygens Professional ( Scientific Volume Imaging ) . Embryonic slice culture was carried out as described previously ( Das et al . , 2012 ) . Briefly , chick neural tubes were electroporated at Hamburger and Hamilton stage 10–12 and incubated for 18 hr . Transverse spinal cord slices were obtained from the trunk region between the wing and leg buds and embedded in collagen ( Corning , 354236 ) ( supplemented with 0 . 1% acetic acid , 5x L-15 medium [ThermoFisher , 41300] and 7 . 5% sodium bicarbonate [ThermoFisher , 25080094] ) in poly-D-lysine coated glass-bottomed petri-dishes ( World Precision Instruments , FD35-PDL-100 ) as described previously . For en face imaging the same region of the neural tube was halved dorso-ventrally along the ventricle . One side was discarded and part of the other intact side ( 4–5 somites long ) including the overlying somites was embedded , with the apical end-feet facing the glass of the dish . Slices embedded in collagen were allowed to recover for three hours in Neurobasal medium ( ThermoFisher , 12348017 ) supplemented with B-27 ( ThermoFisher , 17504044 ) , glutamax ( ThermoFisher , 35050038 ) and gentamicin ( ThermoFisher , 15750037 ) at 37°C before imaging was started . For inhibitor experiments the medium was replaced with warmed medium containing one of the following small molecules at the specified final concentration or their controls: nocodazole ( 8 . 5 μM , Calbiochem , CAS 31340-18-9 ) , taxol ( 10 μM , Sigma , T7191 ) , ML-7 ( 20 μM , Sigma , I2764 ) , DMSO ( Sigma ) or H2O at the start of imaging . Time-lapse imaging of embryo slices was performed using a Deltavision Core microscope system in a WeatherStation environmental chamber maintained at 37°C . ( GE Healthcare ) . Imaging was limited to minimal exposure times ( 50–100 milliseconds ) to detect low fluorescence levels ( Das and Storey , 2014; Das et al . , 2012 ) . Image acquisition was performed using an Olympus 40 × 1 . 3 NA oil immersion objective or an Olympus 40 × 1 . 25 NA silicone oil immersion objective , a solid stated LED light source and a CoolSnap HQ2 cooled CCD camera ( Photometrics ) . Unless otherwise stated , 33–34 optical sections spaced 1 . 5 μm apart were acquired for each slice at 5–10 min intervals ( exposure time 5–50 milliseconds for each channel , 512 × 512 pixels , 2 × 2 binning ) . For en face imaging of EB3-GFP comets , 5–8 optical sections spaced 0 . 5 μm apart were acquired at ~1 . 5–3 . 0 s intervals ( exposure time of 150–200 milliseconds for the EB3-GFP comets ) . For the KillerRed and its control experiments , each slice was exposed to a total of 15 min of green light irradiation . Images were deconvolved using the SoftWorx image processing software . The position of the apical surface at each time point was monitored by acquiring a bright-field reference image at the middle of the z-stack . Trail movies of EB3-GFP comets were generated out using the SoftWorx image processing software . For the measurement of EB3-GFP and F-tractin-mKate2 inter-peak distance , a line of 1 μm was drawn across the EB3-GFP comet and fluorescent intensities measurements were carried for both GFP and mKate2 using the FIJI version of the ImageJ software suite ( Schindelin et al . , 2012 ) . The data were then fitted to Guassian curves on FIJI ( Analyse→ Tools→ Curve Fitting ) and the distance between each EB3-GFP and F-tractin-mKate2 pair calculated where fluorescent intensity was the highest ( inter-peak distance , Figure 2D–F ) . All measurements of fluorescent intensities were carried out using the FIJI software ( Schindelin et al . , 2012 ) . For proper comparison of fluorescence intensities in Figure 3 , the same exposure times were used for DMSO control and the small molecule treatments . For the measurement of the grey scale values of N-Cadherin or actin fluorescence intensity in Figure 3 , a straight line of 2 μm ( Latrunclin-A experiments ) or 4 μm ( Nocodazole experiments ) μm was drawn across the adhesion belt of two cells . Background fluorescence , using the freehand tool , was obtained by measuring the mean grey scale value of the area of one of the cells , defined by the N-Cadherin localisation ( excluding the adhesive belt region ) . The same N-Cadherin defined area was used to obtain the measurement of the mean grey scale value of tubulin fluorescence . Furthermore , the N-Cadherin localisation was used , including the adhesive belt region , to measure the end-foot area ( polygon tool ) . Mean background fluorescence for tubulin was obtained by taking measurements within mitotic cells , before reaching the mitotic microtubules along the Z-axis . For Figure 1—figure supplement 1 and Figure 7C , a straight line ( of 2 and 1 μm , respectively ) , across the region of interest was used for the measurement of the grey scale values . The values for each channel were then normalised to the highest value set as 1 . Graphs were plotted accordingly . To calculate the area under the curve in Figure 3C , D , H and I the following formula was used , ( Y1 + Y2 ) /2 *dx where Y1 is the normalised fluorescence intensity at one point , Y2 is the normalised fluorescence intensity of the following point and dx is the distance , defined by the pixel size . For each cell , the total area under the curve is calculated by adding all the values obtained . For the area that corresponds to the adhesion belt , the middle ten values were added . For Figure 6 , presumptive neurons in the right configuration for abscission , mis-expressing pCIG-Neurog2 , were used for the fluorescence intensity measurements . As established , the majority of such cells , treated with Taxol or ML-7 do not abscise and the fluorescence intensity levels of EMTB-GFP or F-tractin-mKate2 were compared to cells in DMSO conditions . The mean grey value of fluorescence intensity , on maximum intensity projections , was measured every thirty minutes and normalised to background levels . For control treatments , the seventh measurement corresponded to the abscission time ( 0 min ) . The mean inter-peak distance ( Figure 2F ) was compared between time-points using the paired t-test . The t-test was used to compare the mean area under the curve , the normalised tubulin fluorescence intensity and the apical end-foot area between treatments for Figure 3 . The values obtained for each of the above measurements are expected to follow a normal distribution ( continuous data ) . In Figure 6 , comparisons of normalised fluorescent intensity trends between small molecule treatments and their respective controls , over time , were performed on SigmaPlot software using the 2-way ANOVA test . | The brain and spinal cord begin as a tube that runs the length of the developing embryo . This tube made from cells called neural progenitors , which can divide to generate adult nerve cells . As nerve cells are born they detach from their neighbours , in a process called delamination before migrating away . Though the delamination of nerve cells is important for the formation of the nervous system , scientists do not fully understand how proteins inside cells work together to release the newborn nerve cell from its neighbours . Two major components of the process are proteins called actin and tubulin , which form complex structures known as acto-myosin cables and microtubules respectively . Acto-myosin cables must contract during delamination , but the role of the microtubules is unclear . Kasioulis et al . examined the microtubules in chick and mouse neural tube cells during delamination using fluorescent labels to mark key molecules and small molecule inhibitors to selectively block different activities . A combination of live tissue and super-resolution imaging were used to reveal the dynamics of the delamination process . The experiments revealed a wheel-like configuration of microtubules that lined up with the acto-myosin cable . Actin maintained the microtubules , which in turn maintained the acto-myosin cable . As newborn neurons delaminated , the actin cable constricted and the microtubules condensed , forming a tunnel that allowed a structure that organises the microtubules – the centrosome – to move , and the cell to detach . A protein called Drebrin , which links actin to microtubules , was identified as a potential coordinator of the process . These findings not only further our understanding of nervous system development , but may also shed light on the development of human diseases . Failure of delamination can lead to a spectrum of disorders , including epilepsy , dyslexia and intellectual disability . Cell detachment is also important in other developmental processes , as well as in the spread of cancer cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2017 | Inter-dependent apical microtubule and actin dynamics orchestrate centrosome retention and neuronal delamination |
Local regulation of synaptic efficacy is thought to be important for proper networking of neurons and memory formation . Dysregulation of global translation influences long-term memory in mice , but the relevance of the regulation specific for local translation by RNA granules remains elusive . Here , we demonstrate roles of RNG105/caprin1 in long-term memory formation . RNG105 deletion in mice impaired synaptic strength and structural plasticity in hippocampal neurons . Furthermore , RNG105-deficient mice displayed unprecedentedly severe defects in long-term memory formation in spatial and contextual learning tasks . Genome-wide profiling of mRNA distribution in the hippocampus revealed an underlying mechanism: RNG105 deficiency impaired the asymmetric somato-dendritic localization of mRNAs . Particularly , RNG105 deficiency reduced the dendritic localization of mRNAs encoding regulators of AMPAR surface expression , which was consistent with attenuated homeostatic AMPAR scaling in dendrites and reduced synaptic strength . Thus , RNG105 has an essential role , as a key regulator of dendritic mRNA localization , in long-term memory formation .
The formation of long-term memory , but not short-term memory , requires protein translation in neurons ( Bramham and Wells , 2007; Costa-Mattioli et al . , 2009 ) . Gene knockout and administration of drugs for global translational regulators influence , that is , enhance or impair , long-term memory formation ( Costa-Mattioli et al . , 2009 ) . Translation in neurons is regulated not only globally but also locally in dendrites near stimulated postsynaptic sites ( Aakalu et al . , 2001; Yoon et al . , 2016 ) . This local translation is involved in the regulation of synaptic plasticity and functions , and mediated by dendritic mRNA transport by ‘RNA granules’ , membrane-less macromolecular assemblies of mRNAs , ribosomes , and RNA-binding proteins ( Kiebler and Bassell , 2006; Weber and Brangwynne , 2012; Kedersha et al . , 2013 ) . Some components of RNA granules have been reported to be associated with brain functions in disease , for example , fragile X mental retardation by FMRP deficiency and neurodegenerative diseases such as amyotrophic lateral sclerosis ( ALS ) and frontotemporal lobar degeneration ( FTLD ) with aggregation of FUS/TLS and TDP-43 in RNA granules ( Lenzken et al . , 2014; Santos et al . , 2014; Ling et al . , 2013 ) . However , little influence on long-term memory formation of knockout in mice for RNA granule components , for example , FMRP , CPEB , Staufen1 , Pumilio-2 , and G3BP1 , has left inconclusive the primary question whether regulation of local translation is required for long-term memory formation ( Consortium TD-BFX , 1994; Berger-Sweeney et al . , 2006; Vessey et al . , 2008; Siemen et al . , 2011; Martin et al . , 2013 ) . RNA granule protein 105 ( RNG105 , also known as caprin1 ) is a major RNA-binding protein in RNA granules . RNG105 promotes the assembly of RNA granules and is responsible for the transport of its binding mRNAs in cultured cells ( Shiina et al . , 2005; Kedersha et al . , 2016; Shiina et al . , 2010 ) . Knockdown and knockout ( KO ) of RNG105 in cultured neurons causes a reduction in the synaptic connections on dendrites and the density of neural networks ( Shiina et al . , 2010; Shiina and Tokunaga , 2010 ) . A heterozygous nonsense mutation in the Rng105/caprin1 gene has been found in a human patient with autism spectrum disorder ( ASD ) , and heterozygous KO of Rng105 gene in mice causes ASD-like behavior ( Ohashi et al . , 2016; Jiang et al . , 2013 ) . These studies suggested the involvement of RNG105 in higher-order brain functions . However , RNG105 homozygous KO mice are neonatally lethal because of respiratory failure ( Shiina et al . , 2010 ) , which has hampered the analysis of the physiological impact of severe RNG105 deficiency on learning and memory in adult mice . Here , we generated RNG105 conditional deletion mice using the Cre/loxP system . The conditional deletion mice ( Camk2a-Cre;Rng105f/f ) were viable , and subjected to in vivo analyses of synaptic function , behavioral tests for learning and memory , and genome-wide profiling of somato-dendritic localization of mRNAs in the hippocampus . The results demonstrated that RNG105 was an essential element of RNA granules for establishing long-term memory , and suggested RNG105-mediated dendritic localization of mRNAs as an underlying mechanism for AMPA receptor ( AMPAR ) scaling , synaptic strength and plasticity , and long-term memory formation .
To investigate physiological functions of RNG105 in adult mice , we generated RNG105 conditional deletion mice by crossing floxed Rng105 mice and Camk2a-Cre transgenic mice for gene deletion in the central nervous system ( Figure 1A–C ) . In Camk2a-Cre;Rng105f/f mice , exons 5–6 were flanked by loxP sequences and deleted by Cre expression ( Figure 1A ) . Because frame shift occurs in the exon 5–6-deleted mRNA , most of the functional domains of RNG105 ranging from the N-terminal coiled-coil domain to the C-terminal RG-rich domain ( a . a . 123–707 ) are deleted ( Shiina et al . , 2005 ) . In addition , most of the exon 5–6-deleted mRNA appeared to be degraded by nonsense-mediated mRNA decay ( NMD ) : in the hippocampus where Cre was highly expressed , expression of Rng105 transcripts from all exons was reduced to the comparable level to that of exon 5–6 transcripts as judged by RNA-seq analysis ( Figure 1—figure supplement 1A ) . Camk2a-Cre;Rng105f/f mice were born in the Mendelian ratio and showed no apparent abnormalities just after birth . Although their growth was retarded during the lactation period , their body weight recovered thereafter to 80–90% of that of control ( Rng105f/f ) mice ( Figure 1D ) . In addition to growth retardation , Camk2a-Cre;Rng105f/f mice were susceptible to death . However , death was spontaneous , and more than 40% of Camk2a-Cre;Rng105f/f mice survived for more than 4 months ( Figure 1E ) . Thus , RNG105 conditional deletion overcame the neonatal lethality of RNG105 conventional KO . Western blotting revealed that the expression of RNG105 protein was markedly reduced in the cerebral cortex and hippocampus , but not so much in the cerebellum of Camk2a-Cre;Rng105f/f mice ( Figure 1F ) . Although the anti-RNG105 antibody can recognize truncated RNG105 protein ( a . a . 1–122 ) encoded by the exon 5–6-deleted mRNA , it did not detect any truncated form of RNG105 in the cerebrum of Camk2a-Cre;Rng105f/f mice ( Figure 1—figure supplement 1B ) . This supported the notion that the exon 5–6-deleted mRNAs were degraded by NMD and hardly any truncated RNG105 protein was expressed in Camk2a-Cre;Rng105f/f mice . Immunostaining of hippocampal slices showed that RNG105 was markedly reduced in the somatic layer ( stratum pyramidale [SP] ) and dendritic layer ( stratum radiatum [SR] ) of pyramidal neurons , confirming the reduction of RNG105 expression in neurons of Camk2a-Cre;Rng105f/f mice ( Figure 1G ) . To investigate the impact of RNG105 deficiency on synaptic function , we first examined the morphology of pyramidal neurons and dendritic spines in the hippocampal CA1 region . The density of hippocampal neurons , as judged from nuclear staining , was not affected in Camk2a-Cre;Rng105f/f mice ( Figures 1G and 2A ) . To trace the morphology of pyramidal neurons , Camk2a-Cre;Rng105f/f mice were crossed with Thy1-GFP transgenic mice . Fluorescence imaging revealed that the length and branching of dendrites were comparable between Rng105f/f and Camk2a-Cre;Rng105f/f mice ( Figure 2B−D ) . The density of spines on dendrites was also equivalent between the genotypes , but the size of spines was smaller in Camk2a-Cre;Rng105f/f mice ( Figure 2E−G ) . Furthermore , the number of mushroom spines was significantly reduced , which suggested that synaptic strength and/or stimulation-dependent plasticity were attenuated in Camk2a-Cre;Rng105f/f mice ( Figure 2E , H ) . We then analyzed structural plasticity of dendritic spines using a two-photon glutamate uncaging technique . Hippocampal neurons from floxed RNG105 ( Rng105f/f ) mice were cultured in dishes and co-transfected with CMV-Cre and mCherry in order to delete the Rng105 gene and trace the morphology of the neurons , respectively . Immunostaining of the neurons indicated that the expression of RNG105 was significantly reduced in mCherry-positive neurons compared with neighboring mCherry-negative neurons ( Figure 3A , B ) . RNG105 deletion did not affect spine density on dendrites , but reduced the size of spines ( Figure 3C , D ) , which was similar to the results in the hippocampal slices . In response to stimulation by glutamate uncaging , single spines close to the uncaging spots were transiently increased in their volume to ~3 fold and sustained an increased state of ~2 . 5 fold over 60 min in Rng105f/f neurons ( Figure 3E , F ) . The increase in spine volume during the sustained phase is translation-dependent ( Nishiyama and Yasuda , 2015; Tanaka et al . , 2008 ) , which was confirmed by cycloheximide addition ( Figure 3F , G ) . In contrast to Rng105f/f neurons , CMV-Cre;Rng105f/f neurons significantly reduced the spine volume during the sustained phase ( Figure 3E−G ) . These results indicated that RNG105 was required for the stimulation-induced structural plasticity of spines in the translation-dependent late-phase LTP . To examine whether RNG105 regulates the function of synapses , we measured electrophysiological responses of hippocampal CA1 neurons to stimulation . First , basal synaptic transmission at CA3-CA1 synapses was recorded . The relationship between the magnitude of input stimulation and the postsynaptic responses ( amplitude and slope of field excitatory postsynaptic potential [fEPSP] ) indicated that RNG105 deficiency reduced both the amplitude and slope of fEPSP to about half of those of control mice ( Figure 4A , B ) . Paired-pulse ratio ( PPR ) was significantly decreased after LTP induction in Camk2a-Cre;Rng105f/f mice , but it was not significantly different between the genotypes ( Figure 4C , D ) . These results indicated that postsynaptic responses to stimulation were markedly reduced in Camk2a-Cre;Rng105f/f mice , suggesting downregulation of AMPARs by RNG105 deficiency . We also measured LTP at CA3-CA1 synapses . Theta-burst-induced LTP in CA1 neurons revealed that fEPSP was increased from baseline by ~100% , which was comparable between Rng105f/f and Camk2a-Cre;Rng105f/f mice ( Figure 4E , F ) . However , because the absolute amplitude and slope of basal fEPSP in Camk2a-Cre;Rng105f/f mice were about half of those in Rng105f/f mice , fEPSP after LTP induction in Camk2a-Cre;Rng105f/f mice was also about half of that in Rng105f/f mice ( Figure 4E , F ) . As a result , the absolute amplitude and slope of fEPSP in Camk2a-Cre;Rng105f/f mice reached , even after LTP induction , only the same level as the basal fEPSP in Rng105f/f mice ( Figure 4E , F ) . After the induction of LTP , we often observed epileptic-like repetitive patterns in fEPSP waveforms , which may be a back-propagation of repetitive action potentials , suggesting aberrant neuronal excitation after LTP induction in Camk2a-Cre;Rng105f/f mice ( Figure 4C , inset ) . Together , in Camk2a-Cre;Rng105f/f mice , fEPSP amplitude was reduced both in the steady state and after LTP induction , which may be related to abnormal neuronal excitation and impaired structural plasticity of spines . Next , we examined whether RNG105 is required for normal behavior in learning and memory tasks . In open-field test , there were no differences in exploratory horizontal locomotion among Rng105f/f , RNG105 hetero-deletion ( Camk2a-Cre;Rng105f/+ ) and Camk2a-Cre;Rng105f/f mice in the initial trial ( Figure 5A ) . The exploratory activity of Rng105f/f and Camk2a-Cre;Rng105f/+ mice was decreased over 3 days , suggesting that the mice were habituated to the novel place with trials . In contrast , Camk2a-Cre;Rng105f/f mice did not show such experience-dependent reduction in exploratory activity , suggesting that Camk2a-Cre;Rng105f/f mice had problems in becoming habituated to a novel environment ( Figure 5A ) . Habituation is a complex behavior , which is impaired by several factors such as memory deficits , incomplete initial exploration of the entire area because of high level of anxiety , and low level of locomotor activity ( Bolivar , 2009 ) . Among them , the latter two factors were not likely reasons for the impaired habituation in Camk2a-Cre;Rng105f/f mice , because the initial exploratory activity in the open field test was normal and anxiety-like behavior in the light/dark transition test was also normal in Camk2a-Cre;Rng105f/f mice ( Figure 5A; Figure 5—figure supplement 1A ) . We further conducted a novel object recognition test , which is another test to assess habituation ( Figure 5—figure supplement 1B ) . In the first session , mice were habituated to two identical objects and showed no biased preference for either object . In the second session in which one of the objects was replaced by a novel one , Rng105f/f mice showed an increased preference for the novel object . In contrast , Camk2a-Cre;Rng105f/f mice did not show such an increased preference for the novel object ( Figure 5—figure supplement 1B ) . These results supported the notion that habituation , a form of learning and memory , was impaired in Camk2a-Cre;Rng105f/f mice . In the rotarod test , Rng105f/f and Camk2a-Cre;Rng105f/+ mice showed increasing retention time on the rod and decreasing number of falls from the rod over 3 days , indicating that the mice learned the rotarod skill day by day ( Figure 5B ) . Camk2a-Cre;Rng105f/f mice showed indistinguishable performance from the other genotypes , indicating that RNG105 conditional deletion did not affect motor skill learning . This was consistent with a mild reduction in RNG105 proteins in the cerebellum of Camk2a-Cre;Rng105f/f mice , as well as a notion that RNG105 may be simply not required for this cerebellum-dependent form of learning and memory . To test spatial learning and memory , Morris water maze was conducted . Mice were first subjected to a visible platform test . Escape latency of Camk2a-Cre;Rng105f/f mice was longer than that of Rng105f/f and Camk2a-Cre;Rng105f/+ mice , but significantly reduced over trials ( Figure 6A ) . These results indicated that Camk2a-Cre;Rng105f/f mice required more training than the other genotypes , but they had escape motivation , vision , and motor skills sufficient to accomplish the task . The mice were then subjected to a hidden platform test , which necessitates hippocampus-dependent long-term spatial memory ( Figure 6B ) . Repeated trials enabled Rng105f/f and Camk2a-Cre;Rng105f/+ mice to learn the platform location and escape on the platform faster than before the trials . In contrast , the escape latency of Camk2a-Cre;Rng105f/f mice did not shorten at all over the trials ( Figure 6B ) . Following the last trial , a probe test was conducted without the platform . Rng105f/f and Camk2a-Cre;Rng105f/+ mice intensively searched around the target place ( Figure 6C ) . In contrast , Camk2a-Cre;Rng105f/f mice showed a circular swimming path along the wall and reduced the search time around the target place ( Figure 6C ) . The time that Rng105f/f and Camk2a-Cre;Rng105f/+ mice spent in the target quadrant was significantly longer than in the other quadrants , whereas Camk2a-Cre;Rng105f/f mice markedly reduced the time in the target quadrant compared to the other genotypes ( Figure 6D ) . These severe phenotypes raised a concern that Camk2a-Cre;Rng105f/f mice might be incapable of learning the task . However , if outliers were eliminated ( two Camk2a-Cre;Rng105f/f mice showing maximum values in L and O quadrants [Figure 6D] , p<0 . 01 using Smirnov-Grubbs test ) , statistical significance was detected in the quadrant effect in Camk2a-Cre;Rng105f/f mice ( one-way repeated measures ANOVA , F[3 , 45]=3 . 234 , p=0 . 0309 ) . Consistently , the density plot of swim path showed weak preference of Camk2a-Cre;Rng105f/f mice for the target quadrant ( Figure 6C ) , suggesting Camk2a-Cre;Rng105f/f mice were able to learn the task . Together , these results indicated that RNG105 was critical for the formation of spatial long-term memory . Finally , a contextual fear conditioning test ( passive avoidance ) was conducted . Before receiving foot shock in a dark chamber , all genotypes spent 60–70% of the test time in the dark chamber ( Figure 6E ) . At 5 min after the foot shock , none of the genotypes stayed in the dark chamber , indicating that fear-conditioned learning had took place . However , at 1 day after the foot shock , Camk2a-Cre;Rng105f/f mice spent a significantly longer time in the dark chamber than the other genotypes . Furthermore , at 1 week after the foot shock , Camk2a-Cre;Rng105f/f mice spent as long a time in the dark chamber as before the foot shock ( Figure 6E ) . Because Camk2a-Cre;Rng105f/f mice spent a comparable time to the other genotypes in the dark chamber before the foot shock , as well as showing normal anxiety-like behavior in the light/dark transition test ( Figure 5—figure supplement 1A ) , the increased time in the dark chamber at 1 day and 1 week was considered not to be due to increased anxiety , but reduced long-term memory in Camk2a-Cre;Rng105f/f mice . In another contextual fear conditioning test , in which mice received foot shock in a single chamber and their freezing responses were measured in the same chamber at 5 days after the foot shock , Camk2a-Cre;Rng105f/f mice showed less freezing behavior than Rng105f/f mice ( Figure 6—figure supplement 1 ) . These results indicated that RNG105 was required for long-term fear conditioning memory formation . Besides the impairment in memory formation , Camk2a-Cre;Rng105f/f mice sometimes exhibited seizures during and just after the Morris water maze and contextual fear conditioning tests ( Video 1 ) . This phenotype was reminiscent of the epileptic-like fEPSP appeared after relatively intense stimulation of LTP ( Figure 4C ) . Taken together , behavioral analyses demonstrated an essential role of RNG105 in the formation of long-term memory . To understand the underlying mechanisms by which RNG105 regulates synaptic strength , synaptic structural plasticity and long-term memory formation , we investigated the impact of RNG105 deficiency on mRNA localization in neurons . Although RNG105 is involved in mRNA transport to dendrites in vitro ( Shiina et al . , 2010 ) , it is unknown whether RNG105 regulates somato-dendritic mRNA localization in vivo , and if so , which mRNAs change the dendritic localization in RNG105-deficient mice . First , we extracted somatic and dendritic mRNAs from the hippocampal CA1 region in adult mice . The somas of the CA1 pyramidal neurons are aligned in the SP , from which apical dendrites elongate in the SR ( Figures 1G and 7A ) . We microdissected and isolated each layer for the preparation of soma and dendrites of pyramidal neurons ( Figure 7B , C ) , as demonstrated in previous studies ( Cajigas et al . , 2012; Ainsley et al . , 2014 ) . Next , mRNAs extracted from SP and SR of Rng105f/f mice and Camk2a-Cre;Rng105f/f mice were subjected to RNA-seq analysis ( Supplementary file 1 ) . For each mRNA , relative concentration ( FPKM: read counts normalized by transcript length ) in SR to SP , which we termed the ‘dendritic accumulation index ( DAI ) " , was calculated . Statistical analysis identified SP-enriched ( low DAI ) mRNAs and SR-enriched ( high DAI ) mRNAs . Because the hippocampus contains not only pyramidal neurons but also other types of cells such as interneurons and glial cells , mRNAs also expressed in these cell types were eliminated from the mRNA lists as in the previous study ( Cajigas et al . , 2012 ) ( Supplementary file 2A , B ) . As a result , we identified 1122 dendritically enriched mRNAs ( D-mRNAs ) and 2106 somatically enriched mRNAs ( S-mRNAs ) in control mice ( Figure 7D; Supplementary file 3A−C ) . The D-mRNAs included already known dendritic mRNAs such as Camk2a , Eef1a1 , Dlg4 , Iptr1 , Arc , Shank2 , Homer2 , and Limk1 ( Figure 7D; Supplementary file 3A ) , suggesting that the strategy was appropriate to detect somato-dendritic mRNA distribution pattern . Comparison between Rng105f/f and Camk2a-Cre;Rng105f/f mice revealed that the somato-dendritic distribution pattern of mRNAs was different between the genotypes; the variance of DAI of mRNAs was smaller in Camk2a-Cre;Rng105f/f mice than in Rng105f/f mice , and the gap in DAI between D- and S-mRNAs was narrower in Camk2a-Cre;Rng105f/f mice ( Figure 8A ) . Variance values ( s2 ) of the DAI of D- and S-mRNAs were 1 . 945 for Rng105f/f mice and 1 . 092 for Camk2a-Cre;Rng105f/f mice , which were significantly different ( n = 3 , 228 , F0 = 1 . 781 , p<0 . 005 ) . These results indicated an in vivo role of RNG105 to establish the asymmetric localization of mRNAs in the soma and dendrites . We further analyzed whether changes in the somato-dendritic localization of mRNAs in Camk2a-Cre;Rng105f/f mice were mRNA-selective . For each mRNA , the ratio of DAI in Camk2a-Cre;Rng105f/f mice to Rng105f/f mice was calculated ( Figure 8B ) . The ratio was lower than one for majority of , but not all , the D-mRNAs , indicating that RNG105 deficiency reduced the dendritic localization of many , but specific D-mRNAs . To further address mRNA-selectivity , relation between DAI ( Rng105f/f ) and the ratio of DAI ( Camk2a-Cre;Rng105f/f/Rng105f/f ) for D- and S-mRNAs was analyzed ( Figure 8C ) . D-mRNAs showed large variance and low correlation ( |r| = 0 . 3455 ) . In contrast , S-mRNAs showed smaller variance and high correlation ( |r| = 0 . 7269 ) ( Figure 8C ) . These results indicated that the effect of RNG105 deficiency on dendritic localization of mRNAs varied among D-mRNAs , suggesting RNG105 targets selective D-mRNAs . On the other hand , the relatively uniform effect of RNG105 deficiency on S-mRNAs indicated that the effect was mRNA-non-selective . The apparent increase in DAI of S-mRNAs in Camk2a-Cre;Rng105f/f mice may be mainly a secondary effect of the decrease in DAI of D-mRNAs because the sum of total mRNAs' log ( DAI ) should be nearly zero . To identify the biological categories in which the D-mRNAs are involved , gene ontology ( GO ) enrichment analysis was conducted . First , D- and S-mRNAs in control mice were classified into GO categories ( Figure 8D; Supplementary file 4A−I ) . Major categories of D-mRNAs were ‘regulation of Arf protein signal transduction’ , which included GTPase-activating proteins ( GAPs ) and guanine nucleotide exchange factors ( GEFs ) of small G protein ADP-ribosylation factor ( Arf ) , and ‘structural constituent of ribosomes’ which included ribosomal subunit proteins ( Figure 8D; Supplementary file 4A , B ) . D-mRNAs included in these categories were plotted on MA plots , which indicated that the dendritic accumulation of these mRNAs was comparable to that of well-known dendritic mRNAs ( Figure 9 , cf . Figure 7D ) . Arf is known to regulate membrane trafficking between the cell surface and endosomes , and in particular , Arf6 participates in the surface expression of AMPARs ( D'Souza-Schorey and Chavrier , 2006; Jaworski , 2007; Oku and Huganir , 2013; Zheng et al . , 2015 ) . Arf is also known as a regulator of actin dynamics and dendritic spine formation via Rac1 activation ( D'Souza-Schorey and Chavrier , 2006; Jaworski , 2007 ) . There are various Arf GAPs and GEFs possessing or not possessing a membrane-associated pleckstrin-homology ( PH ) domain , among which the Arf GAPs and GEFs identified here were mostly the PH domain-possessing types and classified in ‘pleckstrin homology’ category with other PH domain-containing proteins ( Figure 8D; Supplementary file 4E ) . The Arf GAPs and GEFs were also classified in ‘GTPase regulator activity’ , which included regulators of other small G proteins such as Ras , Rho , and Rac ( Figure 8D; Supplementary file 4G ) , known to be involved in actin reorganization and spine morphogenesis ( Nishiyama and Yasuda , 2015 ) . ‘Fc gamma receptor-mediated phagocytosis’ , ‘leukocyte activation’ , and ‘chemotaxis’ categories also had high-fold enrichment scores , which contained several overlapping proteins such as PI3 kinase pathway proteins and Rac pathway proteins involved in actin regulation , and extracellular membrane proteins ( Figure 8D; Supplementary file 4D , F , H ) . GO enrichment analysis was further conducted on D-mRNAs whose dendritic localization was reduced in Camk2a-Cre;Rng105f/f mice compared with Rng105f/f mice ( the ratio of DAI < 0 . 8 ) . Major categories with high-fold enrichment scores were the same as above , for example , ‘regulation of Arf protein signal transduction’ , ‘structural constituent of ribosomes’ , and ‘GTPase regulator activity’ , suggesting that RNG105 was responsible for the dendritic localization of mRNAs classified in the major categories ( Figures 8E and 9; Supplementary file 4A−I ) . Reduction in the DAI of Arf regulator mRNAs in Camk2a-Cre;Rng105f/f mice was reminiscent of the results that fEPSP , which is mediated by AMPARs , was reduced in Camk2a-Cre;Rng105f/f mice ( Figure 4 ) . We then examined whether the DAIs of mRNAs encoding other regulatory proteins of AMPARs ( Bassani et al . , 2013; Henley and Wilkinson , 2013 ) were also reduced in Camk2a-Cre;Rng105f/f mice . mRNAs such as Cnih2 , Arc , Dlg4 , Pik3r2 , Camk2a , and Cplx2 were identified as D-mRNAs . Furthermore , their DAIs were reduced in Camk2a-Cre;Rng105f/f mice compared with Rng105f/f mice ( Supplementary file 5A ) . These results indicated that various mRNAs , encoding AMPAR regulators involved in AMPAR surface expression and retention to postsynapses , were localized to dendrites in an RNG105-dependent manner . By contrast , mRNAs encoding AMPAR subunits themselves ( Gria1-4 ) were somatically enriched and their DAIs were not reduced in Camk2a-Cre;Rng105f/f mice ( Supplementary file 5A ) . Notably , the DAI of Gria2 mRNA was markedly increased , rather than decreased , in Camk2a-Cre;Rng105f/f mice ( Supplementary file 5A ) , suggesting that RNG105 influences specifically , if not directly , the dendritic localization of Gria2 mRNA . RNG105 deficiency also influenced , if not directly , the total expression level of some mRNAs , as judged from the S-mRNA concentration ( FPKM ) . mRNAs whose expression was reduced in Camk2a-Cre;Rng105f/f mice included Rng105 itself , and notably , a considerable number of immediate early genes ( IEGs ) such as Fos , Btg2 , Egr1 , Egr4 , Dusp1 , and Arc ( Supplementary file 6 ) . Because the expression of these IEGs was reportedly upregulated by neuronal activation ( Saha et al . , 2011; Iacono et al . , 2013 ) , these results suggested a reduction in neuronal activity by RNG105 deficiency . In addition , the drastic increase in Gria2 mRNA localization to dendrites in Camk2a-Cre;Rng105f/f mice may be also attributed to reduced neuronal activity in RNG105-deficient mice ( Grooms et al . , 2006 ) . The reduction in fEPSP amplitude and dendritic localization of mRNAs for AMPAR regulators suggested that RNG105 regulates AMPAR scaling . In particular , given the reduction in steady-state fEPSP , we hypothesized that RNG105 may be important for homeostatic scaling of AMPARs . To test this , we analyzed GluR1 and GluR2 cell surface expression in response to activity deprivation with TTX and APV in primary cultured neurons from wild-type ( Rng105+/+ ) and RNG105 knockout ( Rng105−/− ) mice . Surface and total GluR1 and GluR2 were immunostained before and after cell permeabilization , and quantified by counting the number and measuring the fluorescence intensity of their puncta in dendrites ( Figure 10 ) . After activity deprivation , the number of surface GluR1 puncta normalized to that of total GluR1 was significantly increased , and the intensity of surface GluR1 tended to be increased , in dendrites of Rng105+/+ neurons . By contrast , the activity deprivation-dependent increase in surface GluR1 expression was not observed in Rng105−/− dendrites ( Figure 10A , C ) . Although homeostatic scaling of GluR2 is controversial ( Isaac et al . , 2007; Wierenga et al . , 2005; Gainey et al . , 2009 ) , our results indicated that the number of GluR2 surface puncta was increased by activity deprivation in dendrites of Rng105+/+ neurons ( Figure 10B , D ) . By contrast , this increase in surface GluR2 puncta was not observed in Rng105−/− neurons , similarly to GluR1 . Comparison of Rng105+/+ and Rng105−/− neurons indicated that the number of surface GluR1 in dendrites was significantly reduced in Rng105−/− neurons ( Figure 10A , C ) , confirming the previous results ( Ohashi et al . , 2016 ) . By contrast , GluR2 surface expression in dendrites was not much reduced by RNG105 deficiency ( Figure 10B , D ) , which could be due to the increase in Gria2 mRNA localization to dendrites by RNG105 deficiency ( Supplementary file 5A ) and/or different surface expression pathways of a fraction of GluR2 from that of GluR1 ( Tanaka and Hirano , 2012 ) . We further conducted biotin labeling of cell surface proteins of cultured neurons followed by immunoblot measurement for GluR1 and GluR2 ( Figure 10—figure supplement 1 ) . GluR1/2 in total cell lysates was detected as a major band of ~100 kDa , which was reduced in amount after biotin labeling , accompanied by an increase in the amount of an upper band ( Figure 10—figure supplement 1A ) . The upper band , but not the lower band , bound to avidin agarose beads , indicating that the upper band was biotin-labeled and mobility-shifted surface GluR1/2 , whereas the lower band was non-labeled intracellular GluR1/2 ( Figure 10—figure supplement 1A−C ) . Because the upper band was also detected in control lysates without biotinylation , the band was considered to contain a non-specific protein ( s ) as well as biotinylated GluR1/2 . The most obvious difference between TTX/APV-treated and untreated neurons was the less amount of the lower GluR1 band in TTX/APV-treated neurons from Rng105+/+ mice , but not from Rng105−/− mice , which suggested that TTX/APV treatment reduced intracellular GluR1 in Rng105+/+ neurons , but not in Rng105−/− neurons ( Figure 10—figure supplement 1B ) . Then we quantified the intensity of the lower band in the cell lysate and the upper band in the avidin agarose-bound fraction , and calculated the ratio of surface to intracellular GluR1 and GluR2 ( Figure 10—figure supplement 1D , E ) . The ratio of surface/intracellular GluR1 was increased in TTX/APV-treated neurons compared to untreated neurons from Rng105+/+ mice . Compared to Rng105+/+ neurons , the ratio of surface/intracellular GluR1 was lower and was not significantly increased by TTX/APV treatment in Rng105−/− neurons ( Figure 10—figure supplement 1D ) . We noted that the GluR1 band intensity was lower both in the total lysates and avidin-bound fractions in Rng105−/− neurons than in Rng105+/+ neurons ( Figure 10—figure supplement 1B , upper panel ) , which may be because of the reduced density of neural networks in Rng105−/− primary cultured neurons ( Shiina et al . , 2010 ) . This weaker band intensity in Rng105−/− neurons made it difficult to identify differences in the surface/intracellular ratio between the genotypes at a glance , but when twice the volume of samples from Rng105−/− neurons was loaded on gels , the difference in the surface/intracellular ratio between the genotypes was obvious: the intensity of the lower GluR1 bands was higher , whereas that of the upper GluR1 bands was lower , in Rng105−/− neurons than in Rng105+/+ neurons ( Figure 10—figure supplement 1B , bottom panel ) . As for GluR2 , although statistical significance was not detected , the surface/intracellular ratio of GluR2 tended to be increased by TTX/APV treatment ( Figure 10—figure supplement 1E ) . These results were consistent with the results of the immunofluorescence imaging of GluR1 and GluR2 . Taken together , these results indicated that RNG105 impacts on homeostatic scaling of AMPARs in dendrites , which is coupled to basal synaptic function and also influences synaptic strength in activated state .
Translation in neurons is required for long-term memory formation , but the relevance of the regulation of mRNA transport and local translation to long-term memory formation has remained unclear . This study demonstrated essential roles of an RNA granule protein RNG105 in the regulation of synaptic strength , structural plasticity of spines , and long-term memory formation . The loss of synaptic strength and long-term memory in Camk2a-Cre;Rng105f/f mice was likely attributed to reduced dendritic localization of mRNAs , which included mRNAs for various regulators of AMPAR surface expression and retention to postsynapses . Consistently , AMPAR homeostatic scaling in dendrites was impaired by RNG105 deficiency , which was related to the decreased setting level of fEPSP amplitude in the steady state . The decreased setting level of fEPSP may reduce excitatory transmission in neurons even after LTP induction , because of unchanged ( not upregulated ) LTP efficiency in RNG105-deficient mice , which will impair long-term memory formation . Together , this study revealed physiological roles of RNG105 in mice , and suggested cellular mechanisms linking RNA granule functions with long-term memory formation . Studies on learning and memory in KO mice of other RNA granule components , for example , FMRP , Pumilio2 , CPEB1 , Staufen1 , G3BP1 , and GLD-2 have been reported . However , these KO mice did not show apparent influence on long-term memory formation in the Morris water maze or contextual fear conditioning ( Consortium TD-BFX , 1994; Berger-Sweeney et al . , 2006; Vessey et al . , 2008; Siemen et al . , 2011; Martin et al . , 2013; Mansur et al . , 2016 ) . Although KO mice of Ataxin-2 , another RNA granule component , showed impaired contextual fear conditioning , their long-term memory in the Morris water maze was normal ( Huynh et al . , 2009 ) . These studies have left unanswered whether RNA granules are involved in long-term memory formation . Compared with these KO mice , RNG105 conditional deletion mice displayed remarkable impairment of long-term memory formation . The different phenotypes between RNG105 conditional deletion mice and the other KO mice of RNA granule components may be attributed to the existence and non-existence of alternative factors . Pumilio1 and 2 , Staufen1 and 2 , and G3BP1 and 2 are paralogs , and the paralog proteins have redundant functions and target an overlapping set of mRNA ( Kedde et al . , 2010; Galgano et al . , 2008; Park and Maquat , 2013; Furic et al . , 2008; Matsuki et al . , 2013 ) . CPEB1 has three paralogs , CPEB2−4 , which are not functionally redundant with CPEB1 because they do not bind to the cytoplasmic polyadenylation element ( CPE ) sequence . However , mRNA polyadenylation by CPEB1 may be redundant with other polyadenylation mechanisms ( Villalba et al . , 2011 ) . FMRP ( FMR1 ) has two paralogs , FXR1 and FXR2 . FMR1 and FXR2 function cooperatively and the phenotype of single KO of each gene in mice was less severe than that of double KO mice ( Zhang et al . , 2009 ) . These studies suggest that KO of single genes could be compensated by alternative factors and/or not completely impair the cooperative activity of the factors involved . Another explanation for the different phenotypes between RNG105 conditional deletion mice and the other KO mice may be that the RNA granule components other than RNG105 are not essential for long-term memory formation . However , this is not unlikely because , in Drosophila , mutants of FMRP , Staufen , Pumilio , CPEB1 ( Orb ) , GLD-2 ( Wispy ) , Ataxin-2 ( Atx2 ) , and DDX6 ( Me31B ) have been reported to show defects in long-term memory ( Bolduc et al . , 2008; Dubnau et al . , 2003; Pai et al . , 2013; Kwak et al . , 2008; Sudhakaran et al . , 2014 ) . The long-term memory deficits in Drosophila may be because alternative paralogs of these RNA granule components do not exist in Drosophila . In addition , because family proteins are less in Drosophila than in mice in general , proteins encoded by target mRNAs of RNA granule components could have less alternative family proteins in Drosophila than in mice . Thus , mutations in single genes could lead to severer defects in long-term memory in Drosophila . In contrast to the RNA granule components described above , RNG105 may not have alternatives . Although RNG105 ( caprin1 ) has a paralog , RNG140 ( caprin2 ) , in mice , RNG105 and RNG140 are localized to different kinds of RNA granules , and knockdown phenotypes of RNG105 and RNG140 in cultured neurons were not compensated by each other ( Shiina and Tokunaga , 2010 ) . In addition , because RNG105 targets ( affects the localization of ) more than a thousand mRNAs , the probability may be large that mRNAs encoding long-term memory-related factors and their alternative factors are simultaneously affected by RNG105 deficiency . Thus , RNG105 deficiency may have large impact on the function of RNA granules and therefore the formation of long-term memory . This study applied for the first time the technique of the somato-dendritic mRNA identification with RNA-seq to mutant animals . The RNA-seq revealed that the loss of RNG105 reduced the asymmetric somato-dendritic localization of mRNAs in vivo . Furthermore , the RNA-seq identified multiple mRNAs whose dendritic localization was reduced by RNG105 deficiency , which included mRNAs encoding regulators of the cell surface expression and postsynaptic retention of AMPARs , for example , Arf regulator mRNAs . In addition , the identified mRNAs included those encoding Na+/K+ ATPase subunit isoforms and K+ channel subunits ( Supplementary file 5B ) , which was consistent with the previous in vitro study ( Shiina et al . , 2010 ) . Proteins encoded by these mRNAs are involved in the control of membrane potential , and thereby may be associated with fEPSP amplitude and epileptic-like EPSP . Furthermore , the identified mRNAs included those encoding regulators of Ras , Rho , and the PI3 kinase and Rac pathway proteins , involved in actin reorganization and spine formation ( Nishiyama and Yasuda , 2015; Sala and Segal , 2014 ) , which may be associated with the impaired structural plasticity of spines in RNG105-deficient mice . mRNAs for ribosomal subunit proteins were the major dendritic mRNAs and also reported in the previous studies ( Cajigas et al . , 2012; Ainsley et al . , 2014 ) . However , whether locally translated ribosomal proteins are involved in ribosome biogenesis or in other biological processes , which could be associated with RNG105-deficient phenotypes , remains elusive . Thus , RNG105-dependent dendritic mRNAs included various mRNAs whose encoded proteins are involved in AMPAR localization , membrane potential control , and actin reorganization . Even if the reduction of each mRNA in dendrites could have a small influence on synaptic functions , integration of the reduction of these mRNAs could have a large impact on it . RNG105 deficiency reduced homeostatic scaling of AMPARs and steady-state fEPSP , and also spine structural plasticity and fEPSP amplitude after LTP induction . These lines of evidence indicate that RNG105 is critical for synaptic functions both in the homeostatic phase and during LTP , which is consistent with the involvement of translational regulation in both the processes ( Nishiyama and Yasuda , 2015; Tanaka et al . , 2008; Cajigas et al . , 2010 ) . Although fEPSP amplitude after LTP induction was small , LTP appeared intact in Camk2a-Cre;Rng105f/f mice . This could be because translation may not be inhibited although the material ( mRNA ) is reduced in the dendrites of Camk2a-Cre;Rng105f/f mice , which enables the delivery of locally synthesized proteins , even if at a low level , to synapses in the late phase of LTP . Another , though not mutually exclusive , explanation is that translation deficiency may not necessarily cause the decline of LTP in the late phase , but limit the absolute amplitude of EPSP to a certain level in the late phase of LTP . The mechanism of the EPSP limitation is likely coupled with spine size reduction because spine size is tightly correlated with AMPAR expression level in the spine ( Kopec et al . , 2007; Matsuzaki et al . , 2001 ) . If EPSP amplitude is above the limit in the early-phase LTP , EPSP will decline in the late phase , whereas if EPSP amplitude is already low in the early phase , EPSP will be retained in the late phase . In Camk2a-Cre;Rng105f/f mice , translation deficiency and spine size reduction in dendrites could limit the absolute amplitude of EPSP . However , because the basal EPSP amplitude of Camk2a-Cre;Rng105f/f mice is low , EPSP amplitude after LTP induction could not reach the limit level even if LTP occurs normally . As a result , LTP may appear to be sustained in the late phase , but because of the low absolute amplitude of EPSP , long-term memory may be affected . A similar phenotype to Camk2a-Cre;Rng105f/f mice , that is , long-term memory deficits with low steady-state EPSP and intact LTP , has been reported in other mice such as aged mice , prion-infected mice , chronic stressed mice , and Rett syndrome model ( MeCP2-null ) mice ( Burke and Barnes , 2006; Mallucci et al . , 2007; Kallarackal et al . , 2013; Dani and Nelson , 2009 ) . In these mice , similarly to Camk2a-Cre;Rng105f/f mice , impaired long-term memory formation may be attributed to reduced excitatory transmission in neurons . There are increasing number of evidence that RNA granules are associated with mental disorder and neurodegenerative diseases . However , the primary question whether RNA granules are required for the formation of long-term memory has been unclear . This study demonstrated that an element of RNA granules , RNG105/caprin1 , was required for long-term memory formation , and dendritic localization of mRNAs as an underlying mechanism for AMPAR-dependent synaptic strength and long-term-memory formation .
All animal care , experiments and behavioral testing procedures were approved by the Institutional Animal Care and Use Committee of the National Institutes of Natural Sciences , and performed in accordance with the guidelines from the National Institutes of Natural Sciences , Niigata University and the Science Council of Japan . To generate a loxP-flanked ( floxed ) Rng105 construct , we isolated the Rng105/caprin1 gene by PCR from a genomic DNA library of C57BL/6 mice . A DNA fragment , which carried a 34 bp loxP sequence and a neomycin resistance gene ( Neo ) flanked by two frt sites , was inserted into the site 227 bp upstream of exon 5 . The other loxP site was introduced into the site 282 bp downstream of exon 6 . The targeting vector contained exons 5 and 6 of the Rng105 gene flanked by loxP sequences , 7 . 33 kb upstream and 5 . 8 kb downstream homologous genomic DNA fragments , and the diphtheria toxin ( DT ) gene for negative selection ( Figure 1A ) . The targeting vector was introduced into C57BL/6 ES cells ( RENKA ) , and homologous recombinants were selected by G418 resistance and identified by Southern blotting analysis of genomic DNA after EcoRI digestion . ES cell clones with the correct recombination were injected into eight-cell stage embryos of CD-1 mice to generate chimeric mice , which were mated with C57BL/6 mice to obtain heterozygous offspring ( Rng105f/+ ) . Homozygous floxed Rng105 mice ( Rng105f/f ) were obtained by mating heterozygotes . Floxed Rng105 mice were further crossed with Camk2a-Cre transgenic mice ( C57BL/6-TgN[a-CaMKII-nlCre]/20 , RIKEN RBRC00254 ) to obtain heterozygous conditional deletion mice ( Camk2a-Cre;Rng105f/+ ) . Homozygous conditional deletion mice ( Camk2a-Cre;Rng105f/f ) and control mice ( Rng105f/f ) were obtained by crossing Camk2a-Cre;Rng105f/+ female and Rng105f/f male mice because Cre is expressed in male germ cells as well as in the central nervous system under the control of the Camk2a promoter . Genotyping was performed by PCR with primers 5'-AGATGGCTTTTCTTCTGCCA-3' and 5'-CTGGAAAACACGCTCAACAA-3' , which amplified a 918 bp product from the wild-type Rng105 allele and a 978 bp product from the floxed Rng105 allele; and primers 5'-GTCGATGCAACGAGTGATGA-3' and 5'-AGCATTGCTGTCACTTGGTC-3' , which amplified a 291 bp product from the Cre transgene . The Thy1-GFP transgenic mice ( Tg[Thy1-EGFP]MJrs/J ) were purchased from Jackson Laboratory ( Bar Harbor , ME , USA ) . We crossed Camk2a-Cre;Rng105f/+ female and Thy1-GFP;Rng105f/f male mice to obtain Thy1-GFP;Rng105f/f control mice and Thy1-GFP;Camk2a-Cre;Rng105f/f RNG105 conditional deletion mice . In the glutamate uncaging experiments , dissociated hippocampal neurons were prepared from Rng105f/f embryos at embryonic day 17–18 ( E17–18 ) . Neurons were plated at a density of 1 . 6 × 106 cells/cm2 onto poly-D-lysine-coated coverslips in glass-bottomed dishes ( MatTek , Ashland , MA , USA ) in Neurobasal-A medium ( Thermo Fisher Scientific , Waltham , MA , USA ) containing B-27 supplement ( Thermo Fisher Scientific ) , 0 . 5 mM glutamine and 25% Neuron culture medium ( Wako Pure Chemical Industries , Osaka , Japan ) . Cultures were incubated at 37°C in a 5% CO2 incubator . The neurons were transfected with plasmids at 6 days in vitro ( DIV ) using conventional calcium-phosphate transfection method . In the GluR1 and GluR2 immunostaining and biotinylation experiments , dissociated cerebral cortical neurons were prepared from individual littermates at E17 . 5 . Neurons were cultured in the same way as above at a density of 6 . 4 × 104 cells/cm2 . CHO-K1 cells ( RCB0285 , RIKEN BRC , Tsukuba , Japan ) were cultured in HAM's F-12 ( Wako Pure Chemical Industries ) containing 5% fetal calf serum ( FCS ) at 37°C in the 5% CO2 incubator . The cells were transfected with plasmids using Lipofectamine 2000 ( Thermo Fisher Scientific ) in accordance with the manufacturer's protocol . CHO-K1 cells were not included in the list of commonly misidentified cell lines maintained by the International Cell Line Authentication Committee . The origin of the cells ( Chinese hamster ) was confirmed by PCR in RIKEN BRC ( link of datasheet is http://www2 . brc . riken . jp/lab/cell/detail . cgi ? cell_no=RCB0285 ) . The cells were negative for mycoplasma by both PCR and nuclear staining , which were performed based on protocols by RIKEN BRC ( http://cell . brc . riken . jp/ja/quality/myco_kensa ) . To construct the expression vector for Cre , Cre cDNA was obtained using RT-PCR from RNA isolated from the cerebral cortex of Camk2a-Cre transgenic mouse with primers 5'-GGGGAATTCATGTCCAATTTACTGACC-3' and 5'-CTCGAATTCCTAATCGCCATCTTCCAGC-3' . The product was cloned into the EcoRI site of pEGFP-N1 ( Clontech , Mountain View , CA , USA ) whose GFP coding sequence was deleted by BamHI/NotI digestion . To construct the expression vector for RNG105 ( a . a . 1–122 ) -GFP , Rng105 cDNA was obtained by RT-PCR from RNA isolated from mouse cerebral cortex with primers 5'-GTCGACATGCCCTCGGCCACCAGCCACAG-3' and 5'-GAATTCGGAGCTTTATATCTTGACTTAATG-3' . The product was cloned into the XhoI/EcoRI sites of pEGFP-N1 ( Clontech ) . An expression vector for mCherry ( pCS2-mCherry ) was kindly donated by Dr . N . Kinoshita . Extracts of mouse brains were prepared by homogenization in 50 mM Tris ( pH 8 . 0 ) , 150 mM NaCl , 1% NP-40 , protease inhibitors ( 1 mM PMSF , 10 µg/ml leupeptin , pepstatin , and aprotinin ) and 1 mM dithiothreitol . After centrifugation for 10 min at 10 , 000 × g at 4°C , the supernatant was added to Laemmli sample buffer and boiled . Extracts from cultured CHO cells were prepared in the same way . The extracts were separated by SDS-PAGE , transferred to polyvinylidene fluoride membranes ( Merck Millipore , Billerica , MA , USA ) and probed with an anti-RNG105 polyclonal antibody ( Shiina et al . , 2010 ) , anti-α-tubulin monoclonal antibody ( 1:2 , 000 , DM1A , Sigma-Aldrich , St . Louis , MO , USA ) , or anti-GFP monoclonal antibody ( 1:500 , GF200 , Nacalai Tesque , Kyoto , Japan ) . Biotinylated secondary antibodies ( GE Healthcare , Chicago , IL , USA ) and alkaline phosphatase-conjugated streptavidin ( GE Healthcare ) were used for the detection with a bromochloroindolyl phosphate/nitro blue tetrazolium solution . To immunostain brain slices , adult mouse brains were infused with Tissue-Tek ( Sakura Finetek , Tokyo , Japan ) , frozen in liquid nitrogen and sectioned at 10 µm using a cryostat ( HM500-OM , Carl Zeiss , Oberkochen , Germany ) . The sections were mounted on silane-coated coverslips and dried for ~30 min at room temperature . The samples were fixed with 3 . 7% formaldehyde in phosphate-buffered saline ( PBS; 137 mM NaCl , 8 . 1 mM Na2HPO4 , 1 . 5 mM KH2PO4 , and 2 . 7 mM KCl , pH 7 . 4 ) for 10 min at room temperature and permeabilized with 0 . 5% Triton X-100 in PBS . After blocking with 10% FCS in PBS , the samples were incubated with the anti-RNG105 antibody over night at 4°C . After washing with PBS , the samples were incubated with Alexa488-conjugated anti-rabbit IgG ( 1:400 , Jackson ImmunoResearch , West Grove , PA , USA ) and 1 µg/ml 4' , 6-diamidino-2-phenylindole ( DAPI ) ( Wako Pure Chemical Industries ) for 1 hr at room temperature to label RNG105 and nuclei , respectively . To immunostain cultured neurons , neurons at 12 DIV were fixed and stained in the same way . Fluorescence images were acquired using an A1 confocal laser microscope equipped with a Ti-E inverted microscope ( Nikon , Tokyo , Japan ) with a 10 × objective lens or a PlanApo VC60 × oil objective lens . Brains were removed from the Thy1-GFP expressing Rng105f/f and Camk2a-Cre;Rng105f/f mice and fixed with 3 . 7% formaldehyde in PBS for 2 hr at room temperature . The brains were sectioned at 100 µm using a vibratome VT1200S ( LEICA , Wetzlar , Germany ) , and the sections were mounted in Mowiol ( Merck Millipore ) . Fluorescence images were acquired using the A1 confocal microscope with a 20 × objective lens to measure the length and branching of dendrites , and a 100 × objective lens to measure the size and morphology of spines . The images were analyzed using ImageJ software . Spines were classified as the mushroom type when Wneck/Whead < 0 . 5 and Whead > 0 . 4 µm , where Wneck and Whead are the width of spine neck and spine head , respectively . Time-lapse two-photon imaging of dendritic spines was performed using an FVMPE-RS two-photon laser scanning microscope ( Olympus , Tokyo , Japan ) equipped with an Insight DS Dual-line laser system ( Spectra Physics , Santa Clara , CA , USA ) and a 95% O2/5% CO2 incubator ( Tokai Hit , Shizuoka , Japan ) with a water immersion objective lens XLPLN25XWMP2 ( Olympus ) . The culture medium of dissociated hippocampal neurons ( 12–15 DIV ) was exchanged with modified artificial cerebrospinal fluid ( ACSF ) ( 125 mM NaCl , 2 . 5 mM KCl , 3 mM CaCl2 , 1 . 25 mM NaH2PO4 , 26 mM NaHCO3 , 20 mM glucose , 1 μM tetrodotoxin [TTX] [Wako Pure Chemical Industries] , and 50 µM picrotoxin [Sigma-Aldrich] , gassed with 95% O2/5% CO2 before use ) containing 2 mM MNI-caged L-glutamate ( Tocris , Ellisville , MO , USA ) . MNI-caged L-glutamate was uncaged locally near single spine heads by 120 pluses ( 2 ms pulse duration at 2 Hz ) of 740 nm laser illumination with laser power of 5 mW . For imaging of spines with mCherry , a 1 , 040 nm laser was used . 20 images at 0 . 5 µm focus step were projected by summation , and the fluorescence intensity of spine ( sum of pixel intensity in the spine area ) , which reflects spine volume , was measured using ImageJ software as described previously ( Matsuzaki et al . , 2004 ) . Hippocampal slice preparation and electrophysiology were performed as described previously ( Shinoda et al . , 2011 ) . Briefly , postnatal 8–10 ( P8–10 ) weeks old Camk2a-Cre;Rng105f/f mice or Rng105f/f littermates were deeply anesthetized and decapitated , brains were isolated and cooled rapidly to 4°C . Transverse 400 µm thickness hippocampal slices were prepared using a vibratome RPO7 ( D . S . K , Kyoto , Japan ) and high-sucrose cutting solution ( the formulations of all solutions are described below ) , and maintained in ACSF at room temperature for at least 2 hr . A bipolar stimulation electrode was placed in the CA1 SR region to stimulate CA3 Schaffer collateral fibers . Field excitatory postsynaptic potentials ( fEPSPs ) were recorded using a capillary glass electrode ( Harvard Apparatus , Massachusetts , MA , USA ) filled with ACSF placed in the CA1 SR after a 0 . 05 Hz test pulse generated by a pulse generator Master-8 ( A . M . P . I . , Jerusalem , Israel ) equipped with an isolator ISO-Flex ( A . M . P . I . ) . For LTP recording and theta-burst stimulation , 50% maximum stimulus intensity was used . LTP was induced by theta-burst stimulation ( Four trains with 10 s intervals between trains; each train had five bursts separated by 200 ms and included four pulses delivered at 100 Hz ) . Data were amplified using a MultiClamp 700A ( Molecular Devices , Sunnyvale , CA , USA ) , digitized at 10 kHz and filtered at 2 kHz using a Digidata 1440 system ( Molecular Devices ) with pCLAMP9 software ( Molecular Devices ) . The formulation of the solutions were ( in mM ) : High-sucrose cutting solution; 234 sucrose , 2 . 5 KCl , 1 . 25 NaH2PO4 , 0 . 5 CaCl2 , 10 MgSO4 , 26 NaHCO3 , and 11 D-glucose , gassed with 95% O2/5% CO2 . ACSF; 125 NaCl , 2 . 5 KCl , 1 . 25 NaH2PO4 , 2 CaCl2 , 1 MgCl2 , 26 NaHCO3 , and 11 D-glucose , gassed with 95% O2/5% CO2 . In all behavioral tests , male mice ( 3–4 months old; 2–4 mice were co-housed in a 12 hr dark/light cycle ) were used during the light cycle . The open field test was conducted in a round apparatus ( 85 cm in diameter with a 40 cm wall ) with concentric circles and radial lines drawn on the floor , which divided the floor into 25 blocks . A mouse , naive to the apparatus , was placed in the center of the field and allowed to explore for 5 min . The total number of line crossings was recorded as an index of exploratory activity . The mouse was subjected to the test on three consecutive days to assess habituation to the novel place . The test was conducted with an apparatus consisted of a light chamber and a dark chamber separated by a sliding door ( LDK-M , Melquest , Toyama , Japan ) . A mouse was automatically monitored using a SCANET system ( Melquest ) . The mouse was placed into the light chamber , and 5 s later , the door was opened . The mouse was allowed to move freely between the two chambers for 5 min . Distance traveled , time spent in each chamber , and the number of transition between the chambers were measured . To habituate to the test environment , mice were placed in a chamber ( 43 × 43 × 29 . 5 cm ) and allowed to explore freely for 1 . 5 hr on four consecutive days . In the first session , two identical objects ( object 1 and object 2 , cell culture flasks ) were placed at the opposite corners of the chamber , and the mouse was allowed to explore the chamber for 5 min . In the second session , object two was replaced by a different type of object ( object 3 , a wooden prism ) , and the mouse was allowed to explore the chamber for 5 min . The interval between the sessions was 2 . 5 hr . The number of mouse interactions with the objects was counted , where object interaction was defined as the nose directing toward the object at a distance ≤2 cm . The rotarod test was conducted using ROTA-ROD for mice 7650 ( Ugo Basile , Varese , Italy ) . A mouse was placed on a rod rotating at a fixed speed of 24 rpm . The duration of a trial was 3 min , and if the mouse fell from the rod within 3 min , the mouse was re-placed on the rod . The number of falls and the longest latency without falling within a trial were measured . The mouse was subjected to the test on three consecutive days to assess motor skill learning . The Morris water maze was conducted in a round water tank with a diameter of 110 cm filled with ~24°C water . First , the mouse was subjected to visible platform test . In this test , a black platform ( 6 cm in diameter ) was placed 0 . 5 cm above the water surface . A mouse was placed in the pool and the escape latency to the platform was measured . If the mouse found the platform within a 1 min time limit , the mouse was allowed to stay on the platform for 30 s . If not , the mouse was guided to the platform before staying on the platform for 30 s . Mice were given six trials per day for five consecutive days . The location of the platform and the start position were changed randomly in each trial . Next , the mice were subjected to a hidden platform test , in which a clear platform ( 6 cm in diameter ) was placed 1 cm below the water surface . The water was clouded with non-toxic white paint ( Sakura Color Products , Osaka , Japan ) . Spatial cues ( four different plane figures ) were attached on the interior wall of the tank above the water surface . The test was conducted in the same way as in the visible platform test with a 2 min time limit . The start position was changed randomly , but the location of the platform was fixed . The mice were given six trials per day for 10 consecutive days . One day after the last trial of the hidden platform task , the mice were subjected to a probe test . The platform was removed from the pool and the swimming path of the mouse was tracked for 1 min using a computer-based video tracking system ANY-Maze ( Stoeling , Wood Dale , IL , USA ) . Time spent in the target quadrant , where the hidden platform had been placed , and in the other quadrants was also measured . The test was conducted using the LDK-M chamber ( Melquest ) with the SCANET system ( Melquest ) . One day before the training day , to habituate to the test environment , the mouse was allowed to freely explore both chambers for 30 min with the door opened . On the training day , the mouse was placed in the light chamber and allowed to freely explore the chambers for 15 min after the door opened , and the time spent in the dark chamber during the first 5 min was measured ( Pre-FS ) . After that , when the mouse entered the dark chamber , the door was closed and foot shock ( 0 . 5 mA , 1 s duration , 4 times with 1 min intervals ) was delivered . After the foot shock , the mouse was returned to its home cage . The mouse was replaced in the light chamber at 5 min , 1 day , and 1 week after the foot shock and allowed to freely explore the chambers for 5 min after the door opened , and the time spent in the dark chamber was measured . A mouse was placed in a test chamber ( 15 × 15 × 15 cm , the dark chamber of LDK-M with the door closed ) and allowed to explore freely for 30 s . Then the mouse received a foot shock ( 0 . 5 mA , 2 s ) and was returned to its home cage . Five days after the conditioning , the mouse was replaced in the test chamber and in a control chamber ( 24 . 5 × 15 . 5 × 14 . 8 cm , CL-0112–3 , CLEA Japan , Tokyo , Japan ) for 1 min each . Freezing time in the chambers was measured by ANY-Maze . Mouse brains ( P12 weeks old ) were removed in ice cold sterile PBS , and embedded in a 4 . 5% low-melting point agarose ( Agarose XP , Nippon Gene , Tokyo , Japan ) . Coronal sections ( 500 µm thick ) were sliced in ice-cold sterile PBS using the vibratome VT1200S , and transferred into ice-cold sterile PBS . Brain slices were stained with 1 µg/mL DAPI for 10 min at 4°C . Stratum pyramidale ( SP ) and stratum radiatum ( SR ) in the hippocampal CA1 region were microdissected manually using glass capillaries ( G-1 , Narishige , Tokyo , Japan ) with the use of a stereomicroscope . Glass capillaries were made using a needle puller ( model PB-7 , Narishige ) . The borders between the stratum oriens ( SO ) and SP , and between the SP and SR were cut to isolate the SP , and the border between SP and SR , and between SR and stratum lacnosum-moleculare ( SLM ) were cut to isolate the SR . To avoid contamination of SR with the soma of pyramidal neurons , the isolated SR was checked to confirm it did not contain a high-density DAPI-positive nucleic layer ( SP ) with the use of an inverted fluorescence microscope ( IX83 , Olympus ) with a 40 × objective lens . To obtain sufficient amount of RNA for RNA-seq , left and right hippocampi from three mice were dissected and collected into one sample tube . Triplicate samples were prepared for RNA-seq analysis . The samples were stored at −80°C until RNA extraction . Total RNA was extracted from the tissues ( SP and SR ) using ISOGEN ( Nippon gene ) in accordance with the manufacturer's protocols with minor modifications . The isolated tissue was homogenized in 400 µL ISOGEN by pipetting . After the homogenate was stored at room temperature for 5 min , 100 µL chloroform was added and the sample was shaken vigorously for 30 s . After being stored on ice for 5 min , the sample was centrifuged for 15 min ( 21 , 900 × g , 4°C ) , and about 240 µL of the aqueous phase containing RNA was collected in a new tube . The sample was centrifuged for 15 min ( 21 , 900 × g , 4°C ) again and the aqueous phase was collected into a new tube . To remove phenol and chloroform from the sample , 240 µL diethyl ether was added to the sample and vortexed for 1 min . After the sample was centrifuged for 1 min ( 21 , 900 × g , 4°C ) , the upper phase was removed . This diethyl ether treatment was performed three times . Then , 25 µL 3 M sodium acetate and 250 µL 2-propanol were added to the sample and mixed , and then RNA was precipitated on ice for 30 min . After the sample was centrifuged for 15 min ( 21 , 900 × g , 4°C ) , the supernatant was removed . The RNA precipitate was washed with 500 µL 70% ethanol , dried using a vacuum desiccator and dissolved in 22 µL RNase-free water ( TaKaRa , Shiga , Japan ) . Next , DNA was removed from the sample by DNase treatment . 0 . 5 µL DNase ( RT grade ) ( Nippon Gene ) and 10 × buffer were added to the 22 µL RNA solution , and the sample was incubated for 15 min at 37°C . Then , 25 µL phenol-chloroform mixture was added , and the sample was vortexed for 30 s . After the sample was centrifuged for 10 min ( 21 , 900 × g , 4°C ) , upper phase was collected . Phenol and chloroform were removed from the sample by diethyl ether treatment as described above . Then , 2 . 5 µL 3 M sodium acetate and 62 . 5 µL 100% ethanol were added to the sample and mixed , and then RNA was precipitated at −20°C for 30 min . After the sample was centrifuged for 20 min ( 21 , 900 × g , 4°C ) , the supernatant was removed . The RNA precipitate was washed with 150 µL 70% ethanol , dried , and dissolved in 10 µL RNase-free water . RNA solution was stored at −80°C until use for the preparation of cDNA libraries . Before preparation of cDNA libraries , the quality of the extracted RNA was checked . RNA integrity was measured using an Agilent 2100 bioanalyzer ( Agilent Technologies , Santa Clara , CA , USA ) with RNA 6000 Nano Kit and RNA 6000 Pico Kit ( Agilent Technologies ) in accordance with the manufacturer's protocols . The RNA integrity number was over seven for all samples , which was sufficient quality for RNA-seq . Total RNA ( 200 ng per sample ) was used as the starting material to prepare cDNA using TruSeq RNA sample Preparation Kit v2 ( Illumina , San Diego , CA , USA ) at a half scale compared with the manufacturer's protocols , as follows . Poly-A-containing mRNAs were purified using Oligo dT magnetic beads . After the mRNAs were denatured , they were fragmented at 94°C for 4 min . First-strand cDNA was synthesized using SuperScript II Reverse Transcriptase ( Thermo Fisher Scientific ) , and then second-strand cDNA was synthesized . After the ends of the fragments were blunted , an adenine nucleotide was added to the 3'-end at 37°C for 30 min . RNA adapter index was ligated to the 5'- and 3'-ends of the ds cDNA , and the cDNA was amplified with 10 PCR cycles . cDNA quality was validated using the bioanalyzer with a High Sensitivity DNA Kit ( Agilent Technologies ) . The cDNA libraries were quantified with quantitative PCR ( qPCR ) using a 7500 real-time PCR system ( Thermo Fisher Scientific ) and adjusted to 2 nM . In each experiment , an equal amount of cDNA from SP and SR was analyzed , and three independent experiments were conducted . Twelve cDNA libraries ( four kinds of samples × three biological replicates ) were sequenced with 101 bp paired-end sequencing using a HiSeq1500 ( Illumina ) . The resulting reads were mapped to the mouse genome ( Mus musculus Ensemble NCBIM37 ) using TopHat ( v 2 . 0 . 11 ) . The mapped reads were assembled using Cufflinks ( v 2 . 2 . 1 ) , and differential gene enrichment analysis was conducted using Cuffdiff . S-mRNAs and D-mRNAs in control ( Rng105f/f ) mice were identified as follows . First , mRNAs from SP and SR of Rng105f/f mice were analyzed using Cuffdiff , as described above , which identified mRNAs whose concentration ( FPKM value ) was significantly different between SP and SR , designated as ‘yes’ , and not significantly different between the layers designated as ‘no’ in the ‘all mRNAs’ list ( Supplementary file 1 ) . After this , mRNAs , whose FPKM values were more than zero both in SP and SR , were selected and subjected to subsequent data analysis . ‘Yes’ mRNAs were divided into two groups , candidates for dendritic mRNAs whose DAIs ( relative FPKM value in SR to SP ) were more than 1 , and candidates for somatic mRNAs whose DAIs were less than 1 . Next , to identify mRNAs specifically expressed in pyramidal neurons , we eliminated mRNAs also expressed in other types of cells such as glial cells , interneurons , and endothelial cells ( Cajigas et al . , 2012; Doyle et al . , 2008; Daneman et al . , 2010; Cahoy et al . , 2008 ) from each of the lists ( Supplementary file 2A , B ) . Finally , we identified 1122 dendritic mRNAs , 2106 somatic mRNAs , and 2814 non-significant mRNAs in Rng105f/f mice ( Supplementary file 3A−C ) . Furthermore , to compare the dendritic localization of mRNAs between Rng105f/f and Camk2a-Cre;Rng105f/f mice , the ratio of DAI in Camk2a-Cre;Rng105f/f mice to Rng105f/f mice was calculated for each mRNA and indicated in the ‘DAI ( Camk2a-Cre;Rng105f/f ) /DAI ( Rng105f/f ) ’ columns ( Supplementary file 3A−C ) . Gene ontology enrichment analysis was performed using DAVID functional annotation tools . Significance of overrepresentation of GO terms was assessed using the Benjamini-Hochberg false discovery rate ( FDR ) criterion at p<0 . 05 . Immunostaining of cultured cortical neurons ( 9 DIV ) for GluR1 and GluR2 was conducted as described previously ( Ohashi et al . , 2016 ) . To block neuronal activity , 0 . 7 µM TTX and 20 µM D-2-amino-5-phosphonovaleric acid ( APV ) ( Sigma-Aldrich ) were added to the medium for 24 hr at 37°C in a 5% CO2 incubator . Live neurons were incubated with an anti-GluR1 ( 1:15 , PC246 , Merck Millipore ) or an anti-GluR2 antibody ( 1:100 , MAB397 , Merck Millipore ) at 37°C in a 5% CO2 incubator for 1 hr . After the neurons were washed in Neurobasal-A medium , they were fixed with 3 . 7% paraformaldehyde in PBS for 20 min at 25°C . Fixed neurons were blocked for 30 min in 10% FCS in DMEM ( Sigma-Aldrich ) , and incubated with an Alexa Fluor 488-conjugated anti-rabbit and mouse IgG antibodies ( 1:400 , Thermo Fisher Scientific ) in 10% FCS in DMEM for 3 hr at 25°C to label cell surface GluR1 and GluR2 . After neurons were washed in PBS , they were fixed again and permeabilized with 0 . 25% Triton X-100 in PBS for 10 min . After the neurons were blocked , they were incubated with the anti-GluR1 antibody ( 1:50 ) or anti-GluR2 antibody ( 1:100 ) for 12 hr at 4°C and then with a Cy3-conjugated anti-rabbit and mouse IgG antibodies ( 1:400 , Jackson Immuno Research ) for 3 hr at 25°C to label intracellular and residual surface GluR1 and GluR2 . Neurons were imaged using the IX83 inverted fluorescence microscope ( Olympus ) with a 40 × objective lens and an ORCA-R2 digital CCD camera ( Hamamatsu Photonics , Hamamatsu , Japan ) . Because GluR1 and GluR2 were detected in a punctate manner , they were quantified by counting the number and measuring the fluorescence intensity of the puncta in dendrites . To count the number of GluR1 and GluR2 puncta , the images were converted into binary images by selecting dendritic regions and using the MaxEntropy threshold algorithm in ImageJ software . The number of GluR1 and GluR2 puncta in dendrites was counted using the magic wand tool and analysis tool in Adobe Photoshop software . To count total GluR1 and GluR2 puncta , the binary images of before and after permeabilization were merged and used . To measure the fluorescence intensity of GluR1 and GluR2 puncta in dendrites , the binary image and original image were layered in Adobe Photoshop software . ROIs were selected in the binary layer using the magic wand tool , and fluorescence intensity in the ROIs in the original layer was calculated by multiplying mean pixel intensity by area of the ROIs . The sum of fluorescence intensity was normalized by dendrite length and by the intensity of GluR1 and GluR2 in the soma . GluR1 and GluR2 intensity in the soma was calculated using the layered images by multiplying mean pixel intensity by area of ROIs in the soma and normalized by the area of the soma . Fluorescence intensity of GluR1 and GluR2 puncta immunostained before permeabilization was normalized by that immunostained after permeabilization . Biotinylation assay was conducted as described previously ( Chung et al . , 2000; Snyder et al . , 2001; Aoto et al . , 2008 ) . Primary cultured cortical neurons ( 9 DIV ) in two 30 mm dishes were treated with or without 1 μM TTX and 100 μM APV in the culture medium for 24 hr prior to surface biotinylation ( Aoto et al . , 2008 ) . The dishes were placed on ice and washed three times with ice-cold ACSF ( 124 mM NaCl , 5 mM KCl , 1 . 25 mM NaH2PO4 , 26 mM NaHCO3 , 0 . 8 mM MgCl2 , 1 . 8 mM CaCl2 , and 10 mM D-glucose , gassed with 95% O2/5% CO2 ) . Then the neurons were incubated with ACSF containing 1 mg/ml sulfo-NHS-LC biotin ( Thermo Fisher Scientific ) for 30 min on ice . After washing once with ice-cold 100 mM glycine in ACSF and three times with ice-cold TBS ( 50 mM Tris , 150 mM NaCl , pH 7 . 5 ) , the neurons were lysed in 100 μl of modified RIPA buffer ( 1% Triton X-100 , 0 . 1% SDS , 0 . 5% deoxycholic acid , 50 mM NaHPO4 [pH 7 . 2] , 150 mM NaCl , 2 mM EDTA , 25 mM β-glycerophosphate , 1 mM PMSF , 10 µg/ml leupeptin ) . The lysate was centrifuged at 14 , 000 × g for 15 min at 4°C , and 85 μl of supernatant was incubated with 40 μl of NeutraAvidin Agarose beads ( Thermo Fisher Scientific ) for 3 hr at 4°C with gentle rocking . After washing three times with modified RIPA buffer , biotinylated proteins were eluted from the beads with 80 μl of SDS sample buffer and boiled for 5 min . The total lysate and biotinylated eluate were analyzed by western blotting with the anti-GluR1 ( 1:50 , PC246 , Merck Millipore ) , and anti-GluR2 ( 1:1000 , MAB397 , Merck Millipore ) antibodies . Alkaline phosphatase-conjugated secondary antibodies ( 1:5000 , 711-055-152 and 115-055-146 , Jackson Immuno Research ) and Can Get Immunoreaction Enhancer Solution ( TOYOBO , Osaka , Japan ) were used for the detection with a bromochloroindolyl phosphate/nitro blue tetrazolium solution . Quantification of the band intensity was conducted as previously described ( Ohashi et al . , 2016 ) . A standard dilution series of the lysate from cultured neurons was loaded on the same gel , and used to generate a standard curve and calculate the relative intensity of upper and lower bands of GluR1 and GluR2 . The band intensity was measured using ImageJ software . Triplicate from three mice ( nine samples ) were analyzed for each group . Sample numbers and experimental repeats are indicated in figure legends . Statistical significance was determined using Student's t-test , paired t-test , one-way ANOVA , one-way repeated measures ANOVA , two-way ANOVA , two-way repeated measures ANOVA , post-hoc Tukey-Kramer test , and Bonferroni post-hoc t-test as indicated in the figure legends . Statistical analysis was performed in R , Excel , or an Excel add-in software Statcel ( The Publisher OMS Ltd . , Saitama , Japan ) . Exact F , t , and p-values are indicated in Statistical reporting table ( Supplementary file 7 ) . Post-hoc power analysis was performed with G*Power ( http://www . gpower . hhu . de/ ) and statistical power is indicated in the Statistical reporting table . No blinding method was used in this study . In behavioral tests , animals that died before the experimental endpoint were excluded from the data analysis . The Gene ontology term enrichment was analyzed using DAVID functional annotation tools ( https://david . ncifcrf . gov/ ) . Raw and processed data files for the RNA-seq analysis have been deposited in the NCBI Gene Expression Omnibus ( GEO ) under series accession number GSE96552 ( https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE96552 ) . | Messages pass from one nerve cell to the next across gaps called synapses . The first neuron releases chemical signals from the end of its long , thin nerve fiber . The second receives the message at receptors on branching structures known as dendrites . Each connection has a corresponding bump called a dendritic spine . As animals learn , these can grow larger , strengthening the connection . This is the basis of how memories form . To strengthen a synapse , the cell must transport the materials to the dendritic spine . The cell makes copies of the genetic instructions to strengthen the synapse in the form of messenger RNA ( often shortened to mRNA ) . But , this happens in the body of the cell , a long way from the dendrites themselves . The mRNA travels from the cell body to the dendrites in collections of molecules referred to as ‘RNA granules’ . One of the key components of the RNA granule system is a protein called RNG105/caprin1 . Now , Nakayama , Ohashi et al . have engineered mice to delete the gene for RNG105/caprin1 , revealing its effect on memory . Mice lacking RNG105/caprin1 struggled to make long-term memories . Unlike their normal counterparts , these mutant mice did not become accustomed to new environments or objects . They also found it more challenging to learn the position of a hidden platform in a water-based maze . Lastly , over time , the mutant mice forgot to be fearful of a dark chamber where they had received a small electric shock . Memories form in a part of the brain called the hippocampus and the dendritic spines in this region were smaller in mice lacking RNG105/caprin1 . Furthermore , when the nerve cells from this part of the brain were grown in Petri dishes , they did not respond normally to stimulation . The dendritic spines of normal cells increased in size , but those on the cells lacking RNG105/caprin1 got smaller compared to normal cells . A closer look revealed that the distribution of mRNA in brain cells from mice lacking RNG105/caprin1 differed from that of normal mice . Some pieces of genetic information failed to make it from the cell body to the dendrites . This included mRNA involved in making regulators of a component of dendritic spines called the AMPA receptor . The AMPA receptor detects the chemical messenger , glutamate , and is crucial for memory formation . These findings further our understanding of long-term memory and open the way for future research into human disease . Mutations in RNA granule components , including RNG105/caprin1 , have links to conditions such as amyotrophic lateral sclerosis ( ALS ) and autism spectrum disorder ( ASD ) . Further investigation could reveal new targets for drug treatment . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2017 | RNG105/caprin1, an RNA granule protein for dendritic mRNA localization, is essential for long-term memory formation |
Cellular adhesion is a key ingredient to sustain collective functions of microbial aggregates . Here , we investigate the evolutionary origins of adhesion and the emergence of groups of genealogically unrelated cells with a game-theoretical model . The considered adhesiveness trait is costly , continuous and affects both group formation and group-derived benefits . The formalism of adaptive dynamics reveals two evolutionary stable strategies , at each extreme on the axis of adhesiveness . We show that cohesive groups can evolve by small mutational steps , provided the population is already endowed with a minimum adhesiveness level . Assortment between more adhesive types , and in particular differential propensities to leave a fraction of individuals ungrouped at the end of the aggregation process , can compensate for the cost of increased adhesiveness . We also discuss the change in the social nature of more adhesive mutations along evolutionary trajectories , and find that altruism arises before directly beneficial behavior , despite being the most challenging form of cooperation .
Some of the most peculiar behaviors observed in biological populations arise from conflicts between individual interests and collective function . The existence of different levels of organization generates situations where a behavior that is beneficial for others is associated with individual costs , in terms of reproductive success . The ubiquity of genetically encoded traits that underpin cooperative behavior has thus long been considered an evolutionary paradox , as such traits should be purged by natural selection acting at the individual level . Such paradoxical situations appear in what are known as 'major evolutionary transitions' ( Maynard-Smith and Szathmáry , 1995 ) , an instance of which is the transition from autonomously replicating unicellular to multicellular organisms ( Michod and Roze , 2001; Wolpert and Szathmáry , 2002; Sachs , 2008; Rainey and Kerr , 2010; Ratcliff et al . , 2012; Hammerschmidt et al . , 2014 ) . Evolutionary transitions all face the possibility that a part of the population free rides and exploits the collective benefit produced by the action of others . For instance , cancerous cells that divide faster disrupt the survival of their host ( Frank , 2007 ) . Mathematical models , mostly based on population genetics or game theory , have pointed out several solutions to this evolutionary conundrum . In most game-theoretical and trait-group models , collective-level processes derive from individual-level behavior , and cooperation can be sustained by mechanisms that affect either population structure or individual decisions . The former mechanisms produce assortment between cooperators , arising for instance during group formation ( Avilés , 2002; Fletcher and Zwick , 2004; Santos et al . , 2006; Guttal and Couzin , 2010; Powers et al . , 2011 ) or due to limited dispersal ( Nowak and May , 1992; Pfeiffer and Bonhoeffer , 2003 ) . The emergent topology of interaction makes cooperation profitable on average despite cooperators being outcompeted within every group , as exemplified by the so-called Simpson’s paradox ( Chuang et al . , 2009 ) . The latter mechanisms are instead effective in random population structures and alter individual strategy enough to offset cooperation’s cost . These mechanisms typically rely on players modifying their behavior conditionally to interactants’ types . For instance , reciprocity ( Trivers , 1971; Nowak and Sigmund , 1998 ) , peer recognition ( Antal et al . , 2009 ) and 'green beard' mechanisms ( Brown and Buckling , 2008 ) or policing ( Boyd et al . , 2003 ) are all means by which individuals preferentially direct cooperation toward other cooperators , or retaliate against defectors . Such mechanisms require players to collect and interpret information about others , and are therefore better suited to model elaborate forms of cooperation , rather than the origin of cooperative groups themselves . Ultimately , all explanations for the evolutionary success of cooperative strategies rely on the fact that individual fitness depends on the social context , whether it is shaped by the population structure or the nature of the cooperative act . The interest in knowing which mechanisms are effective in the evolution of pristine modes of collective organization has recently spawned numerous theoretical efforts to model microbial assemblages ( Nadell et al . , 2013; Levin , 2014 ) . Although capable of thriving in isolation , most unicellular species participate—in parts of their life cycle , at least—to multicellular aggregates that offer instances of ‘intermediate’ integration of cells into higher levels of organismality ( Smukalla et al . , 2008; Nadell et al . , 2009; Queller and Strassmann , 2009; Rainey and Kerr , 2010; Ratcliff et al . , 2012; Celiker and Gore , 2013 ) . Microbial societies can be highly regulated , such as in the multicellular fruiting bodies of slime moulds and myxobacteria , where cells within the collective commit to a developmental program analogous to that of permanent multicellular organisms , including the soma-like 'suicide’ of a part of the population . At a lesser degree of sophistication , many other forms of collectives are known in microbial organisms , ranging from colonies to biofilms ( Grosberg , 2007; Nadell et al . , 2009; Niklas , 2014 ) . A central feature of most microbial aggregates is their ability to persist long enough to affect the survival of their composing units . Such persistence is mediated by various forms of adhesion ( Jiang et al . , 1998; Gresham , 2013 ) ensuring local cohesion between cells of the same or different kind . For instance , cells that fail to completely separate at the moment of division are selected for in regimes favoring increased aggregate size ( Ratcliff , 2013 ) , a feature reinforced by genetic homogeneity within groups ( Nadell et al . , 2010 ) . Although apparently an achievable initial step to foster multicellularity ( Kirk , 2005 ) , incomplete division is however unlikely to be the only mechanism at play in the emergence of multicellular organization in the tree of life , as some biological populations form aggregates from dispersed cells and can be composed of multiple genetically distinct strains ( Nanjundiah and Sathe , 2011 ) . Such cases turn out to be particularly challenging for the different theoretical solutions to the evolution of collective function ( Tarnita et al . , 2013 ) . Here , we focus on populations where groups disperse after individual reproduction and emerge anew from a well-mixed pool at the next generation . By affecting group formation as well as the performance of a group , differential attachment is an established means to create positive assortment , which enables the concomitant evolution of adhesive traits and sizeable groups in idealized , well-mixed populations ( Simon et al . , 2013; Garcia and De Monte , 2013 ) as well as more realistic , spatially explicit settings ( Garcia et al . , 2014 ) . These models address the binary competition between more adhesive ‘cooperators’ and less adhesive ‘defectors’ and demonstrate that segregation between types can be generated even if encounters among players occur at random . The stickier type is favored as soon as its frequency exceeds a threshold ( Garcia and De Monte , 2013 ) , in spite of the costs associated with increased adhesion . In actual biological populations , however , social propensity needs not result from individual features that abide by a binary logic . For instance , the FLO1 gene governing flocculation in yeast is highly variable: the level of adhesion changes as a function of the number of tandem repeats within FLO1 ( Smukalla et al . , 2008 ) . This molecular mechanism may underpin a diversity of cell-cell interaction strengths that are better described as a trait with continuous , rather than discrete , values . In this context , the process of adaptation is mathematically formalized with the theory of adaptive dynamics ( Geritz et al . , 1998; Waxman and Gavrilets , 2005 ) , describing evolution as a mutation-substitution process where mutants successively invade a monomorphic resident population . Since mutants are each time initially rare , the aforementioned frequency threshold can potentially hinder the evolution of stickiness as a gradually increasing trait . In the following , we show that , even in the absence of selective pressures favoring large groups , natural selection can promote and sustain costly adhesion nearly from scratch , and we discuss the theoretical and applied implications of our mechanism in puzzling out the evolutionary origins of social behavior .
We study the evolution of adhesiveness in the context , developed by Garcia and De Monte ( 2013 ) , where the trait , associated with a fitness cost , plays a role both in the emergence of population structure—through group formation—and in the collective function of groups . Benefits derived by excreted extracellular glues , for instance , are potentially shared with neighboring individuals , while being individually costly . At the same time , glues affect the way a population clusters in groups . Here , we will consider adhesiveness as a continuous trait described by the real number z∈[0 , 1] . Individuals undergo successive life cycles defined by the alternation of a dispersed and a grouped phase , as illustrated in Figure 1 . In the Aggregation Phase ( AP ) , initially dispersed individuals form groups in a process that is influenced by the values of their trait . In the Reproductive Phase ( RP ) , individuals leave offspring according to the benefits collected within their groups . Finally , all individuals are dispersed anew into a global pool in the Dispersal Phase ( DP ) . 10 . 7554/eLife . 08595 . 003Figure 1 . Life cycle used in the model . At each generation , individuals undergo a succession of three steps: an aggregation phase ( AP ) during which they form groups depending on their adhesiveness trait; a reproduction phase ( RP ) in which they leave offspring with a probability dependent on their strategies and their payoffs in groups; a dispersal phase ( DP ) when all individuals are scattered anew for the next generation . Such life cycle is consistent with those observed in facultative multicellular microorganisms such as dictyostelids and myxobacteria . DOI: http://dx . doi . org/10 . 7554/eLife . 08595 . 003 These assumptions , that include as a special case trait-group models with random group formation ( Wilson , 1975 ) , reflect actual life cycles found in species such as dictyostelids ( Li and Purugganan , 2011; Strassmann and Queller , 2011 ) and myxobacteria ( Xavier , 2011 ) . Such modeled life cycle constitutes a worse-case scenario for the evolution of cooperative traits . Indeed , when aggregates are allowed to persist for more than one individual generation , positive assortment between cooperators is amplified throughout successive reproductive episodes . This can ultimately favor cooperation when population re-shuffling occurs in time scales shorter than that of defectors’ takeover within groups ( Wilson , 1987; Fletcher and Zwick , 2004; Killingback et al . , 2006; Traulsen and Nowak , 2006; Cremer et al . , 2012 ) . The individual trait under the effect of natural selection is adhesiveness z . Higher values of z increase the probability that an individual joins a group ( in AP ) , and at the same time enhance group cohesion , hence group-related benefits ( in RP ) . Adhesiveness is costly , and the cost is assumed proportional to its value: C ( z ) =cz . In Section 4 . 4 , we consider the more realistic case when the cost of glue production increases faster than linearly , for instance because of metabolic or physical constraints . The cost of adhesiveness is here assumed to be context-independent , thus it does not change conditionally to individuals belonging or not to a group . This choice reflects the standard assumption in quantitative genetics that the trait is genetically encoded . The context-dependent part is thus restricted to the benefit term . Situations where increased attachment has no cost have been modelled by Avilés ( 2002 ) . Assuming that adhesive ungrouped individuals do not undergo adhesion costs , or that they earn direct benefits just as grouped individuals do , would relax the social dilemma and promote even more efficiently increased adhesion . In the RP , each individual in a group is assigned a net payoff according to a Public Goods Game ( PGG ) ( Kollock ( 1998 ) ; Doebeli and Hauert ( 2005 ) ) , that models in the simplest terms the reproductive success of individuals taking part in a social enterprise . The net payoff is the sum of two terms: the cost of adhesiveness , and a benefit B drawn from belonging to a group , and therefore equal for all its members . This second term encapsulates the cohesion of the group and depends on the average adhesiveness z̄ of its members: B=B ( z̄ ) ( Brännström et al . , 2011 ) , with B an increasing function of z̄ . In the following , we opt for a linear function B ( z̄ ) =bz̄ . This choice is conservative , since nonlinear ( e . g . saturating ) functions alleviate the constraints on the evolution of social traits ( Archetti and Scheuring , 2012 ) . Ultimately , a z-individual in a group of average adhesiveness z̄ gets a net payoff bz̄-cz . If an individual does not belong to any group , it does not get any group-related benefit and its payoff is merely -cz . The average payoff of one strategy , determining its reproductive success , is obtained by averaging over all social contexts experienced , meaning that the probabilities of occurrence of each possible group composition must be known . Specifying the realized group structure in populations with multiple traits is a daunting task even under simple rules of group formation , so that the evolution of the trait z can only be known by explicitly simulating the aggregation process . Since microbial populations are vast , and their interactions complex , this kind of numerical simulations can be extremely time-consuming . If mutations on z occur seldom with respect to the demographic time scale of trait substitution , however , the payoff of a mutant can be assessed in the background structure provided by the resident , monomorphic population . In this case , the realized repartition of players inside groups can be deduced from specific rules of group formation , and the framework of adaptive dynamics allows to study the gradual evolution of the trait in general settings . Section 3 discusses the adaptive dynamics of the trait in infinitely large populations where its value is associated with a given group size distribution . The general results will be applied in Section 4 to a specific aggregation model where adhesiveness underpins the probability of attachment among cells , that we introduce in the next paragraph . The effect of adhesiveness on group formation can be exemplified by a simple model where the trait underpins physical attachment . Such model , introduced by Garcia and De Monte ( 2013 ) , will be used later in Section 4 to illustrate the general results of Section 3 . During AP , individuals belonging to an initially dispersed , infinite population are distributed at random into patches of size T , akin for instance to the attraction domains observed in Dictyostelium discoideum aggregation ( Goldbeter , 2006 ) . Within every patch , one group is nucleated by a randomly drawn recruiter . Each focal individual among the remaining T-1 is given one opportunity to attach to the recruiter , and does so with a probability that depends on both the recruiter and the focal individual’s adhesivenesses . If it fails to stick to the recruiter , the individual remains alone . So as to preclude any assortment a priori between individuals with same adhesiveness , the probability that an individual of trait z1 attaches to a recruiter of trait z2 must be the geometric mean of the two adhesivenesses: pattach ( z1 , z2 ) =z1z2 ( see Garcia and De Monte ( 2013 ) for an explanation ) . Any other choice for pattach ( z1 , z2 ) entails more positive assortment of the stickier type , hence further facilitates an increase in adhesiveness . Note that this model is not spatially structured . Unlike works based on individuals playing on lattices ( Nowak and May , 1992; Doebeli and Hauert , 2005; Perc et al . , 2013 ) or on an explicit continuous space ( e . g . Meloni et al . ( 2009 ) ; Chen et al . ( 2011 ) ; Garcia et al . ( 2014 ) ) , positive phenotypic assortment is not correlated with spatial proximity . In Garcia and De Monte ( 2013 ) ’s model , players possessed a binary attachment strategy with fixed cost , according to the standard choice of modeling ‘cooperator’ and ‘defector’ strategies . With this assumption , cooperation typically spreads once a threshold frequency of cooperators is overcome in the population , so that it is impossible for an initially rare cooperative mutant to be successful . Moreover , this threshold increases when adhesivenesses of the two types get closer , to such extent that even demographic stochasticity would be insufficient to favor mutations of small phenotypic effect . We will show that if attachment is based on a continuous trait , both these limitations are overcome . In a monomorphic population of trait z , the aggregation process produces groups of various sizes ( smaller than or equal to T ) and a component of ungrouped individuals , as illustrated in Figure 2 . Increasing the adhesiveness level z leads to a rise in average group size , and a decrease in the fraction of individuals that remain ungrouped . These group size distributions define the population structure associated with a given value z of adhesiveness . In the next Section , we provide a condition for increased adhesiveness to be selected , given the group size distributions that characterize a particular group formation process . 10 . 7554/eLife . 08595 . 004Figure 2 . Group size distribution experienced by individuals in a momomorphic population with trait value z , for the aggregation process based on adhesion . The size of each patch is T=100 . The distribution is composed of a fraction 1-z of ungrouped individuals ( n=1 ) and a binomial distribution of grouped individuals centered on n=zT . Here , we display this distribution for 5 distinct values of z . DOI: http://dx . doi . org/10 . 7554/eLife . 08595 . 004
Predicting the evolutionary outcome on adhesiveness in specific biological populations requires to determine how it influences the population structure . In this Section , we show that general conclusions can be drawn , nevertheless , from qualitative properties of the aggregation phase . Notably , the evolutionary success of adhesion critically depends on how it affects the recruitment of individuals within groups . Let us consider a monomorphic resident population composed only of individuals of trait ẑ , challenged with the appearance of a rare mutant of trait z=z^ + dz , where dz is small . In this context , at most two trait values—resident and mutant—are present in the population at a given point in evolutionary time . Determination of average payoffs only relies on the knowledge of the distribution g ( n , z , z^ ) of group sizes experienced by a z-mutant in a population composed of individuals with trait z^ , i . e . the probability that a mutant belongs to a group of size n . In particular , we call u ( z , z^ ) =g ( 1 , z , z^ ) the probability that a z-mutant remains alone . These distributions can be derived—analytically , numerically , or by direct observation—for any given group formation process , and summarize the effect of adhesiveness on group formation . In line with the adaptive dynamics framework , we assume that the evolutionary dynamics of the trait z is fully determined by the selection gradient ( Geritz et al . , 1998 ) . In Section S1 of the Supplementary Information ( SI ) , we calculate the invasion fitness of mutants S ( z , z^ ) for a group formation process characterized by its group size distribution g ( n , z^ , z^ ) and by the proportion u ( z , z^ ) of ungrouped individuals . The selection gradient is the variation of the invasion fitness with a change in adhesiveness: ( 1 ) ∇S ( z^ ) =∂S ( z , z^ ) ∂zz=z^=bz^h ( z^ ) + b∑n≥21ng ( n , z^ , z^ ) -c , where: ( 2 ) h ( z^ ) =-∂u ( z , z^ ) ∂zz=z^ Since higher adhesiveness makes individuals more likely to join groups , u ( ⋅ , z^ ) is a decreasing function of z and h ( z^ ) >0 . The change dS in invasion fitness associated with an infinitesimal increase of the adhesiveness trait has three components . The third , negative term is the additional cost -cdz . The second , positive term , corresponds to direct benefits associated with the change in adhesiveness . It becomes negligible when groups are large , consistent with the observation that sociality gets established more easily in small groups ( Olson , 1971; Powers et al . , 2011 ) . Even when this term cannot offset the cost of increased adhesiveness , the first component can compensate . This component corresponds to the increased chance h ( z^ ) dz for z-individuals to join a group , multiplied by the payoff bz^ drawn from it , and thus reflects the gain associated with the new interaction neighborhood created by a change in adhesiveness . The adaptive dynamics framework states that a more adhesive mutant z>z^ will replace the resident if the selection gradient in Equation 1 is positive . A sufficient condition for an increase in adhesiveness , that holds locally for any z^ , is then: ( 3 ) bz^h ( z^ ) >c The function h ( z ) measures the impact of adhesiveness on the fraction of cells that remain ungrouped at the end of aggregation . Depending on its variation , Equation 3 can provide a global solution to the long-term evolutionary outcome . For instance , when h increases ( i . e . if u is concave ) , or decreases slowly with z , then if the inequality is satisfied for one particular value of the trait , it also holds for larger values . Adhesiveness then steadily increases until it reaches its maximal value z=1 . On the contrary , evolutionarily stable coexistence between grouped and ungrouped cells is only possible if ∇S ( z^ ) switches from positive to negative for some value z^∈]0 , 1[ . A necessary condition for this to happen is that the first term in Equation 1 goes below c . A first possibility is that z^ is small , but then , arguably , groups would be small too and direct benefits large , so that the second term might keep ∇S ( z^ ) positive . A second possibility is that h ( z^ ) is small , meaning that the effect of adhesiveness on the proportion of ungrouped cells is limited . This emphasizes qualitatively the importance of ungrouped individuals on the evolutionary trajectory , and how their proportion might be considered a first-order proxy of population structure . If all individuals are randomly distributed among groups of fixed size , or even group size distribution is assigned a priori ( as considered e . g . by Peña ( 2012 ) ) , the fraction of ungrouped cells is independent of z^ . According to Equation 3 , then , the evolution of increased adhesiveness is still possible , but it will be the consequence of direct , group-derived benefits ( second term of Equation 1 ) . Life cycles consisting of alternate phases of dispersion and aggregation are more likely to leave ungrouped cells than life cycles based on single-cell bottlenecks followed by clonal growth of groups . Equation 3 implies that a subpopulation of nonaggregated cells might boost the evolution of collective function even when cells are genetically unrelated and benefits are weak , and suggests a possible route to the evolution of multicellularity in genetically heterogeneous populations .
For any b>2c , the evolutionary dynamics is bistable . The actual evolutionary outcome depends on whether the initial adhesiveness value is smaller or larger than a threshold value z* , where the selection gradient vanishes . Its analytical expression is computed in Section S2 of the SI . If initial adhesiveness is lower than z* , additional group-related benefits are too small to offset the cost of increased adhesion; the resident trait z^ gradually declines and all individuals are eventually ungrouped . This population structure corresponds to permanently unicellular organisms . If initial adhesiveness is greater than z* , instead , adhesiveness keeps increasing until its maximum value z^=1 and all individuals belong to groups of identical size T . This outcome corresponds to the usual setting of multiplayer game-theoretical models . The threshold z* increases with T due to diminishing direct benefits , and converges to a maximum value ( 4 ) z∞*=2cb in the limit of infinite T . Figure 3 displays the threshold z* as a function of the benefit-to-cost ratio b∕c for two different values of T ( blue and red full lines ) and in the limit case T→+∞ ( black full line ) . Overcoming adhesiveness level z∞* guarantees the selection of increased adhesiveness even in cases when groups are allowed to be very large , as commonly happens in microbial populations . The equilibrium z* can be classified as a 'garden of Eden' , i . e . a noninvasible equilibrium that is not convergence-stable ( Nowak , 1990 ) ( see Section S2 of the SI ) . 10 . 7554/eLife . 08595 . 005Figure 3 . Threshold adhesiveness value z* required for the evolution of increased adhesion . In the case of group formation by attachment , the theoretical value of z* in the limit of infinite T is z*=2c∕b , as demonstrated in Section 4 . 1 . Analytical thresholds ( full lines ) as well as numerical estimations ( circles ) are displayed for small ( 20 ) and large ( 100 ) values of T . As T decreases , threshold values decrease too because of enhanced direct benefits . Numerical results are consistent with analytical predictions . Error bars indicate the variability—associated with the finite size of the population—in the estimation of the threshold across multiple computations of the aggregation process . DOI: http://dx . doi . org/10 . 7554/eLife . 08595 . 005 In the following section , we focus on the evolutionary trajectories leading to an increase in adhesiveness , and discuss how the social nature of the trait changes in the course of evolution . Cooperation , defined as a behavior that increases other individuals’ fitness at a personal cost , has been classified in two categories , depending on the sign of the net effect of the behavior on the cooperator ( West et al . , 2007; Wilson , 1975 , 1990 ) . Indeed , marginal gains retrieved from a cooperator’s own contribution to the common good might be large enough to compensate its systematic cost: in this case , cooperation is termed mutualistic or directly beneficial , otherwise , it is coined altruistic ( West et al . , 2007 ) . Note that in both cases , cooperators fare worse than the defective members of the same group . The distinction in the status of cooperative acts is also known as weak vs . strong altruism ( Wilson , 1975 , 1990; Fletcher and Doebeli , 2009 ) . In randomly assorted populations , directly beneficial cooperation is favored by natural selection while altruistic cooperation is not ( Wilson , 1975 ) . Therefore , directly beneficial cooperation can be expected to establish first , as it is more easily obtained , potentially providing the substrate for altruistic forms of cooperation to spread later . This idea seems to be supported by the common observation of directly beneficial behavior in microbes . In invertase-secreting yeast , a small proportion of the hydrolized glucose is retained by the producer , providing advantage to cooperator cells at low frequencies ( Gore et al . , 2009 ) . Similarly , the bacterium Lactococcus lactis expresses an extracellular protease that helps transform milk proteins into digestible peptides: Bachmann et al . ( 2011 ) showed that such cooperative behavior can persist owing to a small fraction of the peptides being immediately captured by the proteolytic cells . In Pseudomonas aeruginosa colonies grown on solid substrates , the diffusion of pyoverdine is locally confined as cells are densely packed , relaxing the burden of public good production ( Julou et al . , 2013 ) . On the other hand , extremely sacrificial behavior such as altruistic suicide seems restricted to Myxobacteria and cellular slime moulds and its evolution appears to require mechanisms of reinforcement , though the peculiar life cycle of such organisms reduces the influence of genetic relatedness . It is important to notice that the status of a cooperative trait ( whether directly beneficial or altruistic ) can be defined only relatively to a given population structure . In the case of linear PGGs within groups of fixed size N , direct benefits of a cooperator amount to b∕N and cooperation is mutualistic whenever b∕N>c . Group size is thus crucial to define the status of a cooperative trait; the maintenance of cooperation in several models actually owes to alternating phases when it is altruistic or directly beneficial , generally because of group size variations ( Hauert et al . , 2002; Fletcher and Zwick , 2004; Killingback et al . , 2006 ) . In our model , the population structure depends on the value of the social trait itself . At any point along an evolutionary trajectory , we can compute the average 'gain from switching' ( Peña et al . , 2014 ) of a more adhesive mutation , that quantifies the balance between the additional cost of increased adhesion , and the marginal return from its additional contribution to group cohesion . Here , the gain from switching is calculated in the social context established by less adhesive residents , characterized by the group size distribution g ( n , z^ , z^ ) . We call 'altruisitic’ any positive mutation on adhesiveness such that its gain from switching is negative , and ’directly beneficial’ any mutation whose gain from switching is positive . These definitions extend those for groups of fixed size and encapsulate the concept of altruism as a situation in which individual costs are not immediately recovered by marginal benefits , all other things unchanged ( but see Kerr et al . ( 2004 ) for other individual-based interpretations of altruism ) . The condition for a small positive mutation to be altruistic in a resident population of trait z^ is , as showed in Section S3 of the SI: ( 5 ) ralt ( z^ ) b<c where: ( 6 ) ralt ( z^ ) =∑n≥2g ( n , z^ , z^ ) n . This condition is comparable to that found in the classical PGG with fixed group size ( b<Nc ) , as 1/ralt ( z^ ) is homogeneous to a group size and can be linked to the average group size in the population . Indeed , for the differential attachment model of Section 2 . 2 , ( 7 ) ralt ( z^ ) = z^γ^ z^ , where γ^z^ is the average group size ( across groups , not individuals—not to be confused with insider’s view of group size or crowding ( Jarman , 1974; Reiczigel et al . , 2008 ) ) . In highly adhesive populations ( z^≈1 ) with groups of size T and no ungrouped individuals , Equation 5 is then simply b∕c<T; while in poorly adhesive populations ( z^≈0 ) where all individuals are ungrouped , no direct benefit is possible , so that ralt vanishes and Equation 5 is always satisfied . Figure 4 displays , for various values z^ of the resident trait , the benefit-to-cost ratio 1∕rmin above which an increase in adhesiveness is favored and the benefit-to-cost ratio 1∕ralt below which such a mutation is classified as altruistic . Let us suppose that the benefit-to-cost ratio b∕c remains fixed all along an evolutionary trajectory as the resident trait varies . Depending on its values , an increase in adhesiveness can be attributed a different social status: ( 1 ) if b∕c>T , social mutations are altruistic at the onset of social evolution , and directly beneficial when z^ becomes larger; ( 2 ) if 2<b∕c<T , social mutations are altruistic throughout the evolutionary trajectory; ( 3 ) if b∕c<2 , social mutations never invade . 10 . 7554/eLife . 08595 . 006Figure 4 . Status of social mutations . For any resident adhesiveness value z^between 0 and 1 , we display , in black: the minimal benefit-to-cost ratio 1/rmin=2/z^ for a social ( or positive ) mutation to be selected; in red: the maximal benefit-to-cost ratio such that this mutation is altruistic . Let us choose a fixed b∕c ( i . e . an horizontal line in the graph ) . According to the value of b∕c , the fate and the social status of positive mutations change . For low b∕c ( <2 ) , all social mutations are altruistic but none of them is ever selected: the population is doomed to full asociality . For intermediate b∕c ( between 2 and T ) , social mutations are favored as soon as z^ overcomes a threshold ( crossing of the black line with the horizontal line y=b∕c ) , and are altruistic all along the evolutionary dynamics . For large b∕c ( >T ) , once the threshold is overcome and z^ increases , social mutations are altruistic until some value of z^ ( crossing of the red line with the horizontal line y=b∕c ) ; afterwards , social mutations turn directly beneficial . DOI: http://dx . doi . org/10 . 7554/eLife . 08595 . 006 The first case in particular demonstrates that the status of a social mutation can change along an evolutionary trajectory , and that this change is more likely to occur the smaller the maximal size groups can attain . Contrary to what we expected , the social dilemma raised by costly mutations in adhesiveness relaxes over the course of evolution , leading from altruistic to mutually beneficial behavior . Indeed , even when the benefit-to-cost ratio is large ( greater than the maximal group size T ) , social mutations are not directly beneficial at start . This is because , adhesiveness being small , chances are high to be left outside groups and miss any group-related benefit whatsoever . This suggests that altruism may play a more important role in the origin of social behavior than it is currently assumed , and that it could be a first attainable step even when it involves large costs , eventually paving the way to the evolution of less sacrificial behavior . The generality of the analytical results obtained in Sections 3 and 4 rely on a number of hypotheses that are not strictly realized in actual biological populations: infinite population size; infinitesimal mutational steps; the equivalence between positive growth rate ( when rare ) and the fixation of the trait; the linear dependence of group-related benefits on group average composition . Whereas the soundness of adaptive dynamics equations as limits of individual-based processes has been mathematically proved ( Champagnat and Lambert , 2007 ) , the introduction of potential nonlinearities in the payoff function of our model quickly makes analytical calculations intractable . We thus implemented the life cycle described in Section 2 . 2 in an individual-based model ( detailed in Section 4 of the SI ) . This way , we could check the validity of the results of Section 4 . 1 , here summarized by Pairwise Invasibility Plots ( PIP ) ( Geritz et al . , 1998 ) , and repeat the analysis also for biologically interesting , non-linear payoff functions . Figure 5 displays the PIP for the individual-based model , representing the sign of the invasion fitness for each pair of resident z^ and mutant z trait values . A single interior singular strategy z* can be observed and characterized as a 'garden of Eden' in accordance with the analytical results of Section 4 . 1: in a neighbourhood of z* , the population can be successively invaded either by mutants with decreasing trait values until z^=0 , or by mutants with increasing trait values until z^=1 . Figure 3 confirms the consistence between the thresholds obtained with the theoretical analyses ( full lines ) with those obtained by numerical simulations ( circles ) for two values of T . However , stochastic fluctuations due to a finite population size can , when the maximal group size T is small , decrease such threshold to the extent that increased adhesiveness evolves for scratch ( blue circles ) . 10 . 7554/eLife . 08595 . 007Figure 5 . Pairwise invasibility plot obtained by simulation of the toy model for differential attachment . A positive invasion fitness ( gray ) means that the mutant can invade the population and replace the resident trait whereas a negative invasion fitness ( white ) means that the mutant is outcompeted . A singular point is found around 0 . 1=2c∕b=z* , consistently with analytical predictions . This equilibrium can be characterized as a 'garden of Eden' ( non-invasible repellor ) , which means that , depending on the position of the initial value z^0 of z^ with respect to z* , evolutionary dynamics leads to the selection of either z^=0 ( when z^0<z* ) or z^=1 ( when z^0>z* ) , i . e . either full asociality or full sociality . Parameters: T=100 , N=5000 , b∕c=20 . DOI: http://dx . doi . org/10 . 7554/eLife . 08595 . 007 While a linear PGG is a parsimonious—and tractable—way to depict social dilemmas occurring in multicellular aggregates , it is reasonable to imagine that actual cost and benefit functions are more convoluted . Notably , excessive group sizes and investments in adhesion might be constrained by intrinsic properties of microorganisms . By numerical simulations , we thus explored two alternative , nonlinear functional forms of the benefit or the cost function . First , we considered the situation when there is a physical upper limit to the size of functional groups . For instance , fruiting bodies of Dictyostelium become unstable and topple over if they are too large ( Savill and Hogeweg , 1997 ) . We model this by adding a group size threshold above which group benefits become null ( Figure 6 ) . Second , we modelled possible trade-offs between the production of the glue and other cell functions , so that producing large amounts of glue results in a disproportionate decrease in fitness ( Goymer et al . , 2006 ) . This scenario corresponds to a cost function that is linear for small values of the adhesiveness z but diverges as z gets closer to 1 ( Figure 7 ) . 10 . 7554/eLife . 08595 . 008Figure 6 . Pairwise invasibility plot obtained when group-related benefits are null above group size αT . Here , the change in the benefit function leads to the appearance of a second equilibrium z+ that is convergence-stable and non-invasible . As soon as the initial value of the adhesiveness trait ẑ is larger than the adhesiveness threshold z* , selection favors adhesion level z+ at equilibrium . Parameters: T=100 , N=5000 . DOI: http://dx . doi . org/10 . 7554/eLife . 08595 . 00810 . 7554/eLife . 08595 . 009Figure 7 . Pairwise invasibility plot obtained when individual cost diverges for large adhesiveness values . As in the previous case , an other equilibrium appears that is a CSS ( convergence-stable strategy ) , hence the evolutionary endpoint as soon as the initial adhesiveness value overcomes the threshold z* . Parameters: T=100 , N=5000 . DOI: http://dx . doi . org/10 . 7554/eLife . 08595 . 009 Both cases lead to the appearance of a second internal singular strategy z+ <1 that can be characterized as a continuously stable ( i . e . convergence-stable and noninvasible ) strategy . This adaptive equilibrium is associated with populations structured in groups of sizes distributed around an average z+ T and a proportion 1-z+ of solitary individuals . Such an evolutionary outcome is more in line with observations of cellular slime moulds , where groups coexist with a fraction of nonaggregated cells ( Dubravcic et al . , 2014; Tarnita et al . , 2015 ) .
Social behavior is expected to occur in populations already endowed with some minimal ability of attachment before the multicellular stage evolved . Is this condition relevant to biologically realistic settings ? Adhesiveness is a trait that is highly variable among microbial species , depending on the physical properties of the cell surface , as well as on the amount and quality of compounds they excrete . Directed selection experiments show that the secretion of extracellular products providing some kind of collective advantage—such as polysaccharides improving the strength of aggregates and stress resistance—can be achieved on a relatively fast time scale ( Rainey and Rainey , 2003; Xavier and Foster , 2007 ) . Overcoming a threshold in adhesiveness should thus not be a strong constraint on the evolution of social behavior , as long as the collective function provides a sufficient benefit . Cells with no or little capacity of motion can rely on spatial patterning through clonal growth to ensure a high degree of local genetic and phenotypic homogeneity . Adhesion , on the other hand , can play such an assortative role for cells that actively move . Adhesiveness may therefore significantly contribute to the evolution of multicellular organization by causing both assortment during group formation , and cell sorting at a later , developmental , stage ( Savill and Hogeweg , 1997 ) , thus underpinning both collective function and division of labor . A change in adhesion per se , and not specifically to other cells , may however have drawbacks , if it is associated with a modification in dispersal properties . In D . discoideum at the onset of the aggregation phase , for instance , enhanced chemotaxis upon starvation ( Schäfer et al . , 2013 ) increases adherence to both the substrate and the neighboring cells . If higher adhesiveness drives more effective crawling , increased dispersal and the subsequent genetic heterogeneity in multicellular chimeras might oppose the selection for collective function . The emergence of sociality in spite of spatial mixing has been addressed in a simple model where cells are described as interacting persistent random walkers ( Garcia et al . , 2014 ) . In this model , however , cell velocity and adhesiveness were independent parameters . Their concomitant variation might lead , in a spatially explicit setting , to insights on the conditions under which stable polymorphism is possible in a population with gradually evolving adhesiveness . In the aggregation phase , populations of social amoebae D . discoideum do not only form multicellular groups , they also leave behind solitary cells . Recent experiments show that the existence of an ungrouped component is widespread in lab and wild strains . Theoretical arguments support the idea that nonaggregated cells play an essential role in the evolution of some peculiarities of those organisms ( Dubravcic et al . , 2014; Tarnita et al . , 2015; Rainey , 2015 ) . Neglecting cells that are outside groups alters population-level statistics , such as the average fitness of a given strategy , and thus affect the prediction of evolutionary outcomes . Although widely disregarded in experiments , typically focused on the multicellular body , the possible role of solitary cells in the evolution of cooperative or social behavior has been a recurrent theme in theoretical works . Early simulations modeling the evolution of cellular slime moulds suggested that individuals ’diffusing’ out of groups may contribute to the maintenance of cooperative behavior by forgoing the systematic cost of grouping , together with its potential benefits ( Armstrong , 1984 ) . Hauert et al . ( 2002 ) stressed that a 'loner' strategy—whereby individuals adopt an autarkic lifestyle—can provide a way out of the deadlock of the tragedy of the commons by a periodic soaring of individuals that opt out of the collective enterprise . Even though the predicted oscillations in group size have never , to our knowledge , been observed in microbial populations , they might be conceived as primitive life cycles including a collective phase . Solitary behavior has also been explained as an adaptation to variable environmental conditions ( Dubravcic et al . , 2014; Tarnita et al . , 2015; Rainey , 2015 ) : cells hedge their evolutionary bets between commitment to a developmental program with uncertain outcome ( they may end up in the stalk or in the spores ) and the wait for the re-establishment of conditions favorable to growth . The theoretical prediction in this case is that the fraction of lonely cells reflects the optimal probability of remaining outside groups given the pattern of environmental variation . The evolutionary model discussed here also supports the claim that lonely cells , which happen to remain outside groups by chance , are of primary importance for the onset of sociality . As discussed in Section 3 for a general group formation process , and illustrated in Section 4 with a toy model for aggregation , nonaggregated individuals might tip the balance in favor of stickier types . This conclusion is reinforced by the fact that , in our model , the fitness of an individual depends on the composition of the group it belongs to—as opposed to the assumption in Dubravcic et al . ( 2014 ) and Tarnita et al . ( 2015 ) that , when in groups , the probability of becoming spores is constant . At the evolutionary equilibrium , however , the ungrouped component of the population vanishes , unless nonlinearities are introduced in the functional dependence of collective benefits and/or individual costs on adhesiveness , as observed in Section 4 . 3 . Some important differences between Dubravcic et al . ( 2014 ) and Tarnita et al . ( 2015 ) and our model should be noted . When group formation is based on adhesiveness , lonely individuals occur also in the absence of temporal variation of the environment and of selective advantage due to dispersal . Moreover , in our model , being ungrouped is not a strategy determined by a fixed probability , as failure to join a group is contingent on the emergent population structure . A pivotal issue is thus when the decision to join groups is made by microbial organisms . Populations of facultatively multicellular species face two ’lotteries’ where stochasticity affects the fitness of an individual: joining groups and , once in a group , reproducing or dying . If the trait under selection is , as in Dubravcic et al . ( 2014 ) and Tarnita et al . ( 2015 ) , the probability of staying alone , cells decide to be in a group before the fitness cost of giving up reproduction is possibly assigned . If it is adhesiveness , as in our model , social behavior in groups is decided at the same time as aggregative behavior . Experimental evidence suggests that conditions preceding the aggregation phase influence the partition between lonely and aggregated cells ( Dubravcic et al . , 2014 ) , and between stalk and spore cells ( Gomer and Firtel , 1987; Jang and Gomer , 2011 ) . The role of adhesiveness itself in influencing such decisions is unknown . Further experiments are needed , together with models that explore the possible role of competition within aggregates , to clarify to what extent social interactions within the multicellular phase are linked to group size distribution , including singletons . Directly beneficial cooperation is by definition more easily established than altruistic cooperation ( West et al . , 2007 ) . Example of microbial populations where individuals draw direct benefits from the production of extracellular compounds are numerous ( Gore et al . , 2009; Bachmann et al . , 2011; Julou et al . , 2013; Zhang and Rainey , 2013 ) . One could thus expect that , by sustaining the onset of cooperation , directly beneficial behavior sets the scene for the later establishment of altruistic traits , of which 'suicide for the good of the group' is emblematic . Deciding the nature ( altruistic or not ) of a given social mutation is not just a matter of costs and benefits , but also depends on the way the population is structured . As a consequence , the nature of a directional change in the trait value can vary along an evolutionary trajectory , if it entails modifications in the average local interaction environment . We show that the status of mutations increasing adhesiveness can change along with the gradual evolution of the trait , even when benefit and cost parameters are fixed . When adhesion starts to increase , mutations are always altruistic , meaning that they entail a net cost to the individual compared to the resident trait . Nonetheless , these altruistic mutants thrive because of their role in the earlier aggregation phase . For large enough benefit-to-cost ratios , though , positive mutations become directly beneficial at a later time along the evolutionary trajectory . Altruistic behavior emerges first , and direct benefits only appear afterwards . This means that directly beneficial interactions , while frequently observed in nature , might not necessarily be the stepping stone for the evolution of cooperation , but rather an evolutionary endpoint . The gradual increase in adhesiveness is a mechanistic—and seemingly readily achievable—way by which unicellular organisms may have made the transition to complex , multicellular organization ( De Monte and Rainey , 2014; Rainey and De Monte , 2014 ) . Such a mechanism not only accounts for assortment , but potentially underpins spatial segregation in tissues ( Steinberg , 2007; Marée and Hogeweg , 2001 ) , through which evolutionary forces might have moulded the developmental program of the multicellular body . The relevance of adhesion in structuring the collective phase of organisms that aggregate from sparse cells however requires more experimental observations , that will hopefully shed light on the biological and physical processes involved in the major evolutionary transition to multicellularity . | Throughout the living world , organisms work together in groups and help each other to survive . Indeed , multicellular organisms such as plants and animals owe their existence to cooperation . Life on Earth was initially made up of single cells , some of which evolved the ability to stick to each other and work together to form tissues and organs . However , developing the ability to adhere to other cells costs energy that could otherwise be used by the cell to ensure its own survival and proliferation . How multicellularity emerged , despite such costs , remains puzzling , in particular in groups of cells that do not share a common ancestor . Now , Garcia , Doulcier and De Monte have produced a mathematical model that shows how large cohesive groups of cells can evolve . Over long periods of time , these groups can emerge from a population of non-adhesive cells through a series of small mutations that increase the overall adhesiveness of the cells in the group . Furthermore , the evolution of cohesive groups can arise just through the cells randomly interacting . By contrast , previous models that investigated how social groups form have tended to assume that particular cell types preferentially interact with each other . The model also suggests that the costs associated with developing adhesiveness can be partially compensated for in groups that contain cells with different abilities to adhere to each other . This means that individual cells that do not join any groups also play a crucial role in the development of cohesive groups . Finally , Garcia , Doulcier and De Monte challenge the popular belief that social behavior arises primarily because it is beneficial to the individual performing those actions . Instead , the model suggests that selfless cooperation may occur first , and only afterwards lead to the evolution of behavior that is mutually beneficial for the individuals involved . In the future , the plausibility of the evolutionary path suggested by the model could be tested in experiments using single-celled organisms such as some amoebae and bacteria , that , along their life cycle , alternatively live alone and in cohesive groups . | [
"Abstract",
"1",
"Introduction",
"2",
"Adhesiveness",
"as",
"a",
"quantitative",
"trait",
"affecting",
"group",
"formation",
"and",
"function",
"3",
"Adaptive",
"dynamics",
"of",
"adhesiveness",
"4",
"The",
"evolution",
"of",
"adhesiveness",
"by",
"attachment",
"5... | [
"evolutionary",
"biology",
"computational",
"and",
"systems",
"biology"
] | 2015 | The evolution of adhesiveness as a social adaptation |
The phylogenetic placement of the morphologically simple placozoans is crucial to understanding the evolution of complex animal traits . Here , we examine the influence of adding new genomes from placozoans to a large dataset designed to study the deepest splits in the animal phylogeny . Using site-heterogeneous substitution models , we show that it is possible to obtain strong support , in both amino acid and reduced-alphabet matrices , for either a sister-group relationship between Cnidaria and Placozoa , or for Cnidaria and Bilateria as seen in most published work to date , depending on the orthologues selected to construct the matrix . We demonstrate that a majority of genes show evidence of compositional heterogeneity , and that support for the Cnidaria + Bilateria clade can be assigned to this source of systematic error . In interpreting these results , we caution against a peremptory reading of placozoans as secondarily reduced forms of little relevance to broader discussions of early animal evolution .
The discovery ( Schulze , 1883 ) and mid-20th century rediscovery ( Grell and Benwitz , 1971 ) of the enigmatic , amoeba-like placozoan Trichoplax adhaerens did much to ignite the imagination of zoologists interested in early animal evolution ( Bütschli , 1884 ) . As microscopic animals adapted to extracellular grazing on the biofilms over which they creep ( Wenderoth , 1986 ) , placozoans have a simple anatomy suited to exploit passive diffusion for many physiological needs , with only six morphological cell types discernible even to intensive microscopical scrutiny ( Grell and Ruthmann , 1991; Smith et al . , 2014 ) , albeit a greater diversity of cell types is apparent through single-cell RNA-seq ( Sebé-Pedrós , 2018a ) . They have no conventional muscular , digestive , or nervous systems , yet show tightly-coordinated behaviour regulated by peptidergic signaling ( Smith et al . , 2015; Senatore et al . , 2017; Varoqueaux , 2018; Armon et al . , 2018 ) . In laboratory conditions , they proliferate through fission and somatic growth . Evidence for sexual reproduction remains elusive , despite genetic evidence of recombination ( Srivastava et al . , 2008 ) and descriptions of early abortive embryogenesis ( Eitel et al . , 2011; Grell , 1972 ) , with the possibility that sexual phases of the life cycle may occur only under poorly understood field conditions ( Pearse and Voigt , 2007; McFall-Ngai et al . , 2013 ) Given their simple , puzzling morphology and dearth of embryological clues , molecular data are crucial in placing placozoans phylogenetically . The position of Placozoa in the animal tree proved recalcitrant to early standard-marker analyses ( Kim et al . , 1999; Silva et al . , 2007; Wallberg et al . , 2004 ) , although this paradigm did reveal a large degree of molecular diversity in placozoan isolates from around the globe , clearly indicating the existence of many cryptic species ( Pearse and Voigt , 2007; Eitel et al . , 2013; Signorovitch et al . , 2007 ) with up to 27% genetic distance in 16S rRNA alignments ( Eitel and Schierwater , 2010 ) . An apparent answer to the question of placozoan affinities was provided by analysis of a nuclear genome assembly ( Srivastava et al . , 2008 ) , which strongly supported a position as the sister group of a clade of Cnidaria + Bilateria ( sometimes called Planulozoa ) . However , this effort also revealed a surprisingly bilaterian-like ( Dunn et al . , 2015 ) developmental gene toolkit in placozoans , a paradox for such a simple animal . As metazoan phylogenetics has pressed onward into the genomic era , perhaps the largest controversy has been the debate over the identity of the sister group to the remaining metazoans , traditionally thought to be Porifera , but considered to be Ctenophora by Dunn et al ( Dunn et al . , 2008 ) . and subsequently by additional studies ( Hejnol et al . , 2009; Moroz et al . , 2014; NISC Comparative Sequencing Program et al . , 2013; Whelan et al . , 2015; Whelan et al . , 2017 ) . Others have suggested that this result arises from artifacts with potentially additive effects , such as inadequate taxon sampling , flawed matrix husbandry ( undetected paralogy or contamination ) , and use of poorly fitting substitution models ( Philippe et al . , 2009; Pick et al . , 2010; Pisani et al . , 2015; Simion et al . , 2017; Feuda et al . , 2017 ) . A third view has emphasized that using different sets of genes can lead to different conclusions , with only a small number sometimes sufficient to drive one result or another ( Nosenko et al . , 2013; Shen et al . , 2017 ) . This controversy , regardless of its eventual resolution , has spurred serious contemplation of possibly independent origins of several hallmark traits such as striated muscles , digestive systems , and nervous systems ( Moroz et al . , 2014; Dayraud et al . , 2012; Hejnol and Martín-Durán , 2015; Liebeskind et al . , 2017; Moroz and Kohn , 2016; Presnell et al . , 2016; Steinmetz et al . , 2012 ) . Driven by this controversy , new genomic and transcriptomic data from sponges , ctenophores , and metazoan outgroups have accrued , while new sequences and analyses focusing on the position of Placozoa have been slow to emerge . Here , we provide a novel test of the phylogenetic position of placozoans , adding draft genomes from three putative species that span the root of this clade’s known diversity ( Eitel et al . , 2013 ) ( Table 1 ) , and critically assessing the role of systematic error in placing of these enigmatic organisms ( Laumer , 2018 ) .
Orthology assignment on sets of predicted proteomes derived from 59 genome and transcriptome assemblies yielded 4294 gene trees with at least 20 sequences each , sampling all five major metazoan clades and outgroups , from which we obtained 1388 well-aligned orthologues . Within this set , individual maximum-likelihood ( ML ) gene trees were constructed , and a set of 430 most-informative orthologues were selected on the basis of tree-likeness scores ( Misof et al . , 2013 ) . This yielded an amino-acid matrix of 73 , 547 residues with 37 . 55% gaps or missing data , with an average of 371 . 92 and 332 . 75 orthologues represented for Cnidaria and Placozoa , respectively ( with a maximum of 383 orthologues present for the newly sequenced placozoan H4 clade representative; Figure 1 ) . Our Bayesian analyses of this matrix place Cnidaria and Placozoa as sister groups with full posterior probability under the general site-heterogeneous CAT + GTR + Г4 model ( Figure 1 ) . Under ML inference with the C60 +LG + FO + R4 profile mixture model ( Wang et al . , 2018 ) ( Figure 1—figure supplement 1 ) , we again recover Cnidaria + Placozoa , albeit with more marginal resampling support . Both Bayesian and ML analyses show little internal branch diversity within Placozoa . Accordingly , deleting all newly-added placozoan genomes from our analysis has no effect on topology and only a marginal effect on support in ML analysis ( Figure 1—figure supplement 2 ) . Quartet-based concordance analyses ( Zhou , 2017 ) show no evidence of strong phylogenetic conflicts among ML gene trees in this 430-gene set ( Figure 1 ) , although internode certainty metrics are close to 0 for many key clades including Cnidaria + Placozoa , indicating that support for some ancient relationships may be masked by gene-tree estimation errors , emerging only in combined analysis ( Gatesy and Baker , 2005 ) . Compositional heterogeneity of amino-acid frequencies along the tree is a source of phylogenetic error not modelled by even complex site-heterogeneous substitution models such as CAT+GTR ( Blanquart and Lartillot , 2008; Foster , 2004; Lartillot and Philippe , 2004; Lartillot et al . , 2013 ) . Furthermore , previous analyses ( Nosenko et al . , 2013 ) have shown that placozoans and choanoflagellates in particular , both of which taxa our matrix samples intensively , deviate strongly from the mean amino-acid composition of Metazoa , perhaps as a result of genomic GC content discrepancies . As a measure to at least partially ameliorate such nonstationary substitution , we recoded the amino-acid matrix into the 6 ‘Dayhoff’ categories , a common strategy previously shown to reduce the effect of compositional variation among taxa , albeit the Dayhoff-6 groups represent only one of many plausible recoding strategies , all of which sacrifice information ( Feuda et al . , 2017; Nesnidal et al . , 2010; Rota-Stabelli et al . , 2013; Susko and Roger , 2007 ) . Analysis of this recoded matrix under the CAT + GTR model again recovered full support ( pp = 1 ) for Cnidaria + Placozoa ( Figure 2 ) . Indeed , under Dayhoff-6 recoding , the only major change is in the relative positions of Ctenophora and Porifera , with the latter here constituting the sister group to all other animals with full support . Similar recoding-driven effects on relative positions of Porifera and Ctenophora have also been seen in other recent work ( Feuda et al . , 2017 ) , and have been interpreted to indicate a role for compositional bias in misplacing Ctenophora as sister group to all other animals Many research groups , using good taxon sampling and genome-scale datasets , and even recently including data from a new divergent placozoan species ( Whelan et al . , 2017; Feuda et al . , 2017; Eitel , 2017 ) , have consistently reported strong support for Planulozoa under the CAT + GTR model . Indeed , when we construct a supermatrix from our predicted peptide catalogues using a different strategy , relying on complete sequences of 303 pan-eukaryote ‘Benchmarking Universal Single-Copy Orthologs’ ( BUSCOs ) ( Simão et al . , 2015 ) , we also see full support in a CAT + GTR + Г analysis for Planulozoa , in both amino-acid ( Figure 3a ) and Dayhoff-6 recoded alphabets ( Figure 3b ) . Which phylogeny is correct , and what process drives support for the incorrect topology ? Posterior predictive tests , which compare the observed among-taxon usage of amino-acid frequencies to expected distributions simulated using the sampled posterior distribution and a single composition vector , may provide insight ( Feuda et al . , 2017; Lartillot and Philippe , 2004 ) . Both the initial 430-gene matrix and the 303-gene BUSCO matrix fail these tests , but the BUSCO matrix fails it more profoundly , with z-scores ( measuring mean-squared across-taxon heterogeneity ) scoring in the range of 330–340 , in contrast to the range of 176–187 seen in the 430-gene matrix ( Table 2 ) . Furthermore , inspecting z-scores for individual taxa in representative chains from both matrices shows that a large amount of this global difference in z-scores can be attributed to placozoans , with additional contributions from choanoflagellates and select isolated representatives of other clades ( Figure 3C ) . As a final measure to describe the influence of compositional heterogeneity in this dataset , we applied a null-simulation test for compositional bias to each alignment in our set of 1388 orthologues . This test , which compares the real data to a null distribution of amino-acid frequencies simulated along assumed gene trees with a substitution model using a single composition vector , is less prone to Type II errors than the more conventional X ( Grell and Benwitz , 1971 ) test ( Foster , 2004 ) . Remarkably , at a conservative significance threshold of α = 0 . 10 , the majority ( 764 genes or ~55% ) of this gene set is identified as compositionally biased by this test , highlighting the importance of using appropriate statistical tests to control this source of systematic error , rather than applying arbitrary heuristic cutoffs ( Kück and Struck , 2014 ) . Building informative matrices from gene sets on either side of this significance threshold , and again applying both CAT + GTR mixture models and ML profile mixtures , we see strong support for Cnidaria + Placozoa in the test-passing supermatrix , and conversely , strong support for Cnidaria + Bilateria in the test-failing supermatrix ( Figure 4 , Figure 4—figure supplement 1 , Figure 4—figure supplement 2 ) . Interestingly , in trees built through CAT + GTR + Г4 analysis of the test-failing supermatrix ( Figure 4A , C ) , in both amino-acid and Dayhoff-6 alphabets , we also observe full support for Porifera as sister to all other animals . In contrast , analysis of this amino acid matrix under a profile mixture model recovers support for Ctenophora in this position ( Figure 4—figure supplement 1 ) , indicating that , at least for this alignment , compositional heterogeneity need not be invoked to explain why outcomes differ among analyses , as some have argued ( Feuda et al . , 2017 ) : both CAT + GTR and the C60 +LG + FO + R4 profile mixture model assume a single composition vector over time , but the CAT + GTR model is better able to accommodate site-heterogeneous substitution patterns ( Lartillot et al . , 2013; Quang et al . , 2008 ) . In the context of this experiment , Dayhoff-6 recoding appears impactful only for the test-passing supermatrix ( Figure 4B , D ) , where it obviates support for Ctenophora-sister ( Figure 4B , Figure 4—figure supplement 2 ) in favour of ( albeit , with marginal support ) Porifera-sister ( Figure 4D ) , and also diminishes support for Placozoa + Cnidaria ( in contrast to the 430-gene matrix; Figure 2 ) , perhaps reflecting the inherent information loss of using a reduced amino-acid alphabet for this relatively shorter matrix . A possible hidden variable related to the phylogenetic discordance we describe , the precise significance of which remains unclear , is mean trimmed alignment length: both the test-passing and the original 430-gene matrix are composed of considerably shorter alignments than the test-failing and the 303-gene BUSCO matrix ( see Materials and methods ) . Indeed , alignment length has been previously shown to be predictive of a number of other metrics of phylogenetic relevance ( Shen et al . , 2016 ) ; the generality and directionality of such relationships in empirical datasets at varying scales of divergence is clearly worthy of further investigation . The previously cryptic phylogenetic link between cnidarians and placozoans seen in gene sets less influenced by compositional bias will require further testing with other analyses and data modalities , such as rare genomic changes , which should be ever more visible as highly contiguous assemblies continue to be reported from non-bilaterian animals ( Eitel et al . , 2018; Kamm et al . , 2018; Jiang , 2018; Leclère , 2018 ) . However , if validated , this relationship must continue to raise questions on the homology of certain traits across non-bilaterians . Many workers , citing the incompletely known development ( Eitel et al . , 2011; Pearse and Voigt , 2007 ) and relatively bilaterian-like gene content of placozoans ( Srivastava et al . , 2008; Eitel , 2017 ) , presume that these organisms must have a still-unobserved , more typical development and life cycle ( DuBuc et al . , 2018 ) , or else are merely oddities that have experienced wholesale secondary simplification , having scant significance to any evolutionary path outside their own . Indeed , it is tempting to interpret this new phylogenetic position as further bolstering such hypotheses , as much work on cnidarian models in the evo-devo paradigm is predicated on the notion that cnidarians and bilaterians share , more or less , many homologous morphological features , viz . axial organization ( Genikhovich and Technau , 2017; DuBuc et al . , 2018 ) , nervous systems ( Liebeskind et al . , 2017; Moroz and Kohn , 2016; Kelava et al . , 2015; Kristan , 2016; Arendt et al . , 2016 ) , basement-membrane lined epithelia ( Fidler et al . , 2017; Leys and Riesgo , 2012 ) , musculature ( Steinmetz et al . , 2012 ) , embryonic germ-layer organisation ( Steinmetz et al . , 2017 ) , and internal digestion ( Presnell et al . , 2016; Putnam et al . , 2007; Hejnol and Martindale , 2008; Martindale and Hejnol , 2009 ) . While we do not argue , as some have done ( Schierwater , 2005; Syed and Schierwater , 2002 ) , that placozoans resemble hypothetical metazoan ancestors , we hesitate to dismiss them a priori as irrelevant to understanding early bilaterian evolution in particular: although apparently simpler and less diverse , placozoans nonetheless have equal status to cnidarians as an immediate extant outgroup . Rather , we see value in testing assumed hypotheses of homology , character by character , by extending pairwise comparisons between bilaterians and cnidarians to include placozoans , an agenda which demands reducing the large disparity in embryological , physiological , and molecular genetic knowledge between these taxa , towards which recent progress has been made using both established methods such as in situ hybridization ( DuBuc et al . , 2018 ) and image analysis ( Varoqueaux , 2018 ) , as well as new technologies such as single-cell RNA-seq ( Sebé-Pedrós , 2018a; Sebé-Pedrós et al . , 2018b ) . Conversely , we emphasize another implication of this phylogeny: characters that can be validated as homologous at any level between Bilateria and Cnidaria must have originated earlier in animal evolution than previously appreciated , and should either cryptically occur in modern placozoans or else have been lost at some point in their ancestry . In this light , paleobiological scenarios of early animal evolution founded on inherently phylogenetically-informed interpretations of Ediacaran fossil forms ( Cavalier-Smith , 2018; Cavalier-Smith , 2017; Dufour and McIlroy , 2018; Sperling and Vinther , 2010; Evans et al . , 2017 ) and molecular clock estimates ( Cunningham et al . , 2017; dos Reis et al . , 2015; Dohrmann and Wörheide , 2017; Erwin et al . , 2011 ) may require re-examination .
Haplotype H4 and H6 placozoans were collected from water tables at the Kewalo Marine Laboratory , University of Hawaii-Manoa , Honolulu , Hawaii in October 2016 . Haplotype H11 placozoans were collected from the Mediterranean ‘Anthias’ show tank in the Palma de Mallorca Aquarium , Mallorca , Spain in June 2016 . All placozoans were sampled by placing glass slides suspended freely or mounted in cut-open plastic slide holders into the tanks for 10 days ( Pearse and Voigt , 2007 ) . Placozoans were identified under a dissection microscope and single individuals were transferred to 500 µl of RNAlater , stored as per manufacturer’s recommendations . DNA was extracted from 3 individuals of haplotype H11 and 5 individuals of haplotype H6 using the DNeasy Blood and Tissue Kit ( Qiagen , Hilden , Germany ) . DNA and RNA from three haplotype H4 individuals were extracted using the AllPrep DNA/RNA Micro Kit ( Qiagen ) , with both kits used according to manufacturer’s protocols . Illumina library preparation and sequencing was performed by the Max Planck Genome Centre , Cologne , Germany , as part of an ongoing metagenomics project in marine symbiosis . In brief , DNA/RNA quality was assessed with the Agilent 2100 Bioanalyzer ( Agilent , Santa Clara , USA ) and the genomic DNA was fragmented to an average fragment size of 500 bp . For the DNA samples , the concentration was increased ( MinElute PCR purification kit; Qiagen , Hilden , Germany ) and an Illumina-compatible library was prepared using the Ovation Ultralow Library Systems kit ( NuGEN , Leek , The Netherlands ) according the manufacturer’s protocol . For the haplotype H4 RNA samples , the Ovation RNA-seq System V2 ( NuGen , 376 San Carlos , CA , USA ) was used to synthesize cDNA and sequencing libraries were then generated with the DNA library prep kit for Illumina ( BioLABS , Frankfurt am Main , Germany ) . All libraries were size selected by agarose gel electrophoresis , and the recovered fragments quality assessed and quantified by fluorometry . For each DNA library 14 – 75 million 100 bp or 150 bp paired-end reads were sequenced on Illumina HiSeq 2500 or 4000 machines ( Illumina , San Diego , U . S . A ) ; for the haplotype H4 RNA libraries 32 – 37 million single 150 bp reads were obtained . For assembly , adapters and low-quality reads were removed with bbduk ( https://sourceforge . net/projects/bbmap/ ) with a minimum quality value of two and a minimum length of 36 and single reads were excluded from the analysis . Each library was error corrected using BayesHammer ( Nikolenko et al . , 2013 ) . A combined assembly of all libraries for each haplotype was performed using SPAdes 3 . 62 ( Bankevich et al . , 2012 ) . Haplotype four and H11 data were assembled from the full read set with standard parameters and kmers 21 , 33 , 55 , 77 , 99 . The Haplotype H6 data was preprocessed to remove all reads with an average kmer coverage <5 using bbnorm and then assembled with kmers 21 , 33 , 55 and 77 . Reads from each library were mapped back to the assembled scaffolds using bbmap ( https://sourceforge . net/projects/bbmap/ ) with the option fast = t . Scaffolds were binned based on the mapped read data using MetaBAT ( Kang et al . , 2015 ) with default settings and the ensemble binning option activated ( switch –B 20 ) . The Trichoplax host bins were evaluated using metawatt ( Strous et al . , 2012 ) based on coding density and sequence similarity to the Trichoplax H1 reference assembly ( NZ_ABGP00000000 . 1 ) . The bin quality metrics were computed with BUSCO2 ( Simão et al . , 2015 ) ( Table 1 ) and QUAST ( Gurevich et al . , 2013 ) . Both the stringent metagenomics binning procedure ( a procedure also expedient in other holobiont organisms ( Celis et al . , 2018 ) ) and the very low proportion of multiple orthologue hits in the BUSCO2 assessment ( Table 1 ) attest to the lack of evidence for residual non-placozoan contamination within the scaffolds used for gene prediction . Predicted proteomes from species with published draft genome assemblies were downloaded from the NCBI Genome portal or Ensembl Metazoa in June 2017 . For Clade A placozoans , host metagenomic bins were used directly for gene annotation . For the H6 and H11 representatives , annotation was entirely ab initio , performed with GeneMark-ES ( Ter-Hovhannisyan et al . , 2008 ) ; for the H4 representative , total RNA-seq libraries obtained from three separate isolates ( BioProject PRJNA505163 ) were mapped to genomic contigs with STAR v2 . 5 . 3a ( Dobin et al . , 2013 ) under default settings; merged bam files were then used to annotate genomic contigs and derive predicted peptides with BRAKER v1 . 9 ( Hoff et al . , 2016 ) under default settings . Choanoflagellate proteome predictions ( Simion et al . , 2017 ) were provided as unpublished data from Dan Richter . Peptides from a Calvadosia ( previously Leucosolenia ) complicata transcriptome assembly were downloaded from compagen . org . Peptide predictions from Nemertoderma westbladi and Xenoturbella bocki as used in Cannon et al 2016 ( Cannon et al . , 2016 ) were provided directly by the authors . The transcriptome assembly ( raw reads unpublished ) from Euplectella aspergillum was provided by the Satoh group , downloaded from ( http://marinegenomics . oist . jp/kairou/viewer/info ? project_id=62 ) . Predicted peptides were derived from Trinity RNA-seq assemblies ( multiple versions released 2012–2016 ) as described by Laumer et al ( Laumer et al . , 2015 ) . for the following sources/SRA accessions:: Porifera: Petrosia ficiformis: SRR504688 , Cliona varians: SRR1391011 , Crella elegans: SRR648558 , Corticium candelabrum: SRR504694-SRR499820-SRR499817 , Spongilla lacustris: SRR1168575 , Clathrina coriacea: SRR3417192 , Sycon coactum: SRR504689-SRR504690 , Sycon ciliatum: ERR466762 , Ircinia fasciculata , Chondrilla caribensis ( originally misidentified as Chondrilla nucula ) and Pseudospongosorites suberitoides from ( https://dataverse . harvard . edu/dataverse/spotranscriptomes ) ; Cnidaria: Abylopsis tetragona: SRR871525 , Stomolophus meleagris: SRR1168418 , Craspedacusta sowerbyi: SRR923472 , Gorgonia ventalina: SRR935083; Ctenophora: Vallicula multiformis: SRR786489 , Pleurobrachia bachei: SRR777663 , Beroe abyssicola: SRR777787; Bilateria: Limnognathia maerski: SRR2131287 . All other peptide predictions were derived through transcriptome assembly as paired-end , unstranded libraries with Trinity v2 . 4 . 0 ( Haas et al . , 2013 ) , running with the –trimmomatic flag enabled ( and all other parameters as default ) , with peptide extraction from assembled transcripts using TransDecoder v4 . 0 . 1 with default settings . For these species , no ad hoc isoform selection was performed: any redundant isoforms were removed during tree pruning in the orthologue determination pipeline ( see below ) . Predicted proteomes were grouped into top-level orthogroups with OrthoFinder v1 . 0 . 6 ( Emms and Kelly , 2015 ) , run as a 200-threaded job , directed to stop after orthogroup assignment , and print grouped , unaligned sequences as FASTA files with the ‘-os’ flag . A custom python script ( ‘renamer . py’ ) was used to rename all headers in each orthogroup FASTA file in the convention [taxon abbreviation] + ‘@’ + [sequence number as assigned by OrthoFinder SequenceIDs . txt file] , and to select only those orthogroups with membership comprising at least one of all five major metazoan clades plus outgroups , of which exactly 4300 of an initial 46 , 895 were retained . Scripts in the Phylogenomic Dataset Construction pipeline ( Yang and Smith , 2014 ) were used for successive data grooming stages as follows: Gene trees for top-level orthogroups were derived by calling the fasta_to_tree . py script as a job array , without bootstrap replicates; six very large orthogroups did not finish this process . In the same directory , the trim_tips . py , mask_tips_by_taxonID_transcripts . py , and cut_long_internal_branches . py scripts were called in succession , with ‘ . / . tre 10 10’ , ‘ . / . /y’ , and ‘ . / . mm 1 20 . /’ passed as arguments , respectively . The 4267 subtrees generated through this process were concatenated into a single Newick file and 1419 orthologues were extracted with UPhO ( Ballesteros and Hormiga , 2016 ) . Orthologue alignment was performed using the MAFFT v7 . 271 ‘E-INS-i’ algorithm , and probabilistic masking scores were assigned with ZORRO ( Wu et al . , 2012 ) , removing all sites in each alignment with scores below five as described previously ( Laumer et al . , 2015 ) . 31 orthologues with retained lengths less than 50 amino acids were discarded , leaving 1388 well-aligned orthologues . A full concatenation of all retained 1388 orthogroups was performed with the ‘geneStitcher . py’ script distributed with UPhO available at https://github . com/ballesterus/PhyloUtensils . However , such a matrix would be too large for tractably inferring a phylogeny under well-fitting mixture models such as CAT + GTR; therefore we used MARE v0 . 1 . 2 ( Misof et al . , 2013 ) to extract an informative subset of genes using tree-likeness scores , running with ‘-t 100’ to retain all taxa and using ‘-d 1’ as a tuning parameter on alignment length . This yielded our 430-orthologue , 73 , 547 site matrix , with a mean partition length of 202 . 24 ( s . d . 116 . 96 ) residues . As a check on the above procedure , which is agnostic to the identity of the genes assigned into orthologue groups , we also sought to construct a matrix using complete , single-copy sequences identified by the BUSCO v3 . 0 . 1 algorithm ( Simão et al . , 2015 ) , using the 303-gene eukaryote_odb9 orthologue set . BUSCO was run independently on each peptide FASTA file used as input to OrthoFinder , and a custom python script ( ‘extract . py’ ) was used to parse the full output table from each species , selecting only those entries identified as complete-length , single-copy representatives of each BUSCO orthologue , and grouping these into unix directories , facilitating downstream alignment , probabilistic masking , and concatenation , as described for the OrthoFinder matrix . This 303-gene BUSCO matrix had a total length of 94 , 444 amino acids , with 39 . 6% of sites representing gaps or missing data , with mean partition length 311 . 70 ( standard deviation 202 . 78 ) . Within the gene bins nominated by the test of compositional heterogeneity ( see below ) , matrices were constructed again by concatenating and reducing matrices with MARE , using ‘-t 100’ to retain all taxa and setting ‘-d 0 . 5’ to yield a matrix of an optimal size for inferring a phylogeny under the CAT + GTR model . This procedure gave a 349-gene matrix of 80 , 153 amino acids ( mean partition lengths 228 . 67 ± s . d . 136 . 19 , 41 . 64% gaps ) within the test-failing gene set , and a 348-gene matrix of 55 , 426 amino acids ( mean partition lengths 158 . 27 ± s . d . 79 . 06 , 38 . 92% gaps ) , within the test-passing set ( Figure 4 ) . Individual ML gene trees were constructed on all 1388 orthologues in IQ-tree v1 . 6beta , with ‘-m MFP -b 100’ passed as parameters to perform automatic model selection and 100 standard nonparametric bootstraps on each gene tree . For inference on the initial 430-gene matrix , we proceeded as follows: ML inference on the concatenated matrix ( Figure 1—figure supplement 1 ) was performed with IQ-tree v1 . 6beta , passing ‘-m C60 +LG + FO + R4 bb 1000’ as parameters to specify a profile mixture model and retain 1000 trees for ultrafast bootstrapping; the ‘-bnni’ flag was used to incorporate NNI correction during UF bootstrapping , an approach shown to control misleading inflated support arising from model misspecification ( Hoang et al . , 2018 ) . ML inference using only the H1 haplotype as a representative of Placozoa ( Figure 1—figure supplement 2 ) was undertaken similarly , albeit using a marginally less complex profile mixture model ( C20 +LG + FO + R4 ) . Bayesian inference under the CAT + GTR + Г4 model was performed in PhyloBayes MPI v1 . 6j ( Lartillot et al . , 2013 ) with 20 cores each dedicated to four separate chains , run for 2885–3222 generations with the ‘-dc’ flag applied to remove constant sites from the analysis , and using a starting tree derived from the FastTree2 program ( Price et al . , 2010 ) . The two chains used to generate the posterior consensus tree summarized in Figure 1 converged on exactly the same tree in all MCMC samples after removing the first 2000 generations as burn-in . Analysis of Dayhoff-6-state recoded matrices in CAT + GTR + Г4 was performed with the serial PhyloBayes program v4 . 1c , with ‘-dc -recode dayhoff6’ passed as flags . Six chains on the 430- gene matrix were run from 1441 to 1995 generations; two chains showed a maximum bipartition discrepancy ( maxdiff ) of 0 . 042 after removing the first 1000 generations as burn-in ( Figure 2 ) . QuartetScores ( Zhou , 2017 ) was used to measure internode certainty metrics including the reported EQP-IC , using the 430 gene trees from those orthologues used to derive the matrix as evaluation trees , and using the amino-acid CAT + GTR + Г4 tree as the reference to be annotated ( Figure 1 ) . For inference on the BUSCO 303 gene set , we ran 4 chains of the CAT + GTR + Г4 mixture model with PhyloBayes MPI v1 . 7a , applying the -dc flag again to remove constant sites , but here not specifying a starting tree; chains were run from 1873 to 2361 generations . Unfortunately , no pair of chains reached strict convergence on the amino-acid version of this matrix ( with all pairs showing a maxdiff = 1 at every burn-in proportion examined ) , perhaps indicating problems mixing among the four chains we ran . However , all chains showed full posterior support for identical relationships among the five major animal groups , with differences among chains assignable to minor differences in the internal relationships within Choanoflagellata and Bilateria . Accordingly , the posterior consensus tree in Figure 3A is summarized from all four chains , with a burn-in of 1000 generations , sampling every 10 generations . For the Dayhoff-recoded version of this matrix , we ran six separate chains again with CAT + GTR + Г4 with the -dc flag , for 5433 – 6010 generations; two chains were judged to have converged , giving a maxdiff of 0 . 141157 during posterior consensus summary with a burn-in of 2500 , sampling every 10 generations ( Figure 3B ) . For inference on the 348 and 349 gene matrices produced within gene bins defined by the null-simulation test of compositional bias ( see below ) , we ran six chains each for the amino acid and recoded versions of each matrix , under CAT + GTR + Г4 with constant sites removed . In the amino-acid matrix , chains ran from 2709 to 3457 and 1423 – 1475 generations for the test-failing and test-passing matrices , respectively . In the recoded matrix , chains ran from 3893 to 4480 and 4350 – 4812 generations for the test-failing and test-passing matrices , respectively . In selecting chains to input for posterior consensus summary tree presentation ( Figure 4A–D ) , we chose pairs of chains and burn-ins that yielded the lowest possible maxdiff values ( all <0 . 1 with the first 500 generations discarded as burn-in , except for the amino-acid coded test-failing matrix , whose most similar pair of chains gave a maxdiff of 0 . 202 with 1000 generations discarded as burn-in ) . We emphasize that the topologies and supports displayed in Figure 4A–D are similar when all chains ( and conservative burn-in values ) are used to generate consensus trees . For ML trees using profile mixture models for the test-failing ( Figure 4—figure supplement 1 ) and test-passing ( Figure 4—figure supplement 2 ) gene matrices , we used IQ-tree 1 . 6rc , calling in the same manner ( with C60 +LG + FO + R4 ) as used on our 430-gene matrix ( see above ) . For posterior predictive tests of compositional heterogeneity and residue diversity using MCMC samples under CAT + GTR ( Table 2 ) , we used PhyloBayes MPI v1 . 8 to test two chains from the initial 430-gene matrix , three chains from the 303-gene BUSCO matrix , and six chains each from the 348 ( test-failing ) and 349 ( test-passing ) gene matrices , removing 2000 generations from the first matrix and 1000 from the others as burn-in . Results from tests on representative chains were selected for plotting in Figure 3C and summary in Table 2; however , results from all chains tested are deposited in the Data Dryad accession . For the per-gene null simulation tests of compositional bias ( Foster , 2004 ) , we used the p4 package ( https://github . com/pgfoster/p4-phylogenetics ) , inputting the ML trees inferred by IQ-tree for each of the 1388 alignments , and assuming an LG+Γ4 substitution model with a single empirical frequency vector for each gene; this test was implemented with a simple wrapper script ( ‘p4_compo_test_multiproc . py’ ) leveraging the python multiprocessing module . We opted not to model-test each gene individually in p4 , both because the range of models implemented in p4 are more limited than those tested for in IQ-tree , and because , as a practical matter , LG ( usually with variant of the FreeRates model of rate heterogeneity ) was chosen as the best-fitting model in the IQ-tree model tests for a large majority of genes , suggesting that LG+Γ4 would be a reasonable approximation for the purposes of this test . We selected an α-threshold of 0 . 10 for dividing genes into test-passing and -failing bins as a conservative measure; however , we emphasize that even at a less conservative α = 0 . 05 , 47% of genes would still be detected as falling outside the null expectation . SRA accession codes , where used , and all alternative sources for sequence data ( e . g . individually hosted websites , personal communications ) , are listed above in the Materials and methods section . A DataDryad accession is available at https://doi . org/10 . 5061/dryad . 6cm1166 , which makes available all helper scripts , orthogroups , multiple sequence alignments , phylogenetic program output , and raw host proteomes inputted to OrthoFinder . Metagenomic bins containing placozoan host contigs and gene annotations from H4 , H6 and H11 isolates are also provided in this accession . PhyloBayes . chain files , due to their large size , are separately accessioned at in Zenodo at https://doi . org/10 . 5281/zenodo . 1197272 . | Filter-feeding sponges and tiny gliding , pancake-like animals called placozoans are the only two major groups of animals that lack muscles , nerves and an internal gut . Sponges have historically been seen as the first to have branched off in animal phylogeny – the family tree of living organisms that shows how species are related . This is because it is assumed that they split from the other animals before features including muscles , nerves and internal guts evolved . Sequences of their genetic material ( the genome ) support this view , although some argue that jellyfish-like animals called ctenophores branched first . One explanation for this disagreement is that ctenophores use different proportions of amino acids in their proteins , known as compositional heterogeneity . Computer algorithms that assume amino acid usage is the same universally throughout evolution may therefore place ctenophores incorrectly . In contrast , so far the only genome from a placozoan shows that they are equally closely related to jellyfish and corals ( cnidarians ) and bilaterians , which includes worms , insects and vertebrates . To test whether this view of the first branches of the animal tree of life is correct , Laumer et al . included the genomes from several undescribed species of placozoans in a phylogenetic analysis . These analyses showed a relationship that had not previously been seen . The placozoans were the closest living relative to cnidarians . However , when looking at the level of genes rather than whole genomes , the more usual relationship of placozoans being equally related to cnidarians and bilaterians re-emerged . To resolve this conflict , Laumer et al . focused on the genes that had the least compositional heterogeneity . When doing this , the relationship appeared to be the newly identified one of placozoans being most closely related to cnidarians . Researchers studying cnidarians often hope to find some clues as to how the complex features they seem to share with bilaterians originated . The findings of Laumer et al . may suggest that the ancestors of the placozoans did in fact have muscles , nerves and guts , but they lost these traits in favor of a simpler lifestyle . An alternative , but controversial possibility is that the ancestor of cnidarians and bilaterians was a simple organism like a placozoan , and the two evolved their complex traits independently . The findings show a complex picture of early animal evolution . Further study of placozoans may well clarify this picture . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"short",
"report"
] | 2018 | Support for a clade of Placozoa and Cnidaria in genes with minimal compositional bias |
The sense of smell in vertebrates is detected by specialized sensory neurons derived from the peripheral nervous system . Classically , it has been presumed that the olfactory placode forms all olfactory sensory neurons . In contrast , we show that the cranial neural crest is the primary source of microvillous sensory neurons within the olfactory epithelium of zebrafish embryos . Using photoconversion-based fate mapping and live cell tracking coupled with laser ablation , we followed neural crest precursors as they migrated from the neural tube to the nasal cavity . A subset that coexpressed Sox10 protein and a neurogenin1 reporter ingressed into the olfactory epithelium and differentiated into microvillous sensory neurons . Timed loss-of-function analysis revealed a critical role for Sox10 in microvillous neurogenesis . Taken together , these findings directly demonstrate a heretofore unknown contribution of the cranial neural crest to olfactory sensory neurons in zebrafish and provide important insights into the assembly of the nascent olfactory system .
The last few decades have witnessed a surge of new information regarding the mature olfactory system , ranging from extensive and elegant descriptions of molecular and cellular functionality to physiological data on odorants , receptors , and transduction pathways . In contrast , far less is known about initial development of the olfactory epithelium during embryogenesis . This knowledge is critical for understanding not only disorders such as anosmia—the loss of smell—but also for clarifying the remarkable ability of the olfactory epithelium to regenerate neurons in adult organisms ( Graziadei and Graziadei , 1979; Bermingham-McDonogh and Reh , 2011 ) . In addition to providing critical clues for comprehending de novo differentiation of nascent stem cells into olfactory neurons , studies of olfactogenesis may offer mechanistic insights into neurogenesis as a whole . The peripheral nervous system of the vertebrate head is derived from two sources: cranial ectodermal placodes and neural crest cells . Whereas placodes form the sense organs ( nose , ears , and lens ) of the head and contribute to neuronal portions of cranial ganglia , the glia are solely neural crest derived ( Northcutt and Gans , 1983; Baker and Bronner-Fraser , 2001; Barraud et al . , 2010 ) . Sensory neurons are the primary functional players of the olfactory epithelium and have long been thought to be similar to neurons of the ear or eye . The latter arise within the developing sensory structures , the otic placode and neural retina , respectively . By analogy , the olfactory epithelium arises from the olfactory placode , a swathe of thickened ectoderm that initially resides within the anteriormost portion of the neural tube and subsequently moves laterally within the plane of the ectoderm . These cells invaginate to form the olfactory vesicle , which in turn becomes encapsulated by neural crest cells that form the surrounding nasal capsule ( Osumi-Yamashita et al . , 1994; Le Douarin and Kalcheim , 1999 ) . In the zebrafish embryo , the olfactory placode first becomes apparent around 18 hr post-fertilization ( hpf ) ( Hansen and Zeiske , 1993; Miyasaka et al . , 2007 ) and subsequently gives rise to sensory neurons and supporting cell populations . Within the olfactory epithelium , there are two predominant types of olfactory sensory neurons: ciliated and microvillous neurons . These differ in roles , receptors , location , and patterns of innervation . Whereas more basally located ciliated sensory neurons detect volatile odorants , more apically located microvillous sensory neurons can detect pheromones , nucleotides , and/or amino acids , depending on the species ( Buck , 2000; Sato and Suzuki , 2001; Hansen and Zielinski , 2005 ) . The possible role of the neural crest in development of olfactory-related derivatives has been the subject of vigorous debate . Recent literature demonstrates that olfactory ensheathing glia , nonneuronal supporting cells previously assumed to derive from the olfactory epithelium , are actually neural crest derived ( Barraud et al . , 2010 ) . These findings are consistent with the neural crest origin of all other peripheral glia ( Woodhoo and Sommer , 2008 ) . In neuronal lineages , there has been considerable controversy over the cell type of origin . As one example , some studies ( Whitlock et al . , 2003; Forni et al . , 2011 ) have suggested that neural crest cells give rise to GnRH cells of the terminal nerve , which is close to though not part of the olfactory nerve and epithelium . However , many other studies refute this notion and support a placodal origin for GnRH cells ( Schwanzel-Fukuda and Pfaff , 1989; Wray et al . , 1989; Dubois et al . , 2002; Palevitch et al . , 2007; Metz and Wray , 2010; Sabado et al . , 2012 ) . Within the olfactory epithelium itself , Harden et al . ( 2012 ) recently examined the interactions of Sox10:eGFP-positive cells with six4b:mCherry-expressing placodal precursors in zebrafish . The authors interpreted their negative results to indicate a lack of contribution of neural crest cells to the olfactory epithelium . However , using the same Sox10:eGFP line , here we definitively demonstrate the neural crest origin of microvillous neurons within the olfactory epithelium . To address the role of the neural crest in the development of olfactory sensory neurons , we have employed a combination of live confocal imaging , photoconvertible fate mapping , and laser ablation using zebrafish as a model . Interestingly , we find that neural crest gives rise to microvillous , but not ciliated , sensory neurons in the developing olfactory epithelium . Specifically , we demonstrate that a neurogenin1 ( Ngn1 ) reporter and Sox10 protein are expressed in the subset of neural crest-derived nasal cavity cells that undergo dynamic ingression into the olfactory epithelium to form microvillous neurons . If the neural crest is ablated , the olfactory epithelium cannot compensate for the loss of microvillous neurons . Finally , we show that the neural crest specifier gene Sox10 is required for microvillous neurogenesis as these cells are ingressing into the olfactory epithelium . These results definitively demonstrate for the first time a Sox10-dependent neural crest origin for microvillous sensory neurons that take residence within the placode-derived olfactory epithelium .
To analyze the contribution of neural crest cells to the forming olfactory organ , we examined fluorescent neural crest cells in live transgenic zebrafish embryos at high resolution via confocal microscopy using a Sox10:eGFP line ( Wada et al . , 2005; Carney et al . , 2006; see ‘Materials and methods' for full annotation of all used lines ) . Visualization and analysis with Imaris software ( Bitplane , Zurich , Switzerland ) demonstrated that cranial neural crest cells moved anteriorly from the dorsal neural tube toward the future location of the olfactory pit . Some neural crest cells migrated and coalesced as a cup-like structure around the olfactory vesicle ( Figure 1A , B ) to form the mesenchymal nasal cavity , consistent with descriptions in mouse , chick , and medaka ( Langille and Hall , 1988; Osumi-Yamashita et al . , 1994; Le Douarin and Kalcheim , 1999 ) . 10 . 7554/eLife . 00336 . 003Figure 1 . Sox10:eGFP+/Ngn1:nRFP+ cells migrate and differentiate into microvillous neurons . ( A ) Sox10:eGFP+ neural crest migrates dorsal to the eye and toward the olfactory region at 19 hpf as the olfactory placode is first apparent . ( B ) – ( D ) Time-lapse confocal microscopy of live embryos ( Sox10:eGFP+; Ngn1:nRFP+ ) demonstrates that a subset of nasal cavity cells that express Ngn1:nRFP ( B , boxed area arrowheads; z = 3 . 5 μm ) ingress into the olfactory epithelium . ( E ) – ( J ) Selected time point and z-plane excerpts from full z stacks are shown from ∼29 hpf ( C ) to ∼33 hpf ( D ) and are 40' apart; z = 17 . 5 μm ( consisting of five 3 . 5 μm slices ) . Arrowheads indicate two ingressing cells , one from the top and the other directly behind the olfactory epithelium . All ingressing cells express Ngn1:nRFP ( F , inset arrowhead ) . Sox10:eGFP , green; Ngn1:nRFP , red . ( K ) At 53 hpf , all Sox10:eGFP+ cells in the olfactory epithelium are TRPC2:Venus+ microvillous neurons . ( because eGFP signal bleeds into the Venus channel , overlap was confirmed via image processing; see Figure 1—figure supplement 1 ) . Only a small number of microvillous neurons are not Sox10:eGFP+ ( arrowheads ) . A single 3 . 5-μm-thick z-plane slice of the boxed area more clearly shows colocalization . Sox10:eGFP: green; TRPC2:Venus: red; Ngn1:nRFP: blue . e: eye; y: yolk; LOE: left olfactory epithelium; ROE: right olfactory epithelium . Orientation arrows: A: anterior; P: posterior; D: dorsal; V: ventral; L: lateral . Scale bars: 50 μm ( A ) ; 30 μm ( B–D ) ; 20 μm ( E–K ) . See also Figure 1—figure supplements 1 and 2 and Videos 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 00336 . 00310 . 7554/eLife . 00336 . 004Figure 1—figure supplement 1 . Panel ( K ) from Figure 1 with Sox10:eGFP ( green ) channel removed ( bottom ) to better illustrate the large population of ciliated neurons present basally that are not Sox10:eGFP+ but are Ngn1:nRFP+ ( blue ) . High magnification inset ( right ) has Sox10:eGFP signal artificially underexposed and decreased to better view the colocalization with membrane TRPC2:Venus expression . DOI: http://dx . doi . org/10 . 7554/eLife . 00336 . 00410 . 7554/eLife . 00336 . 005Figure 1—figure supplement 2 . ( A ) and ( A' ) Sox10:eGFP+ microvillous neurons in fixed embryos stained with anti-GFP antibody colocalize with anti-HuC/D antibody staining at 53 hpf , confirming post-mitotic neuronal identity . Sox10:eGFP: green; HuC/D: red; nuclear stain: blue . Orientation arrows: D: dorsal; V:ventral; L: lateral . z = 2 . 5 μm . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00336 . 005 Subsequently , beginning at ∼24 hpf and continuing until at least day 2 , a subset of the Sox10:eGFP-positive cells in the nasal cavity delaminated and intercalated into the olfactory epithelium where placode-derived ciliated neurons and support cells were already present . Concomitantly , these cells increased their eGFP levels and changed their morphology from mesenchymal to the ‘teardrop' shape characteristic of neurons ( Figure 1C–J and Videos 1 and 2 ) . 10 . 7554/eLife . 00336 . 006Video 1 . Sox10:eGFP , Ngn1:nRFP from ∼26 . 5 hpf to ∼33 hpf , 20' intervals . Over time , multiple Sox10:eGFP+ , Ngn1:nRFP+ cells ( bright green-yellow ) appear in the apical portion of the olfactory epithelium and adopt neuronal morphology . See also Figure 1 . Videos assembled and registered using Imaris software . DOI: http://dx . doi . org/10 . 7554/eLife . 00336 . 00610 . 7554/eLife . 00336 . 007Video 2 . Higher magnification view of the right olfactory epithelium , Sox10:eGFP channel only , taken from Video 1 . Individual cell movements can now be clearly observed . Several Sox10:eGFP+ cells ingress from the nasal cavity into the olfactory epithelium and take on the teardrop shape characteristic of neurons while extending axonal projections toward the olfactory bulb . See also Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00336 . 007 To characterize the nature of the ingressing neurons , we crossed the Sox10:eGFP line with OMP:RFP and TRPC2:Venus lines , which bear markers of ciliated and microvillous neurons , respectively ( Sato et al . , 2005 ) . The results show that all of the Sox10:eGFP-positive cells within the olfactory epithelium are microvillous neurons , as revealed by their TRPC2:Venus expression and apical position ( Figure 1K and Figure 1—figure supplement 1 ) . In contrast , none are OMP:RFP-positive ciliated neurons ( data not shown ) . A small number of microvillous neurons are Sox10:eGFP-negative ( Figure 1K and Figure 1—figure Supplement 1 , arrowheads ) , raising the possibility that a minor subset of these cells has a separate origin . To further confirm that the ingressed cells are neurons , we performed immunostaining with an antibody against HuC/D , a marker for newly born post-mitotic neurons ( Kim et al . , 1996 ) , which colocalized in the apically located Sox10:eGFP-positive neurons ( Figure 1—figure supplement 2 ) . All neurogenic placodes in zebrafish , including the olfactory placode , express the proneural bHLH transcription factor Ngn1 , which is known to be required for sensory neuron differentiation in mice ( Andermann et al . , 2002; Bertrand et al . , 2002; Cau et al . , 2002 ) . In addition , Ngn1 is expressed in neural crest cells that become sensory neurons in the dorsal root ganglia in zebrafish ( McGraw et al . , 2008 ) . Using a Ngn1:nRFP transgenic line ( Blader et al . , 2003 ) that has been well-characterized as representative of endogenous Ngn1 expression ( Blader et al . , 2003 , 2004; Madelaine et al . , 2011 ) , we noted Ngn1:nRFP expression in a subset of Sox10:eGFP-positive cells in the nasal cavity ( Figure 1B ) . Interestingly , only cells that upregulate Sox10:eGFP and coexpress Ngn1:nRFP ingress into the olfactory epithelium to differentiate into microvillous neurons ( Figure 1F ) . Although Sox10-driven lines mark neural crest cells , Sox10 is not entirely neural crest specific and becomes downregulated with time in some lineages , whereas highly stable eGFP often persists . To substantiate our real-time confocal imaging , we followed the migration and differentiation of small groups of neural crest cells by making use of the PhOTO-N transgenic zebrafish line ( Dempsey et al . , 2012 ) expressing a nuclear localized version of the highly stable photoconvertible ( green to red ) protein Dendra2 ( Gurskaya et al . , 2006 ) . By crossing this line to Sox10:eGFP fish ( Figure 2A ) , we were able to photoconvert small numbers of neural crest cells ( Figure 2B , B' ) , ensuring specificity of labeling . We then followed their movement from the neural rod to the nasal cavity . For some experiments , large blocks of cells ( Figure 2C , C' ) were photoconverted to achieve broad coverage and identify all possible neural crest contributions . In both cases , all photoconversion was performed between 14 hpf ( 10-somite stage ) and 16 hpf ( 14-somite stage ) , always unilaterally . 10 . 7554/eLife . 00336 . 008Figure 2 . Lineage tracing by photoconversion demonstrates neural crest origin of microvillous neurons . ( A ) Nuclear Dendra2 ( low level green ) was photoconverted in Sox10:eGFP+ ( bright green ) neural crest cells at 14–16 hpf ( 10–14 somite stage ) in live embryos . ( B ) – ( C' ) Representative example z-planes show unilateral photoconversion of a few cells ( B and B' ) or a large number of cells ( C and C' ) . ( B and C ) show ubiquitous Dendra2 ( low level green ) and Sox10:eGFP ( bright green ) , and ( B' and C' ) show photoconverted Dendra2 ( red ) . ( D ) Photoconverted cells are visible in the contralateral nasal cavity at 29 hpf; ( E and E' ) show a 2 . 5-μm-thick z-plane slice of the boxed area with photoconverted cells ( red , arrowheads ) . ( F and F' ) At 55 hpf , photoconversion of a small number of cells ( B and B' ) results in a subset of Sox10:eGFP+ microvillous neurons being labeled by photoconverted Dendra2 on the contralateral side . ( G and G' ) Large-scale photoconversion ( C and C' ) labels most Sox10:eGFP+ microvillous neurons . Directly adjacent ciliated neurons are never labeled . ( F–G' ) z = 2 . 5 μm; Sox10:eGFP: green; Dendra2Green: blue; Dendra2Red: red; histology in brightfield . Orientation arrows: A: anterior; P: posterior; D: dorsal; V: ventral; L: lateral . Scale bars: 50 μm ( A ) ; 20 μm ( B–D ) ; 10 μm ( F–G' ) . See also Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00336 . 00810 . 7554/eLife . 00336 . 009Figure 2—figure supplement 1 . Shown are a few 55 hpf microvillous neurons that are Sox10:eGFP negative ( arrowheads ) and were not photoconverted , suggesting a possible non-neural crest origin . DOI: http://dx . doi . org/10 . 7554/eLife . 00336 . 009 Prior to day 1 of development , no photoconverted cells were observed in the olfactory epithelium ( data not shown ) , consistent with our previous findings from the Sox10 reporter line on the timing of microvillous ingression ( see Figure 1 ) . At 29 hpf , a few photoconverted cells were present in the nasal cavity ( Figure 2D–E' ) . At 55 hpf , by which time ingression/differentiation was nearly complete , photoconverted neural crest cells had differentiated into apically localized microvillous neurons . These results are consistent across embryos having either ‘small' or ‘large' numbers of photoconverted cells ( Figure 2F–G' ) . In contrast , no photoconverted cells differentiate into the basally located ciliated neurons , consistent with their placodal origin and confirming the specificity of our labeling . Interestingly , many labeled neural crest cells cross the midline and migrate to the contralateral side ( shown in Figure 2 ) where they differentiate into microvillous neurons . The small subset of microvillous neurons that are not Sox10:eGFP-positive show no overlap with the photoconverted Dendra2Red ( Figure 2—figure supplement 1 ) . Since our findings suggest that neural crest cells contribute the vast majority of microvillous neurons to the olfactory epithelium , we asked whether elimination of specific subpopulations of these cells would alter development within the olfactory epithelium and whether the olfactory epithelium could compensate for the loss of this source population . Two-photon laser ablation was performed at 25 hpf unilaterally on small ( 2–7 cells , not shown ) or large ( ≥20 cells ) ( Figure 3A , B ) groups of neural crest-derived nasal cavity cells that were Sox10:eGFP positive and either Ngn1:nRFP negative or positive . These manipulations left cells in the olfactory epithelium untouched . By 35 hpf , both the small ( Figure 3C ) and large ( Figure 3D ) ablations of Ngn1:nRFP-positive cells in the right-side nasal cavity had highly deleterious effects on the subsequent number of microvillous neurons . By comparison , the unablated left-side nasal cavity and olfactory epithelium appeared normal . Large ablations often produced a striking phenotype with the near-complete absence of microvillous neurogenesis post-ablation ( Figure 3D ) , whereas placode-derived ciliated neurons ( Sox10:eGFP negative and Ngn1:nRFP positive in olfactory epithelium ) continued to form . Ablations of Ngn1:nRFP-negative nasal cavity cells had a minor phenotype ( data not shown ) , which may be due to the fact that some ablated cells were destined to turn on Ngn1 but were not yet expressing Ngn1:nRFP at detectable levels . 10 . 7554/eLife . 00336 . 010Figure 3 . Laser ablation of neural crest in the nasal cavity inhibits microvillous neurogenesis . ( A ) Immediately before and after ablation of Sox10:eGFP+/Ngn1:nRFP+ cells in the right nasal cavity in live embryos; example z-plane slice shows six regions of ablation ( asterisks ) . ( B ) 3D z-stack post-ablation of several z-planes with total of ≥20 Sox10:eGFP+/Ngn1:nRFP+ ablated cells . ( C ) and ( D ) 35 hpf embryos show a significant decrease in the number of Sox10:eGFP+/TRPC2:Venus+ microvillous neurons within the right olfactory epithelium as compared to the unablated left side in both small ( C ) and large ( D ) ablation experiments . Directly adjacent ciliated neurons are only marginally affected . Sox10:eGFP: green; Ngn1:nRFP: red; TRPC2:Venus: blue . LOE: left olfactory epithelium; ROE: right olfactory epithelium . Orientation arrows: A: anterior; P: posterior; D: dorsal; V: ventral; L: lateral . Scale bars: 30 μm ( A and B ) ; 40 μm ( C and D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00336 . 010 Taken together , these results reveal an essential role for a specific subset of neural crest–derived cells in the nasal cavity that upregulate Ngn1 shortly before ingressing into the olfactory epithelium . Interestingly , the placode-derived olfactory epithelium is unable to compensate for the loss of the neural crest–derived population of microvillous neurons . The Sox10:eGFP line exhibits a striking upregulation of eGFP expression in cells concomitant with their intercalation into the olfactory epithelium ( see Figure 1 ) . To examine if the increased eGFP reflects an upregulation of endogenous Sox10 protein , we performed immunostaining with a Sox10 antibody ( Park et al . , 2005 ) . These experiments reveal a marked increase in Sox10 protein levels in differentiating neural crest–derived microvillous neurons as they move from the nasal cavity ( arrowhead ) into the olfactory epithelium and transition from mesenchymal to the teardrop morphology characteristic of neurons at both 26 and 34 hpf ( Figure 4A–B' ) . Directly adjacent ciliated neurons ( unstained ) do not upregulate Sox10 . 10 . 7554/eLife . 00336 . 011Figure 4 . Sox10 protein is preferentially upregulated in neural crest cells that differentiate into microvillous neurons . ( A ) – ( B' ) Sox10:eGFP+ differentiating microvillous neurons in fixed embryos stained with anti-GFP antibody colocalize with anti-Sox10 antibody staining at 26 hpf ( A and A' ) and 34 hpf ( B and B' ) . Levels of eGFP and Sox10 mirror each other , both increasing as cells ingress and become neurons . Arrowhead in ( B and B' ) indicates a nasal cavity cell likely about to ingress that has increased Sox10 protein expression . Sox10:eGFP: green; Sox10 protein: red; nuclear stain: blue; LOE: left olfactory epithelium . Orientation arrows: D: dorsal; V: ventral; L: lateral . z = 2 . 5 μm . Scale bars: 15 μm ( A–B ) ; 30 μm ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00336 . 011 The rapid upregulation of Sox10 levels during microvillous neurogenesis raises the intriguing possibility that Sox10 protein may play a vital role in this process . To test this hypothesis , we performed loss-of-function analysis with a morpholino that has been well characterized as specifically blocking Sox10 protein production ( Dutton et al . , 2001a , 2001b; Prasad et al . , 2011; Whitlock et al . , 2005 ) . After morpholino injection , olfactory development was followed via real-time confocal imaging of transgenic lines . The results show that loss of Sox10 leads to marked depletion of microvillous neurons in the olfactory epithelium ( Figure 5B , C' , E' , F' , H ) in comparison to control morpholino-treated embryos ( Figure 5A' , D' , G ) . Neighboring ciliated neurons , in contrast , develop with only mild disruption ( Figure 5B , C , E , F ) , suggesting that downregulation of Sox10 protein specifically affects formation of microvillous neurons . This phenotype was apparent as early as 29 hpf ( Figure 5A–C' ) and remained consistent at 33 hpf , 55 hpf , and beyond ( Figure 5D–H and Figure 5–figure supplement 1 ) . By 55 hpf , ciliated neurons were present in robust numbers comparable to those in the control embryos , while microvillous neurons remained scarce and those that did form appeared disorganized . Furthermore , the few remaining microvillous neurons were often TRPC2:Venus positive but Sox10:eGFP negative ( Figure 5—figure supplement 1 ) , suggesting that they were likely the small proportion of microvillous neurons not derived from the neural crest . 10 . 7554/eLife . 00336 . 012Figure 5 . Morpholino knockdown of Sox10 selectively inhibits microvillous neurogenesis . ( A ) and ( A' ) Control morpholino-injected live embryos at 29 hpf have ciliated ( red ) and microvillous ( green ) neurons . ( B ) – ( C' ) Embryos with mild and moderate phenotypes ( assayed histologically ) display only slight changes in ciliated neurons , whereas microvillous neurons are significantly decreased in number and organization . ( D ) – ( F' ) Similar results are seen at 33 hpf , with almost no Sox10:eGFP+ ( green ) cells becoming microvillous ( blue ) neurons . Ciliated ( red ) neurons remain relatively unaffected . ( G ) and ( H ) Antibody staining with anti-GFP against TRPC2:Venus in fixed embryos demonstrates a persistent decrease in the number of microvillous cells and disorganization at 60 hpf in Sox10 morpholino-treated embryos ( H ) in comparison to control embryos ( G ) . All images were captured at identical settings to facilitate direct comparison of control and experimental embryos . ( I ) – ( K ) All embryos ( I ) were divided into ‘high pigmentation' ( J ) or ‘low pigmentation' ( K ) to roughly correlate Sox10 levels ( higher and lower , respectively ) with degree of olfactory phenotype . As expected , all control morpholino-treated embryos have high pigmentation , and all Sox10 morpholino-treated embryos with high pigmentation have only a mild phenotype . In all cases ( J and K ) , Sox10 morpholino treatment results in most embryos having either no microvillous neurons or a decreased number . Ratios correlate with degrees of pigmentation and phenotype . ( A–C' , G , and H ) TRPC2:Venus: green; OMP:RFP: red . ( D–F' ) Sox10:eGFP: green; OMP:RFP: red; TRPC2:Venus: blue . Orientation arrows: D: dorsal; V: ventral; L: lateral . Scale bars: 50 μm . *p<0 . 001 . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00336 . 01210 . 7554/eLife . 00336 . 013Figure 5—figure supplement 1 . At 33 hpf , severe phenotype embryos ( top right ) have major malformations in both microvillous and ciliated neurons and often perish , likely reflecting non-specific effects . At 55 hpf ( bottom ) are shown the same embryos as in Figure 5 ( D–F' ) . At 55 hpf ( bottom ) are shown the same embryos as in Figure 5 ( D–F' ) . While there is some recovery of microvillous neurons in ‘mild' embryos , their numbers remain decreased in comparison to controls . ‘Moderate' embryos have near complete lack of neural crest-derived microvillous neurons , whereas ciliated neurons are present and project to the olfactory bulb . DOI: http://dx . doi . org/10 . 7554/eLife . 00336 . 013 Since Sox10 is necessary for differentiation of many neural crest lineages , including the melanocyte lineage , its loss leads to a known ‘colorless' phenotype ( Dutton et al . , 2001a , 2001b ) . We were able to use this decrease in pigmentation as a surrogate marker for the degree of Sox10 knockdown . All assayed embryos ( Figure 5I ) were scored as ‘high pigmentation' ( Figure 5J ) or ‘low pigmentation' ( Figure 5K ) and then further separated into ‘mild' , ‘moderate' , or ‘severe' morphological phenotypes . Surprisingly , microvillous neuron reduction is a consistent and common feature across all categories after knockdown of Sox10 . These data suggest that microvillous neuron formation is exquisitely sensitive to reductions in Sox10 protein levels , whereas other Sox10-dependent processes such as pigmentation may be more resilient . The transcription factor Sox10 plays a variety of roles in neural crest development that differ in time and cell type . To determine if Sox10 is required specifically during microvillous neurogenesis , as opposed to only during its well-established earlier role in neural crest specification/migration , we took advantage of a new photocleavable morpholino system ( Gene Tools , LLC; Tallafuss et al . , 2012 ) . First , using a second translation-blocking antisense morpholino targeting Sox10 ( previously used in Dutton et al . , 2001a ) , we confirmed our previous results that Sox10 was necessary for microvillous neuron formation ( data not shown ) . We then utilized a sense photo-morpholino specifically targeted to the second antisense morpholino and containing a photocleavable moiety . Coinjection of the two oligonucleotides sequestered the antisense Sox10 morpholino , rendering it inactive until illumination at 365 nm resulted in cleavage of the photo-morpholino and release of the antisense morpholino . This process allowed for the inhibition of Sox10 protein production from a specific time point onward and had the advantage of leaving Sox10-related processes such as early neural crest development intact . In addition , this experimental setup allowed for the injection of a single oligonucleotide cocktail into both control ( not photocleaved ) and experimental ( photocleaved ) embryos , reducing the variability inherent in morpholino-based experiments . Injected embryos were subjected to photocleavage at the time of olfactory placode formation ( ∼17 . 5 hpf ) or at the onset of ingression into the olfactory epithelium ( ∼24 hpf ) , with a subset of controls left unilluminated . In addition , some uninjected embryos were illuminated to control for any effects of UV light ( data not shown ) . While control embryos developed normal numbers and organization of olfactory sensory neurons at 38 hpf ( Figure 6A , A' ) and 54 hpf ( Figure 6—figure supplement 1 ) , knockdown of Sox10 expression at 24 hpf specifically impeded microvillous neuron formation and organization ( Figure 6B , B' and Figure 6—figure supplement 1 ) . Knockdown at the earlier 17 . 5 hpf time point had an even more pronounced effect ( data not shown ) , since it allowed more time for degradation of existing Sox10 protein before the start of ingression . Importantly , the later 24 hpf time point of photocleavage occurs well after neural crest cells have populated the nasal cavity , and results from that experiment decouple microvillous neurogenesis from earlier , critical roles of Sox10 in neural crest development and migration . Taken together , these results demonstrate that Sox10 is required for ingression/differentiation from the nasal cavity into the olfactory epithelium to generate microvillous neurons . 10 . 7554/eLife . 00336 . 014Figure 6 . Photo-morpholino knockdown of Sox10 demonstrates its necessity during the ingression/differentiation process . ( A ) and ( A' ) Control ( antisense morpholino + sense photo-morpholino injected but not photocleaved ) live embryos have robust numbers of ciliated ( red ) and neural crest–derived microvillous ( green ) neurons at 38 hpf . In contrast , identically injected embryos subjected to photocleavage at 24 hpf go on to develop ciliated neurons but significantly lack microvillous neurons at 38 hpf ( B and B' ) in comparison to control embryos . Sox10:eGFP: green; OMP:RFP: red . Orientation arrows: D: dorsal; V: ventral; L: lateral . Scale bars: 30 μm . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00336 . 01410 . 7554/eLife . 00336 . 015Figure 6—figure supplement 1 . Shown are more details of the phenotype in Figure 6: ( A and A' ) control ( antisense morpholino + sense photo-morpholino injected but not photocleaved ) live embryos have robust numbers of ciliated ( red ) and neural crest–derived microvillous ( green ) neurons at 38 hpf . In contrast , identically injected embryos subjected to photocleavage at 24 hpf go on to develop ciliated neurons but have slight ( B and B' ) or , much more commonly , large ( C and C' ) decreases in microvillous neuron numbers and organization in comparison to control embryos . Photocleavage may have been less efficient in some embryos , resulting in a range of phenotypic severity . ( D–F' ) similar results are seen in the same embryos at 54 hpf , now with an even greater difference in microvillous neuron numbers between control and cleaved embryos . Sox10:eGFP: green; OMP:RFP: red . Orientation arrows: D: dorsal; V: ventral; L: lateral . Scale bars: 30 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00336 . 015
The data presented here reveal a previously unrecognized neural crest origin for one of the two main types of olfactory sensory neurons , microvillous sensory neurons , and implicate Sox10 as a critical player in microvillous neuron formation in zebrafish . Our conclusions are derived from several complementary lines of experimentation ranging from live imaging of migration to fate mapping via photoconversion and laser ablation to morpholino knockdown at specific time points . The cumulative results demonstrate that neural crest cells migrate anteriorly to form the mesenchymal nasal cavity . A subset of these cells expresses Ngn1 and upregulates Sox10 to high levels during ingression into the olfactory epithelium to form microvillous neurons . Laser ablation data confirm that the neural crest contribution is necessary for microvillous neurogenesis and that the olfactory placode is unable to compensate for the loss of this precursor population . Our loss-of-function studies suggest that high levels of the transcription factor Sox10 are necessary for microvillous neurogenesis to proceed . Notably , the reduction in microvillous neurons after the loss of Sox10 is accompanied by only minor effects on the placode-derived ciliated neurons with which they intermix . Therefore , there is no general loss of neurogenesis in the olfactory placode but rather a microvillous-specific phenotype . An observed slight decrease in ciliated neuron numbers and thinning of projections in morpholino-treated embryos , similar to that seen in the laser ablation experiments , is likely a secondary effect due to the absence of normally adjacent microvillous neurons . Ciliated neurons largely recover by 48 hpf development , whereas neural crest–derived microvillous neurons do not . Interestingly , the role of Sox10 in microvillous neurogenesis is consistent with its role in regulating neuronal differentiation from neural crest of other peripheral sensory and autonomic neurons across species . Sox10 is not only required for formation of sensory , sympathetic , and enteric neurons but also for glia and Schwann cells ( Britsch et al . , 2001; Sonnenberg-Riethmacher et al . , 2001; Carney et al . , 2006; Kelsh , 2006 ) . It has been a matter of debate how much of this role is direct or indirect via Sox10's regulation of glial development ( Britsch et al . , 2001; Carney et al . , 2006 ) . Data shown here , namely Sox10's expression in and specific effect on microvillous neurons within the first several hours of olfactogenesis , demonstrate a direct effect on neurogenesis at least in the context of zebrafish olfactory development . The origins of various olfactory components have been the subject of considerable controversy in the literature . It was long assumed that olfactory sensory neurons derive solely from the olfactory placode ( Hansen and Zeiske , 1993; Whitlock and Westerfield , 2000 ) . Although two lineage analysis studies in mice suggested that a small subpopulation of unspecified olfactory sensory neurons may have a neural crest origin ( Forni et al . , 2011; Katoh et al . , 2011 ) , they relied upon Wnt1-Cre and P0-Cre , both of which suffer from lack of neural crest specificity . Furthermore , conflicting work by another group also using Wnt1-Cre lineage tracing found no contribution to neurons but rather a neural crest contribution only to olfactory ensheathing glia ( Barraud et al . , 2010 ) . Interpretation of these experiments is made difficult by the fact that the Wnt1 marker is not entirely neural crest specific , as it is also expressed in the dorsal neural tube , otic placode , and several other regions ( Riccomagno et al . , 2005; Liu et al . , 2006; Freyer et al . , 2011; Valenta et al . , 2011 ) . In zebrafish , Harden et al . ( 2012 ) suggest the idea of a possible neural crest contribution to olfactory ensheathing glia but not to sensory neurons , although they provide no direct evidence to substantiate this possibility . Here , by using photoconversion of Sox10-expressing neural crest cells , we were able to unequivocally follow individual neural crest cells from their site of origin in the dorsal neural tube to their ingression into the olfactory epithelium and differentiation into microvillous sensory neurons . Olfactory sensory neurons are notable for their regenerative capacity . Hence , analysis of their lineage , paths of differentiation , and molecular determinants may provide important insights into general mechanisms of self-renewal . Since processes of renewal may be lineage-dependent , understanding the embryonic origins of different classes of olfactory neurons has distinct value . In the case of placode-derived ciliated sensory neurons , their replenishment throughout the lifespan likely occurs within the mature olfactory vesicle . Similarly , there may be a subpopulation of microvillous sensory neurons derived from the olfactory vesicle . Consistent with this possibility , we see a small number of microvillous neurons that are Sox10:eGFP-negative . Alternatively , given our evidence of large-scale neural crest contribution to microvillous sensory neurons , it is intriguing to consider that their replacement during adulthood may take place via a latent neural crest stem cell population derived from the nasal cavity . In summary , we have utilized complementary techniques in zebrafish to directly demonstrate a neural crest contribution to olfactory neurogenesis that depends critically on the transcription factor Sox10 . These data not only contradict existing dogma but open new avenues for better understanding sensory neurogenesis in the embryo and renewal in the adult . During vertebrate evolution , a myriad of changes have occurred in olfaction , including the emergence of the vomeronasal organ , which fish and birds appear to lack . Given the diverse nature of olfaction across species , it will be interesting to examine possible neural crest contributions from a comparative perspective as new lineage tracing tools become available in other model systems .
Wild-type and transgenic lines were maintained according to Institutional Animal Care and Use protocols . Embryos were grown , staged , and harvested as previously described ( Kimmel et al . , 1995; Westerfield , 2000 ) . Treatment with 1-phenyl-2-thiourea ( PTU ) to prevent pigmentation from interfering with imaging was done for all experiments except morpholino knockdowns ( to preserve observation of the ‘colorless' phenotype ) . Lines used and their abbreviations are Tg ( –4 . 9sox10:eGFP ) ( Wada et al . , 2005; Carney et al . , 2006 ) = Sox10:eGFP; Tg ( -8 . 4neurog1:nRFP ) ( Blader et al . , 2003 ) = Ngn1:nRFP; Tg ( OMP2k:lyn-mRFP ) /rw035 ( Sato et al . , 2005 ) = OMP:RFP; Tg ( TRPC24 . 5k:gap-Venus ) /rw037 ( Sato et al . , 2005 ) = TRPC2:Venus; Tg ( bactin2:memb-Cerulean-2A-H2B-Dendra2 ) pw1 ( Dempsey et al . , 2012 ) = PhOTO-N . Fish were mated in various combinations to yield compound heterozygote embryos . Due to the spectral overlap between Venus and eGFP , fate identification crosses were done only with TRPC2:Venus confirmed homozygote adults . Embryos were mounted in custom-designed embryo-shaped 1% agarose/30% Danieau molds to provide support while immersed in 2–3% methylcellulose/30% Danieau solution with standard working concentrations of tricaine anesthetic and PTU ( to prevent pigmentation ) added . Embryos were maintained at 28 . 5°C during imaging . Prior to and during experimentation , embryo staging was carefully monitored . A Zeiss LSM 510 Meta laser scanning confocal microscope was used with an LD C-Apochromat 40x/1 . 1 W Corr objective . Collected time-lapse data were registered using Imaris software ( Bitplane ) and exported as TIF images and AVI videos . Sox10:eGFP ( and sometimes also TRPC2:Venus ) fish were crossed with PhOTO-N fish . Resultant dual- or triple-positive embryos were mounted as above for live imaging . Seven to ten embryos from a single clutch were photoconverted unilaterally in a D-V or near-lateral orientation per individual experiment , in four independent experiments . Photoconversion was performed as previously described ( Dempsey et al . , 2012 ) , with the following modifications: a Zeiss LSM 5 Exciter with an LD C-Apochromat 40x/1 . 1 W Corr objective and Zen 2009 software's ‘Regions' tool were used for photoconversion at 405 nm ( 200 cycles at 30% laser power ) and immediate confirmation of conversion . Subsequent live imaging was done on a Zeiss LSM 510 Meta with an LD C-Apochromat 40x/1 . 1 W Corr objective of unconverted monomeric Dendra2Green excited at 458 nm and converted Dendra2Red excited at 561 nm . Each step ( before conversion , conversion of selected cells , and after conversion ) was captured as a single plane image , and full z-stacks were captured at the beginning and end for each photoconverted embryo . Embryos were imaged at several time points over multiple days to track the progress of photoconverted cells . Sox10:eGFP+/Ngn1:nRFP+/TRPC2:Venus+ embryos were mounted in a near-lateral angled orientation to allow ablation of olfactory epithelium cells ( data not shown ) or the nasal cavity surrounding the olfactory epithelium . There were 8–10 embryos per experiment , with two experiments directed at olfactory epithelium ablation and four at nasal cavity ablation , always unilaterally . Laser ablation was done with a two-photon laser at 840 nm for 20 cycles at 100% power on a Zeiss LSM 510 Meta with an LD C-Apochromat 40x/1 . 1 W Corr objective and Zen 2009 software's ‘Regions' tool for cell selection . Subsequently , embryos were imaged to determine degree of ablation and overall health of both targeted cells and surrounding untargeted cells and were assayed with nuclear markers in some cases . Each step ( before ablation , ablation of selected cells , and after ablation ) was captured as a single plane image , and full z-stacks were captured at the beginning and end for each experimental embryo . Embryos were then imaged at several time points over multiple days to ascertain the post-ablation phenotype . Embryos were collected at specific time points , fixed with 4% PFA , and washed in PBST . They were then washed in PBDT ( 1% BSA , 1% DMSO , 0 . 1% Triton X-100 in PBS , pH 7 . 4 ) , blocked in 10% normal donkey or goat serum/PBDT , and incubated overnight at 4°C with primary antibodies to HuC/D ( 1:100; Invitrogen # A-21271 ) , GFP ( 1:250; Abcam # 6673 ) , and/or zSox10 ( Park et al . , 2005 ) ( 1:500 ) . Further PBST washes and blocking were followed by secondary antibodies ( Invitrogen ) overnight at 4°C . Hoechst 34580 was added to stain nuclei . After further PBDT and PBS washes , embryos were mounted for confocal imaging on a Zeiss LSM 5 Exciter with a Zeiss LD C-Apochromat 40x/1 . 1 W Corr objective . Each experiment was done at least in triplicate . All morpholinos were obtained from Gene Tools , LLC . A previously used translation-blocking antisense morpholino against Sox10 ( ATGCTGTGCTCCTCCGCCGACATCG; Dutton et al . , 2001a , 2001b; Prasad et al . , 2011; Whitlock et al . , 2005 ) targets the −23 to +2 region of the Sox10 sequence . Based on previously used amounts in these papers , several concentrations in the 5–15 ng per embryo range were tested . All data reported here were obtained from injections of 7 ng ( in 2 . 3 nl ) of the Sox10 morpholino or control morpholino into one- to two-cell embryos . On day 1 , embryos were sorted based on health and any aberrations ( mild , moderate , and severe ) , the number of melanocytes ( high and low ) , and the expression profile of Sox10:eGFP and/or TRPC2:Venus ( normal , decreased/abnormal , and none ) . Multiple representative embryos were imaged on a Zeiss LSM 510 Meta with an LD C-Apochromat 40x/1 . 1 W Corr objective . Some embryos were fixed and stained with antibodies against GFP/Venus and/or Sox10 as outlined above . Numbers of control and experimental embryos are indicated in Figure 5 . A second previously used translation-blocking antisense morpholino against Sox10 ( GCCACAGGTGACTTCGGTAGGTTTA; Dutton et al . , 2001a ) targets the −43 to −19 region of the Sox10 sequence . A sense photo-morpholino ( AACCTACCGAPGTCACCTGTG ) targets the antisense morpholino , with ‘P' at position 11 indicating a photocleavable moiety . The antisense morpholino was used at 7 ng per embryo as described above for the first antisense morpholino . A premixed solution was made at room temperature of antisense morpholino and sense photo-morpholino at a 1:1 . 1 molar ratio and injected into one- to two-cell embryos . After testing several permutations , photocleavage described here was done at 17 . 5 or 24 hpf on a Zeiss Axiovert 35 fluorescent microscope with a 5× objective and illumination via an Opti-Quip Model 1200 mercury lamp paired with a G365/FT395/LP420 filter set for 15 min , three to four embryos at a time . Some control embryos were injected but not photocleaved , and other control embryos were uninjected but illuminated under the same conditions as injected embryos to control for any effects of UV light illumination . Embryos were then incubated until later time points , and those appearing healthiest were selected for imaging with a laser scanning confocal microscope as before . Three independent experiments were performed with 12–15 embryos imaged per experiment . Calculations to determine statistical significance of Sox10 morpholino experiments were performed using Student's t-test of variables ( two-sample t-test assuming unequal variances ) . | Neurons have crucial roles in both the peripheral and central nervous systems . The role of the neurons in the sensory organs ( the eyes , ears , and nose ) is to sense stimuli—including light , sound , and odor—and transmit this sensory information to the neurons of the central nervous system for processing . The first step in sensing an odor relies on the peripheral nerves of the olfactory epithelium . This tissue , which lines the inside of the nasal cavity , includes two main types of olfactory sensory neurons: ciliated sensory neurons that detect volatile or easily evaporated substances and microvillous sensory neurons that detect pheromones , nucleotides , and/or amino acids . During vertebrate embryogenesis , an embryo develops three distinct germ layers , the ectoderm , mesoderm , and endoderm , each of which gives rise to the different tissues of the body . The ectoderm has three parts—the external ectoderm , the neural crest , and the neural tube—and together they give rise to the neurons of the peripheral and central nervous systems . The neurons within the eye and ear are known to originate from a thickened portion of the ectoderm , and it has been proposed that olfactory neurons develop in a similar manner . Now , Saxena et al . show that , unlike what happens in the eye and ear , some olfactory sensory neurons originate from the neural crest . By studying the development of the olfactory system in zebrafish , Saxena et al . discovered that microvillous neurons , but not ciliated neurons , develop from neural crest cells , and that the transcription factor Sox10 is critical for the development of microvillous neurons . By establishing that neural crest cells are involved in the development of a substantial proportion of olfactory sensory neurons , this work sets the stage for future studies of olfactory nerve growth and regeneration . It may also assist researchers working on anosmia ( the inability to smell ) . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2013 | Sox10-dependent neural crest origin of olfactory microvillous neurons in zebrafish |
Transcription factor IIH ( TFIIH ) is a heterodecameric protein complex critical for transcription initiation by RNA polymerase II and nucleotide excision DNA repair . The TFIIH core complex is sufficient for its repair functions and harbors the XPB and XPD DNA-dependent ATPase/helicase subunits , which are affected by human disease mutations . Transcription initiation additionally requires the CdK activating kinase subcomplex . Previous structural work has provided only partial insight into the architecture of TFIIH and its interactions within transcription pre-initiation complexes . Here , we present the complete structure of the human TFIIH core complex , determined by phase-plate cryo-electron microscopy at 3 . 7 Å resolution . The structure uncovers the molecular basis of TFIIH assembly , revealing how the recruitment of XPB by p52 depends on a pseudo-symmetric dimer of homologous domains in these two proteins . The structure also suggests a function for p62 in the regulation of XPD , and allows the mapping of previously unresolved human disease mutations .
Transcription factor IIH ( TFIIH ) is a 10-subunit protein complex with a total molecular weight of 0 . 5 MDa that serves a dual role as a general transcription factor for transcription initiation by eukaryotic RNA polymerase II ( Pol II ) , and as a DNA helicase complex in nucleotide excision DNA repair ( NER ) ( Compe and Egly , 2016; Sainsbury et al . , 2015 ) . Mutations in TFIIH subunits that cause the inherited autosomal recessive disorders xeroderma pigmentosum ( XP ) , trichothiodystrophy ( TTD ) , and Cockayne syndrome ( CS ) are characterized by high incidence of cancer or premature ageing ( Cleaver et al . , 1999; Rapin , 2013 ) . Furthermore , TFIIH is a possible target for anti-cancer compounds ( Berico and Coin , 2018 ) and therefore of great importance for human health and disease . The TFIIH core complex is composed of the seven subunits XPB , XPD , p62 , p52 , p44 , p34 , and p8 , and is the form of TFIIH active in DNA repair ( Svejstrup et al . , 1995 ) , where TFIIH serves as a DNA damage verification factor ( Li et al . , 2015; Mathieu et al . , 2013 ) and is responsible for opening a repair bubble around damaged nucleotides . This activity depends on both the SF2-family DNA-dependent ATPase XPB , and the DNA helicase activity of XPD ( Coin et al . , 2007; Evans et al . , 1997; Kuper et al . , 2014 ) . TFIIH function in transcription initiation requires the double-stranded DNA translocase activity of XPB to regulate opening of the transcription bubble ( Alekseev et al . , 2017; Fishburn et al . , 2015; Grünberg et al . , 2012 ) , and additionally the CdK activating kinase ( CAK ) complex , which harbors the kinase activity of CDK7 as well as the Cyclin H and MAT1 subunits ( Devault et al . , 1995; Fisher et al . , 1995; Fisher and Morgan , 1994; Shiekhattar et al . , 1995; Svejstrup et al . , 1995 ) . Targets of human CDK7 include the C-terminal heptapeptide repeat domain of the largest subunit of Pol II , as well as cell-cycle regulating CDKs ( Fisher and Morgan , 1994; Shiekhattar et al . , 1995 ) . MAT1 serves as a bridging subunit that promotes CAK subcomplex formation by interacting with Cyclin H and CDK7 ( Devault et al . , 1995; Fisher et al . , 1995 ) , recruits the CAK to the core complex by interactions with XPD and XPB ( Abdulrahman et al . , 2013; Busso et al . , 2000; Greber et al . , 2017; Rossignol et al . , 1997 ) , and also aids in Pol II-PIC formation by establishing interactions with the core PIC ( He et al . , 2013; He et al . , 2016; Schilbach et al . , 2017 ) . The presence of MAT1 inhibits the helicase activity of XPD ( Abdulrahman et al . , 2013; Sandrock and Egly , 2001 ) , but the mechanism of this inhibition is not fully understood . While the enzymatic activity of XPD is not required for transcription initiation , it is critical for the DNA repair function of TFIIH ( Dubaele et al . , 2003; Evans et al . , 1997; Kuper et al . , 2014 ) . Therefore , NER requires the release of the CAK subcomplex from the core complex ( Coin et al . , 2008 ) . The activities of both XPB and XPD are regulated by interactions with additional TFIIH components , including that of p44 with XPD ( Coin et al . , 1998; Dubaele et al . , 2003; Kim et al . , 2015 ) , and those of the p52-p8 module with XPB ( Coin et al . , 2007; Coin et al . , 2006; Jawhari et al . , 2002; Kainov et al . , 2008 ) . These interactions are likely to be crucial for TFIIH function , as some are affected by disease mutations ( Cleaver et al . , 1999 ) , but they have been only partially characterized mechanistically . Our previous structure of the TFIIH core-MAT1 complex at 4 . 4 Å resolution ( Greber et al . , 2017 ) allowed modeling of TFIIH in the best-resolved parts of the density map , but several functionally important regions remained unassigned or only partially interpreted because reliable de novo tracing of entire domains in the absence of existing structural models was not possible . Here , we present the complete structure of the human TFIIH core complex in association with the CAK subunit MAT1 , determined by phase plate cryo-electron microscopy ( cryo-EM ) at 3 . 7 Å resolution . Our structure reveals the complete architecture of the TFIIH core complex and provides detailed insight into the interactions that govern its assembly . Additionally , our cryo-EM maps define the molecular contacts that control the regulation of the XPB and XPD subunits of TFIIH , including the critical p52-XBP interaction , and an extensive regulatory network around XPD , formed by XPB , p62 , p44 , and MAT1 .
To determine the complete structure of the human TFIIH core complex , we collected several large cryo-EM datasets ( Supplementary file 1 ) of TFIIH immuno-purified from HeLa cells using an electron microscope equipped with a Volta phase plate ( VPP ) ( Danev and Baumeister , 2017 ) and a direct electron detector camera mounted behind an energy filter . From a homogeneous subset of approximately 140 , 000 TFIIH particle images identified by 3D classification ( Scheres , 2010 ) , we reconstructed a 3D cryo-EM density map at 3 . 7 Å resolution ( Figure 1—figure supplements 1 and 2A–C ) . This VPP-based cryo-EM map was substantially improved compared to our previous maps obtained without phase plate , both in resolution and interpretability ( Figure 1—figure supplement 2D–G ) , and enabled building , refinement , and full validation of an atomic model of the TFIIH core complex and the MAT1 subunit of the CAK subcomplex ( Figure 1A–C , Figure 1—figure supplement 2B , C , Supplementary file 2 , 3 ) , while the remainder of the CAK subcomplex is invisible in our map because it is flexibly tethered to the TFIIH core complex . Tracing and sequence register assignment of protein components modeled de novo was facilitated by density maps obtained from focused classification and multibody refinement ( Figure 1—figure supplements 3–5 ) ( Bai et al . , 2015; Nakane et al . , 2018 ) , which resulted in density maps of improved interpretability for all three sub-volumes and a slightly improved resolution of 3 . 6 Å for the XPD-MAT1 region . Both the overall and multibody-refined maps showed clear side chain information ( Figure 1—figure supplement 6A–D ) . Furthermore , our model was corroborated by existing chemical crosslinking-mass spectrometry ( CX-MS ) data of human TFIIH ( Luo et al . , 2015 ) and site-specific crosslinks from yeast TFIIH ( Warfield et al . , 2016 ) ( Figure 1—figure supplement 6E–I , Supplementary file 4 ) . Our structure of the TFIIH core complex shows its horseshoe-like overall shape ( Figure 1A–C , Video 1 ) , as observed in previous lower-resolution reconstructions of free and PIC-bound TFIIH ( Gibbons et al . , 2012; Greber et al . , 2017; He et al . , 2016; Murakami et al . , 2015; Schilbach et al . , 2017 ) , and allows us to define the complete set of inter-subunit interactions that lead to the formation of the TFIIH core complex directly from our structure ( Figure 1D ) . The largest subunits of the complex , the SF2-family DNA-dependent ATPases XPB and XPD , both containing two RecA-like domains ( RecA1 and RecA2 ) , interact directly ( Greber et al . , 2017 ) , are on one side of the complex , and are additionally bridged by MAT1 ( Figure 1B ) , which has been shown to interact with either ATPase in isolation ( Busso et al . , 2000 ) . On the side facing away from MAT1 , XPD interacts with the von Willebrand Factor A ( vWFA ) domain of p44 ( Coin et al . , 1998; Dubaele et al . , 2003; He et al . , 2016; Kim et al . , 2015 ) , which in turn forms a tight interaction with p34 via interlocking eZnF domains ( Schilbach et al . , 2017 ) and a p44 RING domain interaction ( Radu et al . , 2017 ) ( Figure 1B–D , Figure 1—figure supplement 6J , K ) , consistent with the formation of a multivalent interaction network between p34 and p44 ( Radu et al . , 2017 ) . The vWFA domain of p34 recruits p52 by a three-way interaction that involves the most N-terminal winged helix domain in p52 and a helical segment of p62 ( Schilbach et al . , 2017 ) ( Figure 1—figure supplement 6L ) . The p52 C-terminal region comprises two domains; first , the ‘clutch’ that interacts with XPB ( Jawhari et al . , 2002 ) and second , a dimerization module that binds p8 ( Kainov et al . , 2008 ) , thereby recruiting XPB to TFIIH and cradling XPB RecA2 ( see below ) . In addition to this structural framework that is formed by folded domains , our cryo-EM map reveals several interactions involving extended protein segments , including several interactions formed by p62 ( Figure 2 ) , and an interaction between the p44 N-terminal extension ( NTE ) and the N-terminal domain ( NTD ) of XPB ( Figure 1B ) . To form this interaction , approx . 15 residues of p44 span the distance between the p44 vWFA domain and the XPB NTD , where a small helical motif in p44 contacts XPB residues 72–75 , 95–102 , and 139–143 , in agreement with CX-MS data ( Luo et al . , 2015 ) ( Figure 1—figure supplement 6E ) . Partial deletion of the p44 NTE in yeast causes a slow-growth phenotype , suggesting a functional role for this p44-XPB interaction ( Warfield et al . , 2016 ) . The p62 subunit is almost completely resolved in our structure and exhibits a complex beads-on-a-string-like topology . It fully encircles the top surface of TFIIH ( Figures 1C and 2A , Figure 2—figure supplement 1 ) , interacting with XPD , p52 , p44 , and p34 , in agreement with previous structural findings ( Greber et al . , 2017; Schilbach et al . , 2017 ) . Based on these interactions , p62 can be subdivided into three functional regions: ( i ) the N-terminal PH-domain , disordered in our structure , is responsible for mediating interactions with components of the core transcriptional machinery ( Di Lello et al . , 2008; He et al . , 2016; Schilbach et al . , 2017 ) , transcriptional regulators ( Di Lello et al . , 2006 ) , and DNA repair pathways ( Gervais et al . , 2004; Lafrance-Vanasse et al . , 2013; Okuda et al . , 2017 ) ; ( ii ) residues 108–148 and 454–548 of p62 , including the first BSD ( BTF2-like , synapse-associated , DOS2-like ) domain ( BSD1 ) and the C-terminal 3-helix bundle , play an architectural role by binding to p34 and the extended zinc finger ( eZnF ) domain of p44 ( Figure 2A , Figure 1—figure supplement 6J–L , Figure 2—figure supplement 1B , C ) ; and ( iii ) residues 160–365 , including the BSD2 domain , are responsible for interactions with and regulation of XPD ( Figure 2B–D , Figure 2—figure supplement 1D–F ) . Specifically , p62 residues 160–365 form three structural elements that interact with XPD ( Figure 2B , Video 1 ) , in agreement with previous biochemical , structural , and CX-MS data ( Figure 1—figure supplement 6G ) ( Jawhari et al . , 2004; Luo et al . , 2015; Schilbach et al . , 2017 ) . First , an α-helix formed by p62 residues 295–318 binds directly to XPD RecA2 and thereby recruits residues 160–258 of p62 , comprising the BSD2 domain and adjacent sequence elements , to this surface of XPD RecA2 ( Figure 2A , B , Figure 2—figure supplement 1D ) . Second , p62 residues 266–287 are inserted into the DNA-binding cavity of XPD ( Figure 2B , D ) , in agreement with previous observations ( Schilbach et al . , 2017 ) . This inserted p62 segment directly blocks a DNA-binding site on XPD RecA1 ( Figure 2D ) and localizes near the access path to a pore-like structure between the XPD FeS and ARCH domains . While p62 does not directly contact the DNA-binding surface on XPD RecA2 , it may still sterically interfere with DNA binding or access to the helicase elements of XPD in this region ( Figure 2—figure supplement 1E ) . Therefore , this segment of p62 may need to move away when XPD binds and unwinds DNA . Third , p62 residues 350–358 form a short α-helix that binds in a cleft between the two RecA-like domains of XPD ( Figure 2C ) , so that it not only closes the entrance to the nucleotide binding pocket in XPD RecA1 ( Figure 2—figure supplement 1F ) , but also partially overlaps with the predicted location of the nucleotide itself ( Figure 2C ) , strongly suggesting a role for this p62 sequence element in XPD regulation . The density for these structural elements of p62 ( residues 260–300 and 346–365 ) in our cryo-EM map is weaker than for the remainder of the complex , suggesting a dynamic interaction with XPD that enables them to modulate the access to the nucleotide-binding pocket , the DNA-binding cavity , and the DNA-translocating pore of XPD , depending on the functional state of TFIIH . 3D reconstructions of TFIIH classified for these regions of p62 ( Figure 1—figure supplement 3 ) show globally intact TFIIH , both in the presence and absence of the p62 segments at these XPD sites ( Figure 2—figure supplement 1G–J ) , supporting our hypothesis of dynamic regulation , rather than the alternative hypothesis of p62 binding to XPD as a requirement for TFIIH stability ( Luo et al . , 2015 ) . Our structure of TFIIH resolves the structure and interactions of all four folded domains of human XPB – two RecA-like domains that form the SF2-family type helicase cassette , a DNA damage recognition domain ( DRD ) -like domain , and an N-terminal domain ( NTD ) ( Figure 3A , Figure 3—figure supplement 1 ) – and reveals the molecular basis of XPB recruitment by p52 . The XPB NTD encompasses residues 1–165 , with the first 55 residues forming an N-terminal extension ( NTE ) , and the remainder assuming a mixed α/β-fold with four α-helices and five β-strands ( Figure 3A ) . The side chain densities in the cryo-EM map ( Figure 1—figure supplement 6A ) and CX-MS data ( Luo et al . , 2015 ) ( Figure 1—figure supplement 6F ) both confirm our assignment of this domain . Existing biochemical data show that the XPB NTD is required for integration of XPB into TFIIH ( Jawhari et al . , 2002 ) by forming an interaction with p52 that has been referred to as the ‘clutch’ ( Schilbach et al . , 2017 ) . In our structure , the p52 contribution to the clutch encompasses p52 residues 306–399 , which , strikingly , assume the same overall fold as the XPD NTD ( Figure 3B ) , as hypothesized previously ( He et al . , 2016; Luo et al . , 2015 ) , thereby forming a pseudo-symmetric dimer of structurally homologous domains . The two domains interact through their β-sheets , via both hydrophobic and charged interactions ( Figure 3—figure supplement 2A–C ) , and with the most N-terminal β-strand emanating from the XPB NTD extending the p52 β-sheet by additional lateral interactions ( Figure 3A , C ) . Our structural findings rationalize biochemical data that show that deletion of XPB residues 1–207 , but not deletion of residues 1–44 , impairs the p52-XPB interaction ( Jawhari et al . , 2002 ) ( Figure 3C ) . Our structure is also consistent with data indicating that p52 residues 304–381 are critical for the XPB-p52 interaction ( Coin et al . , 2007; Jawhari et al . , 2002 ) , but does not show any contacts that could explain that reported binding of XPB to p52 residues 1–135 or 1–304 ( Jawhari et al . , 2002 ) ( Figure 3—figure supplement 2D , E ) . The interaction between p52 and XPB not only recruits XPB to TFIIH , but also stimulates its ATPase activity in vitro ( Coin et al . , 2007 ) . Because our structure does not shown any elements of p52 approaching the XPB nucleotide-binding pocket , we propose that this effect is likely induced by the interactions of p52 with the XPB NTD and RecA2 , which may , together with p8 ( Coin et al . , 2006 ) , properly arrange the XPB helicase cassette to bind and hydrolyze ATP ( Figure 3D ) ( Grünberg et al . , 2012 ) . The XPB NTD is the site of the two human disease mutations F99S and T119P , which cause XP and TTD , respectively ( Cleaver et al . , 1999 ) . Or structure shows that neither of these residues is in direct contact with p52 or the RecA-like domains of XPB , suggesting that the F99S and T119P mutations exert their detrimental effects through structural perturbation of the XPB NTD ( Figure 3—figure supplements 1B and 2F–I ) . Specifically , T119 is located near a turn at the end of a β-strand ( Figure 3—figure supplement 2G ) , where its side chain points towards the solvent . Nevertheless , this residue is highly conserved in eukaryotic XPB from ciliates to humans , in some archaeal and bacterial XPB homologs ( Figure 3—figure supplement 1B ) , and in the structurally homologous clutch domain in p52 ( Figure 3—figure supplement 2F–H ) . This conservation suggests that a threonine at this position is important for the efficient folding of this domain in general , and that the T119P mutation may cause its destabilization , resulting in lower levels of active enzyme in TTD patients . Lower overall levels of properly assembled TFIIH have been shown to be a hallmark of TTD ( Botta et al . , 2002; Dubaele et al . , 2003; Giglia-Mari et al . , 2004 ) and could explain the disease-causing effect of T119P in vivo even though recombinant TFIIH carrying this mutation retains some activity in both transcription initiation and NER ( Coin et al . , 2007 ) . A less likely alternative , given the conservation of the equivalent residue in the p52 clutch , is that T119 is involved in an interaction with a factor that is critical for cellular function , for example in NER . The F99S mutation affects a residue that is conserved throughout eukaryotic XPB ( Figure 3—figure supplement 1B ) , is inserted into a conserved hydrophobic pocket , and localizes to an α-helix at the XPB contact site with the p44 N-terminal extension ( Figure 3—figure supplement 2I ) . This mutation is likely to impair the stability and folding of the XPB NTD . Unlike T119P , this mutation leads to impaired DNA opening in NER assays , reduced interaction with p52 , reduced ATPase activity ( Coin et al . , 2007 ) , and strong impairment in DNA damage repair ( Riou et al . , 1999 ) , suggesting a severe effect on the structure of the XPB NTD . Natural and synthetic mutations in the Drosophila melanogaster homolog of p52 that lead to disease-like phenotypes in flies and have similar defects when introduced into human cells ( Fregoso et al . , 2007 ) map directly to the p52-XPB interface , explaining their detrimental phenotypes ( Figure 3C , Figure 3—figure supplement 2A ) . Our structure assigns XPB residues 165–300 to a DRD-like domain that connects the NTD to the RecA-like domain ( Figure 3A , Figure 3—figure supplement 1A ) , the deletion of which is lethal in yeast ( Warfield et al . , 2016 ) . The DRD is a DNA-binding domain found in DNA repair enzymes and chromatin remodelers ( Mason et al . , 2014; Obmolova et al . , 2000 ) and has been implicated in DNA damage recognition in archaeal XPB ( Fan et al . , 2006; Rouillon and White , 2010 ) . Our 3 . 7 Å-resolution map of TFIIH reveals that in eukaryotic XPB , one β-strand of the DRD of archaeal XPB is replaced by an insertion of approximately 70 residues that exhibits relatively low sequence conservation ( Figure 3E , Figure 3—figure supplement 1B ) and shifts the domain boundaries of the human XPB DRD-like domain with respect to previous sequence alignments ( Fan et al . , 2006; Oksenych et al . , 2009 ) . The part of this insertion resolved in our map consists of a negatively charged linker and an α-helical element that contacts XPD directly ( Figure 3E , F ) . The surface on XPD involved in this interaction has been implicated in the initial step of DNA substrate binding by XPD ( Constantinescu-Aruxandei et al . , 2016; Kuper et al . , 2012 ) . Density features and secondary structure prediction indicate the presence of several aromatic side chains of XPB near the interface ( Figure 3E ) , where they might form contacts resembling those of nucleoside bases of XPD-bound DNA substrates ( Figure 3F ) . Thus , XPB may modulate substrate binding by XPD , further reinforcing the idea that XPD activity is regulated by several other components of TFIIH . In order to investigate the dynamics of TFIIH , we analyzed the conformational landscape of the particles in our cryo-EM dataset ( Figure 4A , Figure 4—figure supplement 1; see Materials and methods for details ) ( Nakane et al . , 2018 ) . This analysis revealed the relative motions of the two ATPases and their domains ( Figure 4A ) . The major mode of motion , which involves the breaking of the interaction between XPB and XPD , resembles the conformational transition of TFIIH when it enters the Pol II-PIC and binds to DNA ( Greber et al . , 2017; He et al . , 2016; Schilbach et al . , 2017 ) ( Figure 4—figure supplement 2A–C , Video 2 ) . Analysis of our TFIIH structure and comparison with that of the complex within the Pol II-PIC maps ( He et al . , 2016; Schilbach et al . , 2017 ) allowed us to identify a specific rearrangement at the interface between the clutch and adjacent winged helix domain in p52 ( Figure 4B ) as the basis of this conformational change . A structural unit composed of XPB , p8 and the clutch domain of p52 undergoes a downward motion upon DNA-binding within the Pol II-PIC ( Figure 4B , Figure 4—figure supplement 2B , C ) . This conformational change in TFIIH upon PIC entry also appears to break the interaction between MAT1 and the XPB DRD-like domain ( Figure 4C ) , which in turn might serve to enable positioning of the CDK7-cyclin H dimer within the CAK subcomplex at the appropriate location for Pol II-CTD phosphorylation in the mediator-bound Pol II-PIC ( Figure 4D ) ( Robinson et al . , 2016; Schilbach et al . , 2017 ) . Our structural comparison also reveals that a TFIIE-XPB interaction that has been implicated in XPB regulation ( Schilbach et al . , 2017 ) may depend on the existence of the open conformation of TFIIH , as there would be steric hindrance in a complex involving the closed conformation of TFIIH ( Figure 4—figure supplement 2D–F ) . The structure of XPD shows the conserved domain arrangement of two RecA-like domains ( RecA1 and RecA2 ) , with the FeS and ARCH domain insertions in RecA1 ( Constantinescu-Aruxandei et al . , 2016; Fan et al . , 2008; Kuper et al . , 2012 ) . The quality of the map allowed us to interpret the density for the N- and C-termini of XPD , which closely approach each other near the nucleotide-binding site within RecA1 ( Figure 5A ) . The N-terminus of XPD forms a short two-stranded β-sheet near the nucleotide-binding site in XPD RecA1 ( which is empty in our structure ) and may contribute to the stabilization of the bound nucleotide via the aromatic side chains of Y14 and Y18 , the latter being affected by the Y18H mutation in an XP/TTD patient ( Kralund et al . , 2013 ) ( Figure 5—figure supplement 1A–C ) . The XPD C-terminal segment runs along the side of XPD RecA2 and interacts with the linker between RecA1 and RecA2 ( Figure 5—figure supplement 1D ) . The C-terminal segment includes the site of the K751Q polymorphism ( Figure 5—figure supplement 1D ) , and deletion of this terminal segment causes XP in human patients ( Cleaver et al . , 1999 ) . Before XPD-bound DNA reaches the helicase motifs in the RecA like domains , it passes through a pore-like structure next to the 4FeS cluster at the interface between the FeS and ARCH domains ( Figure 5—figure supplement 1E , F ) ( Cheng and Wigley , 2018; Constantinescu-Aruxandei et al . , 2016; Kuper et al . , 2012; Liu et al . , 2008; Wolski et al . , 2008 ) . This region was poorly defined in previous TFIIH reconstructions , but our cryo-EM map now shows side-chain densities for the aromatic residues Y158 , F161 , and F193 , which are critical for the DNA-binding , ATPase , and helicase activities of XPD ( Kuper et al . , 2014 ) , as well as for residues Y192 and R196 , which form part of a DNA lesion recognition pocket ( Mathieu et al . , 2013 ) ( Figure 5B ) . This functionally important region is only partially conserved in archaeal XPD homologs ( Figure 5—figure supplement 1G–I ) ( Fan et al . , 2008; Kuper et al . , 2012; Wolski et al . , 2008 ) . A eukaryotic-specific loop insertion in the XPD ARCH domain ( Greber et al . , 2017; Schilbach et al . , 2017 ) closely approaches this binding pocket ( Figure 5B ) and may serve to regulate the binding of DNA in the lesion recognition pocket such as to prevent untimely access of substrates to the XPD pore . Our structure of TFIIH shows that XPD forms architectural and regulatory interactions with four other TFIIH subunits: XPB , p62 , p44 , and MAT1 , which together form a cradle-like structure around XPD ( Figure 5C ) . We described above two interactions that could potentially regulate XPD activity: the newly defined interaction of an insertion element in the XPB DRD with a DNA-binding site in XPD ( Figure 3E , F ) ; and XPD-p62 interactions involving the XPD nucleotide binding pocked and DNA binding cavity ( Figure 2 , Figure 2—figure supplement 1D–F ) that implicate p62 , as well as XPB , in XPD regulation . Additionally , it is known that the helicase activity of XPD is inhibited by the CAK subcomplex ( Araújo et al . , 2000; Sandrock and Egly , 2001 ) . The contacts we see between MAT1 and XPD localize to the ARCH domain of XPD and the N-terminal RING domain and helical bundle of MAT1 ( residues 1–130 ) ( Figure 5C ) , in agreement with previous structural ( Greber et al . , 2017; Schilbach et al . , 2017 ) and biochemical analysis ( Abdulrahman et al . , 2013; Luo et al . , 2015; Warfield et al . , 2016 ) . The interaction between the XPD ARCH domain and the MAT1 helical bundle is characterized by charge complementarity ( Figure 5—figure supplement 1J–M ) . This interface is highly dynamic , enabling the release of MAT1 and the entire CAK subcomplex from TFIIH during NER , as well as its subsequent re-association to regenerate a transcription-competent TFIIH ( Coin et al . , 2008 ) . Insertion of substrate DNA into the pore between the XPD ARCH and FeS domains requires the flexibility of the XPD ARCH domain ( Constantinescu-Aruxandei et al . , 2016 ) , and large domain motions have been observed in the structure of the DNA-bound homologous helicase DinG upon nucleotide binding ( Cheng and Wigley , 2018 ) . This suggests a role for the mobility of the ARCH domain in both DNA loading and DNA translocation by the XPD helicase . Our structure suggests that binding of the MAT1 helical bundle and RING domain to the ARCH domain may prevent such motion and therefore the subsequent substrate loading and XPD helicase activity ( Figure 5D ) , in agreement with biochemical data that show XPD inhibition upon MAT1 binding ( Sandrock and Egly , 2001 ) , as well as reduced single-stranded DNA affinity of TFIIH in the presence of the CAK ( Li et al . , 2015 ) . Conversely , release of MAT1 from XPD might allow the ARCH domain to move more freely , thereby de-repressing XPD . Furthermore , displacement of the MAT1 α-helix that connects XPD to XPB may allow XPB to move away from XPD , thereby unmasking the substrate-binding site on XPD RecA2 that is otherwise occluded by the DRD insertion element ( Figure 5D ) . This latter conformational change would be similar , overall , to that seen for TFIIH upon incorporation into the Pol II-PIC , where XPD and XPB move apart and density for the MAT1 helix is missing ( Figure 4C ) ( He et al . , 2016; Schilbach et al . , 2017 ) . We propose that the combined unmasking of the XPD substrate binding site and the enhanced flexibility of the XPD ARCH domain may both contribute to de-repression of the XPD helicase upon release of MAT1 . This mechanism of XPD inhibition by MAT1 does not exclude the possibility of additional repression of NER activity by the CAK subcomplex through phosphorylation of NER pathway components ( Araújo et al . , 2000 ) . Our structure also resolves in detail the XPD-p44 interaction , a known regulatory interface ( Dubaele et al . , 2003; Kim et al . , 2015; Kuper et al . , 2014 ) affected by numerous disease mutations ( Cleaver et al . , 1999; Greber et al . , 2017; Kuper et al . , 2014 ) ( Figure 5E ) . The relatively small interaction surface between p44 and XPD , of just 940 Å2 ( Figure 5—figure supplement 2A ) , contrasts with the much larger buried surface of 3300 Å2 between XPD and p62 , or 1580 Å2 for the p52-XPB interaction . This smaller interaction surface may result in higher sensitivity to mutations that localize at the XPD-p44 interface . Our structure , thus , rationalizes the deleterious effect of a number of natural and synthetic mutations in this interface ( see Appendix 1 and Figure 5—figure supplement 2B–E ) , including mutations L174W and T175R in the β4-α5 loop of p44 ( Figure 5E ) ( Kim et al . , 2015; Seroz et al . , 2000 ) , which may lead to steric clashes in the densely packed interface ( Figure 5—figure supplement 2B ) , and the XPD R722W mutation ( Kuper et al . , 2014 ) , which disrupts the salt bridge with D75 in p44 and may additionally cause steric clashes with neighboring p44 residues due to the bulky tryptophan side chain ( Figure 5—figure supplement 2C ) . Our structure also shows that , in contrast to a previously proposed model ( Luo et al . , 2015 ) , the XPD R616P , D673G , and G675R disease mutations act either via disruption of the XPD structure or the XPD-p44 interface , but not via disruption of the interaction with p62 ( see Appendix 1 and Figure 5—figure supplement 2D ) . Notably , the p44-dependet stimulation of XPD activity does not depend on the presence of p62 ( Dubaele et al . , 2003; Kim et al . , 2015; Kuper et al . , 2014 ) . We were also able to map a number of disease mutations onto our XPD structure ( Figure 6A , Video 3 , Appendix 2 ) and analyze in detail the interactions involving the affected residues ( example shown in Figure 6B ) . Our analysis confirms that XP mutations mostly localize near the helicase substrate-binding or active sites , while TTD mutations predominantly localize to the periphery of XPD ( Figure 6 , Figure 6—figure supplement 1 ) ( Fan et al . , 2008; Liu et al . , 2008 ) , where they disrupt TFIIH assembly and cause the transcription defects that are a hallmark of this disease ( Dubaele et al . , 2003 ) ( Appendix 2 ) .
Our study reveals the complete structure of the TFIIH core complex and provides mechanistic insights into the regulation of its two component helicases . Specifically , it shows XPD wrapped by numerous interactions with XPB , p62 , p44 , and MAT1 ( Figure 5C , D ) , indicating how its activity can be tightly controlled and de-repressed only when its enzymatic function is needed . XPD activity is not needed and most likely inhibited during transcription initiation , but it may also be tightly controlled during NER , when repair bubble opening and lesion verification need to be coordinated with the recruitment and activation of the damage recognition and processing machinery ( Figure 7 ) . While the regulation of XPD by MAT1 and p44 has been studied in some detail , and the domain motions in TFIIH suggest a straightforward mechanism for liberating the substrate-binding site on XPD RecA2 , less was known about the interplay between XPD and p62 . Our structure now shows how p62 is able to impede both substrate and nucleotide binding in XPD RecA1 , and hints at dynamic structural changes of p62 during de-repression and enzymatic activity of XPD , possibly regulated by other components of the transcription or NER pathways . Our results allow us to put extensive biochemical data on the NER pathway into a structural context ( Figure 7 ) . Depending on whether XPB binds to the damaged ( Figure 7A ) or undamaged ( Figure 7B ) strand , the combined action of XPD and XPB could lead to the extrusion of a DNA bubble ( Figure 7A ) or to the tracking of the entire complex towards the lesion ( Figure 7B ) , which is initially located 3’ of the TFIIH binding site ( Sugasawa et al . , 2009 ) . The latter hypothesis is attractive in the context of biochemical data that show that XPD tracks along the damaged strand in the 5’ to 3’ direction until it encounters the DNA lesion in order to verify the presence of a bona fide NER substrate ( Buechner et al . , 2014; Li et al . , 2015; Mathieu et al . , 2013; Naegeli et al . , 1993; Sugasawa et al . , 2009; Wirth et al . , 2016 ) . It is worth noting that the length of DNA fragments excised during NER is approx . 29 nt , with 22 nt located 5’ and 5 nt located 3’ of a thymine dimer lesion ( Huang et al . , 1992 ) . According to our structure , the 22 nt 5’-fragment corresponds well to the estimated 20 nt of DNA that are required to span the distance from the DNA damage verification pocket in XPD ( Mathieu et al . , 2013 ) to the helicase elements of XPB . This proposal is compatible with a model in which TFIIH sitting on the open repair bubble might track towards the lesion , where it would stop due to inhibition of XPD ( Li et al . , 2015; Mathieu et al . , 2013; Naegeli et al . , 1993 ) , at which point double incision could be initiated . However , this model ( Figure 7B ) would require strong DNA bending before both XPB and XPD could be loaded . Additionally , it has not been fully resolved whether XPB participates in DNA translocation or unwinding during TFIIH activity in NER ( Li et al . , 2015 ) , which would be required in the tracking model ( Li et al . , 2015 ) , or whether it exclusively acts to anchor the complex in the vicinity of the DNA lesion ( Coin et al . , 2007; Oksenych et al . , 2009 ) . Independently of the orientation of the repair bubble , our structural data are compatible with literature data introduced above and a model ( Figure 7C ) that localizes XPG near XPD and p62 ( site of 3’-incision ) , XPF-ERCC1 near XPB ( site of 5’-incision ) , and with RPA binding the non-damaged strand ( Fagbemi et al . , 2011 ) . We have currently not included XPA in this model because its interactions with distinct partners or participation in various processes , such as involvement in CAK release ( Coin et al . , 2008 ) , binding to p8 ( Ziani et al . , 2014 ) , and participation in helicase stalling after lesion recognition ( Li et al . , 2015 ) , suggest its localization to various , often distant sites on TFIIH , or the repair bubble in general ( Sugitani et al . , 2016 ) . In summary , our structure of the human TFIIH core complex reveals the interactions that govern the architecture and function of this molecular machine , provides new insights into the regulation of its enzymatic subunits , and thus constitutes an excellent framework for further mechanistic studies of TFIIH in the context of larger DNA repair and transcription assemblies .
TFIIH was purified and cryo-EM grids were prepared on carbon-coated C-flat CF 4/2 holey carbon grids ( Protochips ) using a Thermo Fisher Scientific Vitrobot Mk . IV , as previously described ( Greber et al . , 2017 ) . To improve on our previous 4 . 4 Å cryo-EM map of human TFIIH ( Greber et al . , 2017 ) , which was based on four cryo-EM datasets ( 3 of which were retained in the 4 . 4 Å reconstruction , datasets 8–10 in Supplementary file 1 ) from a low-base Titan microscope ( Thermo Fisher Scientific ) equipped with a side-entry holder ( Gatan ) and a K2 Summit direct electron detector ( Gatan ) , we collected new data ( dataset seven in Supplementary file 1 ) on a Titan KRIOS microscope ( Thermo Fisher Scientific ) operated at 300 kV extraction voltage and equipped with a CS-corrector , a K2 Summit direct electron detector ( Gatan ) operated in super-resolution counting mode , and a Quantum energy filter ( Gatan ) . This dataset was collected under the same imaging conditions as our previous data ( i . e . 37 , 879 x magnification resulting in 1 . 32 Å pixel size , and at a total exposure of 40 e- Å−2 ) , except for the change of microscope . Datasets 7–10 could be combined to yield a cryo-EM map at 4 . 3 Å resolution ( not shown ) , however , this did not lead to a substantial improvement in map quality , suggesting that particle alignment quality was limiting . We therefore opted to collect further data on a Titan KRIOS electron microscope ( Thermo Fisher Scientific ) operated at 300 kV acceleration voltage and equipped with a Volta Phase Plate ( VPP ) , a Gatan Quantum energy filter ( operated at 20 eV slit width ) , and a Gatan K2 Summit direct electron detector ( operated in super-resolution counting mode ) . VPP data ( datasets 1–6 in Supplementary file 1 ) were collected according to the defocus acquisition technique ( Danev and Baumeister , 2017; Khoshouei et al . , 2017 ) at 43 , 478 x magnification , resulting in a physical pixel size of 1 . 15 Å on the object scale , with a total electron exposure of 50 e- Å−2 at an exposure rate of 6 . 1 e- Å−2 s−1 during an exposure time of 8 . 25 s , dose fractionated into 33 movie frames ( 50 frames for dataset 6 ) . Data collection was monitored on-the-fly using FOCUS ( Biyani et al . , 2017 ) to ensure proper evolution of the VPP-induced phase shift . Initially , we used data collected in 10 microscopy sessions , six sessions using the VPP and four sessions without the VPP , resulting in >30’000 total micrographs , of which approx . 16 , 000 were retained after inspection of the quality of Thon rings and CTF fitting ( for details , see Supplementary file 1 ) . Movie stacks were aligned and dose weighed using MOTIONCOR2 ( Zheng et al . , 2017 ) . The aligned , dose weighed sums from the datasets collected at 1 . 32 Å pixel size ( datasets 7–10 ) were up-sampled to 1 . 15 Å per pixel to match the scale of the micrographs collected using the VPP ( datasets 1–6 ) after calibrating the two magnifications to each other based on 3D reconstructions computed from the two types of data . CTF parameters were estimated using GCTF ( Zhang , 2016 ) and particles were picked using GAUTOMATCH ( Kai Zhang , MRC Laboratory of Molecular Biology , Cambridge UK ) or RELION ( Scheres , 2015 ) using templates generated from a preliminary run without reference templates . All subsequent data processing was performed in RELION 2 ( Kimanius et al . , 2016; Scheres , 2012 ) or RELION 3 ( Nakane et al . , 2018; Zivanov et al . , 2018 ) . To remove false positive particle picks and broken particles , an initial 3D classification at low resolution ( 7 . 5° angular sampling ) was performed on each dataset individually ( datasets 3 , 4 , 5 ) , or on a few pooled datasets if appropriate ( datasets 1 and 2 were pooled as they used the same batch of specimen; the non-VPP datasets 7–10 were joined because only few micrographs were retained due to more stringent quality criteria compared to our previous study; and dataset six was initially classified together with particles from dataset four to compensate for particle orientation bias in dataset 6 , see Figure 1—figure supplement 1 ) . In summary , a total of >2 , 000 , 000 initial particle picks were subjected to this initial low-resolution 3D classification , identifying approx . 820 , 000 intact particles that were subjected to further processing . After 3D auto-refinement and another round of 3D classification , performed separately for the VPP and non-VPP data because the two data types were spuriously separated into distinct classes in combined RELION 3D classification runs , the best classes ( one from VPP and non-VPP data each ) resulting from the high-resolution 3D classifications were refined according to the gold-standard refinement procedure ( fully independent half-sets ) , resulting in a 3 . 9 Å-resolution reconstruction according to the FSC = 0 . 143 criterion ( Rosenthal and Henderson , 2003; Scheres and Chen , 2012 ) . Beam tilt refinement in RELION 3 ( Zivanov et al . , 2018 ) improved the map computed from the final subset of VPP data ( 138 , 659 particle images ) to 3 . 7 Å resolution . The non-VPP data no longer improved the reconstruction after beam tilt correction and was therefore discarded at this point . The final map was post-processed by application of a B-factor of −142 A2 and low-pass filtration to the nominal 3 . 7 Å resolution for visualization and later coordinate refinement . We note that even though the final reconstruction comprises only a relatively small fraction of the total particle picks , the first 3D refinement from 786 , 755 VPP particle images ( Figure 1—figure supplement 1 ) resulted in a 4 . 3 Å-resolution map that is in excellent agreement with the final map , except for lower resolution and worse map quality caused by residual heterogeneity that was addressed in the subsequent 3D classification step to yield the final set of 138 , 659 particle images . Therefore , we conclude that our final reconstruction is representative of the overall particle population in the dataset . To facilitate the interpretation of less ordered or only partially occupied parts of the structure , including the p62 BSD2 domain , the MAT1 RING domain , the MAT1 three-helix bundle at the XPD arch domain , and the N-terminus of XPD , we used signal subtracted classification ( Bai et al . , 2015; Nguyen et al . , 2015 ) ( p62 BSD2 domain , Figure 1—figure supplement 3 ) , focused classification ( MAT1 RING domain , Figure 1—figure supplement 4 ) , and multibody refinement ( Nakane et al . , 2018 ) ( MAT1 three-helix bundle and XPD N-terminus , Figure 1—figure supplement 5 ) . For these classification procedures , we used only the VPP data because 3D classification separated VPP and conventional cryo-EM data into distinct classes , rendering combined classification ineffective . Multibody refinement led to only a slight improvement in resolution for the XPD-MAT1 body ( to 3 . 6 Å ) relative to the overall refined best map , and only during the first two iterations , likely due to the relatively small size of the individual bodies and the resulting limited signal for alignment . However , the above-mentioned structural elements showed improved density features ( Figure 1—figure supplement 5B ) and could be more reliably interpreted in the multibody-refined XPD-MAT1 map ( green in Figure 1—figure supplement 5; also used in Figure 1—figure supplement 2E , F ) . Overall , the use of VPP data in this work resulted in substantial improvements both in nominal resolution ( Figure 1—figure supplement 2D ) and map quality ( Figure 1—figure supplement 2E–G ) compared to our previous 4 . 4 Å-resolution structure . The previous structure of the human TFIIH core complex ( Greber et al . , 2017 ) and of yeast TFIIH in the Pol II-PIC ( Schilbach et al . , 2017 ) were docked into the cryo-EM map and used as the basis for atomic modeling in O ( Jones et al . , 1991 ) and COOT ( Emsley et al . , 2010 ) . In addition to these models , the structure of the human p34 VWFA-p44 RING domain complex ( Radu et al . , 2017 ) , the N-terminal RING domain of MAT1 ( Gervais et al . , 2001 ) , the C-terminal RecA-like domain of human XPB ( Hilario et al . , 2013 ) and several homology models for the p52 winged-helix domains generated using the PHYRE2 web server ( Kelley et al . , 2015 ) based on templates PDB IDs 3F6O and 1STZ ( Liu et al . , 2005 ) were used for model building . For the analysis of conformational dynamics of TFIIH , VPP datasets 1 and 2 were subjected to multibody refinement in RELION 3 ( Nakane et al . , 2018 ) using six masks ( Figure 4—figure supplement 1 ) . After completion of multi-body refinement , we used RELION three to run a principal component analysis to identify the principal modes of motion of the bodies relative to each other ( Nakane et al . , 2018 ) . The volume series for the first 12 principal components were reconstructed and difference densities ( green and purple in Figure 4—figure supplement 1 ) were computed between the most extreme states in each series and are shown in Figure 4—figure supplement 1 . Subsequently , roughly 20 , 000 particles corresponding to both ends of the distribution were used for selected principal components and subjected to 3D refinement , resulting in maps of approx . 10 Å resolution . It is important to note that the particles used for these refinements , and the subsequent analysis shown in Figure 4A , were un-subtracted original particle images containing the entire TFIIH . These refinements are therefore not directly affected by any limitations on alignment accuracy that would arise from alignment of smaller sub-volumes of TFIIH . We also repeated this analysis for two different data subsets ( the final 138 , 659 particle-subset that gave rise to the 3 . 7 Å-resolution reconstruction and the complete set of 786 , 755 VPP particles resulting from the initial 3D classification ) using only three bodies for multibody refinement ( providing more signal for alignment per body ) and obtained consistent results overall , with the exception that the ranking of the principal components changed in some instances . The refined atomic model of the TFIIH core complex , subdivided into suitable rigid bodies , was then rigid-body refined into these volumes using PHENIX real space refinement ( Afonine et al . , 2018 ) and coordinate displacement between the two resulting models for each principal component was plotted to obtain an initial assessment of the modes of motion present in the TFIIH dataset ( Figure 4A ) . For actual structural interpretation ( Figure 4B , C ) , the final cryo-EM maps of the TFIIH core complex ( this work ) and TFIIH in the context of the Pol II-PIC ( Schilbach et al . , 2017 ) were used . Figures were created using PyMol ( The PyMOL Molecular Graphics System , Version 1 . 8 Schrödinger , LLC . ) and the UCSF Chimera package from the Computer Graphics Laboratory , University of California , San Francisco ( supported by NIH P41 RR-01081 ) ( Pettersen et al . , 2004 ) . Protein-protein interface statistics were determined using PISA ( Krissinel and Henrick , 2007 ) . Multiple sequence alignments were performed with Clustal Omega ( Sievers et al . , 2011 ) . The cryo-EM map of the human TFIIH core complex at 3 . 7 Å and the refined coordinate model have been deposited to the EMDB and PDB with accession codes EMD-0452 and PDB-6NMI , respectively . Additional cryo-EM maps resulting from the classification of the dataset for presence of the MAT1 RING domain and for the p62 BSD2 domain ( both presence and absence ) have been deposited to the EMDB with accession codes EMD-0587 , EMD-0589 , and EMD-0588 , respectively . The multibody-refined maps for XPD-MAT1 , XPB-p8-p52 ( clutch , CTD ) , and p44-p34-p62-p52 ( N-terminal region ) have been deposited with accession codes EMD-0602 , EMD-0603 , and EMD-0604 , respectively . | The DNA inside a cell carries the instructions it needs to survive . Living cells use many different proteins to read and maintain this store of information . For example , a group of ten proteins collectively called TFIIH is often involved in both reading and repairing the DNA . Proteins in the TFIIH complex include p52 , p62 , XPB and XPD . Understanding the structure of the proteins in TFIIH could reveal much about how it works and how changes to its structure contribute to various medical conditions . Yet TFIIH is a dynamic assembly of molecules and includes many proteins , which makes examining its structure challenging . An ideal protein structure should provide an accurate map of the positions of all the atoms in a protein . Previously , it has not been possible to get this level of detail for TFIIH . Greber et al . used an approach called cryo-electron microscopy ( also called cryo-EM ) to reveal the structure of TFIIH collected from human cells . The structure revealed several new details , including how p52 helps XPB attach to the rest of TFIIH , and that p62 helps to control the activity of XPD . With such a detailed structure , Greber et al . could link changes in TFIIH that are seen in different human diseases to specific parts of the complex . Examining the atomic details of proteins can reveal a lot about how they work and the changes that occur during different diseases . These structures can also help to reveal aspects of how DNA is read and repaired , and may help to design new approaches to treat diseases in the future . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2019 | The complete structure of the human TFIIH core complex |
Neither the disease mechanism nor treatments for COVID-19 are currently known . Here , we present a novel molecular mechanism for COVID-19 that provides therapeutic intervention points that can be addressed with existing FDA-approved pharmaceuticals . The entry point for the virus is ACE2 , which is a component of the counteracting hypotensive axis of RAS . Bradykinin is a potent part of the vasopressor system that induces hypotension and vasodilation and is degraded by ACE and enhanced by the angiotensin1-9 produced by ACE2 . Here , we perform a new analysis on gene expression data from cells in bronchoalveolar lavage fluid ( BALF ) from COVID-19 patients that were used to sequence the virus . Comparison with BALF from controls identifies a critical imbalance in RAS represented by decreased expression of ACE in combination with increases in ACE2 , renin , angiotensin , key RAS receptors , kinogen and many kallikrein enzymes that activate it , and both bradykinin receptors . This very atypical pattern of the RAS is predicted to elevate bradykinin levels in multiple tissues and systems that will likely cause increases in vascular dilation , vascular permeability and hypotension . These bradykinin-driven outcomes explain many of the symptoms being observed in COVID-19 .
The COVID-19 beta-coronavirus epidemic that originated in Wuhan , China in December of 2019 is now a global pandemic and is having devastating societal and economic impacts . The increasing frequency of the emergence of zoonotic viruses such as Ebola , Severe Acute Respiratory Syndrome ( SARS ) , and Middle East Respiratory Syndrome ( MERS ) ( among others ) are of grave concern because of their high mortality rate ( 10%–90% ) . Fortunately , successful containment of those pathogens prevented global-scale deaths . In contrast , the current estimates of mortality for COVID-19 are much lower ( ~4% ) , but the virus has now infected more than nine million people and caused nearly a half a million deaths . The cause of mortality appears to be heterogeneous and although it typically targets older individuals , younger individuals are also at risk . A key to combating the pandemic is to understand the molecular basis of COVID-19 that may lead to effective treatments . Paradoxically , an opportunity that was unavailable with SARS , MERS or Ebola has arisen because of the intense , globally distributed focus of medical and scientific professionals on COVID-19 that is providing a wealth of highly diverse information and data types . Nine bronchoalveolar lavage ( BAL ) samples were originally collected from patients in Wuhan China for RNA sequencing in order to determine the etiological agent for COVID-19 and resulted in the sequence of the first SARS-CoV-2 viral genome . However , the human reads from these samples were discarded 3 . Here , we analyze the human RNA-seq data from these BAL samples alongside 40 controls .
Although pre-existing hypertension is a reported comorbidity for COVID-19 , recent reports indicate hypotension is highly associated with COVID-19 patients once in the hospital ( Rentsch , 2020 ) . The RAS is an important pathway linked to these conditions because it maintains a balance of fluid volume and pressure using several cleavage products of the peptide angiotensin ( AGT ) and their receptors ( Arendse et al . , 2019 , Flores-Muñoz et al . , 2011 , Carey , 2017 ) . The most well-studied peptide is angiotensin II ( Ang II ) , which typically generates vasoconstriction and sodium retention via the AGTR1 receptor and vasodilation and natriuresis when binding to the AGTR2 receptor . The RAS also includes several other lesser known peptides that are highly important; Ang1-7 binds to the MAS1 receptor , generating anti-inflammatory and vasodilatory effects , and Ang1-9 binds to the AGTR2 receptor . Ang II is produced by the enzyme ACE whereas Ang1-7 is generated by the combination of ACE and ACE2 activity and Ang1-9 by ACE2 alone . It is important , therefore , to consider all of these components in the context of the others and not any one in isolation . ACE2 is also the main receptor for the SARS-CoV-2 virus and is not highly expressed in normal lung tissue based on the Genotype-Tissue Expression ( GTEx , gtexportal . org ) six population . However , results from our differential gene expression analysis of RAS genes in cells taken from BAL samples from individuals presenting with severe symptoms of COVID-19 ( Zhou et al . , 2020 ) demonstrates upregulation of ACE2 ( 199 fold ) and disruption of this system compared to controls . In the COVID-19 samples , AGT ( 34 fold ) and the enzyme that activates it ( REN , 380 fold ) are increased compared to controls whereas the enzymes that produce most of the cleavage products , including ACE ( −8 fold ) , are downregulated , which will likely result in a shift of the entire RAS to produce Ang1-9 . In addition , the AGTR1 ( 430 fold ) and AGTR2 ( 177 fold ) receptors are upregulated in BAL COVID-19 samples . Given the central role that the angiotensin and bradykinin ( BK ) peptides play in COVID-19 based on our gene expression analysis from BAL samples , we next focused on the RAS- and BK-related gene pathways in lung tissue from the GTEx population; specifically , the networks of genes that are correlated and ani-correlated with the expression of the angiotensin receptors AGTR2 and AGTR1 . This subset of genes was annotated with functional information and cell type involvement which resulted in a network ( Figure 1 ) that , as would be expected , demonstrates their extensive involvement in arterial and vascular resistance and blood flow via microvascular dilation , flow , and fluid balance . The genes on the left side of the network are extensively involved in vasoconstriction and contain , among others , ACE , AGTR1 , BDKR2 , Nitric Oxide Synthase-1 , and −2 ( NOS1 and NOS2 ) . The right side of the network is extensively involved in decreased arteriolar resistance ( vasodilation ) , increased vascular permeabilization , and altered fluid balance and includes , among other genes , ACE2 , AGTR2 , and the Vitamin D Receptor ( VDR ) . Surprisingly , we find that both sides of the network are also clearly involved in immune system modulation . Although not as widely discussed as angiotensin , BK is another potent regulator of blood pressure and can be considered essentially an extension of the RAS ( Schmaier , 2002 ) . Briefly , similar to the angiotensin peptides , BK is produced from an inactive pre-protein kininogen ( either circulating - HMWK or tissue - LWMK ) through activation by the serine protease kallikrein ( Figure 2A ) . Kallikrein is represented by a cluster of serine proteases ( KLK1-KLK15 ) on chromosome 19 with different tissue distributions; KLKB1 ( on chromosome 4 ) is normally expressed in the pancreas and is responsible for circulating ( plasma ) kallikrein . These proteases are inactivated by zinc and several are known co-receptors for viruses including influenza ( Kalinska et al . , 2016 ) . KLKB1 is activated by FXII of the intrinsic coagulation pathway , which is normally kept in check by the C1-Inhibitor encoded by SERPING1 ( Figure 2A ) . This has the vital ancillary effect of inhibiting the feedback loop of FXII activation by kallikrein ( Kaplan and Ghebrehiwet , 2010 ) . Similar to AGTR2 stimulation , BK induces vasodilation , natriuresis , and hypotension upon activation of the BDKRB2 receptor . BK is tightly integrated with the RAS as BK receptor signaling is augmented by Ang1-9 , likely by resensitization of the BDKRB2 receptor ( Chen et al . , 2005; Marcic et al . , 1999; Erdös et al . , 2002 ) and also because ACE degrades and inactivates BK . Interestingly , ACE has a higher affinity for BK than it does for AGT ( Cyr et al . , 2001 ) and therefore under conditions where ACE is low , the vasopressor system is tilted toward a BK-directed hypotensive axis ( Figure 2A ) . In addition to its role in pressure and fluid homeostasis , BK is a normal part of the inflammatory response after injury and acts to induce pain via stimulation of the BDKRB1 receptor by BK1-8 ( Jacox et al . , 2014 ) , which also causes neutrophil recruitment and increases in vascular permeability ( Stuardo et al . , 2004; Araújo et al . , 2001; Hofman et al . , 2016; Figure 2B ) . BK1-8 is produced by the enzyme carboxypeptidase N ( CPN1 671 fold ) acting on BK . As with the RAS , the BK system is also severely affected in the COVID-19 BAL samples . The expression of the BK precursor kininogen and nearly all of the kallikreins are undetected in controls but expressed in COVID-19 BAL ( Figure 2A ) . Most of the enzymes that degrade BK , including ACE , are downregulated ( −8 fold ) in COVID-19 BAL compared to controls , directing BK1-9 and BK1-8 to the upregulated receptors BKB2R ( 207 fold ) and BKB1R ( 2945 fold ) , respectively . Of note , the pain-receptor BKB1R is normally tightly controlled and expressed only at very low levels in nearly all tissues in GTEx , but in the case of COVID-19 BAL , both BK receptors are expressed whereas they are virtually undetected in controls . Finally , F12 is unchanged but the SERPING1 ( −33 fold ) gene that encodes the C1-Inhibitor that inhibits FXII is highly down-regulated , which would result in even further increases in BK in COVID-19 patients given its role in KLKB1 activation ( Schmaier , 2016 ) . As described below , the resulting Bradykinin Storm is likely responsible for most of the observed COVID-19 symptoms . Hyaluronic acid ( HA ) is a polysaccharide found in most connective tissues . HA can trap roughly 1000 times its weight in water ( Cowman and Matsuoka , 2005 ) and when bound to water the resulting hydrogel obtains a stiff viscous quality similar to ‘Jello’ ( Necas et al . , 2008 ) . HAS1 , HAS2 and HAS3 are genes that encode hyaluronan synthases which are integral membrane proteins responsible for HA production ( Necas et al . , 2008 ) . HA is degraded by hyaluronidases encoded by HYAL1 and HYAL2 . Proteins encoded by other genes in this family ( HYAL3 and HYAL4 ) do not appear to have a hyaluronidase activity ( Harada and Takahashi , 2007; Kaneiwa et al . , 2010 ) . HYAL1 encodes a lysosomal hyaluronidase ( Hyal-1 ) active at low pH and is responsible for intracellular degradation of HA ( Harada and Takahashi , 2007 ) . HYAL2 encodes a membrane-bound hyaluronidase ( Hyal-2 ) responsible for extracellular degradation of HA ( Harada and Takahashi , 2007 ) . Both Hyal-1 and Hyal-2 are dependent on CD44 ( an HA receptor ) for activity ( Harada and Takahashi , 2007 ) . As with the RAS and BK systems , the genes encoding HA synthesis and degradation are also severely affected in the COVID-19 BAL samples . There is significant upregulation of the genes involved in HA synthesis: HAS1 ( 9113 fold ) , HAS2 ( 493 fold ) , and HAS3 ( 32 fold ) . The CD44 gene that encodes the HA receptor required for HA degradation and the gene encoding extracellular hyaluronidase HYAL2 are both downregulated ( −11 and −5 fold respectively ) in the BAL fluid of COVID-19 patients . HYAL1 is not expressed in the BAL fluid of controls or the COVID-19 patients . The result of these shifts in expression would be likely to cause an increase in the amount of HA in the bronchoalveolar space of the lungs which , combined with the vascular hyperpermeability caused by BK , could form a viscous hydrogel that would negatively impact gas exchange ( Figure 3 ) . In fact , HA in BAL fluid has previously been associated with acute respiratory distress syndrome ( ARDS ) where there was a significant anticorrelation between the concentration of HA and the pulmonary oxygenation index ( Modig and Hällgren , 1989; Hällgren et al . , 1989 ) . HA has also been associated with pulmonary thrombosis and/or ground glass opacities in radiological findings ( Bhagat et al . , 2012; Han et al . , 2019; Jang et al . , 2014 ) . Although not the focus of the present study , coagulopathy is commonly reported in cases of COVID- 19 ( The Lancet Haematology , 2020 ) , and there are suggestions in the literature of links between RAS and coagulopathy . The Ang1-9 peptide that is increased in COVID-19 BAL stimulates thrombosis by inhibiting fibrinolysis ( Mogielnicki et al . , 2014 ) . In addition to BK , ACE also degrades the antifibrotic peptide N-acetyl-seryl-aspartyl-lysyl-proline ( AcSDKP ) , which is produced from thymosin beta-4 ( TMSB4X , −130 fold ) ( Kanasaki , 2020 ) . Increased fibrinolysis could therefore be achieved by increasing ACE , or by administering thymosin beta-4 , which is currently in clinical trials for the treatment of cardiovascular disorders ( Timbetasin ) . If TMSB4X is , in fact , protective , it could explain the lower incidence of COVID-19 induced mortality in women ( Jin et al . , 2020 ) because it is found on the X chromosome and escapes X-inactivation . Women therefore would have twice the levels of this protein than men , which is supported by our BAL analysis ( −207 fold in males , −131 fold in females ) . In addition , both the RAS and BK pathways have previously been tied to HA . It was found that Angiotensin II increased CD44 expression and hyaluronidase activity ( Bai et al . , 2016 ) . As discussed above , COVID-19 likely significantly downregulates the production of Angiotensin II which is consistent with the decrease in CD44 expression that is seen in the BAL fluid of SARS-CoV-2 infected patients . Furthermore , IL2 was recently reported to be highly upregulated in symptomatic but not asymptomatic COVID-19 patients ( Long et al . , 2020; Paegelow et al . , 1995; Mustafa et al . , 2002 ) and is upregulated ( 21 fold ) in the BAL samples compared to controls . This cytokine is induced by BK in the lung , and causes vascular leakage syndrome ( VLS ) , which appears to be mediated through CD44 . Interestingly , CD44 knockout mice displayed reduced IL2-induced VLS , suggesting this may be a valuable target for COVID-19 intervention . According to the CDC , the majority of SARS-CoV-2 infections are asymptomatic or mild . Those that proceed to more severe forms present with fever , a non-productive cough that may result in hemoptysis and shortness of breath . Other common symptoms are myalgia , fatigue , sore throat , nausea , vomiting , diarrhea , conjunctivitis , anorexia , and headache ( cdc . gov/coronavirus/2019-ncov/hcp/clinical-guidance-management-patients . html ) . Reports from blood studies include leukopenia , eosinopenia , neutrophilia , elevated liver enzymes , C-reactive protein , and ferritin ( Fan et al . , 2020; Huang et al . , 2020; Goyal et al . , 2020 ) . Furthermore , autopsies have reported extensive hyaline membrane formation in the lungs of COVID-19 patients ( Barton et al . , 2020; Xu et al . , 2020; Adachi et al . , 2020; Mong et al . , 2020 ) . Specifically , histological analysis of the lungs of a deceased COVID-19 patient showed organizing hyaline membranes in the early stages of alveolar lesions and prominent hyaline membranes in the exudative phase of diffuse alveolar damage ( Adachi et al . , 2020 ) . In a seperate post mortem study of lung tissue from COVID-19 patients , microscopic examination found ‘numerous hyaline membranes without evidence of interstitial organization’ ( Barton et al . , 2020 ) . Furthermore , in another autopsy study of a COVID-19 patient , histological analysis found extensive hyaline membranes , which the authors interpreted as indicative of ARDS ( Xu et al . , 2020 ) . Finally , a meta-analysis showed that there was a statistically significant 4 . 6 fold difference in lung weight of COVID-19 patients versus controls , which they conclude is consistent with the HA-hydrogel formation known to occur in ARDS ( Mong et al . , 2020 ) . Although much focus has been on the lung due to the need for ventilator support of end-stage disease , COVID-19 also affects the intestine , liver , kidney , heart , brain , and eyes ( Wadman , 2020 ) . Nearly one-fifth of hospitalized patients experience cardiac injury ( Shi et al . , 2020 ) , many of whom have had no history of cardiovascular problems prior to infection . Responses include acute myocardial injury , myocarditis , and arrhythmias ( Driggin et al . , 2020 ) that may be due to viral infection directly , which is consistent with high expression of the SARS-CoV-2 receptor ACE2 in cardiac tissue ( gtexporta . org ) . An important extension of the RAS in controlling cardiac contraction and blood pressure is the potent inotrope apelin ( APLN ) , which acts as an NO-dependent vasodilator when its receptor ( APLNR ) heterodimerizes with BDKRB1 ( Bai et al . , 2014 ) . APLN ( 98 fold ) , APLNR ( 3190 fold ) and BDKRB1 ( 2945 fold ) are all upregulated in COVID-19 BAL . As with BK and ANG derived peptides , APLN is inactivated by Neprilysin ( MME ) , which is significantly downregulated in the BAL samples from COVID-19 individuals ( −16 fold ) . Therefore , increased APLN-signaling can be added to the imbalanced RAS . In addition to cardiac dysfunction , neurological involvement in COVID-19 was revealed after an MRI assessment of COVID-19-positive patients with encephalopathy symptoms in France identified enhancement in leptomeningeal spaces and bilateral frontotemporal hypoperfusion ( Helms et al . , 2020 ) which are consistent with increased vascular permeabilization in the brain . Furthermore , earlier reports from China indicate high frequencies of dizziness , headache , as well as taste and smell impairment ( Mao et al . , 2020 ) . The most recent reports from the United States and China indicate that 30–50% of COVID-19 patients experience adverse gastrointestinal symptoms ( Cholankeril , 2020; Pan et al . , 2020 ) . Direct infection by the virus and damage to the kidney was also observed , specifically in the proximal tubules ( Su et al . , 2020 ) . These latter two findings are not surprising given the higher expression of ACE2 in these tissues compared to tissues overall ( gtexportal . org ) , which would facilitate infection by the virus . Finally , COVID-19 patients also frequently display skin rashes including ‘covid-toe’ that appear to be related to dysfunction of the underlying vasculature . Based on previous work in SARS-CoV-1 and SARS-CoV-2 , it is likely that this new coronavirus enters host cells in nasal passages where the receptor ACE2 is moderately expressed . Migration to throat tissues and passage through the stomach is then possible given that SARS-CoV-2 can survive the extreme pH of the gastric tissues ( Chin et al . , 2020 ) and infection could then expand into the intestines where ACE2 levels are high ( GTEx Consortium , 2013 ) . Initial infection might not occur in the lung epithelium given that ACE2 is undetectable or expressed at extremely low levels there ( GTEx Consortium , 2013 ) . Following infection , the single polypeptide that is translated from the virus’ positive-strand RNA genome is cleaved into active proteins by the non-structural protein 3CLpro protease . This protein is then repurposed by the virus to inactivate the host cells’ first line of defense , interferon , most likely by degrading the NFkappaB activating factor IKK-gamma as has been shown to happen in the porcine coronavirus PEDV ( Wang et al . , 2016 ) . Aside from self-protection , the suppression of NFkappaB ( −9 fold reduced in BAL samples ) directly affects the RAS as NFkappaB normally induces the expression of ACE by binding to its promoter and increasing transcription ( Garcia et al . , 2016; Figure 2A ) . This likely relates to the role of ACE in the innate immune response that is independent of its actions on the vascular system ( Bernstein et al . , 2018 ) . The virus therefore acts pharmacologically as an ACE inhibitor by reducing its RNA expression more than 10-fold , which is supported by our BAL RNA-seq analysis . Additionally , ACE2 expression is normally downregulated in-part by Ang II ( Patel et al . , 2016 ) . As Ang II is the catalytic product of ACE , it would seem that the virus’s ability to decrease ACE expression would have the effect of upregulating ACE2 ( 199 fold in our BAL analysis ) . In some patients , severe pulmonary involvement could occur when the virus is introduced into the intestinal lymph vessels and moves up the lymphatic system ( Chen , 2020 ) , enters the bloodstream at the thoracic duct and moves through the heart and into the lung microvasculature where it could attack cells in the lungs that now express ACE2 due to virus-induced upregulation . Given that the high levels of ACE in the vascular bed of the lung are the major producer for circulating angiotensin-derived peptides ( Studdy et al . , 1983 ) , establishment of SARS-CoV-2 in the lung will have profound effects . Downregulation of ACE here ( confirmed in BAL samples from COVID-19 patients ) will result in the diversion of the RAS to produce the BK-augmenting peptide Ang1-9 , exacerbating BK-effects , such as pain sensitization and increased vascular permeability on a system-wide level . Expansion of this imbalance as described above ( Figure 2 ) , increases levels of BK and will result in increased vascular permeability in tissues that have been infected by SARS-CoV-2 and be most severe in those that are normally regulated by ACE . ACE may also provide a key diagnostic point as half of the variation amongst individuals can be explained by an insertion/deletion polymorphism of the gene ( Rigat et al . , 1990 ) . As mentioned above , the combination of vascular permeability and HA build up in the lungs could produce a hydrogel that significantly inhibits gas exchange in bronchoalveolar spaces . This is consistent with the autopsy reports of hyaline membranes in the lungs of deceased COVID-19 patients as well as other acute respiratory distress conditions ( e . g . , SARS , MERS , ARDS ) ( Barton et al . , 2020; Xu et al . , 2020; Adachi et al . , 2020 ) Although this likely represents a late-stage event in severe cases of COVID-19 , if the cause is overproduction of HA as a result of disruption of the RAS , it is also a potentially valuable intervention point because the condition is easily identified , and treatment could have rapid and significant beneficial effects . In addition , increased levels of the vasodilating peptide APLN that are produced in COVID-19 patients could have spillover effects on cardiac function . APLN upregulates the expression of ACE2 ( Sato et al . , 2013 ) and directly affects cardiac contraction and vasodilation . Increased levels of APLN are known to be associated with cardiac arrhythmia ( Salska et al . , 2018 ) and in the case of hyper-stimulated BK output , could be causing cardiac events in COVID-19 patients . In addition , increased levels of APLN could lead to more ACE2 receptors for SARS-CoV-2 in the heart and thus stimulate further infection . Furthermore , excess BK can lead to hypokalemia ( Zhang et al . , 2018 ) , which is associated with arrhythmia and sudden cardiac death ( Kjeldsen , 2010 ) , ( Bielecka-Dabrowa et al . , 2012; Skogestad and Aronsen , 2018 ) , both of which have been reported in COVID-19 patients ( Huang et al . , 2020; Guo et al . , 2020 ) , ( Wang et al . , 2020 ) ; a recent report confirms that hypokalemia is occurring in severe cases of COVID-19 ( Lippi et al . , 2020 ) . It is also notable that many of the other symptoms being reported for COVID-19 ( myalgia , fatigue , nausea , vomiting , diarrhea , anorexia , headaches , decreased cognitive function ) are remarkably similar to other hyper-BK-conditions that lead to vascular hyper-permeabilization such as angioedema as was recently noted ( van de Veerdonk et al . , 2020 ) . In agreement with that report , our results indicate that the pathology of COVID-19 is likely the result of Bradykinin Storms rather than cytokine storms ( although given the induction of IL2 by BK , the two may be intricately linked ) . This model predicted that a loss of ACE2 would exacerbate the BK-induced pathogenesis ( van de Veerdonk et al . , 2020 ) . However , the BAL fluid expression data indicate that the Bradykinin Storm is instead caused by upregulation of ACE2 and reduced degradation of BK by ACE . Based on this data-driven model , an individual’s symptomatology is likely directly related to the specific tissue distribution of viral infection around the body ( Figure 4 ) and should be viewed in the context of an overactive bradykinin response . The majority of circulating BK is degraded in the lungs by ACE and therefore heterogeneous symptoms of COVID-19 could also be the result of systemic effects of increased levels of circulating bradykinin and the eight-fold reduction of ACE in the lung microvasculature that would normally degrade it . Given this model , factors that affect RAS balance should be further investigated in the framework of diagnosis and treatment . For example , another well-documented regulator of RAS is Vitamin D ( Vaidya and Williams , 2012 ) as the liganded Vitamin D receptor ( VDR ) suppresses REN expression . Patients who are deficient in Vitamin D are at-risk for ARDS in general ( Dancer et al . , 2015 ) and Vitamin D deficiencies have recently been associated with severity of illness in COVID-19 patients ( Alipio , 2020 ) . Our BAL gene expression analysis shows that VDR is 2-fold down-regulated and enzymes [CYP24A1 ( 465 fold ) , CYP3A4 ( 208 fold ) ] that catabolize Vitamin D ( 1 , 25 ( OH ) 2D ) and its precursor ( 25OHD ) ( Bikle , 2014 ) are up-regulated in COVID-19 patients compared to controls , which will likely result in further increases in REN . Furthermore , our analysis of ChipSeq experiments from a VDR study Tuoresmäki et al . , 2014 have determined that , in addition to REN , the following genes in the RAS-Bradykinin system have a VDR binding site within 20 kilobases: BDKRB1 , BDKRB2 , CYP24A1 , DPP4 , IKBKG ( regulates NFkappaB ) , KLK1 , KLK2 , KLK4 , KLK6 , KLK7 , KLK9 , KLK10 , and MME . Six of these binding sites can be tied to the following genes via chromatin structure with the use of H-MAGMA and Hi-C data ( see Materials and methods ) : DPP4 , BDKRB2 , KLK6 , KLK7 , KLK10 , and IKBKG . VDR binds to many sites in the genome with tissue-specific binding patterns so these putative associations to other genes in the RAS and BK pathways will require further investigation . Several interventional points ( most of them already FDA-approved pharmaceuticals ) could be explored with the goal of increasing ACE , decreasing BK , or blocking BK2 receptors ( Table 1 ) . Icatibant is a BKB2R antagonist ( Dubois and Cohen , 2010 ) whereas Ecallantide acts to inhibit KLKB1 , reducing levels of BK production ( Farkas and Varga , 2011 ) . Androgens ( danazol and stanasolol ) increase SERPING1 , although the side effects likely make these undesirable ( Wilkerson , 2012 ) , but recombinant forms of SERPING1 ( Berinert/Cinryze/Haegarda ) could be administered to reduce BK levels . It should be noted that any intervention may need to be timed correctly given that REN levels rise on a diurnal cycle ( Gordon et al . , 1966 ) , peaking at 4AM which corresponds with the commonly reported worsening of COVID-19 symptoms at night . Another approach would be the modulation of REN levels via Vitamin D supplementation when warranted . 4-methylumbelliferone ( Hymecromone ) is a potent inhibitor of HAS1 , HAS2 , and HAS3 gene expression and results in the suppression of the production of hyaluronan in an ARDS model ( McKallip et al . , 2003; McKallip et al . , 2013 ) . Hymecromone ( 4-methylumbelliferone ) is approved for use in Asia and Europe for the treatment of biliary spasm . However , it can cause diarrhea with subsequent hypokalemia , so considerable caution should be used if this were to be tried with COVID-19 patients ( NCATS Inxight , 2020 ) . As mentioned above , Timbetasin may reduce COVID-19 related coagulopathies by increasing fibrinolysis . The testing of any of these pharmaceutical interventions should be done in well-designed clinical trials . Given the likely future outbreaks of zoonotic viruses with a similar outcome , it would be in the best interest long-term to invest in the development of small molecules that can inhibit the virus from replicating or suppressing the host immune system such as a 3CLpro inhibitor . However , to date , no large multi-centered , randomized , placebo controlled , blinded clinical trials have been done with 3CLpro inhibitors ( Sisay , 2020 ) . In the meantime , our analyses suggest that prevention and treatment centered on vascular hyper-permeability and the suppression of hyaluronan may prove beneficial in fighting the pathogenesis of COVID-19 . Given the fact that two recent studies have validated our model’s predictions of hypokalemia ( Lippi et al . , 2020 ) and Vitamin D deficiency ( Alipio , 2020 ) in COVID-19 patients , we suggest that rapid testing of the pharmaceutical interventions discussed above is warranted .
FASTQ files were downloaded from the Sequence Read Archive ( PRJNA605983 and PRJNA434133 , metadata: Supplementary file 1 ) at the NCBI and trimmed using the default parameters in CLC Genomics Workbench ( 20 . 0 . 3 ) . RNA-Seq analysis was performed using the latest version of the human transcriptome ( GRCh38_latest_rna . fna , 160 , 062 transcripts to which we appended the SARS-CoV-2 reference genome , MN908947 ) . Mapping parameters were set with a mismatch cost of two , insertion and deletion cost of three , and both length and similarity fraction were set to 0 . 985 . TPMs were generated for all 160 , 063 transcripts for the nine COVID-19 samples and the 40 controls ( Supplementary file 2 ) . The resulting transcript mappings for genes of interest were manually inspected to account for any expression artifacts , such as reads mapping solely to repetitive elements such as the Alu transposable element or all reads mapping to a UTR or pseudogene therein . Transcripts whose counts came solely from ( or were dominated by ) reads at repetitive elements were removed from the analysis . For the controls cases we ran an outlier analysis using the prcomp function in the R package factoextra . Input data were TPM for transcripts that averaged greater than one across all samples ( 30 , 102 , Supplementary file 2 ) . To test the hypothesis that gene transcripts were differentially expressed in the COVID-19 patients vs controls , the edgeR package was used ( Robinson et al . , 2010; McCarthy et al . , 2012 ) . Briefly , normalization factors were determined and the count data were scaled to account for library size according to the package instructions . Then , dispersion was estimated and a negative binomial model was fit to the read counts for each gene . Genewise tests were then performed to test for differential expression . As described below , manual inspection of isoforms determined when there was isoform switching and differential isoform or gene expression was calculated as described above . The p-values were adjusted for multiple comparisons using the Benjamini-Hochberg method . For each of the differentially expressed genes , we plotted transcript-level TPMs across COVID-19 and control individuals for visual inspection and annotation . We collapsed transcripts to the gene-level if ( 1 ) all but one of the transcripts had low TPM , ( 2 ) different transcripts coded for the same protein , ( 3 ) and none of the transcripts were substantially truncated or otherwise altered in any functional domains . Fold change ratios were calculated with mean TPM values . A geneset from the RAS and BK pathways was extracted from log2 transformed GTEx expression lung data . Pearson correlation values were calculated among these genes and the resulting values clustered using hierarchical and k-means clustering of both genes and samples to identify patterns . Four k-means was sufficient to partition all of the genes . Two of the four clusters were highly anti-correlated: AGTR1 identified one pattern and AGTR2 identified an anti-correlated cluster . One of the two remaining intermediate clusters was partially correlated to and was subsequently merged with the AGTR1 cluster and the other remaining cluster was partially correlated to and therefore was merged with AGTR2 cluster . The resulting two clusters were annotated with functional terms and cell types in order to create the annotation network seen in Figure 1 . AThe chromosomal coordinate for SNPs for each each VDR binding site waswere used to test for chromatin contact with RAS-BK genes of interest in a synthetic GWAS study using H-MAGMA , which allows integration of chromatin interaction data for gene-set analysis ( Sey et al . , 2020 ) Each coordinate SNP within a VDR binding site was assigned a large population size of 500 , 000 and a large p-value of 1 × 103-35 . The gene annotation file mapped all coordinates SNPs to genes that were either in the exonic or promoter region of said gene , or within a related chromatin region in intronic and intergenic regions using Hi-C data from lung tissue ( Schmitt et al . , 2016 ) . With these two data sets , H-MAGMA returned a list of genes in contact with the coordinates of VDR binding sites related to the synthetic GWAS SNPs . H-MAGMA identified six RAS-BK genes of interest , namelyin particular: , DPP4 , BDKRB2 , KLK6 , KLK7 , KLK10 , and IKBKG . | In late 2019 , a new virus named SARS-CoV-2 , which causes a disease in humans called COVID-19 , emerged in China and quickly spread around the world . Many individuals infected with the virus develop only mild , symptoms including a cough , high temperature and loss of sense of smell; while others may develop no symptoms at all . However , some individuals develop much more severe , life-threatening symptoms affecting the lungs and other parts of the body including the heart and brain . SARS-CoV-2 uses a human enzyme called ACE2 like a ‘Trojan Horse’ to sneak into the cells of its host . ACE2 lowers blood pressure in the human body and works against another enzyme known as ACE ( which has the opposite effect ) . Therefore , the body has to balance the levels of ACE and ACE2 to maintain a normal blood pressure . It remains unclear whether SARS-CoV-2 affects how ACE2 and ACE work . When COVID-19 first emerged , a team of researchers in China studied fluid and cells collected from the lungs of patients to help them identify the SARS-CoV-2 virus . Here , Garvin et al . analyzed the data collected in the previous work to investigate whether changes in how the body regulates blood pressure may contribute to the life-threatening symptoms of COVID-19 . The analyses found that SARS-CoV-2 caused the levels of ACE in the lung cells to decrease , while the levels of ACE2 increased . This in turn increased the levels of a molecule known as bradykinin in the cells ( referred to as a ‘Bradykinin Storm’ ) . . Previous studies have shown that bradykinin induces pain and causes blood vessels to expand and become leaky which will lead to swelling and inflammation of the surrounding tissue . In addition , the analyses found that production of a substance called hyaluronic acid was increased and the enzymes that could degrade it greatly decreased . Hyaluronic acid can absorb more than 1 , 000 times its own weight in water to form a hydrogel . The Bradykinin-Storm-induced leakage of fluid into the lungs combined with the excess hyaluronic acid would likely result in a Jello-like substance that is preventing oxygen uptake and carbon dioxide release in the lungs of severely affected COVID-19 patients . Therefore , the findings of Garvin et al . suggest that the Bradykinin Storm may be responsible for the more severe symptoms of COVID-19 . Further experiments identified several existing medicinal drugs that have the potential to be re-purposed to treat the Bradykinin Storm . A possible next step would be to carry out clinical trials to assess how effective these drugs are in treating patients with COVID-19 . In addition , understanding how SARS-Cov-2 affects the body will help researchers and clinicians identify individuals who are most at risk of developing life-threatening symptoms . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"medicine",
"computational",
"and",
"systems",
"biology"
] | 2020 | A mechanistic model and therapeutic interventions for COVID-19 involving a RAS-mediated bradykinin storm |
The cytomatrix at the active zone ( CAZ ) is a macromolecular complex that facilitates the supply of release-ready synaptic vesicles to support neurotransmitter release at synapses . To reveal the dynamics of this supply process in living synapses , we used super-resolution imaging to track single vesicles at voltage-clamped presynaptic terminals of retinal bipolar neurons , whose CAZ contains a specialized structure—the synaptic ribbon—that supports both fast , transient and slow , sustained modes of transmission . We find that the synaptic ribbon serves a dual function as a conduit for diffusion of synaptic vesicles and a platform for vesicles to fuse distal to the plasma membrane itself , via compound fusion . The combination of these functions allows the ribbon-type CAZ to achieve the continuous transmitter release required by synapses of neurons that carry tonic , graded visual signals in the retina .
Neurons communicate at synapses , where chemical neurotransmitter is released from the presynaptic terminal by fusion of transmitter-containing synaptic vesicles with the plasma membrane . A macromolecular complex , the cytomatrix at the active zone ( CAZ; Südhof , 2012 ) is thought to capture and organize synaptic vesicles to support neurotransmitter release by presynaptic terminals ( Sigrist and Schmitz , 2011; Gundelfinger and Fijtova , 2012; Hallermann and Silver , 2013 ) , but the dynamics of this process at living synapses remain largely elusive . In order to follow the trafficking of vesicles in real time at the CAZ , methods are needed to label and image both vesicles and the active zone at the same time in living synapses . To label the living active zone , we used fluorescent Ribeye-binding peptide ( RBP; Zenisek et al . , 2004 ) , which marks the synaptic ribbon found at the active zones of retinal bipolar cells ( BPCs ) and other neurons that release neurotransmitter continuously in response to graded changes in membrane potential ( Matthews and Fuchs , 2010 ) . However , monitoring the trafficking of tiny synaptic vesicles ( 30–50 nm in diameter ) is a technical challenge , which we were able to meet by targeting a photoactivatable fluorescent protein to synaptic vesicles and using super-resolution photoactivated localization microscopy ( Betzig et al . , 2006 ) to track movements of single vesicles in voltage-clamped synaptic terminals . We also targeted the exocytosis reporter pHluorin ( Sankaranarayanan et al . , 2000 ) to synaptic vesicles ( Granseth et al . , 2006; Voglmaier et al . , 2006 ) in order to detect vesicle fusion . These approaches then allowed us to analyze the trafficking and fusion of vesicles at the active zone of BPC ribbon synapses during ongoing neurotransmitter release . Previously , studies of synaptic vesicle trafficking and fusion at ribbon synapses of BPCs have been carried out using total internal reflection fluorescence microscopy ( TIRFM ) to image single vesicles labeled with FM dye ( Zenisek et al . , 2000 , 2002; Holt et al . , 2004; Midorikawa et al . , 2007; Zenisek , 2008; Joselevitch and Zenisek , 2009 ) . Some of the conclusions from this work are: 1 ) vesicles stably associate with ribbons in the absence of stimulation ( ‘residents’ ) , 2 ) these resident vesicles rapidly undergo exocytosis in response to depolarization , and 3 ) new vesicles ( ‘newcomers’ ) appear , move toward the membrane , and fuse during sustained depolarization . Although TIRFM is a powerful approach to study membrane-associated phenomena , it is limited to imaging labeled vesicles within ~100 nm of the plasma membrane ( e . g . , Zenisek et al . , 2000 ) , which is insufficient to provide coverage of all the ribbon-associated vesicles . The method also requires tight adherence of the plasma membrane to a planar substrate , which restricts observations to a small part of the terminal and eliminates the natural membrane curvature in that observable region . Furthermore , to our knowledge , only Midorikawa et al . ( 2007 ) and Zenisek ( 2008 ) have combined vesicle imaging with ribbon labeling , and then only in sequentially acquired images separated by some time , which were intended to test whether observed hotspots of fusion coincided with ribbons . Because of these limitations , we used two-color laser scanning methods that allowed single labeled vesicles to be observed throughout the full extent of the ribbon in voltage-clamped synaptic terminals , while the ribbon and cell border were imaged with a second fluorescent label . Since the positions of the ribbon and a labeled vesicle were known accurately , we were able to detect vesicle movements on the ribbon prior to fusion and determine where vesicles resided on the ribbon when they fused . Therefore , our experiments complement and significantly extend the previous studies based on TIRFM .
What happens to the immobile ribbon-associated vesicles during transmitter release ? To control release , isolated BPCs were voltage-clamped via a whole-cell pipette placed directly on the synaptic terminal ( Figure 1—figure supplement 1B ) and stimulated by brief depolarization from -60 to 0 mV to activate voltage-gated calcium channels ( Figure 2A , inset ) , causing calcium influx that triggered synaptic vesicle fusion . BPCs exhibit two kinetically distinct components of transmitter release during depolarization ( for zebrafish BPCs , see Vaithianathan and Matthews , 2014 ) : a fast component depleted within 10 ms , and a slower component lasting hundreds of ms . We first determined how individual paRFP-labeled vesicles at the ribbon responded when the system was perturbed by releasing just the fast component , which is thought to represent the cohort of vesicles tethered to the ribbon nearest the plasma membrane ( Matthews and Fuchs , 2010 ) , where voltage-gated calcium channels reside . 10 . 7554/eLife . 13245 . 008Figure 2 . Vesicle release , replenishment , and movement elicited by brief depolarization . ( A ) Example of an increase in paRFP fluorescence after a 10-ms depolarization from -60 mV to 0 mV , which evoked the Ca2+ current shown in the inset . Data points show fluorescence for individual frames , and the thick lines show the average ( ± sem ) fluorescence over 8 frames before and after the stimulus . ( B ) An instance where depolarization caused a loss of paRFP fluorescence . ( C ) Representative examples of averaged images after depolarization for trials in which paRFP-labeled vesicles appeared post-stimulus . ( D ) Examples of averaged images before depolarization for trials in which paRFP-labeled vesicles disappeared post-stimulus . ( E ) Loci of paRFP-labeled vesicles with respect to the center of ribbon for 59 disappearances ( green circles ) and 71 appearances ( red circles ) . ( F ) Histogram of displacement amplitude after 10-ms stimulation ( red line ) for 64 paRFP-labeled vesicles that were present before and after depolarization , compared with displacement histogram for 437 paRFP-labeled vesicles in the absence of stimulation ( black line ) . Displacement = ( ΔΔx2 + ΔΔy2 ) 0 . 5 , where ΔΔx and ΔΔy are changes in paRFP locus relative to the center of the ribbon . ( G ) Displacement vectors without stimulation ( N=437 ) , shown on an expanded scale in the inset . The black dot in the center is the average vector . ( H ) Displacement vectors after 10-ms depolarization ( N = 64 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13245 . 008 On some trials after a 10-ms stimulus , a new stable paRFP spot appeared at the ribbon ( Figure 2A , C ) , while on other trials , a previously present paRFP spot disappeared after the stimulus ( Figure 2B , D ) , representing single-vesicle replenishment and loss , respectively . When the loci of gained and lost paRFP spots were determined as illustrated in Figure 1 and plotted with respect to the center of the ribbon ( Figure 2E ) , there was no difference in the distribution of appearances and disappearances along the axis parallel to the membrane ( the y-axis; p = 0 . 4 , Wilcoxon rank test ) . However , disappearances clustered significantly closer to the plasma membrane than appearances ( Figure 2E ) along the axis perpendicular to the membrane ( the x-axis; p<10-8 , Wilcoxon rank test ) . This pattern revealed by single-vesicle imaging is consistent with selective fusion of membrane-proximal synaptic vesicles during brief depolarization , but preferential recruitment of new vesicles at the opposite pole of the ribbon to replace those lost , as suggested previously from population behavior of vesicles labeled with FM dye ( Vaithianathan and Matthews , 2014 ) . A simple interpretation is that loss of membrane-proximal vesicles creates the opportunity for vesicles remaining on the ribbon to rearrange and occupy the vacated region , opening up 'slots' for new vesicles at more distal positions on the ribbon . To determine whether vesicles remaining on the ribbon indeed changed position after depolarization , we measured the relative positions of single paRFP-labeled synaptic vesicles that were present on the ribbon both before and after 10-ms stimuli , or under the same conditions without stimulation . In the absence of stimulation , the position of labeled vesicles on the ribbon was stable , as shown by the histogram of displacement amplitude in Figure 2F . The great majority of such 'displacements' were less than a vesicle diameter and are therefore attributable to noise . However , after 10-ms stimulation , displacements were larger ( Figure 2F; p≈0 , Wilcoxon rank test ) , which is consistent with the prediction that the loss of ribbon-associated vesicles allows movement of remaining vesicles . We next examined the direction of motion by plotting the vesicle displacement as a vector from the starting position . Without stimulation ( Figure 2G ) , vector amplitudes were mostly within the range expected for imaging noise ( 93% were <50 nm; see inset in Figure 2G for an expanded view to better reveal the large number of small vectors ) , and the direction of the vector was equally likely to be toward or away from the membrane ( positive and negative ΔΔx , respectively , indicated by green and red vectors in Figure 2G ) . The average vector amplitude was therefore zero ( black dot at the center of vector clusters ) . Rare larger excursions were also observed , which may reflect a low rate of genuine vesicle movement occurring spontaneously on the ribbon at the holding potential of -60 mV . After 10-ms stimuli ( Figure 2H ) , vector amplitudes were significantly larger ( 77% were >50 nm; p<10-6 , Wilcoxon rank test ) , and there was a significant tendency for movement toward the membrane , with 70% of vectors having a positive ΔΔx ( p=0 . 0008 that + and – ΔΔx were equally likely , by sign test ) . Amplitudes of vectors both toward the membrane ( green vectors in Figure 2H ) and away from the membrane ( red vectors ) were significantly larger after stimulation than in the absence of stimulation ( +ΔΔx: 89 ± 8 nm vs . 17 ± 1 nm , p≈0 , Wilcoxon rank test; -ΔΔx: 59 ± 6 nm vs . 16 ± 1 nm , p<10-12 , Wilcoxon rank test ) . Overall , the average vector after stimulation ( black arrow , Figure 2H ) was a displacement by 33 nm toward the membrane , or approximately one vesicle diameter . Since movements both toward and away from the plasma membrane were enhanced by stimulation , the results suggest a mechanism of passive diffusion of tethered vesicles along the ribbon , such as the recently proposed 'crowd-surfing' model ( Graydon et al . , 2014 ) , rather than a directed motor . In such a passive mechanism , the net movement toward the membrane that we observed after stimulation is due to preferential release of vesicles near the membrane by 10-ms depolarization , creating the opportunity for more distal vesicles to diffuse into the vacated positions . In addition to the rapid burst of release at the onset of depolarization , ribbon synapses also release neurotransmitter continuously during sustained depolarization . To examine vesicle trafficking during this sustained component of release evoked by depolarization from -60 mV to -10 mV for >0 . 5 s , we measured fluorescence along a short scan-line perpendicular to the plasma membrane , which yielded x-t raster plots that were analyzed as shown in Figure 3 to determine the x-axis position of paRFP spots relative to the synaptic ribbon and the plasma membrane ( also see Image analysis section of Materials and methods ) . Although paRFP fluorescence was detectable in individual scan lines ( Figure 3E ) , the noise level precluded accurate localization of the emitter by fitting the x-axis intensity profile of single lines . In the example shown in Figure 3E , for instance , the peak of the fitted Gaussian varied over a range of 170 nm in 16 consecutive individual lines . However , averaging across 4 scan lines reduced the range of variation of fitted positions to 20 nm , as shown in Figure 3F . As a result , we routinely averaged over 4–10 lines , depending on the noise level in a particular experiment , in order to achieve localization along the x-axis with a resolution of approximately one vesicle diameter . After such averaging , the position of a labeled vesicle in the absence of stimulation was stable for many seconds within a range of ± 25 nm ( Figure 3G , H ) , which we take as the spatial precision of the line-scan method . 10 . 7554/eLife . 13245 . 009Figure 3 . Generation and analysis of line scan data . ( A ) Scan lines were positioned perpendicular to the plasma membrane , extending from the intracellular side of the ribbon into the extracellular space . For illustrative purposes , a second scan line is shown at a non-ribbon location . Lower-case letters show line positions for panels B–D . ( B ) Intensity profile of green RBP fluorescence at the non-ribbon location . The parameter x1/2 from the sigmoid fit ( black line ) is taken as the position of the membrane ( see Materials and methods ) . ( C ) Intensity profile of green RBP fluorescence at the ribbon . The parameter x0 is the peak of the Gaussian fit , giving the x-position of the center of the ribbon ( see Materials and methods ) . ( D ) Intensity profile of paRFP fluorescence at the ribbon , with the x-position of the labeled vesicle given by x0 . ( E ) Example of x-t image of paRFP spot at a ribbon , consisting of 64 line scans ( right ) taken over 472 ms . X-axis intensity profiles are plotted ( left ) for the first 16 line scans to illustrate the noise within and across individual line scans . The superimposed black line shows the average of all 64 lines . ( F ) Data from E were averaged over four lines to reduce noise and allow more precise localization of the paRFP spot along the x-axis . The intensity profiles show the first four of the temporally averaged line scans , with the average for all lines superimposed in black . ( G ) Example x-t images averaged over five line scans , showing a stable paRFP-labeled synaptic vesicle associated with a ribbon , which was labeled in green with RBP . Membrane potential was voltage-clamped at -60 mV . ( H ) Intensity profiles of RBP ( green ) and paRFP ( red ) taken in successive 406-ms intervals from the images in G . The red Xs show the position of x0 for each paRFP trace , which varied over a range of 48 nm during the 15-s recording . DOI: http://dx . doi . org/10 . 7554/eLife . 13245 . 009 We next examined changes in paRFP-labeled vesicles at ribbons during activation of calcium current elicited by depolarizing voltage-clamp steps from -60 to -15 mV for 500–5000 ms . During sustained depolarization , previously stable paRFP spots often disappeared from the ribbon , either without a change in position before loss ( e . g . , Figure 4A ) or after moving toward the membrane ( e . g . , Figure 4B , C ) . Whether vesicles moved or not before disappearance depended on starting position with respect to the membrane ( Figure 4D ) : vesicles nearer the plasma membrane ( positive starting positions in Figure 4D ) did not consistently move before disappearing , while distal vesicles ( negative starting positions ) moved toward the membrane before loss . Since membrane-proximal vesicles did not move while distal vesicles moved toward the ribbon center , the histogram of final paRFP positions just prior to disappearance was broad ( Figure 4E ) and covered the full range of ribbon-associated vesicle positions ( cf . , Figure 1F ) . During disappearance , paRFP fluorescence did not cease abruptly , as would occur for bleaching , but instead declined along a Gaussian time course , which is consistent with diffusion of the emitter away from the scanned region ( Figure 4—figure supplement 1 ) . It is uncertain at present whether this diffusion reflects movement of the Vglut1-paRFP protein itself , as would occur for example if the protein incorporated into the plasma membrane after vesicle fusion , or dissociation of an intact vesicle from the ribbon . 10 . 7554/eLife . 13245 . 010Figure 4 . Movement and loss of vesicles during sustained depolarization . ( A ) Example x-t images in which a previously stable paRFP-labeled vesicle disappeared from view during sustained depolarization . The timing of depolarization and the evoked Ca2+ current are shown to the right . Analysis of x-t line scan images is described in Figure 3 . ( B ) Example showing a paRFP-labeled vesicle that appeared distal to the center of the ribbon during sustained depolarization , moved toward the membrane , and disappeared along a Gaussian time course ( Figure 4—figure supplement 1 ) . ( C ) Fluorescence intensity profiles along the x-axis for the example in panel B , showing ribbon position ( green ) and paRFP positions at appearance and disappearance ( red ) . Black lines are fits described in Figure 3 . The dotted line shows the estimated position of the plasma membrane estimated from x1/2 obtained from the fit to RBP fluorescence . ( D ) Displacement amplitude for 88 paRFP-labeled vesicles along the x-axis prior to disappearance , as a function of initial starting position relative to the center of the ribbon . Positive and negative displacements are movements toward and away from the membrane , respectively . Open circles show the average of groups of five points binned by starting position . Error bars: ± 1 sem . Positive starting positions are nearer the membrane , and negative positions are farther away . ( E ) Histogram of final positions of paRFP-labeled vesicles along the x-axis just before disappearance . DOI: http://dx . doi . org/10 . 7554/eLife . 13245 . 01010 . 7554/eLife . 13245 . 011Figure 4—figure supplement 1 . Fluorescence of an emitter declines along a Gaussian time course as the emitter moves away from the region of a line scan . ( A ) The fluorescence of a 27-nm fluorescent bead sampled by a line scan as the bead moved progressively away from the line position . The dashed red line is the best-fitting Gaussian decline . ( B ) Time course of fluorescence decline for 11 paRFP-labeled synaptic vesicles that disappeared from the scanned line during sustained depolarization . The dashed red line is the best-fitting Gaussian decline . DOI: http://dx . doi . org/10 . 7554/eLife . 13245 . 011 Are vesicles that disappear from distal positions on the ribbon during sustained depolarization simply shed from the ribbon without participating in release , or do they somehow fuse and release their contents without approaching the plasma membrane ? To detect the fusion of ribbon-associated vesicles , we generated transgenic zebrafish that express the exocytosis reporter SypHy ( Granseth et al . , 2006 ) ( Synaptophysin-pHluorin fusion protein ) or Vglut1-pHluorin ( Voglmaier et al . , 2006 ) under control of heat-shock promoter . Since pHluorin-fusion proteins mixed with pre-existing native vesicle proteins , pHluorin events were sparse during depolarization , allowing detection and localization of single events with respect to the ribbon , in the same manner as paRFP-labeled vesicles . For 2-color imaging , ribbons were labeled with deep-red CF633-RBP , which did not interfere with pHluorin fluorescence . Under these conditions , unitary pHluorin events representing vesicle fusion were observed in x-t raster plots during sustained depolarizing voltage-clamp steps from -60 to -15 mV for 500–5000 ms ( Figure 5A , B ) , but not in the absence of stimulation . However , the signal-noise ratio for pHluorin fluorescence was poorer than for paRFP , and this necessitated averaging over a greater number of scan lines to achieve spatial resolution in the range of a vesicle diameter for pHluorin events , as illustrated in Figure 5—figure supplement 1 . The fluorescence intensity profile of pHluorin events perpendicular to the membrane was then analyzed as described earlier for paRFP to determine the position of pHluorin events relative to the ribbon and the plasma membrane ( Figure 5C , D ) . Events sometimes arose near the plasma membrane , suggesting fusion of membrane-proximal vesicles ( e . g . , Figure 5C ) , but events also commonly originated at the distal pole of the ribbon , at a distance from the plasma membrane ( e . g . , Figure 5D ) . The histogram of event positions relative to the center of the ribbon ( Figure 5E ) showed that pHluorin signals were broadly spread across the ribbon ( Figure 5E ) and were not confined to positions near the plasma membrane , as would be expected if vesicles must first approach the plasma membrane in order to fuse . 10 . 7554/eLife . 13245 . 012Figure 5 . Synaptic vesicle exocytosis reported by pHluorin occurs at a distance from the plasma membrane . ( A ) Example of x-t images showing a pHluorin event from SypHy ( green ) at a ribbon shortly after onset of depolarization from -60 mV to -15 mV , shown by the trace to the right . Ribbon fluorescence from deep-red CF633-RBP is pseudocolored red . ( B ) Example of x-t images showing a pHluorin event from SypHy ( green ) on the membrane-distal side of a ribbon ( red ) in a different cell during depolarization to -15 mV . ( C ) ( D ) Fluorescence intensity profiles along the x-axis for RBP ( red ) and pHluorin ( green ) for the examples shown in A and B . Black lines are fits as described in Figure 3A–D and in Materials and methods . The x-axis position of the pHluorin event was taken to be the peak of the Gaussian from the fit ( x0 , shown by the dotted green line ) , and the position of the plasma membrane was taken to be the parameter x1/2 ( dotted black line ) from the fit to the fluorescence profile of the ribbon ( red traces ) . ( E ) Histogram of the x-axis position of 127 pHluorin events from SypHy and Vglut1-pHluorin during sustained depolarization . The solid arrow shows the average relative position of the plasma membrane , estimated as described in Materials and methods ( error bar: ± 1 sem ) . The dashed arrow shows the estimated position of the membrane at the top and bottom of the optical section , arrived at by assuming that membrane curvature observed along the y-axis for each x-y image also applied in the z-axis . Most of the pHluorin events fell outside the range of membrane positions between the two arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 13245 . 01210 . 7554/eLife . 13245 . 013Figure 5—figure supplement 1 . Noise level during pHluorin events . Green traces show x-axis intensity profiles during the example pHluorin events shown in Figure 5 . ( A , B ) Event from Figure 5A . ( C , D ) Event from Figure 5B . ( A , C ) The fluorescence profile was averaged over successive groups of four scan lines during the pHluorin event . The black line shows the average over the entire duration of the event ( A: 48 scan lines; C: 128 scan lines ) . The noise level after averaging over four scan lines was insufficient to allow precise localization of the pHluorin emitter . ( B , D ) The same data from A and C averaged over successive groups of 16 scan lines ( B ) or 32 scan lines ( D ) , which reduced noise and allowed more precise localization of the event along the x-axis . DOI: http://dx . doi . org/10 . 7554/eLife . 13245 . 01310 . 7554/eLife . 13245 . 014Figure 5—figure supplement 2 . Variation in ribbon position in different focal planes . ( A ) X-axis profiles of RBP fluorescence were measured after displacing the objective in the z-axis by the indicated relative amounts , spanning the full extent of the labeled ribbon in the z-axis . The gray Xs show fitted values of x0 for the six traces ( positions 200–1200 nm ) when the ribbon was visible . This is an example of a ribbon that showed little variation in x0 along the z-axis . ( B ) X-axis profiles of RBP fluorescence obtained in the manner described for A , but for a ribbon that showed larger variation in x0 along the z-axis . ( C ) Summary of the variation in x0 of the ribbon along the z-axis for 13 experiments . Data were normalized for each experiment by measuring the change in x0 ( Δx0 ) relative to x0 at the brightest focal position for that experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 13245 . 014 Because the z-axis thickness of our optical section was appreciable on the scale of our x-y resolution ( see Materials and methods ) , one concern is that curvature of the plasma membrane in the z-axis could make fusions that actually occurred at the membrane near the top or bottom of the optical section appear to be nearer to the center of the ribbon . However , the majority of pHluorin events still arose at a substantial distance from the estimated membrane position , even after correcting for z-axis curvature ( dashed arrow , Figure 5E ) . Another source of uncertainty in localizing pHluorin events could be changes in ribbon position along the z-axis within the optical section , since we normalized x-axis positions of pHluorin events with respect to the center of the ribbon in constructing the histogram shown in Figure 5E . Major errors seem unlikely to arise from such changes in ribbon position within the optical section , because light from the entire optical section contributes to the x-axis intensity profile used to localize the ribbon in a given focal position . To estimate the extent of error from this source , we measured the change in x-axis position of ribbons as the focal plane was varied within a z-axis range from appearance to disappearance of the ribbon . As shown in Figure 5—figure supplement 2 , the center of the Gaussian component of the fit to the RBP profile varied somewhat with focal position , especially near the upper and lower z-axis bounds , where fluorescence intensity of the ribbon was lower ( e . g . , sample traces in Figure 5—figure supplement 2A , B ) . However , the estimated x-axis position of the ribbon varied by less than ± 50 nm throughout the central part of the sampled z-axis range ( Figure 5—figure supplement 2C ) . Therefore , the observed uncertainty in the x-axis position of ribbons within the optical section is insufficient to account for the wide range of pHluorin-event locations ( Figure 5E ) . We conclude that pHluorin events associated with ribbons arose both near the membrane and at more distal locations , at a substantial distance from the membrane . A simple explanation is that the distal pHluorin events represent fusion of synaptic vesicles with other ribbon-attached vesicles , which have themselves already undergone exocytosis and provide a path to the extracellular space . This model is also consistent with previous work using electron microscopy ( Matthews and Sterling , 2008 ) , which suggested the existence of such vesicle-vesicle compound fusion at ribbon synapses of goldfish BPCs .
In this study , we exploited the special features of ribbon synapses of retinal BPCs to observe the trafficking of single synaptic vesicles associated with the CAZ during ongoing transmitter release . Prior studies of synaptic vesicle trafficking at BPC ribbon synapses using TIRFM focused on a narrow region extending ~100 nm from the plasma membrane and revealed vesicle movements within this region , as well as vesicle exocytosis at the membrane ( Zenisek et al . , 2000 , 2002; Holt et al . , 2004; Midorikawa et al . , 2007; Zenisek , 2008 ) . By contrast , our imaging approaches allowed vesicle motion and fusion to be detected across the entire population of ribbon-associated vesicles , albeit at a spatial resolution insufficient for detection of small movements of less than a vesicle diameter , like those reported in TIRFM experiments ( Zenisek et al . , 2000 ) . Therefore , our studies target a different set of questions and provide results that are complementary to the previous studies of synaptic vesicle dynamics at ribbon active zones . Our results suggest a dual role for the synaptic ribbon in neurotransmitter release at the active zone , with the ribbon serving as both a conduit for diffusion of tethered synaptic vesicles and a platform for vesicles to fuse distal to the plasma membrane . A likely mechanism for such distal fusions is compound fusion of synaptic vesicles with other vesicles . Diffusion appeared to dominate for ribbon-associated vesicles farthest from the plasma membrane , which moved toward the center of the ribbon prior to loss during sustained depolarization . Synaptic vesicle fusion detected by pHluorin events occurred most frequently on the membrane-proximal half of the ribbon , perhaps because calcium levels are higher nearer the membrane during depolarization ( Zenisek et al . , 2003; Francis et al . , 2011; Vaithianathan and Matthews , 2014 ) . These fusion events include some that likely occurred directly at the plasma membrane , which possibly correspond to the fusions seen in TIRFM experiments . However , many pHluorin events arose at distances >100 nm from the estimated position of the plasma membrane , suggesting that compound fusion contributes significantly to ongoing transmitter release at ribbon synapses during sustained depolarization . Overall , our results set the stage for models of synaptic vesicle trafficking and fusion at the ribbon active zone that incorporate calcium signaling , the kinetics of tethered diffusion of vesicles , and vesicle-vesicle fusion at a distance from the plasma membrane .
All animal procedures were in accord with NIH guidelines and followed protocol 247885 approved by the Institutional Animal Care and Use Committee of Stony Brook University . To generate transgenic zebrafish , transgenes were assembled using the Gateway-based Tol2kit ( Kwan et al . , 2007 ) , by constructing appropriate middle-entry and 3’-entry vectors for combination with hsp70 5’-entry vector into a destination vector that included flanking Tol2 transposons and a reverse-orientation cmlc2:EGFP cassette for identifying transgenic embryos . To construct Vglut1 fused at the C-terminus with PATagRFP , cDNA for full-length Vglut1a ( GenBank NM_001098755 ) was obtained by RT-PCR from zebrafish retinal total RNA and cloned into a middle-entry vector . PATagRFP cDNA was a gift from Dr . Vladimir Verkhusha ( Albert Einstein College of Medicine ) and was cloned into a 3’-entry vector . Zebrafish SypHy was generated as described ( Zhu et al . , 2009 ) , using cDNA for Synaptophysin-b ( GenBank NM_001030242 ) obtained by RT-PCR from zebrafish retinal total RNA and cDNA encoding super-ecliptic pHluorin ( Sankaranarayanan et al . , 2000 ) obtained by PCR from a SypHluorin vector ( Addgene 37003; Zhu et al . , 2009 ) . The resulting zebrafish SypHy construct was then cloned into a middle-entry vector . Zebrafish VGlut1-pHluorin ( Voglmaier et al . , 2006 ) was produced in a similar manner . The assembled hsp70 transgenes were injected into single-cell zebrafish embryos along with Tol2 mRNA , and transgenic embryos were selected at 24 hours postfertilization based on GFP fluorescence in the heart driven by cmlc2:EGFP . Transgenic fish were tested for germline transmission of the transgene at 3 months , and the positive progeny were then used to establish transgenic lines . For experiments , transgenic fish >3 months old of both sexes were exposed to 37° C water for 2 hr , kept overnight for protein expression , and used the next day . Bipolar neurons were isolated from adult zebrafish retina as described previously ( Vaithianathan and Matthews , 2014; Vaithianathan et al . , 2013 ) , and whole-cell recordings were made within 2 hr of dissociation , using a patch pipette placed directly on the synaptic terminal . Pipette and bath solutions were similar to those used for goldfish bipolar cells ( Heidelberger and Matthews , 1992 ) , except the pipette solution included 3 mM reduced glutathione as a free-radical scavenger and fluorescent RBP peptide to mark ribbons ( Zenisek et al . , 2004 ) . Although it was concentrated at ribbons , fluorescent RBP also filled the entire cell , allowing the border of the synaptic terminal to be visualized with fluorescence imaging ( see Figure 3 ) . Voltage-clamp data were acquired using a HEKA EPC-9 amplifier controlled by PatchMaster software ( HEKA , Lambrecht/Pfalz , Germany ) . Fluorescence images were acquired using Olympus FluoView software controlling an Olympus FV1000/IX-81 laser-scanning confocal microscope equipped with a 60X 1 . 42 NA oil-immersion objective . Separate xy-scanners allowed simultaneous imaging and photoactivation . The focal plane was carefully adjusted to bring labeled ribbons into sharp focus , avoiding the region of high curvature near the top of the terminal and the plane of adherence of the membrane to the glass coverslip at the bottom of the terminal . This procedure minimized curvature of the plasma membrane in the z-axis within the optical section and therefore facilitated localization of the plasma membrane with respect to the ribbon . Either x-y raster scans or x-t line scans were acquired of a zoomed region near a selected ribbon , depending on the experimental goals . Logic pulses exchanged between patch-clamp and imaging computers synchronized acquisition of electrophysiological and imaging data , and precise timing of imaging relative to voltage-clamp stimuli was established using PatchMaster to digitize horizontal-scan synch pulses from the imaging computer in parallel with the electrophysiological data . Electron microscopy of isolated bipolar cells was performed as described previously ( Matthews and Sterling , 2008; Vaithianathan et al . , 2013 ) . FluoView images of x-y and x-t scans were imported into ImageJ for initial processing and analysis , and subsequent analysis was performed in IGOR Pro ( Wavemetrics , Portland OR ) . For analysis of changes elicited by brief stimuli , 4–10 paRFP images were collected , followed by 10-ms depolarization or no stimulation , and the sequence was repeated at 18-s intervals . Images of the ribbon marked with RBP were taken either just before the series of RFP images , or in sequential alternation with each RFP image of the series . Images were then averaged for each repeat , and the positions of paRFP spots and the ribbon were determined from the peak of a 2D Gaussian fitted using IGOR Pro . Positions of paRFP spots were then expressed as Δx and Δy relative to the center of the ribbon for each repeat . Changes in relative RFP position across repeats were then calculated as ΔΔx and ΔΔy . For analysis of changes during sustained stimuli , a line perpendicular to the membrane at a ribbon location was scanned at intervals of ~1 . 4 ms , with alternation between RBP and paRFP or pHluorin channels on successive lines , or on successive frames . The resulting x-t images were typically averaged in ImageJ over 2 pixels in the x-axis and 5–8 lines in the t-axis to reduce noise . Positions of the ribbon and paRFP spots along the scanned line were determined by fitting the x-axis intensity profile with the equation f ( x ) = s ( x ) + g ( x ) , where s ( x ) is a sigmoid describing the transition from intracellular to extracellular background fluorescence at the edge of the cell , given by s ( x ) = b- ( c/ ( 1-exp ( ( x1/2-x ) /d ) ) ) , and g ( x ) is a Gaussian representing the fluorescence of RBP , paRFP , or pHluorin , given by g ( x ) = a ( exp ( - ( x-x0 ) 2/w2 ) ) . The parameters x1/2 and x0 were taken as the x-axis positions of the plasma membrane and the fluorescence emitter , respectively . The parameter b is intracellular background fluorescence , c is extracellular background fluorescence , d is the slope factor of the sigmoid , a is the peak amplitude of emitter fluorescence , and w is √2* the standard deviation of the Gaussian . In practice , the latter parameters were highly constrained by the data or by the measured PSF , essentially leaving only x1/2 and x0 as free parameters in the fitting . No statistical method was used to predetermine sample size . Non-parametric tests ( Wilcoxon rank test; sign test ) were used to determine statistical significance . Variance in estimates of the population mean is reported as ± sem . | Neurons communicate with one another through junctions known as synapses . When a neuron is activated , it triggers the release of chemicals called neurotransmitters at the synapse , which bind to and activate neighbouring neurons . Neurons involved in vision , sound and balance contain “ribbon” synapses , which are able to release neurotransmitters steadily over longer periods of time than other types of synapse . Neurotransmitters inside neurons are packaged into small structures called vesicles , which can then fuse with the cell’s surface membrane to release the neurotransmitters from the cell . Unlike other types of synapse , ribbon synapses are able to release these vesicles in a continuous fashion . How vesicles move around at the synapses remains poorly understood because monitoring the vesicles in living cells is technically difficult and previous studies were limited to tracking vesicles in a small part of the synapse . Now , Vaithianathan et al . overcome these technical hurdles to follow the movement of vesicles across whole ribbon synapses in zebrafish eyes . The experiments use fluorescent proteins to track the movement of the vesicles under a microscope . Vaithianathan et al . find that vesicles at ribbon synapses move very little when the neurons are not active . However , when the neurons are activated , the vesicles that are near the cell membrane fuse with it and release their neurotransmitters . Other vesicles that are further away from the membrane then move to fill in the spaces vacated by the fusing vesicles . Further experiments show that some of the vesicles that are further away from the membrane can fuse with vesicles that have already released their neurotransmitter but remain in place at the membrane . This process – known as compound fusion – allows neurotransmitters to be released over a longer period of time by providing a path for vesicles to release neurotransmitters without having to approach the membrane first . The next challenge is to develop a computational model using the data from this study to better understand how ribbon synapses work . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Nanoscale dynamics of synaptic vesicle trafficking and fusion at the presynaptic active zone |
Hutchinson-Gilford progeria ( HGPS ) is a premature ageing syndrome caused by a mutation in LMNA , resulting in a truncated form of lamin A called progerin . Progerin triggers loss of the heterochromatic marker H3K27me3 , and premature senescence , which is prevented by telomerase . However , the mechanism how progerin causes disease remains unclear . Here , we describe an inducible cellular system to model HGPS and find that LAP2α ( lamina-associated polypeptide-α ) interacts with lamin A , while its interaction with progerin is significantly reduced . Super-resolution microscopy revealed that over 50% of telomeres localize to the lamina and that LAP2α association with telomeres is impaired in HGPS . This impaired interaction is central to HGPS since increasing LAP2α levels rescues progerin-induced proliferation defects and loss of H3K27me3 , whereas lowering LAP2 levels exacerbates progerin-induced defects . These findings provide novel insights into the pathophysiology underlying HGPS , and how the nuclear lamina regulates proliferation and chromatin organization .
The nuclear lamina , a proteinaceous meshwork consisting of A-type and B-type lamins , underlies the inner nuclear membrane and is important for maintaining interphase nuclear architecture . In addition , it provides a structural scaffold for factors involved in DNA repair , replication and transcription ( Burke and Stewart , 2006; Dechat et al . , 2008 ) . Mutations in the LMNA gene are responsible for a variety of human genetic disorders , collectively called the laminopathies ( Burke and Stewart , 2006; Worman et al . , 2010 ) . Laminopathies include forms of muscular dystrophy , cardiomyopathy , lipodystrophy and the premature aging disease , Hutchinson-Gilford Progeria Syndrome ( HGPS ) . HGPS patients appear normal at birth , but by 12–18 months begin to exhibit features associated with accelerated ageing , and usually die in their teens due to cardiovascular failure . HGPS is caused by an autosomal dominant C to T nucleotide substitution at position 1824 ( G608G ) in LMNA , which activates a cryptic splice site and results in a truncated and constitutively farnesylated version of lamin A called progerin ( De Sandre-Giovannoli et al . , 2003; Eriksson et al . , 2003 ) . HGPS fibroblasts have a greatly reduced proliferative capacity , abnormal nuclear architecture , persistent activation of DNA damage checkpoints and shortened telomeres ( Allsopp et al . , 1992; Bridger and Kill , 2004; Goldman et al . , 2004; Liu et al . , 2005 , 2006; Decker et al . , 2009 ) . Critically shortened telomeres elicit a DNA damage response and trigger senescence , resulting in irreversible growth arrest ( d'Adda di Fagagna et al . , 2003 ) . Previous results revealed that ectopic expression of telomerase reverse transcriptase ( TERT ) extends the proliferative capacity of HGPS fibroblasts and rescues progerin-induced DNA damage ( Kudlow et al . , 2008; Benson et al . , 2010 ) . However , it remains unknown to what extent ectopic expression of TERT rescues all progerin-induced phenotypes , and whether physiological levels of TERT are sufficient . It also remains unclear how progerin triggers senescence and why specific tissues in HGPS patients are more affected than others . Here , we describe a regulatable cellular model of progeria and show that upon induction in primary human fibroblasts , progerin leads to increased DNA damage , cellular senescence , senescence-associated reduction of lamin B1 , nuclear morphology defects and altered expression of H3K27me3 , in a dose-dependent manner . Exogenous TERT prevents the proliferative defects , DNA damage , lamin B1 reduction and gene expression differences induced by progerin , although nuclear morphology defects and altered deposition of H3K27me3 are not prevented by TERT . To determine how progerin may induce these defects , we compared the protein interactome between lamin A and progerin . This revealed that the physical association between progerin and the α-isoform of the lamina-associated polypeptide 2 ( LAP2α ) is disrupted . Increased levels of LAP2α , but not LAP2β , suppressed many of the progerin-induced defects , including the inhibition of cell proliferation and reduction in heterochromatin , revealing that LAP2α plays a central role in the molecular pathology of progeria .
To investigate the mechanism of progerin's effects , we developed a tractable experimental system utilizing primary ( telomerase-negative ) and telomerase-positive ( expressing pBABE-Neo-hTERT ) ( TERT+ ) human fibroblasts . These were then either induced to express V5-tagged lamin A or progerin to model the pathophysiology of HGPS fibroblasts . We used a doxycycline ( DOX ) inducible lentiviral based system to quantitatively induce the expression of progerin and lamin A to levels comparable to those present in HGPS fibroblasts ( Figure 1A , B; Figure 1—figure supplement 1A , B , H ) . The advantage of such a system is that it accurately tracks the replicative history of isogenic cell lines , removing the uncertainty in using HGPS-patient derived cells where passage number and telomere length may be unknown , as described ( Decker et al . , 2009 ) . 10 . 7554/eLife . 07759 . 003Figure 1 . Telomerase rescues dose dependent progerin-induced proliferation defects , DNA damage and gene expression changes without alleviating chromatin changes . ( A ) Immunofluorescence microscopy using v5-tag antibody showing doxycycline-dependent inducible expression of v5-progerin and its localization to the nuclear periphery . DAPI staining is shown on the bottom panels . Scale bar: 100 μm . ( B ) Western blotting showing doxycycline-dependent progerin expression in primary ( left panel ) and TERT+ ( right panel ) fibroblasts . Progerin migrates between lamin A and C as indicated ( red arrowhead ) . Doxycycline concentrations ( 0–2000 ng/ml ) are indicated under each lane . ( C ) Quantification of progerin-induced proliferation defects . Relative growth rates of primary ( left panel ) and TERT+ cells ( right panel ) according to progerin expression levels ( *p < 0 . 05; ***p < 0 . 001 compared to control 0 ng/ml doxycycline , error bars represent SEM , 2-way ANOVA with Bonferroni's post-test ) . ( D ) Quantification of progerin-induced 53BP1 DNA damage foci in response to progerin expression levels , in primary ( left panel , p < 0 . 01 , χ2 test ) and TERT+ cells ( right panel ) . 350–500 cells were counted for each condition . ( E ) Scatter plot analysis of primary ( blue , red ) and TERT+ ( cyan , orange ) cells showing an inverse correlation between H3K27me3 and progerin expression in each cell nucleus using immunofluorescence microscopy ( Pearson r = −0 . 43 and −0 . 24 for TERT negative and TERT+ cells expressing progerin , respectively , p < 0 . 001 , n > 9800 ) . Inset: box plot of the same data , whiskers represent 10–90 percentile ( ***p < 0 . 001 , *p < 0 . 05 , one way ANOVA with Bonferroni's post-test ) . ( F ) Scatter plot analysis of H3K27me3 and progerin expression in single nuclei of two primary HGPS lines using immunofluorescence microscopy ( Pearson r = −0 . 70 and −0 . 52 for HGPS AG01972 and HGPS AG11513 respectively , p < 0 . 001 , n > 4000 ) . ( G ) Illustration showing the number of genes whose expression changed more than twofold after 28 days of lamin A or progerin expression ( I , induced . N . I . , non-induced ) . No significant changes were observed upon expression of lamin A . In primary and TERT+ cells , 5 and 142 genes were differentially regulated upon progerin expression , respectively . ( H ) Heatmap representation of the 142 differentially regulated genes in the presence or absence of progerin and TERT , in human fibroblasts . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 00310 . 7554/eLife . 07759 . 004Figure 1—figure supplement 1 . Progerin induced senescence , lamin B1 loss , DNA damage , and telomere shortening are prevented by TERT in primary and HGPS fibroblasts , control experiments . ( A ) Western blotting showing inducible expression of progerin or wild type lamin A in primary and TERT+ human fibroblasts ± doxycyclin ( DOX ) as indicated . ( B ) Immunofluorescence microscopy using V5-tag antibody ( top panels ) showing doxycyclin-dependent expression of v5-lamin A and v5-progerin in TERT+ human fibroblasts and localization to the nuclear periphery . Inset: higher magnification image of different field . DAPI staining is shown in bottom panels . Scale bar: 50 μm; scale bar inset: 20 μm . ( C ) Growth curve of TERT+ and primary cells in the presence or absence of progerin ( ±DOX ) . Dotted lines indicate SEM ( n = 5 ) . Inset: growth rate after 6 days , error bars indicate SEM ( **p < 0 . 01 , one-way ANOVA with Tukey's post-test ) . ( D ) Quantification of lamin B1 levels upon progerin or lamin A expression in primary and TERT+ cells . Error bars indicate standard deviation ( n = 4 , *p = 0 . 05 , Student's t-test ) . Values were normalized to no DOX control . ( E ) Quantification of progerin protein levels upon induction with doxycycline ( ±DOX ) in primary and TERT+ human fibroblasts . Levels were normalized to GAPDH loading control ( n = 3 , error bars indicate standard deviation ) . ( F ) Growth rate of primary and TERT+ fibroblasts in the presence ( +DOX ) or absence ( −DOX ) of exogenous lamin A ( n = 3 , error bars indicate SEM ) . ( G ) Percentage of senescence-associated β-gal-positive cells in the presence or absence of progerin induction ( ±DOX ) in primary or TERT+ fibroblasts ( n = 3 , error bars indicate standard deviation , ***p < 0 . 001 , two-way ANOVA with Tukey's post-test ) . ( H ) Western blotting showing expression of lamin A , lamin C and progerin in TERT negative and TERT+HGPS cells . ( I ) Telomere blot showing telomere length in the parental AG11498 HGPS cells , and subsequent re-elongation upon ectopic expression of TERT . ( J ) Percentage of DNA-damage associated 53BP1-foci , in the presence or absence of TERT , in wild type primary fibroblast and two different HGPS fibroblasts lines ( AG11498 , AG11513 ) . Error bars indicate standard deviation ( n = 4 , ***p < 0 . 001 , one-way ANOVA with Tukey's post-test ) . ( K ) Scatter plot of H3K27me3 and progerin levels in TERT+HGPS AG01972 by immunofluorescence microscopy ( Pearson r = −0 . 5 , n > 1400 nuclei ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 00410 . 7554/eLife . 07759 . 005Figure 1—figure supplement 2 . Expression of hTERT or LAP2α does not alleviate nuclear abnormalities in HGPS cells . ( A ) Percentage of cells with nuclear abnormalities before and after transduction with hTERT . P16 control ( ctrl ) , HGPS AG11498 and HGPS AG11513 were transduced with hTERT and propagated for an additional 20 to 28 passages . ( B ) Percentage of cells with nuclear abnormalities in the presence or absence of ectopic LAP2α . HGPS AG11513 were transduced with hTERT and pTRIPZ-LAP2α . Nuclear abnormalities were scored by immunofluorescence microscopy , numbers of scored cells are indicated within each bar . Error bars represent standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 005 Progerin expression leads to misshapen nuclei ( Figure 1—figure supplement 1B ) , inhibition of cell proliferation ( Figure 1C , Figure 1—figure supplement 1C ) , premature senescence ( as measured by expression of senescence-associated β-galactosidase , [SA-β-gal] ) , a reduction in lamin B1 levels , ( Figure 1—figure supplement 1A , D , G ) and the induction of 53BP-1 DNA-damage foci ( Figure 1D ) . In addition , western blot analysis from three independent experiments demonstrated that expression of TERT did not reduce progerin levels ( Figure 1B , E and Figure 1—figure supplement 1E ) . All of these changes are consistent with previous findings from HGPS and senescent fibroblasts ( Scaffidi and Misteli , 2005; Taimen et al . , 2009; Shimi et al . , 2011; Freund et al . , 2012; Dreesen et al . , 2013 ) . The expression of exogenous TERT prevented progerin-induced proliferative defects , loss of lamin B1 , and reduced the number of SA-β-gal positive cells to background levels ( Figure 1B , D; Figure 1—figure supplement 1A , C , D , G ) . Consistent with these results , expression of TERT in HGPS patient derived fibroblasts increased telomere length , restored their proliferative capacity and reduced the number of cells with DNA damage foci ( Figure 1—figure supplement 1H–J ) . In contrast , expression of TERT did not ameliorate the nuclear abnormalities of HGPS cells ( Figure 1—figure supplement 2A ) . Lastly , expression of normal lamin A did not significantly affect the proliferation rates of primary- or TERT+ human fibroblasts ( Figure 1—figure supplement 1F ) . One of the most intriguing aspects of HGPS and other premature aging syndromes is the tissue-specific manifestation of the disease . Different lineages derived from reprogrammed HGPS induced pluripotent cells ( iPSC ) expressed varying amounts of progerin , with neural lineages showing the lowest levels , consistent with the fact that they remain unaffected in HGPS patients ( Zhang et al . , 2011 ) . These results suggested that progerin's detrimental effects depend on its levels of expression in a given tissue . To test this , we expressed increasing levels of progerin in primary and TERT+ fibroblasts by increasing the concentration of DOX ( 0–2000 ng/ml ) . This resulted in progerin inhibiting the proliferation of primary fibroblasts in a dose-dependent manner ( Figure 1C ) . Defective proliferation was accompanied by a gradual increase in DNA damage levels , as quantified by the number of 53BP1 foci per nucleus ( Figure 1D , left panel , p < 0 . 01 ) . The levels of progerin required to induce a phenotype corresponded to ∼30–40% of endogenous lamin A levels ( at 100–250 ng/ml DOX; Figure 1B–D ) . Progerin therefore must reach a certain threshold to induce DNA damage and inhibit proliferation . Both of these effects were suppressed by exogenous TERT ( Figure 1B–D , right panel ) . Previous studies reported that progerin leads to a decrease in repressive histone marks including H3K27 trimethylation and loss of peripheral heterochromatin ( Shumaker et al . , 2006; McCord et al . , 2013 ) . To determine whether TERT would prevent progerin-induced chromatin alterations , we measured the levels of progerin and H3K27me3 in single cells by immunofluorescence . As shown in Figure 1E , a scatterplot analysis of >9800 nuclei revealed an inverse correlation between progerin and H3K27me3 levels ( Pearson r = −0 . 43 and −0 . 24 for TERT negative and TERT+ cells expressing progerin , respectively , p < 0 . 001 , Figure 1E , inset ) . However , progerin mediated loss of H3K27me3 was not prevented by TERT . The inverse correlation between progerin expression and loss of H3K27me3 is also apparent in two primary HGPS cell lines ( Figure 1F , Pearson r = −0 . 70 and −0 . 52 for HGPS 01972 and HGPS 11513 respectively , p < 0 . 001 ) , as well as in one TERT+HGPS cell line ( Figure 1—figure supplement 1K , Pearson r = −0 . 5 ) . To determine whether this altered chromatin state was sufficient to affect genome-wide gene expression , and whether TERT would prevent these changes , we performed a microarray analysis on primary and TERT+ cells expressing either lamin A or progerin at 4 weeks after induction . The expression of 142 genes was increased or decreased more than twofold in progerin vs lamin A expressing cells ( Figure 1G ) . Many of these gene expression changes were associated with senescence , including a reduction of Wnt2 ( Ye et al . , 2007 ) , increased expression of matrix metalloproteinases ( Kang et al . , 2003 ) and plasminogen activator inhibitor-1 ( PAI-1 ) ( Kortlever et al . , 2006 ) . Expression of TERT prevented nearly all these changes in the differentially expressed genes , and restored the gene expression profile to that seen in cells not expressing progerin ( Figure 1H ) . These results show that the inducible system reliably phenocopies HGPS cell characteristics in isogenic cell lines , and that ectopic expression of TERT prevents dose-dependent DNA damage , premature cellular senescence and senescence-associated changes in gene expression induced by progerin , independent of its impact on H3K27 methylation . To determine whether endogenous physiological levels of TERT would recapitulate the effects of exogenous TERT expression , we expressed lamin A and progerin in mouse embryonic stem cells ( ESC ) . Endogenous TERT expression is a hallmark of ESCs and enables them to perpetually self-renew . Both lamin A and progerin were expressed in the ESC nuclei upon addition of DOX and localized to the nuclear periphery ( Figure 2—figure supplement 1A , B ) . Expression of the exogenous progerin or lamin A did not impair the proliferation of the pluripotent ESCs ( Figure 2A ) , induce significant changes in gene expression ( Figure 2B ) , alter nuclear lamina structure , as measured by lamin B1 and emerin expression , nor affect the expression of the pluripotency markers Nanog , Oct-4 and Sox-2 ( Figure 2C ) . TERT is expressed in undifferentiated ESC , but is repressed during differentiation ( Armstrong et al . , 2000 ) . To determine whether ESC would become susceptible to progerin expression upon differentiation , we aggregated ESC into embryoid bodies ( EB ) , plated them in tissue culture dishes and measured the size of the differentiating EB outgrowth upon plating . While the total EB size did not vary significantly between the different conditions ( Figure 2E ) , the EB outgrowth of differentiated cells was significantly reduced in progerin expressing cells ( Figure 2F , p < 0 . 01 ) . As in primary fibroblasts , we observed an increase in 53BP-1 foci in the differentiated progerin expressing cells ( Figure 2—figure supplement 1C ) . To further investigate whether TERT is necessary to prevent progerin-induced defects in pluripotent ESC , we expressed progerin in Tert−/− mouse ESC ( Figure 2—supplement 1D , E ) . Expression of progerin in Tert−/− ESC led to a reduction in cell number ( Figure 2G ) , rapidly induced differentiation and significantly impaired ability of ESC to form embryoid bodies ( Figure 2H , I ) . Taken together , these results demonstrate that physiological expression levels of TERT are necessary and sufficient to prevent progerin-induced defects . 10 . 7554/eLife . 07759 . 006Figure 2 . Physiological levels of telomerase prevent progerin-induced defects in mouse ESC . ( A ) Growth curve of mouse ESC expressing progerin ( PG ) or lamin A ( LA ) upon DOX induction ( n = 3 , error bars indicate SEM ) . ( B ) Heatmap showing the number of genes whose expression changed more than twofold after 8 days of lamin A or progerin expression ( I , induced . N . I . , non-induced ) . ( C ) Immunofluorescence microscopy using Oct-4 , emerin , lamin B1 and Sox2 antibodies in the presence or absence of v5-lamin A and v5-progerin expression . ( D ) Embryoid body ( EB ) formation upon removal of leukemia inhibitory factor ( LIF ) . The orange line indicates the total size of the differentiated EB , while the pink line indicates the differentiated cell outgrowth . ( E ) Quantification of total embryoid body size in ESC expressing lamin A ( LA+DOX ) or progerin ( PG+DOX ) , compared to EBs differentiated from ESC LA non induced controls ( one-way ANOVA , n > 80 , p > 0 . 05 ) . ( F ) Quantification of the size of the differentiated cell layer , in percentage of the total EB size for each EB , compared to EBs differentiated from non-induced ESC LA controls ( p < 0 . 01 , n > 80 , one-way ANOVA with Tukey's post-test ) . ( G ) Cell counts of Tert−/− ESC in the presence ( PG+DOX ) or absence ( PG ) of progerin . Cells were induced for 5 days prior to cell counting ( p < 0 . 05 , n = 3 , Student's t-test ) . ( H ) Brightfield microscopy images of Tert−/− ESC ± progerin . Pictures were taken 7 days after induction with progerin ( PG+DOX ) or non-induced controls ( PG ) . ( I ) Total size of EBs differentiated from Tert−/− ESC expressing progerin ( PG+DOX ) or controls ( PG ) ( p < 0 . 001 , n > 160 , Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 00610 . 7554/eLife . 07759 . 007Figure 2—figure supplement 1 . ( A ) Western blot showing inducible expression of v5-progerin or v5-lamin A in primary mouse ESC ± doxycycline ( DOX ) as indicated . V5 , nanog , GAPDH and actin are shown . ( B ) Immunofluorescence microscopy using v5-tag antibody ( top panels ) showing doxycycline-dependent expression of v5-lamin A and v5-progerin and localization to the nuclear periphery . ( C ) Immunofluorescence staining of embryoid body outgrowth . V5-tagged progerin ( v5 , green ) and DNA damage foci ( 53BP-1 , red ) are shown . ( D ) Expression of v5-progerin in telomerase-deficient ESC . Western blot showing DOX-regulated expression of v5-progerin . Antibodies: v5-tag , lamin B1 , GAPDH . ( E ) Immunofluorescence microscopy of Tert−/− ESC in the presence of absence of v5-progerin . Antibody: v5-tag ( red ) , DAPI ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 007 Cellular senescence is considered to be a key factor in HGPS , as well as during normal ageing in humans ( Kuilman et al . , 2010 ) . To determine how progerin may trigger senescence , we compared the protein interactomes of lamin A and progerin using BioID ( Roux et al . , 2012 ) . The Myc-tagged promiscuous biotin ligase BirA* was fused to the N-termini of lamin A or progerin , and expressed in fibroblasts by DOX-induction . To avoid complications from senescence-associated secondary consequences of progerin expression , we performed the comparison in TERT-expressing cells . Upon induction , BirA*-lamin A and BirA*-progerin were expressed ( Figure 3A ) , localized at the nuclear periphery ( Figure 3B ) , with BirA*-progerin inducing lobulated and misshapen nuclei ( Figure 3B ) . Protein biotinylation by the BirA*-lamin A and progerin fusion proteins occurred exclusively upon addition of biotin and DOX ( Figure 3—figure supplement 1A ) . Biotinylated proteins were purified and analyzed by mass spectrometry . As expected , self-biotinylated BirA*-lamin A , BirA*-progerin , endogenous lamin A/C and biotinylated lamin B1 , previously shown to interact with A-type lamins , were identified ( Figure 3—figure supplement 1B , C ) ( Kubben et al . , 2010 ) . Mass spectrometry analysis of pull-down fractions revealed several known components of the nuclear envelope/lamina , including lamin A , LAP2 , emerin , lamin B1 and B2 ( Figure 3—figure supplement 1C ) ( Roux et al . , 2012 ) . We compared the interactome of lamin A vs progerin , and quantified the differential interactions using the exponentially modified protein abundance index ( emPAI ) ( Ishihama et al . , 2005 ) . We observed a decreased interaction of the nuclear pore complex protein TPR with progerin , consistent with a previous report describing impaired nuclear import of TPR in HGPS cells ( Snow et al . , 2013 ) . A list of the 11 identified nuclear proteins and their respective interaction index with lamin A or progerin is shown in Figure 3—figure supplement 1C . 10 . 7554/eLife . 07759 . 008Figure 3 . BioID analysis reveals differential interaction of lamin A and progerin with lamina-associated polypeptide 2 ( LAP2 ) . ( A ) Western blot showing doxycycline-dependent expression of myc-BirA*-progerin ( BirA-PG ) and myc-BirA*-lamin A ( BirA-LA ) fusion constructs in primary and TERT+ cells . Antibodies are indicated: myc; lamin A , lamin C , LAP2α , actin , GAPDH . ( B ) Immunofluorescence microscopy confirms doxycycline-dependent induction and localization of BirA*-lamin A/BirA*-progerin fusion constructs to the nuclear periphery ( green , myc tag; blue , DAPI staining ) . Scale bar: 20 μm . ( C ) Impaired interaction of LAP2 with progerin . Quantitative interactome of lamin A ( black bars ) or progerin ( striped bars ) with nuclear proteins lamin A , LAP2 , lamin B1 and B2 . Control: non-induced BirA*-lamin A ( grey bars ) . BioID ( emPAI ) index: quantification based on the number of peptides for each protein detected by mass spectrometry error bars represent SEM ( n = 3 , ***p < 0 . 001 , one-way ANOVA with Tukey's post-test ) . ( D ) Interaction of lamin A or progerin with LAP2α or emerin by co-immunoprecipitation . In vitro transcribed and translated v5-tagged lamin A , v5-tagged progerin , LAP2α and emerin ( antibodies: v5-tag , LAP2α , emerin are indicated ) . Top panel: recombinant v5-tagged progerin and lamin A , myc-LAP2α and HA-emerin were efficiently immunoprecipitated using anti-v5-tag or anti-myc antibodies , respectively ( input lanes two , three and four ) . Bottom panel: LAP2α or emerin immunoprecipitated by either v5-lamin A or v5-progerin . Quantification of LAP2α and emerin pulled down by v5-lamin A or v5-progerin is shown below ( normalized to respective v5-signal , *p < 0 . 05 , Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 00810 . 7554/eLife . 07759 . 009Figure 3—figure supplement 1 . BirA*-dependent biotinylation of proteins in human fibroblasts , control experiments and protein list . ( A ) Western blot showing specific biotinylation of proteins by BirA*-laminA and BirA*-progerin fusion proteins , upon DOX induced expression and in the presence of biotin . Conditions are indicated above each lane . Staining: HRP-streptavidin . ( B ) Western blot analysis of pulled down biotinylated proteins ( prior to mass spectrometry analysis ) . Antibodies: lamin A/C antibody recognizes both endogenous lamin A ( LA ) , lamin C ( LC ) and fusion constructs BirA*-lamin A and BirA*-progerin , lamin B1 ( LB1 ) , GAPDH and actin . ( C ) Comparative interactome of lamin A and progerin after BioID assay using the exponentially modified protein abundance index ( emPAI ) . emPAI index quantifies the abundance of a protein identified by mass spectrometry in a protein mixture . DOX untreated cells in the presence of biotin were used as negative controls accounting for endogenous biotinylation . Upper panel: top row: experimental conditions ( ±doxycyclin ) and expressed constructs ( BirA*-lamin A or BirA*-progerin ) . Lower panel ( shaded row ) : list of proteins identified by mass spectrometry and their respective emPAI index . p-values and SEM from three independent experiments are indicated ( one-way ANOVA with Tukey's post-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 00910 . 7554/eLife . 07759 . 010Figure 3—figure supplement 2 . BioID analysis of lamin A or progerin in pluripotent ESC . ( A ) Western blot showing doxycyclin-dependent expression of myc-BirA*-progerin and myc-BirA*-lamin A fusion constructs in ESC . Antibodies are indicated: myc , GAPDH . ( B ) Doxycyclin-dependent induction and localization of BirA*-lamin A/BirA*-progerin fusion constructs to the nuclear periphery by immunofluorescence microscopy . Antibodies: myc , lamin B1 ( LB1 ) , emerin , SUN1 and DAPI staining . ( C ) Comparative interactome of lamin A and progerin with LAP2 , lamin B1 and Lamin B2 after BioID using emPAI . Cells in the absence of DOX but in the presence of biotin were used as negative controls ( *p < 0 . 05 , one-way ANOVA with Tukey's post-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 010 We observed a significantly decreased interaction of progerin with the lamina-associated polypeptide 2 ( LAP2 ) ( Figure 3C ) . LAP2 exists as several alternatively spliced isoforms ( Dorner et al . , 2006 ) , among which LAP2α and β were identified by BioID . Since LAP2α forms nucleoplasmic complexes with lamin A ( Dechat et al . , 2000 ) , and its levels decline with progerin expression ( Scaffidi and Misteli , 2005; Zhang et al . , 2011 ) or during senescence ( Dreesen et al . , 2013 ) , we focused on the α-isoform . To avoid complications associated with cellular senescence , we used TERT+ cells expressing BirA*-progerin , in which total LAP2α levels remained stable in protein extracts ( Figure 3A , lanes 3 + 4 bottom panel ) . This confirmed that the reduced interaction between LAP2α we observed by BioID was not due to a global decrease in the LAP2α levels in the protein samples . In addition , we expressed BirA*-lamin A and BirA*-progerin in pluripotent ESC . Both constructs correctly localized to the nuclear periphery and did not lead to any alterations in the nuclear lamina , as judged by emerin , lamin B1 and SUN1 staining ( Figure 3—figure supplement 2A , B ) . As expected , mass spectrometry analysis of pull-down fractions identified the nuclear lamina constituents lamin A/C , lamin B1 and B2 , suggesting that the BirA*-constructs interact with endogenous proteins similarly in ESC and in human fibroblasts . We also noted that , progerin also showed a significantly decreased interaction with LAP2 in the ESC ( Figure 3—figure supplement 2C ) . To determine whether LAP2α physically interacts with lamin A and progerin , we examined the interaction of in vitro transcribed/translated v5-tagged lamin A or v5-progerin with LAP2α and emerin by co-translation followed by co-immunoprecipitation ( Figure 3D , upper panel ) . Progerin consistently pulled down ∼40–60% less LAP2α than lamin A , while its interaction with emerin was unaffected ( Figure 3D , lower panel ) . These results demonstrate that the weakened binding between progerin and LAP2α , suggested by the BioID screen , is due to a reduction of the association between progerin and LAP2α . A comparison of the lamin A and progerin interactomes has been described using other procedures ( Kubben et al . , 2010; Liu et al . , 2011 ) , but it is unclear whether any of the previously identified differential interactors had any functional role in the pathophysiology of HGPS . However , a previous report indicated that LAP2α may directly interact with chromatin and telomeres ( Dechat et al . , 2004 ) . In addition , since exogenous telomerase suppresses progerin-induced defects , we investigated whether LAP2α localization to the telomeres was altered in TERT+HGPS cells . To address this with sufficient resolution , we used 3D-structured illumination microscopy ( Schermelleh et al . , 2010 ) to compare nuclei from TERT+ wild-type and TERT+HGPS cells . In normal nuclei , LAP2α was present as discrete foci distributed throughout the nucleoplasm ( Figure 4A , B ) , with many foci closely localized with telomeres , visualized by staining for TRF1 , a component of the telomere-associated shelterin complex ( De Lange , 2005 ) . We then measured the distribution profile of LAP2α along a ∼400 nm axis from the center of each telomere in 3D-SIM nuclear sections ( 544 nm thick ) , at 45° angle intervals ( Figure 4B ) . To eliminate any localization bias due to the observed differences in LAP2α amounts between WT and HGPS nuclei ( Figure 4C , p < 0 . 001 , n = 17 ) , we normalized the signal intensity to the maximum/minimum LAP2α intensity for each nucleus . We found that in normal nuclei , the highest average LAP2α signal intensity is within ∼200 nm of each telomere . In contrast , in HGPS nuclei ( Figure 4—figure supplement 1A , B ) , the distribution of LAP2α in relation to the telomeres was significantly altered , and reached its maximum value roughly 360–400 nm from the center of each telomere ( Figure 4D ) . We observed a similar profile of LAP2α localization to telomeres in TERT+ wild type cells expressing progerin ( Figure 4—figure supplement 1H , I ) . 10 . 7554/eLife . 07759 . 011Figure 4 . LAP2α association with telomeres . ( A , B ) Projection of an extended section of a wild-type fibroblast nucleus showing lamin A ( blue ) , LAP2α ( red ) and TRF1 ( green ) staining . Magnified section indicated by the white frame is shown in panel ( B ) . The eight radiuses used to measure the distribution profile of LAP2α around each telomere are indicated by the dotted white lines . ( C ) Average LAP2α intensity in WT or TERT+HGPS nucleus ( n = 17 , ***p < 0 . 001 , errors bars indicate SEM , Student's t-test ) . ( D ) Intensity profile of LAP2α ( red ) near telomeres ( TRF1 , green ) in WT ( left part , blue ) or TERT+HGPS ( right part , red ) nuclei along ∼400 nm axis , relative to the minimum and maximum LAP2α and TRF1 signal per nucleus . Dotted lines indicate SEM , n ≥ 88 . ( E ) Quantification of telomere distance to the nuclear lamina ( a total number of 2891 telomeres from 34 nuclei were analyzed ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 01110 . 7554/eLife . 07759 . 012Figure 4—figure supplement 1 . LAP2α association with telomeres , control experiments and details of analysis . ( A , B ) Projection of an extended section of a HGPS nucleus showing lamin A ( blue ) , LAP2α ( red ) and TRF1 ( green ) staining . Scale bar ( A ) : 3 μm; scale bar ( B ) : 2 μm . ( C ) Average surface occupied by LAP2α in wild type and HGPS nuclei ( errors bars indicate SEM , p < 0 . 001 , n = 25 , Student's t-test ) . ( D ) Processed section of a wild-type nucleus showing LAP2α ( pink ) , TRF1 ( green ) and lamin A . The yellow line marks the borders of the area used to quantify the average LAP2α surface coverage per nucleus ( LAP2c = 15 . 3% in this nucleus ) . Scale bar: 3 μm . ( E ) Higher magnification of identified telomeres objects shown in ( D ) . The red line marks the borders of the area used to quantify the LAP2α surface coverage for each telomere . LAP2α surface coverage for each telomere is indicated in brackets: telomere 1: 12 . 5% category >LAP2c¯±σ , telomere 2: 85% category >LAP2c¯±σ , telomere 3: 36% category: >LAP2c¯+σ , telomere 4: 0% category: >LAP2c¯−σ . ( F ) Proportion of the surface of a telomere associated with LAP2α in WT or TERT+HGPS nuclei . The LAP2α/TRF1 colocalization is expressed in relation to the average surface occupied by LAP2α for each nucleus ( LAP2c: average nuclear surface covered by LAP2α in one nucleus , [n = 25 , *p < 0 . 05 , ***p < 0 . 001 , errors bars indicate SEM , two-way ANOVA with Bonferroni's post-test] ) . ( G ) 3D rendering of a wild type nucleus from super resolution imaging data . The nuclear lamina ( light blue ) has been rendered partially to allow visualizing the position of telomeres within the nucleus . Telomeres are represented as spheres and their proximity to the nuclear lamina is indicated as follows: within 250 nm ( red ) or outside 250 nm ( green ) of the nuclear lamina . Scale bar: 2 μm . ( H ) Intensity profile of LAP2α ( red ) near telomeres ( TRF1 , green ) in TERT+ human fibroblasts , control ( left part ) or progerin expressing nuclei ( right part ) along a ∼400 nm axis , relative to the minimum and maximum LAP2α and TRF1 signal per nucleus . Dotted lines indicate SEM , n ≥ 38 nuclei . ( I ) Average LAP2α intensity in control or progerin induced TERT+ human fibroblasts ( n ≥ 38 , **p < 0 . 01 , errors bars indicate SEM , Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 012 To confirm this loss of proximity between LAP2α and the telomeres in HGPS cells , we quantified the surface area of telomeres co-localizing with LAP2α for each telomere in normal or HGPS nuclear sections . Consistent with our previous results , we found a decrease in the surface area of HGPS nuclei associated with LAP2α ( 6 . 7% of the surface of each HGPS nuclei was covered by LAP2α vs 15 . 7% for wild-type nuclei [p < 0 . 001 , n = 25 , Figure 4—figure supplement 1C] ) . To take into account this difference in nucleoplasmic LAP2α levels between HGPS and wild type nuclei , we grouped telomeres into three categories ( i ) low co-localization TRF1/LAP2α at the telomeres: telomeres with a percentage of their surface co-localizing with LAP2α below the average percentage of surface covered by LAP2α in the nucleus , minus one standard deviation ( noted < LAP2c¯−σ ) , ( ii ) average co-localization TRF1/LAP2α at the telomeres: telomeres with a percentage of their surface co-localizing with LAP2α within the average surface covered by LAP2α in the nucleus ( noted LAP2c¯±σ ) and ( iii ) high co-localization TRF1/LAP2α at the telomeres: telomeres with a percentage of their surface co-localizing with LAP2α above the average surface area covered by LAP2α in the nucleus plus one standard deviation ( noted > LAP2c¯+σ ) . Examples of these quantified images used for telomere/LAP2α co-localization are shown in Figure 4—figure supplement 1D , E . In agreement with an impaired LAP2α localization at telomeres , we observed a lower co-localization of telomeres with LAP2α in HGPS cells . In addition , we observed fewer telomeres with average or high LAP2α association in HGPS cells as indicated by categories LAP2c¯±σ and >LAP2c¯+σ , respectively ( Figure 4—figure supplement 1F ) . Taken together , these results demonstrate that the close physical proximity between LAP2α and telomeres is disrupted in HGPS nuclei . Depletion of , or mutations in LMNA can alter telomere distribution within the nucleus ( Gonzalez-Suarez et al . , 2009; Taimen et al . , 2009; De Vos et al . , 2010 ) . Telomeres may also transiently localize to the nuclear periphery during the G1 phase of the cell cycle ( Crabbe et al . , 2012 ) . To determine whether telomeres were mis-localized in HGPS nuclei , we used 3D-SIM and 3D rendering to measure the distance between telomeres and the nuclear lamina ( Figure 4—figure supplement 1G ) . We found that ∼50% of the telomeres localized to within 250 nm of the nuclear lamina in interphase nuclei , but we did not observe any change in telomere distribution between normal- and HGPS+TERT fibroblasts ( Figure 4E ) . This suggests that progerin expression or mis-localization of LAP2α does not affect telomere distribution within the nucleus . To investigate whether the impaired association of LAP2α with telomeres and progerin was functionally relevant to the pathophysiology of HGPS , we modulated LAP2 levels in wild type and progerin expressing cells using our doxycyclin-inducible system . First , we depleted the α , β and γ isoforms of LAP2 using lentiviral delivered shRNA , and observed , in agreement with previous reports , enhanced proliferation of primary and TERT+ fibroblasts ( Figure 5—figure supplement 1A–C ) ( Dorner et al . , 2006; Naetar et al . , 2008 ) . To determine the consequences of LAP2 depletion in progerin expressing cells , we introduced v5-tagged progerin or lamin A into LAP2-depleted fibroblasts ( Figure 5B ) . Surprisingly , and in contrast to normal cells , the loss of LAP2 enhanced progerin-induced proliferation defects ( Figure 5C , red arrowhead ) . However , this enhanced reduction in proliferation was rescued by TERT expression ( Figure 5—figure supplement 1D ) . From these findings we conclude that LAP2 depletion potentiates the detrimental effect of progerin on cell proliferation . 10 . 7554/eLife . 07759 . 013Figure 5 . LAP2 depletion exacerbates the progerin-induced proliferation defect whereas specific overexpression of LAP2α rescues it . ( A ) Growth curve of normal ( ctrl shRNA ) and LAP2-depleted ( shLAP2 ) primary fibroblasts expressing progerin or lamin A . Dotted lines indicate SEM ( n = 3 ) . Inset: growth rate after 8 days , error bras indicate SEM ( *p < 0 . 05 , Student's t-test ) . ( B ) Western blot of primary and TERT+ cells expressing v5-tagged lamin A ( v5-LA ) or progerin ( v5-PG ) with or without LAP2 silencing by shRNA ( shLAP2 ) . V5-tag , lamin A , lamin C , progerin ( PG ) , LAP2α , LAP2β and GAPDH are indicated . ( C ) Growth curve of fibroblasts expressing control vector or progerin in the presence or absence of doxycycline-inducible LAP2α . LAP2α expression was induced by addition of 0 . 25 µg/ml doxycycline . Dotted lines indicate SEM ( n = 3 ) . Inset: growth rate after 5 days , error bras indicate SEM ( **p < 0 . 01 , Student's t-test ) . ( D ) Western blot of primary fibroblasts carrying doxycycline-inducible v5-LAP2α , expressing control vector or progerin . LAP2α expression was induced by addition of 0 . 25 µg/ml doxycycline . LAP2α , progerin and actin are indicated . ( E ) Western blot showing doxycycline-dependent induction of v5-LAP2α in TERT+HGPS cells . ( F ) Box plot of H3K27me3 levels in human fibroblasts expressing progerin in the presence ( red ) or absence ( blue ) of ectopically expressed LAP2α ( Student's t-test , p < 0 . 05 , n > 7500 cell analyzed , whiskers represent 10–90 percentile . ( G ) Scatter plot analysis of H3K27me3 levels in TERT+HGPS cells in the presence ( red ) or absence ( blue ) of ectopically expressed LAP2α showing increased levels of H3K27me3 upon LAP2α induction ( Student's t-test , p < 0 . 01 , n > 4000 cell analyzed . Inset: box plot of the same data , whiskers represent 10–90 percentile , ***p < 0 . 001 ) . ( H ) Growth curve of control ( ctrl ) or progerin expressing fibroblasts ( +progerin ) in the presence or absence of doxycycline-inducible v5-LAP2β . Dotted lines indicate SEM ( n = 3 ) . Inset: growth rate after 6 days , error bars indicate SEM ( **p < 0 . 01 , Student's t-test ) . ( I ) Western blot of fibroblasts transduced with doxycycline-inducible LAP2β and expressing control vector or progerin . LAP2β , progerin and actin are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 01310 . 7554/eLife . 07759 . 014Figure 5—figure supplement 1 . Depletion and expression of LAP2 in wild type and progerin expressing fibroblasts , control experiments . ( A ) Growth curve of primary and TERT+ control and LAP2 shRNA expressing fibroblasts . Dotted lines indicate SEM ( n = 3 ) . Inset: growth rate after 7 days . ( B ) Immunofluorescence microscopy analysis of lamin A/C and LAP2 levels in cells expressing scrambled ( ctrl ) or LAP2-specific shRNA ( shLAP2 ) . Scale bar: 20 μm . ( C ) Western blot of wild type cells expressing scrambled control shRNA ( ctrl ) or LAP2 shRNA ( shLAP2 ) . The LAP2 antibody recognizes both α and β-isoforms of LAP2 . Lamin A , lamin C and actin loading control are indicated . ( D ) Growth curve of normal ( ctrl ) and LAP2-depleted ( shLAP2 ) TERT+ fibroblasts expressing progerin or lamin A . Dotted lines indicated SEM ( n = 3 ) . Inset: growth rate after 7 days . ( E ) Immunofluorescence microscopy showing dose-dependent induction of v5-LAP2α and its nucleoplasmic localization . Antibodies: v5 , lamin B1 ( LB1 ) , merged + DAPI . Doxycyclin concentrations are indicated on the left . Scale bar: 50 μm . ( F ) Growth curve of primary fibroblasts expressing varying amounts of v5-LAP2α ( doxycyclin concentration: 0 , 0 . 25 , 0 . 5 and 1 µg/ml ) . Dotted lines indicate SEM ( n = 3 ) . Inset: growth rate after 7 days ( *p < 0 . 05 , **p < 0 . 01 , errors bars indicate SEM ) . ( G ) Western blot showing dose ( doxycyclin ) -dependent induction of v5-tagged LAP2α in normal dermal fibroblasts . Antibodies recognizing v5-tag , LAP2α and GAPDH are shown . ( H ) Box plot of H3K27me3 levels in human TERT+ fibroblasts expressing LAP2α ( or non induced control ) , whiskers represent 10–90 percentile ) . ( I ) Western blot showing doxycyclin-dependent induction of v5-tagged LAP2β in normal fibroblasts . Western blot was probed with an antibody recognizing LAP2α and LAP2β , GAPDH ( upper panel ) and v5-tag ( lower panel ) . ( J ) Immunofluorescence microscopy showing doxycyclin-dependent induction of v5-LAP2β and its localization to the nuclear periphery . Antibodies: v5-tag ( green ) , lamin B1 ( LB1 , red ) , merged + DAPI . Scale bar: 20 μm . ( K ) Growth curve of normal fibroblasts in the presence ( +DOX ) or absence ( no DOX ) of ectopic LAP2β . Dotted lines indicate SEM ( n = 3 ) . Inset: growth rate after 7 days ( *p < 0 . 05 , errors bars indicate SEM , Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 01410 . 7554/eLife . 07759 . 015Figure 5—figure supplement 2 . Expression of LAP2α prevents progerin induced DNA damage and premature senescence . ( A ) Quantification of progerin-induced 53BP1 DNA damage foci in response to expression of v5-LAP2 α levels ( p < 0 . 001 , n > 300 , χ2 test ) . ( B ) Percentage of senescence-associated β-gal-positive cells in control or progerin expressing cells in the presence or absence of ectopic LAP2α ( n = 3 , error bars indicate SEM , *p < 0 . 05 , **p < 0 . 01 , two-way ANOVA with Tukey's post-test ) . ( C ) Brightfield ( BF ) and phase contrast ( PH ) images of control or progerin expressing cells ± ectopic LAP2α . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 01510 . 7554/eLife . 07759 . 016Figure 5—figure supplement 3 . Effects of valproic acid treatment on proliferation of control ( non-induced WT , green lines ) and progerin-expressing ( red lines ) normal dermal fibroblasts . Concentrations of valproic acid 0 . 025–0 . 5 μM are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 016 Since depletion of LAP2 enhances the progerin-induced reduction in proliferation , we investigated whether ectopic expression of specific LAP2 isoforms would ameliorate progerin-induced defects . In normal cells , increased expression of LAP2α impairs cell proliferation in a dose-dependent manner ( Figure 5—figure supplement 1E–G ) , as previously reported ( Dorner et al . , 2006 ) . We then introduced progerin into cells expressing exogenous LAP2α ( Figure 5C , D ) . Strikingly , a moderate increase in LAP2α levels almost completely restored the proliferative capacity of cells expressing progerin ( Figure 5C , red arrowhead ) , reduced the number of cells with DNA damage foci ( Figure 5—figure supplement 2A ) and prevented premature senescence in these cells ( Figure 5—figure supplement 2B , C ) . This is in contrast to wild-type cells ( or vector control cells ) , in which increased LAP2α impaired proliferation ( Figure 5—figure supplement 1E–G ) and increased premature senescence ( Figure 5—figure supplement 2B ) . To determine to which extent LAP2α prevented progerin-induced defects , we used immunofluorescence microscopy to measure LAP2α and H3K27me3 levels in single cells upon LAP2α induction . As shown in Figure 5F , expression of LAP2α in progerin expressing cells increased H3K27me3 levels as compared to non-induced controls ( n = 3 , p < 0 . 05 ) . We further confirmed these results in TERT+HGPS fibroblasts ectopically expressing LAP2α ( Figure 5E , G ) . Scatterplot analysis of ≈5500 nuclei revealed a significant increase in H3K27me3 upon induction of LAP2α ( Pearson r = 0 . 1835 , p < 0 . 01 ) ( n = 3 , p < 0 . 001 ) . However , overexpression of LAP2α did not ameliorate progerin induced nuclear abnormalities in TERT+HGPS fibroblasts ( Figure 1—figure supplement 2B ) . Lastly , overexpression of LAP2α in wild type cells , not expressing progerin , did not result in a significant increase in H3K27me3 levels ( Figure 5—figure supplement 1H ) . Taken together , these results suggest that increasing LAP2α levels prevents progerin-induced proliferation defects and alleviates the progerin-induced reduction in the heterochromatin mark H3K27me3 . A reduction in heterochromatin renders DNA vulnerable to increased damage ( Di Micco et al . , 2011 ) . To determine whether ‘open chromatin’ renders cells more susceptible to progerin , we treated progerin-expressing , and their non-induced ( wild type ) controls , to increasing concentrations of the histone deacetylase inhibitor valproic acid ( 0 . 025–0 . 5 μM ) . As shown in Figure 5—figure supplement 2 , valproic acid treatment exacerbated progerin-induced proliferation defects , at concentrations that had no discernible effect on non-induced wild type cells ( Figure 5—figure supplement 3 ) . Based on these results we speculate that a reduction in heterochromatin may render cells more susceptible to progerin-induced proliferation defects . LAP2 exists in many isoforms . To determine whether the rescue by LAP2α is specific to this isoform , we expressed the β-isoform of LAP2 in normal- and progerin expressing cells ( Figure 5H , I; Figure 5—figure supplement 1I–K ) . In contrast to LAP2α , ectopic expression of LAP2β led to a reduction in the rate of proliferation in both normal and progerin expressing fibroblasts ( Figure 5H , red arrow ) . Together , these results demonstrate that expression of the α-isoform of LAP2 specifically ameliorates progerin-induced proliferation defects and increases the levels of heterochromatin associated H3K27me3 .
HGPS is described as a ‘segmental ageing syndrome’ , but it remains unclear why specific tissues are more affected , in particular those of mesenchymal origin , while others , such as neural lineages , are seemingly spared ( Zhang et al . , 2011 ) . To address this , we developed a DOX-inducible expression system to regulate the levels and timing of progerin expression in isogenic primary and TERT-positive cells . By using this system , we find that progerin inhibits proliferation , causes DNA damage and entry into senescence in a dose-dependent fashion . These results suggest that progerin's detrimental effects become apparent only when levels reach a critical threshold . This provides a compelling explanation as to why tissues expressing relatively high levels of lamin A/progerin are central to the pathology of HGPS ( Jung et al . , 2012; Nissan et al . , 2012 ) . Previous studies showed that progerin-induced defects can be rescued by ectopic expression of TERT ( Kudlow et al . , 2008; Benson et al . , 2010 ) . However , it remained unclear whether physiological levels of TERT would suffice , and to what extent TERT prevents progerin's defects . Our results confirm that ectopic expression of TERT prevents progerin-induced DNA damage , proliferation defects , premature senescence and senescence-associated loss of lamin B1 ( Kudlow et al . , 2008; Benson et al . , 2010 ) . Moreover , by expressing progerin and lamin A in ESCs which express endogenous levels of TERT , and Tert−/− ESC , we demonstrate that physiological levels of TERT are sufficient to prevent progerin-induced proliferation defects and changes in gene expression . These results may be relevant to HGPS patients , as TERT expression during embryogenesis or in adult stem cell compartments may protect these cells from the detrimental consequences of progerin ( Wright et al . , 1996 ) . In this respect , HGPS may be quite different from dyskeratosis congenita ( DC ) , a premature ageing disorder that is caused by defects in telomerase that particularly affects stem cell maintenance ( Vulliamy et al . , 2004 ) . Here we have expanded the analysis as to what physiological and biochemical parameters are effected by progerin , and to what extent they are restored to normal levels by the simultaneous expression of TERT . Importantly , the microarray analysis revealed that the persistent loss of H3K27me3 in TERT-positive progerin expressing cells did not result in significant changes in gene expression . However in determining the levels of the heterochromatin marker H3K27me3 , in primary and TERT-positive cells , we found that TERT does not prevent progerin-induced loss of heterochromatin . This , in turn , suggests that the changes in gene expression in primary fibroblasts expressing progerin are a consequence of premature senescence , rather than a direct consequence of progerin expression . To determine the direct effects of progerin on other nuclear proteins , with the consequent impairment of cell proliferation and premature senescence , we compared the protein interactomes of lamin A and progerin . Previous studies showed that lamin A interacts with LAP2α ( Dechat et al . , 2000; Kubben et al . , 2010 ) . Our interactome analysis confirmed this interaction , but demonstrated that the physical interaction between LAP2α and progerin was significantly reduced . Furthermore , it had been suggested that LAP2α may transiently associate with telomeres during mitosis ( Dechat et al . , 2004 ) , and cells with a disrupted nuclear lamina show abnormal telomere localization ( Gonzalez-Suarez et al . , 2009; Taimen et al . , 2009 ) . We therefore compared telomere distribution , and their association with LAP2α , in HGPS and normal nuclei . We found that a significant proportion of telomeres can be found at the nuclear periphery , and that their distribution is not perturbed by progerin . This suggests that other progeroid mutations ( LMNA E145K ) , or loss of LMNA may have different effects on telomere localization ( Gonzalez-Suarez et al . , 2009; Taimen et al . , 2009 ) . Although telomere distribution was unaffected , we find that the association of LAP2α with telomeres is reduced in HGPS . To determine the functional relevance of these changes , we modulated LAP2α levels in normal and progerin-expressing fibroblasts . Depletion of LAP2 exacerbated the progerin-induced defects whereas increasing nuclear LAP2α levels prevented the proliferation defects , DNA damage and premature senescence resulting from progerin expression . This rescue was specific to the α-isoform of LAP2 , as increasing LAP2β levels failed to restore normal rates of proliferation . Moreover , increased or decreased levels of LAP2α led to diametrically opposed effects between normal and progerin-expressing cells . Taken together , our data provide mechanistic evidence that LAP2α plays a key role in the HGPS pathophysiology and that progerin-induced defects are rescued by ectopic expression of LAP2α . Changes in the levels of LAP2α are emerging as a consistent feature in at least some of the laminopathies . Loss of Lmna , that causes cardiomyopathy and muscular wasting/dystrophy results in a transient increase in LAP2α in muscle precursors and myoblasts ( Melcon et al . , 2006 ) . However , ablation of LAP2α in Lmna null mice significantly reduces the severity of the disease and increases longevity ( Cohen et al . , 2013 ) , a result similar to that observed following elimination of another nuclear envelope protein Sun1 ( Chen et al . , 2012 ) . Together these observations , and those presented here , suggest that the lamina/nuclear envelope forms an integrated and mutually regulated stoichiometric network of proteins . When the network is disrupted , for example by LMNA mutations , one of the consequences is that the levels of other protein components change , with this change being a major contributing factor to the consequent pathology . Within this context , and based on the results presented here , we propose a model linking progerin and LAP2α to human telomeres ( Figure 6 ) . In this model , telomeres in normal cells are surrounded by LAP2α complexes , which in turn interact with nuclear lamins ( Figure 6A ) . In progeria , the impaired association between LAP2α and progerin disrupts this organization and ultimately leads to premature senescence ( Figure 6B ) . This can be prevented by expression of either telomerase or by increasing the levels of LAP2α ( Figure 6C , D ) . However , it is not clear if the effect of TERT mechanistically differs from the rescue by LAP2α . Our data suggest that this indeed might be the case: in contrast to LAP2α , TERT expression does not ameliorate the progerin-induced reduction in H3K27me3 levels . However , it remains to be investigated whether progerin affects H3K27me3 specifically at telomeric- or subtelomeric regions , and how expression of LAP2α ameliorates this loss . Telomeres are fragile sites and a reduction in heterochromatin has been associated with increased susceptibility to DNA damage and cell cycle arrest ( Sfeir et al . , 2009; Di Micco et al . , 2011 ) . In agreement with this hypothesis , progerin expression destabilizes chromosome ends , causing DNA damage and premature senescence ( Benson et al . , 2010; Wood et al . , 2014 ) . By elongating telomeres , TERT counteracts such telomere loss and prevents the telomeric DNA damage associated with HGPS ( Figure 6E; Figure 1—figure supplement 1I , J ) ( Kudlow et al . , 2008; Benson et al . , 2010 ) . In contrast to TERT , we found that increased LAP2α alleviates the progerin-induced loss of H3K27me3 ( Figure 5F , G ) . Chromatin decondensation by treatment with valproic acid enhanced progerin-dependent proliferation defects . Taken together , our results suggest that increased LAP2α stabilizes chromatin structure by increasing H3K27me3 and prevents progerin-associated DNA damage that resulted in premature senescence ( Figure 6E ) . In support of this notion , both LAP2α and H3K27me3 decline during normal ageing in human cells ( Scaffidi and Misteli , 2006; Naetar et al . , 2007; Dreesen et al . , 2013 ) , and it has been suggested that sustaining heterochromatin levels may extend lifespan by protecting against DNA damage ( Larson et al . , 2012 ) . Our results have important implications in understanding the role of the nuclear architecture , in particular the lamina and nuclear envelope , in regulating cell proliferation , chromatin organization and in providing novel insights into the molecular pathology of progeria . They may also be relevant to other laminopathies that are associated with perturbations or mutations of components of the nuclear lamina . 10 . 7554/eLife . 07759 . 017Figure 6Putative model of lamin A/progerin::LAP2α interaction mechanism with telomeres . ( A ) Normal formation of lamin A::LAP2 complexes allows proper positioning of LAP2α near telomeres ( green ) . ( B ) Perturbed progerin::LAP2α interaction impairs LAP2α localization at telomeres . ( C ) Telomeric damage resulting from this impaired interaction can be rescued by expression of telomerase ( gray ) or ( D ) by supplementing cells with exogenous LAP2α ( orange ) . ( E ) Progerin-induced H3K27me3 loss is prevented by ectopic LAP2α expression . TERT expression does not prevent progerin-induced loss of H3K27me3 but rescues telomere dysfunction by telomere elongation . DOI: http://dx . doi . org/10 . 7554/eLife . 07759 . 017
Normal primary dermal fibroblasts were a gift from Dr Bruno Reversade ( Institute of Medical Biology , A*STAR , Singapore ) . Fibroblasts were grown under standard culture conditions ( 37°C; 5% CO2 ) in minimum essential medium ( MEM; Invitrogen , Carlsbad , CA ) supplemented with 50 U/ml penicillin and streptomycin ( Invitrogen ) , 15% fetal calf serum ( FBS , Invitrogen ) , 0 . 2 mM non-essential amino acids ( NEAA , Invitrogen ) and 2 mM glutamine ( Invitrogen ) . For immunofluorescence microscopic analysis , fibroblasts were grown on Lab Tek Chambered Coverglass slides ( Nalge Nunc International , Rochester , NY ) for 4–7 days , fixed 10 min in 4% paraformaldehyde , washed in PBS , permeabilized using PBS + 0 . 3% Triton-X and blocked in PBS + 5% FBS + 1% BSA . Primary antibodies in blocking solution were incubated overnight at 4°C , washed in PBS , probed with secondary antibodies for 30–45 min at room temperature ( RT ) and DAPI ( 4 , 6 diamidino-2-phenylindole ) stained in PBS for 5 min . Images were acquired using a Zeiss Axiovert 200M inverted microscope ( Carl Zeiss International , Singapore ) using 10× NA 0 . 3 ZEISS Plan-NeoFluar , 40× NA 0 . 60 Ph2 Korr LD-Plan-NeoFluar or 63× NA 1 . 4 oil DIC ZEISS Plan-Apochromat objectives and a AxioCam MRm . Images were processed and exported using AxioVision LE software 4 . 5 SP ( 2006 ) . Images were cropped and figures assembled using Adobe Photoshop CS4 and Adobe Illustrator CS3 . DNA damage ( by 53BP and γH2A-X staining ) was quantified by scoring 350–500 cells for each cell line and condition . Confocal images were acquired on an upright Olympus FV-1000 confocal microscope using a 100× oil objective . pBABE-Neo-hTERT ( Counter et al . , 1998; Hahn et al . , 1999 ) were obtained from Addgene . Full length lamin A , progerin and LAP2β were amplified from a cDNA library ( human embryonic stem cell line H9 ) . LAP2α was amplified from pTD15 ( gift from Dr Roland Foisner , Vienna ) . cDNAs were cloned into retroviral vector pBABE-hygro ( Addgene , Cambridge , MA ) or doxycyclin-inducible lentiviral vector pTRIPZ ( Open Biosystem , Singapore ) . Restriction sites and v5-tag were introduced during PCR amplification step . Retroviruses were generated and fibroblast cultures infected using standard procedures . Lentiviruses were generated according to manufacturer's protocol ( OpenBiosystem ) . Doxycline-dependent expression was verified by western blotting , immunofluorescence and FACS analysis . DNA damage foci were detected using antibodies against 53BP1 ( Novus Biologicals; NB100-304 ) and anti-phospho-Histone H2A-X ( Ser139 ) ( Millipore; 05-636 ) , lamin B1 ( YenZym ) , lamin A/C ( Millipore; MAB3211 ) , progerin ( Santa Cruz , SC 81611 ) , LAP2 ( Santa Cruz , H-130 ) , LAP2α ( Abcam , Ab 5162 ) , TRF1 ( Abcam 10579 ) , V5-tag ( Invitrogen; 37-7500 ) , myc ( Santa Cruz , sc-40 ) , GAPDH ( Sigma; G9545 ) , β-tubulin ( Covance; MRB 435P ) , β-actin ( Sigma; A5441 ) . DOX-induced and non-induced cells were seeded in triplicates , grown for 3–5 days , trypsinized and counted using a Scepter cell counter ( Millipore ) . Experiments were repeated 2 , 3 and 4 weeks after doxycycline induction . Growth curves were performed at least in triplicates using the xCELLigence System ( Roche , Basel , Switzerland ) . Cell Index was monitored at hourly intervals . Tert−/− embryonic stem cells were derived by crossing heterozygous TERT-deficient mice ( Jackson Laboratories; B6 . 129S-Terttm1Yjc/J ) . Embryonic stem cells were isolated from day 4 blastocyst stage embryos according to previously published protocols ( Wong et al . , 2010 ) and allowed to hatch out for 5 days . Outgrowths were dissociated and embryonic stem cells were expanded in KO-DMEM media supplemented with leukemia inhibitory factor ( LIF ) , under standard culture conditions ( 37°C; 5% CO2 ) . Two wild-type , two homozygous and four heterozygous mouse ESC lines were genotyped using primers and conditions provided on the Jackson Lab website ( http://jaxmice . jax . org/strain/005423 . html ) . Bruce4 mouse embryonic stem cells ( derived from C57BL/6 mouse strain ) were grown under standard culture conditions ( 37°C; 5% CO2 ) in 90% Knockout DMEM high glucose medium ( Gibco , Waltham , MA ) , supplemented with 10% FBS , L-glutamine ( 2 mM , Gibco ) , pen strep ( 100 U/ml , Gibco ) , mercapto-ethanol ( 100 μM , Gibco ) , human LIF ( 10 ng/ml , Millipore ) , BIO GSK3-I ( 2 μM final , Calbiochem , Singapore ) . Differentiation was induced by removing leukemia inhibitory factor ( LIF ) from the culture medium as previously described ( Keller , 1995 ) . Embryoid bodies were generated using the hanging drop method , by aggregating 400 cells in 20 μl drops as described previously ( Dang et al . , 2002 ) . To induce differentiation , embryoid bodies were grown for 6 days prior to plating onto gelatin coated dish in standard ESC growth medium without LIF . This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health . All of the animals were handled according to approved institutional animal care and use committee ( IACUC ) protocols ( 140960 ) of the Institute of Medical Biology , A*STAR , Singapore . Whole cell lysates were isolated using Complete Lysis-M solution kit ( Roche ) , quantified using the Pierce Microplate BCA protein assay kit ( Thermo Scientific , Waltham , MA ) , separated by SDS-PAGE and transferred onto nitrocellulose membranes . Membranes were blocked for 1 hr in Odyssey Blocking Buffer:PBS ( 1:1 ) ( LI COR Biosciences , Lincoln , NE ) and hybridized with antibodies overnight at 4°C . Membranes were washed in PBS and two color detection was carried out using Odyssey Infrared ( IR ) -labeled secondary antibodies . A LI-COR Odyssey scanner was used to scan membranes and quantify signals . Primary and TERT+ fibroblasts stably expressing pTRIPZ-v5-lamin A or pTRIPZ-v5-progerin were grown in triplicates for 28 days in the presence or absence of 1 μg/ml doxycycline . At each time point , cells were grown to confluency and serum starved for 24 hr . Total mRNA was isolated using RNAeasy mini kit ( Qiagen , Singapore ) and integrity of the RNA was verified using Agilent 2100 Bioanalyzer ( Agilent , Singapore ) . cRNA was synthesized using the Ambion Target Amp kit ( Ambion ) according to the manufacturer's protocol , and cRNA from each sample was hybridized to BeadChip v2 chips ( Illumina ) . Lamin A and progerin were N-terminally tagged with the myc-BirA* biotinylation enzyme and cloned into the pTRIPZ lentiviral vector . Primary and TERT+ fibroblasts stably expressing either construct were generated by lentiviral transduction and selected with 1 . 0 μg/ml puromycin . The myc-BirA*-fusion constructs were expressed upon induction with doxycycline for at least 6 days prior to analysis . Induction was verified by western blotting and by immunofluorescence microscopy using an anti-myc antibody ( Santa Cruz , sc40 ) . 50 µM biotin was added to the medium for 24 hr prior to lysing cells under denaturing conditions ( M-lysis buffer; Roche ) . Control cells not induced with doxycycline , or without addition of biotin were processed in parallel . Biotinylated proteins were purified using streptavidin-coupled magnetic beads ( Invitrogen ) . After reduction and alkhylation , purified proteins were separated by SDS-PAGE electrophoresis and analyzed by mass spectrometry . Proteins were quantified using the Exponentially Modified Protein Abundance Index ( emPAI ) , which is directly proportional to the abundance of a protein in a mixture ( Ishihama et al . , 2005 ) . V5-tagged lamin A , v5-tagged progerin and LAP2α cDNAs were cloned into pcDNA 3 . 1 ( Invitrogen ) , and corresponding proteins were individually translated in vitro using the TnT quick coupled transcription/translation system ( Promega ) according to manufacturer's protocol . After translation , 20 μl of each produced protein was mixed with respective partners as indicated in the figure legend . 4 μg of anti-v5 antibody ( Invitrogen ) was added to each protein mix and incubated for 12 hr at 4°C with 50 μl of protein–G coupled Dynabeads ( Invitrogen ) in PBS . Beads were washed twice for 10 min in 0 . 5% sodium deoxycholate , 150 mM NaCl , 1% NP-40 , 0 . 1% SDS , 50 mM TRIS pH7 . 4 with proteinase inhibitors , and once for 10 min in 20 mM TRIS HCl pH 7 . 4 with proteinase inhibitors . Protein complexes retained by the anti-v5 coupled beads were then eluted in Laemmli BioRad buffer at 95°C , run on SDS page gels and analyzed by western blotting using antibodies against v5-tag and LAP2α . Recovered amounts of LAP2α were quantified and normalized to the v5-tagged proteins ( lamin A and progerin ) . Cells were fixed in 2% formaldehyde and 0 . 2% glutaraldehyde for 5 min at room temperature , washed twice in PBS and incubated for 6 hr in 5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside as described previously ( Dimri et al . , 1995 ) . Cells were grown on microscopy cover glasses in 6-well plates , fixed in 2% paraformaldehyde in PBS for 20 min at room temperature and incubated in 50 mM NH4Cl/PBS ( 5 min ) and 1% Triton X-100/0 . 1% SDS ( 5 min ) according to a previously published protocol ( Dechat et al . , 2004 ) . Samples were blocked in 0 . 2% gelatin/PBS for 30 min prior to antibody incubation . Primary and secondary antibodies were applied in gelatin/PBS for 1 hr at room temperature , washed in PBS and post-fixed in 2% paraformaldehyde in PBS for 20 min at room temperature . Acquisition was performed using a DeltaVision OMX v4 Blaze microscope ( GE Healthcare , Singapore ) , with the BGR-FR filter drawer for acquisition of 3D-SIM images . Olympus Plan Apochromat 100×/1 . 4 PSF oil immersion objective lens was used , with liquid-cooled Photometrics Evolve EM-CCD cameras for each channel . 15 images per section per channel were acquired with a z-spacing of 0 . 125 µm ( Gustafsson et al . , 2008; Schermelleh et al . , 2008 ) . Structured illumination reconstruction and wavelength alignment was completed using the SoftWorX ( GE Healthcare ) program . 3D image rendering and analysis was performed using Imaris version 7 . 6 ( Bitplane an Oxford Instruments Company ) , Tango ( Ollion et al . , 2013 ) and 2D image analysis using Fiji ( Schindelin et al . , 2012 ) , and CellProfiler ( Carpenter et al . , 2006 ) . Data and statistical analyses were performed using Excel and Graphpad Prism software . Results are shown as mean ± S . E . M/SD and box plots whiskers indicate 10–90 percentile , unless otherwise indicated . Data were analyzed using one or two way ANOVA and Bonferroni's/Tukey's post-hoc test if required , as well as two tailed Student's t-test and Pearson correlation coefficients , as appropriate . p-values below 0 . 05 were considered significant . | Hutchinson-Gilford Progeria Syndrome ( HGPS ) is a rare genetic disease in which individuals age prematurely . Newborns appear normal at birth , but start ageing rapidly when they are around a year old . Symptoms of the disease include stunted growth and joint stiffness , and individuals often die of heart failure during their teens . A mutated version of a protein called lamin A causes HGPS; this mutant is known as progerin . In cells that produce progerin , the ‘telomeres’ that protect the ends of chromosomes ( the structures that contain most of the cell's DNA ) from damage , are unusually short . Every time a cell divides , the telomeres get shorter . If they get too short , the DNA is damaged and the cell stops dividing and enters a state known as senescence . HGPS affects some of the tissues in the body more severely than others , and these tissues tend to produce high levels of progerin . By gradually raising the levels of progerin in human cells , Chojnowski et al . found that DNA damage and cell senescence only occur when the amount of progerin in a cell exceeds a particular threshold . Moreover , the expression of telomerase—a complex that can elongate telomeres—prevented progerin-induced DNA damage and premature senescence . To find out how progerin affects cells , Chojnowski et al . compared how lamin A and progerin interact with other proteins . This revealed that progerin interacts with a protein called LAP2α more weakly than lamin A . LAP2α normally associates with telomeres , but using super-high resolution microscopy , Chojnowski et al . observed that this association is less likely to occur in the cells of people with HGPS . Importantly , increasing the amount of LAP2α in progerin-expressing cells prevented DNA damage and senescence and enabled these cells to continue dividing . Chojnowski et al . propose that in HGPS , the weak interaction between LAP2α and progerin disrupts how LAP2α interacts with telomeres , which prevents cells from dividing . Understanding this process may help to design new ways of treating HGPS , and may also help us to understand other diseases that are caused by mutations in lamin proteins . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2015 | Progerin reduces LAP2α-telomere association in Hutchinson-Gilford progeria |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.